1.14k likes | 1.81k Views
Computational Toxicology. Richard Judson. The views expressed in this presentation are those of the author[s] and do not necessarily reflect the views or policies of the U.S. Environmental Protection Agency. UNC, November 2012. Big Ideas. Understand chemical toxicity at a molecular level
E N D
Computational Toxicology Richard Judson The views expressed in this presentation are those of the author[s] and do not necessarily reflect the views or policies of the U.S. Environmental Protection Agency. UNC, November 2012
Big Ideas • Understand chemical toxicity at a molecular level • Understand using as few animal as possible • Build predictive models • Screening and prioritization • Assess many chemicals – deal with the data gaps
Problem Statement Too many chemicals to test with standard animal-based methods • Cost, time, animal welfare • Exposure is as important as hazard • Need for better mechanistic data • - Determine human relevance • - What is the relevant Mode of Action (MOA) or Adverse Outcome Pathway (AOP)?
Benefits Less expensive More chemicals screened Fewer animals Solution oriented Innovative Multi-disciplinary Collaborative Catalytic Transparent Cancer ReproTox DevTox NeuroTox PulmonaryTox ImmunoTox Computational Toxicology in vitro testing Bioinformatics/ Machine Learning
Initial Objective: Risk-based Prioritization Initial Exposure Evaluation: Use Categories Chemical Universe >100,000 Chemicals with likely exposure potential Mixtures Chemicals w/o HTS or structural similarity HTS Chemical Library Structure Similarity Modeling Structural neighbors to HTS library AOP / MOA Targeted High-throughput testing Active chemicals and structural neighbors Inactive chemicals and structural neighbors Detailed Exposure and Toxicokinetics Evaluation Very Low priority bin High, Medium, Low priority bins
AOP / MOA Targeted High-throughput testing Hazard-based Approach • Identify molecular targets or biological pathways linked to toxicity • MOA / AOP • Chemicals perturbing these can lead to adverse events • Develop assays for these targets or pathways • Assays probe “Molecular Initiating Events” or “Key Events” [MIE / KE] • Develop predictive models: in vitro → in vivo • “Toxicity Signature” • Extend to inform biomarkers or bioindicators for key events • Use signatures: • Prioritize chemicals for targeted testing (“Too Many Chemicals” problem) • Suggest / distinguish possible AOP / MOA for chemicals
Chemical Receptors / Enzymes / etc. Direct Molecular Interaction Pathway Regulation / Genomics Cellular Processes Tissue / Organ / Organism Tox Endpoint Toxicity Pathways
AOP / MOA Targeted High-throughput testing AOP / MOA Targeted HTS Data AOP / MOA Development • International workgroups developing frameworks and models • OECD – AOP • WHO – MOA • Key Concepts • Molecular Initiating Events or Key Events – measureable in vitro • Causal evidence for downstream effects • AOP includes effects up to the population level Ankley et al. 2010
AOP / MOA Targeted High-throughput testing AOP / MOA Targeted HTS Data Proposed AOP: Embryonic Vascular Disruption Knudsen and Kleinstreuer. Birth Def Res C. 2012
ToxCast • Combine High-throughput screening with computer models
Key Research and Tools Toxicity Forecaster (ToxCast) • 500 fast, automated chemical screens (in vitro) • Builds statistical and computer models to forecast potential chemical toxicity • Phase 1: Screened over 300 well characterized chemicals • Phase 2: 700 more chemicals representing broad structures • Multi-year, multi million dollar effort • Tox21 collaboration utilizes ToxCast
AOP / MOA Targeted High-throughput testing AOP / MOA Targeted HTS Data Tox21 qHTS 10K Library • EPA • Pesticides actives and inerts • Industrial chemicals • Endocrine Disruptor Screening Program • OECD Molecular Screening Working Group • FDA Drug Induced Liver Injury Project • Failed Drugs • NTP • NTP-studied compounds • NTP nominations and related compounds • NICEATM/ICCVAM validation reference compounds for regulatory tests • External collaborators (e.g., Silent Spring Institute, U.S. Army Public Health Command) • Formulated mixtures NCGC • Drugs • Drug-like compounds • Active pharmaceutical ingredients
LTS MTS HTS uHTS Gene-expression High-Throughput Screening Assays batch testing of chemicals for pharmacological/toxicological endpoints using automated liquid handling, detectors, and data acquisition 1000s/day 10s-100s/yr 10,000s-100,000s/day 10s-100s/day Human Relevance/ Cost/Complexity Throughput/ Simplicity
96-, 384-, 1536 Well Plates Pathway Chemical Exposure Cell Population Assay Target Biology (e.g., Estrogen Receptor) High Throughput Screening 101 HTS Robotic Platform HTS: High Throughput Screening
Protein super-families GPCR Kinase Phosphatase Protease Ion channel Nuclear receptor Other enzyme CYP P450 inhibition Various formats: Radioligand receptor binding Fluorescent receptor binding Fluorescent enzyme substrate-intensity quench Fluorescent enzyme substrate-mobility shift Initial screening: 25 mM in duplicate 10 mM in duplicate (CYPs) Normalize data to assay window % of control activity (central reference – scalar reference) Biochemical Assays
What do biochemical assays measure? • Mainly direct effects of chemical on target protein • Enzyme activity • Ligand binding • False positives: • Fluorescent compounds—fluorescing and quenching • Reactive compounds/covalent modification of target • Physical effects—colloid aggregation of target • Operational • False negatives: • Solubility • Inappropriate assay conditions • Operational • Target protein not physiological • Lack of biotransformation
Biochemical Concentration-Response Testing • Retest actives: • Median absolute deviation (MAD) median Ιx-xmedΙ two MADs or 30% activity • 8 conc/3-fold serial dilutions • 50 mM high conc • 25 mM high conc for CYPs • Normalize to assay window • Fit % Activity data to 3- or 4-parameter Hill function • Sometimes had to fix top or bottom of curve • Did not extrapolate beyond testing range • Manual or automated removal of obvious outliers
Example Curve Fits rAdrRa2B hCYP 2C9 hERa hLynA Activator hM1 hKATPase
Real Time Cell Growth Kinetics • Cytotoxicity with potential mechanistic interpretation • Human A549 lung carcinoma cell line • ACEA experience with line • Reference compound effects • Concentration-response testing • 8 conc/3-fold serial dilutions • Duplicate wells • Real-time measuremens during exposure (0-72 hr) • IC50 and LELs calculated
Data examples Replicate Analysis: Example Plots:
Multiplexed Transcription Factor Assays • Modulation of TF activity in human hepatoma HepG2 cells • Multiplexed reporter gene assay • cis 52 assays (response element driving reporter) • trans 29 assays (GAL4-NR_LBD driving reporter) “ligand detection” • IC50 for cytotoxicity measured first in HepG2 • High concentration either 100 mM or 1/3 calculated IC50 for cytotoxicity • Seven concentrations, 3-fold serial dilutions, 24 hr exposure • Cells harvested, RNA isolated, processed for reporter gene quantitation • LEL provided in data set
Cis: AhR Multiplexed Reporter Gene Technology
Corresponding cis and trans assays Bisphenol A HPTE trans: ERa cis: ERE
Assays Profile Database Informatics LPS Human primary cells Disease-like culture conditions Biological responses to drugs and stored in the database Specialized informatics tools are used to mine and analyze biological data SM3C BF4T BioSeek: BioMAP® Technology Platform Primary Human Cell-Based Assay Platform for Human Pharmacology
Stress Pathway Activation Organelle Functions DNA Damage Oxidative Stress Cell Cycle CSK Integrity High-Content Screening of Cellular Phenotypic Toxicity Parameters • Technology: automated fluorescent microscopy • Objective: Determine effects of chemicals on toxicity biomarkers in a cell culture of HepG2 and primary rat hepatocytes • Panel 1 design*: • Multiple mechanisms of toxicity • Acute, early & chronic exposure • 384-well capacity • HepG2
Data Examples Cell Loss Mitochondrial Membrane Potential DNA Damage
XME Gene Expression in Primary Human Hepatocytes • Primary human hepatocytes from two donors used • Cells exposed for 6, 24, and 48 hr; medium/chemical refreshed daily • Concentrations tested: 40, 4, 0.4, 0.04, and 0.004 µM • 16 Genes measured in multiplexed RNAse protection assay (qNPA) • Genes targeted XME and transporters
Data Examples CYP1A1-AhR HMGCS2-PPARα CYP2B6-CAR
NCGC Reporter Gene Assays • Nuclear Receptors • GAL4 System (ligand detection assay) • 11 human receptors • 1 rat (PXR) • b-lactamase reporter gene assays except: • PXR assays are luciferase reporter gene assays • p53 Reporter Gene assay • b-lactamase reporter gene assay • Parental cell lines mostly HEK293 (also HeLa and DPX-2) • 12-15 point concentration-response curves (single replicate)
ERa NCGC: Data Calculations • Data normalized to reference compound effect • Curves fit to 3- or 4-parameter Hill equation • Artifacts removed where obvious fluorescence or cytotoxity detected • Required at least 25% efficacy of control compound to calculate AC50 • AC50 values provided • Antagonist format assays challenging due to effects of cytotoxicity • LXR assay problematic—contaminated with GR reporter line? PPARg
Published Predictive Toxicity Models • Predictive models: endpoints liver tumors: Judson et al. 2010, EnvHlthPersp 118: 485-492 hepatocarcinogenesis: Shah et al. 2011, PLoS One 6(2): e14584 cancer: Kleinstreuer et al. 2012, submitted rat fertility: Martin et al. 2011, BiolReprod 85: 327-339 rat-rabbit prenatal devtox: Sipes et al. 2011, Toxicol Sci 124: 109-127 zebrafish vs ToxRefDB: Sipes et al. 2011, Birth Defects Res C 93: 256-267 • Predictive models: pathways endocrine disruption: Reif et al. 2010, EnvHlthPersp 118: 1714-1720 microdosimetry: Wambaugh and Shah 2010, PLoS Comp Biol 6: e1000756 mESC differentiation: Chandler et al. 2011, PLoS One 6(6): e18540 HTP risk assessment: Judson et al. 2011, Chem Res Toxicol 24: 451-462 angiogenesis:Kleinstreuer et al. 2011, EnvHlthPersp 119: 1596-1603 • Continuing To Expand & Validate Prediction Models • Generally moving towards more mechanistic/AOP-based models
Reproductive Rat Toxicity Model Features Martin et al 2011
Reproductive Rat Toxicity Model Features + 36 Assays Across 8 Features Balanced Accuracy Training: 77% Test: 74% - Martin et al 2011
Example: Cancer SignaturesNon-genotoxic carcinogens • Use insights from Hallmarks of Cancer • Hanahan and Weinberg 2000, 2011 • Cancer is a multi-step progressive disease • Virtually all cancers display all hallmark processes • We observe that most chemicals perturb multiple pathways • Hypothesis: • A chemical that perturbs many pathways related to cancer hallmark processes will be more likely to cause cancer in the lifetime of an animal than a chemical that perturbs few such pathways • Chemicals can increase cancer risk through many different patterns of pathway perturbations
Hallmarks of Cancer Hanahan and Weinberg (2000) PPARa p53 ICAM1 CCL2
Hallmarks of Cancer Hanahan and Weinberg (2011) IL-1a IL-8 CXCL10
Hallmark-related ADME-related Endpoint Pathway Hits Raise Risk of Multiple Cancer Types Level 2: Preneoplastic Level 3: Neoplastic
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) Systems Models
Beyond in vitro to in vivo signatures In vitro Assays Structure Clusters Chemical Categories Pharmacokinetics Adverse Outcome In Vitro-In Vivo Signatures
Combining Chemical Structure and In Vitro Assays • Structure clustering based on chemical fragments • FP3, FP4, MACCS, PADEL, PubChem (~2700 total) • Hierarchical clustering and then set variable cutoffs • For examples: ~12 chemicals / cluster • Goals • Find clusters that are highly predictive of each assay (read-across) • Assay structure alerts: alternatives assessments • Assay QC Cluster Assay Endpoint
Data Set Incomplete Clusters 80% predictive of assay hit Chemical Set 1 Chemical Set 2 ER Assays Assays Estrogens Steroids Endosulfans CYP Binding Assays Inflammation Assays Azoles Alkyl Phenols Conazoles Tetracycline … Surfactants Alachlor … Surfactants Captan … GPCR Binding Assays
Adding Pharmacokinetics Reverse ToxicoKinetics (rTK) Human Hepatocytes (10 donor pool) Add Chemical (1 and 10 mM) Remove Aliquots at 15, 30, 60, 120 min Analytical Chemistry Hepatic Clearance Plasma Protein Binding Human Plasma (6 donor pool) Add Chemical (1 and 10 mM) Analytical Chemistry Equilibrium Dialysis Combine experimental data with PK Model to estimate dose-to-concentration scaling Collaboration with Thomas et al.., Hamner Institutes Publications: Rotroff et al, ToxSci 2010, Wetmore et al, ToxSci 2012
Combining in vitro activity and dosimetry Range of in vitro AC50 values converted to humanin vivo daily dose Triclosan Pyrithiobac-sodium log (mg/kg/day) margin Actual Exposure (est. max.) Rotroff, et al. Tox.Sci 2010 Wetmore et al Tox Sci 2012
Application: Endocrine Disruption • Prioritization • Screening thousands of chemicals • Developing activity thresholds of concern • Dose-relevance • Combining in vitro data with PK modeling • Refining activity thresholds of concern • Investigating the broader range of phenotypes of concern • Use many available in vitro tests and computer models as complement to EDSP animal tests
Initial Prioritization Application: EDSP21 Use high-throughput in vitro assaysand modeling tools to prioritize chemicals for EDSP Tier 1 screening assays
ER / AR Focus: EDSP21 • Endocrine Disruptor Screening Program • FQPA, SDWA 1996 contain provisions for screening for chemicals and pesticides for possible endocrine effects • Test pathways: estrogen, androgen, thyroid, steroidogenesis (EATS) • Universe of chemicals: 5000-6000 • Tier 1 screening battery (T1S): 11 in vitro &in vivo assays • Development and validation > 10 years • >$1 M per chemical • Current throughput < 100 chemicals / year • EDSP21 goal: • Prioritize chemicals for T1S • Hypothesis: EATS (in vitro)+ more likely to be T1S+ • Use many EATS in vitro assays • Combine with modeling, use, occurrence and exposure information
Characterizing chemicals for estrogen signaling pathway activity • Active vs. inactive • Potency and efficacy spectrum across assays • Agonist … Antagonist • Partial … full Agonist / Antagonist • ERa vs. ERb • Metabolically activated or deactivated • Cell type specificity • ER-mediated or not All Data is preliminary and unpublished