Genomic Dose Response

Genomic Dose Response:

The Picture

NTP Genomic Dose Response Modeling Expert Panel Meeting October 23, 2017

Russell Thomas Director National Center for Computational Toxicology

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 Scott May Want To Rethink Asking Me To Be The “Big Picture Guy”…

National Center for Computational Toxicology It is a Well Known Fact that Toxicology Continues to Have a Data Problem

US National Research Council, 1984

Size of Estimate Mean Percent Category Category In the Select Universe

Pesticides and Inert Ingredients of Pesticides 3,350 Formulations 10 24 2 26 38

Cosmetic Ingredients 3,410

2 14 10 18 56

Drugs and Excipients Used in Drug Formulations 1,815 18 18 3 36 25

Food Additives 8,627 • Major challenge is too many 5 14 1 34 46 Chemicals in Commerce: chemicals and not enough At Least 1 Million 12,860 Pounds/Year data 11 11 78 Chemicals in Commerce: • Total # chemicals = 65,725 Less than 1 Million 13,911 Pounds/Year • Chemicals with no toxicity 12 12 76 data of any kind = ~46,000 Chemicals in Commerce: Production Unknown or 21,752 Inaccessible 10 8 82

Complete Partial Minimal Some No Toxicity Health Health Toxicity Toxicity Information Hazard Hazard Information Information Available Assessment Assessment Available Available 2 Possible Possible (But Below Minimal)

National Center for Computational Toxicology It is a Well Known Fact that Toxicology Continues to Have a Data Problem

Percent of Chemicals <1% 10

0 Acute Cancer

Gentox Dev Tox

Repro Tox EDSP Tier 1

Modified from Judson et al., EHP 2009 3

National Center for Computational Toxicology For Those With Data, Have We Been Truly Predictive or Just Protective?

…data compiled from 150 compounds with 221 human toxicity events reported. The results showed the true positive human toxicity concordance rate of 71% for rodent and nonrodent species, with nonrodents alone being predictive for 63% of human toxicity and rodents alone for 43%.

National Center for Computational Toxicology Budding Field of Toxicogenomics Promised to Change All That…

Rapid Mode Hazard Of Characterization Action

National Center for Computational Toxicology Initial Focus of Toxicogenomics Was on Inferring MOA

-Log P-value 0 2 4 6 8 10

1. Neurophysiological process_Circadian rhythm

2. Cytoskeleton remodeling_Cytoskeleton remodeling • Initial 3.a NRF2pplications regulation of oxidative stress response of toxicogenomics to identify MOA have generally 4. Cytoskeleton remodeling_TGF, WNT and cytoskeletal lacked a systematicremodeling approach 5. Development_BMP signaling

• More 6.art Cell adhesion_Role than of tetraspanins science in the integrin- mediated cell adhesion

• Development7. Signal transduction_JNK pathway of large reference databases difficult due to cost ME Cellular Effects Tissue Effect Organ Effect constraints5 day, 90 ppm 8. Development_EGFR signaling via PIP3 5 day, 200 ppm 9. Cell adhesion_Integrin-mediated cell adhesion and Hepatocyte Cell Liver 15 day, 30 ppm migration ER Stress Cell Death Hyperplasia • Most expert committees/reports defaultedInjury to usingProliferation it as partTumors of an 15 day, 90 ppm overall15 day, 200 ppm 10. Methionine weight-cysteine-glutamate- metabolismof-evidence • Not very satisfying

MOA Generator

National Center for Computational Toxicology Focus Shifted Towards Supervised Classification Approaches

• Most studies show 60-85% accuracy for predicting cancer- related endpoints • Only a limited number of tissues have been evaluated • Requires >20 compounds with adequate redundancy and diversity in mechanisms to have a robust training set (Thomas et al., 2009) • >30 organs show tumor responses in NTP database with ~50% having >10 chemicals in at least one species/sex • Difficult to justify as a comprehensive screen for rodent carcinogenicity

National Center for Computational Toxicology Continued Evolution Toward More Quantitative Applications

TP53 BMDL NFkB RfD UFs

BMR Apical POD Apical

BMDL BMD BMDL10 RSD 100,000 Transcriptional POD

ShortSubchronic-term Animaland TranscriptionalMost Sensitive FitFit Each Data Genewith IdentifyIdentifyGroup PointPoint Genes ofof By DeriveCompare Safe In Vivo ChronicStudies Animal PathologicalResponses withStatistical Statistical DepartureSignaling (i.e., BMD) Pathway ExposureApical Level and Studies Response Models and Calculate Transcriptional Considered Summary Value PODs Adverse (usually the median pathway BMD) Thomas et al., Tox Sci., 2011 Thomas et al., Mut Res., 2012 Thomas et al., Tox Sci., 2013

National Center for Computational Toxicology In Vivo Study to Assess Transcriptional and Apical Correlation

Chemical Route Dosesc Rodent Model Time Point Target Tissue 1,2,4-Tribromobenzenea Gavage 2.5, 5, 10, 25, 75 mg/kg Male Sprague Dawley rats 5 d, 2, 4, 13 wks Liver Bromobenzenea Gavage 25, (50), 100, 200, 300, 400 mg/kg Male F344 rats 5 d, 2, 4, 13 wks Liver 2,3,4,6- Gavage 10, 25, 50, 100, 200 mg/kg Male Sprague Dawley rats 5 d, 2, 4, 13 wks Liver Tetrachlorophenola 4,4'-Methylenebis (N,N- Feed 50, 200, 375, 500, 750 ppm Male F344 rats 5 d, 2, 4, 13 wks Thyroidb dimethyl) benzenamineb N-Nitrosodiphenylamineb Feed 250, 1000, 2000, 3000, 4000 ppm Female F344 rats 5 d, 2, 4, 13 wks Bladderb HydrazobenzeneMeasuredb apical (histologicalFeed 5, and 20, 80, organ 200, 300 weight; ppm n = 10) andMale F344gene rats expression5 d, 2, changes4, 13 wks (nLiver = 5)b at each dose and time point in the target tissue. 1,4-Dichlorobenzeneb Gavage 100, 300, 400, 500, 600 mg/kg Female B6C3F1 mice 13 wks Liver Propylene glycol mono-t- Inhalation 25, 75, 300, 800, 1200 ppm Female B6C3F1 mice 13 wks Liver butyl etherb 1,2,3-Trichloropropaneb Gavage 2, 6, 20, 40, 60 mg/kg Female B6C3F1 mice 13 wks Liver Methylene Chlorideb Inhalation 100, 500, 2000, 3000, 4000 ppm Female B6C3F1 mice 13 wks Liver, Lung Naphthaleneb Inhalation 0.5, 3, 10, 20, 30 ppm Female B6C3F1 mice 13 wks Lung 1,4-Dichlorobenzeneb Gavage 100, 300, 400, 500, 600 mg/kg Female B6C3F1 mice 13 wks Liver aChemicals in IRIS database for non-cancer endpoints only bChemicals previously tested by the U.S. National Toxicology Program cUnderlined doses used in NTP two-year rodent bioassay or IRIS database

National Center for Computational Toxicology Temporal Changes Between Transcriptional and Non-Cancer PODs

1000 1000 5 Days 2 Weeks Bladder Liver Thyro id

100 100

10 10 Lowest Apical BMD (mg/kg/d) BMD Apical Lowest Lowest Apical BMD (mg/kg/d) BMD Apical Lowest

r = 0.909 r = 0.987 1 1 1 10 100 1000 1 10 100 1000 Lowest Pathway Transcriptional BMD Lowest Pathway Transcriptional BMD (mg/kg/d) (mg/kg/d) 1000 1000 4 Weeks 13 Weeks

100 100

10 1 Lowest Apical BMD (mg/kg/d) BMD Apical Lowest Lowest Apical BMD (mg/kg/d) BMD Apical Lowest

r = 0.985 r = 0.982 1 0.1 1 10 100 1000 0.1 1 10 100 1000 Lowest Pathway Transcriptional BMD Lowest Pathway Transcriptional BMD (mg/kg/d) (mg/kg/d) Thomas et al., Toxicol Sci, 2013

National Center for Computational Toxicology Combined Correlation Between Non- Cancer and Transcriptional PODs

10000 Bladder Median Log2 Ratio Apical:Transcriptional BMD = 0.068 (1.05) Liver Thyro id 1000 Lung

Rat Mouse 100

10 Lowest Apical BMD (mg/kg/d) BMD Apical Lowest 1

r = 0.827 (p = 0.0031) 0.1 0.1 1 10 100 1000 10000 Lowest Pathway Transcriptional BMD (mg/kg/d) Thomas et al., Toxicol Sci, 2013

National Center for Computational Toxicology Temporal Changes Between Transcriptional and Cancer PODs

1000

Bladder

100

Thyroid

10 Liver Tumor BMD (mg/kg/d) BMD Tumor

1 5 d

2 wk

4 wk

13 wk 0.1 0.1 1 10 100 1000 Thomas et al., Toxicol Sci, 2013 Lowest Pathway Transcriptional BMD (mg/kg/d) National Center for Computational Toxicology Combined Correlation Between Cancer and Transcriptional PODs

10000 Median Log2 Ratio Apical:Transcriptional Bladder BMD = 0.524 (1.44) Liver Thyro id 1000 Lung

Rat Mouse 100

10 Tumor BMD Tumor BMD (mg/kg/d)

r = 0.940 (p = 0.0002) 0.1 0.1 1 10 100 1000 10000 Lowest Transcriptional BMD (mg/kg/d) Thomas et al., Toxicol Sci, 2013 National Center for Computational Toxicology Why Could this Be True?

• Most histological changes do not occur without upstream or downsream changes in the transcriptome

• Most environmental chemicals are highly non-selective in their interactions with biological systems

Thomas et al., Tox Sci., 2013 National Center for Computational Toxicology Integration of TGX in a Tiered 21st Century Toxicity Testing Framework

Protective Predictive Protective Predictive 15

National Center for Computational Toxicology The Next Frontier for Toxicogenomics

ToxCast ToxCast ~600 assays ~1,000 chemicals Tox21 Tox21 ~50 assays ~10,000 chemicals Response

16 Concentration National Center for Computational Toxicology Developing a Portfolio of High- Throughput Toxicogenomic Tools

High-Throughput Transcriptomic Screening Platform (TempOSeq) • Low cost, 384-well, cell lysate compatible • Whole transcriptome (EPA), S1500+ (NTP) • Workflow integration of reference materials and controls, development of performance standards • Portable platform/workflow for collaborative data generation

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 A MAQC-A (Us) 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 non-treated B MAQC-A (Us) 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 non-treated Platform SelectionC MAQC-B (Us) 10 10 10 10 10 10 10 10Workflow10 10 10 10 Verification10 10 10 10 10 Study10 10 10 10 10 non-treated D MAQC-B (Us) 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 DMSO E Bulk Lysate (DMSO) •1 1Half1 1 way1 1 through1 1 1 1 first1 1 large1 1 scale1 1 1 screen1 1 1 1of 12,200DMSO Technical PerformanceF Bulk Lysate (DMSO) 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 DMSO G Bulk Lysate (Trichostatin) 0.1 0.1chemicals0.1 0.1 0.1 0.1 0.1(ToxCast0.1 0.1 0.1 0.1 I/II/III)0.1 0.1 0.1 in0.1 single0.1 0.1 0.1 cell0.1 0.1 type0.1 0.1 DMSO [No Label] H Bulk Lysate (Trichostatin) 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03Trichostatin0.03 0.03 0.03 0.03 A0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 Trichostatin (1 µM) I Lysis Buffer (Us) •100 100Additional100 100 100 100Raw 100screens 100 100 100 100across100 100 Processed100 multiple100 100 100 100 cell100 100 types/lines100 100 Trichostatin (1 µM) EPAJ Lysis Buffer (Us) 30 30 30 30 30 30 30 30 Org30 30 #130 30 30 30 30 30 30 30 30 Ref30 30 30 Trichostatin (1 µM) K MAQC-A (Them)Lysate 10 10 10 10 10 Data10 10 10 10 10 10 10 10 10 Data10 10 10 10 10 10 10 10 Genistein (10 µM) MAQC-A (Them) •3 3Tox213 3 3 cross3 3 -3partner3 3 3 project3 3 3 3to3 add3 3 reference3 3 3 3 Genistein (10 µM) L Chems MOrg #2MAQC-B (Them) 1 1 Assay1 1 1 1 1 1 1 Org1 1 #21 1 1 1 1 1 1 1 1 1 1 Genistein (10 µM) N MAQC-B (Them) 0.3 0.3chemical0.3 0.3 0.3 0.3 database0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 Sirolimus (0.1 µM) O OrgLysis #3 Buffer (Them) 0.1 Provider0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 Org0.1 0.1 #30.1 0.1 Corr 0.1 0.1 0.1 0.1 0.1 0.1 Other0.1 0.1 Sirolimus (0.1 µM) P Lysis Buffer (Them) •0.03 0.03G0.03enetic0.03 0.03 0.03 perturbations0.03 0.03 0.03 0.03 0.03 0.03 (RNAi/CRISPR/cDNA)0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 Sirolimusfor (0.1 µM)

Screen Designr2 0.75 targets/pathways with no reference chemicals • Cytogenetically and functionally characterizedStandard cells Treatment Standard Analysis Data Repository •FunctionalConcentration PerformanceProtocols response Protocols Treatment Concentration • ParallelCMAP HCI Match screen • RNA and4/5 lysate(80%) reference materials CVs ~25-30% Randomized FC Exp Rep Corr ~0.8 17• Positive and negative controls treatment National Center for Josh Harrill, Unpublished Computational Toxicology Developing a Portfolio of High- Throughput Toxicogenomic Tools

High-Throughput Transcriptomic Screen • Low cost, 384-well, cell lysate compatible • Whole transcriptome (EPA), S1500+ (NTP) • Workflow integration of reference materials and controls, development of performance standards • Portable platform/workflow for collaborative data generation

Mode of Action/MIE Analysis • Refined CMAP tool and machine learning approaches • Curating reference chemical database for MIE and Karmaus, directional response Unpublished • >60 MIEs and growing

Concentration Response Analysis • BMDExpress 2.0 BMR • Tcpl

BMDL BMD

National Center for Computational Toxicology Expanding the Tiered 21st Century Toxicity Testing Framework

ERa Antagonist Aromatase

EC50 = 1.5 – 2.5 uM ?

Protective Predictive Protective Predictive 19

National Center for Computational Toxicology International Case Study Evaluating Bioactivity as a Conservative POD

ExpoCast PODToxCast (PODTraditional PODEFSA PODHC)

< Total = 95% 95%

ToxCast 379 chemicals POD ExpoCast httk, ToxCast data, and POD value(s) currently available

For ~87% of the chemicals, PODToxCast was conservative. = conservative ToxCast POD

Missing an < important trad ToxCast component POD POD of biology?

National Center for Computational Toxicology What is the Regulatory Need?

National Center for Computational Toxicology But, A Necessary Piece Is Convergence and Acceptance of Analysis Approaches

National Center for Computational Toxicology Acknowledgements and Questions

Tox21 Colleagues: EPA’s National Center for Computational Toxicology NTP FDA NCATS

EPA Colleagues: NERL NHEERL NCEA

Collaborative Partners: Unilever A*STAR ECHA EFSA Health Canada

National Center for Computational Toxicology