Adverse Outcome Pathways (AOPs) for Target Organ Effects: The Role of Structural Alerts and Chemotypes for Liver Toxicity to Group Compounds and Apply Read-Across

Mark Cronin, Liverpool John Moores University, England

Elena Fioravanzo, S-IN Soluzioni Informatiche srl, Italy

Alexandre Péry, INERIS, France

Lothar Terfloth, Molecular Networks GmbH, Germany

Ivanka Tsakovska, CBME-BAS, Sofia,Bulgaria

Andrew Worth, European Commission JRC, Ispra, Italy

Chiahe Yang, Altamira LLC, Columbus OH

The Allure of the Adverse Outcome Pathway (AOP):

A Framework from Interaction to Mechanism to Adverse Effect

Molecular Interaction

Cellular Effects

Organ Effects

Organism Effects

Taken from Vinken M

(2013) Toxicology

312: 158-165

An Adverse Outcome Pathway (AOP) for Cholestasis:

Putting the Pieces Together

Molecular Interaction

Cellular Effects Organ Effects

Organism Effects

Taken from Vinken M

(2013) Toxicology

312: 158-165

Molecular interactions are

described by chemistry

An Adverse Outcome Pathway (AOP) for Cholestasis:

Putting the Pieces Together

Molecular Interaction

Cellular Effects Organ Effects

Organism Effects

BSEP Inhibition

Taken from Vinken M

(2013) Toxicology

312: 158-165

Molecular interactions are

described by chemistry

and linked to the

adverse event

An Adverse Outcome Pathway (AOP) for Cholestasis:

Putting the Pieces Together

Molecular Interaction

Cellular Effects Organ Effects

Organism Effects

BSEP Inhibition

Cholestasis

A Muddle of Acronyms

IATA

AOP

MoA

(Q)SAR

MIE

ITS KER

AOP-wiki

AOP-KB

ECHA

EPA

OECD

Tox21

SEURAT

Making Sense of the Acronyms

IATA

AOP

MoA

(Q)SAR MIE

ITS KER

AOP-wiki

AOP-KB

ECHA EPA OECD

Tox21 SEURAT

Everything You Wanted to Know About

AOPs but Were Too Embarrassed to Ask

• OECD Guidance and templates

• AOPs 101: The How and Why of Development and Use

– 5.00 – 7.00 pm, 25 March

– Regency Ballroom A, Hyatt Regency

– http://www.ascctox.org

• Development of a Knowledge Base for Quantitative Modelling of Adverse Outcome Pathways: Stakeholder Input Sessions

– 25, 26, 27 March

The Molecular Initiating Event (MIE)

• The required first “Key Event”

• Anchors the AOP

• MIE provides the basis of structural alerts and categories

• Alerts based on the MIE have a direct mechanistic linkage and justification

Types of Initiating Events:

Relevance to Chemistry

• Basal cytotoxicity (baseline narcosis)

• Receptor binding

• (Electrophilic) covalent reactivity

• Free radical formation

• Others: physical (abrasion, corrosion)

Why Organ Level Toxicity?

• The paradigm of one-to-one replacement of in vivo chronic testing is flawed

• Safety is determined from the no, or lowest, observed level

• N / L OELs are driven by toxicity to organs or physiological systems

• Replacement must be at that level

– AOPs provide the framework for computational and in vitro modelling

Evaluating Liver Toxicity

Granuloma

Cholestasis

Steatosis

Hepatitis Necrosis

Vascular Lesions

Neoplasm

Others

Why Computational Prediction?

• … when testing is not possible…

• Product design

• Green chemistry

• … part of a “testing strategy”

Assessing Cosmetic Ingredients

• COSMOS cosmetics inventory contains over 19,000 cosmetics-related substances

– Only a few have reliable toxicity data

• Do we need to test (with alternatives) all of these?

• How will we assess new ingredients?

• Computational predictions are a major tool in the alternatives toolbox

• Grouping and read-across provides a rational possibility to evaluation of chemicals

Using the MIE to Link Chemistry to Adverse

Outcome: A New Paradigm in Toxicity Prediction

Phenotypic

Adverse Outcome

In Vitro

Using the MIE to Link Chemistry to Adverse

Outcome: A New Paradigm in Toxicity Prediction

Describing the MIE

Development of Alerts

Review and Update

Molecular Initiating

Event

Phenotypic

Adverse Outcome

In Vitro

951 Liver Toxicants / Non-Toxicants

Profiled On Mechanism of Action

• Approx. 34% of liver toxicants were protein reactive (computational prediction)

• Approx. 25% of liver non-toxicants were protein reactive – acylation, (pre) Michael addition, SN2 and Schiff base

formation

• Approx. 25% of liver toxicants were profiled as being capable of producing phosolipidosis

951 Liver Toxicants / Non-Toxicants

Profiled On Mechanism of Action

• Approx. 34% of liver toxicants were protein reactive (computational prediction)

• Approx. 25% of liver non-toxicants were protein reactive – acylation, (pre) Michael addition, SN2 and Schiff base

formation

• Approx. 25% of liver toxicants were profiled as being capable of producing phosolipidosis

The poor predictivity was expected:

• large number of mechanisms

• potential toxicity of metabolites not being

addressed.

New Alerts for Liver Toxicity With

Mechanistic Basis

Hewitt M et al (2013) Hepatotoxicity: A scheme for generating chemical categories for read-across, structural alerts and insights into mechanism(s) of action. Crit Rev Toxicol 43: 537–558

Examples of Alerts for Liver Toxicity

Effect Molecular Initiating Event

Type(s) of Structural

Alerts

# of Alerts

Examples

Phosopholiposis Trapping of

molecules within lysosomes

Generalistic > 30

Reactive Hepatotoxicity incl. Fibrosis

Disruption of cellular function

Covalent Binding

> 100

Mitochondrial Toxicity (Poster 2266)

Disruption of proton gradient

Redox Cycling 9

Steatosis Binding to LXR Toxicophores < 5

N

R1

R2R3

OO

Screening of COSMOS Cosmetics

Inventory Using Structural Alerts

• Does not imply toxicity

• Includes banned substances

Grouping Similar Compounds:

A Vision of the Future or Over-Simplification?

Teratogen Sedative

Activity cliffs demonstrate need for better

ways to predict toxicity

ToxPrint Chemotypes: The Next

Generation of Structural Alerts

• A new public tool to capture structural knowledge

• From a collaboration between US Food and Drug Administration (FDA), Molecular Networks, and Altamira

• Structural fragments with embedded physicochemical properties

– Atoms, bonds, electron systems and whole molecule

O

OH

Molecular-level property

constraints can also be

defined; e.g., find

molecules containing

this substructure and

having logP > 4

An atom in a

substructure can be

defined based on

values of a physico-

chemical property,

e.g., partial charge ≤ 0

Chemotype Matches in ChemoTyper

carbon has partial total

charge > 0.5

carbon has sigma

partial charge > 0.0

*More details will be presented by AS Mostrag-Szlichtyng (Poster #

2254), M Nelms (Poster # 2266), AN Richarz (Poster # 2273x)

• There are many openly available toxicity databases

– ACToR, eChemPortal, ChEMBL…

• Some lack quality control and may contain errors

• Few, if any, data on cosmetic ingredient

• Not designed for modelling, integration or intelligent queries

– “What are the structural characteristics of compounds that cause steatosis?”

– “Which organ level toxicity is important for this group of compounds?”

Category Formation: Databases of Toxicity Data

• New databases being developed

• COSMOS repeat dose toxicity database:

– Open access format

– Over 800 compounds

– Better querying capabilities

– Available December 2013

• eTox database: repeat dose tests

– Over 2,000 study records for >800

compounds

– More than 70% confidential

Category Formation: A New Generation of

Databases for Repeat Dose Toxicity Data

KNIME Workflows

Reducing Uncertainty:

Supporting Read-Across

• For regulatory purposes, read-across predictions may need support beyond “the chemicals look the same”

• The AOP provides the framework for gathering evidence to

– Support category membership

– (Possibly) provide a quantitative prediction

• Role of alerts

– Predictors of toxicity

– Grouping for read-across

Can We Predict Toxicity of

Cosmetics Ingredients?

Can We Predict Toxicity? Yes…

Understanding of

Mechanisms (AOPs)

Good

Chemistry

Kinetics (inc. Metabolism)

Databases

Flexible Software

Implementation

In Vivo Toxicity Prediction

Acceptance

Conclusions

• In silico techniques are vital in 21st Century toxicology and increasingly widely applied

• Move away from one QSAR fits all to intelligent use for organ level effects

• The MIE, through the AOP, provides the stimulus, justification and validation for models

• Alerts created will support grouping and read-across through within “testing strategies”

Acknowledgements

• The European Community’s Seventh Framework Program (FP7/2007-2013) COSMOS Project under grant agreement n° 266835 and Cosmetics Europe

• Co-workers in Liverpool, EU, USA