Office of Research and Development
Using Tox21 Data for Risk Assessment and Alternatives AssessmentRichard JudsonU.S. EPA, National Center for Computational ToxicologyOffice of Research and Development
The views expressed in this presentation are those of the author and do not necessarily reflect the views or policies of the U.S. EPA
Mid-Atlantic SOT, May 2014
Office of Research and DevelopmentNational Center for Computational Toxicology
Problem Statement
2
Too many chemicals to test with standard animal-based methods
–Cost, time, animal welfare
Need for better mechanistic data- Determine human relevance
- What is the Mode of Action (MOA) or Adverse Outcome Pathway (AOP)?
Office of Research and DevelopmentNational Center for Computational Toxicology
ToxCast / Tox21 Overall Strategy
• Identify targets or pathways linked to toxicity (AOP focus)• Develop high throughput assays for these targets or pathways• Develop predictive systems models
– in vitro → in vivo
–in vitro → in silico
• Use predictive models (qualitative):–Prioritize chemicals for targeted testing –Suggest / distinguish possible AOP / MOA for chemicals
• High Throughput Risk Assessments (quantitative)• High Throughput Exposure Predictions
3
Office of Research and DevelopmentNational Center for Computational Toxicology 4
Testing under ToxCast and Tox21Chemicals, Data and Release Timelines
Set Chemicals Assays Endpoints Completion Available
ToxCast Phase I 293 ~600 ~700 2011 Now
ToxCast Phase II 767 ~600 ~700 03/2013 10/2013
ToxCast Phase IIIa 1001 ~100 ~100 Just starting 2014
E1K (endocrine) 880 ~50 ~120 03/2013 10/2013
Tox21 8,193 ~25 ~50 Ongoing Ongoing
Chemicals
Ass
ays
~600
~8,2000
Office of Research and DevelopmentNational Center for Computational Toxicology
ToxCast Phase II:1051 Chemicals x 791 Assay Readouts
5
ACEA: redAttagene: orangeApredica: blackBioSeek: greenNovascreen: grayTox21: violetOT: blue
Assays
Chem
icals
Office of Research and DevelopmentNational Center for Computational Toxicology 6
ToxCast and the Endocrine Disruptor Screening Program
http://www.epa.gov/endo/pubs/edsp21_work_plan_summary%20_overview_final.pdf
EPA Research provides basis for improving the suite of assays and models to advance chemical
prioritization and screening
The universe of chemicals passes through each version of the HTS/in silico
pipeline to evaluate chemicals in refined tests, or for new pathways, to evaluate improve and validate methods.
Chemical PrioritizationIncludes registration review timeline, physico-chemical properties, exposure estimates, in vitro assays and computer models (QSAR, expert systems, systems biology models).
Screening DecisionsNear Term = Incorporates HTS/in silico prioritization methods for post EDSP List 2Intermediate = Run subset of T1S assays indicated by HTS and in silico predictions Long Term = Full replacement of EDSP T1S Battery
Chemicals Of Regulatory Interest
in vitro HTS/ in silico (P1)Current EDSP T1S Battery
Test+
Test-
Near Term(<2 yrs)
FocusedEDSP Tier 2 Tests
WOE+
WOE-
Test-
in vitro HTS/ in silico (P2)in vitro/in silico focuses
subset of EDSP T1STest+Intermediate
Term (2-5 yrs)
WOE+
WOE-
in vitro HTS/ in silico (full replacement of Tier 1)Longer Term (>5 yrs) WOE+
WOE-
Office of Research and DevelopmentNational Center for Computational Toxicology 7
Office of Research and DevelopmentNational Center for Computational Toxicology
Computational Model
8
Office of Research and DevelopmentNational Center for Computational Toxicology
Example 1 – BPA: true agonist (AUC=0.66)
9
Office of Research and DevelopmentNational Center for Computational Toxicology
Example curves
10
True Agonist True Antagonist
Negative-BAI Negative-NAI
Office of Research and DevelopmentNational Center for Computational Toxicology
Reference Chemical Classification
11
AUC heat map for Reference chemicals
Office of Research and DevelopmentNational Center for Computational Toxicology
Major theme – all assays have false positives and negative
Much of this “noise” is reproducible, i.e. it is “assay interference”
Result of interaction of chemical with complex biology in the assay
Our chemical library is only partially “drug-like”-Solvents-Surfactants-Intentionally cytotoxic compounds-Metals-Inorganics
Assays cluster by technology,suggesting technology-specific
non-ER activity
Office of Research and DevelopmentNational Center for Computational Toxicology
Example illustrating assay data
13
Office of Research and DevelopmentNational Center for Computational Toxicology
Most chemicals display a “burst” of activity at same concentration as cytotoxicity
14
Most chemicals cause activity in many assays near the cytotoxicity threshold
Cell-stress related assay interference
“Hit” (AC50) in burst region is less likely to result from specific activity (e.g. binding to receptor or enzyme)
Z-score: # of SD from burst center-High Z: more likely to be specific-Low Z: less likely to be specific
Office of Research and DevelopmentNational Center for Computational Toxicology
Examine Z-scores by assay
15
Bimodal
Cytotox / Cell Stress“True” activity
Office of Research and DevelopmentNational Center for Computational Toxicology
Non-specificity with cytotox is general
16
Having cytotoxicity @<100 uM greatly increases number of hits
Chemicals with cytotoxicity @<100 uM have many hits, but few are outside of burst
Office of Research and DevelopmentNational Center for Computational Toxicology 17
* =Reference chemicals
- These chemicals should be near the right of the gene score distribution
- Most assays show reference chemicals to be potent and specific
- Gives confidence that novel chemicals active in the assay are perturbing that pathway
“Weak Validation”
After removing burst, most assays detect reference chemicals - 95% success
Office of Research and DevelopmentNational Center for Computational Toxicology
Promiscuity: Highest for chemicals designed to be bioactive
Least Promiscuous Chemical Classes
0.1-0.3% of assays are active
None designed to be bioactive
Most Promiscuous Chemical Classes
2-3% of assays are active
All designed to be bioactive
Category NchemMean Hit Ratio p-cold
alcohol primary 10 0.0011 0.00021phthalate 17 0.0032 0.00084carboxylate di 15 0.0028 0.0029carboxylate 7 0.0015 0.0042
Category NchemMeanHit Ratio p-hot
conazole (triazoles) 13 0.034 3.5E-06Pharma Class 4.86 10 0.031 1.1E-05Pharma Class 4.58 11 0.029 4.1E-05conazole (imidazoles) 6 0.031 0.003Pharma Class 3.292 5 0.039 0.0049steroid P 5 0.022 0.0052Pharma Class 4.43 7 0.020 0.0067
Office of Research and DevelopmentNational Center for Computational Toxicology
High throughput pharmacokinetic (HTPK) in vitro methods have been developed by pharmaceutical industry for predicting efficacious doses in clinical trials
In Wetmore et al. (2012) the same methods are used to approximately convert ToxCast in vitro bioactive concentrations (µM) into daily doses needed to produce similar levels in a human (mg/kg BW/day)
These doses can then be directly compared with exposure data, where available
Egeghy et al. (2012) and National Academy Report: “Exposure Science in the 21st
Century” points out that not much exposure information is out there
e.g. Judson et al., (2011)
Potential Exposure from
ExpoCast
mg/kg BW/day
Potential Hazard from ToxCast with
Reverse Toxicokinetics
LowRisk
MedRisk
HighRisk
High Throughput
Pharmacokinetics and Exposure
Office of Research and DevelopmentNational Center for Computational Toxicology20
ExpoCast Coverage of the ToxCast
Phase II Chemicals
ToxCast Oral Equivalents based on unpublished data from Barbara Wetmore
Ora
l Eq
uiv
alen
t D
ose
s an
d E
stim
ated
Exp
osu
res
(mg
/kg
/day
)
Predictions from Wambaugh et al. (2013) ExpoCast model with USEtox, RAIDAR, and near field/far field heuristic
Office of Research and DevelopmentNational Center for Computational Toxicology
Chemical use information informs exposure scenarios
• CPCat : Chemical and Product categories– Public use information on 40,000 chemicals– http://actor.epa.gov/cpcat
Office of Research and DevelopmentNational Center for Computational Toxicology
Approach for Rapid Validation and Application of Alternatives
22
Office of Research and DevelopmentNational Center for Computational Toxicology
Understanding Success and Failure
• Why In vitro to in vivo can work:–Chemicals cause effects through direct molecular interactions that
we can measure with in vitro assays
• Why in vitro to in vivo does not always work:–Pharmacokinetics issues: biotransformation, clearance (FP, FN)–Assay coverage: don’t have all the right assays (FN)–Tissue issues: may need multi-cellular networks and physiological
signaling (FN)–Statistical power issues: need enough chemicals acting through a
given MOA to be able to build and test model (FN)–Homeostasis: A multi-cellular system may adapt to initial insult
(FP)– In vitro assays are not perfect! (FP, FN)– In vivo rodent data is not perfect! (FP, FN) 23
SystemsModels
Office of Research and DevelopmentNational Center for Computational Toxicology
AcknowledgementsEPA NCCTRusty ThomasKevin CroftonKeith HouckAnn RichardRichard JudsonTom KnudsenMatt MartinWoody SetzerJohn WambaughMonica LinnenbrinkJim RabinowitzSteve Little
Agnes ForgacsJill FranzosaChantel NicolasBhavesh AhirNisha SipesLisa TruongMax LeungKamel MansouriEric WattCorey Strope
EPA NCCTNancy BakerJeff EdwardsDayne Filer Jayaram KancherlaParth KothiyaJimmy PhuongJessica LiuDoris SmithJamey VailHao TruongSean WatfordIndira ThillainadarajahChristina Baghdikian
NIH/NCATSMenghang XiaRuili HuangAnton Simeonov
NTPWarren CaseyNicole KleinstreuerMike DevitoDan ZangRay Tice
EPA CollaboratorsKathie DionisioKristin IsaacsPeter EgeghyDavid DixAlan DixonScott LynnPatience BrownDon BergfeltLes Touart