evolution of computational toxicology
Post on 23-Jan-2022
3 Views
Preview:
TRANSCRIPT
Evolution of Computational ToxicologyFrom Primitive Beginnings to Sophisticated Application
Rusty ThomasDirectorNational Center for Computational Toxicology
Health Canada Seminar
November 15, 2016
The views expressed in this presentation are those of the presenter and do not necessarily reflect the views or policies of the U.S. EPA
Comp Chem, HTTK, HTS, Systems Tox, HT Exposure
National Center forComputational Toxicology
1
Early Beginnings of Computational Toxicology…
• Application of computational modeling to toxicological endpoints began with development of QSARs in 1970s.
• In early 1970s, first physiologically-based pharmacokinetic (PBPK) models were developed
• In 1980s, significant growth in computer modeling in QSAR and PBPK modeling
• In 1990s, development of physiologically-based pharmacodynamic (PBPD) models for AChE inhibition and cell death/proliferation/mutation
• In the late 1990s, use of the term ‘computational toxicology’ appeared in the literature
0
5
10
15
20
25
30
35
No.
Pub
Med
Cita
tions
[ ] = f ( )
National Center forComputational Toxicology
2
Rapid Evolution of Computational Toxicology
c. 1970s - 2002 2003 - 2006 2007 - 2010 2011 - 2016c.1500 – 1960s…
• The field of computational toxicology has since undergone punctuated and rapid evolution
• What started out as predicting toxicity based on (Q)SAR has rapidly evolved to use computational modeling to integrate diverse data streams
• Diverse regulatory needs, complex scientific challenges, expert panel reports, and large influx of data have added selection pressure
Comp Chem, HTTK, HTS, Systems Tox, HT Exposure
Comp Chem, HTTK, HTS, Systems Tox, HT Exposure
National Center forComputational Toxicology
3
What Traits Have Been Under Selection?
• Increased throughput and biological coverage
• Interrogation of effects at the molecular and pathway level
• Increasingly relevant test systems
• Putting results in a dose/exposure context
• Characterization of uncertainty
• Computational modeling to integrate experimental data
• Delivery of data and models through decision support tools
• Transparency and performance-based validation
National Center forComputational Toxicology
4
Throughput Spurred by Regulatory Demand and Expert Guidance
• In 1976, the U.S. Toxic Substances Control Act put burden on EPA to demonstrate “unreasonable risk” for industrial chemicals
• Increased recognition of lack of safety data for many environmental/industrial chemicals– 1984 U.S. NRC report
• Guidance from internal strategy reports and expert committees highlighted need for increased throughput– 2005 EPA CompTox Report – 2007 U.S. NRC report
0
10
20
30
40
50
60
70
Perc
ent o
f Che
mic
als
Acute CancerGentox Dev ToxRepro Tox
Judson, et al EHP (2010)
National Center forComputational Toxicology
5
Increased Throughput Required Shift to Molecular/Pathway Approaches
ToxCast
Concentration
Resp
onse
~600 Cell & biochemical
assays
~1,000 Chemicals
Tox21
~30 Cell & biochemical
assays
~8,000 Chemicals
Set Chemicals Assays Completion
ToxCast Phase I 293 ~600 2011
ToxCast Phase II 767 ~600 2013
ToxCast Phase III 1001 ~100 Ongoing
E1K (endocrine) 880 ~50 2013
National Center forComputational Toxicology
6
Broad Success Derived from High-Throughput Screening Approaches
Provide Mechanistic Support for Hazard ID
Group Chemicals by Similar Bioactivity and Predictive Modeling
Prioritization of Chemicals for Further Testing
Assays/Pathways
Che
mic
als
IARC Monographs 110, 112, 113
FIFRA SAP, Dec 2014
National Center forComputational Toxicology
7
Continuing Pressure Towards Increased Biological Coverage
Picture of Everything - Howard Hallis
Whole Transcriptome Technologies
Diverse Cell Types & Culture
Systems
National Center forComputational Toxicology
8
Searching for a Platform
Low Coverage RNA-seq
MCF7
6 chemicals1 concentration
5 Replicates@ 6 hours
TempO Seq & TruSeq
MCF7
6 chemicals1 concentration
5 Replicates@ 6 hours Low Coverage
RNA-seq
MAQCRNA Stds
TempO Seq & TruSeq
MAQCRNA Stds
4 RNA Pools
5 ReplicatesLow Coverage
RNA-seq
4 RNA Pools
5 Replicates
Targeted RNA-seq
Requirements:• Whole genome • 384 well • Automatable• Low cost
Technical Comparison
Technical Comparison
Functional Comparison
Functional Comparison
National Center forComputational Toxicology
9
Technical Performance of the Three Sequencing Platforms
TruSeqr2 0.74
TempO-Seqr2 0.75
Low Coverager2 0.83
National Center forComputational Toxicology
10
Functional Performance of the Three Sequencing Platforms
TrichostatinGenistein
Low Coverage TruSeq TempOSeqCorrectCMAP 0/5 (0%) 0/5 (0%) 4/5 (80%)
• Large scale screen of 1,000 chemicals (ToxCast I/II) in single cell type
• Additional screens across multiple cell types/lines• Additional reference chemicals and genetic
perturbations (RNAi/CRISPR/cDNA)
National Center forComputational Toxicology
11
What Traits Have Been Under Selection?
• Increased throughput and biological coverage
• Interrogation of effects at the molecular and pathway level
• Increasingly relevant test systems
• Putting results in a dose/exposure context
• Characterization of uncertainty
• Computational modeling to integrate experimental data
• Delivery of data and models through decision support tools
• Transparency and performance-based validation
National Center forComputational Toxicology
Increasing Relevance of Test Systems Necessary at Multiple Levels
Relevant Metabolism
Relevant Tissue and Organ Responses
Relevant Species Responses
National Center forComputational Toxicology
13
Assessing Cross-Species Differences in Response
Houck et al., Unpublished
Multispecies Attagene Trans Reporter Assay
• Host cell: human HepG2• Stimulation with EC20 of 6a-
fluorotestosterone for detection of androgen receptor antagonists
• 100 chemicals with ER, AR, PPAR activity tested in concentration-response
• Data calculated as fold-change over control (6a-fluorotestosterone/DMSO)
National Center forComputational Toxicology
14
Cross-Species Differences in Nuclear Receptor Responses
-log AC50 (uM)
= agonist= antagonist
National Center forComputational Toxicology
15
Adding Relevant Metabolic Activity
-3 -2 -1 0 1 2 30
50
100
150
CYP1A2
[Compound] log µM
Perc
ent A
ctiv
ity
-3 -2 -1 0 1 2 30
50
100
150
CYP2B6
[Compound] log µM
Perc
ent A
ctiv
ity
-3 -2 -1 0 1 2 30
50
100
150
CYP2C9
[Compound] log µM
Perc
ent A
ctiv
ity
-3 -2 -1 0 1 2 30
50
100
150
CYP3A4
[Compound] log µM
Perc
ent A
ctiv
ity
3
Furafylline
Thio-TEPA
Tienilic Acid
Ketoconazole
DeGroot and Simmons, Unpublished
National Center forComputational Toxicology
16
What Traits Have Been Under Selection?
• Increased throughput and biological coverage
• Interrogation of effects at the molecular and pathway level
• Increasingly relevant test systems
• Putting results in a dose/exposure context
• Characterization of uncertainty
• Computational modeling to integrate experimental data
• Delivery of data and models through decision support tools
• Transparency and performance-based validation
National Center forComputational Toxicology
The Need for High(er) Throughput TK Approaches
0
100
200
300
400
500
600
700
800
Total Number Chemicals with InVivo TK
HTTK (Hamner &EPA)
HTTK (In Process)
Num
ber o
f Che
mic
als
ToxCast Phase I ToxCast Phase II
National Center forComputational Toxicology
Incorporating a High-Throughput Toxicokinetic Approach
Rotroff et al., Tox Sci., 2010Wetmore et al., Tox Sci., 2012
Reverse Dosimetry
Oral Exposure
Plasma Concentration
ToxCast AC50Value
Oral Dose Required to Achieve Steady
State Plasma Concentrations
Equivalent to In VitroBioactivity
Human Liver Metabolism
Human Plasma Protein Binding
Population-Based IVIVE Model
Upper 95th Percentile CssAmong 100 Healthy
Individuals of Both Sexes from 20 to 50 Yrs Old
EPA ToxCast Phase I and II Chemicals • Models available through ‘“httk” R package
(https://cran.r-project.org/web/packages/httk/)
National Center forComputational Toxicology
Comparing Dosimetry Adjusted Bioactivity with Exposure
Wetmore et al., Tox Sci., 2012
Fent
in H
ydro
xide
Clo
prop
Qui
noxy
fen
Spiro
xam
ine
Endo
sulfa
nIp
rodi
one
Nic
losa
mid
eH
alos
ulfu
ron-
met
hyl
Cyp
rodi
nil
Prom
eton
Emam
ectin
ben
zoat
eA
ciflu
orfe
nPa
rath
ion
Etox
azol
eFe
noxy
carb
Lind
ane
Etha
lflur
alin
2,4-
DB
Tri-a
llate
Fenb
ucon
azol
eC
hlor
pyrif
os-m
ethy
lB
ensu
lfuro
n-m
ethy
lC
hlor
etho
xyfo
sIs
oxab
enD
iclo
fop-
met
hyl
Prop
etam
phos
Trifl
oxys
ulfu
ron-
sodi
umIn
doxa
carb
Dic
hlor
prop
MC
PAB
enta
zone
Qui
nclo
rac
Dic
ofol
Pros
ulfu
ron
Iodo
sulfu
ron-
met
hyl-s
odiu
mPy
rithi
obac
-sod
ium
Esfe
nval
erat
e2,
4-D
Dic
hlor
anIm
azal
ilC
lofe
ntez
ine
Prod
iam
ine
PFO
SN
apro
pam
ide
Bife
nthr
inPr
omet
ryn
Din
icon
azol
eTh
idia
zuro
nPi
clor
amFi
pron
ilPr
opaz
ine
Nitr
apyr
inTe
bufe
npyr
adR
oten
one
Bro
mac
ilFe
narim
olD
iclo
sula
mPi
rimip
hos-
met
hyl
Etha
met
sulfu
ron-
met
hyl
Forc
hlor
fenu
ron
0.00001
0.0001
0.001
0.01
0.1
1
10
100
1000
10000
100000
Ora
l Equ
ival
ent D
ose
or E
stim
ated
Exp
osur
e(m
g/kg
/day
)
Lact
ofen
Dith
iopy
rA
nila
zine
Chl
orpr
opha
mD
iazi
non
Flum
etra
linPy
racl
ostr
obin
Pyrid
aben
Clo
roph
ene
Oxa
diaz
onC
oum
apho
sTe
trac
onaz
ole
Thio
benc
arb
Flum
etsu
lam
Prop
yzam
ide
Mon
o-n-
buty
l Pht
hala
teM
esos
ulfu
ron-
met
hyl
Am
etry
nC
yclo
ate
Feni
trot
hion
Hex
ythi
azox
Triti
cona
zole
Met
hoxy
feno
zide
Fent
hion
Peno
xsul
amC
yrom
azin
eA
traz
ine
Prop
anil
Tria
dim
enol
Flud
ioxo
nil
Milb
emec
tinFl
uoxa
stro
bin
Pipe
rony
l but
oxid
eTr
iclo
pyr
Imaz
apyr
Cyp
roco
nazo
leB
utac
hlor
Nov
alur
onIm
azaq
uin
But
ylat
ePe
ndim
etha
linO
xasu
lfuro
nPh
osal
one
Perm
ethr
inFl
urox
ypyr
Terb
acil
Sim
azin
eB
utra
linR
esm
ethr
inB
upro
fezi
nM
ethy
l Par
athi
onFl
uom
etur
onB
enflu
ralin
Bos
calid
Ace
tam
iprid
Flut
olan
ilC
inm
ethy
linPr
ochl
oraz
Trifl
ural
inN
orflu
razo
n
0.00001
0.0001
0.001
0.01
0.1
1
10
100
1000
10000
100000
Ora
l Equ
ival
ent D
ose
or E
stim
ated
Exp
osur
e(m
g/kg
/day
)
Tebu
pirim
fos
Trifl
umiz
ole
Tric
losa
nFo
sthi
azat
ePy
rimet
hani
lA
mitr
azH
PTE
MG
KD
iuro
nM
etho
xych
lor
Tria
sulfu
ron
Qui
ntoz
ene
Fora
msu
lfuro
nFl
uazi
nam
Rim
sulfu
ron
Dife
noco
nazo
leB
enom
ylPr
opox
urM
etsu
lfuro
n-m
ethy
lTh
iabe
ndaz
ole
Isaz
ofos
Myc
lobu
tani
lM
alat
hion
PFO
ATe
fluth
rinTe
bufe
nozi
deIs
oxaf
luto
leEP
TCFl
usila
zole
Hex
acon
azol
eZo
xam
ide
Fena
mid
one
Ald
icar
bD
isul
foto
nFl
umio
xazi
nEt
ridia
zole
Dic
hlob
enil
Tepr
alox
ydim
2-Ph
enyl
phen
olD
imet
hoat
eTh
iazo
pyr
zoqu
at m
ethy
l sul
fate
Dic
roto
phos
Aba
mec
tinM
olin
ate
Ben
sulid
eD
imet
hom
orph
-Des
isop
ropy
latr
azin
ete
trac
yclin
e di
hydr
ate
Thia
met
hoxa
mIm
azet
hapy
rC
loth
iani
din
Dip
heny
lam
ine
Feno
xapr
op-e
thyl
Azo
xyst
robi
nO
ryza
linC
lom
azon
eSe
thox
ydim
Met
ribuz
inD
ieth
yl to
luam
ide
0.00001
0.0001
0.001
0.01
0.1
1
10
100
1000
10000
100000
Chl
oron
ebPa
clob
utra
zol
Pyrip
roxy
fen
Mes
otrio
neD
icam
baO
xyflu
orfe
nC
acod
ylic
aci
dA
ceph
ate
Chl
orid
azon
Linu
ron
Imaz
amox
Met
hida
thio
nC
arbo
xin
Imaz
apic
Pirim
icar
bO
xam
ylEt
hopr
opC
yana
zine
Fena
mip
hos
Tria
dim
efon
Cyc
lani
lide
Thia
clop
ridlu
roxy
pyr-
mep
tyl
Phen
oxye
than
olB
isph
enol
-ATe
tram
ethr
inM
etal
axyl
hyhe
xyl p
htha
late
Fenh
exam
idIc
arid
inTr
iflus
ulfu
ron
Prop
amoc
arb
HC
lB
endi
ocar
bTe
buth
iuro
nVi
nclo
zolin
Trib
ufos
Bife
naza
teIm
idac
lopr
idM
etol
achl
orEt
hofu
mes
ate
ioph
anat
e m
ethy
lFl
ufen
acet
Daz
omet
Bro
mox
ynil
Sulfe
ntra
zone
Dim
ethe
nam
idS-
Bio
alle
thrin
Car
bary
lA
lach
lor
Hex
azin
one
Azi
npho
s-m
ethy
lA
ceto
chlo
ris
,tran
s- A
lleth
rinPy
met
rozi
neFo
rmet
anat
e H
Cl
Flua
zifo
p-P-
buty
lD
ibut
yl p
htha
late
imet
hyl p
htha
late
Dia
zoxo
n
0.00001
0.0001
0.001
0.01
0.1
1
10
100
1000
10000
100000
Ora
l Equ
ival
ent D
ose
or E
stim
ated
Exp
osur
e(m
g/kg
/day
)
Assay Bioactivity
Exposure Range
National Center forComputational Toxicology
20
What Traits Have Been Under Selection?
• Increased throughput and biological coverage
• Interrogation of effects at the molecular and pathway level
• Increasingly relevant test systems
• Putting results in a dose/exposure context
• Characterization of uncertainty
• Computational modeling to integrate experimental data
• Delivery of data and models through decision support tools
• Transparency and performance-based validation
National Center forComputational Toxicology
Incorporating Uncertainty in the Pharmacodynamic Modeling
21
Watt et al., Unpublished
Bootstrapped Modeling of CR Curves
National Center forComputational Toxicology
Incorporating Uncertainty Into the Pharmacokinetic Modeling
Bayesian Modeling of Plasma Protein Binding and Intrinsic Clearance
Propagation of Experimental Uncertainty to Steady State Plasma Concentrations
Wambaugh, Unpublished
National Center forComputational Toxicology
Impact of Incorporating Uncertainty and Variability is Chemical Specific
Wambaugh, Unpublished
Chem 1 Chem 100
Variability Predominates
Uncertainty Predominates
National Center forComputational Toxicology
24
What Traits Have Been Under Selection?
• Increased throughput and biological coverage
• Interrogation of effects at the molecular and pathway level
• Increasingly relevant test systems
• Putting results in a dose/exposure context
• Characterization of uncertainty
• Computational modeling to integrate experimental data
• Delivery of data and models through decision support tools
• Transparency and performance-based validation
National Center forComputational Toxicology
Relatively Simple Models Used to Reduce Assay Interference
ER Receptor Binding(Agonist)
Dimerization
CofactorRecruitment
DNA Binding
RNA Transcription
Protein Production
ER-inducedProliferation
R3
R1
R5
R7
R8
R6
N1
N2
N3
N4
N5
N6
A1
A2
A3
A4
A5
A6
A7
A8
A9
A10
A12
A13
A14
A15
A16
ε3
A11
Receptor (Direct Molecular Interaction)
Intermediate Process
Assay
ER agonist pathway
Interference pathway
Noise Process
ER antagonist pathway
R2
N7
ER Receptor Binding
(Antagonist)
A17
A18
Dimerization
N8
N9DNA Binding
CofactorRecruitment
N10AntagonistTranscriptionSuppression
R4
R9
Computational Modeling of Estrogen Receptor Pathway
Accuracy 0.93 (0.95)
Sensitivity 0.93 (0.93)
Specificity 0.92 (1.0)
Accuracy 0.86 (0.95)
Sensitivity 0.97 (0.97)
Specificity 0.67 (0.89)
Che
mic
als
Assays cluster by technology,suggesting technology-specific non-ER bioactivity
Assays
In Vitro Reference Chemicals* In Vivo Reference Chemicals*
*Values in parentheses exclude inconclusive chemicalsJudson et al., Tox Sci. 2015Browne et al., ES&T. 2015
18 In vitro HTS assays for ER bioactivity
Jundson, Unpublished
National Center forComputational Toxicology
Complex Systems Models to Predict Phenotypic Responses to Chemicals
Leung et al., Repro Toxicol, 2016
Signaling Network Underlying Virtual Genital Tubercle Model (Mouse)
Cell field - androgenSHH field
FGF10 field no androgen
Simulation of Genital Tubercle Closure
Embryonic GT Abstracted GT
GD13.5 – 17.5
National Center forComputational Toxicology
27
What Traits Have Been Under Selection?
• Increased throughput and biological coverage
• Interrogation of effects at the molecular and pathway level
• Increasingly relevant test systems
• Putting results in a dose/exposure context
• Characterization of uncertainty
• Computational modeling to integrate experimental data
• Delivery of data and models through decision support tools
• Transparency and performance-based validation
National Center forComputational Toxicology
28
Significant Pressure to Deliver Data and Models in Useful Format
Initial Data Delivered as Flat Files
Progress to Dashboard with Limited Search,
Visualization, and Export Functionality
Currently Providing Cross-Functional and Decision
Support Dashboards
National Center forComputational Toxicology
29
New Chemistry Dashboard Delivers Structural and Property Data
https://comptox.epa.gov/dashboard
National Center forComputational Toxicology
30
RapidTox Decision Support Dashboard in Development
• Semi-automated decision support tool with dashboard interface for high-throughput risk assessments
• Combining diverse data streams into quantitative toxicity values with associated uncertainty
National Center forComputational Toxicology
31
Regulatory Applications Require More Focus on Quality and Transparency
• Public release of Tox21 and ToxCast data on PubChem and EPA web site (raw and processed data)
• Transparent ToxCast data analysis pipeline• Data quality flags to indicate concerns with chemical
purity and identity, noisy data, and systematic assay errors
• Publicly available as an R package
• Tox21 and ToxCast chemical libraries have undergone analytical QC and results publicly available
• Public posting of ToxCast procedures• Chemical Procurement and QC• Data Analysis • Assay Characteristics and Performance
• External audit on ToxCast data and data analysis pipeline
FLAGS:Only one conc above baseline, activeBorderline active
National Center forComputational Toxicology
32
Next Phase… Evolution Towards a Truly Predictive Science
c. 1970s - 2002 2003 - 2006 2007 - 2010 2011 - 2016c.1500 – 1960s
Comp Chem, HTTK, HTS, Systems Tox, HT Exposure
2016 - ?
National Center forComputational Toxicology
33
Acknowledgements and Questions
Tox21 Colleagues:NTP CrewFDA CollaboratorsNCATS Collaborators
EPA Colleagues:NERLNHEERLNCEA
EPA’s National Center for Computational Toxicology
top related