The Emerging Paradigm of Network Medicine in the Study of Human Disease

Stephen Y. Chan and Joseph Loscalzo

Circ Res. 2012;111:359-374

doi: 10.1161/CIRCRESAHA.111.258541
Circulation Research is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231
Copyright © 2012 American Heart Association, Inc. All rights reserved.
Print ISSN: 0009-7330. Online ISSN: 1524-4571

The online version of this article, along with updated information and services, is located on the
World Wide Web at:
http://circres.ahajournals.org/content/111/3/359

Permissions: Requests for permissions to reproduce figures, tables, or portions of articles originally published
in Circulation Research can be obtained via RightsLink, a service of the Copyright Clearance Center, not the
Editorial Office. Once the online version of the published article for which permission is being requested is
located, click Request Permissions in the middle column of the Web page under Services. Further information
about this process is available in the Permissions and Rights Question and Answer document.

Reprints: Information about reprints can be found online at:

http://www.lww.com/reprints

Subscriptions: Information about subscribing to Circulation Research is online at:

http://circres.ahajournals.org//subscriptions/

Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014

 Review

The Emerging Paradigm of Network Medicine in the Study

of Human Disease
Stephen Y. Chan, Joseph Loscalzo

Abstract: The molecular pathways that govern human disease consist of molecular circuits that coalesce into
complex, overlapping networks. These network pathways are presumably regulated in a coordinated fashion, but
such regulation has been difficult to decipher using only reductionistic principles. The emerging paradigm of
“network medicine” proposes to utilize insights garnered from network topology (eg, the static position of
molecules in relation to their neighbors) as well as network dynamics (eg, the unique flux of information through
the network) to understand better the pathogenic behavior of complex molecular interconnections that
traditional methods fail to recognize. As methodologies evolve, network medicine has the potential to capture the
molecular complexity of human disease while offering computational methods to discern how such complexity
controls disease manifestations, prognosis, and therapy. This review introduces the fundamental concepts of
network medicine and explores the feasibility and potential impact of network-based methods for predicting
individual manifestations of human disease and designing rational therapies. Wherever possible, we emphasize
the application of these principles to cardiovascular disease. (Circ Res. 2012;111:359-374.)
Key Words: network medicine ! systems biology ! cardiovascular disease ! systems pharmacology

F or centuries, reductionistic thinking has dominated West-

ern scientific theory and has been rigorously applied to
biomedicine. This construct argues that crucial biological
factors operate in simple mechanistic association to control
disease pathobiology. Consequently, current classification of
disease phenotype (pathophenotype) is the result of inductive
generalization from clinicopathological evidence founded on
the law of parsimony. This paradigm has been helpful to
clinicians as it formalizes syndromic patterns that limit the
number of potential pathophenotypes. While quite useful in
an earlier era, classifying disease in this way overgeneralizes
pathophenotypes and does not usually consider individual-
ized nuances in disease expression, susceptibility states, or
preclinical disease manifestations. Underlying those varia-
tions in predisposition to disease and subsequent disease
manifestation is the competing concept that complex biolog-
ical processes driving human disease rarely result from an
abnormality in a single molecular effector. Rather, they are
nearly always the net result of multiple pathobiological
pathways that interact through an interconnected network.
Network theory offers a tractable structure for distilling
relevant insight from the growing sets of molecular “-omics”
data. In its simplest form, a network can be generated by
mapping a set of molecular entities, called “nodes,” with their
positions based on their functional interconnections, called
“links” or “edges,” with one another (Figure 1A). In network
medicine, a “node” may represent a biological factor (eg,
gene, chromosome, metabolite, noncoding RNA, or even a
specific disease/phenotype). “Edges” may represent physical
interactions, transcriptional induction, enzyme activation,
enzymatic conversion of one metabolite to another, or even
shared genes/traits among distinct disease manifestations. As
a result of rapidly progressing technology and high-
throughput data sets, the first representations of a compre-
hensive network, or “interactome,” of relevant functional
molecular interactions in human tissue is now possible.
Currently, data accumulation for network construction has
primarily relied on an integration of genetic studies (linkage
analyses and genome-wide association studies, GWAS), so-
called high-throughput –OMIC analyses (genomic sequenc-
ing, transcriptional expression analyses, proteomic analyses,
etc), and biochemical studies (ie, metabolomic analyses),
coupled with cellular, physiological, and clinical data.
As a result, application of network theory has begun to
permeate the scientific study of a large number of defined
biomedical systems, ranging from simple signaling circuits to
complex human diseases, such as cancer. The science of
studying disease network behavior is just now emerging and
has been termed “network medicine” (Figure 1B).1 In this
review, we will define the challenges of building the current
human “interactome” and the fundamental principles of
biological network theory in medicine. In doing so, we will

Original received April 5, 2012; revision received June 11, 2012; accepted June 19, 2012. In May 2012, the average time from submission to first
decision for all original research papers submitted to Circulation Research was 12.0 days.
From the Division of Cardiovascular Medicine, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA.
Correspondence to Joseph Loscalzo, MD, PhD, Department of Medicine, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02115. E-mail
jloscalzo@partners.org
© 2012 American Heart Association, Inc.
Circulation Research is available at http://circres.ahajournals.org DOI: 10.1161/CIRCRESAHA.111.258541

359
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
360 Circulation Research July 20, 2012
Non-standard Abbreviations and Acronyms
CALM1 calmodulin
CETP cholesterol ester transfer protein
ChIP-ChIP chromatin immunoprecipitation followed by microarray
analysis
ChIP-Seq ChIP followed by sequencing
FDA-AERS Adverse Events Reporting System of the FDA
FDA Food and Drug Administration
GWAS genome-wide association studies
HCMV human cytomegalovirus
LCMS likelihood-based causality model selection test
LQTS long-QT syndrome
MCA metabolic control analysis
mRNA messenger RNA
miRNA microRNA
ODEs ordinary differential equations
PPAR peroxisome proliferator-activated receptor
PPI protein-protein interactions
QTL quantitative trait loci
explore how the study of network topology and network dynam-
ics can accelerate discovery of novel disease genes and path-
ways, offer new insights into the systems-wide effects of drug
therapy, and reshape disease classification. In the course of this
presentation, we will highlight key studies, cardiovascular and
otherwise, that have successfully integrated network theory with
experimental validation and thus have demonstrated the power
of these approaches for enhancing our understanding of molec-
ular and cellular interconnectedness and its potential regulation
in the control of disease phenotype.
Modeling of Biochemical Networks
Although most areas of modern medicine have relied nearly
exclusively on reductionist models of human pathogenesis,
efforts to establish and apply a detailed understanding of
networks for predicting “downstream” effects on cellular or
physiological states have been attempted but have not yet
been routinely adopted. While catalogs have been curated in
recent years describing the global regulation of mammalian
genes, detailed biochemical and kinetic data associated with
enzymatic pathways controlling cellular metabolism and
energy production have been available for decades. As a
result, mathematical frameworks exist for quantifying how
metabolic variables (eg, metabolite concentrations) depend
on biochemical network parameters,2– 4 thus setting the stage
for attempts at utilizing in silico simulation and modeling to
understand complex biochemical processes.
Methodology involves use of either deterministic [eg,
systems of ordinary differential equations (ODEs) to assess
the effects of perturbations on associated nodes in the
network context] or stochastic [eg, accounting for random
effects on associated nodes by calculation of effect probabil-
ities after network perturbation] simulation of reaction net-
works. Such simulation has allowed for the computation of
metabolite steady-state concentrations and their stability5 or
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
for the analysis of transcriptional network dynamics, respec-
tively.6 “Sensitivity analysis,” or so-called “metabolic control
analysis” (MCA), is used to compute sensitivity coefficients,
which generally reflect the ratios between the change in the
biochemical network behavior (as monitored by some mea-
sure of global system behavior) and the perturbation on
specific system nodes.3,4 Resulting parameters with large
sensitivity magnitude are deemed to be biologically important
and hence considered to be the controlling factors in the
functional regulation of a system.7 Various software packages
for kinetic modeling of biochemical networks are available
and continue to grow in number.8 Although most software
shares a common core of functionality, the specific capabil-
ities and user interfaces of different packages differ with
respect to their functionality, reliability, efficiency, user-
friendliness, and compatibility.
Validation of the predictions arising from such MCA have
mostly centered on relatively narrow biochemical networks
linked by defined rate and time constants that are consistent
among a variety of perturbations. Although in many biolog-
ical circumstances such rate constants fluctuate and depend
on regulation of the catalyzing enzyme, the tight homeostatic
regulation that is seen in many biochemical networks, in part,
allows for consistent approximations of rate constants with-
out substantially altering the network output. Moreover,
probabilistic estimations of dynamically changing rate con-
stants have been attempted.9 Such methodologies have vali-
dated important molecular features in metabolic and signaling
cascades at the single cell level. Consequently, one of the
uses of MCA is to infer the importance of cellular processes
or pathways and to provide mechanistic explanations for
biological behavior.10 –14 As the technology improves for
detecting dynamic changes in metabolites via high through-
put metabolomics in a variety of clinically relevant scenar-
ios,15 the potential capabilities of such biochemical kinetic
modeling should continue to expand.
Modeling of Transcriptional Gene
Regulatory Networks
The success of such biochemical modeling strategies has
paved the way for modeling other biological systems. For
example, attempts to map transcriptional gene regulatory
networks have been pursued. Similar to metabolic networks,
the first versions of these networks were constructed from the
simple known regulatory pathways mapped via traditional
molecular biology studies. With the advent of high-
throughput expression array technology and the ability to
quantify and correlate alterations in mRNA expression at the
transcriptomic level to various stimuli, the ability to construct
gene “coexpression networks” has advanced considerably,
allowing for mapping of gene modules that are coregulated
under a shared stimulus or are associated with a relevant
biological condition or genetic polymorphism (Figure 2A). In
this context, novel computational procedures have been
developed, such as “expression screening,” which can inte-
grate information from thousands of microarray expression
data sets to identify genes that are consistently coexpressed
with a target pathway, such as those in oxidative phosphoryla-
tion,16 across biological contexts. By additionally accounting for
 Chan and Loscalzo
predictive algorithms to identify promoter binding sites for
transcription factors (TRANSFAC and B-cell interactome) as
well as incorporating data from high-throughput assays demon-
strating actual physical binding of transcription factors to pro-
moters [chromatin immunoprecipitation followed by microarray
analysis (ChIP-ChIP) and ChIP followed by sequencing (ChIP-
Seq) as found in databases including the Universal Protein
Binding Microarray Resource for Oligonucleotide Binding Eval-
uation (UniPROBE) and the open access database of transcrip-
tion factor-binding profiles, JASPAR], further coexpression
networks have been verified and, in turn, used to identify novel
factors and direct regulatory connections that influence substan-
tially various cellular phenotypes.
Network Modeling of Protein-Protein Interactions
and Posttranslational Modifications
More recently, there have been considerable advances made
in cataloguing the immense set of human protein-protein
interactions (PPI). While curation from the scientific litera-
ture alone has revealed substantial obstacles17 (Figure 2B),
more promising methodologies using yeast 2-hybrid and
3-hybrid screens as well as immunoprecipitation and high-
throughput mass spectrometry have greatly increased the
likelihood of attaining a more comprehensive PPI cata-
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
Network Medicine and Human Disease 361
Figure 1. An introduction to network-based
strategies in biomedical research. A, Hypotheti-
cal biological network. Essential genes (called
“hubs”) demonstrate dense interconnectivity with
other genes and hold a central topographical posi-
tion in the network; nonessential genes (called
“nodes”) tend to be placed at the network periph-
ery and are less well connected to other genes.
Genes that are responsible for normal biological
variability tend to be nonessential, whereas defi-
ciencies of “hub” genes typically cause embryonic
lethality.1 Many human disease genes (arrow) are
located at the periphery and display a preference
to interact with other disease genes (the “local
hypothesis”). Adapted from Chan et al,107 with per-
mission. B, The universe of “network medicine”
links genetic, “-OMIC,” biochemical, cellular, physi-
ological, and clinical data to create an intercon-
nected graph that can be used to model predic-
tively a (patho)biological event(s). Adapted from
Morel et al,108 with permission (illustration credit:
Ben Smith).
log.18 –22 To date, novel insight of biological function from
these PPI networks has resulted largely from analysis of static
networks. There has been thought of establishing more
dynamic modeling via incorporation of reported association
constants among protein pairs. However, it remains to be seen
if mathematical modeling can adequately capture the sheer
complexity of these networks, and, more importantly, if such
modeling is accurate when considering the potential need to
define the precise and changing concentrations of each
protein component involved in competitive binding to
multiple protein partners in different cellular compart-
ments. Similarly, regulatory relationships that coordinate
activating and inhibitory posttranslational modifications
(eg, phosphorylation, acetylation, S-nitrosylation, redox
modifications, etc) have begun to be described in databases
such as PhosphoELM, PhosphoSite, phosphorylation site
database (PHOSIDA), NetPhorest, and the CBS prediction
database. While phosphorylation sites in the proteome have been
reasonably mapped and catalogued, the systematic cataloging of
alternative modifications has yet to be completed. When con-
sidering the more than 300 posttranslational modifications of the
proteome that are now recognized as well as associated protein
binding complexes with bioactive inorganic ions (eg, iron,
iron-sulfur clusters, selenium, etc), such network mapping re-
362 Circulation Research July 20, 2012

Figure 2. Network modeling of func-

tional interactions among biological
molecules. A, Topology of a gene coex-
pression module shared by developing
fetal myocardium and hypertrophied/fail-
ing myocardium. Adapted from Dewey et
al,53 with permission. B, Venn diagram
displaying the overlaps of the total
curated number of reported binary yeast
PPIs, with or without removing the larg-
est high-throughput yeast PPI reports
(numbers in parentheses). These data
suggest that that literature-curated data-
sets, at least for yeast, are far from com-
plete. Adapted from Cusick,17 with
permission.

mains an essential but daunting task. For instance, while sub- the systematic mapping of these modifications by mass spec-
stantial evidence supports the notion that redox modifications troscopy has begun by both academic23 and commercial biotech-
regulate specific enzymatic activities, identification and predic- nology laboratories.
tion of redox modifications throughout the proteome presents a
complicated challenge, as the exact identity of relevant modifi- Network Modeling of Epigenetic Modifications and
cations depends not only on protein site but also on the oxidative Interactions With Noncoding RNA
state of redox-sensitive amino acids, such as cysteines (eg, Epigenetic alterations of the genome (eg, DNA methylation)
sulfonic versus sulfenic versus thiolate moieties). Nonetheless, and the transcriptome (eg, repression by noncoding RNA)
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
 Chan and Loscalzo
have recently been recognized as powerful and ubiquitous
gene regulatory mechanisms. These regulatory patterns are
increasingly becoming incorporated into more comprehen-
sive gene regulatory networks. For example, the MethDB
database24 contains a catalog of genomic DNA methylation
encompassing 5382 methylation patterns or profiles for 48
species, 1511 individuals, 198 tissues and cell lines, and 79
phenotypes.25 Thus, it may soon be possible to integrate such
granularity in this database into other networks to reveal
novel functional connections through DNA methylation.
Numerous atlases also have been compiled recently detail-
ing the systems-wide functions of noncoding RNA, such as
microRNA (miRNA). MiRNA are conserved, non–protein-
coding RNA molecules that have been identified as essential
mediators of a variety of genes and cellular processes. Inside
the cell, miRNA negatively regulate gene expression primar-
ily by binding to the 3! untranslated regions of messenger
RNA (mRNA) transcripts to repress translation and/or de-
grade mRNA. Efficient binding relies on Watson-Crick
base-pairing between the seven-nucleotide “seed sequence”
of a given miRNA and its mRNA target, and several
algorithms (TargetScan, miRBase, PicTar, miRDB, and oth-
ers) have accordingly been developed to predict accurately
mRNA targets of each miRNA.26 In concert, high-throughput
analyses using transcriptional expression arrays and/or pro-
teomics have allowed for verification and identification of
additional miRNA targets,27 which have been incorporated
into a number of these databases (eg, DIANA and TarBase28).
Although the utility of these prediction algorithms and
high-throughput methods has been reflected in their promi-
nent use in identifying key mRNA targets of specific miRNA,
their use in more comprehensive network-based approaches
is just beginning.29 –31 It is estimated that most messenger
RNA transcripts are regulated by miRNA,32,33 with each
miRNA ("700 identified mammalian miRNA) typically
repressing expression of multiple mRNA (at times, "100
targets simultaneously). It has been proposed that miRNA
appear to act by regulating multiple interacting targets in the
same functional network to generate robust actions in vivo.34
A deeper exploration of the implications of such networks of
miRNA regulating networks of mRNA will certainly neces-
sitate advances in computational analyses as well as experi-
mental validation. Nonetheless, the possibility of mapping a
comprehensive epigenetic regulatory network for complex
human diseases is increasingly becoming realized.
Topological Characteristics of the
Human Interactome
Building on the advances of mapping biochemical, gene
regulatory, protein-protein, and epigenetic networks, we are
now able to construct a more expansive network that more
comprehensively reflects the multitude of molecular connec-
tions affecting human health and disease. Depending on
which sets of databases are used, the human “interactome”
currently encompasses #25 000 protein-coding genes,
#1000 metabolites, and an increasing number of posttrans-
lationally modified proteins and functional RNA molecules,
together exceeding 100 000 participants and substantially
greater functional interactions that variably overlap with one
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
Network Medicine and Human Disease 363
another.1 The resulting interactome displays significant clus-
tering, with pockets of especially dense interconnectedness.
As a result, at most, a few links separate each node from any
other node. Thus, perturbation of a single node can induce
widespread effects throughout multiple modules and, thus,
can substantially alter multiple functional systems.
Importantly, the human interactome is a “scale-free” net-
work in which gene interactions do not occur randomly, with
the degree distribution following a power-law tail (Figure
3A).1,35 Thus, this type of network contains a few highly
connected “hubs,” which are nodes that link substantially to
multiple other nodes, and typically are essential for maintain-
ing the integrity of the entire network. Hub proteins are
probably encoded by evolutionarily conserved genes that
function in crucial cellular activities. Some hubs appear to
direct intrinsic gene cluster function, whereas others preserve
the integrity of the entire interactome by linking distinct
clusters to one another. Scale-free networks exhibit properties
that offer substantial insight into biological systems. Scale-
free properties minimize transition time between states of the
system; they facilitate transfer of molecular information
across the system; they accommodate perturbations to the
system with minimal adverse effects; and they promote
biological diversity.1 Notably, scale-free systems exhibit
“overdetermined” features, implying that the complete defi-
nition of all components of the network is not necessary to
understand how the network functions as a whole. Finally,
scale-free systems display emergent behavior, as defined by
the fact that intimate knowledge of one node cannot be used
to predict its behavior in a network nor the behavior of the
entire network (Figure 3B). It is this emergent property of
biological networks that emphasizes the significant pitfalls in
the reductionistic studies guiding biomedical research pres-
ently—that is, erroneously attempting to extrapolate the
functional study of a single gene alone to its in vivo role(s)
within the context of the endogenous biological network
within which it operates.
Disease-relevant genes and related pathways are certainly
present within the human interactome but have been incom-
pletely defined. Currently, only 10% of human genes have
been associated with a known disease (www.omim.org), and
many have been found to correlate more often with nonhub
nodes.36 Disease genes may also exist as hubs or essential
genes, but since dysfunction of essential genes often results in
embryonic lethality, the association of disease genes and hubs
has been more difficult to prove, especially for chronic
diseases of adulthood. According to the “local hypothesis,”
disease genes also tend to interact directly with other disease
genes in “disease modules” that induce a common pathophe-
notype36 (Figure 3C). Indeed, expansive analysis of gene
expression arrays has confirmed the consistent and pervasive
nature of these disease “signatures” of gene expression even
across vastly different tissue sources.37 These disease mod-
ules can be challenging to map. Nonetheless, they may
substantially overlap with “topological modules”—identifi-
able by unbiased network-clustering tools—and with already-
defined “functional modules”— clusters of nodes of similar
or related function that have been defined previously by
independent experimental validation38 (Figure 3C). Accord-
364 Circulation Research July 20, 2012

Figure 3. Properties of biological networks. A, The connectivity of random networks is reflected by a Poisson distribution, whereas a
scale-free (or clustered) network is defined by a power law distribution. As a result, there exist nodes that are sparsely linked as well as
nodes that are more highly linked (hubs). B, As depicted by a hypothetical complex electric circuit, the detailed analysis of 1 or 2 resis-
tors in isolation allows for little insight into the behavior of the circuit as a whole. Thus, similar to scale-free networks, this circuit dis-
plays emergent behavior. Moreover, detailed analysis of every resistor in the circuit is not necessary but rather can be inferred to
understand the circuit as a whole. Thus, scale-free networks and this circuit are overdetermined. C, Depiction of the 3 modularity con-
cepts of biological networks. Topological modules (left) are defined by locally dense neighborhoods of the interactome. Functional
modules (gray nodes, center) are defined by network neighborhoods in which there is a statistically significant segregation of nodes of
related function. Disease modules are groups of nodes whose perturbation (mutations, deletions, copy number variations, or expression
changes) can be linked to a particular disease phenotype and can cross functional molecules, shown as red nodes. Adapted from
Barabasi et al,1 with permission.

ingly, human disease manifestation may result from several
perturbations of a single disease module which may overlap
with the same components as related, but perhaps more easily
identifiable, topological and functional modules.39 Conse-
quently, the topographical position of a single disease gene in
the human interactome can offer unique insight into its novel
associated partners, their connected roles in pathogenesis, and
connected modules that regulate disease manifestation and
response to therapeutic interventions.
Network Topology and Human Disease
Network medicine can accurately predict novel disease genes,
based on their topographical position in relation to known
disease-relevant factors (Figure 4A). First, the so-called
“linkage” method enables the identification of unique factors
important in human disease via their direct connection to
known disease genes. In cardiovascular disease, successful
examples have included identification of a network of candi-
date susceptibility genes for coronary artery disease based on
disease genes selected from GWAS followed by analyses of
directly connected neighboring genes via a protein-protein
interaction database.40 Ghazalpour et al identified a cadre of
genes that are associated with obesity in the mouse based on
their links in a coexpression network with disease genes
genetically associated with an obese phenotype.41 Second, the
so-called “disease module” method (with some variants
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
called “neighborhood” or “clustering” methods) relies on
network mapping of genes altered in transcriptional coexpres-
sion analyses of diseased tissue, thus defining previously
unrecognized factors carried within these “disease modules”
that influence pathogenesis. For cardiovascular diseases,
relevant modules have been defined for atherosclerosis,42– 45
in-stent restenosis,45,46 type 2 diabetes mellitus,47 and cardiac
hypertrophy and failure,48 among others. We have recently
identified a disease module important in the progression of
pulmonary hypertension, allowing for the identification of a
set of microRNA that control the pathophenotype based on
the predominance of their targets in that gene module29
(Figure 4B). Finally, delineation of modules that topolog-
ically neighbor but do not overlap with a given disease
gene or module has facilitated the delineation of new
disease pathways. Based on the mean topological distance
from a given set of disease nodes, so-called “random walk”
algorithms have been useful in defining such neighboring
modules and linking them to diseases such as diabetes
mellitus.49 Linkage methods and mapping of disease mod-
ules have become much more prevalent, probably due to the
relative ease of performing these analyses. Conversely,
diffusion methods appear to carry the best predictive
power,50 probably stemming from their ability to integrate
a greater amount of relevant topographical information
into the final analysis.
 Chan and Loscalzo
Network medicine may also provide insight into the
presumed but otherwise unappreciated molecular links
among separate human diseases. Goh et al first demonstrated
the existence of these widespread connections after construct-
ing a comprehensive map of disease modules in the human
interactome36 (Figure 4C). Similarly, analyses of coexpres-
sion networks have revealed novel molecular connections
between Alzheimer disease and cardiovascular illnesses.51
More recently, Rende et al constructed a cardiovascular
disease functional linkage network (CFN) that carries signif-
icant interconnections among modules representing cardio-
vascular diseases with other complex disorders such as
infection by Listeria monocytogenes, myasthenia gravis,
hemorrhagic diatheses, and protein S deficiency.52 Gene
coexpression analyses have also identified shared candidate
genes that link normal myocardial development to myocar-
dial hypertrophy and failure.53 Thus, such analyses can be
used to identify links among developmental and pathological
contexts in the same tissue. As a result, these molecular
interconnections may drive the clinical cosegregation of
specific disease phenotypes. Furthermore, identification of
these shared links may lead to novel, shared therapeutic
targets for multiple diseases or suggest a “repurposed” use of
medications already approved for separate disease conditions.
Network Dynamics and Human Disease
Although network topology certainly provides unique in-
sights, more substantial advances in our comprehension of
biological function in complex systems depend on under-
standing dynamic signal flux through these networks. To do
so, as with any complex circuitry, direct measurement or
inference of three key components is essential: the initial
concentrations of each component or node, the overall struc-
ture of the network (eg, the interconnectivity), and the
strength and direction of the nodal interactions. Although
great advances have been made in defining the first two of
these components in a variety of human disease conditions,
accurate determination of the dynamically changing strength
of interactions is a challenging task. Nonetheless, rapidly
advancing methodologies have been proposed to define or
infer the details of those interconnections.
As discussed previously, understanding the dynamics of
biochemical networks has been fruitful given the previously
defined details regarding the network structure and the
enzymatic rate constants of the reactions that interconnect
within the network. Especially in the setting of the study of
single-celled prokaryotes (eg, bacteria) or eukaryotes (eg,
yeast) where initial metabolite concentrations are readily
measured, precise calculations of sensitivity coefficients re-
flect the predicted effects on the biochemical network after
specific perturbations. Application of such modeling has been
powerful when performed in the context of bacterial patho-
genesis and has emphasized the importance of global meta-
bolic alterations54 that contribute to bacterial growth, antibi-
otic susceptibility, and infective capabilities in humans.
Modeling of metabolic networks in human tissue and human
pathogenesis has been more challenging. First, complete
metabolomic profiles have only recently become available
from clinically relevant tissue samples, thus increasing the
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
Network Medicine and Human Disease 365
difficulty in inferring accurately initial or final concentrations
of metabolites in a given condition. Second, the human
network of metabolic reactions is more complex than that of
single-celled organisms. Finally, owing to the complexity of
overlapping pathways and inputs in human tissue, rate con-
stants are variable and are subject to alterations in enzyme
activity, expression, and time-dependent relationships. None-
theless, various mathematical models have been developed to
reduce the impact of these constraints, leading to significant
insight into the impact of metabolic dysfunction in human
disease.
Defining and applying the dynamic flux through gene
regulatory networks presents an even greater challenge. Gene
coexpression networks alone falter as compared to more
sophisticated approaches to simulate reaction networks,
which stems from several causes. First, as generated from
expression array data, coexpression networks are relatively
large and tend to encompass numerous subsets of molecular
networks. Coexpression alone is typically insufficient to map
the direct regulatory connections among all relevant genes,
and thus gene regulatory maps tend to remain incomplete.
Furthermore, unlike the previously described biochemical
networks in which each connection can typically be described
by a precise rate constant, connections functionally linking
gene regulatory networks together have not been as precisely
defined, as binding affinity of transcription factors to gene
promoters do not necessarily always quantitatively reflect the
relative level of transcriptional activation. Finally, modeling
dynamic changes in gene expression generally requires sto-
chastic, rather than deterministic, approaches given the fre-
quently low concentrations of transcription factors and the
limited number of promoter binding sites (as few as 2).
Despite these obstacles, methodologies have been estab-
lished to infer better the exact nature of the relationships and
the rudimentary dynamics of gene regulatory networks. One
method involves applied human genetics analysis. Presently,
chromosomal linkage mapping and GWAS are the most
prevalent analytic methods used to reveal common and rare
genetic mutations and polymorphisms linked with human
disease. These are powerful methods by which to identify
novel genetic associations with human disease, but, when
used alone, these approaches are limited in their ability to
prove a causal or directed pathogenic mechanism. In contrast,
when coupled with an appreciation of where those genes
lie in the interactome network, these genetic alterations can
be integrated to infer the direction and dynamics of informa-
tion involving associated factors throughout the network to
determine causality among those molecular relationships. For
example, when used in an iterative and systematic process to
map how variations in DNA associate with relative transcript
abundances in a relevant coexpression network, statistical
algorithms have been successful in inferring an independent,
causative, or reactive function of disease-gene candidates
relative to a specific human disease phenotype.55 This ap-
proach has been used for the identification of 3 new genes in
the susceptibility to human obesity and potentially could be
applied to human mutations that segregate with overall
cardiovascular health risk (Figure 4D). More broadly, infer-
ence of network dynamics through genetic study may help to
366 Circulation Research July 20, 2012

Figure 4. Discovery of novel disease genes using network-based approaches. A, Identification of novel disease genes using
network-based approaches. Adapted from Chan et al,107 with permission. B, A network-based strategy for identifying miRNA important
in pulmonary hypertension (PH) elucidates a target network of 7 miRNA groups at the crucial intersection of hypoxia, inflammation, and
TGF/BMP signaling. Such an approach also reveals genes that may represent points of coordinated miRNA regulation in PH. Circles
indicate predicted gene targets; triangles, miRNA; blue lines, predicted associations of miRNA and targets; dotted gray lines, gene
interactions documented in the consolidated interactome. Circle size is proportional to the number of miRNA groups (among these 7)
predicted to target that particular gene. Node colors represent functional pathways (ie, endothelin signaling, mitochondrial metabolism,
etc). Adapted from Parikh et al,29 with permission. C, A human “diseasome” displays molecular interconnections among separate cate-
gories of human disease. An edge denotes the involvement of a shared disease factor(s) or pathways common to multiple different dis-
eases. Adapted from Goh et al,36 with permission. D, Schadt et al used a likelihood-based causality model selection test (LCMS), using

Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014

 Chan and Loscalzo Network Medicine and Human Disease 367

answer the question of “missing heritability” in human traits.
Specifically, despite efforts to identify additional genetic loci
to explain the vastly differing patterns of heritability in
human disease, recent computational analyses by Zuk et al56
suggest that variations in heritability need not invoke missing
genetic variants but could adequately be explained by com-
plex variations in genetic interactions among different mo-
lecular pathways. Thus, these findings emphasize the intrigu-
ing potential of studying network dynamics to reveal a
widespread “network-based” origin of disease heritability.
Separate from human genetic studies, an alternative meth-
ods by which to study network dynamics in human disease
involves “reverse engineering” regulatory networks.57 In this
approach, a discrete and typically small module of relevant
genes is systematically perturbed with measurement of the
consequent changes of their associated genes. The regulatory
relationships among the genes can then be inferred, yielding
novel predictions of direct regulatory connections that can be
verified experimentally. Indeed, such processes have been
validated in small sets of discrete gene networks active in
organisms, such as bacteria,58 yeast,59 and Drosophila,60 as
well as for mapping networks of small noncoding RNA in
bacteria.61 This approach has recently been applied to human
disease by Ergun et al, who inferred a gene regulatory
network important in prostate cancer that led to validation of
the androgen receptor, in the context of this network, as a
marker to detect relatively aggressive primary prostate can-
cers (Figure 4E).62 The advantage of such reverse engineering
is its unbiased approach to network mapping,63 as a precon-
ceived framework of regulatory connections is not necessary;
however, this approach is hampered by the technical chal-
lenges of applying the algorithms to larger and much more
complicated regulatory networks. Additionally, because in-
ferred relationships can only be depicted dichotomously as
direct “activating” or “inhibiting” conditions, more complex
network relationships that involve connections beyond tran-
scriptional regulation (ie, posttranslational modifications,
protein-protein oligomerization, etc) require different types of
representation, and thus the accuracy of any simulation of
network perturbation could suffer. Future iterations of anal-
ysis of the dynamics of gene regulatory networks may benefit
from utilizing the known topology of the existing consoli-
dated interactome coupled with statistical algorithms that
incorporate the results of perturbations of genetic coexpres-
sion networks via genetic associations and/or alternative
stimuli and thus iteratively annotate and optimize the regu-
latory map. Consequently, a more sophisticated and compre-
hensive network representation is possible that could be
applied to further systems-level discoveries of how patho-
genic or therapeutic stimuli may alter overall disease
manifestation.
Network Medicine and Drug Development
(Systems Pharmacology)
In addition to aiding discovery of novel molecular aspects of
human disease, network medicine has the potential to revo-
lutionize the process of drug discovery.64 Historically, drug
design has been based on key observations of pharmacolog-
ical or genetic manipulation of putative targets or pathways
that, when perturbed, alter a quantifiable (patho)phenotype.
Unwanted pharmacological side effects have been attributed
to an interaction of a medication with alternative factors in
separate pathways. However, in complex human diseases,
both therapeutic and adverse consequences of drug treatment
may depend on shared effectors and pathways that, when
manipulated in distinct combinations involving specific cel-
lular or tissue contexts, may drive disparate pathophenotypes.
It has been estimated that approximately 400 to 600 proteins
of the presumed 100 000 participants in the consolidated
interactome are targets of the 1200 Food and Drug Admin-
istration (FDA)-approved drugs65; however, a large propor-
tion of these established drugs do not target actual disease-
associated proteins but merely neighboring proteins.66 Taken
together, these drugs may only modestly affect disease
phenotype. In addition, there likely exist a larger number of
potential therapeutic targets that await identification.
Furthermore, despite the rapid advances in our knowledge
regarding complex biological processes, pharmaceutical re-
search and development productivity has been in decline
since the mid-1990s (Figure 5A). In striking statistics drawn
from the pharmaceutical industry (2000 to 2008), only #25 to
30 new molecular entities per year are identified as viable
therapeutic agents. Clinical development of these compounds
necessitates an average of 13.9 years, and new agents carry a
probability of success of approximately 2.01% per year.67
Reasons for such declining productivity include a more
intense regulatory environment, the exhaustion of “easy” or
more obvious targets, an increasing attrition rate for devel-
oping drugs secondary to their unanticipated side effects, and,
from the perspective of this review, an intrinsically flawed
reductionistic approach to drug development centered on the
perceived need to identify a single drug target that entirely
reverses a given disease phenotype (ie, an Erlichian “magic
bullet”). Conversely, the failures of single-target drugs in
treating complex human disease have already suggested their
severe limitations when used alone and have prompted the
exploration of potent combinatorial drug therapies for a
variety of disease phenotypes (ie, tuberculosis, cancer, HIV/
AIDS, heart failure, etc). Furthermore, given the advances in
pharmacogenomics and improved understanding of how
drugs may be differentially effective based on individual
genotype, development of optimal drug regimens based on

Figure 4 (Continued). quantitative trait loci (QTL) data, to infer relationships between RNA levels and complex traits that are best sup-
ported by the data. Possible relationships (M1-M5) are shown here among QTL (L), RNA levels of a gene (R), and a complex trait (C).
Adapted from Schadt et al,55 with permission. E, Strategy of reverse engineering to infer a gene regulatory network in nonrecurrent pri-
mary and metastatic prostate cancer. In phase 1, high-throughput microarray data obtained from cancer cell lines or patient samples
are incorporated by the MNI algorithm to infer a model of regulatory interactions between genes (blue-filled circles indicate genes;
arrows indicate regulatory effects). In phase 2, test condition expression data are filtered using the reconstructed network to distin-
guish the genes affected by a condition (red-filled circles) from the remaining genes that exhibit changes in expression. Adapted from
Ergun et al,62 with permission.

Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014

368 Circulation Research July 20, 2012

Figure 5. Rationale for systems pharmacology. A, The declining productivity of pharmaceutical research is reflected by the steep
decline in the number of newly approved medications over the past decade. B, Comparison of the models used in conventional and
systems pharmacology to study the effects of drugs on a given system. C, Proposed approach for network-based drug discovery
involving first defining a disease module and then iteratively refining the map based on its emergent properties to identify appropriate
drug targets. Adapted from Loscalzo64 and Schadt et al,109 with permission.

their individualized effects is under further investigation and
of premium priority.68,69
Rapidly evolving methods of “systems pharmacology”70
facilitate a more rational and efficient approach to identifying
the most relevant “network-based” targets based on their
presence in relevant disease modules (Figure 5B).64 Given the
expansive small molecule libraries that have been mapped to
direct targets through large scale chemical biology reposito-
ries,71 network theory presents the opportunity not only to
discern previously unidentified targets for disease therapy
and to map novel pathogenic pathways but also to predict
how those targets can be manipulated, in isolation or in
combination. A generalized methodology has recently been
proposed for a systems-based method for drug discovery that
involves defining a disease module and iteratively refining
the map based on its emergent properties (Figure 5C).64 In
doing so, drugs can be predicted and/or produced that
rationally target key disease nodes followed by iterative
refinement in silico that account for its effects on the
emergent behavior of the disease module. Therefore, contrary
to traditional drug design, systems pharmacology may offer
elegant and cost-effective solutions to identifying multiple
drug combinations (“rational polypharmacy”)64 that maxi-
mize the therapeutic efficacy, minimize adverse side effects,
and minimize drug-drug interactions. As a result, we may
expect a much lower attrition rate as these agents are
clinically evaluated, decreased time from drug conception to
approval, and substantially higher rate of success for all
compounds that enter clinical testing (provided regulatory
bodies can adapt to this change in drug development strategy
and clinical trial design strategies [adaptive] follow suit72).
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
Identification and Targeting Novel
Drug-Host Interactions
Application of systems pharmacology to the identification of
novel drug-host interactions in complex human disease has
been reported with success. To identify a larger pool of
potential therapeutic targets in human disease that are not
limited by preexisting knowledge of disease genes, systems
pharmacology has been employed to generate a quantitative
framework within which to compare and contrast diseases by
an integrated analysis of disease-related mRNA expression
data and the human protein interaction network. For example,
Lamb et al have devised a process (“Connectivity Map”) of
identifying drug-dependent gene expression profiles that
correlate with known disease-dependent profiles,73 thus al-
lowing for a method for in silico screening of drugs that may
have salubrious effects in multiple disease contexts. A variant
of that approach has been pursued by Hu and Agarwal,74 in
which gene expression data are not prospectively collected
but can be directly obtained from an existing database, Gene
Expression Omnibus.75 A “guilt by association” method has
also been proposed by Chiang and Butte,76 whereby predic-
tions of novel associations between drugs and diseases can be
made by assuming that if two diseases are treated by the same
drug, alternative drugs treating only one of them might also
treat the other. More recent studies have attempted to make
predictions on a broader scale. Suthram et al identified 4620
functional modules in the human protein network and pro-
vided a quantitative metric to record their responses in 54
human diseases leading to 138 significant similarities be-
tween diseases77 (Figure 6A). Interestingly, 14 of the signif-
icant disease correlations also shared common drugs, support-
 Chan and Loscalzo Network Medicine and Human Disease 369

Figure 6. Current examples of drug discovery by systems pharmacology. A, Network of 138 significant disease correlations, 14 of
which encompass shared drug targets. The colors correspond to significant disease correlation categories after hierarchical clustering.
Adapted from Suthram et al,77 with permission. B, Reverse-engineered pathways related to aminoglycoside lethality. On comparing
treatment with a lethal aminoglycoside versus the bacteriostatic ribosome inhibitor, spectinomycin, gene expression changes were fil-
tered through an E coli gene connectivity map; the resulting gene networks were analyzed for functional enrichment. The strains exhib-
iting decreased growth or increased growth found from a high-throughput screen of an E coli single-gene deletion library treated with
gentamicin were overlaid on these networks. Adapted from Kohanski et al,87 with permission. C, Network map displaying the relation of
LQTS seed nodes (red) with 2 distinct drugs (blue), the Src inhibitor dasatinib (a cancer drug), and loperamide (an antidiarrheal) that
interacts with calmodulin (CALM1). Adapted from Berger et al,99 with permission. D, Differential side effects of CETP inhibitors (torce-
trapib, anacetrapib, and JTT705 or dalcetrapib) are predicted by their complex interactions with the PPAR network. Adapted from Xie
et al,101 with permission.

ing the hypothesis that similar diseases can be treated by the
same drugs, allowing for predictions for new uses of existing
drugs. Finally, the authors identified 59 modules that are
significantly enriched for genes that are known to be drug
targets and are dysregulated in at least half of the studied
diseases, representing a common disease-state “signature”
and highlighting the importance of these core modules as
prime therapeutic opportunities. Intriguingly, Gottlieb et al
have described a novel method (PREDICT) for the large-
scale prediction of drug indications. This approach is predi-
cated on the idea that similar drugs are indicated for similar
diseases. Thus, by quantifying and considering the known
similarities among drugs (ie, chemical structure, predicted
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
side effects, etc), networks of drug targets, and clinical
pathophenotypic similarities, a more comprehensive compen-
dium can be developed of all relevant drug indications for
human disease.78 Consequently, these types of studies may
form the computational foundation for future development of
personalized therapies, in which variations in gene expression
signatures and specific phenotypes from individual patients
could be considered in order to dictate the course of care.
Advances in “repurposing” or “repositioning” existing
drugs for alternative diseases have also benefited from
network methodology. High-throughput methods of literature
curation have proven effective in defining novel drug-disease
connections that can guide repurposing of known medica-
370 Circulation Research July 20, 2012
tions.79 For example, Li et al have constructed disease-
specific (eg, Alzheimer disease, in this case) drug-protein
connectivity maps derived solely from literature database
searches as a way to uncover a more comprehensive set of
known drugs acting in the same disease context.80 More
recently, commercial bioinformatics companies specializing
in the analysis of scientific literature, such as Ingenuity,
GeneGo, and Adriadne Genomics, have offered proprietary
pathway- and network-based solutions to the pharmaceutical
industry for drug repurposing.81 Beyond literature curation,
other groups have further developed a systematic computa-
tional approach to predict novel therapeutic indications on the
basis of comprehensive testing of molecular signatures in
drug-disease pairs.82 For instance, Sirota et al integrated gene
expression measurements from 100 diseases and gene expres-
sion measurements on 164 drug compounds, yielding pre-
dicted therapeutic potentials for these drugs. Previously
validated drug and disease relationships were predicted, as
well as a number of novel indications for these 164 drugs.
Importantly, one such prediction was experimentally vali-
dated for the antiulcer drug cimetidine as a candidate thera-
peutic in the treatment of lung adenocarcinoma, demonstrat-
ing its efficacy in vivo using mouse xenograft models.83 In a
separate set of experiments, another prediction regarding the
potential for the anticonvulsant topiramate in treatment of
inflammatory bowel disease was successfully validated in a
rodent model of inflammatory bowel disease.84 As a result,
the use of existing gene coexpression networks in combina-
tion with the known molecular interactions of existing drugs
has yielded more comprehensive insight into the pleiotropic
effects of various medicinal compounds.
A variation on the theme of repurposing drugs includes the
notion of rationally “rewiring” molecular networks such that
cells resistant to certain medications can become responsive
to pharmacological exposure. Recently, Lee et al undertook
such a project, utilizing an expansive systems-level approach
using comprehensive time-dependent measurements of sig-
naling networks, gene expression profiles, and cell pheno-
typic responses in combination with computational network
modeling of breast cancer cells resistant to genotoxic drugs.85
In doing so, the authors constructed and validated a molecular
roadmap to reprogram sequentially oncogenic signaling path-
ways and resensitize breast cancer cells to those drugs. Such
results indicate the feasibility of not only studying, but more
importantly manipulating, large-scale network dynamics to-
ward a desired cellular phenotype.
Perhaps the most striking successes of network pharmacol-
ogy to date are derived from studies of bacterial metabolism.
Owing to the fine detail of both the topology and rate
constants connecting metabolic networks in the bacterial
interactome, in silico predictions have been possible regard-
ing the effects on metabolite flux and consequent phenotypic
alterations after perturbation of specific nodes in those
pathways. As a result, by combining phenotypic and genetic
experiments with microarray analyses to identify relevant
bacterial networks important in antibiotic response, Kohanski
et al have demonstrated that all classes of bactericidal
antibiotics, regardless of their specific target, promote the
generation of lethal hydroxyl radicals.86 Kohanski et al have
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
further perturbed these bacterial networks leading to mis-
translation and misfolding of membrane proteins central to
aminoglycoside-induced oxidative stress and cell death, thus
expanding our understanding of the common mechanism of
killing induced by known bactericidal antibiotics (Figure
6B).87 Furthermore, potent new antimicrobial agents have
been identified and validated de novo based on their predicted
system-based metabolic responses on bacterial survival.54
Our understanding of systems-wide metabolic alterations
in diseased mammalian cells has also led to novel potential
therapeutic targets. For instance, by mapping the dynamic
fluxes of labeled metabolites in mammalian cells, Munger et
al noted an especially prominent increase in fatty acid
biosynthesis during human cytomegalovirus (HCMV) infec-
tion.88 Subsequent pharmacological targeting of fatty acid
biosynthesis suppressed the replication of HCMV and influ-
enza A, thereby confirming that fatty acid synthesis is
essential for the replication of two divergent viruses and,
importantly, that systems-level metabolic flux profiling can
identify metabolic targets for antiviral therapy.
In the field of cancer biology, Otto Warburg first described
a half a century ago that tumors exhibit enhanced glycolysis
coupled with a reduction in oxidative phosphorylation, de-
spite adequate oxygen availability. This “Warburg effect”
was proposed to be a key driver of tumor progression.89 Since
that initial observation, we now recognize that there exists a
systems-wide “reprogramming” of metabolic pathways asso-
ciated with alterations in bioenergetics and mitochondrial
function in transformed cells; for this reason, metabolic
interventions are now emerging as potential therapeutic
antineoplastic strategies.90 –92 Accordingly, the fine detail of
systems-level profiling available in cancer biology93,94 has
allowed for a growing number of network-based studies
reporting novel metabolic pathways important in driving
cellular transformation and thus identifying crucial metabolic
targets for pharmacological repression.95,96 Conversely, in the
future, pharmacological rescue of metabolic dysfunction
could be attempted in genetic metabolic diseases as a novel
alternative to gene therapy.
Identification of “Off-Target” Effects and
Drug-Drug Interactions
In addition to guiding the rational targeting of specific
pathways in complex diseases, network-based approaches
have better informed us about the systems properties of a full
drug interaction network that encompass currently unwanted
but otherwise unanticipated (so-called “off-target”) effects
and drug-drug interactions. For example, Yeh et al created a
network map of drug-drug interactions using a unique biolu-
minescence technique to provide quantitative measurements
of pairwise interactions among 21 antibiotics that affect
growth rate in Escherichia coli.97 Interestingly, the investi-
gators reported that the drug interaction network revealed 2
classes of drug that interact either purely synergistically or
purely antagonistically—network properties that correspond
directly to the cellular functions affected by the drugs. This
network approach introduced a new conceptual framework
for understanding the functional mechanisms of drugs and
their cellular targets, which could be expanded to other
 Chan and Loscalzo
FDA-approved medications or small molecule libraries. In-
terestingly, Torella et al have extended these studies of
antibiotic synergy by utilizing mathematical modeling of the
in vivo dynamics of bacterial infection to determine the
optimal strategies for utilizing synergism and antagonism to
suppress better the evolution of multi-drug resistance.98
Analysis of the consolidated interactome can yield even
greater insight into these pleiotropic and complicated molec-
ular interactions among various drugs. Recently, such a
pharmacological approach was applied to the long-QT syn-
drome (LQTS) to determine whether a network-based anal-
ysis could predict arrhythmic side effects of known FDA-
approved medications (Figure 6C).99 It is already established
that mutations in any of 12 genes encoding ion channels are
linked with congenital LQTS and potentially fatal ventricular
arrhythmias, such as torsades de pointes. In an elegant study,
Berger et al reasoned that drugs targeting proteins located in
the network “neighborhood” of bona fide LQTS genes should
also reveal arrhythmic side effects. On construction of a
protein-protein interactome, a LQTS “neighborhood” of
genes was identified by use of a mean first passage time
algorithm to reveal proteins that interact directly or with close
neighbors of seed LQTS genes. Importantly, the known
targets of certain FDA-approved drugs that are associated
with LQTS as an adverse event were enriched in the LQTS
neighborhood. By screening the remaining cache of FDA-
approved medications for those that target the LQTS neigh-
borhood, additional drugs were identified that probably in-
fluence QT length. These predictions were then validated by
cross-referencing with the Adverse Events Reporting System
of the FDA (FDA-AERS). Thus, identification and utilization
of disease-selective neighborhoods within the human interac-
tome can provide specific insight into predicting adverse
event susceptibility for new and “repurposed” drugs. Re-
cently, Tatonetti et al have developed comprehensive,
network-driven databases of drug effects and of drug-drug
interaction side effects and demonstrated novel predicted
adverse effects among cardiovascular drug classes.100
Alternatively, network approaches offer substantial insight
into the elucidation of the unanticipated “off-target” effects of
otherwise presumed direct targeted therapies. Pharmacologi-
cal examples of this scenario are plentiful, such as the
well-documented clinical failure of the cholesteryl ester
transfer protein (CETP) inhibitor torcetrapib.101 Despite its
efficacy in augmenting HDL levels and reducing triglyceride
levels, the use of torcetrapib surprisingly incurred risk of
substantial hypertension, prompting a halt to its development
following phase III clinical trials.102,103 Based on in silico
predictions of protein-ligand binding, Xie et al identified a
panel of off-targets for torcetrapib and other CETP inhibitors
from the human structural genome and mapped those targets
to a network of biological pathways curated from the litera-
ture.101 The predicted protein-ligand network was consistent
with experimental results revealing that the side effect of
hypertension by torcetrapib could be predicted by its target-
ing of the peroxisome proliferator-activated receptor (PPAR)
system. Importantly, second-generation CETP inhibitors are
not predicted to have the same magnitude of action on the
PPAR system and thus a lower likelihood of inducing
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
Network Medicine and Human Disease 371
hypertension (Figure 6D). Furthermore, using these networks
as a guide, the adverse drug effects of second generation
CETP inhibitors could be minimized by fine-tuning multiple
off-target interactions using single or multiple therapies.
Similarly, network approaches may also be effective in
predicting drug-drug interactions. Hansen et al have de-
scribed their preliminary experience with studying the antidi-
abetic thiazolidinedione, rosiglitazone, which has been with-
drawn from the market due to serious adverse side effects.65
According to the FDA-AERS, rosiglitazone is associated with
significant thrombotic complications as well as peripheral
edema and total body volume overload. Consequently, these
authors have constructed protein-protein interaction networks
based on Gene Ontology terms that relate to thrombosis and
fluid regulation. To determine the degree of overlap, these
subnetworks were then cross-referenced with direct gene
targets of PPAR!, a well-established direct target of rosigli-
tazone. Via utilization of FDA-AERS, drugs were then
identified, which, when used in combination with rosiglita-
zone, should increase or decrease associated complications of
thrombosis or edema by analyzing alterations in the effectors
shared as targets of rosiglitazone and as members in the above
subnetworks. These authors continue to develop algorithms to
predict whether combinations of drugs may target the same
pathway and whether they could potentiate therapeutic or
adverse consequences.
Conclusions
Network medicine offers insight into the integrative nature of
human pathogenesis by considering the dense interconnec-
tions among disease genes rather than merely considering
those genes in isolation. It allows for novel predictions of
disease genes, the interconnected actions of those genes in
pathogenesis, the molecular relationships among separate
disease conditions, and the concerted response to therapeutic
interventions. Although still in its early stages, this evolving
discipline offers a potentially powerful set of concepts and
tangible methods by which to improve rapidly our molecular
understanding of the networked nature of human disease and
to design medical therapies rooted in the complexity of these
biological connections. Nonetheless, despite benefiting from
the overdetermined property of biological networks, the
fidelity of network medicine is limited by the currently
incomplete nature of the interactome maps. Yet, while many
predictions derived from network methodology await confir-
mation, specific fundamental network principles have been
consistently supported by independent experimental proof in
human and animal models of disease. Such methods are even
beginning to permeate the commercial biotechnology sector,
as exemplified by companies such as Proteostasis (www.
proteostasis.com), which was founded on the concept of
using systems pharmacology to target efficiently the “protea-
some” gene network. In the very near future, we expect a
more widespread expansion of network methodology in
biomedicine. Consequently, “systems pathophysiology” may
become more of a reality as we explore not only intracellular
molecular networks but also the even less well understood
spatio-temporal interconnections among distinct cells, tis-
372 Circulation Research July 20, 2012
Figure 7. Network-based classification of human disease. In
this model, a specific disease module (white nodes), associated
with myocardial ischemia (module 1) and containing a known
disease gene (arrow), is functionally linked to 2 disease mod-
ules (modules 2 and 3; gray nodes) by shared nodes (black) and
edges. These shared nodes may regulate intermediate (patho)
phenotypes important in other disease processes, such as
inflammation or thrombosis. Influenced by a combination of
environmental and genetic events that also regulate these dis-
ease networks, individualized manifestations of myocardial
dysfunction result (early myocardial infarction, ventricular
tachycardia, or ischemic cardiomyopathy). By incorporating an
understanding of network function in these biological circuits,
these manifestations may be better predicted, treated, and pre-
vented. Adapted from Chan et al,107 with permission.
sues, and organ systems that have just begun to be explored
in cardiac morphogenesis.104
Thus far, network medicine has benefited from the exper-
tise of computer scientists, engineers, statistical physicists,
theoretical biologists, and biostatisticians to generate a range
of predictions regarding the systems-wide behavior of com-
plex biological systems that otherwise had not been accessi-
ble. Such work has led to the recent establishment of a
number of highly regarded journals that specialize in compu-
tational biological approaches, such as PLoS Computational
Biology, BMC Bioinformatics, and Molecular Systems Biol-
ogy, among others. However, most contemporary reports of
applying network theory to biomedical research have resulted
in predictions that remain theoretical and are yet to be
validated. Most likely, this shortcoming stems from limita-
tions in effective communication among computational sci-
entists, molecular and biomedical scientists, and translational
physician-scientists. Thus, theories are published that are not
perceived as relevant to biomedical or translational science,
are not easily validated using experimental biology, and,
perhaps most importantly, are not described in terms that are
easily accessible to colleagues with vastly different scientific
backgrounds. A more facile ability to integrate these currently
diverse scientific fields is warranted, and efforts are under-
way to train more formally the next generation of biologists
and biomedical scientists in network and systems ap-
proaches.105 In the meantime, the great potential of network
biology in translational medicine fundamentally relies on the
close collaboration of diverse scientists who can integrate
their expertise to coordinate the vast information contained in
biological networks into hypotheses that are testable by the
experimentalist.
Finally, by integrating the dense interconnections among
disease networks, a new framework has been proposed to
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
reclassify human disease and thus allow for a more expansive
consideration of the pathogenic mechanisms that are truly at
play (Figure 7).106 By incorporating the relationships of
individual disease genes, their connections within a disease
module, and that module’s shared links with separate disease
modules, insight can be garnered regarding the coordinated
actions of several disease modules as well as specific molec-
ular links that represent “intermediate” phenotypes (endop-
athophenotypes) shared among various disease conditions (ie,
inflammation, thrombosis, etc). Furthermore, genetic, envi-
ronmental, and even behavioral “nodes” can be represented
accurately by this approach. As a result, this type of disease
classification could predict a highly individualized clinical
pathophenotype that ultimately represents a more accurate
molecular schema on which to base future discovery in
personalized medicine.
Acknowledgments
We thank Stephanie Tribuna for expert administrative assistance and
Sol Chan for assistance with figures.
Sources of Funding
This work was supported in part by the NIH, the Pulmonary
Hypertension Association, Gilead Sciences, and the Lerner, Harris,
and Watkins funds (S.Y.C.); and NIH grants HL061795, HL48743,
HL107192, HL70819, and HL108630 (J.L.).
Disclosures
None.
References
1. Barabasi AL, Gulbahce N, Loscalzo J. Network medicine: a network-based
approach to human disease. Nat Rev Genet. 2011;12:56–68.
2. Salter M, Knowles RG, Pogson CI. Metabolic control. Essays Biochem.
1994;28:1–12.
3. Heinrich R, Schuster S. The Regulation of Cellular Systems. New York,
NY: Chapman and Hall; 1996.
4. Fell D. Understanding the Control of Metabolism. London, UK:
Portland Press; 1997.
5. Schuster S, Heinrich R. The definitions of metabolic control analysis
revisited. Biosystems. 1992;27:1–15.
6. Friedland AE, Lu TK, Wang X, Shi D, Church G, Collins JJ. Synthetic
gene networks that count. Science. 2009;324:1199 –1202.
7. Perumal TM, Gunawan R. Understanding dynamics using sensitivity
analysis: caveat and solution. BMC Syst Biol. 2011;5:41.
8. Alves R, Antunes F, Salvador A. Tools for kinetic modeling of bio-
chemical networks. Nat Biotechnol. 2006;24:667– 672.
9. Chen WW, Niepel M, Sorger PK. Classic and contemporary approaches
to modeling biochemical reactions. Genes Dev. 2010;24:1861–1875.
10. Ihekwaba AE, Broomhead DS, Grimley RL, Benson N, Kell DB. Sen-
sitivity analysis of parameters controlling oscillatory signalling in the
NF-kappaB pathway: the roles of IKK and IkappaBalpha. Syst Biol.
2004;1:93–103.
11. Ihekwaba AE, Broomhead DS, Grimley R, Benson N, White MR, Kell
DB. Synergistic control of oscillations in the NF-kappaB signalling
pathway. Syst Biol. 2005;152:153–160.
12. Chu Y, Jayaraman A, Hahn J. Parameter sensitivity analysis of IL-6
signalling pathways. IET Syst Biol. 2007;1:342–352.
13. Iwamoto K, Tashima Y, Hamada H, Eguchi Y, Okamoto M. Mathe-
matical modeling and sensitivity analysis of G1/S phase in the cell cycle
including the DNA-damage signal transduction pathway. Biosystems.
2008;94:109 –117.
14. Adler P, Peterson H, Agius P, Reimand J, Vilo J. Ranking genes by their
co-expression to subsets of pathway members. Ann N Y Acad Sci.
2009;1158:1–13.
15. Mardinoglu A, Nielsen J. Systems medicine and metabolic modelling.
J Intern Med. 2012;271:142–154.
 Chan and Loscalzo
16. Baughman JM, Nilsson R, Gohil VM, Arlow DH, Gauhar Z, Mootha
VK. A computational screen for regulators of oxidative phosphorylation
implicates SLIRP in mitochondrial RNA homeostasis. PLoS Genet.
2009;5:e1000590.
17. Cusick ME, Yu H, Smolyar A, Venkatesan K, Carvunis AR, Simonis N,
Rual JF, Borick H, Braun P, Dreze M, Vandenhaute J, Galli M, Yazaki
J, Hill DE, Ecker JR, Roth FP, Vidal M. Literature-curated protein
interaction datasets. Nat Methods. 2009;6:39 – 46.
18. Rual JF, Venkatesan K, Hao T, et al. Towards a proteome-scale map of
the human protein-protein interaction network. Nature. 2005;437:
1173–1178.
19. Stelzl U, Worm U, Lalowski M, et al. A human protein-protein inter-
action network: a resource for annotating the proteome. Cell. 2005;122:
957–968.
20. Ewing RM, Chu P, Elisma F, et al. Large-scale mapping of human
protein-protein interactions by mass spectrometry. Mol Syst Biol.
2007;3:89.
21. Venkatesan K, Rual JF, Vazquez A, et al. An empirical framework for
binary interactome mapping. Nat Methods. 2009;6:83–90.
22. Dreze M, Monachello D, Lurin C, Cusick ME, Hill DE, Vidal M, Braun
P. High-quality binary interactome mapping. Methods Enzymol. 2010;
470:281–315.
23. Zhao C, Sethuraman M, Clavreul N, Kaur P, Cohen RA, O’Connor PB.
Detailed map of oxidative post-translational modifications of human
p21ras using Fourier transform mass spectrometry. Anal Chem. 2006;
78:5134 –5142.
24. Grunau C, Renault E, Rosenthal A, Roizes G. MethDB: a public
database for DNA methylation data. Nucleic Acids Res. 2001;29:
270 –274.
25. Negre V, Grunau C. The MethDB DAS server: adding an epigenetic
information layer to the human genome. Epigenetics. 2006;1:101–105.
26. Bartel DP. MicroRNAs: target recognition and regulatory functions.
Cell. 2009;136:215–233.
27. Watanabe Y, Kanai A. Systems biology reveals microRNA-mediated
gene regulation. Front Genet. 2009;2:29.
28. Vergoulis T, Vlachos IS, Alexiou P, Georgakilas G, Maragkakis M,
Reczko M, Gerangelos S, Koziris N, Dalamagas T, Hatzigeorgiou AG.
TarBase 6.0: capturing the exponential growth of miRNA targets with
experimental support. Nucleic Acids Res. 2011;40(Database issue):
D222–D229.
29. Parikh VN, Jin RC, Rabello S, et al. MicroRNA-21 integrates
pathogenic signaling to control pulmonary hypertension: results of a
network bioinformatics approach. Circulation. 2012;125:1520 –1532.
30. Lin CC, Chen YJ, Chen CY, Oyang YJ, Juan HF, Huang HC. Crosstalk
between transcription factors and microRNAs in human protein inter-
action network. BMC Syst Biol. 2012;6:18.
31. Gennarino VA, D’Angelo G, Dharmalingam G, Fernandez S, Russolillo
G, Sanges R, Mutarelli M, Belcastro V, Ballabio A, Verde P, Sardiello
M, Banfi S. Identification of microRNA-regulated gene networks by
expression analysis of target genes. Genome Res. 2012;22:1163–1172.
32. Berezikov E, Guryev V, van de Belt J, Wienholds E, Plasterk R, Cuppen
E. Phylogenetic shadowing and computational identification of human
microRNA genes. Cell. 2005;120:21–24.
33. Friedman RC, Farh KK, Burge CB, Bartel DP. Most mammalian
mRNAs are conserved targets of microRNAs. Genome Res. 2009;19:
92–105.
34. Shirdel EA, Xie W, Mak TW, Jurisica I. NAViGaTing the micronome:
using multiple microRNA prediction databases to identify signalling
pathway-associated microRNAs. PLoS One. 2011;6:e17429.
35. Loscalzo J, Barabasi AL. Systems biology and the future of medicine.
Wiley Interdiscip Rev Syst Biol Med. 2011;3:619 – 627.
36. Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabasi AL. The
human disease network. Proc Natl Acad Sci U S A. 2007;104:
8685– 8690.
37. Dudley JT, Tibshirani R, Deshpande T, Butte AJ. Disease signatures are
robust across tissues and experiments. Mol Syst Biol. 2009;5:307.
38. Reimand J, Tooming L, Peterson H, Adler P, Vilo J. GraphWeb: mining
heterogeneous biological networks for gene modules with functional
significance. Nucleic Acids Res. 2008;36(Web Server issue):
W452–W459.
39. Loscalzo J, Kohane I, Barabasi AL. Human disease classification in the
postgenomic era: a complex systems approach to human pathobiology.
Mol Syst Biol. 2007;3:124.
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
Network Medicine and Human Disease 373
40. Jensen MK, Pers TH, Dworzynski P, Girman CJ, Brunak S, Rimm EB.
Protein interaction-based genome-wide analysis of incident coronary
heart disease. Circ Cardiovasc Genet. 2011;4:549 –556.
41. Ghazalpour A, Doss S, Zhang B, Wang S, Plaisier C, Castellanos R,
Brozell A, Schadt EE, Drake TA, Lusis AJ, Horvath S. Integrating
genetic and network analysis to characterize genes related to mouse
weight. PLoS Genet. 2006;2:e130.
42. Wang SS, Schadt EE, Wang H, Wang X, Ingram-Drake L, Shi W, Drake
TA, Lusis AJ. Identification of pathways for atherosclerosis in mice:
integration of quantitative trait locus analysis and global gene expression
data. Circ Res. 2007;101:e11– e30.
43. Diez D, Wheelock AM, Goto S, Haeggstrom JZ, Paulsson-Berne G,
Hansson GK, Hedin U, Gabrielsen A, Wheelock CE. The use of network
analyses for elucidating mechanisms in cardiovascular disease. Mol
Biosyst. 2010;6:289 –304.
44. Romanoski CE, Che N, Yin F, Mai N, Pouldar D, Civelek M, Pan C, Lee
S, Vakili L, Yang WP, Kayne P, Mungrue IN, Araujo JA, Berliner JA,
Lusis AJ. Network for activation of human endothelial cells by oxidized
phospholipids: a critical role of heme oxygenase 1. Circ Res. 2011;109:
e27– e41.
45. Civelek M, Manduchi E, Riley RJ, Stoeckert CJ Jr, Davies PF. Coronary
artery endothelial transcriptome in vivo: identification of endoplasmic
reticulum stress and enhanced reactive oxygen species by gene connec-
tivity network analysis. Circ Cardiovasc Genet. 2011;4:243–252.
46. Ashley EA, Ferrara R, King JY, et al. Network analysis of human
in-stent restenosis. Circulation. 2006;114:2644 –2654.
47. Sengupta U, Ukil S, Dimitrova N, Agrawal S. Expression-based network
biology identifies alteration in key regulatory pathways of type 2
diabetes and associated risk/complications. PLoS One. 2009;4:e8100.
48. Dewey FE, Wheeler MT, Ashley EA. Systems biology of heart failure,
challenges and hopes. Curr Opin Cardiol. 2011;26:314 –321.
49. Vanunu O, Magger O, Ruppin E, Shlomi T, Sharan R. Associating genes
and protein complexes with disease via network propagation. PLoS
Comput Biol. 2010;6:e1000641.
50. Navlakha S, Kingsford C. The power of protein interaction networks for
associating genes with diseases. Bioinformatics. 2010;26:1057–1063.
51. Ray M, Ruan J, Zhang W. Variations in the transcriptome of Alzhei-
mer’s disease reveal molecular networks involved in cardiovascular
diseases. Genome Biol. 2008;9:R148.
52. Rende D, Baysal N, Kirdar B. A novel integrative network approach to
understand the interplay between cardiovascular disease and other
complex disorders. Mol Biosyst. 2011;7:2205–2219.
53. Dewey FE, Perez MV, Wheeler MT, Watt C, Spin J, Langfelder P,
Horvath S, Hannenhalli S, Cappola TP, Ashley EA. Gene coexpression
network topology of cardiac development, hypertrophy, and failure. Circ
Cardiovasc Genet. 2010;4:26 –35.
54. Shen Y, Liu J, Estiu G, Isin B, Ahn YY, Lee DS, Barabasi AL, Kapatral
V, Wiest O, Oltvai ZN. Blueprint for antimicrobial hit discovery tar-
geting metabolic networks. Proc Natl Acad Sci U S A. 2010;107:
1082–1087.
55. Schadt EE, Lamb J, Yang X, et al. An integrative genomics approach to
infer causal associations between gene expression and disease. Nat
Genet. 2005;37:710 –717.
56. Zuk O, Hechter E, Sunyaev SR, Lander ES. The mystery of missing
heritability: genetic interactions create phantom heritability. Proc Natl
Acad Sci U S A. 2012;109:1193–1198.
57. Li H, Xuan J, Wang Y, Zhan M. Inferring regulatory networks. Front
Biosci. 2008;13:263–275.
58. Gardner TS, di Bernardo D, Lorenz D, Collins JJ. Inferring genetic
networks and identifying compound mode of action via expression
profiling. Science. 2003;301:102–105.
59. Lorenz DR, Cantor CR, Collins JJ. A network biology approach to aging
in yeast. Proc Natl Acad Sci U S A. 2009;106:1145–1150.
60. Tegner J, Yeung MK, Hasty J, Collins JJ. Reverse engineering gene
networks: integrating genetic perturbations with dynamical modeling.
Proc Natl Acad Sci U S A. 2003;100:5944 –5949.
61. Modi SR, Camacho DM, Kohanski MA, Walker GC, Collins JJ. Func-
tional characterization of bacterial sRNAs using a network biology
approach. Proc Natl Acad Sci U S A. 2011;108:15522–15527.
62. Ergun A, Lawrence CA, Kohanski MA, Brennan TA, Collins JJ. A
network biology approach to prostate cancer. Mol Syst Biol. 2007;3:82.
63. Guido NJ, Wang X, Adalsteinsson D, McMillen D, Hasty J, Cantor CR,
Elston TC, Collins JJ. A bottom-up approach to gene regulation. Nature.
2006;439:856 – 860.
374 Circulation Research July 20, 2012
64. Loscalzo J. Personalized cardiovascular medicine and drug devel-
opment: time for a new paradigm. Circulation. 2012;125:638 – 645.
65. Hansen J, Zhao S, Iyengar R. Systems pharmacology of complex
diseases. Ann N Y Acad Sci. 2011;1245:E1–E5.
66. Campillos M, Kuhn M, Gavin AC, Jensen LJ, Bork P. Drug target
identification using side-effect similarity. Science. 2008;321:263–266.
67. Pammolli F, Magazzini L, Riccaboni M. The productivity crisis in
pharmaceutical R&D. Nat Rev Drug Discov. 2011;10:428 – 438.
68. Ashley EA, Butte AJ, Wheeler MT, et al. Clinical assessment incorpo-
rating a personal genome. Lancet. 2011;375:1525–1535.
69. Suhre K, Shin SY, Petersen AK, et al. Human metabolic individuality in
biomedical and pharmaceutical research. Nature. 2011;477:54 – 60.
70. Sorger PK, Allerheiligen SRB, Abernethy DR, et al. Quantitative and
systems pharmacology in the post-genomic era: new approaches to
discovering drugs and understanding therapeutic mechanisms. An NIH
White Paper by the QSP Workshop Group; October, 2011. Bethesda,
MD: NIH; 2011:1– 47.
71. Zanders ED. Overview of chemical genomics and proteomics. Methods
Mol Biol. 2011;800:3–10.
72. Antman E, Weiss S, Loscalzo J. Systems pharmacology, pharmacoge-
netics, and clinical trial design in network medicine. Wiley Interdiscip
Rev Syst Biol Med. 2012;4:367–383.
73. Lamb J, Crawford ED, Peck D, et al. The Connectivity Map: using
gene-expression signatures to connect small molecules, genes, and
disease. Science. 2006;313:1929 –1935.
74. Hu G, Agarwal P. Human disease-drug network based on genomic
expression profiles. PLoS One. 2009;4:e6536.
75. Edgar R, Domrachev M, Lash AE. Gene Expression Omnibus: NCBI
gene expression and hybridization array data repository. Nucleic Acids
Res. 2002;30:207–210.
76. Chiang AP, Butte AJ. Systematic evaluation of drug-disease rela-
tionships to identify leads for novel drug uses. Clin Pharmacol Ther.
2009;86:507–510.
77. Suthram S, Dudley JT, Chiang AP, Chen R, Hastie TJ, Butte AJ.
Network-based elucidation of human disease similarities reveals
common functional modules enriched for pluripotent drug targets. PLoS
Comput Biol. 2010;6:e1000662.
78. Gottlieb A, Stein GY, Ruppin E, Sharan R. PREDICT: a method for
inferring novel drug indications with application to personalized
medicine. Mol Syst Biol. 2011;7:496.
79. Loging W, Harland L, Williams-Jones B. High-throughput electronic
biology: mining information for drug discovery. Nat Rev Drug Discov.
2007;6:220 –230.
80. Li J, Zhu X, Chen JY. Building disease-specific drug-protein connec-
tivity maps from molecular interaction networks and PubMed abstracts.
PLoS Comput Biol. 2009;5:e1000450.
81. Deftereos SN, Andronis C, Friedla EJ, Persidis A. Drug repurposing and
adverse event prediction using high-throughput literature analysis. Wiley
Interdiscip Rev Syst Biol Med. 2011;3:323–334.
82. Achenbach J, Tiikkainen P, Franke L, Proschak E. Computational tools
for polypharmacology and repurposing. Future Med Chem. 2011;3:
961–968.
83. Sirota M, Dudley JT, Kim J, Chiang AP, Morgan AA, Sweet-Cordero A,
Sage J, Butte AJ. Discovery and preclinical validation of drug indi-
cations using compendia of public gene expression data. Sci Transl Med.
2011;3:96ra77.
84. Dudley JT, Sirota M, Shenoy M, Pai RK, Roedder S, Chiang AP,
Morgan AA, Sarwal MM, Pasricha PJ, Butte AJ. Computational repo-
sitioning of the anticonvulsant topiramate for inflammatory bowel
disease. Sci Transl Med. 2011;3:96ra76.
85. Lee MJ, Ye AS, Gardino AK, Heijink AM, Sorger PK, Macbeath G,
Yaffe MB. Sequential application of anticancer drugs enhances cell
death by rewiring apoptotic signaling networks. Cell. 2012;149:
780 –794.
86. Kohanski MA, Dwyer DJ, Hayete B, Lawrence CA, Collins JJ. A
common mechanism of cellular death induced by bactericidal anti-
biotics. Cell. 2007;130:797– 810.
Downloaded from http://circres.ahajournals.org/ by guest on May 9, 2014
87. Kohanski MA, Dwyer DJ, Wierzbowski J, Cottarel G, Collins JJ. Mis-
translation of membrane proteins and two-component system activation
trigger antibiotic-mediated cell death. Cell. 2008;135:679 – 690.
88. Munger J, Bennett BD, Parikh A, Feng XJ, McArdle J, Rabitz HA,
Shenk T, Rabinowitz JD. Systems-level metabolic flux profiling iden-
tifies fatty acid synthesis as a target for antiviral therapy. Nat Biotechnol.
2008;26:1179 –1186.
89. Warburg O. On the origin of cancer cells. Science. 1956;123:309 –314.
90. Vander Heiden MG, Christofk HR, Schuman E, Subtelny AO, Sharfi H,
Harlow EE, Xian J, Cantley LC. Identification of small molecule inhib-
itors of pyruvate kinase M2. Biochem Pharmacol. 2009;79:1118 –1124.
91. Michelakis ED, Sutendra G, Dromparis P, Webster L, Haromy A, Niven
E, Maguire C, Gammer TL, Mackey JR, Fulton D, Abdulkarim B,
McMurtry MS, Petruk KC. Metabolic modulation of glioblastoma with
dichloroacetate. Sci Transl Med. 2010;2:31ra34.
92. Wise DR, Thompson CB. Glutamine addiction: a new therapeutic target
in cancer. Trends Biochem Sci. 2010;35:427– 433.
93. Kreeger PK, Lauffenburger DA. Cancer systems biology: a network
modeling perspective. Carcinogenesis. 2009;31:2– 8.
94. Laubenbacher R, Hower V, Jarrah A, Torti SV, Shulaev V, Mendes P,
Torti FM, Akman S. A systems biology view of cancer. Biochim
Biophys Acta. 2009;1796:129 –139.
95. Wellen KE, Lu C, Mancuso A, Lemons JM, Ryczko M, Dennis JW,
Rabinowitz JD, Coller HA, Thompson CB. The hexosamine biosyn-
thetic pathway couples growth factor-induced glutamine uptake to
glucose metabolism. Genes Dev. 2010;24:2784 –2799.
96. Gaglio D, Metallo CM, Gameiro PA, Hiller K, Danna LS, Balestrieri C,
Alberghina L, Stephanopoulos G, Chiaradonna F. Oncogenic K-Ras
decouples glucose and glutamine metabolism to support cancer cell
growth. Mol Syst Biol. 2011;7:523.
97. Yeh P, Tschumi AI, Kishony R. Functional classification of drugs by
properties of their pairwise interactions. Nat Genet. 2006;38:489 – 494.
98. Torella JP, Chait R, Kishony R. Optimal drug synergy in antimicrobial
treatments. PLoS Comput Biol. 2010;6:e1000796.
99. Berger SI, Ma’ayan A, Iyengar R. Systems pharmacology of arrhyth-
mias. Sci Signal. 2010;3:ra30.
100. Tatonetti NP, Ye PP, Daneshjou R, Altman RB. Data-driven prediction
of drug effects and interactions. Sci Transl Med. 2012;4:125ra31.
101. Xie L, Li J, Bourne PE. Drug discovery using chemical systems biology:
identification of the protein-ligand binding network to explain the side
effects of CETP inhibitors. PLoS Comput Biol. 2009;5:e1000387.
102. Nissen SE, Tardif JC, Nicholls SJ, Revkin JH, Shear CL, Duggan WT,
Ruzyllo W, Bachinsky WB, Lasala GP, Tuzcu EM. Effect of torcetrapib
on the progression of coronary atherosclerosis. N Engl J Med. 2007;
356:1304 –1316.
103. Barter PJ, Caulfield M, Eriksson M, et al. Effects of torcetrapib in
patients at high risk for coronary events. N Engl J Med. 2007;357:
2109 –2122.
104. Lage K, Mollgard K, Greenway S, et al. Dissecting spatio-temporal
protein networks driving human heart development and related dis-
orders. Mol Syst Biol. 2010;6:381.
105. Sobie EA, Lee YS, Jenkins SL, Iyengar R. Systems biology: biomedical
modeling. Sci Signal. 2011;4:tr2.
106. Loscalzo J, Kohane I, Barabasi A. Human disease classification in the
postgenomic era: a complex systems approach to human pathobiology.
Mol Syst Biol. 2007;3:124.
107. Chan SY, White K, Loscalzo J. Deciphering the molecular basis of
human cardiovascular disease through network biology. Curr Opin
Cardiol. 2012;27:202–209.
108. Morel NM, Holland JM, van der Greef J, Marple EW, Clish C, Loscalzo
J, Naylor S. Primer on medical genomics, part XIV: introduction to
systems biology: a new approach to understanding disease and
treatment. Mayo Clin Proc. 2004;79:651– 658.
109. Schadt EE, Friend SH, Shaywitz DA. A network view of disease and
compound screening. Nat Rev Drug Discov. 2009;8:286 –295.