QSAR21: Linking biology to chemistry g gy y

through chemical inherency and mode-of-action pathwaysmode of action pathways

Chihae Yang14 June, 2012

• How chemistry helpsHow chemistry helps• Practical approaches

– QSAR 21QSAR 21• Mode of action QSAR• Weight of evidence

• What regulators may accept

Does chemistry help?

“Bi l i i dibl li t d Wh ?”“Biology is incredibly complicated.  Why?” – Richard Dawkins

Does chemistry help?

4

In vitro to in vivo correlation of ToxCastTM Phase I data before and after chemical structure consideration

Chemical inherency

Exposure

In use•Risk assessment

•Structural alerts •QSAR predictions

•Categoriessupervised

Biology Predictions

•Metabolic potential •Chemical reactivity

Categories

interactionlinking layer

Chemical interactions

• PhysChem propertiesSt t l f t unsupervised

linking layer

Intrinsic chemistry

5

• Structural features unsupervised

In collaboration with Ann Richard, US EPA NCCT

Role of cheminformatics in systems biology

Chemical library serves to probe in vitro biology• need sufficient diversity of structures to sample wide range of t t i t ti th d MOAtarget interactions, pathways, and MOAs• need sufficient representation of structures & features to enable QSAR inferences

In Vitro/HTS

In Vivo

Structures

In Vivo

Existing knowledgein collaboration with Ann Richard (US EPA)

Tox21 inventory - sources

NTP studied compounds of all typesNTP-studied compounds of all types, nominations and related compounds, ICCVAM & NICEATM validation and reference compounds, collaborations

NTP

US EPA

NCGC (3458)

FDA CDER Orangebook, OTC, NDC, Green Book, D @FDA B it i

p

US EPA (3726)

Drugs@FDA, Britain NHS, EMEA, Health Canada, Japan NHI

EPA high interest compounds, pesticidalactives, failed drugs, antimicrobials, green alternatives, HPV, MPV,, endocrine ref cmpds, pcompds with tox data from ToxRefDB and NTP in vivo, FDA CFSAN compds

Chemical roles in Tox21 inventory

Covering chemical spaces of:• Pharmaceuticals

– NCGC, Drugs@FDA.gov&• Food ingredients & additives

– US FDA PAFA, FCS• Biocide• Biocide

– EPA pesticide, FIFRA, PAN.

• Cosmetics– EU COSING, US PCPC

• General or industrialtotal of 8204 compounds

Structural features variationTox21 Structural coverage across the

Inventory sources

Feature

y

enrichment

libr

ary

Feature diversityra

gmen

tuc

tura

l fr

Stru

courtesy of Ann Richard, US EPA NCCT

Structural features across the chemical rolesDrugs Biocides Food Ing. & Add. Cosmetics

Pri-alkyl amines

Benzothiazine, Beta-lactam,,Carbamate

AzepinBenzimidazole

t lib

rary

Benzothiazinebeta-lactam

frag

men

t

Carbohydrate

uctu

ral f

PurinPyrazole

Steroid

Stru Purin

Steroid

Frequency of each fragment in dataset (log scale)

Principal components scores plot using structural featuresstructural features

drugsdrugsbiocidesfoodscosmetics

PC are extracted using a basis set of MOSES f t lib fi i t St t l f tfragment library as fingerprints. Structural features found in more than 5 compounds were included.

Principal components scores plot using physicochemical propertiesp y p p

4

6

8

10 drugfoodCOSMOSbiocide

drugsbiocidesfoodscosmetics

-8

-6

-4

-2

0

2

p 4

cosmetics

-10

-5

0

5

10

-10

-5

0

5

10-10

p2p3

logP, water solubility, dipole moment, complexity, h d b d d d t l lhydrogen bond donors and acceptors, molar volume, Lipinski score violation, diameter, radius of gyration

Structural features and physicochemical propertiesphysicochemical properties

• drugsdrugs• biocides• foods

ti• cosmetics

2-D Clustering of compounds toxicity endpoints vs. the structural features

7 toxicity endpoints

endpoints vs. the structural features

s

agm

ents

rope

rtie

s

ctur

e fr

a

chem

pr

f t t

stru

c

phys

surfactant

Toxicity endpoints include mutation (salmonella), rodent tumorigenicity, and developmental effects (e.g., cleft palate, microthalmia, renal tube defects)

Structural alertsExample of structural correlation with toxicity endpointswith toxicity endpoints

cyclic-conjugate-dienej di

Drugs Biocides Food Ing. & Add. Cosmetics

conjugate-diene_oxySBI_Pryimidine,oxo

SBI_TriazoleSBI_Dioxolane

St id 6 6 6 5 iSteroids_6,6,6,5-ringPhthalic ester

aromatic amine(NH2)aromatic azo

ti itaromatic nitroepoxide

furan, 2-nitroreactive alkyl halides

di lfiddisulfideshydrazine

aliphatic estersalpha-halocarbonyl

i h l tmichael acceptors

% frequency in log scale

MoA QSAR

• Applied to endpoints where chemical roles areApplied to endpoints where chemical roles are better understood.– Genetic toxicityGenetic toxicity– Genotoxic tumorigenicity– Skin sensitization– Skin sensitization– Skin irritation

Chemical inherency

Exposure

In use•Risk assessment

•Structural alerts •QSAR predictions

•Categoriessupervised

Biology Predictions

•Metabolic potential •Chemical reactivity

Categories

interactionlinking layer

Chemical interactions

• PhysChem propertiesSt t l f t unsupervised

linking layer

Intrinsic chemistry

17

• Structural features unsupervised

In collaboration with Ann Richard, US EPA NCCT

Biology representation: modes‐of‐action in toxicity & risk evaluationmodes of action in toxicity & risk evaluation

Toxicity pathways

Training setOrgans & Cellular M l l Organs &

OrganismsCellular

ResponseMolecularInitiating

EventOHCH3CH3

CH3CH3

chemical MoA categoryO

CH3

e.g. Retinoic Acid

Modes of action (MoA)

Adverse OutcomesModes of action (MoA)

MoA categories: genotoxic turmorigenicity

Chemical class Biological events

• Aliphatic halide

• Aromatic amine

genotoxictumorigenicity

• PAH

tumorigenicity

• Flavonoid

• Steroid

• Retinoid

MoAs: features for biology and chemistry

Biological features

Assays Toxicity

In vitro In vivoMoA

y Toxicity

Chemical features

From known pathways for cleft palate

• WNT• SHh• Growth factors• …

to Chemistry

Possible Mode-of-Actions for cleft palate

Chemical classes Biological key events  Phenotype 

• Glucocorticoid• Steroid

during embryogenesis effects

• SHh antagonist– Cyclopamin

• Sterol biosynthesis

Wnt signaling SHh signalingGlucocorticoid receptors disruptions in EMT

• Sterol biosynthesis inhibitors– triazinei id l

Retinoic acid receptorsP‐glycoprotein 

Glycogen synthase kinaseNuclear receptor

‐ cell proliferation‐ palate growth

‐ fusion

– imidazole– pyrimidine …

• Retinoids

Nuclear receptorAndrogen receptor

22

• Phthalates…

Data remodeling by tox ontology

Phenotype effects ChemicalPhenotype effects Chemical

Sites

System/Organ

Molecular entity

System/Organ

Region/TissuesChemical groups

Ph h tiC ll

Proteins   

Physchem propertiesCells

Chemical role 

Genes                        Biological role

23

Molecular pathways

Developmental database and ontology

• Tox data (1154 chemicals) merged from public sources*EPA T R fDb FDA CDER (d @fd ) FDA CFSAN (f d t t– EPA ToxRefDb, FDA CDER (drugs@fda), FDA CFSAN (food contact substance), ILSI DevTox, NTP

• Ontology for dose level effects• Ontology for dose‐level effects – Maternal toxicity– Fetal survival and growthFetal survival and growth– Fetal morphology

• e.g., cleft palate, microthalmia, kidney/renal/ureter anomalies

• Dysmorphogenesis data– 644 chemicals 

* Singh, AV et. al. Computational Toxicology, Eds Knudsen T et. al. Elsevier Limited 2010, 307-337.

Cleft palate chemotypes

Steroids, glucocorticolds, cyclopamine

Triazole, imidazole, pyridines, pyrimidine,

*substituent

imparting positive -charge at * carbon atom;

piperazole, dioxolane…

Retinoids, conjugated dienes

O

* charge at carbon atom;x=nitrogen or carbon

Phthalic ester (aliphaticchain) O

O

O

Thiocarbamates

O

O

Thiocarbamates

Organofluorides

Structural features and cleft palate chemotypesp yp

AlcoholCarbamate

3.3 ‐0.8 ‐0.1 0.8 0.1 5.8

‐0.4 2.7 ‐0.5 ‐1 ‐0.2 ‐1Halides_aliphatic

Halides_aliphatic_fluorideHalides_aromatic

d i

‐1.5 0 0.3 1 ‐0.3

‐1.5 ‐0.2 ‐0.5 0.7 1.5 0.3

‐1.8 0 3.2 ‐0.4 ‐1.6 1.9Hydrazine

UreaRetinoid_conjugate‐diene

R ti id j t di

‐0.8 0.1 ‐0.1 1.3 ‐2.3 0.4

‐0.1 0.1 0.7 1.2 0.3 1.3

‐1.1 ‐0.4 ‐0.2 0.1 2.4 0.5Retinoid_conjugate‐diene_oxy

Sterol_BSI_Pryimidine,oxoSterol_BSI_Triazole

Sterol BSI Dioxolane

‐1.2 ‐0.4 0.7 0.7 2.8 0.7

‐0.2 ‐0.8 1.6 ‐0.9 0.7 ‐0.9

0.7 2.3 0.8 0.6 0Sterol_BSI_DioxolaneSteroids_6,6,6,5‐ring

Phthalic ester

2.5 1.8 0.5 0.2 0.8 0.9

2.6 1.5 0.9 1.1 2.6 1.2

0.4 0.1 0.75 0.3 0.6 3.3

Molecular features for modeling

Structural features Calculated properties

• Genericized features 

• FDA Redbook categories*

• Electronic*‐ Dipole moment, Hydrogen bond acceptors/donors

• EPA pesticide MoA classes

• MOSES metabolic reaction l

bond acceptors/donors• Steric‐ Topological complexity*, 

rules

• Reactive chemicals (RECAP)ǂ

Cl ft P l t h t

Shape descriptors• Hydrophobicity*

LogP Water solubility• Cleft Palate chemotypes ‐ LogP, Water solubility, Topological polar surface area

• qHTS assay (ToxCast)ǂ Hann et. al. J. Chem. Inf. Comput. 

‐ TGF1Sci. 1999, 39, 897‐902. 

*Publicly available from Molecular Networks

MoA/WoE modeling approach for cleft palatep

maternal toxicityQuery

Alerts

not maternal tox dose

alerts (chemotypes)N

MoACategories

N

N

TriazolesRetinoidsGlobal DioxolanesConjug DienesTGF Triazoles…RetinoidsGlobal DioxolanesConjug. DienesTGF

WOE model Cleft palate!28

i ifinal

i

w pp

w

MoA/WoE models for rat cleft palate

WOE (Bayesian) % Concordance % Sensitivity % Specificity

87 73 94

%  %  % sets (pos/neg)

( y )87 73 94

Concordance Sensitivity Specificity PLS factors Weights

84 71 93 5 0.8

84 92 71 1 1Retinoids (5/14)

Global (81/167)

sets (pos/ eg)

84 92 71 1 1

92 80 96 1 0.5

89 78 94 5 1

( / )

Conjugated alkene (11/27) 

Sterol BSI – N (31/62)

86 82 89 1 1

84 73 90 1 1

Sterol BSI ‐ O (13/16)

Organofluoride (13/18)

29

Summary

• QSAR 21 B i h i t t T 21 d bi l t QSAR– Bring chemistry to Tox21 and biology to QSAR

– MoA and weight of evidence (WoE)

Biological features Chemical

molecular featuresMoA

A li d t d l t l ff t

molecular features

• Applied to developmental effects– Cleft palate (rat, mouse), Eye, Kidney, ureter

Reality of safety/risk assessment

Safety/Risk assessment deals with extremely diverse and disparate data and information!

• Complex data– Experimental data: In vivo, in vitro …– Exposure information

• Lack of data or knowledge• Knowledge from past

– Structural alerts– Read‐acrossTh h ld f i l i l (TTC)– Threshold of toxicological concern (TTC)

– QSAR predictions  31

Regulators needs

• Qualified data at their fingertips– has it been regulated before?– where has it been regulated?g– what is the current status

• Mechanistic rationaleMechanistic rationale– strong desire to bring pathway knowledge into workflowworkflow

• Make their decision easier, not harder

ifti th h th il f i f ti d lt f– sifting through the pile of information and results from numerous methods and models

What regulators wanted for CFSAN CERES

Chemical evaluation and risk estimation system (CERES)

• Transparent rationale

• Adopt pragmatic approaches reflecting systems biology– e.g., Biological analogs based on MoA

• Quantitative weight of evidence assessment

• Dashboard, War room, Red flags…

Acknowledgment

• James F Rathman, the Ohio State University• Ann Richard, US EPA NCCT• Andrew Worth, JRC• US FDA CFSAN

– Kirk Arvidson, Annette McCarthy, Jason Aungst, Kristi Jacobs Mary Shackelford Yan Gu Deborah Hansen (NCTR)Jacobs, Mary Shackelford, Yan Gu, Deborah Hansen (NCTR)

• Funding: European Community’s Seventh Framework Program (FP7/2007‐2013) and from Cosmetics Europe (Grant agreement n° 266835) in SEURAT cluster.