US EPA CSS-HERA Board of Scientific Counselors Chemical Safety Subcommittee Meeting February 2-5, 2021 CSS Session 1A Slides


Note: this document may contain some
elements that are not fully accessible to
users with disabilities. If you need
assistance accessing any
information in this document, please
contact ORD_Webmaster@epa.gov.

US EPA CSS-HERA
Board of
Scientific
Counselors
Chemical Safety
Subcommittee
Meeting

US EPA CSS-HERA BOSC Meeting - February 2-5, 2021

SW(

$

<

3J

\

SB



ro

z

111

p*

PR0^°

The work presented within represents US EPA Office of Research and Development research
activities. Material includes both peer reviewed, published results and work-in-progress
research. Please do not cite or quote slides.

-------
Table of Contents

High-Throughput Phenotypic Profiling (Joshua Harrill)	3

High-Throughput Transcriptomics (Logan Everett)	26

Retrofitting in vitro Systems with Metabolic Competence (Chad Deisenroth)	47

The work presented within represents US EPA Office of Research and Development research activities. Material
includes both peer reviewed, published results and work-in-progress research. Please do not cite or quote slides.

-------
oEPA

High-Throughput Phenotypic Profiling

Joshua A. Harrill, Ph.D.

Rapid Assay Development Branch
Biomolecular and Computational Toxicology Division
Center for Computational Toxicology and Exposure
Office of Research and Development, U.S. EPA

-------
oEPA

Disclaimer

The views expressed	in this pres

authors	and do not necessarily

of the U.S. Environmental Protection Agency.
product	names	do not constitute

-------
oEPA

Research Drivers

•	There are many chemicals in U.S. commerce with the potential to enter the environment that are poorly
characterized in terms of human health hazards.

•	Traditional toxicity testing approaches in laboratory animals are expensive and time-consuming and therefore
cannot be used to efficiently address this large data gap.

•	Animal-free New Approach Methods (NAMs) provide a means for accelerating the pace of chemical hazard
assessment using models anchored in human biology.

•	EPA has been tasked with and is committed to reducing the use of animals in toxicity testing and expanding
the use of NAMs in chemical risk assessment

•	( jne 16) Frank R. Lautenberg Chemical Safety for the 21st Century Act (15 U.S.C. §2601)

•	( jne '18) US EPA Strategic Plan to Promote the Development and Implementation of Alternative Test
Methods within the TSCA Program (EPA-740-R1-8004).

•	(	) Administrator's Directive to Prioritize Efforts to Reduce Animal Testing (Wheeler 2019)

•	( jne 20) US EPA New Approach Methods Work Plan (EPA 615B2000)

-------
oEPA

NAMs-Based, Tiered Hazard Evaluation Strategy

New Approach Methodologies (NAMs) are any
technology, methodology, approach or combination
thereof that can be used to provide information on
chemical hazard and risk that avoids the use of
intact animals.

US EPA CompTox Blueprint advocates the use of

high throughput profiling (HTP) assays as the first
tier in a NAMs-based hazard evaluation strategy.

HTP assay criteria:

1.	Yield bioactivity profiles that can be used for

potency estimation, mechanistic prediction
and evaluation of chemical similarity.

2.	Compatible with multiple human-derived
culture models.

3.	Concentration-response screening mode.



f \



Tier 1

Chemical Structure

¥ Broad Coverage, \

Multiple cell types

and Properties

' u J

I High Content Assay(s) J

+/- metabolic competence



1

1	





No Defined Biological
Target or Pathway

r

DefinedBiologicalTarget
or Pathway

Select In Vitro
Assays

}

Tier 2

Orthogonal confirmation

r

ExistingAOP

I

In Vitro
Assays for other KEs
and Systems Modeling

Tier 3

NoAOP

Organotypic Assays and
Microphysiological
Systems

Identify Likely Tissue,
Organ, or Organism Effect
and Susceptible Populations

Estimate Point-of-Departure Estimate Point-of-Departure Estimate Point-of-Departure
Based on Biological Pathway or	Based on AOP	Based on Likely Tissue-or

CellularPhenotype Perturbation	Organ-level EffectwithoutAOP

The NexGen Blueprint of CompTox at USEPA, Tox. Sci. 2019; 169(2):317-322

-------
a rnA	Imaging-Based High-Throughput Phenotypic Profiling

VfrtrA	(HTPP)

Healthy and	High-throughput

diseased patient	staining and imaging:

cell lines	e.g. Cell Painting assay

Drugs or genetic

perturbations

(optional)

%£}_
©o#-

®o'8o

EK

Nucleolus

iS)

Nucleus	=¦

F-actin	U

Mitochondrion
Cytoplasmic RNA
Golgi apparatus
Plasma membrane

Image analysis and
feature extraction

Features

Morphological profiles

Cluster 1

Downstream analysis:
mapping relationships

Chondrosekoron et ol. Not Rev Drug Discov. 2020 Dec 22:1-15

•	A high-throughput testing strategy where rich information present in biological images is reduced to
multidimensional numeric profiles and mined for information characteristic to a chemical's biological activity.

•	Originated in the pharmaceutical sector and has been used in drug development to understand disease
mechanisms and predict chemical activity, toxicity and/or mechanism-of-action

-------
oEPA

HIPP with the Cell Painting Assay

Cell Painting is a profiling method that
measures a large variety of phenotypic
features in fluoroprobe labeled cells in vitro.

•	High-throughput

•	Cost-effective (0/well)

•	Scalable

•	Reproducible

•	Amenable to lab automation

•	Deployable across multiple human-
derived cell types.

•	Infrastructure investment

•	High volume data management

Laboratory & bioinformatics workflows for
conduct of this assay have been established
at CCTE.

OPEN 3 ACCESS Frwdy available online	SPLOSH

Multiplex Cytological Profiling Assay to Measure Diverse
Cellular States

Sigrun M. Gustafsdottir% Vebjorn Ljosa*, Katherine L. Sokolnicki™ J. Anthony Wilson°b, Deepika
Walpita, Melissa M. Kemp, Kathleen Petri Seiler00, Hyman A. CarreIDd, Todd R. Golub, Stuart L. Schreiber,
Paul A. demons*1, Anne E. Carpenter'11, Alykhan F. ShamjP

Broad Institute of Harvard and MIT, Cambridge, Massachusetts, United States of America

Golgi + membrane

DNA	RNA+ER	+ actin skeleton	Mitochondria

VntetvM

1300 features

\0

ca\W^°n

te»Ure 6

-------
xv EPA

Imaging & Phenotypic Feature Extraction

NUCLEUS

RING

5 Compartments

CYTOPLASM MEMBRANE

CELL



f

fr

'

n

V'



bO

O „

Radial distribution

Axial

Intensity

sl*lw









Bright

Dark

Compactness

Shape

PerkinElmer Opera Phenix
Modality: Confocal (single z)
Objective: 20X Water
Plate:	CellCarrier-384 Ultra

Fields: 5 or 9

With illustrations from Per kin Elmer

Profile

DMA

RNA

ER

AGP

Mito

49 Feature Categories

(ex. MITO_Texture_Cytoplasm)

\ I

1300 features / cell

Not associated
with a channel

Module

Position
[7]

Nuclei
Cell

Basic
morph-
ology [5]

Nuclei
Cell

SCARP morphology

Symmetry
[80]

Nuclei

Nuclei

Cell

Cell

Cell

Compactness
[40]

Nuclei

Nuclei

Cell

Cell

Cell

Axial
[20]

Nuclei

Nuclei

Cell

Cell

Cell

Radial
[28]

Nuclei
Cell

Nuclei

Cell

Cell

Cell

Profile
[20-30]

Nuclei
Cytoplasm

Nuclei

Cytoplasm

Nuclei
Cytoplasm

Nuclei
Cytoplasm

Intensity

[9]

Nuclei

Nuclei

Ring
Cytoplasm

Ring
Cytoplasm
Membrane

Ring
Cytoplasm

Texture
[14]

Nuclei

Nuclei

Ring
Cytoplasm

Ring
Cytoplasm
Membrane

Ring

Cytoplasm

-------
oEPA

Examples of Chemical Induced Phenotypes

Solvent control (0 5% DMSO) Berbenne chlonde (10 pM)

Solvent control (0 5% DM SO) Ca-074-Me (1 pM)

¦ .>

9 ***

i s v _









§«*

"¦m 4*

mS

DNA

fc fl



| •>

_ ft



V

__ ^

Solvent control (0 5% DMSO) Rapamycin (100 pM)

z

a:

<
z

Q

Mitochondrial Compactness

Golgi Texture

RNA

: r



• *

¦ W 2





~
*

Cell Swelling
AGP

Cell Compaction

Mito

o

¦II

¦a

ui

o

£
UJ

- J



if

.. 1-

	

1

III

P f Mr it m

infl1

1

9

m

Strong phenotypes are observed qualitatively and produce distinct profiles when measured quantitatively.

8

Adapted from Nyffeler etal. Toxicol Appl Pharmacol. 2020 Jan 15;389:114876

Solvent control (0 5% DMSO) Etoposide (3 pM)

-------
oEPA

HIPP Data Analysis Pipeline

Data reduction

cell-level data

Normalization
MAD normalization

cell value - median

DMSO

1.4826 MAD

DMSO

normalized

cell-level data

Aggregation
median

well-level data

Standardization

Z transformation

Cell Count Info

Cone. > 50% cell loss

scaled
well-level data

0



-20-15-10 -5 0 5 10 15 20

Concentration Response Modeling

clipped
well-level data

Calculate Response
Metrics

1300
"¦260

T

MMJIM





!Ur.fc

ll

<* - i—' —i



1 111

ZZ1I " "». " 1



v

Best Model
Selection

if

BMC

See Nyffeler et a I. SLAS Discov. 2020 Aug
29: doi: 10.1177/24 72555220950245

Berberine chloride

Mito_Cells_Morph_STAR

-------
oEPA

Pheriotype Altering Concentration (PACs)

Mahalanobis Distance (DM):

•	A multivariate distance metric that measures the distance between a point (vector) and a distribution.

•	Accounts for unpredictable changes in cell states across test concentrations and inherent correlations in profiling data.

Global Mahalanobis



y

derive a Mahalanobis distance
(relative to control wells)

1300 features







Feature-level
fitting

group them in
49 categories

—~

derive a Mahalanobis distance
(relative to control wells)

1 BMC

49 BMCs

BPAC

Category-level Mahalanobis

concentration (mM)

•	Chemicals where a BMC can be determined using either the global or category DM approach are considered active.

•	The minimum of the global or most sensitive category BMC is the Phenotype Altering Concentration (PAC)

10

-------
xv EPA

v

Concentration-Response Modeling Example

all-trans-Retinoic acid

DTXSID7021239 | 302-79-4 | RA

0.001	0.01

concentration (\iM)

I J

4

l 1



' ! f 1

T





o

« 10 -

¦i	*

15-

£¦
o

o 10

k-

a;

>
o
a.
o

>

o

03
O

E

LU

fp

J

10-

o

3

o

a)
>
o

Q.
O

>»
O
03
O

£
LU

5-

0-

-5-

-10

0.01

concentration (ijM)

0.001	0.010	0.100

Benchmark concentration [jjM]

1.000

2020-07-27

i

0.001	0.010	0.100	1.000

Benchmark concentration [jjM]

Shape / Position
0 DNA
RNA
• ER

AGP
Q M to

Cell / Cytoplasm

Nuclei

Ring

Membrane

I	Intensity

T	Texture

M	Morphology

S	Symmetry

C	Comptadness

A	Axial

R	Radial

P	Profile

•	Phenotypic effects can be observed below the threshold for cytotoxicity and in the absence of cytostatic effects.

•	Category and feature-level modeling can reveal which organelles exhibit treatment-related changes in morphology.

-------
oEPA

ToxCast Chemical Screen - Experimental Design (1)

Parameter

Multiplier

Notes

Cell Type(s)

1

U-2 OS

Culture Condition

1

DMEM + 10% HI-FBS

Chemicals

1,202

Selected from US EPA ToxCast chemical collection

Includes 179 chemicals with annotated molecular targets
Includes 462 APCRA case study chemicals

Time Points:

1

24 hours

Assay Formats:

2

High Throughput Phenotypic Profiling (Cell Painting)

High Throughput Transcriptomics (TempO-Seq)

Concentrations:

8

3.5 log10 units; ~half-log10 spacing

Biological Replicates:

4

--

Accelerating the Pace of Chemical Risk Assessment

APCRA

Kavlock et a!. (2018)
Chem. Res. Tox; 31(5): 287-290

International collaboration of regulatory scientists focused on next generation chemical risk
assessment including deriving quantitative estimates of risk based on NAM-derived potency
information and computational exposure estimates.

PK parameters necessary for in vitro to in vivo extrapolation (IVIVE)
in vivo toxicity data

APCRA Chemicals

o

12

-------
oEPA

ToxCast Chemical Screen - Experimental Design (2)

Label

Reference Chemicals:

Molecular Mechanism-of-Action

Test Concentrations

A

Etoposide

DNA topoisomerase inhibitor

0.03 -10 nM

B

all-trans-Retinoic Acid

Retinoic acid receptor agonist

0.0003 -1 |iM

C

Dexamethasone

Glucocorticoid receptor agonist

0.001-3 nM

D

Trichostatin A

Histone deacetylase inhibitor

1 |iM

E

Staurosporine

Cytotoxicity control

1 |iM

F

DMSO

Vehicle control

0.5%

-------
oEPA

Assay Performance / Reproducibility

29 J

1 "T

c/>

§¦ 10 i

o

k.

O) 20 -

(1)

+-»

J2 29 J

f- 1f

TO
D

T3 10 -
>

^ 20 -

29 -
1 "

—



i

.¦

P

1 f

fl |

"

J
¦

' T

Global Mahalanobis: PAC of reference chemicals

Oragenelle/
Channel



Shape
Positio



DNA



RNA



ER



AGP



Mito

normalized
value
(Level 5)

¦

20
10

¦

o o
5 V

DEX-

as
o

E

0)

-C

o

0) ETOP

o

c

CD

CD

4—

(D

a:

RA

8^

o

o (

,uc

o

^ (ft

cDoC

O o
%

cP

1300 features (ordered by organelle/channel)

-2	-1

PAC log10(uM)

Reference chemicals produce reproducible and distinct profiles.
Reference chemicals produce reproducible potency estimates (PACs).

14

-------
oEPA

ToxCast Chemical Screening Results

inactive

active

50-

40 H

o
fz

>30

Co
(/)

« 20
O
x
o

10-

n =429

inactive -

Q
O

-------
oEPA

In Vitro to In Vivo Extrapolation (I VIVE;

Predicted exposure

Mew approach methodologies (NAMs)

in vivo point-of-departure

Exposure predictions

(EPA ExpoCast)

• Systematic Empirical Evaluation
of Models (SEEM) version 3

• inferred from human
biomonitoring data, production
volume and use categories
(industrial / consumer use)

Toxicological
threshold of
concern
(TTC)

Toxcast BPAC
(M-M)

HTPP BPAC
(|iM)

In vitro-to-in vivo
extrapolation (IVIVE)
high-throughput toxicokinetics (httk)

Toxcast AED
(mg/kg bw/day)

HTPP AED

(mg/kg bw/day)

Database of in vivo effect values (EPA
- ToxValDB)

• Mammalian species

• oral exposures

• Various study types

• NOEL, LOEL, NOAEL, LOAEL.

• mg/kg/day

j-i

5%
~

5% 50% 95%
i 0 1

5% 50% 95%
I O 1

95%
*

POD: point-of-departure

AED: administered equivalent dose

-------
xv EPA

Bioactivity to Exposure Ratio (BER) Analysis

Bioactivity exposure ratio (BER) =

lower bound of HTPP bioactivity
upper bound of exposure estimate

= logio

/HTPP AED 5th*

SEEM3 95th

80-

60-

o 40

-------
oEPA

Contextual Response of Nuclear Receptor Modulators

Comparison to ToxCast potencies

Profile Similarity

17-Methyitestosterone
2,2-Bis(4-hydroxyphenyl)-1,1,1-tricnloroethane **
4-Androstene-3,17-dione
Bisphenol A "
Cyproterone acetate
Danazol
Flutamide
Linuron

Testosterone propionate
Lithochollc acid
1,4-Bis[2-{3,5-dichloropyridvloxy)lbenzene
17aIpna-Ethinylestradiol
17beta-Estradiol

2,2-Bis(4-hydroxyphenvl)-1.1,1-trichloroethane "*
4-(1,1.3,3-Tetramethylbutyl)phenol
4-Nonyiphenol
Benzyl butyl pmhalate
Bisphenol A **
Butylparaben
Diethvlstilbestrol
Endosulfan
Genistein
o,p-DDE

target

AR
BAR

CAR

ESR

PGR
PPAR
PXR

RAR
VDR

Zearalenone
Betamethasone
Budesonide
Dexamethasone
Fluorometholone
Methylprednisolone
Prednisolone
Triamcinolone
Medroxyprogesterone acetate
Mifepristone
Norethindrone
Progesterone
Bisphenol A dialycidyl ether
Clofibric acid
Fenofibrate
GW0742*
Indomethacin
L-165041*
Perfluorooctanoic acid
Pirinixic acid
T roglitazone
Ketoconazole
Pregnenolone carbonitrile
Ritampicin
all-trans-Retinoic acid
AM580
Arotinoid acid
Bexarotene
Ergocalciferol
Vitamin D3

Profile
similarity
i— ¦ 1

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2

-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1

Gene expression in U-2 OS

AR
BAR
CAR
ESR1
ESR2
ESRRB
ESRRG
GR
PGR
PPARA^
PPARD
PPARGi
PXR
RARA
RARB
RARGi
RXRA
RXRB-
RXRG-
VDR

~;
Z]

4 8

Gene expression [NX]

For three receptor systems that are expressed in U-2 OS cells (GR, RAR/RXR, VDR) potencies were comparable with ToxCast.
Phenotypic profiles for chemicals that affect these receptor systems are similar.

-4 -3-2-10 1 2 inactive
HTPP PAC log10 (|jM)

-------
oEPA

Structurally Similar Environmental Chemicals Can
Produce Similar HTPP Profiles

Endrin (LAB-000007)

Endosulfan (LAB-000013)

tested range

0.1 1 10 100

Benchmark Concentration [pM]

Dieldrin (LAB-000016)

m %

tested range

0.1 1 10 100

Benchmark Concentration [pM]

E o

TJ
O

£ -2

£ o

-1

tested range

—r~

0.1

—i—

1 10 100

Benchmark Concentration [pM]

tested range

0.1 1 10 100

Benchmark Concentration [|jM]

? 2

o
c

Heptachlor (LAB-000011)

tested range

0.1 1 10 100

Benchmark Concentration [pM]

u~i

m
E

Shape / Position

DNA

Intensity

RNA

Texture

Morphology

*
m

AGP
Mito

S
c

Symmetry
Comptactness

•

Cell / Cytoplasm

Axial

Nuclei

Radial

¦O

~
A

Ring

Membrane

Profile

Preliminary results. Do riot cite or quote.

-------
oEPA

HIPP is Compatible with Biologically Diverse Cell Lines

Preliminary results. Do not cite or quote.

TIME

Microvascular Endothelium

HTPP is compatible with
many human-derived cell
culture models.

Enables characterization
of chemical effects across
different domains of
human biology.

-------
£EPA V Summary

• Assay Reproducibility: Demonstrated high assay reproducibility through the use of phenotypic
reference chemicals and developed experimental designs that allow for evaluation of assay
performance throughout large-scale screening campaigns.

• Potency Estimation: Developed a concentration-response modeling workflow to identify
concentration thresholds for perturbation of cell morphology (e.g. phenotypic altering
concentration, PAC).

• Mechanistic Prediction: Chemicals with strong and specific target mode associations can
produce similar phenotypic profiles in U-2 OS cells. Strength of similarity varies according to
baseline target expression.

• Chemical Similarity: Chemicals with similar chemical structures can also produce similar
phenotypic profiles in U-2 OS cells.

• Bioactivity to Exposure Ratio: Phenotype altering concentrations (PACs) can be converted to
administered equivalent doses (AEDs) and compared to human exposure predictions for
chemical ranking and prioritization.

• Biologically Diverse Cell Lines: Compatibility of HTPP with many human-derived cell models
permits characterization of chemical bioactivity across different domains of human biology.

-------
Acknowledgements

sn,,

\ Office of Research and Development (ORD)
| Center for Computational Toxicology and

prO^°

Z
m

Exposure (CCTE)

Johanna Nyffeler

Clinton Willis

Rick Brockway
Megan Culbreth
Dan Hallinger
Terri Fairley
Ann Richard
Kathy Coutros
Maureen Gwinn
Russell Thomas

Katie Paul-Friedman
Logan Everett
Imran Shah
Richard Judson
Woody Setzer
Grace Patlewicz
Derik Haggard

PerkinElmer

Joe Trask
Dana Hanes
Jim Hostetter

-------
v»EPA References

• Bray MA, et al. Cell Painting, a high-content image-based assay for morphological profiling using multiplexed fluorescent dyes. Nat Protoc.
2016 Sep;ll(9):l757-74. doi: 10.1038/nprot.2016.105. Epub 2016 Aug 25. PMID: 27560178; PMCID: PMC5223290.

• Chandrasekaran SIM, et al. Image-based profiling for drug discovery: due for a machine-learning upgrade? Nat Rev Drug Discov. 2020 Dec
22:1-15. doi: 10.1038/s41573-020-00117-w. Epub ahead of print. PMID: 33353986; PMCID: PMC7754181.

• Gustafsdottir SM, et al. Multiplex cytological profiling assay to measure diverse cellular states. PLoS One. 2013 Dec 2;8(12):e80999. doi:
10.1371/journal.pone.0080999. PMID: 24312513; PMCID: PMC3847047.

• Nyffeler J, et al. Bioactivity screening of environmental chemicals using imaging-based high-throughput phenotypic profiling. Toxicol Appl
Pharmacol. 2020 Jan 15;389:114876. doi: 10.1016/j.taap.2019.114876. Epub 2019 Dec 30. PMID: 31899216.

• Nyffeler J, Haggard DE; Wiliis C, Setzer RW, Judson R; Paul-Friedman K, Everett LJ, Harrill JA. Comparison of Approaches for Determining
Bioactivity Hits from High-Dimensional Profiling Data. SLAS Discov. 2020 Aug 29:2472555220950245. doi: 10.1177/2472555220950245.

Epub ahead of print. PMID: 32862757.

• Thomas RS, et al. The Next Generation Blueprint of Computational Toxicology at the U.S. Environmental Protection Agency. Toxicol Sci. 2019
Jun l;169(2):317-332. doi: 10.1093/toxsci/kfz058. PMID: 30835285; PMCID: PMC6542711.

• United States. Frank R. Lautenberg Chemical Safety for the 21st Century Act. Pub.L. 114-182

• USEPA. 2018. Strategic Plan to Promote the Development and Implementation of Alternative Test Methods Within the TSCA Program.

Washington (DC): Office of Chemical Safety and Pollution Prevention.

• USEPA. 2020. New approach methods work plan: Reducing use of animals in chemical testing. Washington (DC): Office of Research and
Development & Office of Chemical Safety and Pollution Prevention.

• Wheeler A. 2019. Memorandum from administrator wheeler. Directive to prioritize efforts to reduce animal testing. Washington (DC):

United States Environmental Protection Agency.

• Willis C, et al. Phenotypic Profiling of Reference Chemicals across Biologically Diverse Cell Types Using the Cell Painting Assay. SLAS Discov.
2020 Aug;25(7):755-769. doi: 10.1177/2472555220928004. Epub 2020 Jun 17. PMID: 32546035.

-------
oEPA

High-Throughput Transcriptomics

Logan J. Everett, Ph.D.

Computational Toxicology and Bioinformatics Branch
Biomolecular and Computational Toxicology Division
Center for Computational Toxicology and Exposure
Office of Research and Development, U.S. EPA

-------
&EPA

Disclaimer

The views expressed in this presentation are
those of the authors and do not necessarily
reflect the views or policies of the
U.S. Environmental Protection Agency.
Company or product names do not constitute

endorsement by US EPA.

-------
&EPA Tiered Chemical Safety Testing Strategy

Tier 1 Primary Goals:

• Prioritize chemicals by
bioactivity & potency

• Predict biological targets
for chemicals

HTTr Key Challenges:

• Curve-fitting on
count-based data

• Summarization at
pathway/chemical level

Chemical Structure

Broad Coverage,

and Properties

High Content Assay(s)

No Defined Biological
Target or Pathway

Multiple cell types
+/- metabolic competence

Tier 1

Defined Biological Target
or Pathway

Flexible & Cost-Efficient
HTTr = ~20k genes
x 1,000s of chems

In Vitro
Assays for other KEs
and Systems Modeling

Organotypic Assays and
Microphysiological
Systems

Identify Likely Tissue,
Organ, or Organism Effect
and Susceptible Populations

Estimate Point-of-Departure Estimate Point-of-Departure Estimate Point-of-Departure TtlOHlQS €t dl

Based on Biological Pathway or Based on AOP Based on Likely Tissue-or _ \

Cellular Phenotype Perturbation Organ-level Effect without AOP Toxicol Sci 2019

Regulatory Drivers: TSCA/Admin Memo Sep 2019; FY18-22 US EPA Strategic Plan, Obj 3.3

-------
oEPA

Automated in vitro Chemical Screening

Dispensing Test
Chemicals

Cryopreserved
Ceil Stocks

Cell Expansion

Cell Plating

ET\

BioTek
MultiFIa ™ FX

LabCyte Echo® 550
Liquid Handler

Standardized Expansion Protocol

Day In Vitro
(DIV):

Action:
Vessel:

0
T

2
T

5
T

7
~

9
~

11
~

13
~

MC = Media Change
P - Fawsatu:

P3 (from Cryo)

Seed

T25

T7S

T225

Tost Plafe(s)

Perforin Experiment

Joshua Harrill

Track 1: Targeted RNA-Seq

Generating Cell Lysates

TempO-Seq WT

TempO Seq"

icHM

| i. r JiVsl fir >¦»! lywBK

S (/asJ'J t'-1 ¦" .Imi

KOI a*Ii U' o*Jrt|rrtrt I

^ : v'

"torwei :rs4* iT"jrr = i

Track 2: Apoptosis / Cell Viability

Reagent Dispensing

High Content
Imaging

; ¦¦

Perkin Elmer
Opera Phenix™

High Content Screening System

-------
oEPA High-Throughput Transcriptomics Assay

Targeted RNA-seq enables
high-throughput profiling of
cell lysates or purified RNA

Probe set for whole human
transcriptome targets ~21,000
human genes

Captures majority of signal
with much lower sequencing
depth

(~3M reads with attenuation)

Barcoding and pooling allows
multiplexing of hundreds of
samples

RNA

Purified RNA or Lysates

Detector Oligo Annealing

Excess Oligo Removal
Detector Oligo Ligation

PCR with Tagged Primers

Sample Tag 2

Sample Tag 1

Pool Library, Concentrate/Purify

Sequence

Yeakley, et ol. PLoS ONE 2017

-------
SEPA HTTr Study Design

Cryopreserved
Ceil Stocks

13-day Cell Expansion
& Plating

QC Samples:

¦ UHRR
HBRR
BL DMSO
BLTSA

Lysis Buffer

Test Samples:

8 Concentrations
1/4 Log10 Spacing
Triplicate Plates

Ref Chemicals:

Untreated
DMSO
Genistein
Sirolirnus
Tricostatin A

~ Chemical Dose Plate

.y> Treatments Randomized to Test Plate

Harrili\ et al. Toxicol Sci in press

High-throughput in vitro screens
performed on 384 well plates

Standardized dilution series for every
test sample

Multiple QC and reference chemicals
included on every plate to track assay
performance

Triplicate Test Plates:

> Randomized
independently

> Separate cell
culture batches

-------
oEPA

H'Tr Bioinformatics Pipeline

¦55

tif

(0
¦§-<
CB

Count matrix

Samples

Raw Reads
(FASTQ)

Alignment

(HISAT2)

(/)
CD
_Q
O

Probe Manifest

Database Layer

"55

re
c
<

(Q
O

E
a)
s:
O
o

O)
c

Veh Incr
Ctrls Dose

Signature
Aggregation

Signature PODs

Signature
Cone-Response

Rapid processing for large
screens

Many data steps performed
independently for each test
chemical:

• Removal of low signal probes

• Normalization

• DESeq2 analysis

Exploring multiple analysis
strategies for curve-fitting
and signature & chemical-
level summarization

Harrill, et al. Toxicol Sci in press

-------
SEPA HTTr Quality Control

QC Failure Rates Across HTTr Screens

05 3
CL

QC Issue Type

Liquid Handling
| Cytotoxicity 4
Assay Quality

QC Standards MCF-7 Cells U-2 OS Cells HepaRG Cells

• 44 Chemical Pilot Study

• Screened 1,577 ToxCast
chemicals

• Screened 1,201 ToxCast
chemicals

• Screened 137 PFAS

Acoustic dispenser logs identify
problems with chemical handling

Apoptosis/cell viability assays identify
cytotoxic concentrations

Bioinformatic QC checks remove:
Low read depth samples
High rate of alignment failure
Samples with low gene coverage
Samples with irregular count
distributions

-------
Global View of Bioactivity

Differential Expression per Chemical

Cell Type l~H MCF-7 l~H U-2 OS h-H HepaRG

10000

^ 1000

5
o

oc
Q

(3
LU
Q

o
%

100

! i

i i I

' I :

! ! !

i i

¦ I i

0.032 0.1 0.32 1 3.2 10

Chemical Concentration (jiM)

100

Count data for single chemical
(vehicle controls + 8 cones x 3 reps)

Veh Iricr
Ctrls Dose

• Statistical model tailored to *-seq data

• Remove plate-level effects

• Smooths noise across depth &
expression levels

(Love, et al. Genome Biol 2014)

(^DESeq2^)

Each boxplot shows distribution of DEG count
per chemical

Primarily interested in transcriptional

changes that:

• Are coordinated across known
pathways/gene sets

• Fit standard curve-models across all
concentrations

-------
Signature Scoring

Count data
per chemical

Veh Incr
Ctrls Dose

(. DESeq2^)

(^ssGSEA^)

Catalog of signatures with toxicological relevance,
annotated for known molecular targets

> Bioplanet (Huang, et al. Front Pharmacol 2019)
y CMap (Subramanian, et al. Cell 2017)

DisGeNET (Pinero, et al. Database 2015)
}P' MSigDB (Liberzon, et al. Cell Syst 2015)

Single-Sample Gene Set Enrichment Analysis
(ssGSEA) (Barbie, et al. Nature 2009)

• Score coordinated responses at each concentration

• Use moderated log2 FC values from DESeq2 as input
(no thresholds)

• Null distributions constructed by resampling log2 FC
values from whole screen

• Alternate scoring function:
mean (gene set log2FC) - mean(background Iog2FC)

-------
Signature Scoring

Count data
per chemical

Veh Incr
Ctrls Dose

(. DESeq2^)

Reference Chemical (Effect Size)

Genistein (Weak)

>•3-

A A

/ v

I \ 1 \

— >1 I J

J yk \

0.00

0.25 s "tl.50 0.75 1.00
~

10.0

Response Type Q log2 Fold Change Q Signature Scores

(%sGSEA^) . Differential expression analysis of 3 reference chemicals replicated 37 times (MCF-7 large screen)

• Computed distribution of correlations between each replicate analysis

-------
Signature Scoring

Count data
per chemical

Veh Incr
Ctrls Dose

(. DESeq2^)

(^ssGSEA^)-

Catalog of signatures with toxicological relevance,
annotated for known molecular targets

> Bioplanet (Huang, et al. Front Pharmacol 2019)
y CMap (Subramanian, et al. Cell 2017)

DisGeNET (Pinero, et al. Database 2015)
}P' MSigDB (Liberzon, et al. Cell Syst 2015)

Digitoxin

CMAP digoxin 5.2e-06 100 2866 100

Concentration-Response
Curve Fitting (tcplFit2)

0
!—
o
o
CO

o
d

mthd AC!

• exp5 0.1<

50 Top BMD ACC Hitcall
3 0.74 -&.069 0.Q85 -4

f "T I> * T

Y i 1 J

-¦

class: ATPase inhibitor
size: 69

method :exp5 B
Cutoff=0.11

1e-03

1e-01

cone (uM)

1e+01

-------
oEPA

HTTr MCF-7 Pilot Analysis

1e+02

c3
c

Li.

1 e+00

le-02

;** \

# ~ ¦¦¦' 4 ~ /
.~ ~

~/ ~

Clad ribine-

/ ~

W /

A *~

~ /

t t4

1e-02 le+OO

ToxCast POD (jiM)

1e+02

Pilot study of 44 well-characterized
chemicals (Harrill, et al

Compared HTTr-derived PODs from
MCF-7 cells to previous ToxCast HTS

assay results

(Paul-Friedman, et al. Toxicol Sci 2020)

Signature-based POD are highly
concordant with ToxCast results for the
majority of test chemicals in pilot study

• 6 chemicals with targets that have
low/absent expression in MCF-7 cells

• 5 chemicals show off-target hit as most
potent assay in ToxCast

• Cladribine is a non-specific DNA synthesis
inhibitor

-------
oEPA

ML Models for MIE Classification

-------
xv EPA

Stress Response Gene Signatures

Goal: Develop NAMs to characterize
non-specific environmental
chemicals that activate stress
response pathways (SRPs)

Approach: Characterize chemical
hazards using HTTr data to assess SRP
gene signature activity

Challenges: Cross-talk in signaling
networks makes it difficult to find
gene signatures of SRPs

Results: We have developed
consensus SRP signatures for
accurately classifying known stressors

Future: Use signatures to identify
cellular states involved in adaptive
stress responses and "tipping points"
that lead to adversity

Use crowd-sourcing strategy to build consensus
signatures from published data

Published
Signatures

Reduce

Overlapping

Genes

Unique

Consensus

Signatures

Highly gene-overlapped
published signatures
I

GENE ASSOCIATION
AGGREGATOR

OOF
o • o

Unique gene space

signatures