Abstract

Multigeneration reproduction studies are used to characterize parental and offspring systemic toxicity, as well as reproductive toxicity of pesticides, industrial chemicals and pharmaceuticals. Results from 329 multigeneration studies on 316 chemicals have been digitized into standardized and structured toxicity data within the Toxicity Reference Database (ToxRefDB). An initial assessment of data quality and consistency was performed prior to profiling these environmental chemicals based on reproductive toxicity and associated toxicity endpoints. The pattern of toxicity across 75 effects for all 316 chemicals provided sets of chemicals with similar in vivo toxicity for future predictive modeling. Comparative analysis across the 329 studies identified chemicals with sensitive reproductive effects, based on comparisons to chronic and subchronic toxicity studies, as did the cross-generational comparisons within the multigeneration study. The general pattern of toxicity across all chemicals and the more focused comparative analyses identified 19 parental, offspring and reproductive effects with a high enough incidence to serve as targets for predictive modeling that will eventually serve as a chemical prioritization tool spanning reproductive toxicities. These toxicity endpoints included specific reproductive performance indices, male and female reproductive organ pathologies, offspring viability, growth and maturation, and parental systemic toxicities. Capturing this reproductive toxicity data in ToxRefDB supports ongoing retrospective analyses, test guideline revisions, and computational toxicology research.

The U.S. Environmental Protection Agency (EPA) and other regulatory agencies are investigating novel approaches for predicting chemical toxicity, with the goal of rapidly screening the thousands of environmental chemicals with limited toxicity data (U.S. EPA, 2009). Building predictive models of chemical toxicity requires high-quality in vivo toxicity data, in order to develop and validate new in vitro and in silico approaches. In support of EPA's ToxCast predictive toxicology effort (Dix et al., 2007), we have developed the Toxicity Reference Database (ToxRefDB) for capturing information from in vivo toxicity studies. ToxRefDB includes endpoints from multiple study types, including chronic rat and mouse carcinogenicity 2-year bioassays that have been previously reported and made publicly available (http://www.epa.gov/ncct/toxrefdb/) (Martin et al., 2009). ToxRefDB is being used to build computational models linking whole animal toxicity, and specific tissue and cellular phenotypes, to specific chemical-biological interactions detected by cellular, genomic and biochemical in vitro assays. The in vivo toxicity data captured in ToxRefDB is facilitating a transition to the National Research Council's vision for “Toxicity Testing in the 21st century: A Vision and a Strategy” (Collins et al., 2008; NRC, 2007), by linking toxicity endpoints from animal studies to molecular targets and pathways relevant to humans.

The multigeneration study data entered into ToxRefDB provides anchoring in vivo reproductive toxicity data for the EPA ToxCast research program (http://www.epa.gov/ncct/toxcast/). Within the ToxCast program, bioactivity profiles for hundreds of environmental chemicals are being derived from hundreds of in vitro assays (Dix et al., 2007; Houck and Kavlock, 2008). Phase I of ToxCast is focused on chemicals with known in vivo toxicity data, supporting the development of in vitro data signatures predictive of these in vivo outcomes (Kavlock et al., 2008). It is worth noting that for environmental chemicals, unlike pharmaceuticals, quantitative in vivo toxicity data is essentially restricted to animal species. Nearly all of the ToxCast Phase I chemicals are food-use pesticide active ingredients that have undergone numerous mammalian toxicity tests, including guideline multigeneration studies. This highly standardized dataset provided in ToxRefDB facilitates profiling ToxCast Phase I chemical toxicity based on parental, offspring and reproductive effects. It was hypothesized that, through the use of ToxRefDB, key effects from the 329 multigeneration reproduction studies would characterize the reproductive toxicity potential of the 316 chemicals, and further differentiate the chemicals and effects with regards to generational and life-stage sensitivities. Subsequently, the well-characterized reproductive effects could be used to phenotypically anchor predictive toxicity models.

Traditional toxicity testing for the risk assessment of environmental compounds or groups of compounds can cost millions of dollars and take years of effort. Since 1970, EPA has accumulated a vast store of high-quality regulatory toxicity information on hundreds of compounds, most of which has been inaccessible to computational analyses. The curation and structuring of this chemical toxicity information into ToxRefDB has created a valuable resource for both retrospective and prospective toxicological studies (Martin et al., 2009). In addition to the chronic/cancer rat and cancer mouse studies (Martin et al., 2009) and multigeneration studies reported here, we are also extracting developmental toxicity studies in the rat and rabbit. The multigeneration reproductive toxicity data set—studies used by EPA in the pesticide registration process to assess the performance and integrity of the male and female reproductive systems (U.S. EPA, 1996) include assessment of gonadal function, the estrous cycle, mating behavior, conception, gestation, parturition, lactation, weaning, and on the growth and development of the offspring. The multigeneration study also provides information about the effects of the test substance on neonatal morbidity, mortality, target organs in the offspring, and data on prenatal and postnatal developmental toxicity.

Two historical test guidelines have been used for the multigeneration studies in ToxRefDB. Multigeneration studies according to the 1982 Reproductive and Fertility Effects guideline (U.S. EPA, 1982) on over 700 chemicals have been conducted and submitted to EPA. Multigeneration studies according to the newer 1998 guideline (U.S. EPA, 1998) on over 90 chemicals have been conducted and submitted, including 40 studies extracted into ToxRefDB. Information on data submissions to EPA was drawn from the Office of Pesticide Programs (OPP) Information Network—the OPPIN database (http://www.epa.gov/pesticides/). The 1998 guideline was harmonized by EPA's Office of Pollution Prevention & Toxic Substances (OPPTS; http://www.epa.gov/opptsfrs/home/guidelin.htm) to meet testing requirements of the EPA's Office of Pollution Prevention and Toxics and OPP, as well as international guidelines published by the Organization for Economic Cooperation and Development. Both of these guidelines call for a two-generation study in which continuously treated male and female rats are mated to produce first generation offspring, and in turn the adult offspring are mated to produce a second generation. Continuing refinement of these test guidelines has been proposed (Cooper et al., 2006), and ToxRefDB is being used to test hypotheses concerning the validity of these refinements.

MATERIALS AND METHODS

Data characteristics.

Reviews of registrant-submitted multigeneration reproductive toxicity studies, known as Data Evaluation Records (DER), were collected for roughly 300 chemicals from EPA's OPP. File types of DER include TIFF, Microsoft Word, Word Perfect and PDF formats, some of which are not directly text-readable. Approximately 500 multigeneration reproductive toxicity DER were reviewed, and based on data quality a subset of 329 was selected for curation into ToxRefDB. The first portion of the DER outlines the test substance, purity, lot/batch numbers, MRID (Master Record Identification), study citation, OPPTS test guideline (U.S. EPA, 1982, 1998) and reviewers of the study. The executive summary captures all of the basic study design information, including species and strain, doses, number of animals per treatment group and any deficiencies in study protocol. All dose levels were stored in ToxRefDB as “mg/kg/day” and, where possible, recorded or calculated from food consumption data as an average over the entire dosing period. The executive summary also describes treatment-related effects observed at various dose levels in the study. The body of the DER provides detailed test material and animal information, and full dose response data in text and tables for all measured and observed endpoints. All treatment-related effects were captured for each study in ToxRefDB.

Multigeneration study DER contain all the information necessary to infer lowest effect level (LEL) values for all treatment-related effects that were statistically or biologically significant. Typically, the DER also designated “critical” effects for each study, and lowest observed adverse effect level (LOAEL) and no observed adverse effect level (NOAEL) for each study. If provided by the DER, ToxRefDB captured these study-level NOAEL, LOAEL, and critical effect data. However, it is important to note that the critical effects used to establish NOAEL, LOAEL, and a reference dose (RfD) for a conventional chemical pesticide active, and to make regulatory risk assessment and management decisions, are based on a toxicological review of multiple studies across many study types.

Treatment-related effects were further identified as either a “Parental,” “Offspring,” or “Reproductive” effect. Consistent with DER, “Parental” endpoints were defined as systemic toxicity observed in the male or female adult parents, and exclude effects directly related to reproduction (e.g., reproductive organ toxicity). “Offspring” endpoints were defined as systemic toxicity observed in the preweaning and juvenile animals, and exclude birthing indices up to postnatal day (PND) 4 (e.g., litter size and live birth index). “Reproductive” endpoints were defined as observed effects on the reproductive performance or capacity of the animals and included all reproductive organ toxicities, effects on estrous cyclicity, sperm parameters, fertility, and mating, and prenatal and early postnatal viability.

A small number of ToxCast Phase I chemicals were not pesticide active chemicals, such as some perfluorinated compounds and phthalates. Though DER and pesticide registration studies were not available for these chemicals, there was often high-quality, standardized reproductive toxicity studies available from the National Toxicology Program, peer-reviewed literature, or other sources. When data from such studies were available, it was crated into ToxRefDB consistent with information taken from DER.

Data model and quality control.

The relational data model for ToxRefDB was previously described (Martin et al., 2009) in a diagram showing the data model and field-level. A Data Entry Tool was developed for database population, including a controlled vocabulary for reproductive and other test data (available for download at http://www.epa.gov/ncct/toxrefdb/). Additional data entry and quality control procedures for ToxRefDB are described in Martin et al. (2009), and on the ToxRefDB homepage.

Full descriptions of the available data and conclusions as to the potential for the pesticides to cause harm to humans or the environment, risk mitigation measures, and the regulation of pesticides can be found at U.S. EPA's OPP websites: http://www.epa.gov/pesticides/regulating/index.htm; http://www.epa.gov/pesticides/reregistration/status.htm; http://www.epa.gov/oppsrrd1/registration_review/; http://www.epa.gov/oppsrrd1/reregistration/index.htm. The study-level critical effects captured in ToxRefDB and taken from individual DER and studies cannot be related directly to regulatory determinations or Reds without additional information and analysis.

Data output and analysis.

The structured toxicity information stored within ToxRefDB can be extracted in various formats utilizing SQL queries. For the purpose of providing computable outputs, that is, quantitative outputs amenable to statistical analysis, a consistent data output was used. The cross-tabulated data output consisted of rows of chemical information (e.g., Chemical Abstracts Service Registry Number, chemical name), by columns of toxicity endpoints with the value entered being the lowest dose at which the endpoint was observed (i.e., LEL) in “mg/kg/day.” Even though NOAEL/LOAEL values for each study's “Parental,” “Offspring,” or “Reproductive” effect can be queried from the database, the current analyses for ToxCast only utilized LEL. Log-transformed potency values were derived using -log2 of LEL. A constant value of 12 was then added to zero-center the data allowing for zero to represent no observed effect. Therefore, a value of 1 would be equivalent to an effect at 2048 mg/kg/day and 18 would be equivalent to 0.015625 mg/kg/day. The log-transformed values are predominantly used in the current analysis. However, mill molar concentrations (mol/kg/day) were calculated for each endpoint using the molecular weight of tested chemical and the LEL in mg/kg/day. These data tables are available on the ToxRefDB homepage: http://www.epa.gov/ncct/toxrefdb/. It should be noted, however, that potency does not necessarily equate to risk because this analysis does not take into consideration levels of exposure, a key element in the determination of risk. Moreover, the potency of a compound in this analysis does not represent that the endpoints identified in the multigeneration toxicity study are the most sensitive in the database or for that matter the most appropriate for the exposure scenarios being evaluated.

Hierarchical clustering across all the chemicals and effects was carried out based on log-transformed potency values. Effects were selected based on an occurrence in five or more chemicals, which was level shown to have minimal predictive capability based on a simulation study performed by Judson et al. (2008a). The clustering analysis, implemented in R version 2.6.1 (Ithaca and Gentleman, 1996), used Pearson's dissimilarity as the distance measure for both chemicals and effects and Ward's method for linkage (Ward, 1963). The chemicals were divided into six groups based on the percent of explained variance (Thorn dike, 1953). The weight for each effect in deriving the chemical groupings was calculated as the ratio of the number of positives within the chemical grouping over the number of positives out of the chemical grouping.

RESULTS

Summary Characterization of Multigeneration Study Results

This analysis focused on reproduction-related endpoints culled from 329 multigeneration rat studies on 316 unique chemicals entered into ToxRefDB. The vast majority of studies (294 of 329) were performed using a two-generation protocol. There were seven one-generation studies, for which four were supplementary studies to longer-term two- or three-generation studies. Of the 28 three-generation studies, only first and second generation effects were used in subsequent analyses, whereas third generation effects were excluded. In total, there were 11 chemicals with more than one study in this dataset. Four chemicals had an additional study run to satisfy study guideline requirements. Two chemicals had an additional study to test at additional dose levels. Five chemicals had two studies performed at similar dose levels and the conclusions between each pair of studies were similar.

Across all studies and treatment groups 12,230 treatment-related effects were observed, corresponding to 458 different, unique types of effects. Each effect was tagged with specific endpoint category, life-stage, and generational information. The distribution of treatment-related effects and positive chemicals across life-stage and generation provide insight into the sensitivities of specific classes of endpoints (Table 1). Parental effects were associated with 275 of the 316 chemicals for both the P1 and F1 generation, whereas reproductive effects were associated with only 100 or 129 chemicals in the P1 and F1 generations, respectively. Besides more chemicals, there were 73% more adult reproductive effects in the F1 generation, than in the P1. A similar number of chemicals and offspring effects were observed in the F1 and F2 generation. The relative generational sensitivity among reproductive effects compared to offspring effects prompted us to investigate the patterns of specific reproductive and offspring toxicities across all chemicals.

TABLE 1

Distribution of Chemicals and Effects Across Life-Stage, Endpoint Category and Generation for 316 Chemicals in ToxRefDB with a Multigeneration Reproductive Study

Life stage
 
Adult
 
Adult
 
Juvenile
 
Endpoint category Parentalc Reproductived Offspringe 
Generation P1 275a (2935)b 100 (376)  
 F1 275 (3265) 129 (648) 255 (2274) 
 F2   247 (1979) 
Life stage
 
Adult
 
Adult
 
Juvenile
 
Endpoint category Parentalc Reproductived Offspringe 
Generation P1 275a (2935)b 100 (376)  
 F1 275 (3265) 129 (648) 255 (2274) 
 F2   247 (1979) 
a

Number of chemicals with at least one effect observed at specified life-stage, endpoint category and generation.

b

Number of effects observed at specified life-stage, endpoint category, and generation.

c

Parental endpoints include adult body weight, mortality, clinical signs, and target-organ weight and pathology effects.

d

Reproductive endpoints include reproductive organ weight and pathology and reproductive indices (e.g., fertility, mating, live birth index).

e

Offspring endpoints include pup weight, offspring survival (e.g., viability and lactation index), and juvenile target-organ weight and pathology, and pubertal delay (e.g., PPS and VO) effects.

Patterns of Reproductive Toxicity

Identification of chemical groups with similar reproductive toxicity profiles was achieved by hierarchical clustering of 75 target-level effects (Fig. 1). These were defined as target-level effects because specific descriptive terms were aggregated to the target organ (i.e., liver) or measured index (e.g., lactation index), rather than all possible outcomes for each target (hypertrophy, hyperplasia, degeneration, etc.). Six groups of chemicals were identified based methods described above in the Methods section. Each chemical grouping was described by the effects that most heavily weighted the formation of the chemical groupings in Figure 1 and does not mean that every chemical in the group causes the endpoint. Group 1 consists of the 14 chemicals with no observed toxicities across the 75 effects in this analysis. Group 2 contains 115 chemicals for which general systemic toxicities are driving the formation of the group. Interestingly, this chemical grouping is also heavily weighted with endpoints relating to sperm counts and morphology, endocrine-related organ pathologies and weight changes, and delays in sexual maturation. Of the 115 chemicals, all five phthalate compounds are found in this group. Group 3 contains 63 chemicals with limited toxicity for which parental and offspring body weight changes are driving the formation of the chemical group. Group 4 formation is heavily weighted with cholinesterase inhibition effects and is comprised of 12 organophosphorus compounds. Groups 5 and 6 contain 48 and 64 chemicals, respectively, and the formation of these groupings were heavily weighted with reproductive toxicity endpoints, including testicular and epididymal pathologies in group 5 and decrements in offspring viability and survival in group 6.

FIG. 1.

Two-way hierarchical clustering of 75 treatment-related effects from multigeneration reproduction tests on 316 chemicals in ToxRefDB. Six chemical groups were identified based on their patterns of reproductive toxicity. Each chemical group description is derived from the mostly heavily weighted endpoints (see Results and http://www.epa.gov/ncct/toxrefdb/ for details) and does not mean that every chemical in the group causes the endpoint.

FIG. 1.

Two-way hierarchical clustering of 75 treatment-related effects from multigeneration reproduction tests on 316 chemicals in ToxRefDB. Six chemical groups were identified based on their patterns of reproductive toxicity. Each chemical group description is derived from the mostly heavily weighted endpoints (see Results and http://www.epa.gov/ncct/toxrefdb/ for details) and does not mean that every chemical in the group causes the endpoint.

The complete listing of chemical groupings and endpoint weights are available for download from the ToxRefDB homepage (http://www.epa.gov/ncct/toxrefdb/). This analysis clearly segmented the chemicals into distinct classes based on their profile of systemic and reproductive toxicities. This analysis also guides endpoint selection process by highlighting groups of related chemicals or endpoints based on the entire profile of toxicological activity rather than a single outcome. Many of the associations between endpoints or chemicals were expected, but others were not. For instance, reproductive performance, reproductive organ and offspring viability effects were segregated slightly from each other and to a greater extent from parental systemic effects and even delays in sexual maturation.

Comparative Analysis with Chronic and Subchronic Systemic Toxicity

Parental, reproductive, and offspring potencies (i.e., inverse log-transformed LEL) from the multigeneration studies were compared to potency values for systemic toxicity from 2-year chronic and 90-day subchronic studies in the rat (Fig. 2). For this comparison, data were available in ToxRefDB for 254 chemicals tested in both multigeneration and 2-year chronic studies, and 207 chemicals tested in both multigeneration and 90-day subchronic studies. The potency values compared rarely correspond to the same treatment-related effect across study type. For the majority of chemicals, potency values between the multigeneration, chronic and subchronic studies were comparable, with a general linear relationship falling within ten-fold of each other. However, for four chemicals (bisphenol A, deltamethrin, flucycloxuron, flufenpyr-ethyl) that caused parental or reproductive effects in the multigeneration study, there was no systemic toxicity observed in either the chronic or subchronic studies. For another five chemicals (cyprodinil, diethyltoluamide, difenoconazole, ethametsulfuron methyl, thiamethoxam) potencies for the most sensitive multigeneration endpoints were more than 10-fold greater than for the most sensitive effects in chronic studies. Of these five chemicals only thiamethoxam was more potent based solely on reproductive endpoints, that is, testicular atrophy. Decreasing the threshold from 10-fold to a 2-fold increase in potency resulted in 37, 7, and 20 chemicals more potent for parental, reproductive, or offspring endpoints, respectively. Of the seven chemicals identified as twofold more potent reproductive toxicants, no reproductive organ toxicity was observed in the rat chronic/cancer or subchronic studies for these chemicals—the multigeneration test detected reproductive toxicity that could have been missed in chronic or subchronic studies. Under the conditions of the 2-year chronic studies, the vast majority of chemicals observed effects at lower doses than in the multigeneration reproductive study. However, even in these cases, the multigeneration test often identified selective reproductive toxicants and endpoints not detected in the chronic study.

FIG. 2.

Parental, reproductive, and offspring LELs (inverse log transformed) from multigeneration rat studies were compared to systemic LEL from chronic/cancer and subchronic rat studies for 254 and 207 chemicals, respectively. Points within gold lines indicate less than twofold difference between multigeneration and chronic studies. Points within orange lines indicate less than 10-fold difference between multigeneration and chronic studies. “NE” stands for not established.

FIG. 2.

Parental, reproductive, and offspring LELs (inverse log transformed) from multigeneration rat studies were compared to systemic LEL from chronic/cancer and subchronic rat studies for 254 and 207 chemicals, respectively. Points within gold lines indicate less than twofold difference between multigeneration and chronic studies. Points within orange lines indicate less than 10-fold difference between multigeneration and chronic studies. “NE” stands for not established.

Comparative Analysis of Parental, Reproductive, and Offspring Endpoints

Chemicals with increased potency in the second generation were identified by comparing P1 and F1, or F1 and F2 LEL across parental, reproductive and offspring endpoint categories for 316 chemicals (Fig. 3). Specific second generation effects (i.e., F1 parental or reproductive, F2 offspring) not observed in the first generation (i.e., P1 parental or reproductive, F1 offspring), or sensitive effects occurring at a lower LEL in the second generation are provided for all 316 chemicals on the ToxRefDB homepage (http://www.epa.gov/ncct/toxrefdb/). For parental effects, 15 chemicals had specific effects in the F1 versus P1, and another 48 were more sensitive in the F1 versus P1 based upon at least a twofold difference in LEL.

FIG. 3.

Comparing LELs across generation and endpoint category. Points within dark orange lines indicate less than twofold difference between generations. Points within light orange lines indicate less than 10-fold difference between generations. “NE” stands for not established.

FIG. 3.

Comparing LELs across generation and endpoint category. Points within dark orange lines indicate less than twofold difference between generations. Points within light orange lines indicate less than 10-fold difference between generations. “NE” stands for not established.

For reproductive toxicity endpoints, 52 chemicals had specific effects in the F1 versus P1, and another 14 were more sensitive in the F1 versus P1 based upon at least a twofold difference in LEL. For offspring toxicity endpoints, 14 chemicals had specific effects in the F2 versus F1, and another 28 were more sensitive in the F2 versus F1 based upon at least a twofold difference in LEL. Across reproductive and offspring effects, a total of 94 chemicals displayed specificity or sensitivity in the second generation. However, the F1 reproductive or F2 offspring LEL was the most sensitive LEL across all endpoint categories for only 16 of these 94 chemicals (Table 2). Of the 16 second generation sensitive chemicals as determined by specific LEL, only three of these chemicals had reproductive or offspring LOAEL based on critical effects that required mating of the F1 adults or were observed in the F2 offspring. Of these three, only fenarimol effects on F2 litter size determined the chronic reference dose in the risk assessment. This analysis in ToxRefDB has identified a subset of reference chemicals for ToxCast predictive modeling that may be more specific or potent reproductive toxicants. However, it is important to note that these ToxRefDB values are LEL for all treatment-related effects, and are in only a small minority of cases critical effects being used for determination of NOAEL/LOAEL.

TABLE 2

Sixteen Chemicals with the Most Sensitive LELs from F1 Reproductive or F2 Offspring Toxicities

 LEL (mg/kg/day)
 
 
 Parental
 
Reproductive
 
Offspring
 
 
Chemical name P1 F1 P1 F1 F1 F2 F1 reproductive/F2 offspring sensitive effect 
2,4-DB 112 NE 112 NE 112 25 Kidney dilationb 
Azoxystrobin 165 165 NE NE 165 32.3 Pup and liver weight changesb 
Bromuconazole 141 141 141 NE 141 15.5 Liver weight changesb 
Carbaryl 92.4 92.4 NE 92.4 92.4 31.3 Offspring viabilitya,b 
Chlorethoxyfos NE 0.78 NE NE 0.6 0.3 Pup weight decreaseb 
Clethodim 263 263 51 NE NE Prostate and seminal vesicle weight 
Desmedipham 20 20 NE 110 20 Liver and kidney weight changesb 
Dicyclohexyl phthalate 402 89.9 NE 17.8 457 457 Prostate weight decreasea 
Epoxiconazole 31.9 22.1 22.1 22.1 22.1 0.85 Offspring viabilityb 
Fenarimol NE NE NE 1.2 NE NE Litter size decreasea,b 
Mepiquat chloride 499 575 499 575 575 48.6 Eye opening delayb 
Propetamphos 2.8 7.1 5.5 0.3 5.5 5.5 Litter size decreaseb 
Stannane, tributylchloro- NE 6.25 NE 0.25 1.25 0.25 Pup, testis, and epididymis weight changesa 
TCMTB NE NE NE NE NE 38.4 Pup weight decreasea,b 
Thiamethoxam 61.3 61.3 NE 1.84 158 158 Testicular atrophya 
Triclosan 50 150 NE NE 150 15 Pup weight decreaseb 
 LEL (mg/kg/day)
 
 
 Parental
 
Reproductive
 
Offspring
 
 
Chemical name P1 F1 P1 F1 F1 F2 F1 reproductive/F2 offspring sensitive effect 
2,4-DB 112 NE 112 NE 112 25 Kidney dilationb 
Azoxystrobin 165 165 NE NE 165 32.3 Pup and liver weight changesb 
Bromuconazole 141 141 141 NE 141 15.5 Liver weight changesb 
Carbaryl 92.4 92.4 NE 92.4 92.4 31.3 Offspring viabilitya,b 
Chlorethoxyfos NE 0.78 NE NE 0.6 0.3 Pup weight decreaseb 
Clethodim 263 263 51 NE NE Prostate and seminal vesicle weight 
Desmedipham 20 20 NE 110 20 Liver and kidney weight changesb 
Dicyclohexyl phthalate 402 89.9 NE 17.8 457 457 Prostate weight decreasea 
Epoxiconazole 31.9 22.1 22.1 22.1 22.1 0.85 Offspring viabilityb 
Fenarimol NE NE NE 1.2 NE NE Litter size decreasea,b 
Mepiquat chloride 499 575 499 575 575 48.6 Eye opening delayb 
Propetamphos 2.8 7.1 5.5 0.3 5.5 5.5 Litter size decreaseb 
Stannane, tributylchloro- NE 6.25 NE 0.25 1.25 0.25 Pup, testis, and epididymis weight changesa 
TCMTB NE NE NE NE NE 38.4 Pup weight decreasea,b 
Thiamethoxam 61.3 61.3 NE 1.84 158 158 Testicular atrophya 
Triclosan 50 150 NE NE 150 15 Pup weight decreaseb 

Note. Underline = parental, reproductive, or offspring LOAEL (study-level LOAEL). NE = not established (no observed effects).

a

Study-level critical effect (F1 reproductive or F2 offspring).

b

F1 mating required.

Selected Multigeneration Study Endpoints for Predictive Modeling

Figure 4 presents the incidence and distribution by generation of effects on reproductive performance, reproductive organs, offspring viability, and parental systemic toxicities selected as anchoring endpoints for ToxCast predictive modeling. Toxicity profiles from multigeneration studies on 316 chemicals were based on a diverse set of 19 selected effects or effect aggregations distributed in various combinations across the P1, F1, and F2 generations. A detailed table listing all 19 of these endpoints for the 316 chemicals, including endpoint descriptions and various transformations of LEL values, is available for download from the ToxRefDB homepage (http://www.epa.gov/ncct/toxrefdb/).

FIG. 4.

Incidence and distribution, by generation, of the 19 endpoints selected for predictive modeling, including reproductive, offspring, and systemic toxicity endpoints from the rat multigeneration reproduction study (see Results and http://www.epa.gov/ncct/toxrefdb/ for details). The light gray bar indicates chemicals observing the endpoint only in the first generation, either P1 adult or F1 juvenile. The medium gray bar indicates chemicals observing the endpoint in both first and second generation treatment groups. The dark gray bar indicates chemicals observing the endpoint only in the second generation, either F1 adult or F2 juvenile.

FIG. 4.

Incidence and distribution, by generation, of the 19 endpoints selected for predictive modeling, including reproductive, offspring, and systemic toxicity endpoints from the rat multigeneration reproduction study (see Results and http://www.epa.gov/ncct/toxrefdb/ for details). The light gray bar indicates chemicals observing the endpoint only in the first generation, either P1 adult or F1 juvenile. The medium gray bar indicates chemicals observing the endpoint in both first and second generation treatment groups. The dark gray bar indicates chemicals observing the endpoint only in the second generation, either F1 adult or F2 juvenile.

These 19 highly prevalent effects identified treatment-related changes to reproductive performance including fertility, mating, gestational interval, implantations, litter size, and live birth index demonstrated effects at different stages of the reproductive cycle. Besides effects of many chemicals on offspring viability at PND4 and PND21 (viability and lactation indices, respectively), pubertal delays were also recorded for some chemicals. Pubertal delays were not part of the ToxCast modeling dataset because only a small subset of chemicals and studies assessed these endpoints. Effects on reproductive performance and offspring viability were observed in 110 (35%) and 108 (34%) of the 316 tested chemicals, respectively. Effects on reproductive organs, both organ weight and pathology, were observed in 98 (31%) of the chemicals with roughly 50% of those chemicals causing the effect only in the second generation (F1 adult). Of the 98 chemicals, 31 caused both male and female reproductive organ effects, 43 male only, and 24 female only. Systemic target organ weight and pathology endpoints were also selected, including the liver, kidney and spleen, along with the endocrine-related adrenal, pituitary, and thyroid glands.

The fairly restricted set of 19 effects characterized 151 of the 152 chemicals that demonstrated any reproductive toxicity. Additionally, these 19 effects identified 229 of the 269 chemicals that caused any offspring toxicity. The remaining 40 chemicals not identified were predominantly affecting pup weight only. This supports the hypothesis that we can extract a small finite set of key reproductive effects from this dataset for use in developing robust predictive signatures in the future stages of ToxCast research as a prioritization tool spanning reproductive toxicity.

DISCUSSION

ToxRefDB is being developed with several applications in mind. One is to provide in vivo toxicity effects as targets for ToxCast predictive models. In this fashion, ToxCast can be established as a cost-effective rapid approach for screening and prioritizing a large number of chemicals for further toxicological testing (Dix et al., 2007). Using data from high-throughput screening (HTS) bioassays developed in the pharmaceutical industry, ToxCast is building computational models to forecast the potential toxicity of chemicals. These hazard predictions should provide EPA regulatory programs with science based information helpful in prioritizing chemicals for more detailed toxicological evaluations, and therefore lead to using fewer animal tests. Target chemicals for such prioritization include pesticidal inerts, antimicrobials, and the many industrial chemicals with limited toxicity information (Judson et al., 2008c). ToxCast is currently in the proof-of-concept phase, wherein over 300 chemicals have been assayed in over 500 different HTS bioassays, creating bioactivity profiles being used to derive signatures predicting the known toxicity for these chemicals (Judson et al., 2009).

The Phase I chemicals are primarily conventional pesticide actives that have been extensively evaluated using traditional mammalian toxicity testing, and hence have known properties representative of a number of toxicity outcomes (e.g., reproductive toxicity). Thus a critical component of ToxCast is ToxRefDB, which is being populated with data from OPP for pesticide active chemicals and being extracted from the evaluations on these studies conducted by OPP scientists. Comparable toxicity data from other toxicity sources (e.g., National Toxicology Program) are also being captured in ToxRefDB. A broader and more diverse set of complementary data on thousands of chemicals is being captured in EPA's Aggregated Computational Toxicology Resource (http://actor.epa.gov/actor; Judson et al., 2008b). Although pesticide toxicity data currently predominates in ToxRefDB, the database is being expanded to a broader range of chemicals, both by category and use.

The underlying data represented in ToxRefDB has been evaluated by EPA in prior pesticide registration decisions, and the presence of effects in high-dose animal studies do not translate directly into significant human risk stemming from registered uses of the pesticide. One major issue to note is that the current analysis of ToxRefDB is not limited to just the critical effects leading to regulatory determinations of LOAEL and NOAEL. In addition, it should be noted that the EPA uses animal toxicology studies, like those entered into ToxRefDB, as well as other sources of information such as effects on wildlife populations, mechanisms of action, use patterns, environmental fate and persistence, food residue levels, and human exposure potential in its determinations to register pesticides, and to establish acceptable levels of pesticide residues for uses in the United States (http://www.epa.gov/pesticides/).

The toxicity data in ToxRefDB (www.epa.gov/ncct/toxrefdb) and the HTS data generated in ToxCast (www.epa.gov/ncct/toxcast) is being made publicly available through EPA websites. The first component of ToxRefDB was recently published (Martin et al., 2009), presenting endpoints to be used for predictive modeling from two-year rodent bioassays on 310 chemicals. The analysis and release of developmental toxicity endpoints on 383 chemicals from ToxRefDB will also provide key endpoints to be used for predictive modeling (Knudsen, 2009). Multigeneration reproduction study data for 316 chemicals was entered into ToxRefDB making the vast library of legacy data computable for the first time. The pattern of reproductive toxicity across these chemicals resulted in groupings of similar chemicals that could be used to match up with HTS bioactivity profiles. In the meantime, the analysis corroborated the distinction between parental, offspring and reproductive effects in downstream analyses based on the distribution of endpoints across the chemical groups.

All 12,230 effects in the multigeneration study dataset were placed into three major classes of effects; parental, reproductive and offspring. The LEL for each class or category of effects were used to identify sensitive or specific reproductive toxicants based on comparisons to chronic and subchronic study data and cross-generational comparisons within the multigeneration reproductive test. In general, chemical exposures under conditions of the multigeneration reproduction study were less sensitive than under the conditions of the 2-year chronic study and comparable to the 90-day subchronic study. The analysis did, however, identify a subset of 94 chemicals with sensitive or specific reproductive or offspring toxicities when compared to systemic effects under longer continuous exposure periods. Additional future analyses comparing, for instance, maternal and fetal toxicity from developmental toxicity studies (Knudsen, 2009) to parental and offspring toxicity from reproductive toxicity studies, will provide additional insight into the role of developmental exposures in the manifestation of specific toxicities.

Similar insight can be gleaned from comparing endpoints occurring at a lower dose or only in the second generation, that is, second generation sensitive or specific effects, respectively. Effects that occur in the first generation and are not corroborated in the second generation can be questioned as to its toxicological relevance. Conversely, effects with consistent increases in second generation sensitivity or specificity might reflect the need for reproductive or developmental exposure to occur. Comparisons across these broad classes of endpoints honed in on specific effects for which to characterize the chemical set. The primary set of effects selected as anchoring endpoints for ToxCast predictive modeling were reproductive indices, offspring viability, and male and female reproductive organ effects, along with a set of parental systemic organ toxicities.

The current study focused on providing endpoints for predictive modeling as part of the ToxCast research program (Dix et al., 2007), but also began to address the importance of specific study design parameters, including differences across generation, life-stage and various classes of endpoints. It has recently been suggested that the reproductive test guidelines for agrichemicals could be refined to make the second generation optional based on results seen in the first generation (Cooper et al., 2006). Consistent with results from Janer et al. (2007), the current analysis of this ToxRefDB dataset supports the hypothesis that the second, F2 generation in these 329 studies would rarely impact either the qualitative or quantitative evaluations of these studies. Of the sixteen second generation sensitive chemicals, carbaryl, fenarimol, and TCMTB observed second generation effects that would have required F1 mating. However, of these three chemicals only fenarimol effects on F2 litter size determined the chronic reference dose determination (U.S. EPA, 2006, 2007a,b). Additional analysis will be performed on this dataset in collaboration with OPP and other international chemical regulatory agencies to expound upon the role of these and other study design parameters with respect to chemical regulation and potential guideline study design changes. For instance, 53 of the 73 chemicals proposed for screening in the Endocrine Disruptor Screening Program (EDSP; http://www.epa.gov/endo/pubs/prioritysetting/draftlist.htm) have multigeneration studies entered into ToxRefDB. Where available, multigeneration study data for the remaining chemicals are now being entered into ToxRefDB. A focused analysis of the EDSP chemical set to assess the ability of the current and previous guidelines to identify reproductive effects related to endocrine disruption would be just one example of the utility of ToxRefDB (Kavlock et al., 2009). The use of ToxRefDB to address many research and regulatory science questions regarding in vivo mammalian toxicity not only provides transparency, but also assists in guiding the next set of questions.

The diverse utility of ToxRefDB as a reference database for research applications such as ToxCast demonstrates the power of curating toxicity information into a relational database. In the current analysis on the multigeneration reproductive toxicity test, six chemical sets were derived and subsequently nineteen specific endpoints were identified to serve anchoring endpoints for eventual predictive modeling. These endpoints are further defined by life-stage or generation, and fully characterize the reproductive toxicity potential of the 316 in this study. Capturing this reproductive toxicity data in ToxRefDB supports ongoing retrospective analyses, test guideline revisions, and computational toxicology research.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci.oxfordjournals.org/.

FUNDING

United States Environmental Protection Agency.

References

Collins
FS
Gray
GM
Bucher
JR
Transforming environmental health protection
Science
 , 
2008
, vol. 
319
 (pg. 
906
-
907
)
Cooper
RL
Lamb
JC
Barlow
SM
Bentley
K
Brady
AM
Doerrer
NG
Eisenbrandt
DL
Fenner-Crisp
PA
Hines
RN
Irvine
LF
, et al.  . 
A tiered approach to life stages testing for agricultural chemical safety assessment
Crit. Rev. Toxicol.
 , 
2006
, vol. 
36
 (pg. 
69
-
98
)
Dix
DJ
Houck
KA
Martin
MT
Richard
AM
Setzer
RW
Kavlock
RJ
The ToxCast program for prioritizing toxicity testing of environmental chemicals
Toxicol. Sci.
 , 
2007
, vol. 
95
 (pg. 
5
-
12
)
Houck
KA
Kavlock
RJ
Understanding mechanisms of toxicity: Insights from drug discovery research
Toxicol. Appl. Pharmacol.
 , 
2008
, vol. 
227
 (pg. 
163
-
178
)
Ihaka
R
Gentleman
R
R: A language for data analysis and graphics
J. Comput. Graph. Stat.
 , 
1996
, vol. 
5
 (pg. 
299
-
314
)
Janer
G
Hakkert
BC
Slob
W
Vermeire
T
Piersma
AH
A retrospective analysis of the two-generation study: What is the added value of the second generation?
Reprod. Toxicol.
 , 
2007
, vol. 
24
 (pg. 
97
-
102
)
Judson
RS
Dix
DJ
Houck
KA
Martin
MT
Kavlock
RJ
Developing predictive signatures using in vitro data from the EPA ToxCast program
Toxicologist
 , 
2009
, vol. 
108
 (pg. 
79
-
80
)
Judson
RS
Elloumi
F
Setzer
RW
Li
Z
Shah
I
A comparison of machine learning algorithms for chemical toxicity classification using a simulated multi-scale data model
BMC Bioinformatics
 , 
2008
, vol. 
9
 pg. 
241
  
doi:10.1186/1471-2105-9-241
Judson
RS
Richard
AM
Dix
DJ
Houck
K
Elloumi
F
Martin
MT
Cathey
T
Transue
TR
Spencer
R
Wolf
M
ACToR—Aggregated computational toxicology resource
Toxicol. Appl. Pharmacol.
 , 
2008
 
doi:10.1016/j.taap.2007.12.037
Judson
R
Richard
A
Dix
DJ
Houck
K
Martin
M
Kavlock
R
Dellarco
V
Henry
T
Holderman
T
Sayre
P
, et al.  . 
The toxicity data landscape for environmental chemicals
Environ. Health Perspect.
 , 
2008
 
doi: 10.1289/ehp.0800168
Kavlock
RJ
Dix
DJ
Houck
KA
Judson
RS
Martin
MT
Richard
AM
ToxCast: Developing predictive signatures for chemical toxicity
Altern. Anim. Test. Exp.
 , 
2008
, vol. 
14
 
Special issue
(pg. 
623
-
627
)
Kavlock
RJ
Dix
DJ
Houck
KA
Martin
MT
Judson
R
Biological profiling of endocrine related effects of chemicals using ToxCast
Toxicologist
 , 
2009
, vol. 
108
 pg. 
226
 
Knudsen
TB
Martin
MT
Kavlock
RJ
Judson
RS
Dix
DJ
Singh
AV
Profiling the activity of environmental chemicals in prenatal developmental toxicity studies using the U.S. EPA's ToxRefDB
Reprod. Toxicol.
 , 
2009
 
doi: 10.1016/j.reprotox.2009.03.016
Martin
MT
Judson
RS
Reif
DM
Kavlock
RJ
Dix
DJ
Profiling chemicals based on chronic toxicity results from the U.S. EPA ToxRef database
Environ. Health Perspect.
 , 
2009
, vol. 
117
 (pg. 
392
-
399
)
National Research Council (NRC)
Toxicity Testing in the 21st Century: A Vision and a Strategy
 , 
2007
Washington, DC
The National Academies Press
Thorndike
RL
Who belongs in the family?
Psychometrika
 , 
1953
, vol. 
18
 (pg. 
267
-
276
)
U.S. Environmental Protection Agency (U.S. EPA)
OPP Guideline 83–84: Reproductive and Fertility Effects. Pesticide Assessment Guidelines, Subdivision F, Hazard Evaluation: Human and Domestic Animals
 , 
1982
 
Office of Pesticides and Toxic Substances, Washington, DC, EPA-540/9-82-025
U.S. Environmental Protection Agency (U.S. EPA)
Reproductive toxicity risk assessment guidelines
Fed. Regist.
 , 
1996
, vol. 
61
 (pg. 
56274
-
56322
)
U.S. Environmental Protection Agency (U.S. EPA)
Health Effects Test Guidelines OPPTS 870.3800 Reproduction and Fertility Effects
 , 
1998
 
Office of Pesticides and Toxic Substances, Washington, DC, EPA 712-C-98-208
U.S. Environmental Protection Agency (U.S. EPA)
Reregistration Eligibility Decision for 2-(Thiocyanomethylthio)-benzothiazole (TCMTB)
 , 
2006
Washington, DC
Office of Pesticides and Toxic Substances
 
Office of pesticides and toxic substances, Washington, DC, EPA 739-R-07-003
U.S. Environmental Protection Agency (U.S. EPA)
Fenarimol Summary Document Registration Review: Initial Docket
 , 
2007
 
Office of Pesticides and Toxic Substances, Washington, DC, EPA-HQ-OPP-2006-0241
U.S. Environmental Protection Agency (U.S. EPA)
Reregistration Eligibility Decision (RED) for Carbaryl
 , 
2007
 
Office of Pesticides and Toxic Substances, Washington, DC, EPA 738-R-07-018
U.S. Environmental Protection Agency (U.S. EPA)
The U.S. Environmental Protection Agency's Strategic Plan for Evaluating the Toxicity of Chemicals
 , 
2009
 
Office of the Science Advisor, Science Policy Council, Washington, DC, EPA 100/K-09/001
Ward
JH
Hierarchical grouping to optimize an objective function
J. Am. Stat. Assoc.
 , 
1963
, vol. 
58
 (pg. 
236
-
244
)

Author notes

Disclaimer: The United States Environmental Protection Agency through its Office of Research and Development funded and managed the research described here. It has been subjected to Agency administrative review and approved for submission and peer review.