high throughput exposure prediction for the expocast project
TRANSCRIPT
John Wambaugh U.S. EPA, National Center for Computational Toxicology
High Throughput Exposure Prediction for the ExpoCast Project
Office of Research and Development
August 23, 2012
The views expressed in this presentation are those of the author and do not necessarily reflect the views or policies of the U.S. EPA
Office of Research and Development 1
Introduction
There are thousands of environmental chemicals, many without enough data for evaluation Risk is the product of hazard and exposure High throughput in vitro methods beginning to bear fruit on hazard for many of these chemicals Methods exist for approximately converting these in vitro results to daily doses needed to produce similar levels in a human (Wetmore et al. (2011)) Without a similar capacity for exposure cannot place risk early into prioritizations
Judson et al., (2011) “Estimating Toxicity-Related Biological Pathway Altering Doses for High-throughput Chemical Risk Assessment” Chemical Research in Toxicology 24 451-462
Strategic Guidance: A Framework for a Computational Toxicology Research Program (November 2003)
Source-to-Outcome Continuum
2 Office of Research and Development
Fent
in H
ydro
xide
Clop
rop
Qui
noxy
fen
Spiro
xam
ine
Endo
sulfa
nIp
rodi
one
Nicl
osam
ide
Halo
sulfu
ron-
met
hyl
Cypr
odin
ilPr
omet
onEm
amec
tin b
enzo
ate
Acifl
uorfe
nPa
rath
ion
Etox
azol
eFe
noxy
carb
Lind
ane
Etha
lflur
alin
2,4-
DBTr
i-alla
teFe
nbuc
onaz
ole
Chlo
rpyr
ifos-
met
hyl
Bens
ulfu
ron-
met
hyl
Chlo
reth
oxyf
osIs
oxab
enDi
clof
op-m
ethy
lPr
opet
amph
osTr
iflox
ysul
furo
n-so
dium
Indo
xaca
rbDi
chlo
rpro
pM
CPA
Bent
azon
eQ
uinc
lora
cDi
cofo
lPr
osul
furo
nIo
dosu
lfuro
n-m
ethy
l-sod
ium
Pyrit
hiob
ac-s
odiu
mEs
fenv
aler
ate
2,4-
DDi
chlo
ran
Imaz
alil
Clof
ente
zine
Prod
iam
ine
PFO
SNa
prop
amid
eBi
fent
hrin
Prom
etry
nDi
nico
nazo
leTh
idia
zuro
nPi
clor
amFi
pron
ilPr
opaz
ine
Nitra
pyrin
Tebu
fenp
yrad
Rote
none
Brom
acil
Fena
rimol
Dicl
osul
amPi
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
100000O
ral E
quiv
alen
t Dos
e or
Est
imat
ed E
xpos
ure
(mg/
kg/d
ay)
Cyprodinil
Prometon Fenbuconazole
Dichloran Quinclorac
Acifluorfen
Spiroxamine
Isoxaben
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
Dife
nzoq
uat m
ethy
l sul
fate
Dic
roto
phos
Aba
mec
tinM
olin
ate
Ben
sulid
eD
imet
hom
orph
6-D
esis
opro
pyla
traz
ine
Oxy
tetr
acyc
line
dihy
drat
eTh
iam
etho
xam
Imaz
etha
pyr
Clo
thia
nidi
nD
iphe
nyla
min
eFe
noxa
prop
-eth
ylA
zoxy
stro
bin
Ory
zalin
Clo
maz
one
Seth
oxyd
imM
etrib
uzin
Die
thyl
tolu
amid
e
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
)
Triclosan
2-phenylphenol
Chl
oron
ebP
aclo
butr
azol
Pyr
ipro
xyfe
nM
esot
rion
eD
icam
baO
xyfl
uorf
enC
acod
ylic
aci
dA
ceph
ate
Chl
orid
azon
Linu
ron
Imaz
amox
Met
hida
thio
nC
arbo
xin
Imaz
apic
Pir
imic
arb
Oxa
myl
Eth
opro
pC
yana
zine
Fena
mip
hos
Tria
dim
efon
Cyc
lani
lide
Thia
clop
rid
Flur
oxyp
yr-m
epty
lP
heno
xyet
hano
lB
isph
enol
-ATe
tram
ethr
inM
etal
axyl
Die
thyh
exyl
pht
hala
teFe
nhex
amid
Icar
idin
Trif
lusu
lfur
onP
ropa
moc
arb
HC
lB
endi
ocar
bTe
buth
iuro
nV
incl
ozol
inTr
ibuf
osB
ifen
azat
eIm
idac
lopr
idM
etol
achl
orE
thof
umes
ate
Thio
phan
ate
met
hyl
Fluf
enac
etD
azom
etB
rom
oxyn
ilS
ulfe
ntra
zone
Dim
ethe
nam
idS
-Bio
alle
thri
nC
arba
ryl
Ala
chlo
rH
exaz
inon
eA
zinp
hos-
met
hyl
Ace
toch
lor
d-ci
s,tr
ans-
Alle
thri
nP
ymet
rozi
neFo
rmet
anat
e H
Cl
Flua
zifo
p-P
-but
ylD
ibut
yl p
htha
late
Dim
ethy
l pht
hala
teD
iazo
xon
0.00001
0.0001
0.001
0.01
0.1
1
10
100
1000
10000
100000
Ora
l Equ
ival
ent
Dos
e or
Est
imat
ed E
xpos
ure
(mg/
kg/d
ay)
Dicamba
Fenhexamid Fluroxypyr-meptyl
Wetmore et al. Tox. Sci (2011)
Oral Doses Equivalent to ToxCast Concentrations
3 Office of Research and Development
Ora
l Equ
ival
ent D
ose
and
Est
imat
ed E
xpos
ure
(mg/
kg/d
ay)
High Throughput Exposure Prioritization
Proof of Concept: Using off-the-shelf models capable of quantitatively predicting exposure determinants in a high throughput (1000s of chemicals) manner To date have found only fate and transport models to have sufficient throughput These models predict the contribution from manufacture and industrial use to overall exposure rapidly and efficiently
Applying and developing new high throughput models of consumer use and indoor exposure
Goal: A high-throughput exposure approach to use with the ToxCast chemical hazard identification.
4 Office of Research and Development
Environmental Fate and Transport
Consumer Use and Indoor Exposure
Framework for High Throughput Exposure Screening
5 Office of Research and Development
Estimate Uncertainty
Space of Chemicals (e.g. ToxCast, EDSP21)
(Bio) Monitoring
Infe
rred
Exp
osur
e
Model 1
Model 2 …
Calibrate models
Apply calibration and uncertainty to other chemicals
Evaluate Model Performance
Joint Regression on Models
Exposure Inference
Dataset 1
Dataset 2 …
USEtox RAIDAR
Off the Shelf Models
Treat different models like related high-throughput assays
6 Office of Research and Development
United Nations Environment Program and Society for Environmental
Toxicology and Chemistry toxicity model Version 1.01
Rosenbaum et al. 2008
Risk Assessment IDentification And Ranking
model Version 2.0 Arnot et al. 2006
Model parameters mostly predicted from structure (SMILES)
Parameterizing the Models Variable Description Unit Source Default QSAR USEtox RAIDAR
Chemical Name ToxCast Yes Yes CAS ToxCast Yes Yes MW Molecular Weight g/mol ToxCast Yes Yes Data
Temperature Degrees C 25 Yes
KOW Octanol:Water Partition
Coefficient 1 Episuite Yes Yes Yes
Koc Organic Carbon:Water
Partition Coefficient L/kg USEtox QSAR Yes Yes
KH25C Henry's Law Coefficient
(25 degrees C) Pa*M^3/mol Episuite Yes Yes Yes
Pvap25 Vapor Pressure (25
degrees C) Pa Episuite Yes Yes Yes
Sol25 Solubility (25 degrees C) mg/L Episuite Yes Yes Yes
KDOC
Dissolved Organic Carbon:Water Partition
Coefficient L/kg USEtox QSAR
Yes Yes
kdegA Degredation Rate in Air 1/s Episuite Yes
Yes Yes
kdegW Degredation Rate in
Water 1/s Episuite Yes
Yes Yes
kdegSd Degredation Rate in
Sediment 1/s Episuite Yes
Yes Yes
kdegSl Degredation Rate in Soil 1/s Episuite Yes
Yes Yes
kdegbiota Degredation Rate in
Biota 1/s 2.40E+12 Yes
kdeghuman Degredation Rate in
Humans 1/s 2.40E+12 Yes
pKa Acid Dissociation
Constant 1 QikProp Yes
Yes
BAF Bioaccumulation Factor L/kg EpiSuite Yes
Yes
LC50 Average Log Aquatic 50%
Lethal Concentration Log(mg/L) EcoSAR Yes Yes
Cl/C(Cl)=C/C3C(C(=O)OCc2cccc(Oc1ccccc1)c2)C3(C)C
7 Office of Research and Development
Chemical Landscape Range of physico-chemical properties for the 1600 chemicals evaluated Principal component one: half-life in soil, water, and sediment Principal component two: octonol-water partition coefficient (logP) Principal component three: half-life in air Larger spheres are those for which NHANES data was available
8 Office of Research and Development
Urban Air Rural Air
Freshwater
Sea water Natural soil
Agricultural soil
Soil Water
Air
Models predict fate depending upon release profile (Level III Fugacity Model) Release profile can be chemical-specific, class-specific, or default depending on data Estimated behavior/consumption can in turn yield human and ecological prediction
USEtox RAIDAR
Partitioning Release into the Environment
9 Office of Research and Development
If we have the data then we would use it, but we don’t Assuming an “average” release profile
Urban Air Rural Air
Freshwater
Sea water Natural soil
Agricultural soil
TSCA / Industrial
Urban Air Rural Air
Freshwater
Sea water Natural soil
Agricultural soil
Food-use Pesticide
Partitioning Release into the Environment
Soil Water
Air
Soil Water
Air
USEtox
RAIDAR
10 Office of Research and Development
Production Volume is an Overall Multiplier
Using EPA Toxic Substances Control Act (TSCA) Chemical Data Reporting (CDR) Rule (Formerly Inventory Update Reporting – IUR) data for production volumes Crop Protection Research Institute has data on many pesticides (which are heavily favored in ToxCast Phase I) although the data is old (2002)
Urban Air Rural Air
Freshwater
Sea water Natural soil
Agricultural soil
Urban Air Rural Air
Freshwater
Sea water Natural soil
Agricultural soil
Production Volume (kg/day)
100,000,000
Production (kg/day)
Release Partitioning (fractional)
11 Office of Research and Development
Clustering 1678 chemicals by the media into which they partition most Could infer behavior of understudied chemicals from similar, well-known counterparts – “fate read-across” Fate predictions not terribly consistent
High Throughput Fate Predictions
12 Office of Research and Development
Population Exposure from Environmental Media
Likely Bioaccumulator Log Kow < -1
13 Office of Research and Development
Both models assume exposure scenarios that relate environmental media to food and inhalation exposure Can calculate intake fraction (population exposure in kg per kg emitted) General agreement for most chemicals – putative bioaccumulators predicted to be highest Issue with accumulation in plants causes larger predictions for RAIDAR in some cases
McKone et al. (2007)
Cowan-Ellsbury et al. (2009) PBDE99
Chlorpyrifos
Diazinon
Dimethoate, Malathion, Oxydemeton-methyl
o Measured x Model Predicted
Literature Ground-truthing Efforts
14 Office of Research and Development
Data Availability for Model Predictions and Ground-truthing
Chemicals Current Models can Handle (1678)
Production / Release Data IUR (6759 compounds with production of >25,000 lbs a year) CPRI (242 pesticides with total lbs applied) NHANES
“Ground-truthing” Chemicals
Chemicals of Interest (2127)
51 33
volatile, insoluble
Ground-truth with CDC NHANES urine data Focusing on U.S. median initially Capable of adding population variability, but will need consumer use models
15 Office of Research and Development
Linking NHANES Urine Data and Exposure
( ) ( )∑=i i
i
MWmg/kg/dayMWmg/kg/day 00
Parent
Products
1
2
3
4
( )day
g*gmg
kg 701mg/kg/day creatine
creatine
ii =
Lakind and Naiman (2008)
Steady-state assumption
Office of Research and Development 16
17
Mapping of NHANES Parents and Urine Products
Office of Research and Development
Parent Compound Product Monitored In Urine
Degeneracy of an Exposure Biomarker
Unknowns: 2 (mass, bioavailability)
Knowns: 1 (mass)
Unknowns: 2
Knowns: 3 (mass)
Unknowns: 3 (fractions), Knowns: 1 (mass balance)
Unknowns: 6
Knowns: 1 (mass)
Unknowns: 3 (fractions), Knowns: 3 (mass balance)
parent
metabolite
18 Office of Research and Development
19 Office of Research and Development
Bayesian Model for fij
The real situation may be even more complicated Further complicated by limit of detection of NHANES chemicals – many chemicals that are checked for are below the LoD
However, we still can predict that N parent exposure are related to P=f*N urine products, and many fij are zero, Use MCMC to explore range of possible parent predictions Also incorporate uncertainty in production volume and use all quantiles of NHANES data
( ) ( )∑=i i
iijjj f
MWmg/kg/dayMWmg/kg/day
Unknown fraction fij for each urine product j
due to parent i:
Framework for High Throughput Exposure Screening
20 Office of Research and Development
Estimate Uncertainty
Space of Chemicals (e.g. ToxCast, EDSP21)
(Bio) Monitoring
Infe
rred
Exp
osur
e
Model 1
Model 2 …
Calibrate models
Apply calibration and uncertainty to other chemicals
Evaluate Model Performance
Joint Regression on Models
Exposure Inference
Dataset 1
Dataset 2 …
21
Calibrate ExpoCast Predictions to CDC NHANES Data
Comparison between model predictions and
biomonitoring data indicates correlation
Indoor/consumer use is
critical: Compounds with near-field applications more
than 100x greater
Office of Research and Development
)log()log(*~ 3221 vrmvumNbbY +++
22
Use Categories from ACToR
The sources for various chemical data were assigned to various use categories. Chemicals from multiple sources were assigned to multiple categories. Four categories – personal care products, consumer use, fragrance, and food additive – were aggregated into a single “near field” indicator variable.
Office of Research and Development
)log()log(*~ 3221 vrmvumNbbY +++
Highest Priority Uncertainty of prediction indicated by
the horizontal confidence interval from the empirical calibration
to the NHANES data
Horizontal dotted line indicates the fiftieth
percentile rank and the vertical dotted line indicates the cutoff
between overlapping top-half and lower half
confidence intervals
23 Office of Research and Development
Exposure Prioritization from ExpoCast
24
Wetmore et al. (2011) ToxCast Oral Equivalents
Lines indicate ExpoCast prediction 95% CI indoor/consumer use in red, blue otherwise
Office of Research and Development
25
Model Appropriateness
Statistical Near Field Model
Further investigating near field use determinants using expanded information
25 Office of Research and Development
26
Conclusions
• ExpoCast can use environmental fate and transport models to make high-throughput exposure predictions
• These prioritizations have been compared with CDC NHANES ground truth
• This biomonitoring data gives empirical calibration and estimate of uncertainty
• Indoor/consumer use is a primary determinant
• Next steps:
• HT models for exposure from consumer use and indoor environment
• Use and evaluate these models as additional HT exposure assays
• ORISE Postdoc position for high throughput modeling of nearfield indirect exposure (e.g. flooring, furniture)
Office of Research and Development
NERL Peter Eghehy Kathie Dionisio Dan Vallero
University of Toronto at Scarborough Jon Arnot University of Michigan Olivier Jolliet USDA FSIS Jade Mitchell-Blackwood
NCCT Elaine Cohen-Hubal David Dix Alicia Frame Sumit Gangwal Richard Judson Robert Kavlock Thomas Knudsen Stephen Little Shad Mosher James Rabinowitz David Reif Woody Setzer Amber Wang
U.S. E.P.A. Office of Research and Development ExpoCast Team
The views expressed in this presentation are those of the author and do not necessarily reflect the views or policies of the U.S. EPA