Computational Modelling for TTC Assessment

Andrew Worth1 and Chihae Yang2 1) European Commission - Joint Research Centre Institute for Health & Consumer Protection, Systems Toxicology Unit

2) Altamira LLC, Columbus Ohio, USA and Molecular Networks GmbH, Erlangen, Germany

Eurotox 2015 Continuing Education Course on

“Thresholds of Toxicological Concern – Basics and Latest Developments”

Porto, Portugal, 13 September 2015

Overview

• Computational tools for the application of the TTC approach

• Chemoinformatics in the development of the TTC approach

• Quality-controlled datasets for modelling

• Investigation of chemical space

• Identification of chemotypes

• Route to route extrapolation

• COSMOS-ILSI decision tree for oral to dermal extrapolation

• Internal TTC approach and biokinetic modelling

• Take home messages

2

Computational tools for the application of the TTC approach

1. Is the substance a non-essential metal or metal containing compound, or is it a polyhalogenated-

dibenzodioxin, -dibenzofuran, or -biphenyl?

3. Is the chemical an aflatoxin-like-, azoxy-, or

N-nit roso- compound?

2. Are there structural alerts that raise

concern for potential genotoxicity?Risk assessment requires

compound-specific toxicity data

4. Does estimated intake exceed TTC of

0.15g/day?

Negligible risk (low probability of a life -time

cancer risk greater than 1 in 106 – see text)

5. Does estimated intake exceed TTC

of 1.5g/day?

6. Is the compound an organophosphate?

10. Is the compound

in Cramer structural

class II?

8. Is the compound in

Cramer structural class

III?

12. Does estimated intake

exceed 1800g/day?

YESNO

NO

7. Does estimated intake exceed

TTC of 18g/day?YES

NO

Substance would not be expected

to be a safety concern

YES

YES

YES

11. Does estimated intake

exceed 540g/day?

NO

9. Does estimated intake

exceed 90g/day?

NO YES

NO

YES

YES

YES

YES

NO

NO

Risk assessment requires

compound-specific toxicity data

Substance would

not be expected to

be a safety concern

YESNOYESRisk assessment requires

compound-specific toxicity data

NO

NO

Substance would not be

expected to be a safety concern

NO

1. Is the substance a non-essential metal or metal containing compound, or is it a polyhalogenated-

dibenzodioxin, -dibenzofuran, or -biphenyl?

3. Is the chemical an aflatoxin-like-, azoxy-, or

N-nit roso- compound?

2. Are there structural alerts that raise

concern for potential genotoxicity?Risk assessment requires

compound-specific toxicity data

4. Does estimated intake exceed TTC of

0.15g/day?

Negligible risk (low probability of a life -time

cancer risk greater than 1 in 106 – see text)

5. Does estimated intake exceed TTC

of 1.5g/day?

6. Is the compound an organophosphate?

10. Is the compound

in Cramer structural

class II?

8. Is the compound in

Cramer structural class

III?

12. Does estimated intake

exceed 1800g/day?

YESNO

NO

7. Does estimated intake exceed

TTC of 18g/day?YES

NO

Substance would not be expected

to be a safety concern

YES

YES

YES

11. Does estimated intake

exceed 540g/day?

NO

9. Does estimated intake

exceed 90g/day?

NO YES

NO

YES

YES

YES

YES

NO

NO

Risk assessment requires

compound-specific toxicity data

Substance would

not be expected to

be a safety concern

YESNOYESRisk assessment requires

compound-specific toxicity data

NO

NO

Substance would not be

expected to be a safety concern

NO

Cancer

endpoints

Non-

cancer

endpoints

Kroes et al. (2004).

Food Chem Toxicol 42, 65-83.

High potency

carcinogen

Structural alert

for genotoxicity

Organophosphate

neurotoxicant

Kroes

decision tree

Is the substance a member of an

exclusion category?

Is there a structural alert for

genotoxicity (including

metabolites) ?

Exposure > 0.3 µg/kg bw/day?

Is substance an OP/Carbamate?

Exposure > 1.5 µg/kg bw/day?

Is substance in Cramer Class II or III?

Exposure

> 0.0025 µg/kg bw/day?

Substance

requires non-TTC approach

(toxicity data, read-across, etc)

Low probability of

health effect

Low probability of

health effect

Exposure > 30 µg/kg bw/day?

No

No

No

Yes

No

Yes

No

Yes

Yes

Yes

Yes

Yes

No

No

No

Yes

Does the substance have a known structure and are exposure data available?

Yes

No TTC approach cannot

be applied

EFSA website: http://www.efsa.europa.eu/en/efsajournal/doc/2750.pdf

EFSA

decision tree

Structural Alerts (SAs)

• 30 SAs for genotoxic carcinogens

• 5 SAs for non-genotoxic carcinogens

+

Benigni-Bossa rules for genotoxicity & carcinogenicity

Three QSAR models

Probability

of effect = F Hydrophobic Electronic Steric + +

Descriptors: logP HOMO, LUMO MR

• Statistically-based

• Support Vector Machine classification method + some SAs from the Benigni-

Bossa rulebase

• Training set: 4225 compounds from the Kazius-Bursi database

http://www.caesar-project.eu

http://www.vega-qsar.eu/

CAESAR mutagenicity model

• Statistically-based

• Classification model based on a Counter-Propagation Artificial Neural Network

(CP-ANN)

• Training set: 805 compounds from the Carcinogenic Potency Database (CPDB )

http://www.caesar-project.eu

http://www.vega-qsar.eu/

CAESAR carcinogenicity model

• Downloadable versions from JRC and Sourceforge (http://toxtree.sourceforge.net)

• Current version 2.6.6 (Jun3 2014) includes Cramer, Cramer with Extensions, genotoxicity and carcinogenicity (Benigni-Bossa, In Vivo Micronucleus, Ames), Kroes

• Toxtree online: http://toxtree.sf.net/predict

Toxtree

•Prediction

•Compound structure

•Compound properties

•Reasoning

• naturally occurring in beer, sweet corn, corn tortillas, milk

• proposed flavouring agent

• EFSA opinion

(2008):http://www.efsa.europa.eu/en/efsajournal/pub/797.htm

• Estimated intake:

Maximised Survey-derived Daily Intake (MSDI) of 0.012 g/p/day

Modified Theoretical Added Maximum Daily Intake (mTAMDI) of 1600

g/p/day

SMILES: O=C(C)c1ccccc1N

2-aminoacetophenone

Example: 2-aminoacetophenone

• Cramer Class III

• 1N,2N,3N,5N,6N,7N,16N,17N,19N,23Y,27Y,28N,30Y,31N,32N,22N,33N

• TTC for CC III is 90 g / person / day

• Estimated intake:

MSDI of 0.012 g/p/day < TTC of 90 g/p/ day for CC III

BUT mTAMDI of 1600 g/p/day > TTC of 90 g/p/ day for CC III

Example: 2-aminoacetophenone

SMILES: O=C(C)c1ccccc1N

2-aminoacetophenone

• Many of the original Cramer rules are written in a confusing and inter-dependent way,

which leads to difficulties in the rationalisation of the predictions they make.

• Two rules are not based on chemical features, but simply make reference to look-up

lists of chemicals (Q1, normal body constituents; Q22, common food components).

• Some rules make ambiguous references to chemical features (e.g. steric hindrance)

which need to be clarified and possibly revised/deleted.

• Several studies have identified outliers (e.g. Class I compounds that have low

NOELs). A revised / alternative classification scheme should be more discriminating

in terms of NOEL values.

→ need to update Cramer classification scheme

Lapenna & Worth (2011). JRC report EUR 24898 EN

Evaluation of Toxtree-Cramer

Cramer classifications:

computer-based predictions vs expert judgement

Toxtree

Class II

QSAR Toolbox

Class III

Cramer classifications:

computer-based predictions vs expert judgement

Chemoinformatics in the development of the TTC approach

COSMOS

Integrated In Silico Models for the Prediction of Human Repeated

Dose Toxicity of Cosmetics to Optimise Safety

• Collection of toxicological data

• Development of the Threshold of Toxicological Concern (TTC) approach

• Development of novel in silico methods

• Multiscale modelling:

mitochondrial (dys)function, virtual cell-based assay, 2D liver,

Physiologically Based Biokinetic (PBBK) models

• In silico workflows based on open-source and open-access tools

17

COSMOS database v1.0

• Open-access

• High-quality toxicity data (quality

control, structure curation)

• User-friendly query builder (chemical

name, structure, toxicity data)

• 44,765 unique chemical structures

• 12,538 toxicity studies for 1,660

compounds across 27 endpoints

Webinar and tutorial:

http://www.cosmostox.eu/what/COSMOSdb/

http://cosmosdb.cosmostox.eu/

18

COSMOS Cosmetics Inventory

• Over 5,500 substances

• 66 unique use functions

Chemical classes

Chemical classes

19

Munro dataset (1996)

TTC

dataset

NOAEL database

Munro 1996

• Study reliability • NOAEL selection criteria

• Substance & Study inclusion criteria

• Study relevance • NOAEL decision • Expert review

V1.8 current version

2. NOAEL database 3. TTC dataset 1. Toxicity database

Filter 1 Filter 2

oRepeatTox DB

COSMOS TTC dataset

Compound classes COSMOS TTC v1.8 (all tentative counts)

Munro

All ̴560 613

Cosmetics inventory 495 190

Cramer Class I: Class II: Class III* 244: 35: 281 (> 40% Class I)

119: 28: 448 (<25 % Class I)

Nutrients (removed) - Lipid soluble vitamins - Essential amino acids

- Vitamin A,D,E,K removed - removed

- retinol - phenyl alanine

Compound categories - Hair dyes - Parabens - Phthalates

110 10 7

13 7 5

* Cramer Classes assigned by Toxtree v2.6.0

Description of COSMOS TTC v1.8

% in dataset (ratio)

alcohol alcohol, phenol

alcohol - 1,1; 1,2; 1,3 aldehyde

amine amine, aromatic halides, organo

ketones phosphorus containing

sulfonyl (S=O) pyran ring, generic

silicon urea

Chain - aliphatic chain >= C8 Chain - oxyalkane (EO-PO)

Surfactants - nonionic Surfactants - anionic Surfactants - cationic

Carbohydrate Steroid ring

Parabens Phthalates

Hair-dyes - amine_ethanol Hair-dyes - amine_bis_ethanol

Hair-dyes - azo Hair-dyes - benzene_amino_nitro_alcohol

Hair-dyes - benzene_diamino Hair-dyes - benzene_nitro

> 4x organohalides

> 3x phosphporus

> 2x urea

steroid ring

Surfactant - cationic

ToxPrint chemotypes

Organo silicon

COSMOS (v1.8) MUNRO

Chemical space –

structural features

Chemotypes

• Structural fragments with atom/bond properties (partial charges, polarizability,

electronegativity, etc.)

• Improved ability to predict reactivity and toxicity (compared with structural

alerts)

• Example: association of diazoles and triazoles with cleft palate formation

X= nitrogen or carbon

Pi charge < zero Sigma charge > zero

24

Chemical space – physicochemical descriptors

25

The Chemotyper

• The Chemotyper • feature searching

• profiling datasets / inventories

• building prediction models

• ToxPrint

• library of public chemotypes

• Generic fragments

• Genotoxic carcinogens

(Ashby-Tennant)

• Cancer TTC (Kroes et al)

Developed by Altamira LLC and Molecular Networks GmbH under FDA contract

Publicly available from:

Chemotyper: https://www.chemotyper.org

ToxPrint: https://toxprint.org 26

AOPs for liver toxicity (fibrosis and steatosis)

Landesmann et al (2012). Description of Prototype Modes-of-Action Related to Repeated Dose

Toxicity. JRC report EUR 25631 EN.

Protein alkylation to fibrosis

MolecularTissueCellularOrganelle

Biological

Organization

Level

PPAR-α

antagonism

binding

Steatosis

> 5–10% by

liver weight

Fatty liver

cells

Cytoplasm

displacement

Nucleus

distortion

Mitochondrial

disruption

TGs

accumulation

Inhibition of the

mitochondrial

b-oxidation

PPAR-γ

activation

ER

binding

Inhibition of

respiration =

NAD+ deplition

AhR

agonism

CD36 up-

regulation

Increase of the

fat influx from

peripheral

tissues

Peroxisomal

AOX inhibition

MIE Intermediate Effects

Inhibition of the

microsomal

b-oxidation

PXR

activationInduction of

CYP3A4

LXR

activationAOP from LXR

De novo FA

synthesis

ChREBP

SREBP-1c FAS

ACC

SCD-1

L-PK

ModulatorsKey

events

Intermediate

events

Adverse

Outcome

Molecular

Initiating Event

Angptl3

PTLP

Inhibition of the

TG excretion

ApoE

LXR activation to steatosis

27

Liver toxicity – pathology and mechanisms

Pathological

changes

Molecular

mechanisms

STEATOSIS STEATOHEPATITIS FIBROSIS

28

COSMOS oRepeatToxDB

• 228 cosmetics-related chemicals

• 340 oral studies

(subacute, subchronic, chronic, reproductive &

developmental)

• Controlled vocabulary for toxicological effects

COSMOS database publicly accessible from:

http://cosmosdb.cosmostox.eu/accounts/login

Liver toxicity – mining the COSMOS database (1)

29

Query for liver effects (chronic, subchronic,subacute)

59 chemicals with liver effects

Liver toxicity – mining the COSMOS database (2)

30

39 chemicals with liver steatosis / steatohepatitis / fibrosis

Liver toxicity – mining the COSMOS database (3)

31

Identifying chemotypes for liver toxicity

O

O

OO

O

Cl

N

O

Cl

Cl Cl

• Alcohols, diols, glycol ethers

• Michael acceptors

• Amino phenols, aromatic amines, aromatic

halides

• Polychlorinated short alkanes

• Halogenated amines

O

N

Cl

N

32

Route to route extrapolation

• To add

COSMOS-ILSI Decision tree

35

External Dose

HTS - In vitro

concentration

Biologically-based modelling for an internal TTC

Internal concentration External dose – internal

response

Internal (cellular) response

Gajewska et al (2014).

Toxicology Letters 227, 189-202.

Take home messages

• Various software tools are available to support the application of TTC

• Computational predictions are not intended to be the solution – need to apply

expert judgement, especially for certain chemical classes

• Computational methods also provide a means of further developing the TTC

approach

• Chemistry-based modelling can be supplemented with biological modelling

• Ongoing research is aiming to:

• refine or replace the traditional Cramer tree

• extend the applicability of the approach to new chemical classes

• extend the applicability of the approach to non-oral routes of exposure

• Aleksandra Mostrag-Szlichtyng; Altamira LLC, Columbus, Ohio, USA

• Vessela Vitcheva; Molecular Networks GmbH, Erlangen, Germany

• Kirk Arvidson; US FDA, Center for Food Safety & Applied Nutrition, Maryland, USA

• Alicia Paini; European Commission, Joint Research Centre, Italy

Acknowledgements

37

Some of the research leading to these results has received funding from the

European Community’s Seventh Framework Program (FP7/2007-2013)

COSMOS Project under grant agreement no. 266835 and from Cosmetics

Europe.