Transcript PPT

LINCS@HMS
Pharmaco-response Signatures
and Disease Mechanism
Timothy Mitchison, Peter Sorger, Caroline Shamu
Harvard Medical School
Nathanael Gray
Dana Farber Cancer Institute
Cyril Benes, Daniel Haber,
Massachusetts General Hospital
Joshua Stuart
University California , Santa Cruz
(Avi Maayan)
Mt. Sinai, NY
LINCS Fall Meeting – October 26, 2011
LINCS@HMS
Outline
1. Data being collected at HMS LINCS
Center
2. Why collect data this way?
3. A typical data set
4. Organizing and accessing the data
5. Promises and conceptual challenges
6. Impact and outreach
LINCS@HMS
Goals of the HMS LINCS Center
Collect rich response data:
Provide rich data sets for validation of protocols and
development of standards, algorithms and informatic systems
Integrate cell-biology and genomics:
Demonstrate that coupling (i) high-throughput biochemical and
cell-based data to (ii) expression/genomic data will uncover
novel biology involved in disease and drug response.
Create signatures and understand mechanism:
Develop pathway-aware signatures of cellular response to
(pharmacological) perturbation.
Show signatures can uncover novel pharmacological mechanism
and explain variation in response.
LINCS@HMS
Rich response data
Single-cell imaging
Biochemical pathways
Multiplex biochemical
data
pAkt
Drug binding to kinome
Cell fate data
Ligand
LINCS@HMS
Approaches in the HMS LINCS Center
1. Focus on drugs and medicinal chemistry:
assemble annotated collection of kinase inhibitors (clinical
and new) and measure biochemical specificity using industry
standard kinome profiling assays.
2. Collect multi-factorial pharmaco-response signatures:
capture the complexity of response in time and space using
imaging, multiplex biochemistry and transcriptional assays.
3. Apply to cancer and other diseases: including rheumatoid
arthritis, liver disease, and mitochondrial disease.
4. Develop informatics standards (ultimately a pipeline): to
collect, analyze and disseminate diverse experimental data
5. Develop pathway-focused mathematical models: at different
levels of resolution as means to create predictive pharmacoresponse-signatures (PRSs).
LINCS@HMS
Key features of the HMS center
1. Adaptive approach:
Support diverse and changing assay and data types; adapt
ongoing data collection to previous results.
2. Integration:
–
–
–
–
–
of methods and reagents across seven laboratories
of biochemical, imaging and expression assays
of chemical and cell-level annotation
of multiple nodes in signaling network
of protein, mRNA and genome data
3. Leverage:
Leverage existing efforts in participating labs by adding LINCS
standards, informatics and assays to ongoing projects
(federally and industrially funded).
LINCS@HMS
Data being collected at HMS
LINCS Center
LINCS@HMS
Typical Experimental Design
signal transduction
possible therapies
Multiple drugs
genotypes
Primary cells
Tumor cell lines
Multiple cytokines
growth factors
20-50 Signaling proteins
cellular responses
Cell State - apoptosis, growth,
senescence etc.)
“microenvironment”
Transcription (L1000 assays)*
Sequence
Transcriptional state
(Stuart Lab)
GI50 Data
* With Broad LINCS Center
LINCS@HMS
Lysate-array, xMAP/Luminex, ELISA assays
to measure mean response
Cell response
LINCS@HMS
Image-based measurements reveal dispersion in response
Bjorn Millard
LINCS@HMS
Linking immediate-early signals to transcriptional response
Ligand
Cell Type
Signal
SKBR3 Cells + EGF
The joint project will generate the only dataset
linking immediate-early signaling to transcription
across diverse perturbations and cell lines
Transcription
LINCS@HMS
Phenotypic data on perturbagen response
200 Compounds-3 doses
Number of lines
~1000 Tumor Cell lines
Tissue of origin
~106 Data Points on Cell Killing
Leveraging the Cell Line Collection of the Center for Molecular
Therapeutics/Wellcome Trust /Sanger Institute
LINCS@HMS
Perturbagen response determinants
Unresponsive states:
Not in M phase during assay
Quiescent
Modulators of
drug availability:
Drug efflux
pumps
Drug
sequestration
Targets:
Plk1
Aurora A,B
Kinesin-5
MPS1
Microtubules
Pathways:
Mitotic entry
Mitotic exit
Mitotic spindle assembly
Spindle assembly checkpoint
Apoptosis
LINCS@HMS
Measuring perturbagen (drug) specificity
Cell level data
phospho-Erk
glial cell line-derived neurotrophic factor
on MCF7 cells
phospho-Erk
Heregulin on MCF7 cells
We will generate the largest public database of small molecule selectivity
profiles based on the multiplex biochemical assays used in pharma
Pattricelli et al (2007) Biochemistry. 46:350-8. Cravatt et al (2010) Nat Rev Cancer. 2010 10:630-8
LINCS@HMS
Structure of HMS LINCS Data
LINCS@HMS
Why collect data this way?
State of the art
Proven utility
Captures critical features of response
LINCS@HMS
Broad LINCS Center
(contemporary genomics)
LINCS@HMS
HMS LINCS Center
(contemporary biochemistry)
LINCS@HMS
Structure of conventional biochemical data
LINCS@HMS
Application: discovery of inducible autocrine cascades
LINCS@HMS
Application: cell specific signaling networks
LINCS@HMS
Application: cell specific signaling networks
LINCS@HMS
Clustering cell –specific networks based on pathway
topology and drug responses
Compared to gene expression clusters
LINCS@HMS
Response features and single-cell data
• Cell-cell variation:
Heterogeniety in response in addition to mean response important when cell cycle is major factor.
•Outlier populations:
Uncover responses of outlier populations including those
corresponding to undifferentiated or “tumor stem cells.”
•Assay protein localization :
Monitor changes in protein localization, such as nuclear
translocation of transcription factors upon perturbation.
•Monitor cell state changes:
Assay states and morphological transitions such as cycle
arrest, induction of apoptosis, senescence, EMT etc.
with D. Flusberg
LINCS@HMS
Characterizing perturbagen responses at a
single-cell level by imaging
Non-treated
DAPI
NucView
LTR
+ drug 1
+ drug 2
Mitotic index
DNA replication
Apoptosis
Cell size
Viability
Phospo-states
Localization
5637 cells
(NB: images of various
control compounds)
LINCS@HMS
Fractional responses of cells to perturbagens (TRAIL)
(survivors)
Treat
(survivors)
Allow cells
to recover
Treat
2-3 day recovery for TRAIL exposure in culture
LINCS@HMS
Dynamic responses of lapatinib sensitive and resistant
breast cancer cells to perturbagens
BT474 [24h]
21MT1 [24h]
21MT1 [1h]
pErk in BT474 cells
pErk in 21MT1 cells
1.2
pERK [Normalized]
pERK [Normalized]
1.2
1
0.8
0.6
0.4
0.2
0
1
0.8
0.6
0.4
0.2
0
0
1
6
24
48
72
0
1
Time [hours]
6
24
Time [hours]
1 mm Lapatinib
1 mM Meki
48
72
LINCS@HMS
Impact of cell-to-cell variability on dose-response
TRAIL
Recover
Fractional killing
LINCS@HMS
LINCS Software System for Quantitative Image Analysis
•Cell images encode large amounts of biological data beyond the obvious
measurements of interest
•Will provide image analysis community with reference data
•Requires validated antibodies
LINCS@HMS
A typical data set
LINCS@HMS
Cue-signal response dataset – breast cancer (70% done)
LINCS@HMS
Key questions
1. How diverse is ligand response? Map to subtypes (Basal A,
Basal B, Her2+ etc)
2. Are receptor protein levels correlated to RNA levels or to
ligand response? What about a protein/gene signature?
3. Among cell lines that show similar ligand responses, how
diverse are the transcriptional responses?
4. Do Her2+ lines show similar responsiveness to diverse
ErbB ligands? Predict Trastuzumab sensitivity or resistance
mechanisms?
5. More generally – does ligand response correlate to
sensitivity or predict combination drug responses?
6. Is variation in response at single-cell level predictive or
resistance mechanisms?
Mario Niepel
LINCS@HMS
Some data: Ubiquitous Responsiveness to ErbB Ligands
p-Erk
p-Erk
Receptor expression levels
HRG
VEGF
Time
Her2 amplified
TrastuzumabR
p-Erk
EGF
LINCS@HMS
Common, sporadic and rare responses
Log10
(fold
change
)
Common: HRG Responders
Rare: PDGF Responders
HS578T – Basal B
MDA MB 157– Basal B
Sporadic: VEGF Responders
pAKT-S(473) at 10 min
Slide 34/44
4/9/2015
LINCS@HMS
Variation in diversity of transcriptional responses
PCA Analysis by Avi Maayan (Mt Sina
LINCS@HMS
Not only cancer cells: CSR analysis of primary
synovial fibroblasts rheumatoid arthritis v. normal
Normal
Cue
RA
5
WNT5A
t
(min)
5
RA_10_WNT5A
N_10_WNT5A
4.5
4.5
WNT3A
RA_10_WNT3A
N_10_WNT3A
N_10_POLYIC
RA_10_POLYIC
N_10_IL6
RA_10_IL6
3.5
N_10_IL1
RA_10_IL1
N_10_TNFa
RA_10_TNFa
POLYIC
IL6
IL1
3.5
3
3
N_10_EGF
RA_10_EGF
N_10_Visfatin
RA_10_Visfatin
2
N_10_Adiponectin
RA_10_Adiponectin
EGF
1
1
Visfatin
0.5
p53(S37)
Erk2
STAT3
Akt
STAT2
TrkA
STAT6
cJun
PDGFRb
p70S6
STAT3
p53(S15)
Src
HSP27
Erk1
MEK1
NFkBp65
p38MAPK
JNK
RA_30_WNT5A
p90RSK
Tyk2
Gsk3
p53(S37)
Erk2
STAT3
Akt
STAT2
TrkA
STAT6
cJun
PDGFRb
p70S6
STAT3
p53(S15)
Src
HSP27
Erk1
MEK1
NFkBp65
p38MAPK
JNK
p90RSK
Tyk2
Gsk3
CREB
N_30_WNT5A
5
CREB
RA_10_Leptin
0.5
N_10_Leptin
4.5
N_30_WNT3A
RA_30_WNT3A
N_30_POLYIC
RA_30_POLYIC
N_30_IL6
RA_30_IL6
3.5
N_30_IL1
RA_30_IL1
5
4
3
3
N_30_TNFa
RA_30_TNFa
N_30_IGF
RA_30_IGF
IGF
N_30_EGF
RA_30_EGF
N_30_Visfatin
RA_30_Visfatin
N_30_Adiponectin
RA_30_Adiponectin
3
3.5
IL6
TNFa
3.5
4.5
4
IL1
4
1.5
1.5
Adiponecti
WNT5A
n
WNT3A
Leptin
POLYIC
4.5
RA_10_IGF
N_10_IGF
2
IGF
10
2.5
2.5
TNFa
5
4
4
30
2.5
2.5
2
2.5
2
1.5
EGF
1.5
1
Visfatin
1
RA_90_WNT3A
4
RA_90_POLYIC
N_90_IL1
RA_90_IL1
N_90_TNFa
RA_90_TNFa
RA_90_IGF
RA_90_EGF
N_90_Visfatin
RA_90_Visfatin
Leptin
p53(S37)
Erk2
STAT3
STAT2
Akt
TrkA
STAT6
PDGFRb
cJun
p70S6
STAT3
p53(S15)
Src
HSP27
MEK1
Erk1
NFkBp65
p38MAPK
p90RSK
JNK
Tyk2
Gsk3
1.5
1
RA_90_Adiponectin
p38MAPK
1
p53(S37)
Erk2
STAT3
STAT2
Akt
TrkA
STAT6
PDGFRb
cJun
p70S6
STAT3
p53(S15)
Src
HSP27
MEK1
Erk1
NFkBp65
p38MAPK
p90RSK
JNK
Tyk2
p53(S37)
Erk2
STAT3
STAT2
Akt
TrkA
STAT6
PDGFRb
cJun
p70S6
STAT3
p53(S15)
Src
HSP27
MEK1
Erk1
NFkBp65
p38MAPK
p90RSK
JNK
Tyk2
RA_90_Leptin
0.5
CREB
Gsk3
Gsk3
CREB
NFkBp65
Tyk2
JNK
p90RSK
Erk1
p38MAPK
NFkBp65
Erk1MEK1
MEK
Src
HSP27 Src
STAT3
p53(S15)
HSP27
p70S6
cJun
STAT3
PDGFRb
STAT6
TrkA
p53(S15)
Akt
STAT2
p70S6
Erk2
STAT3
Gsk3
p53(S37)
CREB cJun
Tyk2
JNK
PDGFRb
p90RSK
p38MAPK
STAT6
NFkBp65
Erk1
MEK
TrkA
Src
HSP27
STAT3 Akt
p53(S15)
p70S6
STAT2
cJun
PDGFRb
STAT6 Erk2
TrkA
Akt
STAT3
STAT2
Erk2
STAT3
p53(S37)
p53(S37)
p90RSK
JNK
N_90_Leptin
n
1
2
1.5
N_90_Adiponectin
Adiponecti
90
2.5
2
EGF
Visfatin
3
2.5
N_90_IGF
IGF
1.5
3.5
3
N_90_EGF
TNFa
4
Gsk3
IL1
RA_90_IL6
3.5
CREB
IL6
N_90_IL6
2
5
4.5
4.5
N_90_WNT3A
N_90_POLYIC
POLYIC
Tyk2
5
RA_90_WNT5A
WNT3A
Leptin
CREB
p53(S37)
Erk2
STAT3
STAT2
Akt
TrkA
STAT6
PDGFRb
cJun
p70S6
STAT3
p53(S15)
Src
HSP27
MEK1
Erk1
NFkBp65
p38MAPK
WNT5A
n
N_90_WNT5A
p90RSK
JNK
Tyk2
Gsk3
CREB
Adiponecti
CREB
0.5
RA_30_Leptin
N_30_Leptin
0.5
0.5
LINCS@HMS
Organizing and accessing the data
LINCS@HMS
Informational Web Site
LINCS@HMS
HMS LINCS Database
LINCS@HMS
SDCubes: a new approach to management of
high content data
SD cubes merge the HDF5 data standard (from remote earth
sensing with XML to achieve efficient file-based storage of
Unlimited size based on OWL-compliant ontologies
LINCS@HMS
Informatics to store and disseminate image data
We will create the largest (only?) public database of single-cell data on
cellular responses to therapeutic drugs.
LINCS@HMS
Preliminary informatics pipeline for collecting, managing
and distributing single-cell signatures and data
Now
2012
2013
LINCS@HMS
Progress on data release: completing an assay
set typically takes 12-24 months
Biochemical Signatures
Acquire compounds
6 mo
Assay in vitro
Format
6 mo
Release
6 mo
Cue-signal response analysis
Single-cell assays
Repeat -QC
12 mo
3 mo
Population-average assays
12 mo
Transcription assays
3 mo
3 mo
Format
6 mo
Release
LINCS@HMS
Promises and challenges
LINCS@HMS
Challenges facing HMS LINCS Center
• Acquisition of large-scale data sets is occurring in the
absence of a robust analytical or informatics platform
– tracking, analyzing, publishing data is tricky
• Value of analyzing immediate-early pathways
biochemically on a large scale is not yet known – how
dense does data need to be to infer pathways?
• Relative merits of single-cell and population average
data sets must be established.
• Relating transcriptional (Broad) and biochemical/celllevel data (HMS) will require new statistical
approaches and modeling tools
LINCS@HMS
Comparing PARADIGM and Biochemical Networks
ERK Component of the Basal
Pathway in PARADIGM
Comparing biochemical pathways to PARADIGM pathway concepts
developed by Josh Stuart and used in the Santa Cruz Genome Browser
LINCS@HMS
Interaction maps are not biochemical networks
But highly discrepant by source
An receptor rich interactome
EREG
CE(i,j
1.0 )
And not interpretable biochemically
With classical bow-tie features
Cytosolic Kinases
HBEGF
2
ERBB4
10
Macrophage
0.8
0.5
I2D
GeneGo
0.3
NCI-PID
0.2
Reactome
0.1
CellMap
Macropha
ge
I2D
GeneGo
Reactome
STRING
NCI-PID
STRING
Pathway Maps
CellMap
1
10
0
10
-7
10
-6
10
-5
10
-4
10
-3
10
-2
10
-1
10
Betweeness (BT)
Receptors and
binding proteins
Transcription
Factors
LINCS@HMS
Impact and outreach
LINCS@HMS
Building links to relevant communities
•Clinicians and cancer biologists: e.g. SU2C breast
Cancer consortium (led by Joe Gray and Dennis
Slamon).
•Medicinal chemists in industry and academe
(LINCS@HMS advisors)
•Computational biologists involved in pathway
engineering (DREAM competition)
•Cell Biologists and microscopists:
LINCS@HMS
Publications
•
•
•
•
•
•
•
•
Yang R, Niepel M, Mitchison TK, and Sorger PK (2010). Dissecting Variability in Responses to Cancer
Chemotherapy through Systems Pharmacology. Clin Pharmacol Ther 88, 34-38. PMC2941986 PMID:
20520606.
Millard BL, Niepel M, Menden MP, Muhlich JL, and Sorger PK (2011). Adaptive Informatics for Multifactorial
and High-Content Biological Data. Nat Methods 8, 487-492. PMC3105758 PMID: 21516115.
Prill RJ, Saez-Rodriguez J, Alexopoulos LG, Sorger PK, and Stolovitzky G (2011). Crowdsourcing Network
Inference: The Dream Predictive Signaling Network Challenge. Sci Signal 4, mr7. PMID: 21900204.
Zhang T, Inesta-Vaquera F, Niepel M, Zhang J, Ficarro S, Machleidt T, Xie T, Marto JA, Kim N, Sim T,
Laughlin JD, Park H, LoGrasso PV, Patricelli M, Sorger PK, Alessi DR, and Gray NS (2011). Discovery of
Potent and Selective Covalent Inhibitors of Jnk. Chemistry & Biology (accepted, in press)
Garnett MJ, Edelman EJ, Heidorn SJ, Greenman CD, Dastur A, Lau KW, Greninger P, Thompson IR, Luo X,
Soares J, Liu Q, Iorio F, Milano RJ, Bignell GR, Tam AT, Davies H, Stevenson JA, Barthorpe S, Lutz SR,
McLaren-Douglas A, Mitropoulos X, Mironenko T, Thi H, Richardson L, Zhou W, Jewitt F, Zhang T, O’Brien P,
Price S, Hur W, Yang W, Deng X, Butler A, Choi HG, Chang JW, Baselga J, Stamenkovic I, Engelman JA,
Sharma SV, Saez-Rodriguez J, Gray NS, Settleman J, Futreal PA, Haber DA, Stratton MR, Ramaswamy S,
McDermott U, and Benes CH The Genomics of Drug Sensitivity in Cancer. Nature (in review)
Alagesan B, Contino G, Guimaraes A, Corcoran R, Deshpande V, Wojtkiewicz G, Greninger P, Brown R, Chu
G, Ying H, Hezel A, Wong KK, Liu Q, DePinho R, Loda M, Weissleder R, Benes C, Engelman J, and
Bardeesy N Combined Mek and Pi3k Inhibition Induces Cell Death and Tumor Regression in Mouse Models
of Pancreatic Cancer. Cancer Discovery (submitted)
Ni J, Liu Q, Xie S, Carlson C, Thanh V, Vogel KW, Riddle SM, Benes CH, Eck M, Roberts TM, Gray NS, and
Zhao JJ Functional Characterization of the Anti-Cancer Potential of a P110β Isoform-Selective Pi3k Inhibitor.
Nature Chemical Biology (in progress)
Tang Y, Xie T, Moerke N, Shamu CE, and Mitchison TJ A Novel Single-Cell Dye-Based Imaging Assay for
Studying Multi-Dimensional Pharmacological Responses in Human Tumor Cell Lines to Small-Molecule AntiCancer Drugs. (in progress)
LINCS@HMS
Impact: Catalyze formation new discipline
Emerging Discipline:
Quantitative and Systems Biology:
Better Integration of Pharmacology and Systems Biology
Slide Courtesy of Francis Collins
LINCS@HMS
Developing new LINCS collaborations
•Pharmaco-response signatures relevant to mitochondrial
disease: Collaboration with Vamsi Mootha
•Pharmaco-response signatures in iPS Cells: Collaboration
with Evan Snyder.
•Pharmaco-response signatures in primary synovial
fibroblasts from normal v. rheumatoid arthritis patients:
Collaboration with Doug Lauffenburger and Boehringer
Ingelheim.
•Using perturbagen responses to map signaling pathways in
normal and diseased liver: Collaboration with Vertex.
LINCS@HMS
Posters from HMS LINCS Center
• HMS LINCS Database
Andrew Tolopko, Sean Erickson, David Wrobel, and Caroline Shamu
(Harvard Medical School)
• HMS LINCS Informatics and Integration
Jeremy Muhlich, Gabriel Berriz, Jay Copeland, and Peter Sorger
(Harvard Medical School)
• LINCS Joint Project: Linking the phosphoproteome and transcriptome
Mario Niepel*, Aravind Subramanian*, Peter K. Sorger#, Todd R. Golub#
• Discovery of candidate biomarkers of anti-cancer drug sensitivity by
high-throughput cell line screening“
Cyril Benes (MGH-CMT)
• Identifying causal paths linking genomic perturbations to expression
states in cancer
Evan Paull and Josh Stuart (UCSC)
• LINCS Data Working Group
Members of the LINCS Data Working Group (by Caroline Shamu)
LINCS@HMS
Acknowledgements
Investigators
• Cyril Benes
• Nathanael Gray
• Tim Mitchison
• Josh Stuart
• Caroline Shamu
Postdocs/Students
•Mario Niepel
•Yangzhong Tang
•MingSheng Zhang
•Bjorn Millard
•Rob Yang
•Dan Kirouc
Collaborators
•Aravind Subramanian
•Todd Golub
•Joe Gray
•Emily Pace
•Birgit Schoeberl
•Avi Maayan
Staff
• Jay Copeland
• Sean Erickson
• Nate Moerke
• Jennifer Nale
• Jeremy Muhlich
• David Wrobel
• Andrew Tolopko
• Lili Xhou
LINCS@HMS
Conflict Statement
Peter Sorger is founder of Merrimack Pharmaceuticals;
Dean of academic affairs has determined collaborative
CSR project with Merrimack is acceptable because
there is no exchange of funds.
Peter Sorger is founder of Glencoe software, commercial
developer of OME; relationship is under active
management by Dean and Becky Ward. All software is
available in the open source.