Developmental toxicity testing for safety assessment: new approaches and technologies

Abstract

The ILSI Health and Environmental Sciences Institute's Developmental and Reproductive Toxicology Technical Committee held a 2-day workshop entitled “Developmental Toxicology—New Directions” in April 2009. The fourth session of this workshop focused on new approaches and technologies for the assessment of developmental toxicology. This session provided an overview of the application of genomics technologies for developmental safety assessment, the use of mouse embryonic stem cells to capture data on developmental toxicity pathways, dynamical cell imaging of zebrafish embryos, the use of computation models of development pathways and systems, and finally, high-throughput in vitro approaches being utilized by the EPA ToxCast program. Issues discussed include the challenges of anchoring in vitro predictions to relevant in vivo endpoints and the need to validate pathway-based predictions with targeted studies in whole animals. Currently, there are 10,000 to 30,000 chemicals in world-wide commerce in need of hazard data for assessing potential health risks. The traditional animal study designs for assessing developmental toxicity cannot accommodate the evaluation of this large number of chemicals, requiring that alternative technologies be utilized. Though a daunting task, technologies are being developed and utilized to make that goal reachable. Birth Defects Res (Part B) 92:413–420, 2011. © 2011 Wiley Periodicals, Inc.

INTRODUCTION

Scope of the Problem

Developmental toxicity is chemical perturbation during formative stages of the reproductive cycle affecting embryo and fetal development as well as general children's health. Although developmental toxicology has been a formal field for many decades, the testing paradigm for the prenatal regulatory safety assessment of drugs and chemicals has remained essentially the same for over 30 years, and is based on extrapolating risk from phenotypic outcomes in high-dose toxicity animal studies (Kimmel et al., 2006). Relatively few human developmental toxicants are known, suggesting either that compounds with developmental toxicity potential are being kept out of commerce (i.e., the traditional testing strategy that forms the basis for our current regulatory system is technically sound) or that compounds are used at levels below which adverse effects occur.

Even if the current testing strategy is sound and effective at preventing false negatives, throughput is low, slow, and costly—both in terms of fiscal and animal resources (Hartung, 2009). These restrictions result in a relatively low number of compounds that have sufficient in vivo data to assess the potential for adverse effects on human development. NIEHS' National Toxicology Program (NTP) online database, for example, provides developmental effects data on only about 3% of the listed chemicals (70 of 2,330). Other databases are similarly sparse for developmental (or reproductive) effects including FDA's Center for Drug Evaluation and Research publicly accessible database (16.3%; 58 of 355 listed compounds), and FDA's Center for Food Safety and Nutrition database (27.2%; 312 of 1,146 listed compounds; provided in Leadscope Databases (Leadscope) (Chihae Yang and Ann Richard, personal communication; see also Singh et al., 2010). The EPA Integrated Risk Information System (IRIS) contains comprehensive reviews for 553 environmental chemicals (as of April 2010), and identifies the most sensitive or “critical effect” as the basis for setting safe exposure levels to protect the public health. The critical effect is the first observed effect deemed adverse that is likely to occur in the most sensitive species as the dose rate of an agent increases (IRIS, 2010). Less than 2% of 533 IRIS assessments report the critical effect for the derivation of a noncancer reference value (i.e., a safe exposure level) as being a developmental (5 of 553) or reproductive (4 of 553) effect [http://www.epa.gov/IRIS/]. This may be due to other effects being more sensitive, but more likely due to a lack of developmental and/or reproductive effects data, which contributed to an increased uncertainty in the database for the choice of the critical effect, and resulted in a lower reference value in 85% of the cases where an uncertainty factor for an inadequate database was used.1 Finally, in one of the largest data compilations from multiple resources to-date, EPA's Aggregated Toxicology Resource (ACToR) identified available developmental toxicity data for less than 30% of the 9,912 chemicals in commerce or of environmental interest, out of a chemical domain of 418,513 generic chemicals (Judson et al., 2009).

The regulatory need to efficiently and effectively assess potential health risks for 10,000 to 30,000 chemical compounds in broad use world-wide is a daunting challenge in light of the small fraction which have been tested to-date (Judson et al., 2009). This an especially complex challenge with respect to assessing the potential for prenatal developmental toxicity where the current paradigm is largely based on traditional animal study outcomes (Hurtt et al., 2003; Carney et al., 2007). Hazards to prenatal development are currently assessed based on guideline studies with standardized animal protocols. These studies are designed to provide general information concerning the effects of prenatal exposure on the pregnant animal and developing conceptus of two species, usually rat and rabbit, in dams/does orally dosed across three dose levels and a control. The dosages are spaced to produce a gradation of toxicities from no evidence of either maternal or developmental toxicity at the lowest dose tested to overt toxicity at a dose high enough to produce effects on the mother without death or severe suffering, or on her unborn litter scored for intrauterine death (resorptions, pregnancy loss), structural abnormalities (malformations, abnormalities), and fetal weight reduction. A number of factors contribute to uncertainty when extrapolating results from these descriptive animal studies to humans. Some of the major causes of uncertainty are unknown species differences in sensitivity to the compound, in the difficulty of extrapolation of the specificity of the effect 1:1 between species, and differences in the level and duration of exposure between the high-dose, short term, exposures test animals receive compared with the low-dose, long term, maternal exposures that are likely to occur in humans from exposure to environmental chemicals (Kimmel et al., 1993). Additional limitations in extrapolating the results from animal studies to predict effects in humans include incomplete understanding of the mechanisms leading to an adverse effect, and species differences in chemical disposition and toxicity targets (Daston, 2007). Because most developmental defects have multifactorial etiology, this understanding is needed for an appropriate intervention and preventive public health strategy.

In the traditional in vivo animal testing paradigm, most of the data come from pregnant rat and rabbit bioassays evaluated at term. The bioassays are not intended to identify molecular targets, mechanisms or pathways of toxicity. Thus, the assessments based on these bioassays draw from science that precedes significant research advances in molecular embryology that have been made over the last 30 years (National Research Council, 2000). More recent in vitro systems take advantage of new technologies for mechanistic evaluation, can be run quickly on many chemical compounds, and are less reliant on animal usage (Chapin et al., 2008). Automated technologies for high-throughput screening (HTS), high-content screening (HCS) and computational (in silico) methods motivate a paradigm shift for testing developmental toxicity that emphasizes drug or chemical interactions with sensitive molecular targets and the identification of the relevant biological pathways that lead to the endpoints of toxicity observed in traditional in vivo bioassays (National Research Council, 2007).

Developmental Complexity

There are unique challenges in implementing the new paradigm to assess developmental toxicity with alternative model systems, HTS and HCS technology platforms, and computational systems biology. High on the list is resolving the complex biology inherent in the dynamics of a system within a system (i.e., the embryo within the mother). Traditional test results are relatively uninformative for understanding the underlying biological complexity and for reducing the uncertainties in predicted outcomes, especially in humans, due to: (1) species differences in response (biological susceptibility), (2) variability in outcomes depending on when exposure occurs (windows of vulnerability, whereby a chemical may or may not be toxic depending upon the time of exposure and whether its target is a critical event at that particular stage in the developing fetus), and (3) uncertainties in extrapolating outcomes from the high-dose, short-term animal bioassay exposure regimens to the more common low-dose chronic exposure scenarios experienced in humans.

Traditional test protocols also call for doses to the pregnant animals sufficient to cause maternal toxicity to assure that the top end of the dose–response curve has been reached, and that sensitivity can be characterized during critical periods of organ system development. The direct effect of toxicant exposure on the pregnant dam and the consequences of maternal disruption to the developing fetus are sometimes considered by comparing the lowest effective dosage having observed maternal and/or fetal consequences. However, some chemical compounds may induce developmental toxicity in a bioassay of one species at a dose level that is primarily toxic to the mother in another species (Janer et al., 2008). Furthermore, positive results may be difficult to extrapolate to the human situation where an embryo may become exposed to low concentrations of the chemical compound for prolonged periods. The biological dynamics that are important to understand, and that are the focus of the newer approaches, can be organized under the following headings presented in Figure 1.

Figure 1. Factors adding to the complexity of testing developmental toxicity.

DEVELOPMENTAL SIGNALING AND GENOMIC RESPONSE

Conservation of Cell Signaling

Morphogenesis and differentiation require precisely orchestrated genetic signals and cellular rearrangements during body plan development. Key elements of fundamental signaling and regulatory pathways are prone to perturbation by genetic errors and environmental disruptions, leading to a variety of disease conditions in humans that are functionally recapitulated by homologous pathways, such as vesicle trafficking in the development of simpler model organisms such as Drosophila (Chai et al., 2008). Dr. Giusy Pennetta, an embryologist from the University of Edinburgh, presented at the workshop recent updates in molecular embryology drawing from the principle of “conservation of cell signaling” across widely divergent species. Research in Drosophila has led to the identification and characterization of genes underlying the mechanisms of patterning early development leading from homogeneous egg to segmented embryo. Formation of the three primary axes, anterior-posterior (A/P), dorsal-ventral (D/V), and medial-lateral (R/L), results from polarized cell signaling. A genome-wide approach identified the archetypical molecules for body plan development in Drosophila. The determination of A/P axis formation is due to a cascade of interacting genes that act serially to subdivide the embryo into segments that then acquire their anatomical identity based on morphogenetic gradients that trans-activate the next layer of transcription factors. Gene regulatory networks regulate the system, and cytoskeletal components actuate the cellular consequences.

Genetic errors of the developmental program induced by specific mutations have been applied to the functional characterization of these pathways and gene regulatory networks in Drosophila. Importantly, corresponding genes have been discovered in mammalian genomes, including human, reflecting the conserved nature of cell signaling pathways important for development. The function of homeobox genes in mammals, for example, is the same as in flies: the knockout of Hoxc8 in mouse transforms the first lumbar vertebra to a thoracic vertebra, forming a supernumerary rib—a common developmental abnormality in rodents. Dr. Pennetta concluded her presentation by showing a few exemplars of chemical-perturbed cell signaling in early development (WNT, TGFβ, SHH, Notch-Delta (ND), receptor tyrosine kinases) with abnormal developmental consequences (Pennetta, ILSI-HESI workshop, 2009).

Signaling Pathways Critical for Embryonic Development

In 2000, the National Academy of Sciences (NAS) issued a report advocating the use of detailed knowledge about cell signaling pathways to help elucidate mechanisms in developmental toxicity (National Research Council, 2000). That report summarized a listing of molecules mediating cell signaling pathways across species and developmental processes as well as the qualitative relationships among the molecular components. A finite number of cell–cell signaling pathways (17) and stress response pathways (2) were named based on conserved roles in animal development. Included were cell–cell signaling pathways such as WNT, TGFβ, SHH, ND, and receptor tyrosine kinases among others. Focus on the canonical signaling pathways provided an initial framework for analysis of developmental processes in general; however, the question arises as to which of the major signaling pathways are relevant to developmental toxicity as potential “toxicity pathways” in the embryo, placenta, or mother.

High-content genomic assays used to profile biological responses at the molecular and cellular levels are increasingly being used as standard tools for analysis of toxicity pathways. Most commonly these technologies include gene arrays to probe the transcriptome and to classify experimental subjects by signatures of biological response that span multiple nonidentical conditions. Toxicogenomics is particularly relevant to assess the potential for developmental toxicity because the molecular targets of drugs and chemicals may come and go as the activity and function of conserved cell signaling pathways varies during gestation. As such, the vulnerability of different pathways and the susceptibility of the embryo to pathway-level perturbations will likely vary in response to the key factors addressed in the fundamental principles of teratogenesis (e.g., chemical, dose, mechanism, genetics, stage, and bioavailability).

Dose–Response Assessment With Toxicogenomics

The genome-wide characterization of embryos reacting to developmental toxicants can provide clues for quantitative dose–response modeling at a systems-level. Dr. George Daston, a developmental toxicologist from Procter & Gamble, discussed how genome-wide technologies can be applied to problems in developmental safety assessment. Using examples from his laboratory of chemicals acting through nuclear receptor pathways (e.g., estrogens and androgens) in the rat fetus, Dr. Daston demonstrated how toxicogenomics can be a strategy for predictive toxicology and a mechanistically based alternative to in vivo studies (Naciff et al., 2005). An “estrogenic fingerprint” of the uterotrophic responses from low-dose exposure to three estrogenic chemicals (ethynyl estradiol, bisphenol A, and genistein) was composed of 66 genes that were consistently and significantly regulated in the same direction by all three chemicals. This estrogenic fingerprint was diagnostic for several weaker environmental estrogens as well (e.g., 31 of 66 genes were affected by DES, 12 of 66 genes by methoxychlor), but was not diagnostic for nonestrogenic chemicals (e.g., 2 of 66 genes affected by cortisol; and none of the genes affected by ethylenethiourea, propylthiouracil, or methimazole). Another example was presented of gene expression profiling in the fetal rat testis and epididymis that demonstrated greater sensitivity to environmental estrogens than morphological or biochemical endpoints with the potential for higher resolution detection of nonlinear responses at low dose exposures (i.e., below the low-effect- and no-effect-levels observed in animal studies).

Dr. Daston's lab further demonstrated the ability to track temporal patterns of morphological changes with a concurrent change in responses of genes. Samples were analyzed from the immature uterus of 20–day-old female rats sampled from 1 to 96 hr after exposure to ethynyl estradiol. The temporal changes in gene expression for different (relevant) suites of genes paralleled a concurrent increase in uterine weight starting with genes functionally annotated for coding transcription factors, cell signaling, vascular permeability, and growth factors; followed by genes for protein synthesis, regulation of cell growth-differentiation-death, tissue remodeling and pro-inflammatory responses. Dr. Daston concluded his presentation with evidence that further validated the estrogenic fingerprint in the Ishikawa cell line, providing further support for transcript profiling as a viable mechanistically based alternative to animal testing (Daston, ILSI-HESI workshop, 2009).

Candidate Signaling Pathways

Although genome-scale analysis has revealed quantitative dose–response relationships for non a priori pathways, would an a priori (candidate) pathway approach also reveal key information to predict developmental toxicity? One potential alternative model for developmental toxicity testing that can be scaled to a level of throughput envisaged in the new testing paradigm is the embryonic stem cell (ESC) system. This assay exploits the advantages of cell–cell interaction and conservation of cell signaling to detect dose–response activity of compounds on biological pathways and cellular processes critically important for morphogenesis and differentiation. The new testing vision is accommodated by providing a system that detects effects of compounds on undifferentiated or differentiating states of cells resembling an embryonic environment (albeit with restricted geometry) with the potential to assess human cell lines (within ethical guidelines) (Chapin and Stedman, 2009). Don Stedman, a Senior Principal Scientist at Pfizer Pharmaceuticals, presented their research demonstrating the mouse ESC system's ability to capture data on disruption of developmental signaling pathways as a potential alternative for assessing developmental toxicity. His example focused on the expression of genes for the 17 + 2 conserved signaling pathways critical to early development (National Research Council, 2000) taking the hypothesis that an abnormal activation or inhibition of signaling pathways can lead to developmental toxicity.

The test system uses murine ESCs cultured 3 days as hanging drops that form “embryoid bodies” with gene expression patterns for ectodermal, mesodermal, and endodermal lineages. Analysis of gene expression at 5 days revealed the top expressed signaling pathways as Cadherin, Wnt/β-catenin, Hedgehog, Integrin, ND, Nuclear Hormone, and Receptor Ser/Thr kinase (Fig. 2A).

Figure 2. Expression profile of canonical (NAS) signaling pathways during mESC differentiation. (A) In the upper heatmap, average quantitative PCR (QPCR) signal is compared for the percentage pathway change on days 3, 5, and 10 of differentiation compared with Day 0 (undifferentiated mESC signal); color scale: weak (green) to strong (red) pathway-level expression. Day 5 may be an excellent sampling time point to monitor perturbations. (B) In the lower heatmap, pathway-level mESC response to four conditions is shown: an endogenous BMP antagonist that promoted cardiomyocyte differentiation (Noggin), a teratogen that perturbs multiple pathways (Retinoic acid), and RNAi knockdown of BMP4 with clones 4.1 (85% knockdown at Day 0) and 4.3 (75% knockdown at Day 0). ESC, embryonic stem cell.

The 5-day interval proved to be optimal for analysis of susceptibility to perturbation of pathway-level gene expression on the basis of gene coverage, low variability, and cost factors. Based on pathway redundancy, Stedman and co-workers collapsed the Ingenuity Pathway Analysis developmental systems library into 253 representative genes, 65% of which were analyzed to test the hypothesis that perturbation of signaling pathways can lead to developmental toxicity of embryoid bodies. Exposure to retinoic acid, RNAi-based knockdown of BMP4, or the BMP antagonist “Noggin” revealed pathway-level disruptions based on treatment and dose. For example, expression of the ND pathway (9 genes) was sensitive to Noggin and BMP4, but not retinoic acid (Fig. 2B). This indicates that a candidate pathway-level analysis of gene expression can reveal the hierarchical activity of compounds on key morphogenetic signaling pathways in the ESC model (Stedman, ILSI-HESI workshop, 2009). A number of data repositories and resources have appeared in recent years that allow automatic retrieval and integration of information for the relevant pathways (Ganter et al., 2008; Bauer-Mehren et al., 2009). This has become particularly relevant in the effort to integrate experimental data on a broad range of cellular pathways, providing an understanding of function by taking the entire biological system into consideration.

CELLULAR PROCESSES AND TISSUE DYNAMICS

As a complex system, the ESC platform is revealing of multiple pathways acting together to precipitate an adverse effect; however, at a higher scale simple lesions may be propagated through the dynamics of embryogenesis into complex phenotypes or combinatorial lesions may converge to simple phenotypes. There are challenges to study development in a spatiotemporal context, and admittedly the ESC platform provides an arbitrary geometry. Free-living zebrafish embryos (ZFE) have been an active area of research due to the success of stable morpholino-based gene knockdown that can replicate developmental phenotypes in simple and combinatorial scenarios. These and newer antisense-RNA based technologies for functional analysis can be tapped to test the impact of activation or repression of key genes in specific pathways. Database projects such as Zebrafish Morpholino Database project and the Mouse Phenotype Ontology browser are key online resources for this kind of research. One might also consider the application of ZFE to assess the extent to which fine-scale phenotypic changes result from experimentally induced lesions or natural biological variation (e.g., developmental noise). In this regard, tracking the movements and history of individual cells in a growing embryo provides the opportunity to assess early biological consequences of genetic errors or environmental disruption on cellular dynamics using sophisticated cell imaging technologies.

Dynamical Cell Imaging

A striking example of mapping cellular dynamics to morphogenesis was presented by Ms. Annette Schmidt, from the European Molecular Biology Laboratory (Keller et al., 2008). She and coworkers tracked global and local cellular movements in the ZFE during cleavage, gastrulation, and early organogenesis by a recombinant histone 2B-enhanced green fluorescent protein nuclear reporter (Keller et al., 2008). Automated microscopic detection of labeled nuclei with digital scanned laser light sheet fluorescence microscopy and a parallelized image segmentation pipeline enabled the reconstruction of a complete “digital embryo” from 400,000 image slices obtained at 60–90 sec intervals across the first 24 hr of ZFE development.

Using this comprehensive database of cell positions, divisions, and migratory tracks, it was possible to establish a “morphogenetic blueprint” of specific systems such as gastrulation or optic vesicle formation. Digital scanned laser light sheet fluorescence microscopy technology in the ZFE permits a unique opportunity to detect and measure cell-level behaviors such as mitosis, cell-polarity, and lineage mapping during complex tissue movements. For example, the global quantification of cell divisions at discrete intervals of development showed the sudden loss of mitotic synchrony before the activation of the zygotic genome and measured the degree of variability of speed and direction of nuclear movements between individual embryos (e.g., 7.0% during somitogenesis). Ms. Schmidt also presented data cell counts during gastrulation in the nodal mutant one-eyed pinhead (oep). Although ∼1550 epiblast cells internalize in wild-type ZFE, this number is reduced to ∼60 cells in the mutant embryo, resulting in an overextended epiblast (Schmidt, ILSI-HESI workshop, 2009). Cell-based imaging provides an important level of detail absent from genome-based technologies that can be used to advance our understanding of cellular mechanisms in developmental toxicity.

HTS AND COMPUTATIONAL MODELING

Although numerous models have been developed for the qualitative representation of developmental signaling pathways, few models exist for quantifying the propagation of signals through biological networks and the regulation of coordinated cell behaviors in normal morphogenesis. Notable examples include quantitative models of the embryonic heart (Olson, 2006), periodic somite formation (Goldbeter and Pourquie, 2008), neural crest migration (Nikitina et al., 2008), and limb-bud development (Bénazet et al., 2009). These complex developmental processes may integrate signals from across several distinct signaling modules to coordinate multicellular activities. Computational models and algorithms have become essential to understanding critical biological processes at the mesoscale.

Computational (in silico) Models

The new testing paradigm outlined by the NAS shifts emphasis in the evaluation of toxicity from an outcome-based to pathway-based approach (National Research Council, 2007). Although we currently have the ability to measure molecular components of cellular and tissue level phenomena in great depth and detail, the vast data must be organized quantitatively into systems-level models that reveal the interaction of multiple components across time and space. As such, integrating HTS data into useful models will require a system-level understanding of developmental processes and toxicities (Knudsen and Kavlock, 2008). Dr. Tom Knudsen, a Developmental Systems Biologist at EPA's National Center for Computational Toxicology (NCCT), presented on challenges in building such models from the HTS and HCS data. The data engine, referred to as the ToxCast™ project (Judson et al., 2010), currently has data on 309 unique chemical compounds tested in over 467 cell-based and cell-free assays (in vitro) and anchored to >$2B worth of legacy data from over 2,073 guideline in vivo studies (www.epa.gov/ncct/toxrefdb/).

The strategy for predictive modeling of developmental toxicity presented by Dr. Knudsen is contingent upon these HTS and bioassay data, a knowledgebase that integrates pathway-level data with knowledge of normal embryology, and a simulation engine that models key morphogenetic events—together known as EPA's “Virtual Embryo.” Preliminary studies have begun to reveal inferred associations (predictions) for chemical-target and chemical-endpoint. For example, a significant association was identified for 12 chemicals that caused cleft palate and 37 in vitro assays mapping to five pathways: AhR, G-protein coupled receptors, glucocorticoid receptor, retinoid RAR/RXR receptor, and WNT signaling. The longer term goal is to incorporate these kinds of data into computer simulations of multicellular behavior that are simple enough so that the underlying mechanism for emergent behaviors from dynamic interactions among the components of the system can be understood and yet elaborate enough to generate novel and informative emergent behaviors (Knudsen, ILSI-HESI workshop, 2009). Such models can be exercised across conditions not practical experimentally due to cost, time, scale or complexity; and may someday help reduce uncertainty in risk assessment for developmental toxicity.

Translation Into Risk Assessment Strategies

The in vitro approaches that have been used in HTS of pharmaceutical compounds for efficacy and safety are now being explored as a future means to identify biological targets and pathways of toxicity for the many thousands of chemicals in broad use worldwide (Judson et al., 2010). Dr. Bob Kavlock, Director of the National Center for Computational Toxicology at the U.S. Environmental Protection Agency, presented this new testing strategy in summarizing a report issued by the National Research Council (2007) in terms of current challenges to improve the assessment of key exposures (life stages) and toxicity outcomes (neurotoxicity), state-of-the-science testing and assessment procedures (genomics, bioinformatics, pharmacokinetics), the efficient experimental design and reduced use of laboratory animals, new and alternative test methods, and the application of computational and molecular techniques in risk assessment.

The NAS report recommended a new paradigm to evaluate the biological activity of large numbers of chemicals utilizing in vitro assays largely derived from human cells (National Research Council, 2007). In this strategy, signatures of biological activity will be catalogued in HTS in vitro assays and correlated with endpoint toxicities to prioritize chemicals for in vivo testing, environmental exposure assessment, and epidemiology studies. Consistent with these goals, EPA's strategic plan for evaluating the toxicity of chemicals advocates “toxicity pathway” identification and screening, toxicity-based risk assessment, and understanding the tradeoffs moving from an outcome-based to a more mechanism-based risk assessment process (www.epa.gov/osa/spc/toxicitytesting/index.htm). The endorsement of an HTS paradigm was evidenced by the formation of a tripartite consortium, known as “Tox21” formed between the EPA's National Center for Computational Toxicology, the NIEHS NTP, and the NIH National Chemical Genomics Center (Collins et al., 2008).

Dr. Kavlock's presentation addressed the charge questions from the broader perspective of Tox21 and in particular the recent achievements in EPA's component, the ToxCast™ project (http://www.epa.gov/ncct/toxcast/). The first phase of this project employed 467 in vitro assays to test 309 unique chemicals, many of which have a rich amount of in vivo data from traditional animal studies (Judson et al., 2010). The HTS paradigm supports a mechanistic-based and pathway-based risk assessment process generating unbiased, inexpensive and broad-based views of chemical–biological interactions, and a new foundation for prioritization and testing. Dr. Kavlock also addressed the tradeoff of loss of detailed phenotypic observations in animal models on a few chemicals for considerably more information on more chemicals as we move from an outcome-based to a more mechanism-based risk assessment process. He also emphasized the important consideration that any new approach be at least as protective of human health as the traditional paradigm.

The new paradigm will address gaps in knowledge between the data generated in in vivo versus in vitro testing by developing approaches and technologies that identify and quantitate the emergent properties of tissues (multicellular models) and that provide appropriate dosimetry (qualitative and quantitative). Stakeholders' expectations must be managed through frequent communications and documentation sufficient to assure that computational models are transparent and address the most relevant data needs (Kavlock, ILSI-HESI workshop, 2009).

PATH FORWARD: CHALLENGES AND OPPORTUNITIES

The current testing strategy for toxicity in general and developmental-reproductive toxicity in particular is low-throughput, costly, and relies on human extrapolation of high-dose effects observed in animal models, mostly pregnant rats and rabbits. Scientific and practical needs for a different way to predict human toxicology have been recognized (National Research Council, 2007), and key matters for debate are:

• identify new testing paradigms that gather data from in vitro assays to profile the cellular and molecular activities of chemical compounds at defined concentrations;

• integrate these HTS/HCS data into kinetic models (PBPK) and tissue-level dynamic models (systems biology) that bridge the gap with key events leading to “apical endpoints” of in vivo toxicity;

• utilize human exposure data whenever available now or in the future.

This workshop examined key opportunities and challenges to implementing the new paradigm, including new approaches and technologies from basic research that can be more efficiently and effectively applied to improve chemical prioritization and risk assessment by increasing throughput (e.g., data on many more chemicals) and critical biology (e.g., building predictive models). Challenges for technology development derive from the need to anchor predictions from in vitro profiling to relevant in vivo endpoints for traditional animal studies or human data where they exist, and to validate pathway-based predictions with targeted studies in whole animal models. These challenges demand the application of newer information technology and management resources for integrative biology, specifically the systematic characterization of canonical signaling pathways that have transformed modern developmental biology into a quantitative science (Lewis, 2008; Oates et al., 2009).

One can reasonably ask to what extent has research devoted to the systematic characterization of canonical signaling pathways been fruitful to understanding developmental toxicity? A framework had been nicely portrayed by the NAS in their 2000 report (National Research Council, 2000). As demonstrated in this workshop, using female reproductive tract development as an example, a nonsupervised genome-scale analysis can reveal quantitative dose–response relationships for the expression of many genes studied in parallel. The altered genes can be systematically characterized in several dimensions. First and foremost is by mapping these genes to standard Gene Ontology terms for molecular function, cellular localization, and biological process. This systems-level analysis reveals the order and timing of pathway-level changes for a specific target tissue. These data, however, can only infer the average response measured across a heterogeneous field of many thousands of cells some of which may be disrupted by a chemical compound and others that could be reacting positively (adaptive) or negatively (bystander) to local tissue injury. This gives rise to a second important dimension for the systematic characterization of signaling pathways, which is to predict the tissue-level behavior of a developing tissue as a complex system. One anticipates local cellular behaviors are at some level governed by conserved signaling pathways and so it follows logically that a better understanding of the signaling pathways could lead to better understanding of developmental toxicity, perhaps understanding the system in sufficient depth and detail for risk assessment purposes.

Although the canonical NAS signaling pathways (National Research Council, 2000) provide a conceptual framework for this interpretation, the difficulty in addressing the role of chemicals on development is evident. One example of this difficulty was addressed in the meeting summary presented by Dr. Paul Foster of the NIH NTP, focusing on the toxicity of male reproductive tract development which is among the most sensitive system to environmental chemical disruption. Although the application of toxicogenomics to this research has gained some ground over the past decade, few studies have systematically characterized the canonical or signaling pathways following toxicant exposure. One study demonstrated similar gene expression patterns following either in utero exposure to the phthalate DBP, or postnatal exposure to the phthalate metabolite, MEHP, suggesting that the initial mechanism of fetal and prepubertal phthalate-induced testicular injury is shared (Lahousse et al., 2006). On the other hand, a major challenge in assessing risk following an in utero or postnatal exposure (chronic or acute) is that adverse outcomes such as decreased fertility or cancer may not be recognized until puberty and/or adulthood. Thus, the plausibility of identifying biological signaling networks associated with an altered developmental phenotype leading to such adverse male (and female) reproductive outcomes that manifest later in life remains a significant but worthwhile challenge for a pathway-based evaluation of risk that is primarily focused on in vitro profiling data. Clearly, there is a continued need for understanding how the HTS/HCS data can be brought into predictive models of developmental defects that may be programmed during prenatal or early life but that do not manifest until puberty or adult life.

How well does the hypothesis hold that genes critical to metabolic and signaling pathways in normal development (“hubs”) function in pathways of developmental toxicity? This forces us to think about the underlying design of biological systems—both in terms of metabolic and regulatory pathways. We are beginning to see signs from early predictive models of chemical-target and chemical-endpoint associations derived by statistical associations and machine learning algorithms. An example is the association between 12 chemicals that induced cleft palate in rats and in 37 in vitro assays in which these compounds were screened. It is interesting to note that the sensitive assays mapped to a number of pathways known to mediate cleft palate during traditional studies in experimental teratogenesis and developmental genetics. This provides early encouragement that in vitro profiling data have the potential to reveal changes in relevant metabolic and regulatory pathways. Then the question arises as to whether we pick these pathways as a starting point for determining toxicity, or adopt a more “open approach” to follow clusters of genes emerging from non a priori studies. An open approach may be more speculative but could identify “hubs” of biological activity that couple diverse signaling pathways, thus revealing how molecular regulatory information potentially flows through the system.

Then we must determine if the flow of molecular regulatory information, once disrupted by a toxicant, would initiate an adverse or adaptive response. We know already that cross-talk between key pathways underlies functional redundancy at a systems-level. As such, would certain pathways play more prominent roles in specific tissues or during exposure to specific chemicals or class of compounds? Different scenarios could be tested in a computer-simulated system, assuming we build biologically informed computer models that can be exercised with synthetic data and validated with real data, to predict when environmental disruption will lead to developmental phenotype in human embryos. To link pathway-level response with adverse outcome we need to know major components of the target “system” in terms of key cells and molecules, the relevant interactions among components (network state relations), and the dynamic behavior of components as the system evolves to an endpoint. In the future, virtual tissue models may allow scientists to analyze a simulated embryo for responses across environmental and genetic conditions not practical experimentally, to predict outcomes for targeted testing and further evaluation.

ACKNOWLEDGMENTS

This manuscript has been subjected to EPA review and ILSI review and approved for publication. The authors declare that they have no competing financial interests. We are grateful to Dr. Rob DeWoskin of EPA's National Center for Environmental Assessment for helpful comments and editing this manuscript.

• 1

  Increased uncertainty in the critical effect in an IRIS assessment due to inadequate or insufficient data is accounted for quantitatively by a database uncertainty factor (UF-D) that when greater than 1 (UF-D>1) further reduces the reference value EPA derives to protect the public health (i.e., the reference dose or reference concentration). Many of the older IRIS assessments had only a qualitative rating on the confidence in the database, instead of a UF-D. In these older assessments, a designation of low or medium confidence was often due to missing reproductive or developmental studies. For assessments where a UF-D was used, a search of the IRIS database resulted in 136 reference values (for 120 of the 553 chemicals) that had been reduced by a UF-D>1 due to insufficient data. An absence of reproductive or developmental effects data was the sole or major contributing factor to 85% of these higher UF-Ds (115 of the 136).