network medicine: a systems approach to cardiovascular disease diagnosis & management joseph...
TRANSCRIPT
Network Medicine:A Systems Approach to Cardiovascular Disease
Diagnosis & Management
Joseph Loscalzo, M.D., Ph.D.Brigham and Women’s Hospital
Harvard Medical SchoolBoston, MA
Outline
•Systems and networks in biomedicine
•Network medicine, disease modules, and disease classification
•Network medicine and drug development
Systems and Networksin
Biomedicine
• Why has the rapidly expanding knowledge of the human genome not (yet) given us the expected epiphanies about complex human disease?
• Why has new drug development been limited despite a remarkable range of technologies that can identify drug targets and design pharmacologically suitable candidate drugs?
Two Burning Questions
• Conventional scientific approach: hold all variables constant except the one of interest, and deduce its importance (Cartesian reductionism after Democritus).
• Inductive generalization can follow.• Reductionist approach often
oversimplifies biological systems and leads to linear thinking. A timely example will follow….
Changing Scientific Paradigm
Homocysteine Theory of Atherothrombosis
• First proposed by McCully (Am.J.Path. 1969; 56:111)
• Evidence from over 30 studies suggests that even mild-to-moderate elevations of plasma homocysteine confer a significant, independent risk for atherothrombosis.
• Hyperhomocysteinemia found in 20-40% of patients with vascular disease, but in only 2% of unaffected individuals
Homocysteine Metabolism
Methionine
Cystathionine
Homocysteine
Cysteine
SAH
SAM
Betaine
DMG
THF
5-Methyl-THF
5,10-Methylene
THF
Sulfate
Me-B12
B6
1
4
2
3
1: Methionine Synthase
2: MTHF Reductase
3: Betaine-homocysteine
Methyltransferase
4: Cystathionine-beta-
Synthase
Vitamin Rx, Homocysteine, & CV Risk
--HOPE-2 Investigators, NEJM, 2006--Bonaa KH, et al., NEJM, 2006
NORVIT Trial HOPE-2 Trial
3749 pts. s/p AMI 5522 pts. w/ CVD
or diabetes
2.5 mg Folic Acid
1.0 mg B12
50 mg pyridoxine
Folic Acid (mg) 0.8 0.8 0 0
B12 (mg) 0.4 0.4 0 0
Pyridoxine (mg) 40 0 40 0
Folate, B12, and Homocysteine
Methionine
Homocysteine
Folate, B12
Cystathionine
B6
HCY Metabolism
Biomedicine in Network Context
Biomedicine in Network Context
Folate Metabolism
HCY Metabolism
Intermediary Metabolism
Folate Metabolism
HCY Metabolism
Biomedicine in Network Context
Methionine
Homocysteine SAH
SAM
Betaine
DMG
THF
N -CH3-THF
N -CH2-THFMe-B12
Folate, B12, and Homocysteine
5
5,10
DHF
dUMP
dTMPAcceptor
Methyl-
Acceptor
DNA Synthesis Cell Proliferation
CpG DNA Methylation
ADMA Synthesis
Modify Gene Expression
Impair NO Synthesis/EC Function
Increased AngiographicRestenosis post-PCI(Lange et al., NEJM, 2004)
• Most biological systems respond to multiple inputs that vary simultaneously and can interact—i.e., these are complex systems that form molecular networks.
• New quantitative approaches can be used to construct these networks, identify changes in them that lead to disease, and examine their responses to perturbations (including drug-induced perturbations).
Changing Scientific Paradigm
Key Properties of Biological Networks
• Frequently clustered or scale-free architecture
• Manifest emergent behavior• Functional modularity follows structural
localization; modules as subnetworks• Minimize transition time between states• Comprise canonical structural and dynamic
motifs that reflect basic organizing principles in biological systems
Generic Network StructuresRandom Network Scale-free Network
P(k) = e-k P(k) = k-g
m = g
Poisson Distribution Power Law Distribution
P(k
)
log
P(k
)
log kk
<k>
Few nodeshighly linked
Many nodessparsely linked
k=degree or # nodal connections
• Recapitulate natural selection and evolution–Define difference between mutable nodes
(weakly connected) that engender diversity and facilitate natural selection, and immutable nodes (hubs), the loss of which is lethal for the organism
• Accommodate pertubations to the network with minimal effect on critical functions of the organism
Scale-free Networks: Biological Implications
Network Medicine,Disease Modules,
and Disease Classification
• Anachronistic—Malpighi (microscopic dx), Osler (pathophysiologic dx)
• Organ-based approach reflects common end-stage phenotypes determined by intermediate pathophenotypes (e.g., inflammation).
• Major therapeutic emphasis is on non-specific intermediate pathophenotypes (e.g., antithrombotics).
• Many distinct diseases have common phenotypes (e.g., hepatitides).
• Common treatments, disease co-occurrence, and GWAS suggest common underlying mechanisms.
Shortcomings of Conventional Pathophenotype Classification
Complex PathophenotypesThere is (genetic) overlap among common complex human pathophenotypes.
--Rzhetsky et al., PNAS 2007;104:11694-9
Genome-wide Association Study14,000 cases of 7 common diseases, 3,000 shared controls, 500,000 SNPs
--Wellcome Trust Case Control Consortium, Nature 2007;447:661-78
Different complex chronic diseases have common GWAS loci.
The Genome:Linear Narrative of Disease
--Wellcome Trust Case Control Consortium, Nature 2007;447:661-78
Manhattan Plot
“All science is either physics or stamp collecting.”
--Ernest Rutherford, 1920
Can Molecular Networks Give Unique Insight into Disease Pathogenesis and Therapy?
Essential vs. Disease GenesDisease genes are largely nonessential and do not encode hubs.
--Barabasi et al., Nat Revs Genet 2011;12:56-68
Interactome
Essential vs. Disease GenesDisease genes are largely nonessential and do not encode hubs.
--Barabasi et al., Nat Revs Genet 2011;12:56-68
Interactome
The Consolidated Interactome
• Protein-protein interactions from yeast 2H screens) & protein complexes (CORUM)
• Regulatory protein-DNA interactions (TRANSFAC)
• Metabolic enzyme-coupled interactions (MCIs)
• Protein kinase network (only PTM incorporated to date)
Disease Modules, Disease Neighborhoods, and the Interactome
Disease Module
Interactome
Disease Neighborhood
Disease Modules in the Interactome
13,470 Nodes141,296 Edges
--Menche et al., Science 2015;347:1257601
• We first compiled a corpus of 299 diseases defined by Medical Subject Headings (MeSH) ontology.
• These 299 diseases also have at least 20 associated genes in the current Online Mendelian Inheritance in Man (OMIM) and GWAS databases.
• This compilation totals 2,436 disease-associated proteins.• The derived disease modules are incomplete and, on
average, comprise only ~20% of the respective disease genes.
• Used DiseAse MOdule Detection Algorithm (DIAMOnD) to expand the seed gene (protein) pool by incorporating additional interactome-linked proteins with a significant number of connections (Sharma et al., BMC Bioinformatics, in
press).
Disease Modules in the Interactome
PeroxisomalDisorders
RheumatoidArthritis
Multiple Sclerosis
13,470 Nodes141,296 Edges
--Menche et al., Science 2015;347:1257601
Topological Localization and Biological Consequences
--Menche et al., Science 2015;347:1257601
The more localized a disease is topologically, the greater the functional similarity of the genes in the disease module.
Disease Overlap and Separation in the Interactome
--Menche et al., Science 2015;347:1257601
Module Overlap and Function
--Menche et al., Science 2015;347:1257601
Disease module overlap is associated with functional similarity.
PAH Disease Module Derivation
• Identify disease phenotype of interest.• Ascertain disease network
components.• Construct disease network (i.e.,
determine the structural or functional linkages among module components).
• Identify disease module(s) within network.
PAH Network DerivationPAH Network Components (131 Nodes, 26 Functional Pathways)
Disease components derived from curated literature, or gene, protein, or metabolite (expression) profiles.
PAH Network Derivation
Consolidated Interactome(11,643 Nodes100,791 Edges)
PAH Network Components (131 Nodes, 26 Functional Pathways)
Consolidated interactome of all known physical interations: PPIs and Protein Complexes (CORUM), Regulatory Protein-DNA Interaction (TRANSFAC),
Metabolic Enzyme-coupled Interactions (MCIs), Kinase Network
PAH Network Derivation
Consolidated Interactome(11,643 Nodes100,791 Edges)
PAH Network Components (131 Nodes, 26 Functional Pathways)
Map
PAH Network Derivation
Consolidated Interactome(11,643 Nodes100,791 Edges)
PAH Network Components(131 Nodes, 26 Functional Pathways)
Map
PAH Network Derivation
PAH Network (115 Nodes, 255 Edges,Largest Connected Component = 82 Nodes)
Interactome-derived PAH Network
--Parikh et al., Circulation 2012;125:1520-1532
PAH and miR-21 Disease Module
--Parikh et al., Circulation 2012;125:1520-1532
miR-21 serves as a negative regulator of pathogenic pulmonary vascular responses in PAH.
miR-21-/- Mice and PH
--Parikh et al., Circulation 2012;125:1520-1532
Network Analysis Informs GWAS
--Menche et al., Science 2015;347:1257601
Network Analysis Informs GWAS
--Menche et al., Science 2015;347:1257601
Disease-Tissue Network
--Kitsak et al., submitted
Construction of tissue-specific interactome
Disease-Tissue Network
--Kitsak et al., submitted
The integrity and completeness of expression of the disease module determines the tissue specificity of disease expression.
Network Medicineand
Drug Development
FDA Approved Drugs: 2000-2012
0
20
40
60
80
100
Ap
pro
ve
d D
rug
s
Reasons for Declining Productivity
• Regulatory environment• Increasing need to explore novel targets• Easy targets have been exhausted.• Increasing attrition rate for developing
drugs• The intrinsically flawed reductionist
approach to drug development, i.e., the need to identify a single drug target with a single “magic bullet”
Disease Modules and Therapeutics
Drug Target
Target Function
Drug targets are typically characterized in isolation from the disease module.
--Modified from Barabasi et al., Nat Revs Genet 2011;12:56-68
Target-based Screening
Facilitated by:• Genomic datasets for target
identification• Structural tools, including protein X-ray
crystallogaphy, NMR spectroscopy, computational modeling
• Large real and virtual compound libraries
• High-throughput screening technologies
--Swinney & Anthony, Nature Rev Drug Disc 2011;10:507-519
Drug Discovery Strategies:Success Rates
Per
cent
age
of N
ME
s
Phenotype
Target-based
Screen
N=83
N=30N=17
N=28
Disease Modules and TherapeuticsDrug targets are better characterized with regard to their effects on phenotype.
Drug TargetFunctionalPhenotype
--Modified from Barabasi et al., Nat Revs Genet 2011;12:56-68
The Promise of Personalized Medicine: Find the Target
Before Rx
--Wagle et al., J Clin Oncol 2011;29:3085-3096
BRAFmut
MEK1C121S
The Promise of Personalized Medicine: Inhibit the Target
Before Rx Vemurafenib—15 wks
--Wagle et al., J Clin Oncol 2011;29:3085-3096
BRAFmut
MEK1C121S
The Peril of Personalized Medicine with Conventional Strategy
Before Rx Vemurafenib—15 wks Vemurafenib—23 wks
--Wagle et al., J Clin Oncol 2011;29:3085-3096
BRAFmut
MEK1C121S
Pathway Targeting: Combination RxCombination therapy: dabrafenib (BRAF inhibitor) & trametinib (MEK inhibitor)
--Flaherty et al., NEJM 2012;367:1694-1703
Cardiovascular Drug Data Sets• [1] F.J. Azuaje etal . Drug-target network in myocardial infarction reveals
multiple side effects of unrelated drugs. Scientific Reports 1, 52, 2011 • From [1] we obtained 38 drugs used for acute myocardial infarction (MI) and
344 non-MI drugs that interact with MI drugs. • These drugs involves 425 drug targets (denoted by eMIDTs), among which
67 (denoted by MIDTs) are targeted specifically by MI drugs. • We further collected 431 myocardial infarction disease genes from
Phenopedia in HuGE Navigator (http://hugenavigator.net). We then mapped eMIDTs and MI disease genes onto the human interactome.
--Wang et al., in preparation
Proximity of eMIDTs to MI Genes in the Interactome
eMIDTs and MIDTs (inset)have significantly more interactions with MI disease genes than random expectation under Null Model I (left) and Null Model II (right)
--Wang et al., in preparation
• (e)MI drug targets have significant overlap with MI disease genes.
• We found 12 drug-target-gene (DTG) modules with more than 5 proteins.
--Wang et al., in preparation
MI Drug Target-Gene Modules
MI Drug Target-Gene Module Example
--Wang et al., in preparation
Blue: MI Disease GeneYellow: MI-related Drug TargetRed: MI Drug or Drug Target
Asthma Module
--Sharma et al., Hum Mol Genet, 2015, e-pub
Novel Regulatory Pathway in Asthma: GAB1
--Sharma et al., Hum Mol Genet, 2015, e-pub
GAB1 Signalsome
Novel inflammatory pathway in asthma derived from disease module
--From Proteostasis Web Site: http://www.proteostasis.com/science/proteostasis_network.php
Systems Pharmacology:Visualizing Therapeutic Actions
• Laszlo Barabasi• Dan Chasman• Zak Kohane• Paul Ridker
• Masanori Aikawa• Stefan Blankenberg• Brad Maron• Marc Vidal• Scott Weiss• Tania Zeller
Acknowledgements
• Dina Ghiassian• Maksim Kitsak• Joerg Menche• Piero Ricchiuto• Amitabh Sharma• Ruisheng Wang