Assessing Carcinogenic Potential of Chemicals Using OncoLogic Cancer Expert System

1. Assessing Carcinogenic Potential of Chemicals UsingOncoLogic Cancer Expert System Yin-tak Woo, Ph.D., DABT Office of Pollution Prevention and Toxics U.S. Environmental Protection Agency Washington, DC 20460 May 19, 2010

2. Outline of the Presentation • Scientific background in development of the OncoLogic system • Brief description of the OncoLogic system • Recent development in updating and expanding OncoLogic system

3. SAR/QSAR: Background & Issues • SAR/QSAR: activity = f (structure) • Given sufficient data/knowledge on related compounds  screen well defined endpoint • Evolution of SAR/QSAR: from human intuition to cyber sophistication • Impact of commercial software • User base: from domain expert to nonscientists • Pressure of reduction in experimentation

4. Approaches to SAR/QSAR • Statistical vs. Mechanistic • Local/Homogeneous/Congeneric vs. Global/General/Heterogeneous • Active/Inactive vs. Ranking vs. Potency

5. Criteria for assessing scientific soundness • Selection of endpoint • Knowledge base/training database/applicability domain • Methodology and descriptor selection • Model validation (predictive accuracy, internal, external, prospective) • Transparency and scientific rationale • Confidence/uncertainty analysis • Strengths, weaknesses and limitations

6. Importance of mechanistic understanding in (Q)SAR modelling • selection of toxicological endpoint • selection of molecular descriptors • coverage of training database • consideration of database stratification • interpretation of outliers • consideration of human relevance • achieving the goal of statistical association with mechanistic backing

7. ADME/Toxicokinetics considerationAbility of toxicant to reach target tissue/molecule • Chemical structure/phys-chem on ADME • Route of administration • Facilitating “carrier” molecule • Protective “carrier” molecule • Biological half-life

8. Mechanistic/Toxicodynamics consideration • Electrophilic • Receptor-mediated • Disruption of homeostasis • Multiple mechanisms

9. Future Trend of (Q)SAR • Critical evaluation of current methods • Expansion of publicly accessible databases/knowledge bases • Expansion of integrative approaches • Utilization of input from emerging predictive technologies

10. Incorporating emerging predictive technologies Chemical In silico ADME, ADME assays ADME/Toxicokinetics Consideration T O X I C G G E N O M I C S Mechanistic/Toxicodynamics Consideration HTS assays Structural Functional Molecular descriptors, Structural features, etc Screening assays, Biomarkers, etc Prediction

11. Major References for (Q)SARas a Screening Tool Woo YT, Lai DY (2003): Mechanism of action of chemical carcinogens and their role in SAR analysis and risk assessment. In: Quantitative Structure-Activity Relationship (QSAR) Models of Mutagens and Carcinogens, R. Benigni, ed., CRC Press, Boca Raton, FL Doull D, Borzelleca J, Becker R, Daston G, DeSesso J, Fan A, Fenner-Crisp P, Holsapple M, Holson J, Llewellyn G, MacGregor J, Seed J, Walls I, Woo Y, Olin S (2007): Framework for use of toxicity screening tools in context-based decision-making. Food Chem.Toxicol. 45:759-796.

12. Development of OncoLogic Cancer Expert System: Scientific Background

13. Introduction to the Cancer Endpoint • Definitions • Uncontrolled dividing and growth of cells • Caused by mutations, ↑ cell proliferation, ↓ cell death, loss of homeostatic control, etc. • Two general mechanisms by which a chemical can induce cancer • Genotoxic (default) • Interaction with DNA to cause mutation(s) in genes • Non-genotoxic • Variety of mechanisms

14. Introduction to the Cancer Endpoint (Cont.) • Carcinongesis is a multistage/multistep process • Initiation • Mutation converts normal to preneoplastic cells • Promotion • Expansion of preneoplastic cells to benign tumors • Progression • Transformation of benign to invasive malignant tumors • A potent carcinogen acts directly on all three stages • A weak carcinogen acts directly on one stage and indirectly on other

15. Initiation Promotion Progression Main event(s) Direct DNA binding Indirect DNA damage Clonal expansion Cell proliferation Apoptosis Differentiation Overcoming suppressions (e.g., p53, immune, angiogenesis) Key mechanistic consideration Electrophile, resonance stabilization, nature of DNA adduct Receptor, cytotoxicity, gene expression Free radical, receptor, gene suppression Signal transduction, homeostasis SAR/QSAR mechanistic descriptors Electrophilicity, HOMO/LUMO, delocalization energies, …… 2D, 3D, docking, biopersistence, methylation, …. Reduction potential, 2D, 3D, ……

16. Difficulties of (Q)SAR of carcinogenicity • Complex, mechanism-dependent (Q)SAR • Local vs. global models • Data scarcity and variability • Feedback and validation issues • Need for integrative approach

17. Historical development of OncoLogic • TSCA and New Chemicals Program (PMN) • Structure-Activity Team approach • Need to provide guidance to industries • OncoLogic Team (Joseph Arcos, Mary Argus, David Lai, Yin-tak Woo) • LogiChem coop version • Current version • Future developments

18. OncoLogic: A mechanism-based expert system for predicting carcinogenic potential • Developed by domain experts in collaboration with expert system developer • Knowledge from SAR on >10K chemicals • Class-specific approach to optimize predictive capability • Consider all relevant factors including biological input when possible • Predictions with scientific rationale and semiquantitative ranking

19. Major Sources of Data/Insight Used to Develop Cancer Knowledge Rules • The OncoLogic Team and members of SAT • Chemical Induction of Cancer monograph series • IARC monograph series • NCI/NTP technical reports • Survey of compounds which have been tested for carcinogenic activity, PHS Publ. 149 • Non-classified EPA submission data from various EPA program offices • Current literature and ad hoc expert panels

20. Profile of most potent carcinogens • Ability to reach target tissue • Reasonable lifetime of ultimate carcinogen • Persistent and site-specific interaction with target macromolecule • Ability to affect all three stages of carcinogenesis

21. Development of rules for each class • Gather all available data and information • Brainstorming to determine key factors • Determine need for subclassification • Assign concern levels to known carcinogens • Determine mechanism-based modification factors for substituents • Develop rationale for conclusion

22. Concern Levels

23. Critical Factors for SAR Consideration • Electronic and Steric Factors • Resonance stabilization • Steric hindrance • Metabolic Factors • Blocking of detoxification • Enhancement of activation

24. Critical Factors for SAR Consideration • Mechanistic Factors • Electrophilic vs. receptor- mediated • Multistage process • Physicochemical Factors • Molecular weight • Physical state • Solubility • Chemical reactivity

25. OncoLogic®Factors Affecting Carcinogenicity of Aromatic Amines • Number of aromatic ring(s) • Nature of aromatic ring(s) - homocyclic vs. heterocyclic - nature and position of heteroatoms • Number and position of amino or amine-generating groups(s) - position of amino group relative to longest resonance pathway - type of substituents on amino group • Nature, number, position of other ring substituent(s) - steric hindrance - hydrophilicity • Molecular size, shape, planarity

26. R-NO2 R-NO R-NH-OH R-NH2 R-NH-OAc [R-N(CH3)2] Some Hydrocarbon Moieties Present in Carcinogenic Aromatic Amines

27. - + + + Carbonium ion Amidonium ion Molecular Mechanism for Generation of Resonance-stabilized Reactive Intermediates from N-acyloxy Aromatic Amines

28. 5’ 6’ 6 5 4’ 4’ 3’ 2’ 2 3 Very active if: OC•CH3 --N OH Active if: --NH•OC•CH3 --N(CH3)2 --NO2 --OCH3 Inactive if: --F R | --CH-- Transition to diphenylmethane and triphenyl- methane amines --CH==CH-- Transition to amino- stilbenes --N==N-- Transition to amino azo dyes Very active if: --F Active if: --NH2 --NH•OC•CH3 --NO2 Weakly active if: --C6H5 Inactive if: --CH3 --Cl --Br Active if: --S-- Inactive if: --NH-- For 4-aminobiphenyl: Very active if: 3-methyl 3,2’-dimethyl 3,3’-dimethyl 3-fluoro 3’-fluoro Active if: 3-chloro 3-methoxy 3,2,5’-trimethyl 3,2’,4’,6’-tetramethyl Weakly active if: 3-hydroxy Inactive if: 2-methyl 2’-methyl 2’-fluoro 3-amino For benzidine: Very active if: 2-methyl 3,3’-dihydroxy 3,3’-dichloro Weakly active if: 3,3’-dimethyl 3,3’-dimethoxy Inactive if: 2,2’-dimethyl 3,3’-bis-oxyacetic acid Synoptic Tabulation of Structural Requirements for Carcinogenic Activity of 4-Aminobiphenyl and Benzidine Derivatives

29. C = Clear evidence of carcinogenicity S = Some evidence of carcinogenicity N = No evidence of carcinogenicity NT = Not tested + = At least one test = C or S Eq = No C or S, and E must appear at least once -- = No C, S, or E OncoLogic® Prediction vs. NTP BioassaysAromatic Amines and Related Compounds

30. Examples of how “Knowledge Rules” can be used in chemical design Strategies to Designing Safer Chemicals: • Steric hindrance • Nonplanarity • Electronic insulation • Hydrophilicity OncoLogic Cancer Concern = High

31. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential

32. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential

33. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential (Cont.)

34. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential (Cont.)

35. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential (Cont.)

36. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential (Cont.)

37. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential (Cont.)

38. Molecular Design of Aromatic Amine Dyes with Lower Carcinogenic Potential (Cont.)

39. C = Clear evidence of carcinogenicity S = Some evidence of carcinogenicity N = No evidence of carcinogenicity NT = Not tested + = At least one test = C or S Eq = No C or S, and E must appear at least once -- = No C, S, or E OncoLogic® Prediction vs. NTP BioassaysAromatic Amines and Related Compounds

40. Conclusions from NTP Cancer Bioassays Predictive Exercises • Most of the best performers are predictive systems that incorporate human expert judgment and biological information • OncoLogic was one of the best performers among more than 15 methods

41. Final results of 2nd NTP predictive exercise(from Benigni and Zito, Mutat. Res. 566, 49, 2004)

43. FDA Validation of Genetic Toxicity and SAR Methods for Predicting Carcinogenicity* *from Mayer et al.: SAR analysis tools: validation and applicability in predicting carcinogens. Regulatory Toxicol. Pharmacol. 50: 50-58, 2008

44. Sensitivity and Specificity of the Genetic Toxicity and SAR Methods for Predicting Carcinogenicity

45. FDA Validation of Genetic Toxicity and SAR Methods for Predicting Potent Carcinogenicity

46. OncoLogic® - Benefits • Allow non-experts to reach scientifically supportable conclusions • Expedites the decision making process • Allows sharing of knowledge • Reduces/eliminates error and inconsistency • Formalize knowledge rules for cancer hazard identification (SAT-style) • Bridge expertise of chemists and toxicologists for most effective hazard evaluation • Provide guidance to industries on elements of concern for developing safer chemicals

47. Some Notable Uses of OncoLogic • OPPT (new chemicals, design for environment, green chemistry, existing chemicals) • Guidance to industries (Sustainable Future program) • OW (disinfection byproduct prioritization) and other EPA program offices • FDA (food contact notification) and other governmental agencies

48. Developed by recognized domain experts Knowledge not just data Local models with strong mechanistic basis Integrates biological input when possible Semiquantitative ranking with scientific rationale Proven performance in prospective and external validations Industrial chemicals Users need to have some organic chemistry background Coverage limited by available knowledge No batch mode Some updates are needed Current coverage mainly on established carcinogen classes Limited receptor-based SAR Pharmaceuticals? Strengths Limitations

49. Overall Structure of The Cancer Expert System

50. Running OncoLogic® • Two methods to predict carcinogenicity • SAR Analysis • Knowledge rules • Functional Analysis • Uses results of specific mechanistic/non-cancer studies