Predicting Biological Response at a Systems Level: The VirtualLiver National Center for Computational Toxicology

1. Imran Shah (shah.imran@epa.gov) Predicting Biological Response at a Systems Level: The VirtualLiver National Center for Computational Toxicology

2. Overview Toxicity Dose Environmental Chemical How does toxicity depend on dose ? Does it cause Toxicity? What are the Mechanisms ? Computational Toxicology ToxCast™ Screening (HTS, -omics) Chemical Prioritization http://www.epa.gov/comptox/toxcast/ VirtualLiver Species differences Dose-dependant response

3. Dose-response Prediction Challenges • Injury is generally measured in laboratory animals, which is extrapolated to predict human dose and injury • Extrapolation issues: • Limited mechanistic knowledge • Interspecies differences • Limited/absent of low dose data Extrapolation response Animal Human dose dose

4. Regression techniques used to fit response as a function of dose Permissible exposure based on LOAEL & NOAEL Permissible exposure based on LOAEL & NOAEL Difficult to extrapolate dose, species, chemicals, etc. Empirical Dose-response Prediction Empirical models Animal Data Regression y Response dose d dose d ? LOAEL y = f(d) NOAEL

5. Physiological Modeling of Chemical Exposure Andersen et. al. 1999

6. Dynamic simulation of response based on mechanism Low-dose behaviour predicted by causal model as opposed to empirical fit Issues: limited mechanistic insight Physiological Dose-response Prediction Mechanistic models y dose d

7. Focus on Mode of Action … Environmental Chemicals Molecular response Cellular response Tissue response Chemicals Molecular Response (Early) Cell fate Proliferation Death Apoptosis Necrosis Adverse Outcome Hyperplasia Tumor Cancer NR-sig Gene-reg. Transcription Pesticides Conazoles Pyrethroids Toxics DE-71 PCBs Phthalates PFOA/PFOS CAR PXR PPARa cis-reg. trans-reg. Xen. Met. Phase I Phase II Phase III NR activators stimulate intracellular processes that lead to hyperplasia Chronic stimulation increases the risk of neoplasms

8. Physiologic State Tissue Cell & Vascular Net Cell State Cells Cellular processes Molecular Networks Molecules Pathways Interactions Synthesizing Physiologic Response: Systems Biology Tissue histomorphometry anchors the components of the intercellular network Cell state changes i.e. proliferation and apoptosis guide the selection of cellular processes Key cellular processes in the mode of action anchor the development of molecular interaction network models Biological Organization Bottom-up Systems Models Anchor Systems Models With Physiologic States

9. VirtualLiver: Multi-scale Biological Simulation Response Pathways Molecules Pathways Dose Interactions Cell State Tissue Histology Cells Lobule Intra/inter cellular modules Cell Networks Databases Literature Experiments Disparate sources of biological information Knowledgebase of components & functions Qualitative Model Description Quantitative Dynamic Simulation

10. Liver Knowledgebase Development RDF-Gravity & Sesame Semantic Repository

11. Gene-expression Proteomics Metabolomics Histomorphometry Activity profiles Expanding the Knowledgebase: -omics Brain Blood Liver Urine Transcripts Brain Proteins Blood Metabolites Liver Kidneys Quant. Path. Urine Stressors Biological System Large-scale Assays Global Perturbation

12. Discovering Conserved Network Motifs Liver Network Inference Conserved Motifs Chemicals -omic Profiles Computational Learning Inferred Molecular Networks

13. Modeling Biological Circuitry Huang, 1999

14. Attractors are stable to minimal perturbation => physiologically important System can go from one attractor to another => memory Simulating Boolean Dynamics Liang et al, 1998 DDLab

15. Nuclear Receptor Mediated Signaling & Gene-regulation

16. Boolean Network Modeling of NR Network Represent molecular interactions though logical (using a truth-table) Compute system trajectory by successive application of these rules Explore trajectory and stable states Reverse engineer network from –omic data Differences between rodents and humans

17. Simulating Dynamics Each point is a configuration of the boolean network attractors Boolean network System dynamics: Trajectory

18. Physiological Relevance • Attractors could represent stable physiological states • Attractors are sensitive to external perturbations • Perturbations can cause a switch from one attractor to another – represent changes in cell states: division/death Huang, 1999

19. Xenobiotic metabolism Cell Cycle Nuclear receptor signaling MAPK signaling Apoptosis Oxidative Stress Model Intracellular Network Hepatocyte

20. Xenobiotic metabolism Cell Cycle Nuclear receptor signaling MAPK signaling Apoptosis Oxidative Stress Model Intercellular Network Kupffer Cell TNF-a IL-1 IL-6

21. Inflammation Proliferation Apoptosis Dose Dose Dose Model Tissue as Cellular & Vascular Network

22. Summary & Future Outlook • Omic assays: can provide insight into conserved biological response across chemicals, species (life stages, etc.) • Need novel in vitro cultures coupled with sophisticated in silico models • Dose-response modeling: Systems biology+physiological modeling (VirtualLiver) • Reduce the need for animal testing through in vitro and in silico approaches

23. Acknowledgements Hugh Barton Jerry Blancato Rory Conolly David Dix Keith Houck Elaine Hubal Richard Judson Robert Kavlock David Reif Woody Setzer Imran Shah John Wambaugh Chris Corton Mike DeVito Stephen Edwards Hisham El-Masri Nicholas Luke Julian Preston Doug Wolf Chris Lau NHEERL NCCT NERL NCEA Rob Dewoskin Paul Schlosser Miles Okino Daniel Chang

24. Cancer ReproTox DevTox NeuroTox PulmonaryTox ImmunoTox Current Approach for Toxicity Testing in vivo testing

25. Cancer ReproTox DevTox NeuroTox PulmonaryTox ImmunoTox Future of Toxicity Testing in vitro testing in silico analysis HTS -omics Bioinformatics/ Machine Learning The ToxCast™ Project

26. ToxCast™: Aligned with NAS Report on Toxicity Testing …

27. Using –omic Data: Treatment Groups

28. Using –omic Data: Mode of Action

29. Non-carcinogens 75 compounds Non-genotoxic hepatocarcinogens 25 compounds Using –omic Data for Toxicity Classification / Hazard ID Iconix Toxicogenomics Example: • 152 Chemicals • Training Set: • 25 Non-genotoxic Hepatocarcinogens • 75 Non Hepatocarcinogens • Use Codelink Arrays Thanks to: Mark Feilden Richard Brennan Jeremy Golub

30. 37 genes in signature Map to IPA 24 genes Create Network Top Network 12 focus genes (grey) Function enrichment Top Functions Cell growth and proliferation Gene expression/TF’s Cell cycle Biomarkers to Mechanisms …

31. -omics data issues • Genomics data limitations: • Measures tissue average snapshot of mRNA content • mRNA content not always correlated with protein levels • To assess broader molecular context miRNA, protein and metabolite assays may also be necessary • Evolving large-scale assaying technologies – reduction in technical variability expected • Large number of features but relatively few obervations – biological variability continues to be problem

32. GTF: Genomics Data & Risk Assessment Page 17

33. Gather data Curate some data manually from literature and from molecular interaction / pathway databases Store data in Sesame / MySQL Networks Pathways Interactions Capturing Biological Knowledge • Develop ontology using existing standards e.g. BioPAX (OWL/RDFS) • Computational problems • Literature: Text mining • Importing public / commercial data • Inference tools Model of NR-signaling Interactions, pathways and network induced by ligands PPARa, CAR, PXR Molecules Components Functions

34. Modeling & Simulating Biological Networks Qualitative Model Building • Knowledge-driven: Infer causal biological network mediated by nuclear receptors from molecular interactions in KB • Data-driven: Infer biological network using –omic profiles due to chemical perturbation Quantitative Simulation • Discrete: Represent molecule levels discretely to evaluate network topology • Continuous: Additional measurements to calibrate models for response prediction