Kimberly Glass, Network model - Ovarian Cancer, fged_seattle_2013
1. A Network Model for Controlling and
Potentially Reversing Angiogenic
Progression in Ovarian Cancer
Kimberly Glass
Functional Genomics Data Society
June 20, 2013
2. • Biological processes are driven not by individual
genes but by the networks linking those genes
• Ultimately, we look to develop models that describe
the interactions driving different biological systems
• We want to find networks using available genomic
data (largely expression data)
• Correlations in gene expression can be considered to
be the result of network interactions
• The question is not “Is this model right?” Rather, the
question is “Is the model useful?”
Why We Care About Networks
7. Another Idea: Message Passing
Transcription Factor
Downstream Target
The TF is Responsible for
communicating with its Target
The Target must be Available
to respond to the TF
GC Yuan, Curtis Huttenhower, John Quackenbush
8. Passing Messages between
Biological Networks
Protein-protein
interactions
Protein-DNA
interactions
Genomic
Data
Gene Expression
Network
Representation
Cooperation
between TFs
Potential
Regulatory Events
genes
genes
Potential Co-
Regulatory Events
Use Message Passing to find a
consensus among the networks
InitialNetwork
Information
Message
Passing
LearnedNetwork
Information
9. Message-Passing Networks: PANDA
(Passing Attributes between Networks for Data Assimilation)
PPI0 Expression0
Network1
Responsibility Availability
Network0
Motif Data
Expression1PPI1
Glass et. al. “Passing Messages Between Biological Networks to Refine Predicted Interactions.” PLoS One. 2013 May 31;8(5):e64832.
Implementation available on sourceforge: http://sourceforge.net/projects/panda-net/
12. A new subtype of ovarian cancer
• mRNA/miRNA and DNA were extracted from 132
well-annotated FFPE samples and profiled on arrays
• A technique called ISIS was used find robust splits in the data
• A major, robust split was associated with expression of
angiogenesis genes
• Published gene expression data was curated and used to
validate the split and signature
13. Genes
Conditions
Expression data
(Angiogenic)
Genes
Conditions
Expression data
(Non-angiogenic)
Application of PANDA to Ovarian Cancer
• Downloaded expression
data from 510 OvCa patients
from TCGA. Normalized
data using fRMA and
mapped probes to EnsEMBL
IDs using BiomaRt
• Assigned subtypes using a
Gaussian Mixture Model
using Mclust: Identified 188
angiogenic, 322 non-
angiogenic patient samples.
• Combined with TF motif and
PPI data and used PANDA to
map out networks.
Network for
Angiogenic Subtype
Network for
Non-angiogenic Subtype
Interaction data
Motif data Compare and
Identify Differences
GC Yuan, Dimitrios Spentzos, John Quackenbush
14. 12631 unique edges,
Including 56 TFs
Targeting 4081 genes
15735 unique edges,
Including 49 TFs
Targeting 4419 genes
Each point:
TF→gene edge
An individual gene can actually be targeted in both subnetworks,
although by different upstream transcription factors.
Gene Overlap
1828 25912253
Genes Targeted in
Angiogenic Subnetwork
Genes Targeted in Non-
Angiogenic Subnetwork
Network Differences are Captured in Edges
15. Key Regulators of Angiogenesis in OvCa
TFTF
Edges from Regulator in
Angiogenic Subnetwork
Edges from Regulator in
Non-Angiogenic Subnetwork
Calculate an
“Edge Enrichment”
and corresponding
significance
16. TF Potential Connection with Angiogenesis/Cancer Publication(s) PMID
NFKB1 important chromatin remodeler in angiogenesis 20203265
ARID3A required for hematopoetic development 21199920
SOX5 involved in prostate cancer progression, responsive to estrogen 19173284, 16636675
TFAP2A increases MMP2 expression and angiogenesis in melanoma 11423987
NKX2-5 regulates heart development 10021345
PRRX2 deletion cause vascular anomalies 10664157
AHR knock-out impairs angiogenesis 19617630
SPIB inhibits plasma cell differentiation 18552212
MZF1 represses MMP-2 in cervical cancer 22846578
BRCA1 inhibits VEGF and represses IGF1 in breast cancer 12400015, 22739988
Key Regulators of Angiogenesis in OvCa
17. TF differential Expression
Target differential Expression
TF differential Methylation
Target differential Methylation
Target genes’ availability to be
regulated is made possible through
epigenetic modifications
Key Regulators of Angiogenesis in OvCa
Some TFs are acting as
transcriptional repressors.
18. A+ A- A+;N- N+;A- N- N+
yes yes yes yes no no
no no yes yes yes yes
yes no yes no yes no
no yes no yes no yes
927 1326 624 1204 982 1609
Gene Group Nickname
Gene targeted in Angiogenic Subnetwork
Gene targeted in non-Angiogenic Subnetwork
Gene’s expression increases in Angiogenic tumors
Gene’s expression increases in non-Angiogenic tumors
Number of Genes in Group
1828 25912253
Genes Targeted in
Angiogenic Subnetwork
Genes Targeted in Non-
Angiogenic Subnetwork
Both Activation and Repression of
Pathways is Important in Angiogenesis
• "A+/A-" genes targeted and more highly/lowly
expressed in angiogenic subtype
• "A+;N-" genes are targeted in both
subnetworks and more highly expressed in
angiogenic subtype
• "N+;A-" genes are targeted in both
subnetworks and more highly expressed in
non-angiogenic subtype
• "N-/N+" genes targeted in the non-angiogenic
subnetwork but are more highly/lowly
expressed in angiogenic subtype
22. • ARNT and ETS1 dimerization with
HIF1a and HIF2a, respectively, play a
role VEGF production. However, AHR
can inhibit this by competing as an
alternate dimerzation partner.
There is a small molecule ligand
that dimerizes with HIF2a and
which may block angiogenesis
AHR agonists may also help to
prevent activation of the
angiogenic pathway
• ARID3A, SOX5 and PRRX2 activate
many genes through CpG-poor
promoters
Therapies altering methylation
may alter some of these
transcriptional programs
Regulatory Patterns Suggest Therapies
X
X
VEGF production
and angiogenesis
HIF1a ARNT
HIF2a ETS1
HIF1a ARNT
HIF2a ETS1 AHR
AHR AHR
AHR
AHR
(1) Prevent ARNT/HIF1a and
ETS1/HIF2a dimerization
(2) Promote ARNT/AHR and
ETS1/AHR dimerization
TREATMENT MODEL
ANGIOGENIC BEHAVIOR
(3) Decrease genome-
wide methylation
ARID3A
SOX5
PRRX2
High levels of CpG methylation
TF2TF1
A+
ARID3A PRRX2 SOX5
N-
AHR ARNT MZF1 ETS1
23. Other Disease Datasets Provide Validation
Standard platinum-based therapies
may actually be priming the cellular
network to towards angiogenesis.
GEO
treatment
conditions
control
conditions
2) RMA-
normalize
3) Compute differential
expression (T-statistic)
T-statistic
#genes
1) Download data
4) Compute summary
statistic for Gene Group
24. Other Disease Datasets Provide Validation
The effects of the VEGF-inhibiting drug
Sorafenib directly correspond to groups of
network-identified genes.
25. Other Disease Datasets Provide Validation
Expression of Network-identified A+ genes
decreases with proposed treatments.
26. What’s Next?
• Represent other genomic data in
the network model
• Investigate networks underlying
other cancers/diseases
• Continue to think about how
biological mechanisms are
represented in network models
• Use network predictions to
hypothesize on treatments
• The question is not “Is this model
right?” Rather, the question is “Is
the model useful?”
27. Acknowledgements
PANDA Development
Guo-Cheng Yuan
John Quackenbush
Curtis Huttenhower
Angiogenic Subtyping
Benjamin Haibe-Kains
John Quackenbush
Ursula Matulonis
Ovarian Network Analysis
Guo-Cheng Yuan
John Quackenbush
Dimitrios Spentzos
Others
Zhen Shao
Jeremy Bellay
Michelle Girvan
Cristian Tomasetti
Emanuele Mazzola
Luca Pinello
Eugenio Marco-Rubio
Matthew Tung
Funding: NIH R01HL111759
PANDA Availability:
sourceforge.net/projects/panda-net/