The Toxicologist - Society of Toxicology

data, both in terms of the chemical space (industrial/agro chemicals vs. drugs) and
in terms of the hCYP activity profiles, which for ToxCast include a large proportion
(80%) of chemicals inactive in all 3 hCYP inhibition assays. The ToxCast inhibition
assay results and the isoCYP metabolic activity data together afford the opportunity
to explore SAR associations for a significantly expanded chemical space,
using both features and chemical property descriptors to discern requirements for
qualitative hCYP activity, as well as hCYP isoform specificity. ToxCast Phase II is
screening an additional 700 chemicals, including expanded sets of drugs and
agro/industrial chemicals, in the same 3 hCYP inhibition assays, providing an ideal
opportunity to prospectively evaluate the performance of the new models, as well as
to iteratively improve upon these models.
This abstract does not necessarily reflect U.S. EPA policy.
142 EFFECTS OF GENETIC VARIATION IN ENZYME
CYTOCHROME P450 2D6 ON XENOBIOTIC
METABOLISM THROUGH IN SILICO MOLECULAR
DOCKING MODELS.
C. B. Tolson 1, 2 , Y. Tie 1 and E. Demchuk 1 . 1 Division of Toxicology &
Environmental Medicine, ATSDR/CDC, Atlanta, GA and 2 Biomedical Engineering,
Georgia Institute of Technology, Atlanta, GA. Sponsor: B. Fowler.
Most drug metabolism performed in the liver is a result of enzymatic activity of the
cytochrome P450 family of enzymes, especially cytochrome P450 3A4 and 2D6.
Genetic variation is a large source of enzyme activity variability. This project determined
whether methods of computational molecular docking can be used to detect
the effects of genetic variation on substrate binding. The project is part of a large
joint program of ATSDR/CDC and NCTR/FDA aimed at modeling liver toxicity.
The project concentrates on enzymatic activity in CYP2D6 variants with the 120
residue substitution (CYP2D6*53 and CYP2D6*49). Computational molecular
docking modeling using FRED, Induced-Fit, and GLIDE was applied. The in silico
docking models analyzed the interactions between individual ligands and the active
sites of the CYP2D6 enzyme variants. The results suggested that computational
molecular docking models could be used as accurate predictors in pre-clinical drug
testing, but may have limited sensitivity to genetic variations beyond the active site.
The FRED docking model best suited for the CYP2D6 enzyme-substrate system.
Since data on human cytochrome P450 metabolism and associated genetic susceptibility
have been obtained almost exclusively in clinical drug testing, better understanding
of the molecular interactions between xenobiotics and the cytochrome
P450 family will aid in the development of better predictive computational models
to be used during pre-clinical drug development. Once a robust human model is
developed, it can be applied in other areas, including chemical and biological defense
and environmental health problems, such as estimation of metabolic parameters
in physiologically-based pharmacokinetic modeling (PBPK) or more specific
recommendations for toxic substance dosages based on genetic variants within a
particular population.
143 INCORPORATION OF TOXCAST IN VITRO ASSAY
DATA AND RELEVANT TOXICITY PATHWAY
INFORMATION IMPROVES THE EXTERNAL
PREDICTION ACCURACY OF QUANTITATIVE
STRUCTURE-ACTIVITY RELATIONSHIP (QSAR)
MODELS OF CHEMICAL HEPATOTOXICITY.
H. Zhu, L. Zhang, J. Staab, A. Sedykh, H. Tang, S. Gomez, I. Rusyn and A.
Tropsha. University of North Carolina at Chapel Hill, Chapel Hill, NC.
The US EPA ToxCast program provides data for 610 in vitro assays, in which 320
substances with known in vivo toxicity measured in 76 assays were tested. We have
explored this complex data with novel computational approaches aimed to link
chemical structures, in vitro responses, pathway information, and in vivo toxicity
effects. We have used data from 22 ToxCast assays linked to Peroxisome
Proliferator-Activated Receptor (PPAR) pathway to build biologically inferred
QSAR models of six chronic animal hepatotoxicity endpoints. The datasets included
228 and 235 compounds for mouse and rat hepatotoxicity subsets, respectively,
and both conventional QSAR models and those using the following novel hierarchical
QSAR modeling workflow were built. First, all chemicals were
partitioned into two classes based on whether a compound was active (Class I) or
inactive (Class II) in any of the PPAR-relevant assays; we found that there were 100
Class I compounds and 128 Class II compounds for the mouse data and 106 Class
I compounds and 129 Class II compounds for the rat data. Second, classification
hepatotoxicity models for each endpoint were developed using both random forest
and MOE binary QSAR approaches for Class I and II compounds, independently.
30 SOT 2011 ANNUAL MEETING
The prediction for external compounds is realized by initially placing them into
Class I or II based on in vitro responses and then using class-specific QSAR models
for in vivo toxicity prediction. The external prediction accuracy of models built
with this two-step workflow was in the range of 57-76% for all six hepatotoxicity
endpoints, while that achieved by conventional QSAR models was only 50-65%
for the same external set. We conclude that methods discussed herein offer novel
ways of exploiting biological screening data to improve the accuracy of predictive
toxicology models.
144 COMPUTATIONAL DOCKING OF THE ISOMERS OF
NONYLPHENOL TO THE LIGAND BINDING DOMAIN
OF THE ESTROGEN RECEPTOR.
J. Rabinowitz and S. Little. NCCT/ORD/Environmental Protection Agency, Research
Triangle Park, NC.
Nonylphenols are environmentally persistent endocrine disrupting chemicals. They
exist in the environment as complex mixtures containing many nonylphenol isomers.
Environmental mixtures of nonylphenols, along with a few single isomers
have been tested for their capacity to interact with the estrogen receptor (ER) and
have been shown to be weakly estrogenic. The few individual isomers tested have
only in rare examples been shown to be more estrogenic than environmental mixtures
of undetermined isomers. The capacity of a molecule to bind to ER depends
on the three dimensional structure of the potential ligand. There are 211 geometric
isomers and 551 stereo isomers of nonylphenol. The potential for each isomer to
bind to ER will depend on the energetics the interaction. Computational molecular
docking has been used to examine the capacity of each nonylphenol isomer to
bind to ER. Computational targets have been created by removing the ligand
atoms from crystal structures of ER-ligand complexes. The isomers of nonylphenol
have been docked sucessfully into these targets. The results indicate that the potential
for binding will be greatest when the molecular structure maximizes both the
strength of the hydrogen bonds between the receptor and the phenolic hydroxyl
group, and the steric interaction between the alkyl group and the hydrophobic core
of the receptor. For some isomers the formation of the necessary hydrogen bonds
forces the ligand-receptor complex into a structure with less favorable steric interactions.
For example, branching at the alpha position decreases the interaction while
branching in the beta and gamma positions often increases the interaction. Analysis
of these considerations can provide insight for the design and manufacture of
nonylphenol (and other alkylphenol) mixtures that minimize their potential estrogenicity.
[This abstract does not necessarily reflect U.S. EPA policy.]
145 ANALYSIS OF RELATIVE ROSIGLITAZONE,
PIOGLITAZONE, AND EPIRUBICIN CARDIOTOXICITY
IN RATS USING HISTORICAL MICROARRAY DATA.
H. Zhou and T. J. Colatsky. CDER/OPS/OTR/DAPR, U.S. FDA, Silver Spring, MD.
Concerns have been raised about increased cardiovascular risk in patients treated
with rosiglitazone (ROSI), but ROSI’s cardiovascular safety profile remains an open
question because of conflicting data on the existence and magnitude of the risk. We
evaluated rat heart microarray data extracted from the Iconix DrugMatrix (IDM)
database for ROSI (5d 1800 mg/kg/day; IDM high dose) and pioglitazone (PIO;
5d 1500 mg/kg/day; IDM high dose), and compared the results to similar data obtained
for epirubicin (EPI; 5d 2.7mg/kg/day; IDM high dose), a drug with well
characterized cardiotoxicity. ROSI treatment induced 1002 differentially expressed
genes (DEGs) in rat heart compared to 360 DEGs for PIO and 986 DEGs for EPI.
Further analysis using Ingenuity Pathways Analysis revealed that ROSI and EPI induced
more than twice the number of cardiotoxicity-related DEGs than PIO treatment
but fewer than EPI (EPI:ROSI:PIO = 100:88:40). All 3 drugs were associated
with NRF-2-mediated Oxidative Stress Response (EPI>ROSI>PIO), while ROSI
and EPI both impacted Oxidative Stress and VDR/RXR activation. In addition,
ROSI treatments altered 20 cardiac hypertrophy related molecules while PIO altered
only 6. These results could be correlated with the increase in heart weight produced
by ROSI and PIO in these studies. Among the cardiotoxicity related molecules
demonstrating altered expression, ROSI and PIO shared 17/111 common
DEGs, ROSI and EPI shared 22/166 common DEGs, and PIO and EPI shared
13/127 common DEGs while all the three shared only 6 common DEGs: CCNA2,
CD36, FOS, HOPX, NCALD and PRKCH. In conclusion, these data suggest
ROSI may be more cardiotoxic than PIO in rats, but the large number of unique
cardiotoxicity related DEGs suggested multiple pathways exist for the cardiovascular
effects of ROSI, PIO and EPI treatment. However, the relevance of these analyses
to the more complex clinical situation still needs to be assessed.