SEAC
Abstract No. 2646 - Poster Board No. P144
SAFETY & ENVIRONMENTAL ASSURANCE CENTRE
Mechanistic model-based approach to Skin Sensitisation Risk Assessment
Gavin Maxwell, Richard Cubberley, Seraya Dhadra, Nichola Gellatly, John Paul Gosling*, Ruth Pendlington, Juliette Pickles, Joe Reynolds, Richard Stark, Dawei Tang and Cameron MacKay
SEAC, Unilever, Colworth Science Park, Sharnbrook, Bedford, MK44 1LQ, UK; *School of Mathematics, University of Leeds, Leeds, LS2 9JT, UK
Abstract
Despite our understanding of the key events that drive skin sensitisation our ability to
combine non-animal hazard data with exposure information to establish a safe level of
human exposure for a sensitising chemical remains a key gap.
Our aim is to apply mechanistic understanding of skin sensitisation to improve our ability
to make risk assessment decisions. Central to our approach is a toxicokinetictoxicodynamic (TKTD), ordinary differential equation (ODE) model covering several of
the key events captured in the skin sensitisation adverse outcome pathway (AOP) that
outputs naïve CD8+ T cell activation as a surrogate measure for sensitisation induction in
humans. Chemical-specific model parameters are derived from bespoke in vitro
experiments designed to measure reactivity rate and skin bioavailability, whilst
biological parameters are taken from the immunological literature.
The model has been used to simulate a study published previously by Friedmann et al. in
which 165 healthy volunteers were exposed to one of five doses of the contact allergen
2,4-dinitrochlorobenzene (DNCB). As a significant proportion of each dose cohort were
sensitised to DNCB within this study, comparison of model simulation results to these
clinical data have provided an opportunity to explore the relationship between naïve CD8+
T cell activation and clinical sensitisation. To do so, the model was parameterised for
DNCB and prediction made for extent of naïve CD8+ T cell activation occurring across
dose. Reverse dosimetry analysis was then performed to calculate the dose-threshold
for sensitisation to DNCB and the probability of the average individual acquiring
contact allergy at any given dose.
Importantly, uncertainty due to both parameter uncertainty (limitations in our
knowledge of parameter values) and model uncertainty (limitations due to validity of
modelling assumptions) has been characterised. The uncertainty in model predictions
for the DNCB case study is significant, therefore additional historical case studies are
underway to address this finding.
Lymph
Node
Epidermis
Epidermis
Induction
Elicitation
5. Posterior parameters by Bayesian inference
1. Model Scope
A simple compartmental model (Davies, 2011) was
applied to skin penetration data (OECD TG 428 modified
to include additional time points and free/bound
measurements - Pendlington, 2008, Reynolds, 2016) to
update prior estimates for diffusion and partitioning
rates using Bayesian analysis.
A similar approach was applied to protein reactivity data
to obtain estimates of DNCB reaction rate and protein
binding kinetics within human skin (data and data
analysis not shown).
Objective: TKTD model scope should be simplest
representation of the chemistry and biology capable of
reproducing the induction of contact allergy to DNCB to
enable prediction of a safe level of skin exposure
2. Model Schematic
PANEL A = TOXICOKINETIC MODEL (Reynolds, 2016): (1) diffusion
and partitioning into the stratum corneum and skin; (2) sensitiser
clearance by dermal capillaries; (3) covalent modification of
protein nucleophiles by hapten.
To predict the likelihood of an allergic immune response to DNCB the following
sequence of events were explicitly modelled: 1. haptenation of nucleophilic
amino acids in the skin; 2. protein degradation by proteasome; 3. presentation
of proteasome-derived peptides by Class I MHC; 4. recognition of peptide-MHC
by T cell receptor (TcR)
A threshold was defined (using literature data) to enable the model output
(average CD8+ TcR triggering rate over time) to be converted into a probability
that an individual would become allergic as a result of DNCB exposure. A
reverse dosimetry approach (Clewell, 2008) was used to back-calculate the
doses of DNCB that would cross the TcR signal threshold and cause allergy.
The model prediction was benchmarked against historical clinical data from
Friedmann and colleagues – ED50 (dose sensitising 50% cohort) for a single
exposure to DNCB, which was found to be 16.5 µg/cm2.
3. Model Assumptions
Clin. Exp. Immunol. (1983) 53, 709-715.
Quantitative relationships between sensitizing
dose of DNCB and reactivity in normal subjects
P. S. Friedmann, C. Moss, S. Schuster & J. M. Simpson
Table documenting model assumptions, evidence and uncertainty
4. Prior parameter uncertainty by Expert Knowledge elicitation
Expert knowledge elicitation is a process for characterising experts’
uncertainty when data/information is lacking (Cooke, 1991; O’Hagan, 2006;
Rowe and Wright, 1999). EKE was used to obtain prior probability
distributions for all TK/TD model parameters. Example of hapten-protein
degradation rate is shown to illustrate this three-step process: 1. experts
capture evidence related to the parameter of interest, including any
sources of uncertainty; 2. experts are asked to make judgements on
possible or potential parameter; 3. a probability distribution is used to
represent this uncertainty.
Clewell, H.J. et al. 2008. Quantitative interpretation of human biomonitoring data. Toxicol. Appl. Pharmacol. 231,
122–33.
OECD. 2012. The Adverse Outcome Pathway for Skin Sensitisation initiated by covalent binding to proteins. OECD Environ.
Heal. Saf. Publ. Ser. Test. Assess. 168. 1-46
Cooke, R.M. 1991. Experts in Uncertainty: Opinion and Subjective Probability in Science. Oxford University Press,
New York.
Oakley J.E. and O’Hagan A. 2004. Probabilistic sensitivity analysis of complex models: a Bayesian approach. J.R. Stat. Soc.
Ser. B. 66. 751-69
Davies M. et al. 2011. Determining epidermal disposition kinetics for use in an integrated non-animal approach
to skin sensitisation risk assessment. Toxicol. Sci. 119. 308-318.
O’Hagan A. et al. 2006. Uncertain Judgements: Eliciting Experts’ Probabilities. John Wiley & Sons, Chichester
Friedmann P.S. et al. 1983. Quantitative relationships between sensitizing dose of DNCB and reactivity in normal
subjects. Clin. Exp. Immunol. 53, 709-15
MacKay C. et al. 2016. Toxicokinetic/Toxicodynamic (TK/TD) modelling of skin sensitisation. Part II:
Toxicodynamics and hazard characterisation. Tox. Sci. Submitted
Pendlington R. U. et al. 2008. Development of a modified in vitro skin absorption method to study the epidermal/dermal
disposition of a contact allergen in human skin. Cutan. Ocul. Toxicol. 27. 283-94.
Reynolds J. et al. 2016. Toxicokinetic/Toxicodynamic (TK/TD) modelling of skin sensitisation. Part I: Toxicokinetics. Tox. Sci.
Submitted
Rowe G. and Wright G. 1999. The Delphi technique as a forecasting tool: issues and analysis. Int. J. Forecast. 15. 353-375.
Final elicited judgement
- 165 healthy human volunteers divided into five dose groups
- Single exposure to between 62.5 - 1000 µg DNCB applied to
7.1cm2 volar forearm in 100μL acetone vehicle
- Sensitisation assessed four weeks later by DNCB challenge
- Sensitisation varied between 8 - 100% across dose groups
7. Model evaluation and next steps
Sensitivity analysis (Oakley, 2004) was performed to determine the
parameters contributing the most uncertainty to simulated
sensitising dose.
Modified from OECD, ‘Adverse Outcome Pathway for Skin
Sensitisation initiated by covalent binding to proteins’ report
Preliminary data analysis
Posterior parameter estimates for diffusion and partitioning rates
6. Predict probability of skin sensitisation to DNCB
PANEL B = TOXICODYNAMIC MODEL (MacKay, 2016): (4)
proteasome processing of protein nucleophiles to form small
peptides and transport to the endoplasmic reticulum (ER); (5)
binding of peptides and hapten-peptide complexes to Class I MHC
and transport to plasma membrane; (6) binding of pMHC and
hapten-pMHC to CD8+ T cell receptors and (7) activation of naïve
specific CD8+ T cells.
Model is based upon the following major assumptions:
1. Extent of naïve CD8+ T-cell receptor (TcR) triggering is the key
determinant of human allergic status
2. Existence of at least one T-cell specific to the ‘antigen’
3. Required T-cell co-stimulatory signals are sufficient
4. Accompanying CD4+ T-cell response is optimal
5. DC migration from exposure site is sufficient
Uncertainty analysis was considered for all model assumptions (major
and minor), and evidence for and against each assumption was
systematically documented.
Prior Distribution
Posterior Samples
Fitted Distribution
Table documenting literature evidence and expert uncertainty
Acknowledgements
All uncertainty analysis was performed in collaboration with John
Paul Gosling (University of Leeds) who was funded by UK NC3Rs
Next Steps:
1. A framework for evaluating a TKTD model-based approach to skin
sensitisation risk assessment is currently under development.
2. Model development is ongoing to enable the T cell memory
response to be simulated in collaboration with University of
Leeds, University of Manchester, Salford Royal NHS Foundation
Trust & University of Southampton
SAFETY SCIENCE IN
THE 21ST CENTURY
For more information visit
www.tt21c.org
Sensitivity analysis identified the following model
parameters as most sensitive: ƙα* = TcR trigger
rate threshold; Ac = average area of contact
between DC and T cell; β = rate of haptenatedprotein turnover; mT = density of peptide-MHC in
DC and T cell contact zone.