Reverse Dosimetry From Human Biomonitoring and In Vitro Concentration- Response Data Lancaster, 11 th July 2014 George Loizou
The Current Environmental Risk Assessment Paradigm Humans are exposed to thousands of chemicals in the environment Relatively few have been adequately evaluated It s a human toxicity data poor arena Reliant on animal testing
What is Required? A wish list Increased precision of risk estimates Understand and take account of uncertainty and inter individual variability Effective interspecies, high to low dose and route-to-route extrapolation Effective in vitro to in vivo extrapolation Decreased reliance on animal data Societal and ethical objections
What Data are Available? Limited human biological monitoring data Blood, breath, saliva, urinary biomarkers Limited in vivo animal data 28 day, 90 day, lifetime studies (mainly rodent) Increasingly in vitro None yet accepted for acute or chronic systemic toxicity
Forward and Reverse Dosimetry Interpretation of the health implications of biological monitoring data requires one of two approaches Forward dosimetry: estimation of internal exposures in studies characterising toxicity of chemical Reverse dosimetry or dose reconstruction: estimation of environmental exposures that would be consistent with measured biological monitoring data
An Ill-posed Problem BM Data No Model Empirical Model Lung Possible Solutions Fat SPT Liver Physiologically-Based Pharmacokinetic Model
An Ill-posed Problem In vitro Data No Model Empirical Model Lung Possible Solutions Fat SPT Liver Physiologically-Based Pharmacokinetic Model
Exposure reconstruction: An ill posed problem Inverting empirical models leads to: a myriad of answers to a single question unstable outcomes (small change in data may lead to large change in output) Linking internal and external exposure requires an appropriate model
What is a PBPK Model? A biologically realistic, simplified model
What is a PBPK Model? A mathematical description of a biological system translated into computer code and solved computationally An independent structural model, comprising the tissues and organs of the body with each perfused by and connected via the vascular system. Chemical-specific experimental data are superimposed over model predictions
C I Q P C X Q C Q C Lung C A C VL Q L Liver C VF Q F Fat C VR Q R Rapidly Perfused (brain, kidneys, etc.) C VS Q S Slowly Perfused (muscle, bone, etc.)
Model Parameter Requirements iv dose Plasma Kidney Rapidly Perfused Skin Slowly Perfused QC QKid QRap QSkn QSlw Model structure anatomy metabolism / transport processes Model parameters physiological data (organ weights, blood flows) biochemical data (partition coefficients, metabolism) Model equations system of mass-balance differential equations one equation for each tissue connected by equation for blood E.g., metabolizing tissue (liver): Liver VMax, KM, KMI QLiv oral dose L L ( C C / P ) V C / P /( K C P ) da / dt = Q max + / A L L L L M L L
Typical PBPK Parameter Values PARAMETER SCALING FUNCTION VALUE FLOW (l/min) Body weight (BW) - 75 (kg) Minute Volume 15 x(bw) 0.74-6.10 Cardiac Output 11.2 x(bw) 0.81-6.16 (%(BW)) (kg) Liver 2.7 2.03 1.48 Fat 19 14.3 0.31 Muscle 40 30.0 0.23 Kidney 0.44 0.33 1.14 Bra in 2.0 1.50 0.68 Richly Perfused 4.0 3.00 1.24 Slowly Perfused 6.75 5.10 0.57
Hidden Information Exhaled alveolar m-xylene concentration (µg/l) 50 40 30 20 10 Simulation is a composite of many processes 0 0 10 20 30 40 50 Time (hr)
Distribution to the Organs Concentration in organ Brain Kidney Muscle Fat Liver
Individual-based modelling Use measured anthropometric parameters e.g., body weight, alveolar ventilation rate, blood to air partition coefficient, body fat mass
Strengths of PBPK modelling Their mechanistic nature explains the very basis of observed data Tissue concentrations of drugs and chemical accurately predicted (tissue dosimetry). Tissue dosimetry described as linchpin of chemical risk assessment
Advantages of PBPK Modelling in Risk Assessment Estimation of tissue dosimetry is a better representation of biologically effective dose than exposure concentration or applied dose Captures the biological basis of a toxic response Description of a dose metric or dose which is causally related to toxic outcome These features provide a platform for a mode-ofaction based risk assessment
Incorporating Inter-individual Differences Need to generate different populations Neonates, pregnancy, healthy working, racial differences, some disease states e.g., diabetes Need to capture age-related changes in anatomy and physiology
PopGen Algorithm Image from Willmann 2007
PopGen Homepage
PopGen Inputs Page
PopGen Output
Incorporating Inter-individual Differences Body weight Brain mass and blood flow Liver mass and blood flow Kidney mass and blood flow Adipose mass and blood flow Cardiac Output Partition Coefficients Metabolic Rates
Population-based modelling Predict anthropometric parameters for a defined human population e.g., black African females with age range 16 to 56 years or a mixed cohort (males + females) elderly (65-80 years) of Caucasians, black Africans and Asians
Bayesian Inference and PBPK Models Bayesian Approach Determines reconstructed exposure as a probability distribution across the population, not just average or worst-case Merges existing knowledge with new knowledge i.e., PBPK model parameters and observations (BM data) are treated as random variables with distributions = Priors Prior distributions updated into a Posterior distributions using a likelihood model
Merging Data
Population Models BW Updated Distributions QP, QC Vmax,Km Lung Posterior Distribution of Exposure PCs Fat 30 SPT 20 10 New Data Liver 0 0 5 10 15 20 25 Time
Bayesian Inference
Advantages of PBPK Modelling PBPK models can translate: BM to target tissue dose metric (e.g., amount metabolized in the liver) to external exposure In vitro concentration response data translated to human exposure/dose PBPK models are appropriate because: All parameters have a physical or biological meaning Solutions are constrained within biologically plausible ranges They account for uncertainty and variability because parameters have biologically plausible central values and measures of dispersion They account for non-linear behaviour e.g., saturable metabolism Correct integration of exposure routes, pre-systemic clearance, flowlimited metabolism
Workflow Build Model Perform Morris Screening Test Perform Global Sensitivity Analysis on Selected Parameters Perform Reverse Dosimetry Ascribe Prior Distributions To Most Sensitive Parameters Posterior Distribution
Model Structure and Exposure Reconstruction 1. Slowly perfused 2. Rapidly perfused 3. Muscle 4. Skin 5. Adipose 6. Liver 7. Kidney 8. Brain 9. Lung 33 parameters 1. Slowly perfused 2. Rapidly perfused 3. Adipose 4. Liver 5. Kidney 6. Brain 7. Lung 27 parameters 1. Slowly perfused 2. Rapidly perfused 3. Adipose 4. Liver 18 parameters
Global Sensitivity Analysis of PBPK Models Quantify 1. Influence of individual parameters 2. Interactions between parameters 3. Interactions between parameters and non-linear processes Reduce dimensionality of models Reduce computational cost of Monte Carlo and Markov chain Monte Carlo analysis
List for Morris Test BW Body weight QCMC Cardiac output QPMC Alveolar ventilation rate Blood flow to slowly perfused QSPDC compartment Blood flow to rapidly perfused QRPDC compartment Morris screening test QFAC QLIC QKIC QBRC QLUC QMUSC QSKINC VSPDC Blood flow to adipose compartment Blood flow to liver Blood flow to kidney Blood flow to brain Bronchial blood flow to lung Blood flow to muscle Blood flow to skin Mass of the slowly perfused compartment 1.0 10-10 A 7.5 10-11 a VRPDC Mass of the rapidly perfused compartment σ 5.0 10-11 VFAC VLIC VKIC VBRC VLUC VMUSC VSKINC Mass of the adipose compartment Mass of the liver Mass of the kidney Mass of the brain Mass of the lung Mass of the muscle Mass of the skin 2.5 10-11 f c d b MPY PBA PSPDA PBRA Microsomal protein yield per gram liver Blood to air partition coefficient Slowly perfused compartment to air partition coefficient Brain to air partition coefficient 0 0 i h 1.0 10-6 g 2.0 10-6 e 3.0 10-6 4.0 10-6 µ 5.0 10-6 6.0 10-6 7.0 10-6 8.0 10-6 9.0 10-6 PLUA Lung to air partition coefficient PKIA Kidney to air partition coefficient PLIA Liver to air partition coefficient PRPDA PFAA PMUSA PSKINA VMAXC KM Rapidly perfused compartment to air partition coefficient Adipose tissue to air partition coefficient Muscle to air partition coefficient Skin to air partition coefficient Limiting rate of metabolism Michaelis Menten constant 35 parameters
The Lowry Plot MPY Microsomal protein yield per gram liver RURINE rate of urine production BW Body weight CREMMOL Urine creatinine KI Inhibition rate constant VFAC Mass of adipose tissue VMAXC Limiting rate of metabolism QFAC Blood flow to adipose tissue QPMC Alveolar ventilation rate PBAXYL Blood to air partition coefficient QSPDC Blood flow to slowly perfused tissue
Exposure/Dose Reconstruction Concentration-response relationships in Alternatives to animals Plausible in vivo dose responses t CSBP model Organ Venous effluent concentration Distribution of plausible reconstructed exposures Probabilistic PBPK model Human BM data
An Example with m-xylene Human volunteer study 8 white Caucasian, British subjects, 7 male, 1 female aged 25 54 years Exposed by inhalation to 40 ppm m-xylene for 4hours in controlled atmosphere exposure facility BM measurements Venous blood and exhaled m-xylene Urinary methyl-hippuric acid Anthropometric measurements Body mass, fat mass, ventilation rate, urine volume, creatinine concentration m-xylene Blood:air partition coefficient
Inhalation Exposure Reconstruction from venous blood m-xylene Lung Fat SPT Liver Measured Inhalation exposure: 4 h to 39 ± 3 ppm m-xylene Anthropometry from 3 volunteers Mean Median Std 36.53 ppm 36.18 ppm 3.69 ppm
Exposure Reconstruction from Urinary Biomarker: Simulating micturition
Exposure Reconstruction from Urinary Biomarker (3-methylhippuric acid) Inhalation exposure: 4 h to 39 ± 3 ppm m-xylene Urine sampling times: 4, 6, 8, 10, 12, 14, 24, 27, 31 h Anthropometry from 4 volunteers Predicted exposure (ppm) Mean 39.3 Median 39.1 2.5% 31.2 97.5% 48.7 McNally, K., et al. (2012) Journal of Toxicology 2012, 18
Dermal Exposure Reconstruction from venous blood 2-Butoxyethanol Lung Fat SPT Liver Measured dermal exposure: 2 h to 50 ± 3 ppm m-xylene Anthropometry from 4 volunteers Mean Median Std 55.13 ppm 53.92 ppm 9.64 ppm
Inhalation Exposure Reconstruction from venous blood m-xylene Exposure to binary mixture m-xylene + ethanol m-xylene Lung Fat SPT Liver ethanol Lung Fat SPT Liver Mechanism: Competitive inhibition Measured Inhalation exposure: 4 h to 45 ± 4 ppm m-xylene Anthropometry from 4 volunteers Ethanol (0.8g/kg) in fruit juice prior to entering controlled atmosphere exposure facility mean median sd 55.0 ppm 54.9 ppm 4.2 ppm
Alternatives To Animals Body Organs Cells Sub-cellular organelles Macromolecules In vitro data In silico data Proteins DNA
Interpreting In Vitro Toxicogenomics Data Known toxic agent NH O O N OH OH O HO P HO O O N N NH 2 N N Suspected toxic agent Gene-expression data Gene-expression data Toxicant signature Match Toxicant signature Toxicity prediction
In vitro Concentration-Response In vitro cell line concentration response data are surrogates for venous effluent concentration from in vivo organ or tissues Arterial input Venous effluent Liver
Point of departure: in vitro to in vivo extrapolation In Vitro PoD In Vivo PoD Lung Fat SPT Liver benchmark dose lower confidence limit benchmark dose benchmark dose lower confidence limit benchmark dose
Margin of Exposure Combine human exposure estimates with PoD to calculate margin of exposure (MoE) ExpoCast, US EPA initiative to develop tools for screening, evaluating and classifying thousands of chemicals based on relevant potential human exposure
Conclusions Venous blood biomarkers should give fairly robust posterior estimates of exposure Level of biology in PBPK model adequate Urinary biomarkers are: exquisitely sensitive to rate of urine production require PBPK models with more realistic description of the urinary elimination process Reverse dosimetry using PBPK modelling has great potential