The Toxicologist - Society of Toxicology

high-content imaging. Thirty-two compounds were confirmed in both fluorescence
plate reader and imaging assay formats. To study the structure-activity relationship
of these mitochondrial disruptors, we clustered these compounds by structural
similarity. This analysis resulted in four structural clusters and 15 singletons.
These clusters may be useful for identifying structural features associated with mitochondrial
toxicity. Our results confirm the robustness of this assay for identifying
mitochondrial membrane-potential disruptors in qHTS format. Supported by
NIEHS Interagency Agreement Y2-ES-7020-01.
494 QUANTITATIVE STRUCTURE-ACTIVITY
RELATIONSHIP (QSAR) MODELING OF ESTROGEN
RECEPTOR (ER) BINDING AFFINITY AND VIRTUAL
SCREENING FOR POTENTIAL ENDOCRINE
DISRUPTING COMPOUNDS (EDCS).
L. Zhang 1 , H. Zhu 1 , A. Afantitis 2 , G. Melagraki 2 , H. Sarimveis 2 , I. Rusyn 3 and
A. Tropsha 1 . 1 Eshelman School of Pharmacy, UNC, Chapel Hill, NC,
2 NovaMechanics Ltd., Nicosia, Cyprus and 3 Environmental Sciences and Engineering,
UNC, Chapel Hill, NC.
Regulatory agencies, both in Europe and in the US, require substantial, and costly,
animal testing of EDCs. The use of computational predictors to prioritize potential
candidates for in vivo endocrine disruption tests could significantly reduce the cost
of experiments and the number of compounds subject to testing. In this study, we
have employed combinatorial QSAR modeling workflow, incorporating rigorous
model validation procedures (Tropsha, Mol. Inf., 2010, 29, 476-488), to the largest
publicly available dataset of ER binding compounds. The dataset containing 436
ERα and 113 ERβ ligands was compiled from public sources and carefully curated.
Each compound was characterized by chemical descriptors calculated with Dragon,
CDK, and Mold2 Software. To evaluate the true predictivity of models, 5-fold external
cross validation procedure was applied. For each modeling set including 80%
of all compounds, k-nearest neighbors (kNN) QSAR approach and each of the
three descriptor sets were used for model development. The external predictive accuracy
of models ranged between R2 values of 0.63-0.80 and 0.66-0.85 for ERα
and ERβ models, respectively. Y-randomization test yielded no statistically significant
models, indicating the robustness of our models. The consensus models generated
by combining predictions from all independent models passing both training
and test set accuracy thresholds had the highest external accuracy, characterized by
R2 values of 0.72 and 0.76 for ERα and ERβ models, respectively. We have applied
our models for virtual screening of several chemical libraries of interest including
the 30K compounds registered under REACH as well as EPA-10K compound set.
Approximately 3,000 compounds were identified and prioritized as potential EDCs
for future in vivo endocrine disruption tests.
495 A COLLABORATIVE PROJECT FOR DATA SHARING
AND HARMONIZATION OF DEVELOPMENTAL
TOXICITY DATABASES.
E. Busta 1 , S. Espada 2 , M. Martin 3 , D. Hristozov 1 , K. Arvidson 1 , D. Mattison 4
and C. Yang 1 . 1 U.S. FDA, College Park, MD, 2 Pennsylvania State University,
College Station, PA, 3 U.S. EPA, Research Triangle Park, NC and 4 NICHD,
Bethesda, MD.
Developmental toxicity is one of the most important non-cancer endpoints for
both environmental and human health. Despite the fact that numerous developmental
studies are conducted as required for regulatory decisions, there are not yet
sufficient data available to develop predictive computational methods. The Food
and Drug Administration’s (FDA) Center for Food Safety and Applied Nutrition
(CFSAN) has been involved in developing ToxML (a toxicity database model) and
toxicity databases. The National Center for Computational Toxicology (NCCT)
team at Environmental Protection Agency (EPA) implemented ToxRefDB with its
data model inspired by ToxML. The new initiative at FDA CFSAN to build an institutional
knowledge-base, the CERES (Chemical Evaluation and Risk Estimation
System) project, has imposed the need to generate another database model for the
Center’s regulatory data. The prospect of creating yet another toxicity data model
was a real concern; hence, as part of the CERES project, an experiment was conducted
at FDA CFSAN to evaluate EPA’s ToxRefDB data entry tool for capturing
developmental effects from FDA’s approved drugs. In collaboration with National
Institute of Child Health and Human Development (NICHD), we have added developmental
data on 80 more drugs and merged with the existing database of 720
chemicals from EPA and FDA. By incorporating pharmaceutical data into the current
agrochemical domain, both the chemical space and biological profiles were ex-
106 SOT 2011 ANNUAL MEETING
panded. This experiment also assisted our database modeling while expanding and
harmonizing the controlled vocabulary. It is our further goal to combine the data
with existing developmental toxicity databases to support computational method
development. This poster will present comparisons of developmental toxicity data
models, profiles of distinct and common effects, and the chemical domains in both
ToxRefDB and CERES. This abstract does not necessarily reflect policies of US
EPA, US FDA, and NICHD.
496 DEHP: COMPARISON OF IN SILICO PREDICTIONS
WITH IN VITRO AND IN VIVO TOXICITY.
M. M. Dingemans 1 , E. Rorije 2 , A. Efremenko 3 , H. Clewell 3 and B. Blaauboer 1 .
1 Institute for Risk Assessment Sciences (IRAS), Utrecht University, Utrecht,
Netherlands, 2 National Institute for Public Health and the Environment (RIVM),
Bilthoven, Netherlands and 3 The Hamner Institutes, Research Triangle Park, NC.
This research was conducted within a larger program to investigate possibilities and
limitations of the use of non-animal data for predicting human risk. In the part described
here, the in silico predicted effects for bis(2-ethylhexyl) phthalate (DEHP;
using knowledge-based systems DEREK and OECD Toolbox, and QSAR-based
TOPKAT®) are compared with those identified in vitro and in vivo, to investigate
whether in vitro studies can confirm predicted effects.
Toxic effects by DEHP are predicted to be teratogenicity and testicular toxicity
(DEREK), carcinogenicity and developmental toxicity (TOPKAT, DEHP in training
set) and DNA binding (metabolism required; OECD Toolbox). In particular
endocrine disruption, reproductive toxicity (including testicular toxicity) and hepatotoxicity
by peroxisome proliferation, and non-genotoxic carcinogenicity are identified
in vitro. The predicted DNA binding is not confirmed in vitro, possibly due
to a lack of metabolism. In vivo toxicity data for rodents after oral exposure were
obtained from North American and European risk assessment agencies, revealing
increased liver weight and testicular toxicity as the most critical effects (