US Army Corps of Engineers
BUILDING STRONG®
Ed Perkins
Senior Scientist
US Army Engineering
Research and Development
Center
Vicksburg, MS
Development and Use of Quantitive
AOPs to Support Decision Making
Reducing complexity to enable decisions
Reducing complexity to enable decisions
MIE KE 1 KE 2 KE 3 KE 4 AO
Receptor Molecular level
Cellular level
Organ level
Tissue level
Individual level
MIE KE 1 KE 2 KE 3 KE 4 AO 3 2 1 3 2
Indirect KER
3
KER1 KER2 KER3 KER4 KER5
AOP
AOP with
weighted KER
MIE KE 1 KE 2 KE 3 KE 4 AO 2 2 3 1
3
2
Evidence for
chemical X
causing or
initiating AOP
BRIEF BACKGROUND OF AOP FRAMEWORK Builds and expands upon prior pathway-based approaches to
predictive toxicology
Motivated, in part, by NAS (2007) “Toxicity Testing in the 21st
Century” emphasis on greater use of mechanistic data
Framework focuses on translation of mechanistic data into
endpoints meaningful to risk assessment
Emphasis initially on ecotoxicology, but equally applicable to
human health
*Ankley et al. 2010. Adverse outcome pathways: A framework to support
ecotoxicology research and risk assessment. Environ. Toxicol. Chem. 29: 730-
741.
MIE KE 1 KE 2 KE 3 KE 4 AO
Receptor Molecular level
Cellular level
Organ level
Tissue level
Individual level
MIE KE 1 KE 2 KE 3 KE 4 AO 3 2 1 3 2
Indirect KER
3
KER1 KER2 KER3 KER4 KER5
AOP
AOP with
weighted KER
MIE KE 1 KE 2 KE 3 KE 4 AO 2 2 3 1
3
2
Evidence for
chemical X
causing or
initiating AOP
STANDARDIZED, REVIEWED AND OECD
PUBLISHED INDIVIDUAL AOPS
Describes how measureable events lead to a regulatory outcome
•2012 launch of OECD AOP development programme•2013 OECD Guidance on Developing and Assessing AOPs
•Introduce standardization and rigor to AOP development•Conventions, terminology and information content•Weight of evidence evaluation based on modified Bradford-Hill criteria•Support development of AOP Knowledgebase
Adopted by EPA and promoted n EU
AOP Development and Description
Case Studies
Users’ handbook supplement to OECD guidance document for developing and assessing AOPs.
•March 2014 – Advancing AOPs for Integrated Toxicology and Regulatory Applications Workshop
Regulatory Impact of AOP Concept
AOP KB AS A CATALOG OF TOXICOLOGICAL
PATHWAYS AND ADVERSE OUTCOMES
Adverse Outcome Pathway Knowledge Base (AOP-KB)
APPLICATION OF TOXICOLOGICAL
MODELS IN CHEMICAL RISK ASSESSMENT
Robust predictive models offer cost-effective option to assessing
chemical risk when empirical data are limited
Occasionally utilized for screening-level assessments (e.g.,
narcosis QSARs), but rarely applied to regulatory decision-making
When higher-level predictive modeling has been employed
involves “data rich” high-visibility chemicals of human health
concern
Lack of use involves multiple factors
Availability of “generic” (as opposed to chemical-specific) BBDR
models
Accessibility of focused biological knowledge (and data) for the
array of endpoints/species of potential concern
Difficulty in formulating transparent, causal models
ROLE OF AOPS IN DEVELOPMENT OF
PREDICTIVE TOXICOLOGICAL MODELS
Provide transparent conceptual framework upon which to base
models
Options for many taxa/pathways/endpoints available through
resources such as the AOP-wiki
Clearly defined, measurable nodes (MIE, KE) directly indicative
of perturbation of pathways of concern
KERs provide basis for quantitative response-response relationship
model selection/parameterization
KERs and overall AOPs subjected to WoE based evaluation of
plausibility/causality
Moving from an AOP to qAOP
Descriptive
A qAOP captures response-response relationships
Hazard Assessment
Risk Assessment
Hazard assessment vs. Risk Assessment
AOPs are conceptual models for qAOPs
KE
1
KE
2
KE
3
KE
4
KR
5
A
O
KE
1
KE
2
KE
3
KE
4
KR
5
A
O
KE
1
KE
2
KE
3
KE
4
KR
5
A
O
EC
a
EC
b
EC
c
EC
d
EC
e
AOP
May not model all details of the AOP
Or they could have
more detail
qAOP
Must incorporate the AOP, but …
Description of Key Event RelationshipRelevant Models
& AnalysesTypical Data Needs
AOP
Directed: 𝐾𝐸𝐴 → 𝐾𝐵e.g. all KERs in AOP represent a causal linkage
Network/graph
analyses
techniques
Graph structure
providing the
connectivity between
KEs
Directed and signed relationship: 𝐾𝐸𝐴 →±𝐾𝐸𝐵
e.g. increasing and decreasing, i.e. ↑ 𝐾𝐸𝐴 ⇒↓ 𝐾𝐸𝐵
[a]
Direction and scalar-weighted relationship:
𝐾𝐸𝐴 →±𝑤𝐴,𝐵
𝐾𝐸𝐵e.g. simple weights
[a] Expert judged weights
qAOP
Direction and functional relationship: 𝐾𝐸𝐴 →𝑓𝐾𝐸𝐵
Probabilistic, e.g. probability of KE activation Bayesian
Networks
Expert judged
probabilities e.g. P(KE1
active given KE2 active)
Experimental data on
the KEs (e.g.
active/inactive)
Non-linear, e.g. saturable response Regression
modelling
Experimental data on
KEs under different levels
of perturbation
Time-dependent, e.g. acute vs chronic KE
activation
ODE, IBMs,
ABMs,
Leslie projection
matrix
Independent parameter
measurement
temporal data
Steatosis qAOP Bayesian Network
Rapid development of qAOP for screening
NFE
2/Nrf
2
LRH-1FX
RSHP
PPAR
-alpha
LX
RFAS PPAR-
gamma
HSD17b
4FA beta
oxidationCytosolic
FA
Stea
-
tosis
Lipo-
genesi
s
mTORC-
1LFAB-
P
AK
TaPKC
PI3K
SREBP
-1SCD
1
mTORC-
2
Insulin
receptor
+ -
+ |95 5
- | 5 95
+ -
+ | 5 95
- | 95 5
+ -
+ |95 5
- | 5 95
+ -
+ |95 5
- | 5 95
+ -
+ |95 5
- | 5 95
+ -
+ |95 5
- | 5 95
+ -
+ | 5 95
- | 95 5
+ -
+ |99 1
- | 1 99
Lipo & LFAB-P
++ -+ +- --
+ |99 99 99 1
- | 1 1 1 99
FXR & SHP & LXR
+++ -++ +-+ --+ ++- -+- +-- ---
+ | 50 50 50 1 99 50 50 1
- | 50 50 50 99 1 50 50 99
CytoFA & Fab-ox.
++ -+ +- --
+ | 1 1 99 99
- | 99 99 1 1
AKT + PI3K
++ +- -+ --
+ |95 5 50 5
- | 5 5 50 95
+ -
+ |99 1
- | 1 99
+ -
+ |100 0
- | 0 100
LRH-1 & LXR & PPAR-g
+++ -++ +-+ --+ ++- -+- +-- ---
FAS + | 95 75 75 50 75 50 50 1
FAS - | 5 25 25 50 25 50 50 99
mTORC1 & aPKC
++ -+ +- --
+ |95 5 95 5
- | 5 95 5 95
+ -
+ |95 5
- | 5 95
+ -
+ |95 5
- | 5 95
Binary State Bayesian
Network qAOP model
Predicting effect of assay measurements of events in an AOP network
Steatosis causal AOP network
Inhibition
here
Steatosis causal AOP network
Inhibition
here Causes
Fadrozole Inhibition Agonism
Estrogen
ReceptorReduced Vtg
production
Hepatocyte
Impair oocyte
development
Ovary
Impairovulation
& spawning
Female
Declining
trajectory
PopulationAromatase
EnzymeReduced
E2 synthesis
Granulosa
CellAromatase
inhibitor
Inhibition of aromatase by fadrozole well
characterized from molecular to individual level
From D. Villeneuve
Mechanistic qAOP development
Fadrozole Inhibition Agonism
EstrogenReceptor
Reduced Vtgproduction
Hepatocyt
e
Impair oocytedevelopment
Ovary
Impairovulation
& spawning
Female
Decliningtrajectory
PopulationAromataseEnzyme
ReducedE2 synthesis
GranulosaCell
Aromataseinhibitor
Fold
ch
an
ge
re
lative
to
co
ntr
ol (l
og
2)
Day
^ ^
^ ^
#
#
#
# #
# #
* *
*
* *
$ $
$ $
$
$
$
$
3 ug/L Fadrozole
30 ug/L Fadrozole
Time series Microarray, PCR, plasma hormone data, plasma protein data
Putative compensatory mechanism
Time
series
Bayesian
Network
analysis
Data driven understanding of Key events in an AOP
Fadrozole Inhibition Agonism
EstrogenReceptor
Reduced Vtgproduction
Hepatocyt
e
Impair oocytedevelopment
Ovary
Impairovulation
& spawning
Female
Decliningtrajectory
PopulationAromataseEnzyme
ReducedE2 synthesis
GranulosaCell
Aromataseinhibitor
Fold
ch
an
ge
re
lative
to
co
ntr
ol (l
og
2)
Day
^ ^
^ ^
#
#
#
# #
# #
* *
*
* *
$ $
$ $
$
$
$
$
Expert guided modelingData driven modeling
Quantitative Prediction of Reproductive/Population Effectsin Fish: Linking Relevant Models Across an AOP
Animal levelOrgan/tissue level
Vtg productionOocyte development,
ovulation and spawningAromatase
inhibition
Population
declining trajectory
Population level
VTG/fecundity correlation
Fadrozole Inhibition Agonism
Estrogen
ReceptorReduced Vtg
production
Hepatocyte
Impair oocyte
development
Ovary
Impairovulation
& spawning
Female
Declining
trajectory
PopulationAromatase
EnzymeReduced
E2 synthesis
Granulosa
CellAromatase
inhibitor
HPG axis model
Molecular level
Oocyte Growth Dynamics
modelPopulation Dynamics
model
Conolly et al. 2017. Quantitative adverse outcome pathways and their application to predictive toxicology.
Fadrozole 21 day, continuous exposure study. HPG axis model (9) predictions
after 21 days of continuous exposure to fadrozole at concentrations of 0, 1.4, 7.3,
and 57 g/L .
plasma E2plasma VTG
fecundity vs FAD FAD effect on population size
Conolly et al Environ Sci Technol 2017
Read-across of response-response plots to illustrate rapid evaluation of the qAOP-
predicted effects of aromatase inhibition on key events and adverse outcome
Aromatase inhibition vs FAD E2 vs Aromatase inhibition
VTG vs E2 Fecundity vs VTG
Pop size vs Fecundity
50% inhibition of
aromatase
2.3 µM
Conolly et al Environ Sci Technol 2017
Conclusion/Recommendations
AOPs provide a standardized approach relevant to regulatory needs.
AOPs are conceptual models that inform design of qAOP models
Define a specific question before developing a qAOP
The complexity and type of qAOP models is driven by resources and
uncertainty:
Time constrained, Data poor
Extensive response-response data
Biological fidelity
Application
Many different qAOP models can be made
Toxicokinetics+qAOP -> internal exposure at MIE+Hazard
Exposure+Toxicokinetics+qAOP-> external exposure+Hazard
qAOPs can be used to to answer a wide range of questions, but developers
should be transparent and document everything
ACKNOWLEDGEMENTS
USACE-ERDC
Lyle Burgoon, Lynn Escalon, Michael
Mayo, Natalia Garcia-Reyero
FUNDING
US EPA
G. Ankley, M. Breen, J. Cavallin, W.Y. Cheng,
R.Conolly, E. Durhan, K. Jensen, M. Kahl, C.
LaLone, E. Makynen, D. Miller, M. Severson,
K. Stevens, D. Villeneuve