Content uploaded by Hiroaki Kitano
Author content
All content in this area was uploaded by Hiroaki Kitano on Jan 09, 2015
Content may be subject to copyright.
Advances in genomic research in the past two
decades fuelled the expectation that agents
that specifically target a single disease-causing
molecule — ‘magic bullets’ — could cure
cancer, diabetes and other complex diseases.
However, although some such agents have
proven to be successful drugs, many others
have been found to be ineffective or to cause
significant side effects. This disappointing
outcome highlights the limitations of the
single-target–drug paradigm
1–3
, and is
considered by some to be the underlying
cause of stagnation in productivity in the
pharmaceutical industry.
Given these limitations, the production
of effective combinatorial drugs, drugs with
multiple targets and advances in systems
biology
3–6
have led to an increased interest
in systems-oriented approaches to drug
discovery. These include multicomponent
approaches
7
, synthetic lethality
8
,
high-throughput screening and data
integration
5,6,9
, cell-based systems biology
assays
3,4
and metabolome-based
approaches
10–12
. As such approaches
continue to emerge, the establishment of a
theoretical and conceptual framework that
can cope with the large volume of biological
data generated by today’s technology could
have a key role in realizing their promise.
An analysis of both successful and failed
drugs reveals that one fundamental factor
that determines the success of a drug is how
the robustness and fragility of the biologi-
cal systems in terms of disease onset and
progression are exploited. Living systems are
generally robust against various perturba-
tions, such as mutation, toxins and environ-
mental changes, but can be extremely fragile
when faced with perturbations for which the
system has not been optimized. Drugs can
be ineffective when the inherent robustness
of the systems in patients (or pathogens in
some cases) that are being targeted compen-
sates for any changes caused by drugs.
By contrast, drug side effects can be the
result of interference with an unexpected
point of fragility of these systems.
Understanding the relationships between
robustness, disease, and drug efficacy
and side effects is the first step towards
the design of drugs that can target robust
systems to achieve the desired therapeutic
goals. With this in mind, using a conceptual
framework that is based on the complexities
of biological systems, as well as control and
communication theories from engineering,
this article considers the success and failure
of drugs from the perspective of biological
robustness, and suggests a basis for strategies
for systems-oriented drug design.
Basics of biological robustness
Robustness is an intrinsic property of
biological systems that enables them to
maintain their functions in the face of vari-
ous perturbations, and has been reviewed
extensively elsewhere
13–16
. Biological systems
have to be robust yet evolvable, which
requires trade-offs between robustness,
fragility, resource demands and perform-
ance
17,18
. This imposes constraints on the
type of architecture and mechanisms that
constitute such systems. Recent research
has identified that robustness is enabled by
four basic mechanisms
13
: systems control;
fail-safe by means of redundancy and diver-
sity; modularity; and decoupling in which
physical-level perturbation is isolated from
functional-level activities of the system.
Systems control introduces negative-
feedback, positive-feedback, feed-forward
regulation and other regulatory loops to
maintain the homeostasis of the system, as
well as bistable behaviours
19,20
, which enable
the system to move between two stable
states. For example, the effects of a drug can
be neutralized if a negative-feedback loop
compensates for changes in the level of the
molecule that the drug targets.
Fail-safe mechanisms enable a system
to continue to function even if one of its
molecules or pathways is disabled, because
there are other molecules or pathways that
have partly overlapping functions that can
compensate. This implies that drugs that
target molecules in such pathways are likely
to have limited efficacy.
Modularity blocks perturbations that are
applied to a specific module from spreading
throughout the system. Although this is
useful to prevent local damage from causing
system-wide pertubations, it can also localize
drug effects.
Decoupling isolates a functional level
from a physical level. An example is digital
systems in which the functional level that is
a bit-stream of on and off is isolated from
the actual voltage fluctuation of electric
circuits. So, the functional layer is robust
against a minor change in the physical layer.
A biological example of decoupling is the
correction of protein folding by heat-shock
proteins (HSP) such as HSP90
(REF. 21).
In addition to these mechanisms, there
is a specific network architecture often
recognized as ‘bow-tie’ networks
16
that
can be observed in molecular-interaction
networks for various functional subsystems,
including metabolic networks
22
, signalling
networks
23
and the immune system
24
.
The bow-tie architecture is composed of a
highly conserved and often robust core part
of the network connected by diverse and
redundant input and output subnetworks
with various feedback control loops
(FIG. 1a).
This network architecture provides robust
and flexible responses to various stimuli and
inhibition of molecules involved owing to
the redundancy of pathways in input and
output modules. However, the network
architecture can also be fragile in response to
the perturbation of molecules that constitute
INNOVATION
A robustness-based approach
to systems-oriented drug design
Hiroaki Kitano
Abstract | Many potential drugs that specifically target a particular protein
considered to underlie a given disease have been found to be less effective than
hoped, or to cause significant side effects. The intrinsic robustness of living systems
against various perturbations is a key factor that prevents such compounds from
being successful. By studying complex network systems and reformulating control
and communication theories that are well established in engineering, a theoretical
foundation for a systems-oriented approach to more effectively control the
robustness of living systems, particularly at the cellular level, could be developed.
Here, I use examples that are based on existing drugs to illustrate the concept of
robustness, and then discuss how a greater consideration of the importance
of robustness could influence the design of new drugs that will be intended to
control complex systems.
PERSPECTIVES
202
|
MARCH 2007
|
VOLUME 6 www.nature.com/reviews/drugdisc
© 2007 Nature Publishing Group
the core network when it is not built to be
robust to such perturbations, perhaps by
not being redundant or by missing regula-
tory feedback. For example, the core of
the bow-tie network found in the Toll-like
receptor signalling cascade is fragile as it has
a non-redundant core with
MYD88 (myeloid
differentiation primary response gene 88)
25
.
Complex systems that have evolved (or
have been designed) to be optimal (or sub-
optimal) have a ‘robust yet fragile’ feature
17,18
.
It has been asserted that the robust-yet-fragile
feature is an inherent systems property, in
which robustness in response to certain
perturbations is inevitably associated with
fragility in response to other perturbations
15
.
Furthermore, evolved systems may be
constrained by trade-offs between robust-
ness, fragility, performance and resource
demands
13
. Evolved complex networks
tend to demonstrate robustness in response
to removal of nodes, but are vulnerable to
sustained malfunction of nodes. An example
of such a network is the internet, which is
robust against the removal of some of its
nodes, such as routers (as such a perturba-
tion would only affect local users), but
fragile against the malfunctions of nodes,
such as the persistent sending of spam and
viruses, which often have global conse-
quences and continue to pose threats to the
system
26
. This type of failure, in which nodes
of the network continue to function but in
an undesirable way, is often called ‘fail-on’
failure, in contrast with ‘fail-off ’ failure in
which failed nodes are removed. Fail-on
failure (also known as Byzantine failure) has
been a major problem for communication
networks, and is considered to be a crucial
problem in complex network systems in
general
27
. Biological networks are expected
to share similar characteristics — namely
that the most serious threats to the system
are unwanted interactions by mutant or
invading components, overexpression
or amplification of genes and uncontrollable
hyperactivation of their regulatory loops.
Robustness and disease
Disease can be viewed as a breakdown of the
robustness of normal physiological systems
and the re-establishment of robust, and
potentially progressive, disease states. From
the aspect of robustness trade-offs, diseases
can be classified into one or more of the fol-
lowing: a simple disease, with a breach in the
robustness that protects the host system from
various perturbations; a parasitic disease, in
which the mechanisms for robustness are
hijacked for the triggering and progression
of the disease state in the face of various
perturbations, including therapeutic inter-
ventions; and a systemic disease, in which
the mechanisms that provide robustness in
a certain environment work unfavourably
under a current environment that is signifi-
cantly different from the conditions that the
system has adapted to
24,28–30
.
Cancer and infectious diseases can be
considered as parasitic diseases, because
tumour cells and pathogens form a parasitic
symbiosis with the host by hijacking the
host’s mechanisms of robustness to protect
and proliferate themselves, as well as to estab-
lish robustness mechanisms of their own. For
example, in cancer, tumours establish robust-
ness against various therapeutic perturba-
tions through various mechanisms, such as
multiple-drug resistance, micro-environment
remodelling, hypoxia-inducible factor 1
(HIF1) upregulation and intra-tumoural
genetic heterogeneity — mechanisms that
are consistent with the general framework of
a Bow-tie architecture in a cellular system
b Genetic diversity of tumour cells and pathogens
c Diversity of signalling pathways
Nutrients Metabolites
Environmental stimuli
Metabolic
bow-tie
Signalling
bow-tie
Regulation
Physiological responses
Perturbation Proliferation
Figure 1 | Basic mechanisms for robustness. Some of the basic mechanisms that contribute to the
robustness of biological systems are illustrated. a | A ‘bow-tie’ structure of networks might exist for
metabolic networks and signalling networks. A similar figure is also depicted by Csete and Doyle
16
.
Bow-tie networks often have an asymmetrical structure with a large ‘fan-in’ wing with diverse and
redundant pathways that converge into the core of the bow-tie, and a ‘fan-out’ wing with moderate
diversity as the number of possible responses are limited. The metabolic bow-tie and the signalling
bow-tie are interlocked to form a global cellular system. b | Pathogens and tumour cells have a high
level of genetic diversity. Therapeutic perturbations might be able to eliminate a subset of patho-
gens or tumour cells that respond to drugs. However, there are survivors that do not respond to
drugs. These drug-resistant survivors proliferate to form a more resistant group of pathogens and
tumour cells. Diversity and redundancy are a main source of robustness for many biological systems.
c | Diversity of signalling pathways, by means of overlapping functions and cross-talk between
different signalling pathways, contributes to the robustness of cells against inhibition of one or more
of their signalling pathways. Some anticancer drugs that target molecules in a single pathway can
be made ineffective owing to other pathways that cross-talk with that pathway, and so proliferation
of tumour cells is not prevented. Diversity of signalling pathways in cellular networks is one of the
main obstacles for anticancer drugs.
PERSPECTIVES
NATURE REVIEWS
|
DRUG DISCOVERY VOLUME 6
|
MARCH 2007
|
203
© 2007 Nature Publishing Group
robust systems
29,30
. The counteraction of the
therapeutic intervention by enhanced genetic
diversity is commonly observed, so that
drugs are only effective for a subpopulation
of tumour cells or pathogens. Both tumour
cells and HIV, for example, are known to
possess a high level of genetic variation, and
the emergence of drug-resistant pathogen
variants and the transfer of drug-resistance
genes through horizontal gene transfer
provide similar mechanisms at a population
level. With these mechanisms in mind,
strategies for drug design include finding
the point of fragility of the target cells,
circumventing robustness mechanisms that
the target cells might exploit and pre-empt-
ing possible enhancement of the robustness
of the target cells against specific drug inter-
ventions, with the aim being to remove the
target cells or induce dormancy.
Diseases such as t
ype 2 diabetes and
autoimmune disorders can be considered
systemic diseases, as malfunctions of
systemic control cause the symptoms of the
disorders. The typical indication of type 2
diabetes is a persistently high plasma-glucose
level due to insulin resistance. However, in
the ancient environment during evolution,
insulin resistance might have been a mecha-
nism for providing robustness against near-
starvation levels of food supply in a highly
pathogenic and hostile environment by
maintaining a higher blood-sugar level (on
which neuronal cells and innate-immune-
related cells are dependent). So, the system
might have adapted on an evolutionary
timescale
28
to provide robustness against
these historical environmental threats, but it
is now working undesirably in the situation
of a totally different lifestyle. In this type of
disease, there is no target cell to be removed,
or induced to be dormant. There are issues
such as β-cell dysfunction that merit cell-
type-specific intervention, but such strategies
tend to provide only a temporary solution if
disorders of the systemic level network are
left untouched, as most dysfunctions at the
level of specific cell types are manifestations
of cascading failures that stem from systemic
disorders. Mechanisms for robustness are
distributed throughout the organism, and
drugs have to be designed to control disease
status in a systemic manner.
Robustness and drug efficacy
The concept of robustness in biological
systems can be valuable for interpreting the
efficacy and side effects of drugs. This con-
cept is valid from the systemic physiology
level to the cellular level, but owing to space
limitations, this article will focus only on
the cellular level. Drugs are often ineffective
owing to the robustness of cellular systems
against the intended interventions, and can
cause side effects when they interact with
fragile points of off-target cells. Several cases
are analysed below to illustrate these points.
Drugs that target a single specific molecule
are most effective when the target molecule is
only observed in cells and pathogens that
trigger disease symptoms and is the single
causative factor that maintains the disease
state. Inhibition of the target molecule alone
can mitigate, if not eliminate, the symptoms
without harming other cells. The target
molecule, therefore, is the point of fragility
of the system that maintains the disease
state. Some anti-infectives are examples
of this type of drug, as most of the target
molecules only exist in pathogens, but not
in the host cells, and inhibition of the target
molecule can significantly reduce disease
symptoms. For example, Vibrio cholerae is
a pathogen that causes epidemics in many
developing countries. It produces a toxin
that interacts with a G-protein and causes
diarrhoea. Cholera toxin is encoded in a
pathogenicity island (PAI), and is consid-
ered to be acquired through horizontal
gene transfer
31
. A small-molecule inhibitor,
4-N-(1,8-naphthalimide)-n-butyric acid
(virstatin), suppresses the expression of the
V. cho le ra e virulence factors, cholera toxin
and toxin-coregulated pilus, by inhibiting
the transcriptional regulator ToxT
32
.
This agent is successful because ToxT only
exists in pathogens, and it is the single
transcription factor that controls virulence
factors, which therefore implies that the
pathogen system is not robust against
the inhibition of ToxT. The situation could
have been different if the virulence factors
were controlled by multiple transcription
factors with no common binding site for
potential drugs — a system that resembles
a fail-safe mechanism through its diversity
of transcription factors.
However, drug-resistant infections involve
the emergence of one or more bacteria
that acquire drug-resistance, followed by
horizontal transfer of genes that are respon-
sible for the drug-resistant phenotype
33
.
This phenomenon represents a robustness of
bacteria against drugs on the basis of genetic
diversity and modular structure of a PAI
that makes horizontal gene transfer efficient.
At the same time, drug resistance could be
Box 1 | Robustness and drug resistance
Case study: gefitinib. Gefitinib (Iressa; AstraZeneca) is a dose-dependent inhibitor of epidermal-
growth-factor receptor (EGFR) autophosphorylation that competitively binds at the ATP site on
EGFR
66
. The rationale behind this drug is that because EGFR is expressed, or overexpressed,
in various solid tumours, such as non-small-cell lung cancer (NSCLC), breast, colorectal, gastric,
ovarian, prostate and bladder cancer, and because it correlates with poor prognosis
67,68
,
inhibition of EGFR activity should provide therapeutic benefits.
However, clinical benefits were demonstrated for only a specific subpopulation of patients with
advanced NSCLC with overexpression of amphiregulin
69
and a specific mutation (L858R and
exon 19 deletion) in EGFR
70,71
. Incidents of interstitial lung disease as a side effect of gefitinib have
been reported, and although gefitinib has been considered to be selective for EGFR, it can also
inhibit downstream-signalling proteins or ATP-dependent kinases other than EGFR, which could
cause toxicity
72
. A study on gefitinib-resistant tumour cells indicates that gefitinib’s
antiproliferative effect is due to the activation of glycogen synthase kinase-3β (GSK3β) that
degrades cyclin D1, and activation of platelet-derived-growth factor receptor (PDGFR) signalling
overrides gefitinib’s effect by activating mitogen-activated protein kinase (MAPK) and by inhibiting
GSK3β activity
73
. Furthermore, gefitinib-resistant tumour cells have a specific mutation in the
tyrosine kinase domain of EGFR that replaces threonine with methionine at position 790 (T790M),
which prevents the access of gefitinib to its binding site
74
. This case highlights the diversity of
mutations and pathways already prevalent in patient populations that give rise to the robustness
of tumour cells against gefitinib intervention, which prevents the drug from being effective.
Case study: trastuzumab. Trastuzumab (Herceptin; Genentech) is a humanized monoclonal
antibody that binds to the extracellular domain of ERBB2 (also called HER2/neu), thereby
downregulating ERBB2 signalling
75
. Its development was inspired by the fact that about 30% of
breast and ovarian cancer cases have higher levels of ERBB2 expression and gene amplification,
and overexpression of this gene is associated with poor prognosis
76
. Although efficacy of
trastuzumab monotherapy has been reported for ERBB2-overexpressing breast cancer
patients
77
, only a limited response rate at around 35% has been attained
78
. Furthermore,
resistance arises possibly owing to the activation of insulin-like growth factor 1 receptor
signalling
79
and the loss of phosphatase and tensin homologue
80
, which enables tumour-cell
proliferation regardless of ERB family signalling. This is another example in which robustness
of tumour cells against specific intervention through genetic diversity can compromise the
effectiveness of a drug.
PERSPECTIVES
204
|
MARCH 2007
|
VOLUME 6 www.nature.com/reviews/drugdisc
© 2007 Nature Publishing Group
mitigated by interrupting horizontal gene
transfer that depends on a modular PAI
region. This example illustrates how the
concept of robustness helps to interpret
efficacy and resistance to drugs.
A further illustrative case is that of
imatinib (Gleevec; Novartis), which is
considered as a successful example of a
drug that targets the protein product of
a disease-causative gene
34
. Using mice, a
characteristic genetic anomaly — a trans-
location between the long arms of chromo-
somes 9 and 22 that produces the BCR–ABL
fused gene observed in most patients with
chronic myeloid leukaemia (CML) — has
been identified as a cause of the disease
35–39
.
Imatinib binds to a specific conformation
of the mutant BCR–ABL kinase to hold it
in the inactive state. The high efficacy of
imatinib for a subset of the patients with
CML can be attributed to the fact that
the rare incidence of BCR–ABL indicates
that the fusion protein is a single cause of
CML at the initial stage and only exists in
target cells. So, similar to the anti-infectives
example above, imatinib hits the point
of fragility for CML. However, imatinib
efficacy is compromised by the emergence
of various mutations of BCR–ABL, as well
as drug resistance that is not dependent on
BCR–ABL mutations
40
. This indicates that
the CML disease state regains its robustness
against the drug through the diversity of
mutations and pathways that maintain the
activity of tumour cells.
The lesson that can be learned from these
examples is that a drug can be effective if it
hits the point of fragility, but can become
less effective if robustness of pathogens or
tumour cells against interventions emerges
through diversity, which is a type of fail-safe
mechanism
(FIG. 1b,c). That biological sys-
tems have evolved to form complex systems
that are robust against a broad range of
external and internal perturbations implies
that they can also be robust against thera-
peutic drug interventions. Examples of such
cases are analysed in
BOX 1.
The side effects of a drug can be most
serious when the drug hits a point of fragil-
ity of the patient’s system
(BOX 2). Although
obvious side effects can be identified and
addressed at the early stages of drug develop-
ment, difficulties occur when side effects
develop only in a specific subpopulation that
has particular genetic polymorphisms and/or
lifestyle that make the patients susceptible to
such interventions. It is therefore important
that extensive data are collected to better
prevent unwanted effects. This requires the
sampling of numerous cells in different
human tissues for the identification of a
functional role of potential unintended drug
targets, and the screening of unusual geno-
types to identify possible points of fragility
in rare subpopulations.
Multicomponent therapies
Each compound of a multicomponent
therapy simultaneously targets and inter-
acts with a different molecule so that the
combined effect can change the cellular
state. This is useful for when such change
cannot be made, or would be less effective,
by using individual drugs
7
. Multicomponent
therapies are increasingly gaining atten-
tion, with several marketing successes
reported — salmeterol/fluticasone (Advair;
GlaxoSmithKline)
41
, nicotinic acid/lovastatin
(Advicor; Kos Pharmaceuticals)
42,43
and
AZT–3TC (Combivir; GlaxoSmithKline)
44
.
A combined effect of atorvastatin and
fenofibrate in lowering cholesterol levels
has also been reported
6
, and a combination
high-throughput screening strategy has
been proposed and implemented in stages
in the drug discovery pipeline
45
. Moreover,
the application of a system-response profile
that captures differences in gene transcripts,
proteins and metabolites between drug-
perturbed and unperturbed cells has been
proposed
6
.
Although the exact mechanisms that
underlie the combinatorial effects of success-
ful multicomponent therapies have yet to
be elucidated, three explanations have been
put forward (examples are highlighted in
BOX 3). First, the effect of each component
converging at a certain part of the pathway
could trigger a more-than-additive effect.
Second, if different components of the
targeted pathway compensate for the inhibi-
tory effects of one drug, the administration
of a combination of drugs could overcome
this neutralization effect, thereby effectively
eliminating robustness caused by a fail-safe
mechanism through diversity. Third, each
component could target a different molecule
or specific mutation site, in such a way
that the result is non-cross-resistance, as
highlighted by the AZT–3TC combination
therapy for HIV described in
BOX 3. Recent
studies related to the effects of multiple weak
perturbations on biological networks might
provide theoretical accounts for the effects of
multicomponent drugs
46,47
.
Properly designed multicomponent
therapies are the first step in controlling
robustness to achieve clinical efficacy, and can
be widely useful. However, the selectivity of
a drug combination depends on the relative
response of target cells over non-target cells.
This is particularly problematic for cancer
chemotherapy. Unless the molecules targeted
are uniquely expressed in the target cells, and
not in off-target cells, such an approach might
not attain a sufficient level of selectivity. Use
of synthetic lethality for multicomponent
anticancer drug discovery might improve
selectivity
8
, but it has yet to be shown whether
practical targets can be found.
An extreme extension of the multi-
component therapy approach involves
using large numbers of components that
target a broad range of molecules, instead
of targeting just one or two ‘major’ genes
or proteins. Traditional Chinese medicine
emphasizes minor interventions over a
broad range of factors
48,49
, although most
components in traditional Chinese medi-
cine are unidentified. It is interesting to
note that the US National Cancer Institute
has screened 35,000 samples of roots and
fruits from 12,000 plant species only to
find 3 new drugs
48
, which implies that the
central issue is patterns of intervention
rather than a single component. Traditional
Box 2 | Fragility and drug side effects
Case study: rofecoxib. Drug side effects can be caused by unwanted interactions with molecules
that expose the fragility of cellular or organ-level functions to specific interventions in both target
cells and off-target cells. For example, the coxibs, such as rofecoxib (Vioxx; Merck) and celecoxib
(Celebrex; Pfizer), were developed to selectively inhibit cyclooxygenase 2 (COX2) for mitigating
inflammation
81
, although rofecoxib has higher selectivity for COX2 over COX1 than celecoxib.
Selectivity for COX2 over COX1 has been considered to be important because COX2 is
selectively expressed in tissues with inflammation. Moreover, the adverse gastrointestinal effects
of some anti-inflammatory drugs have been attributed to the inhibition of COX1, which is
constitutively expressed in gastrointestinal tissues
82
, as well as in cultured endothelial and
vascular smooth-muscle cells
83
. Therefore, the more selective inhibition of COX2 by rofecoxib
compared with other anti-inflammatory drugs was considered to be a desirable property of the
compound
84
. However, rofecoxib was shown to be associated with an increased risk of adverse
cardiovascular events at high doses
85
, which might be due to a reduced production of
antithrombotic products that have cardio-protective functions
86
. This highlights the fact that
selectivity for a molecule in the target cells does not eliminate the risk of side effects, as the target
molecule might have an important role in off-target cells
.
PERSPECTIVES
NATURE REVIEWS
|
DRUG DISCOVERY VOLUME 6
|
MARCH 2007
|
205
© 2007 Nature Publishing Group
Chinese medicine can be an inspiration
to create a drug discovery strategy that
uses large numbers of components to
systematically control a complex network
system. Such therapies could be considered
as ‘long-tail’ drugs because most targets
are on the long tail of a statistical distribu-
tion. Long-tail distribution in this context
does not need to be scale free or follow
an exact power law; it is only necessary
that there are very long tails of statistical
distribution that might be an overlay of
scale-rich distributions. Recent discoveries
related to the scale-richness of molecular
interaction networks, as seen in metabolic
networks
50
, might imply that perturbations
of molecules in different parts of long-tail
distributions could have qualitatively
different impacts on the system.
There are practical implications of a
long-tail distribution. For example, cancer
subtype classification indicates that the
expression profile of large numbers of genes
has to be considered for accurate prediction,
rather than focusing on a small number of
major genes
51
, and a theoretical study using
microarray data for cancer classification
concluded that expression of genes at
the middle range of expression levels,
rather than top-ranking expression levels,
has more importance in improving
classification accuracy
52
. Although highly
speculative, it might be interesting to
investigate whether the systematic control
of complex molecular networks can be
achieved using perturbations of long-tail
components.
Multistage targeting of cellular dynamics
A more sophisticated approach to control-
ling robustness is multistage therapy. The
general concept is to use a sequence of drugs
administered at specific doses and time
intervals so that the dynamic state of target
cells can be selectively perturbed into the
system state that is desired for therapeutic
purposes. This requires a more sophisticated
understanding of the dynamic cell states and
methods to control them.
An effective multidrug therapy could
be composed of one or more series of pre-
emptive interventions and target interven-
tions. Pre-emptive interventions target
molecules that are involved in controlling
cellular states, such as cell-cycle and regula-
tory feedback loops, so that the cell can be
induced into specific states for which the
target intervention will be most effective
or so that compensatory mechanisms that
could nullify the target interventions can
be disabled. Target interventions ultimately
induce desired changes in cellular states,
such as apoptosis and dormancy for tumour
cells, and various functional remedies for
other diseases. The timing of interventions
is crucial, because the target interventions
can often trigger compensatory robustness
mechanisms that offset the effects of the
interventions, and such effects cannot be
easily countered once invoked, although
they can be easily pre-empted. By disabling
the mechanism that provides robustness
against the target intervention first by
pre-emptive intervention, the robustness
of the cell against the target intervention is
significantly reduced, which increases the
chance of success. In such cases, it is impor-
tant that the correct order and individual
dose of the drugs to be administered, as well
as the interval between the administration
of each drug, are examined. The general
concept of the multistage drug is described
in
Supplementary information S1 (figure).
The importance of the order of adminis-
tration of drugs in a multidrug therapy has
been documented in recent years, particu-
larly in relation to the cell cycle of tumour
cells
53
. For example, in a study using colo-
rectal cancer cell lines (DLD1 and SW480),
effects on the order of administration for
paclitaxel (Taxol; Bristol–Myers Squibb)
and oxaliplatin (Eloxatin, Sanofi–Aventis)
have been investigated; a taxol–oxaliplatin
sequence had greater growth-inhibitory
effects than an oxaliplatin–taxol sequence and
than single-reagent treatment
54
. An experi-
ment using the human colon cancer cell line
HCT-116 demonstrated that a sequence of
SN-38 (the active metabolite of irinotecan)
Box 3 | Multicomponent therapies
Case study: combinations with HSP90 inhibitors. The use of imatinib, an inhibitor of the BCR–ABL
kinase, and heat-shock protein 90 (HSP90) inhibitors represents an interesting potential
combination anticancer strategy
87
. 17-allylamino-demethoxygeldanamycin (17-AAG) is an
inhibitor of HSP90, which is a molecular chaperone that stabilizes proteins by correcting
misfolding
88
. 17-AAG reduces BCR–ABL levels and induces apoptosis in in vitro experiments
89
,
and BCR–ABL-point-mutation cells that are imatinib-resistant are still sensitive to 17-AAG
90
.
17-AAG was also reported to reduce ERBB2 levels and to inhibit the proliferation of a
trastuzumab-resistant tumour cell line in an in vitro experiment
91
. Furthermore, a comparison of
responses against HSP90 inhibition using geldanamycins indicated higher apoptosis rates in
tumour cells with epidermal-growth-factor receptor (EGFR) mutations (NCI-H3255, NCI-H1650,
NCI-H1975) than tumour cells without EGFR mutations (A549)
92
. An interesting point in this report
is that even a gefitinib-resistant cell line (NCI-H1975) that harbours the secondary mutation
(T790M) demonstrated a higher apoptosis rate than non-EGFR mutation cells, which implies that
HSP90 inhibition is effective even for gefitinib-resistant tumours.
Overall, these findings imply that the viability of tumour cells is partly dependent on HSP90
function, and at least certain types of BCR–ABL, ERBB2 and EGFR-mutant proteins require
HSP90 to be functional. So, HSP90 is an example of a protein that provides decoupling to enhance
robustness of component functions against mutations and environmental perturbations by means
of fixing protein-structure folding.
Case study: AZT–3TC. A combination of drugs for non-cross-resistant targets can be particularly
effective when one of the components is designed to target the point of fragility that is induced
by the other components in the combination. A successful example of this approach is seen in HIV
therapy with AZT–3TC combination therapy. HIV-1 can acquire resistance to 3′-azidothymidine
(AZT, zidovudine) through stepwise accumulation of 4 out of 5 mutations in reverse transcriptase
at codons 41, 67, 70, 215 or 219
(REF. 93). For HIV-1 to be resistant to L-2′-deoxy-3′-thiacytidine
(3TC, lamivudine), it must develop an allele substitution of valine for methionine at codon 184 in
reverse transcriptase
94
. It was found that AZT-resistant virus is 3TC-sensitive, and 3TC-resistant
virus is AZT-sensitive owing to interactions on these mutations on viral reverse transcriptase
44
.
This is an impressive example in which the trade-off of being robust against AZT is fragility against
3TC, and vice versa.
Case study: multikinase inhibition. The combination of gefitinib and trastuzumab was reported to
inhibit tumour-cell proliferation of non-small-cell lung cancer (NSCLC) cells more effectively
than either drug alone
95
. Multikinase inhibitors can attain an effect in a single drug that is
analogous to that of multicomponent therapies that are based on drugs that target a single kinase.
For example, sunitinib malate (Sutent; Pfizer) is a compound that interacts with several receptor
tyrosine kinases (RTKs) such as platelet-derived growth-factor receptor-α (PDGFRα), PDGFRβ,
vascular endothelial growth-factor receptor 2 (VEGFR2), KIT and fms-related tyrosine kinase 3
(FLT3)
96
. The efficacy of such drugs could be related to the inhibition of multiple pathways, which
lead to effects through interactions with key molecules in each pathway — that is, they attain
therapeutic effects by mitigating the robustness of cellular functions against inhibitory
perturbations that is gained through a diversity of pathways.
PERSPECTIVES
206
|
MARCH 2007
|
VOLUME 6 www.nature.com/reviews/drugdisc
© 2007 Nature Publishing Group
followed by flavopiridol induced apoptosis at a
high rate (about 44%), but a reverse sequence
attained only a modest level of apoptosis
(about 15%) and SN-38 as a single reagent
induced only 1% of apoptosis
55,56
. The use of
SN-38 induces p21 expression and a con-
comitant G2 arrest, which makes the tumour
cells resistant to SN-38 (and therefore
irinotecan) because SN-38 is most effec-
tive in the S phase owing to its mechanism
of action as a topoisomerase-I inhibitor.
Flavopiridol induces G1 and G2 arrest, and
inhibits CDK1, CDK2, CDK4, CDK6
and a series of anti-apoptotic molecules
such as BCL2, p21 and phospho-survivin
56
.
Therefore, the use of SN-38 followed by
flavopiridol is synergistic, as cells arrested in
G2 by SN-38 might be induced to undergo
apoptosis owing to the inhibitory effect of
flavopiridol on anti-apoptotic molecules, or
released from arrest through p21 inhibition
by flavopiridol to progress into the G1–S
phase, when SN-38 is most effective. However,
the use of flavopiridol followed by SN-38
offsets some of the effects of flavopiridol,
such as p21 inhibition, by antagonistic
effects of SN-38, which upregulate p21.
Also, the cell cycle is arrested at G1 and G2
by flavopiridol, and lower numbers of cells
enter S phase when SN-38 is most effective.
The combination of irinotecan followed by
flavopiridol is currently in clinical trials
57
.
Sequential use of cancer drugs can be
extended to involve scheduling to reflect
circadian rhythms, as these might affect cell
cycle and metabolism — cancer chrono-
therapy
58,59
(BOX 4). Cancer chronotherapy is
based on the idea that tumour and normal
cells have different circadian rhythms that
affect drug response and side effects. For
example, if a taxol–oxaliplatin sequence
is effective and the use of oxaliplatin at a
certain time of the day is most effective and
minimizes the side effects, this time of day
shall determine the overall optimal sched-
ule for sequential multiple drug use that
maximizes the difference of drug response
between tumour and normal cells. The
observation that circadian rhythms might
be deregulated in tumour cells, but not in
normal cells, provides one such clue that
might enhance the selectivity of therapeutic
effects
60
, and research into the relationship
between circadian rhythms, the cell cycle
and tumours is making rapid progress
61
.
The next important step for this type of
therapeutic strategy is to firmly establish
the scientific basis of temporal dynamics of
cells in disease and normal conditions, and
of differential temporal responses of cells to
various drugs and their combinations.
Systems-oriented drug design
The question now is how to design systems-
oriented drugs that take into account the
aspects of system robustness discussed
above. How can targets, timing and dosage
be optimized to attain the desired effects?
This requires an in-depth understanding
of cellular- and organism-level dynamics,
combined with advanced high-throughput
screening and computational analysis tools.
The approaches discussed here also call for
the use of multiple drugs in sophisticated
combinations. This adds a considerable
level of combinatorial complexity in the
drug discovery and therapy design process,
because multiple components (often in large
numbers) have to be identified and validated,
and their proper combination, dosage and
timing of use have to be identified, which
might raise the cost of drug development.
Nevertheless, the corresponding growth in
potential strategies to treat presently intrac-
table diseases could represent an excellent
opportunity. So, establishment of effective
and efficient methods to systematically iden-
tify appropriate strategies is the key. A range
of possible options to tackle this challenge
for different degrees of complexity and use
of dynamics discussed here are summarized
in
FIG. 2.
Theoretically, the main challenge, in both
a biological and a mathematical sense, is to
find out how to control complex network
systems. Although the dynamics of simple
feedback systems have been well understood
from classical and modern control theory,
the control of highly distributed nonlinear
network systems has yet to be established.
Mathematics that describe biological
systems, particularly from a control aspect,
have yet to be developed. Questions such
as how can we control a complex network
using components on a long-tail distribu-
tion have yet to be answered. Advanced
communication theories can be a source of
inspiration. For example, spread-spectrum
communication
62
is used for mobile phones
to achieve highly selective and noiseless
communication between users. Perhaps
the strategy used to attain peer-to-peer
communication without interference could
be applied for highly sophisticated drug
design for particular cases, such as a spread-
spectrum drug that is at the extreme end
of complex drug systems (
Supplementary
information S2
(box)). Many intriguing the-
ories and methods exist, although most are
for engineering and physics, and cannot be
simply translated into the biological domain.
Reformulation and invention of new theories
can be a fruitful challenge for researchers in
both mathematics and biology. These studies
might also shed light on the science behind
traditional Chinese medicine, and such a
series of theories could lead to an integration
of Western and Oriental medicine on a
scientific basis.
From a technological and experimental
viewpoint, effective screening methods and
analysis software tools must be developed,
as well as a more in-depth understanding of
cellular systems. As mentioned above, a sys-
tematic screening study for multicomponent
drugs has been reported
45
, and a number of
successful combination drugs are already
on the market. These developments have
Box 4 | Cancer chronotherapy
Case study: 5-fluorouracil, leucovorin and oxaliplatin. Cancer chronotherapy is an example of
when the timing of drug use is synchronized with circadian rhythms, which might be interacting
with the cell cycle and systemic metabolic conditions
58,59,97,98
. Although cancer chronotherapy
does not actively induce certain cellular states, it exploits natural circadian rhythms to attain a
context-dependent efficacy of the drug use. Furthermore, circadian rhythms have been argued to
be deregulated in advanced tumours
99
.
Clinical studies are scarce, but have shown promising results. In a study for patients with colorectal
cancer, 5-fluorouracil (5-FU), leucovorin (LV) and oxaliplatin were administered in a phased manner;
oxaliplatin infusion peaked at 16:00 hours, and 5-FU and LV peaked at 04:00 hours. Patients with
chronomodulatory treatment showed significantly higher objective responsiveness and a threefold
reduction in the incidence of World Health Organization (WHO) Grade 4 toxicity compared with
patients that received the drugs in a standard manner
100
.
A significant difference in the 5-year survival rates in patients with acute lymphoblastic
leukaemia has also been reported between those who had drugs (6-mercaptopurine and
methotrexate) administered during the evening (80%) and those who had drugs administered in the
morning (40%)
101
. However, how robustness and fragility are affected by different phases of circadian
rhythms has yet to be fully elucidated, although correlation with cell-cycle and metabolic changes
are reasonable causes for such effects. While circadian rhythms are not properly maintained in
tumour cells, cell-cycle and metabolic states might also be affected. If this is the case, tolerable doses
of normal cells against cell-cycle-dependent drugs can be increased by optimizing the time of use,
so that overall therapeutic efficacy can be improved.
PERSPECTIVES
NATURE REVIEWS
|
DRUG DISCOVERY VOLUME 6
|
MARCH 2007
|
207
© 2007 Nature Publishing Group
not explicitly taken robustness issues into
account, but they explore a set of factors that
can counteract the robustness of the cellular
systems against drug interventions. The
issues for the next step are to systematically
identify targets in the context of robustness
and fragility, rather than just combinatorial
effects. Recently, my group has developed
a new experimental method called gene
tug-of-war (gTOW) to comprehensively and
quantitatively measure robustness of cellular
functions against quantitative perturbations
of gene dosage
63
. Such an approach could be
scaled up and improved for genome-wide
robustness-based screening and provides a
new approach for drug screening and target
identification.
Techniques that can efficiently identify
new drugs and their targets are essential for
providing options for multicomponent thera-
pies, multistage therapies and, most notably,
the long-tail therapies. Cell-based screening
of both natural compounds and chemical
genomics might provide us with a broad base
of effective drugs
3,4
, and microfluidics tech-
nologies could provide us with opportunities
for high-throughput cell-based screening
64
.
A set of computational and mathematical
tools also has to be developed to support the
robustness-based approach. The robustness-
based approach has a specific goal: to find a
set of targets that has a specific robustness
property for which combinatorial perturba-
tion generates both efficacy and selectivity.
So, in the future, the process can be made
more sophisticated by applying a range of
mathematical and experimental tools geared
to this specific goal. Highly sophisticated
bifurcation and other systems-dynamics
analysis packages that can handle compara-
tive high-dimension analyses for large-scale
models are essential. Such software enables
the application of computational models to
identify differences in dynamical behaviours
between target and off-target cells
65
.
Although the accuracy of computational
simulation today is not at a satisfactory
level, owing to problems of quantitative
data acquisition and modelling techniques,
the situation will improve rapidly. It should
be remembered that computational fluid
dynamics, one of the most successful
computational approaches, took decades
of efforts to become practically applicable.
In addition, high-precision comprehensive
mapping of molecular interactions need to
be created and properly maintained because
such maps uncover robustness and fragility
of networks, as seen in examples of epider-
mal-growth-factor receptor
23
and Toll-like
receptor
25
networks.
Furthermore, to best identify character-
istic dynamics that are unique to target cells
and to avoid side effects, differential systems
biology needs to be developed for which
differences in the interactions and dynamics
of different cell types together with different
genetic backgrounds have to be compared in
an efficient manner.
Conclusion
This article has put forward the proposal
that drug and therapy design should be
based on an in-depth understanding of
robustness and its trade-offs in the organ-
ism, from the cellular level to the systemic
level. With the explosion of knowledge in
genomics, proteomics and systems biology,
the immediate concern should be to attain
an understanding of robustness and fragility
against drug interventions at the cellular
level. This eventually translates to robust-
ness analysis at the systemic level. It is clear
that many successful drugs in the past have
affected a point of fragility of the target sys-
tem so that a single intervention modulating
specific molecules or biological processes
can trigger dramatic effects. However,
cancer, diabetes, autoimmune disorders and
other diseases that are proving challenging
to treat with this simple strategy stem from
the robustness of the human body, from the
cellular level to the systemic level. Given that
robustness is a fundamental property of bio-
logical systems, systems-based approaches to
future drug design should consider robust-
ness as the central framework. A new way of
controlling complex and robust biological
systems through higher-complexity drugs
will need to be established, as complexity has
to be controlled by complexity, but address-
ing these challenges and making robustness-
based approaches to drug design a reality
could cause a fundamental transformation
in the drug industry and in medical practice.
Use of dynamics
Numbers of components and/or targets
Complexity of
perturbations
Simplicity of
perturbations
Spread-spectrum drugs
Multistage drugs
Long-tail drugs
Traditional
Chinese
medicine
Single-
molecular-
target
drugs
Conventional
drugs
Multicomponent
drugs
Chronotherapy
Figure 2 | A range of options for drug design. Different types of therapeutic strategies can be
classified by the numbers of component drugs, the number of their targets and the use of dynamics.
Current conventional drugs often target a small number of molecules, whereas ‘long-tail’ drugs and
Chinese herbal medicines use a large number of components. Drugs that target single molecules
and spread-spectrum drugs (still a speculative concept at present) that would extensively use dynamics
to attain selectivity are two extremes in the diagram. Both strategies could be highly effective in
special cases, but might not be generic enough to be broadly applicable. Multistage drugs, long-tail
drugs and multicomponent drugs, perhaps in combination with chronotherapy, could have broad
applicability and are expected to represent mainstream therapeutic strategies in the future.
PERSPECTIVES
208
|
MARCH 2007
|
VOLUME 6 www.nature.com/reviews/drugdisc
© 2007 Nature Publishing Group
Hiroaki Kitano is at Sony Computer Science
Laboratories Inc., 3-14-13 Higashi-Gotanda,
Shinagawa, Tokyo 141-0022, Japan; The Systems
Biology Institute, Suite 6A M31 6-31-15 Jingumae,
Shuibuya, Tokyo 150-0001, Japan; and the Department
of Cancer Systems Biology, The Cancer Institute,
3-10-6 Ariake, Koutou-ku, Tokyo 135-8550, Japan.
e-mail: kitano@sbi.jp
doi:10.1038/nrd2195
Published online 23 February 2007
1. Kubinyi, H. Drug research: myths, hype and reality.
Nature Rev. Drug Discov. 2, 665–668 (2003).
2. Horrobin, D. F. Modern biomedical research:
an internally self-consistent universe with little contact
with medical reality? Nature Rev. Drug Discov. 2,
151–154 (2003).
3. Butcher, E. C. Can cell systems biology rescue drug
discovery? Nature Rev. Drug Discov. 4, 461–467
(2005).
4. Butcher, E. C., Berg, E. L. & Kunkel, E. J. Systems
biology in drug discovery. Nature Biotech. 22,
1253–1259 (2004).
5. Hood, L. & Perlmutter, R. M. The impact of systems
approaches on biological problems in drug discovery.
Nature Biotech. 22, 1215–1217 (2004).
6. van der Greef, J. & McBurney, R. N. Rescuing drug
discovery: in vivo systems pathology and systems
pharmacology. Nature Rev. Drug Discov. 4, 961–967
(2005).
7. Keith, C. T., Borisy, A. A. & Stockwell, B. R.
Multicomponent therapeutics for networked systems.
Nature Rev. Drug Discov. 4, 71–78 (2005).
8. Kaelin, W. G. Jr. The concept of synthetic lethality in
the context of anticancer therapy. Nature Rev. Cancer
5, 689–698 (2005).
9. Nicholson, J. K., Holmes, E., Lindon, J. C. & Wilson, I. D.
The challenges of modeling mammalian biocomplexity.
Nature Biotech. 22, 1268–1274 (2004).
10. Fernie, A. R., Trethewey, R. N., Krotzky, A. J. &
Willmitzer, L. Metabolite profiling: from diagnostics
to systems biology. Nature Rev. Mol. Cell Biol. 5,
763–769 (2004).
11. Nicholson, J. K. & Wilson, I. D. Understanding ‘global’
systems biology: metabonomics and the continuum of
metabolism. Nature Rev. Drug Discov. 2, 668–676
(2003).
12. Kell, D. B. et al. Metabolic footprinting and systems
biology: the medium is the message. Nature Rev.
Microbiol. 3, 557–565 (2005).
13. Kitano, H. Biological robustness. Nature Rev. Genet.
5
, 826–837 (2004).
14. Stelling, J., Sauer, U., Szallasi, Z., Doyle, F. J. 3rd &
Doyle, J. Robustness of cellular functions. Cell 118 ,
675–685 (2004).
15. Csete, M. E. & Doyle, J. C. Reverse engineering of
biological complexity. Science 295, 1664–1669
(2002).
16. Csete, M. & Doyle, J. Bow ties, metabolism and
disease. Trends Biotechnol. 22, 446–450 (2004).
17. Carlson, J. M. & Doyle, J. Highly optimized tolerance:
robustness and design in complex systems. Phys. Rev.
Lett. 84, 2529–2532 (2000).
18. Carlson, J. M. & Doyle, J. Complexity and robustness.
Proc. Natl Acad. Sci. USA 99 (Suppl. 1), 2538–2545
(2002).
19. Ferrell, J. E. Jr. Self-perpetuating states in signal
transduction: positive feedback, double-negative
feedback and bistability. Curr. Opin. Cell. Biol. 14,
140–148 (2002).
20. Bhalla, U. S. & Iyengar, R. Robustness of the bistable
behavior of a biological signaling feedback loop.
Chaos 11, 221–226 (2001).
21. Rutherford, S. L. Between genotype and phenotype:
protein chaperones and evolvability. Nature Rev.
Genet. 4, 263–274 (2003).
22. Ma, H. W., Zhao, X. M., Yuan, Y. J. & Zeng, A. P.
Decomposition of metabolic network into functional
modules based on the global connectivity structure of
reaction graph. Bioinformatics 20, 1870–1876 (2004).
23. Oda, K., Matsuoka, Y., Funahashi, A. & Kitano, H.
A comprehensive pathway map of epidermal growth
factor receptor signaling. Mol. Syst. Biol. 1, 25 May
2005 (doi:10.1038/msb4100014).
24. Kitano, H. & Oda, K. Robustness trade-offs and
host-microbial symbiosis in the immune system.
Mol. Syst. Biol. 2, 17 Jan 2006 (doi:10.1038/
msb4100039).
25. Oda, K. & Kitano, H. A comprehensive pathway map
of toll-like receptor signaling network. Mol. Syst. Biol.
2, 18 April 2006 (doi:10.1038/msb4100057).
26. Doyle, J. C. et al. The ‘robust yet fragile’ nature of the
internet. Proc. Natl Acad. Sci. USA 102, 14497–14502
(2005).
27.
Lamport, L., Shostak, R. & Pease, M. The Byzantine
Generals problem. ACM Trans. Program. Languages
Syst. 27, 382–401 (1982).
28. Kitano, H. et al. Metabolic syndrome and robustness
tradeoffs. Diabetes 53 (Suppl. 3), 6–15 (2004).
29. Kitano, H. Cancer as a robust system: implications for
anticancer therapy. Nature Rev. Cancer 4, 227–235
(2004).
30. Kitano, H. Cancer robustness: tumour tactics. Nature
426, 125 (2003).
31. Waldor, M. K. & Mekalanos, J. J. Lysogenic conversion
by a filamentous phage encoding cholera toxin.
Science 272, 1910–1914 (1996).
32. Hung, D. T., Shakhnovich, E. A., Pierson, E. &
Mekalanos, J. J. Small-molecule inhibitor of Vibrio
cholerae virulence and intestinal colonization. Science
310, 670–674 (2005).
33. Normark, B. H. & Normark, S. Evolution and spread of
antibiotic resistance. J. Intern. Med. 252, 91–106
(2002).
34. Druker, B. J. et al. Efficacy and safety of a specific
inhibitor of the BCR–ABL tyrosine kinase in chronic
myeloid leukemia. N. Engl. J. Med. 344, 1031–1037
(2001).
35. Rowley, J. D. A new consistent chromosomal
abnormality in chronic myelogenous leukaemia
identified by quinacrine fluorescence and Giemsa
staining. Nature 243, 290–293 (1973).
36. Daley, G. Q., Van Etten, R. A. & Baltimore, D.
Induction of chronic myelogenous leukemia in mice by
the P210bcr/abl gene of the Philadelphia
chromosome. Science 247, 824–830 (1990).
37. Kelliher, M. A., McLaughlin, J., Witte, O. N. &
Rosenberg, N. Induction of a chronic myelogenous
leukemia-like syndrome in mice with v-abl and BCR/
ABL. Proc. Natl Acad. Sci. USA 87, 6649–6653
(1990).
38. Heisterkamp, N. et al. Acute leukaemia in bcr/abl
transgenic mice. Nature 344
, 251–253 (1990).
39. Lugo, T. G., Pendergast, A. M., Muller, A. J. &
Witte, O. N. Tyrosine kinase activity and
transformation potency of bcr–abl oncogene products.
Science 247, 1079–1082 (1990).
40. Daub, H., Specht, K. & Ullrich, A. Strategies to overcome
resistance to targeted protein kinase inhibitors. Nature
Rev. Drug Discov. 3, 1001–1010 (2004).
41. Nelson, H. S. Advair: combination treatment with
fluticasone propionate/salmeterol in the treatment of
asthma. J. Allergy Clin. Immunol. 107, 398–416 (2001).
42. Gupta, E. K. & Ito, M. K. Lovastatin and extended-
release niacin combination product: the first drug
combination for the management of hyperlipidemia.
Heart Dis. 4, 124–137 (2002).
43. Bays, H. E. et al. Comparison of once-daily, niacin
extended-release/lovastatin with standard doses of
atorvastatin and simvastatin (the advicor versus
other cholesterol-modulating agents trial evaluation
[ADVOCATE]). Am. J. Cardiol. 91, 667–672 (2003).
44. Larder, B. A., Kemp, S. D. & Harrigan, P. R. Potential
mechanism for sustained antiretroviral efficacy of
AZT–3TC combination therapy. Science 269,
696–699 (1995).
45. Borisy, A. A. et al. Systematic discovery of
multicomponent therapeutics. Proc. Natl Acad. Sci.
USA 100, 7977–7982 (2003).
46. Csermely, P., Agoston, V. & Pongor, S. The efficiency of
multi-target drugs: the network approach might help
drug design. Trends Pharmacol. Sci. 26, 178–182
(2005).
47. Agoston, V., Csermely, P. & Pongor, S. Multiple
weak hits confuse complex systems: a transcriptional
regulatory network as an example. Phys. Rev. E Stat.
Nonlin. Soft Matter Phys. 71, 051909 (2005).
48. Yuan, R. & Lin, Y. Traditional Chinese medicine: an
approach to scientific proof and clinical validation.
Pharmacol. Ther. 86, 191–198 (2000).
49. Wang, M. et al. Metabolomics in the context of
systems biology: bridging traditional Chinese medicine
and molecular pharmacology. Phytother. Res. 19,
173–182 (2005).
50. Tanaka, R. Scale-rich metabolic networks. Phys. Rev.
Lett. 94, 168101 (2005).
51. Khan, J.
et al. Classification and diagnostic prediction
of cancers using gene expression profiling and artificial
neural networks. Nature Med. 7, 673–679 (2001).
52. Li, W. & Yang, Y. Zipf’s law in importance of genes for
cancer classification using microarray data. J. Theor.
Biol. 219, 539–551 (2002).
53. Shah, M. A. & Schwartz, G. K. Cell cycle-mediated
drug resistance: an emerging concept in cancer
therapy. Clin. Cancer Res. 7, 2168–2181 (2001).
54. Fujie, Y. et al. Oxaliplatin, a potent inhibitor of
survivin, enhances paclitaxel-induced apoptosis and
mitotic catastrophe in colon cancer cells. Jpn. J. Clin.
Oncol. 35, 453–463 (2005).
55. Motwani, M. et al. Augmentation of apoptosis and
tumor regression by flavopiridol in the presence of
CPT-11 in Hct116 colon cancer monolayers and
xenografts. Clin. Cancer Res. 7, 4209–4219 (2001).
56. Schwartz, G. K. & Shah, M. A. Targeting the cell cycle:
a new approach to cancer therapy. J. Clin. Oncol. 23,
9408–9421 (2005).
57. Shah, M. A. et al. A Phase I clinical trial of the
sequential combination of irinotecan followed by
flavopiridol. Clin. Cancer Res. 11, 3836–3845 (2005).
58. Levi, F. Cancer chronotherapy. J. Pharm. Pharmacol.
51, 891–898 (1999).
59. Levi, F. Circadian chronotherapy for human cancers.
Lancet Oncol. 2, 307–315 (2001).
60. Chen, S. T. et al. Deregulated expression of the PER1,
PER2 and PER3 genes in breast cancers.
Carcinogenesis 26, 1241–1246 (2005).
61. Gery, S. et al. The circadian gene PER1 plays an
important role in cell growth and DNA damage control
in human cancer cells. Mol. Cell. 22, 375–382 (2006).
62. Torrieri, D. Principles of Spread Spectrum
Communication Systems (Springer, New York, 2004).
63. Moriya, H., Shimizu-Yoshida, Y. & Kitano, H. In vivo
robustness analysis of cell division cycle genes in
Saccharomyces cerevisiae. PLoS Genet. 2, e111 (2006).
64. El-Ali, J., Sorger, P. K. & Jensen, K. F. Cells on chips.
Nature 442, 403–411 (2006).
65. Borisuk, M. T. & Tyson, J. J. Bifurcation analysis of a
model of mitotic control in frog eggs. J. Theor. Biol.
195, 69–85 (1998).
66. Wakeling, A. E. et al. ZD1839 (Iressa): an orally
active inhibitor of epidermal growth factor signaling
with potential for cancer therapy. Cancer Res. 62,
5749–5754 (2002).
67. Voldborg, B. R., Damstrup, L., Spang-Thomsen, M.
& Poulsen, H. S. Epidermal growth factor receptor
(EGFR) and EGFR mutations, function and possible
role in clinical trials. Ann. Oncol. 8, 1197–1206
(1997).
68. Salomon, D. S., Brandt, R., Ciardiello, F. &
Normanno, N. Epidermal growth factor-related
peptides and their receptors in human malignancies.
Crit. Rev. Oncol. Hematol. 19, 183–232 (1995).
69. Kakiuchi, S. et al. Prediction of sensitivity of advanced
non-small cell lung cancers to gefitinib (Iressa,
ZD1839). Hum. Mol. Genet. 13, 3029–3043 (2004).
70. Lynch, T. J. et al. Activating mutations in the epidermal
growth factor receptor underlying responsiveness of
non-small-cell lung cancer to gefitinib. N. Engl. J. Med.
350, 2129–2139 (2004).
71. Paez, J. G. et al. EGFR mutations in lung cancer:
correlation with clinical response to gefitinib therapy.
Science 304, 1497–1500 (2004).
72. Cohen, M. H. et al. United States Food and Drug
Administration drug approval summary: gefitinib
(ZD1839; Iressa) tablets. Clin. Cancer Res. 10,
1212–1218 (2004).
73. Kassouf, W. et al. Uncoupling between epidermal
growth factor receptor and downstream signals
defines resistance to the antiproliferative effect of
gefitinib in bladder cancer cells. Cancer Res. 65,
10524–10535 (2005).
74. Kobayashi, S. et al. EGFR mutation and resistance of
non-small-cell lung cancer to gefitinib. N. Engl. J. Med.
352, 786–792 (2005).
75. Carter, P. et al. Humanization of an anti-p185HER2
antibody for human cancer therapy. Proc. Natl Acad.
Sci. USA 89, 4285–4289 (1992).
76. Slamon, D. J. et al. Human breast cancer: correlation
of relapse and survival with amplification of the HER2/
neu oncogene. Science 235, 177–182 (1987).
77. Cobleigh, M. A. et al. Multinational study of the
efficacy and safety of humanized anti-HER2
monoclonal antibody in women who have HER2-
overexpressing metastatic breast cancer that has
progressed after chemotherapy for metastatic disease.
J. Clin. Oncol. 17, 2639–2648 (1999).
78. Vogel, C. L. et al. Efficacy and safety of trastuzumab
as a single agent in first-line treatment of HER2-
overexpressing metastatic breast cancer. J. Clin.
Oncol. 20, 719–726 (2002).
PERSPECTIVES
NATURE REVIEWS
|
DRUG DISCOVERY VOLUME 6
|
MARCH 2007
|
209
© 2007 Nature Publishing Group
79. Lu, Y., Zi, X., Zhao, Y., Mascarenhas, D. & Pollak, M.
Insulin-like growth factor-I receptor signaling and
resistance to trastuzumab (Herceptin). J. Natl Cancer
Inst. 93, 1852–1857 (2001).
80. Nagata, Y. et al. PTEN activation contributes to tumor
inhibition by trastuzumab, and loss of PTEN predicts
trastuzumab resistance in patients. Cancer Cell 6,
117–127 (2004).
81. FitzGerald, G. A. & Patrono, C. The coxibs, selective
inhibitors of cyclooxygenase-2. N. Engl. J. Med. 345,
433–442 (2001).
82. Kargman, S. et al. Characterization of prostaglandin
G/H synthase 1 and 2 in rat, dog, monkey, and human
gastrointestinal tracts. Gastroenterology 111,
445–454 (1996).
83. Evett, G. E., Xie, W., Chipman, J. G., Robertson, D. L.
& Simmons, D. L. Prostaglandin G/H synthase
isoenzyme 2 expression in fibroblasts: regulation by
dexamethasone, mitogens, and oncogenes. Arch.
Biochem. Biophys. 306, 169–177 (1993).
84. Flower, R. J. The development of COX2 inhibitors.
Nature Rev. Drug Discov. 2, 179–191 (2003).
85. Bresalier, R. S. et al. Cardiovascular events
associated with rofecoxib in a colorectal adenoma
chemoprevention trial. N. Engl. J. Med. 352,
1092–1102 (2005).
86. Zarraga, I. G. & Schwarz, E. R. Coxibs and heart
disease. J. Am. Coll. Cardiol. 49, 1–14 (2007).
87. Cortes, J. & Kantarjian, H. New targeted approaches
in chronic myeloid leukemia. J. Clin. Oncol. 23,
6316–6324 (2005).
88. Neckers, L. & Ivy, S. P. Heat shock protein 90.
Curr. Opin. Oncol. 15, 419–424 (2003).
89. Nimmanapalli, R., O’Bryan, E. & Bhalla, K.
Geldanamycin and its analogue 17-allylamino-17-
demethoxygeldanamycin lowers Bcr–Abl levels and
induces apoptosis and differentiation of Bcr–Abl-
positive human leukemic blasts. Cancer Res. 61,
1799–1804 (2001).
90. Gorre, M. E., Ellwood-Yen, K., Chiosis, G., Rosen, N. &
Sawyers, C. L. BCR–ABL point mutants isolated from
patients with imatinib mesylate-resistant chronic
myeloid leukemia remain sensitive to inhibitors of the
BCR–ABL chaperone heat shock protein 90. Blood
100, 3041–3044 (2002).
91. Zsebik, B. et al. Hsp90 inhibitor 17-AAG reduces
ErbB2 levels and inhibits proliferation of the
trastuzumab resistant breast tumor cell line JIMT-1.
Immunol. Lett. 104, 146–155 (2005).
92. Shimamura, T., Lowell, A. M., Engelman, J. A. &
Shapiro, G. I. Epidermal growth factor receptors
harboring kinase domain mutations associate with the
heat shock protein 90 chaperone and are destabilized
following exposure to geldanamycins. Cancer Res. 65,
6401–6408 (2005).
93. Larder, B. A. & Kemp, S. D. Multiple mutations in
HIV-1 reverse transcriptase confer high-level
resistance to zidovudine (AZT). Science 246,
1155–1158 (1989).
94. Tisdale, M., Kemp, S. D., Parry, N. R. & Larder, B. A.
Rapid in vitro selection of human immunodeficiency
virus type 1 resistant to 3′-thiacytidine inhibitors
due to a mutation in the YMDD region of reverse
transcriptase. Proc. Natl Acad. Sci. USA 90,
5653–5656 (1993).
95. Nakamura, H. et al. Cooperative cell-growth
in hibition by combination treatment with ZD1839
(Iressa) and trastuzumab (Herceptin) in non-small-cell
lung cancer. Cancer Lett. 230, 33–46 (2005).
96. Mendel, D. B. et al. In vivo antitumor activity of
SU11248, a novel tyrosine kinase inhibitor targeting
vascular endothelial growth factor and platelet-derived
growth factor receptors: determination of a
pharmacokinetic/pharmacodynamic relationship.
Clin. Cancer Res. 9, 327–337 (2003).
97. Hrushesky, W. et al. A recent illustration of some
essentials of circadian chronotherapy study design.
J. Clin. Oncol. 22, 2971–2972 (2004).
98. Eriguchi, M. et al. Chronotherapy for cancer.
Biomed. Pharmacother. 57 (Suppl. 1), 92–95
(2003).
99. Mormont, M. C. & Levi, F. Circadian-system alterations
during cancer processes: a review. Int. J. Cancer 70,
241–247 (1997).
100. Levi, F., Zidani, R. & Misset, J. L. Randomised
multicentre trial of chronotherapy with oxaliplatin,
fluorouracil, and folinic acid in metastatic colorectal
cancer. International Organization for Cancer
Chronotherapy. Lancet 350, 681–686 (1997).
101. Rivard, G. E., Infante-Rivard, C., Hoyoux, C. &
Champagne, J. Maintenance chemotherapy for
childhood acute lymphoblastic leukaemia: better in
the evening. Lancet 2, 1264–1266 (1985).
Acknowledgements
This research was supported in part by the Exploratory
Research for Advanced Technology (ERATO) and the Solution-
Oriented Research for Science and Technology (SORST)
programs (Japan Science and Technology Agency), a New
Energy and Industrial Technology Development Organization
Grant (NEDO) of the Japanese Ministry of Economy, Trade
and Industry (METI), and the Genome Network Project by the
Ministry of Education, Culture, Sports, Science, and
Technology.
Competing interests statement
The author declares no competing financial interests.
FURTHER INFORMATION
Hiroaki Kitano: Sony Computer Science Laboratories, Inc.:
http://www.csl.sony.co.jp
The Systems Biology Institute: http://www.sbi.jp
SUPPLEMENTARY INFORMATION
See online article: S1 (figure) | S2 (box)
Access to this links box is available online.
PERSPECTIVES
210
|
MARCH 2007
|
VOLUME 6 www.nature.com/reviews/drugdisc
© 2007 Nature Publishing Group
SUPPLEMENTARY INFORMATION In format provided by Kitano (MARCH 2007)
NATURE REVIEWS | DRUG D ISCOV ERY www.nature.com/reviews/drugdisc
Figure S1 | Using multistage drugs to achieve efficacy and selectivity. The
multistage approach involves successively administering a set of drugs to
differentiate between target and off-target cells. The assumption behind this
approach is that there might be differences between target and off-target cells
regarding their response to drugs, and that successive intervention, rather than a
single intervention, could better exploit such differences to attain a greater level of
selectivity. The states of target and off-target cells are initially assumed to be at the
steady state. a | Target and off-target cells respond differently to drug 1, but at the
level of the cellular state, there is no difference observed (A1 and A2). Cellular states
might stay the same in both cells, but, in this example, target cells might only be
stable in a sense that is termed mathematically ‘Lyapnov stable’ (the system’s state
can fluctuate but remain within the basin of attraction, which is a mathematically
defined area in which the state of the system tends to stay within its boundary),
whereas the off-target cell can be asymptotically stable (meaning that the state of
the system returns to its original state, a block dot in the centre in the example of
A2). b | Target and off-target cells (B1 and B2) respond identically to drug 2. In this
example, both cells maintain asymptotical stability against drug 2 when the
perturbation is applied at the certain dose and the state of the cell is at its original
state. c | Simultaneous use of two drugs offsets their effects and causes no main
effect on both target and off-target cells (C1 and C2). d | The successful multistage
drug will use two drugs in appropriate order, timing and dosage. In this case, drug 1
is used first at a dosage that can perturb the cell state to some extent, but not
enough to induce the cell into a different stable state (red trajectory in D1 and D2).
Drug 2 might perturb the cell further so that the cell moves into the new steady
state in target cells (blue trajectory in D1), but not in off-target cells (D2). In order
to attain such effects, differences of dynamics against drugs between target and off-
target cells have to be investigated. If there are differences in robustness against
drug interventions, which might, for example, be due to genetic mutations that are
unique in tumour cells, such differences can be explored. e | Such effects might be
dosage- and order-dependent, so that reverse-order administration might not
generate the same effect (E1 and E2).
© 2007 Nature Publishing Group
SUPPLEMENTARY INFORMATION In format provided by Kitano (MARCH 2007)
NATURE REVIEWS | DRUG D ISCOV ERY www.nature.com/reviews/drugdisc
Nature Reviews | Drug Discovery
Target cells
Off-target cells
A1
a
b
c
d
e
Drug 1
A2
Drug 1
B1
Drug 2
B2
Drug 2
C1
Drug 1 and 2
simultaneously
C2
Drug 1 and 2
simultaneously
D1
Drug 1 then Drug 2
D2
Drug 1 then Drug 2
E1
Drug 2 then Drug 1
E2
Drug 2 then Drug 1
© 2007 Nature Publishing Group
SUPPLEMENTARY INFORMATION In format provided by Kitano (MARCH 2007)
NATURE REVIEWS | DRUG D ISCOV ERY www.nature.com/reviews/drugdisc
Box S2 | The spread spectrum control problem and the long-tail control
problem
The spread spectrum control problem and the long-tail control problem are mathematical
problems motivated by the need to discover a highly selective system of drugs for cancer
chemotherapy and other diseases for which systemic interventions are warranted. The
problem that cancer researchers and pharmaceutical companies are facing is to find a set of
drugs, their dosage and timing of administration that can effectively intervene in tumour cells,
but not normal cells. Ineffective interventions would not eradicate the tumour, and unwanted
intervention in normal cells would cause serious side-effects.
Given that cellular systems can be viewed as network of genes, RNAs, proteins, and other
molecules, the problem can be abstracted and reformulated into a more general mathematical
problem. The solution to the problem would be a step towards more effective cancer
chemotherapy, but could also be used widely for other problems in both the medical and
engineering fields.
The spread spectrum method (the original idea was invented by Hedy Lamarr (a Hollywood
actress) and George Antheil (a composer) in the 1940s), is a communication technique that is
now used for mobile phone and satellite-based global positioning systems. Owing to its
unusual origin, there is no paper by the inventor, but there are numerous textbooks that
highlight the importance of the method
62
. It modulates signal in the way it spread over broad
spectrum each of which is at the noise level for unintended recipients, and the target recipients
with a proper decoding key can reassemble signals spread over broad frequencies, and restore
high fidelity signal (Figure). The beauty of spread spectrum approach is that it can attain high
level of selectivity and tolerance against noise; signals will not be heard by non-target and it
cannot even detect signal exits and noise will be dispersed under the noise level.
Figure | A simple conceptual scheme for spread spectrum communication
The challenge is whether such an approach can be used for controlling a network system,
rather than communication of signals.
The spread spectrum control problem can be stated as:
“Given the two or more networks that are not identical, find a set of sequential minor
perturbations on nodes in networks similarly applied to every network that can induce desired
changes in specified unperturbed nodes in one of the networks, but not in the other networks.”
The minor perturbation means that the perturbation in which a node can return to its
original state once the perturbation is removed. When the node changes into another steady
state, it is no longer a minor perturbation as it has incurred a phase transition onto the
network. Such perturbations are called a “major perturbation”. As a result of perturbations
imposed on nodes in the network, states of unperturbed nodes are affected. When a state of a
node has changed, but returned to the original state, then the change is called a minor change.
When the state of a node has been so affected that it caused transition to a new steady state,
such a change is called a “major change”. Networks are classed as being identical only when
their connection topology, interconnection parameters, and node functions are the same. The
“target node” is the node that undergoes the desired change, and other nodes are called “off-
target nodes”. A network where desired change of target nodes is expected is called a “target
network”, and other networks are called “off-target networks.” The term “desired change”
means either a major change which transit to a specific new steady state, a major change
without a new steady state specified, or a minor change above a specific threshold defined.
When the desired change of one or more nodes in the target network requires a major change,
the condition is called “strong”; the condition is called “weak” otherwise. An additional
condition, a “super-weak” condition, can be defined that sets an upper bound on the level of
minor perturbations and minor changes in off-target nodes of the target network and all nodes
in off-target networks. Also, a stochastic version of the problem can be defined and could be
very important for realistic use of solutions to biological problems.
Challenges are to find a set of perturbations with specific nodes to be perturbed, the level of
perturbations, and timing of perturbations that leads to the desired change only on the target
nodes of the target network. It is most likely that such perturbation is possible only in the
Nature Reviews | Drug Discovery
Intensity
Intensity
Intensity
Frequency FrequencyFrequency
Spreading
Despreading
© 2007 Nature Publishing Group
SUPPLEMENTARY INFORMATION In format provided by Kitano (MARCH 2007)
NATURE REVIEWS | DRUG D ISCOV ERY www.nature.com/reviews/drugdisc
specific set of networks and in specific desired changes. Thus, a whole set of theorems has to be
created to solve this problem. The problem is called the spread spectrum control problem
because the aim of perturbation is to control the state of the network, by means of changing
states of nodes in the network, without directly perturbing these nodes. The perturbations are
expected to spread over numbers of nodes, rather than one small number of nodes, to attain
the expected outcome. The possible solution is also analogous to the spread spectrum
technique for wireless digital communications.
The long-tail control problem can be stated as:
“Given the two or more networks that are not identical but interconnected, find a set of
sequential (minor) perturbations on nodes in networks that are on the long-tail of a defined
distribution index, whereas such perturbations are similarly applied to every network to
induce desired global changes networks.”
Here, nodes on the long-tail distribution means nodes that are classified below the defined
rank based on a certain index such as connectivity, expression level, the degree of change in
expression, etc. Such an index can be problem-dependent, and may require further elaboration
of the problem statement. “Desired global change” is another ambiguous term, but it means
that the state of connected network to be induced to a certain attractor in the certain phase
space. Again, the choice of dimensions for the phase space is arbitrary and problem-dependent
requiring further elaboration. It is unclear if “minor” perturbation is required in the problem
statement, but the problem can be subdivided if the solution can be founded only by using
minor perturbations or not.
These two mathematical problems are not only challenging to mathematics, complex
systems, and engineering studies, they are directly relevant to drug discovery.
© 2007 Nature Publishing Group
