On 10 May 2021, the OECD presented the recently published Guidance Document on the Characterisation, Validation and Reporting of Physiologically Based Kinetic (PBK) Models for Regulatory Purposes. This guidance aims to increase the confidence in the use of PBK models parameterised with data derived from in vitro and in silico methods, and help address “unfamiliar” uncertainties associated with these methods.
The webinar introduced the assessment framework for PBK models that was developed to evaluate the attributes and uncertainties of these models, including a dedicated discussion on sensitivity analysis. It also focused on the scientific workflow for characterising and validating PBK models together with a template for documenting PBK models in a systematic manner and a checklist to support model evaluation.
Check out the webinar video recording at: https://youtu.be/PT7w6PB97Ag and access the Guidance Document on the Characterisation, Validation and Reporting of Physiologically Based Kinetic (PBK) Models for Regulatory Purposes at: https://www.oecd.org/chemicalsafety/risk-assessment/guidance-document-on-the-characterisation-validation-and-reporting-of-physiologically-based-kinetic-models-for-regulatory-purposes.pdf.
Gaining acceptance in next generation PBK modelling approaches for regulatory approaches
1.
2.
3. • OECD Guidance Document on the Characterisation, Validation
and Reporting of Physiologically Based Kinetic (PBK) Models
for Regulatory Purposes published in February 2021 that aims to:
– increase the confidence in the use of PBK models parameterised with data
derived from in vitro and in silico methods, and
– help address “unfamiliar” uncertainties associated with these methods.
• Target audience
– the community of PBK model developers,
– the proponents of PBK models in a regulatory submission,
– the regulators who need to assess the applicability of the models in chemical
evaluations
Introduction to the Webinar
4. Topic Speaker
Welcome & Programme of webinar Magda Sachana (OECD)
Introduction to the PBK GD Cecilia Tan (US EPA)
Parametrisation of PBK models using in vitro and in
silico data
Iain Gardner (Certara)
Sensitivity Analysis –Theory and PBK applications Marina Evans (US EPA)
Sensitivity Analysis –example workflow George Loizou (HSE)
Decision tree for data poor chemicals – Read across
approach
Alicia Paini (EC, JRC)
Evaluation Framework Andrew Worth (EC, JRC)
Reporting template and evaluation checklist Cecilia Tan (US EPA)
Q&A 1 Andrew Worth (EC, JRC)
Case Studies examples Alicia Paini (EC, JRC)
Q&A 2 and wrap up Magda Sachana (OECD)
Programme
6. INTRODUCTION OF THE
OECD GUIDANCE ON
PHYSIOLOGICALLY BASED
KINETIC (PBK) MODELING
Cecilia Tan
U.S. Environmental Protection Agency, Office of Pesticide Programs, NC, U.S.
tan.cecilia@epa.gov
The opinions presented in this Technical Presentation are those of the author
and do not necessarily reflect the views or the policies of the U.S. Environmental
Protection Agency
7. • Purpose and scope of the OECD guidance document
• Introduction to PBK modelling
• Comparison with other PBK modelling guidance
documents
• Overview of the OECD guidance document
– Specific aims
– Contents
• PBK modelling workflow
Outline
9. Purpose and scope
• Provide guidance on characterising, reporting, and
evaluating PBK models used in regulatory assessment
of chemicals
• Address challenges associated with developing and
evaluating PBK models for chemicals without in vivo
kinetic data
• Promote the use of PBK models in regulatory risk
assessment and facilitate dialogue between model
developers and users
10. Scope
• The guidance provides contextual information of the scientific
process of PBK model characterisation and validation, but not
a technical guidance on model development or applications
• The level of confidence required for a PBK model is
dependent on the regulatory context of use
• The guidance is applicable to most chemicals and all species,
provided that appropriate methods/data exist to parameterize
a model
• The guidance is a living document that can be updated as
more experience is gained and new technologies, evidence
and applications emerge
12. PBK modelling approach
• A PBK model is a mathematical representation of
kinetic processes in the body, including
Absorption, Distribution, Metabolism, Excretion
• A PBK model predicts plasma/tissue
concentrations, given an external dose, based on
physiologic and anatomic characteristics, as well
as the physiochemical properties of a chemical
• “All models are wrong and some are useful”. – G
Box
13. Why PBK modelling?
• Predict internal exposure under new/inaccessible conditions
– “PBK models are intended to estimate target tissue dose in species and under
exposure conditions for which little or no data exist. If a complete data set
were available, then there would be no need to develop a model” – US EPA
(2006)
• Organise mechanistic data, present state of knowledge, identify
data gaps, suggest new experiments
– “… no model can be said to be ‘correct’. The role of any model is to provide a
framework for viewing known facts and to suggest experiments”. – S.
Moolgavkar
• Quantify uncertainty and variability in kinetics
• Relate bioactive in vitro concentrations to an equivalent
external dose
15. Comparison with other guidance
(characterising PBK models)
U.S. Environmental Protection Agency (2006)
European Food Safety Authority (2014)
WHO (2010)
18. Specific aims
1. A scientific workflow for characterising and validating
PBK models, with emphasis on models that are
constructed without using in vivo data
2. Knowledge sources on in vitro and in silico methods that
can be used to generate model parameters
3. An assessment framework for evaluating PBK models for
intended purposes
4. A template for documenting PBK models
5. A checklist to support the evaluation of PBK model
applicability according to context of use.
19. Contents
1. Introduction and scope
2. PBK modelling workflow
3. Regulatory assessment of PBK models
Annexes
– List of resources for PBK modelling
– Prospective use of microphysiological systems in PBK models
– Sensitivity analysis
– Case studies
23. PBK MODELS –
PARAMETERISATION OF PBK
MODELS USING IN VITRO
AND IN SILICO DATA
Iain Gardner
Certara, Sheffield, UK
iain.gardner@certara.com
24. • Focus mainly on PBK models parameterised with in vitro or in silico input data
– Little or no in vivo data for model verification
– Bottom up PBK model parameterisation rather than top down (fitting) approaches
• Provide a model assessment framework for facilitating dialogue between PBK model developers and
regulators
– “data poor” situations
– Uncertainties underlying the model input data, model structure and model predictions
• Provides guidance on characterisation and reporting of PBK models used in the regulatory
assessment of chemicals
– Template for model documentation
– Checklist to support the evaluation of PBK model applicability according to the context of use
• Considerations for using human in vitro test systems to characterise the
pharmacological/toxicological hazard
• Document is not a technical guidance on PBK model development or best practice
• This is covered elsewhere
PBK model parameterisation
25. Bottom-up PBK modelling workflow using in vitro
and/or in silico model inputs
Focus on
step 3
The chemical of interest and
mechanisms defined in the problem
formulation step will influence both
model structure and which in
silico/in vitro input data are
needed
26. • Depends on
– Problem formulation
– Underlying biokinetic mechanisms
– Accepted/known physiology in the species of interest
– Available data for model building
– Route of exposure (oral vs dermal vs inhalation etc)
• Outcome
– Model of appropriate complexity for question being addressed
• Number of compartments
• Perfusion vs permeability limitations
• Detail of each compartment
– Is metabolism considered in the tissue?
– Is active transport included?
• Parent/metabolite
• Need to account for/reduce mechanistic differences between the in vitro system and in vivo
Step 2 Model conceptualisation – structure and
mathematical representation
27. • Species specific
– Tissue volumes
– Blood flow
– Tissue connectivity
– IVIVE scaling factors
• Literature and online resources for physiological and
anatomical parameters are provided
Step 3 Model parameterisation – anatomical and
physiological parameters for the species of interest
28. • Physicochemical properties
– Lipophilicity, solubility
• Protein/blood binding
• Absorption
– Oral vs dermal vs inhalation routes have different considerations
– (especially for dermal exposure) formulation characteristics will impact absorption and should be reported.
• Tissue distribution
• Metabolic rates (Clearance)
– CLint , Km & Vmax
• Transport
– CLint , Km & Jmax
• Elimination
– Renal and biliary elimination
• For each PBK model parameter needed
– Background information and guidance is provided
Step 3 Model Parameters – Chemical specific
parameters
29. • Principles for generating, evaluating and reporting in vitro and in silico PBK model parameters are explained in
the guidance
– Not intended to be an exhaustive list of every model ever published/used
– Several documents provide guidelines on generating reproducible and reliable in vitro data (OECD TG 428; OECD TG 319a,b; Good in vitro method practices (GIVIMP), 2018)
– OECD has defined 5 principles for validating in silico QSAR models
• A defined endpoint
• An unambiguous algorithm
• Defined domain of applicability
• Appropriate measure of goodness of fit, robustness and predictivity
• (if possible) a mechanistic interpretation
– Pros and Cons (including discussion of applicability domains) for different methods stated
• Residual uncertainty using different approaches is noted where applicable
• In the PBK report details of how the in silico and in vitro data used in PBK model parameterisation were
calculated and measured should be provided.
• When using in vitro pharmacology/toxicology data in risk assessments there should be consideration of
– Any challenges related to measurement/stability of the chemical in the in vitro test system
– different approaches for biokinetic modelling to related nominal (applied) concentration to the free/intracellular concentration of the chemicals of interest
• Consideration of microphysiological systems is also discussed (Annex 2)
Step 3 Model Parameters – Chemical specific
parameters
30. • Useful for trouble shooting and for points to be considered in model review
For each parameter there are pointers for the
modeller and the assessor
31. • Many different packages available for PBK modelling
• Annex 1 contains a list of commonly used software
• Depending on model structure often solvers that can
handle stiff differential equations need to be used
– Numerical methods are well established and if used correctly
are not considered to represent a significant source of
uncertainty in the modelling process
– Not addressed further in the guidance
Step 4 Computer implementation (solving the equations)
32. • PBK guidance provides background information and resources to aid
with construction of PBK models where input parameters are
primarily from in vitro or in silico sources
• Reliability of different approaches for model parameterisation are
discussed
• Principles for in vitro and in silico parameter generation are
highlighted
• Literature and online resources are listed
Conclusions
34. SENSITIVITY ANALYSIS –
THEORY AND PBK
APPLICATIONS
Marina Villafañe Evans
US EPA/ ORD/ CCTE/RTP, NC USA
evans.marina@epa.gov
The opinions presented in this Technical Presentation are those of the author and
do not necessarily reflect the views or the policies of the U.S. Environmental
Protection Agency.
36. Normalised change in dose (measurement) for each
model parameter at a fixed point in time
KB1= Ah binding constant
BM1 =Ah availability
PL = liver PC
PF = fat PC
BM20 = CYP1A2 basal
BM2I = CYP1A2 induction
N = Hill coefficient
KD =complex binding constant
KB2 = CYP1A2 binding constant
38. • Optimisation –
– Take the difference between data and model simulation
– Make sure difference is as small as possible- Least Squares Sum (LSS)
How do we estimate unknowns?
39. SA and global optimization
Use a global optimisation algorithm to find true global minimum
40. PBPK rat optimisation surface with multiple peaks
and valleys. Red dot is Vmax and Km values
41. Normalised SC for different inhalation concentrations
in rat experiments for previous surface
42. e.g.1300 ppm for Vmax and 200 ppm for Km
Normalised SC for given exposure can predict
experimental concentrations
43. • Calculate Sensitivity Coefficients with Chain Rule and
gradients (slopes) for each parameter of interest
• Perform the calculations for each time point
• Largest normalised sensitivity coefficients suggest time
for estimable parameters
• Ranking of most important parameters in model
• Unique values will be obtained if SC do not cancel each
other
How is SA performed?
47. Sensitivity Analysis
Global Sensitivity Analysis most appropriate for PBK models which
describe non-linear biological processes e.g. enzyme saturation,
receptor and transporter binding.
Two steps:
1. Elementary Effects Screening (Morris Test)
2. Extended Fourier Amplitude Sensitivity Test (eFAST)
48. Elementary Effects Screening: Morris Test
• Initial screening of entire model input parameter set
• Ranks parameters to identify most significant contributors
to model output variance
• High µ* = important overall influence on model output
• High σ = high interaction with other parameters or has
non-linear effects
52. • Two types of sensitivity measure that vary with time:
– Main effects (Si)
– Total effects (ST).
• ST usually have a higher variance than main effects
• ST include any interactions (i) between any number of parameters
• It is useful to think of ST as the sum of Si and i
𝑆𝑇 = 𝑆𝑖 + 𝑖
eFAST Sensitivity Indices
56. DECISION TREE FOR DATA
POOR CHEMICAL – READ
ACROSS APPROACH
Alicia Paini
European Commission Joint Research Centre, Ispra, Italy
alicia.paini@ec.europa.eu
57. Step 5 - Assessment of model predictive
capacity by using a read-across approach
58. Background
• Chemical A data poor
• Chemical B, C, D, E data rich
• How to extrapolate data for chemical A using info from other chemicals?
Annex 1. List of resources for PBK modelling
Table 1A. Physicochemical properties
Table 1B. ADME parameterisation
tools/databases
Table 1C. Dedicated PBK modelling software
Table 1D. Mathematical modelling and
simulation tools that can assist PBK modelling
Disclaimer: The following tables are not necessarily exhaustive of all
available resources, and no endorsement is implied. Tables are based
on Madden et al. 2019 and 2020; but expanded to include also
environmental databases.
Filling biokinetic data gaps by read-across
Madden et al., 2019. Computational Toxicology; Madden et al., 2020. ATLA
59. Step 5
• Model Performance Schematic workflow to identify and use analogues in PBK
model development and validation
OECD Annex IV, CS using
Read across approach CS IV & X
68. REPORTING TEMPLATE &
EVALUATION CHECKLIST
Cecilia Tan
U.S. Environmental Protection Agency, Office of Pesticide Programs, NC, U.S.
tan.cecilia@epa.gov
The opinions presented in this Technical Presentation are those of the author
and do not necessarily reflect the views or the policies of the U.S. Environmental
Protection Agency
69. • Section 3.3 PBK Model Reporting Template
• Section 3.4 Checklist for Evaluation of Model
Applicability
Outline
71. PBK Model Reporting Template
• Harmonizing a reporting template can reduce the burden
of preparing different reports for different agencies for
the same analysis
• It can facilitate efficient review and timely decision-
making
• It can be adopted and customised to meet specific needs
• Other guidance: WHO (2010), FDA (2018), EMA (2018)
72. Template sections
• Name of model
• Model developer and contact details
• Summary of model characterisation, development, validation and
regulatory applicability
• Model characterisation
• Identification of uncertainties
• Model implementational details – software details, code, software
verification
• Peer engagement (input/review)
• Parameter tables
• References and background information
73. Model characterisation
1. Scope and purpose of the model
2. Model conceptualisation – model structure and
mathematical representation
3. Model parameterisation – parameter estimation and
analysis
4. Computer implementation
5. Model performance
76. Checklist for Evaluation of Model Applicability
• The evaluation checklist comprises a series of questions
for user to analyse the evidence provided by the modeller
• The guidance does not stipulate how the checklist should
be weighed, since it should be determined by the users
and is context-dependent
• The checklist has two sections:
A. Context/Implementation
B. Assessment of Model Validity
77. Context/Implementation
• A.1. Regulatory Purpose
– Acceptable degree of confidence for the envisaged application?
– Is the degree of confidence/uncertainty greater or less than non-modeling
option?
• A.2. Documentation
– Model documentation adequate?
• A.3. Software Implementation and Verification
– Model code express the mathematical model?
– Model code devoid errors? No numerical errors?
– Parameter units correct?
– Mass balance?
– ODE solver appropriate?
– PBK modelling platform verified?
78. Context/Implementation
• A.4. Peer engagement
– Has the model used previously for a regulatory purpose?
– Is additional review required?
79. Assessment of Model Validity
• B.1. Biological Basis
– Model consistent with known biology?
• B.2. Theoretical Basis of Model Equations
– Are the underlying equations based on established theories?
• B.3. Reliability of Input Parameters
– Has the uncertainty of parameters been characterized?
• B.4. Uncertainty and Sensitivity Analysis
– Has the impact of parameter uncertainty been estimated
– Confident in influential parameter values?
• B.5. Goodness-of-fit and Predictivity
83. Context of use of the PBK model in risk assessment
PBK models converting in vitro concentrations that
elicit cellular/sub-cellular responses (an in vitro
Point of Departure, PoD) to the corresponding in
vivo external doses (QIVIVE)
PBK models within IATA to make better use of
existing toxicity data and inform testing needs
PBK models developed for data poor chemicals
must rely on kinetic data generated by in vitro
and/or in silico methods
An example of PBK model use in assessment of
chemical can be found in table 1.1
84. Thirteen case studies
(listed in Annex 4)
https://www.oecd.org/chemicalsafe
ty/testing/series-testing-
assessment-publications-
number.htm
85. Case Study VIII
PBK model application
in species and route to
route extrapolation
Bessems et al., 2017
Case study IX
Caffeine PBBK model
to predict MoIE for
risk assessment
IATA caffeine CS
Caffeine Case Studies
86. Pictures source: Bessems et al., 2017
Human health risk of exposure to a chemical can be characterised.
• RCR = Risk characterisation ratio approach
human exposure level compared to human limit value (HLV), based on an animal point of departure (POD) divided by
relevant assessment factors.
• MOE = [POD]/SED
POD point of departure : i.e. NOAEL or BMDL
SED systemic exposure dose (SED)
A shortcoming in both
approaches is the apparent
lack of attention for species-
and route-dependent ADME
information.
• MOIE = [SPOD]/SED
• SPOD Systemic point of departure by PBK modelling
• SED Systemic exposure dose (SED) by PBK modelling
Caffeine in cosmetics, in food products.
• PBK model for rat (oral)
• Define the toxicity using fetotoxicity
• LOAEL 10 mg/kgbw/d
• NOAEL < 10 mg/kgmg/kgbw/d
• PBK model for human (dermal & oral)
• Calibration using oral human in vivo data
• Model validation using oral and dermal in vivo data
Margin of Internal Exposure (MOIE)
87. Picture source: Bessems et al., 2017
• Developed in R (http://cran.r-project.org)
• Model equation reported in the appendix
• Model assumptions reported
• Biological assumptions reported
• Literature data and in silico data as input
• Historical in vivo data to calibrate and validate the model
Reporting Template
88. • No model code available,
equation describing the model
available, code request to authors
• Assumptions and uncertainties
are reported
• Most of the input parameter
where taken from literature we
are not sure on the quality
• Local Sensitivity analysis and
Monte Carlo were performed
• Saturable kinetics (by means of
Vmax and Km) included
Checklist Highlights
89. Evaluation Matrix
Overall Evaluation Matrix
Picture source:
Icons from ppt
HIGH
NONE
Model
Structure simulations
of data; predictivity
Biological
basis
Model reproduces
consistently all kinetic
data, including the shape
of time course profiles
for chemical of interest.
The biological basis of
some model parameters,
structural elements or
assumptions is
questionable.
LEVEL OF CONFIDENCE
Variability/
Uncertainty in
Parameter Analysis;
Global Sensitivity
Analysis
Local Sensitivity
Analysis supports the
robustness of the model.
90. Evaluation Toolbox Conclusions
“The PBK model was developed to illustrate an approach to apply MoIE. The model is well
characterised however, the biological basis of some model parameters, model structural elements are
questionable. Assumptions are reported to underline the uncertainties; the Model reproduces
consistently all kinetic data, including the shape of time course profiles for chemical of interest, for
the calibrated person. A local SA with a Monte Carlo simulation were performed. Based on all the
evidence provided the following model will have a medium/standard level of confidence. The model
code is not reported and a full assessment cannot be performed.”
Is, thus, recommended that the PBK model and
relative output, can be used in a regulatory
framework only as supporting information.
91. Cosmetics Europe - OECD IATA CS - Case Study on the use of Integrated Approaches
for Testing and Assessment for Systemic Toxicity Arising from Cosmetic Exposure to
Caffeine
Picture source: Cosmetics Europe OECD IATA CS 2020 and EC 2016.
Integrated Approach to
Testing and Assessment
Extensive revision - simplification
recoded in Berkeley Madonna
New data to parametrize the model
92. Evaluation Toolbox Conclusions
“A PBK model with relatively simple oral and dermal absorption models was developed in order to
conduce cross species extrapolations and route to route extrapolation. The model is built using solid
knowledge on the chemical MoA, and is parametrise using good in vitro measured data well
established. The model reproduces the shape time and dose course of the chemical of interest. A
Local Sensitivity Analysis supports the robustness of the model.”
On the basis of the results available, it was concluded that, the
application of this PBK model for cross-species
extrapolation is a reasonable approach for a
preliminary hazard characterisation. A global SA
would have given a more global overview on the key
parameters that can perturb the model output.
93. Case Study VIII
PBK model application in
species and route to route
extrapolation
Bessems et al., 2017
Case study IX
Caffeine PBBK model to
predict MoIE for risk
assessment
IATA caffeine CS
• Built to answer a more
scientific question as
proof of principle
• Built to answer a risk
assessment question as
an integrative piece of
evidence
Call for more examples /to populate case studies /IATA
Alicia.paini@ec.europa.eu
Caffeine Case Study