Systems Biology: A Brief Overview

Hiroaki Kitano
Science 295, 1662 (2002);
DOI: 10.1126/science.1069492

This copy is for your personal, non-commercial use only.

If you wish to distribute this article to others, you can order high-quality copies for your
colleagues, clients, or customers by clicking here.

Downloaded from www.sciencemag.org on June 3, 2012

Permission to republish or repurpose articles or portions of articles can be obtained by
following the guidelines here.

The following resources related to this article are available online at

www.sciencemag.org (this information is current as of June 3, 2012 ):

Updated information and services, including high-resolution figures, can be found in the online
version of this article at:
http://www.sciencemag.org/content/295/5560/1662.full.html
This article cites 18 articles, 11 of which can be accessed free:
http://www.sciencemag.org/content/295/5560/1662.full.html#ref-list-1
This article has been cited by 779 article(s) on the ISI Web of Science
This article has been cited by 100 articles hosted by HighWire Press; see:
http://www.sciencemag.org/content/295/5560/1662.full.html#related-urls
This article appears in the following subject collections:
Cell Biology
http://www.sciencemag.org/cgi/collection/cell_biol

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright
2002 by the American Association for the Advancement of Science; all rights reserved. The title Science is a
registered trademark of AAAS.
 SYSTEMS BIOLOGY: THE GENOME, LEGOME, AND BEYOND
REVIEW

Systems Biology: A Brief Overview

Hiroaki Kitano

To understand biology at the system level, we must examine the structure periments to identify specific interactions and
and dynamics of cellular and organismal function, rather than the char- conducting extensive literature surveys. Several
acteristics of isolated parts of a cell or organism. Properties of systems, attempts are under way to create a large-scale,
such as robustness, emerge as central issues, and understanding these comprehensive database on gene-regulatory
properties may have an impact on the future of medicine. However, many and biochemical networks (4). Although such
breakthroughs in experimental devices, advanced software, and analytical databases are useful sources of knowledge,
methods are required before the achievements of systems biology can live many network structures remain to be identi-
up to their much-touted potential. fied. Substantial research has been done on
expression profiling, in which clustering analy-
Since the days of Norbert Weiner, system-level must first examine how the individual com- sis is used to identify genes that are coexpressed
understanding has been a recurrent theme in ponents dynamically interact during opera- with genes of known function (5, 6). Although
biological science (1). The major reason it is tion. We must seek answers to questions such clustering analysis provides insight into the
gaining renewed interest today is that progress as: What is the voltage on each signal line? “correlation” among genes and biological phe-
in molecular biology, particularly in genome How are the signals encoded? How can we nomena, it does not reveal the “causality” of

Downloaded from www.sciencemag.org on June 3, 2012

sequencing and high-throughput measure- stabilize the voltage against noise and exter- regulatory relationships. Several methods have
ments, enables us to collect comprehensive data nal fluctuations? And how do the circuits been proposed to automatically discover regu-
sets on system performance and gain informa- react when a malfunction occurs in the sys- latory relationships solely on the basis of mi-
tion on the underlying molecules. This was not tem? What are the design principles and pos- croarray data (7–9). At present, such methods
possible in the days of Weiner, when molecular sible circuit patterns, and how can we modify use information derived from mRNA abun-
biology was still an emerging discipline. There them to improve system performance? dance, so there is limited scope to infer causal-
is now a golden opportunity for system-level A system-level understanding of a biolog- ity based on transcriptional regulation. Posttran-
analysis to be grounded in molecular-level un- ical system can be derived from insight into scriptional and posttranslational mechanisms of
derstanding, resulting in a continuous spectrum four key properties: regulation must be incorporated as large-scale
of knowledge. 1) System structures. These include the net- data become available, but many properties
System-level understanding, the approach work of gene interactions and biochemical have yet to be measured with sufficient accura-
advocated in systems biology (2), requires a pathways, as well as the mechanisms by which cy or in high throughput. Although it is not
shift in our notion of “what to look for” in such interactions modulate the physical proper- possible to incorporate all the desired data into
biology. While an understanding of genes and ties of intracellular and multicellular structures. the automated discovery system, analysis of
proteins continues to be important, the focus is 2) System dynamics. How a system be- transcriptional regulation may provide very
on understanding a system’s structure and dy- haves over time under various conditions can useful information because of the possible hy-
namics. Because a system is not just an assem- be understood through metabolic analysis, potheses it generates to allow us to infer the
bly of genes and proteins, its properties cannot sensitivity analysis, dynamic analysis meth- network structure. In general, when multiple
be fully understood merely by drawing dia- ods such as phase portrait and bifurcation hypotheses are generated by automated discov-
grams of their interconnections. Although such analysis, and by identifying essential mecha- ery analysis, it reflects a lack of information.
a diagram represents an important first step, it is nisms underlying specific behaviors. Bifurca- This type of analysis can be combined with
analogous to a static roadmap, whereas what we tion analysis traces time-varying change(s) in entropy-based decision-making algorithms to
really seek to know are the traffic patterns, why the state of the system in a multidimensional theoretically suggest an experiment that most
such traffic patterns emerge, and how we can space where each dimension represents a par- reduces the number of ambiguous network hy-
control them. ticular concentration of the biochemical fac- potheses. Although such algorithms have yet to
Identifying all the genes and proteins in an tor involved. reach a level of practical application, they may
organism is like listing all the parts of an 3) The control method. Mechanisms that prove useful for determining the optimal order
airplane. While such a list provides a catalog systematically control the state of the cell can of experiments needed to resolve ambiguous
of the individual components, by itself it is be modulated to minimize malfunctions and hypotheses (10). Progress in this area would
not sufficient to understand the complexity provide potential therapeutic targets for treat- lead to an increased emphasis on hypothesis-
underlying the engineered object. We need to ment of disease. driven research in biology (Fig. 1).
know how these parts are assembled to form 4) The design method. Strategies to mod- Once we have attained an understanding of
the structure of the airplane. This is analo- ify and construct biological systems having network structure, we will be able to investigate
gous to drawing an exhaustive diagram of desired properties can be devised based on network dynamics. In reality, analysis of dy-
gene-regulatory networks and their biochem- definite design principles and simulations, namics and structure on the basis of network
ical interactions. Such diagrams provide lim- instead of blind trial-and-error. dynamics are overlapping processes, because
ited knowledge of how changes to one part of Progress in any of the above areas re- dynamic analysis may yield useful predictions
a system may affect other parts, but to under- quires breakthroughs in our understanding of of unknown interactions. For dynamic analysis
stand how a particular system functions, we computational sciences, genomics, and mea- of a cellular system, we need to create a model.
surement technologies, and integration of But first it is important to carefully consider the
Sony Computer Science Laboratories, Inc., 3-14-13 such discoveries with existing knowledge. purpose of model building: Whether it is to
Higashi-Gotanda, Shinagawa, Tokyo 141-0022, Japan, Identification of gene-regulatory logic (3) obtain an in-depth understanding of system be-
and Kitano Symbiotic Systems Project, ERATO, JST,
and the Systems Biology Institute, Suite 6A, M31,
and biochemical networks is a major challenge. havior or to predict complex behaviors in re-
6-31-15 Jingumae, Shibuya, Tokyo 150-0001, Japan. The conventional methods for creating a net- sponse to complex stimuli, we must first define
E-mail: kitano@csl.sony.co.jp work model include performing a series of ex- the scope and abstraction level of the model.

1662 1 MARCH 2002 VOL 295 SCIENCE www.sciencemag.org

 SYSTEMS BIOLOGY: THE GENOME, LEGOME, AND BEYOND
The choice of analytical method used de- the cell cycle and circadian rhythms, and at the calization, and so forth. Model-based exper-
pends on the availability of biological knowl- circuit level, where it operates in alternative iment planning dictates where accuracy is
edge to incorporate into the model. A steady- metabolic pathways in E. coli. Structural stabil- critical and where it is not, so that resources
state analysis can be done using only the net- ity provides insensitivity to parameter changes can be optimally allocated.
work structure, without knowing the rate con- in the network responsible for segment forma- Complete system-level analysis of biolog-
stants for a particular reaction. For example, tion in Drosophila (18). And modularity is ex- ical regulation requires high throughput and
flux balance analysis (FBA) was used to predict ploited at various scales, from the cell itself to accurate measurements, goals that are per-
switching of the metabolic pathway in Esche- compartmentalized yet interacting signal-trans- haps beyond the scope of current experimen-
richia coli under different nutritional conditions duction cascades (19). tal practices. Technical innovations in exper-
based on knowledge of only the metabolic net- To conduct a systems-level analysis, a com- imental devices, single-molecule measure-
work structure; this was experimentally con- prehensive set of quantitative data is required. ments, femto-lasers that permit visualization
firmed (11). With some knowledge of steady- Projects already under way, such as the Alli- of molecular interactions, and nano-technol-
state rate constants, traditional stability analysis ance for Cellular Signaling (AfCS) (20), are ogies are critical aspects of systems biology
and sensitivity analysis provide insights into making large-scale measurements with the ul- research. For example, microfluidic systems,
how systems behavior changes when stimuli timate goal of creating an in-depth simulation also known as micro-TAS (total analysis sys-
and rate constants are modified to reflect dy- model of cells. Exploratory studies on modeling tem), enable minute quantities ( picoliters) of
namic behavior. Bifurcation analysis, in which should be done at the earliest stage of such a samples to be measured more rapidly and
a dynamic simulator is coupled with analysis project to identify where measurement bottle- more precisely. Various prototypes for poly-
tools, can provide a detailed illustration of dy- necks exist in building the final model and to merase chain reaction and electrophoresis
namic behavior (12, 13). This type of analysis avoid acquiring data with little value for model have been developed (21–24). Such methods

Downloaded from www.sciencemag.org on June 3, 2012

has become conventional in dynamic systems building, such as measurements of insufficient not only speed up measurements, but also
and is already used in many studies on biolog- coverage and accuracy. encourage automation.
ical simulation. Comprehensiveness in measurements re- Software infrastructure is another critical
Once both the network structure and its quires consideration of three aspects: (i) fac- component of systems biology research. Al-
functional properties are understood for a large tor comprehensiveness, which reflects the though attempts have been made to build sim-
number of regulatory circuits, studies on clas- numbers of mRNA transcripts and proteins ulation software and to make use of the many
sifications and comparison of circuits will pro- that can be measured at once; (ii) time-line analysis and computing packages originally de-
vide further insights into the richness of design comprehensiveness, which represents the signed for general engineering purposes, there
patterns used and how design patterns of regu- time frame within which measurements are is no common infrastructure or standard to en-
latory circuits have been modified or conserved made; and (iii) item comprehensiveness, able integration of these resources. The Sys-
through evolution. The hope is that intensive which refers to the simultaneous measure- tems Biology Mark-up Language (SBML),
investigation will reveal a possible evolutionary ment of multiple items, such as mRNA and along with CellML, represent attempts to define
family of circuits as well as a “periodic table” protein concentrations, phosphorylation, lo- a standard for an XML-based computer-
for functional regulatory circuits.
Robustness is an essential property of bio- Fig. 1. Hypothesis-driven
logical systems (14). Understanding the mech- research in systems biol-
anisms and principles underlying biological ro- ogy. A cycle of research
begins with the selection
bustness is necessary for an in-depth under- of contradictory issues of
standing of biology at the system level. The biological significance
phenomenological properties exhibited by ro- and the creation of a
bust systems can be classified into three areas: model representing the
(i) adaptation, which denotes the ability to cope phenomenon. Models
with environmental changes; (ii) parameter in- can be created either au-
tomatically or manually.
sensitivity, which indicates a system’s relative The model represents a
insensitivity to specific kinetic parameters; and computable set of as-
(iii) graceful degradation, which reflects the sumptions and hypothe-
characteristic slow degradation of a system’s ses that need to be test-
functions after damage, rather than catastrophic ed or supported experi-
failure. In engineering systems, robustness is mentally. Computational
“dry” experiments, such
attained by using (i) a form of system control as simulation, on models
such as negative-feedback and feed-forward reveal computational ad-
control; (ii) redundancy, whereby multiple equacy of the assump-
components with equivalent functions are intro- tions and hypotheses
duced for backup; (iii) structural stability, embedded in each mod-
where intrinsic mechanisms are built to pro- el. Inadequate models
would expose inconsis-
mote stability; and (iv) modularity, where sub- tencies with established
systems are physically or functionally insulated experimental facts, and
so that failure in one module does not spread to thus need to be rejected
other parts and lead to system-wide catastrophe. or modified. Models that
Not surprisingly, these approaches used in en- pass this test become subjects of a thorough system analysis where a number of predictions may be
gineering systems are also found in biological made. A set of predictions that can distinguish a correct model among competing models is selected for
“wet” experiments. Successful experiments are those that eliminate inadequate models. Models that
systems. Bacterial chemotaxis is an example of survive this cycle are deemed to be consistent with existing experimental evidence. While this is an
negative feedback that attains all three aspects idealized process of systems biology research, the hope is that advancement of research in computa-
of robustness (15–17). Redundancy is seen at tional science, analytical methods, technologies for measurements, and genomics will gradually trans-
the gene level, where it functions in control of form biological research to fit this cycle for a more systematic and hypothesis-driven science.

www.sciencemag.org SCIENCE VOL 295 1 MARCH 2002 1663

 SYSTEMS BIOLOGY: THE GENOME, LEGOME, AND BEYOND
readable model definition that enables models peutic agents, just as plans for all high- 8. S. Imoto et al., Pacific Symposium on Biocomputing
to be exchanged between software tools. Sys- 2002 (World Scientific, Singapore, 2002), pp. 175–
rise building are required to undergo structural 186.
tems Biology Workbench (SBW) is built on dynamics analysis to confirm earthquake 9. C. Yoo et al., Pacific Symposium on Biocomputing 2002
SBML and provides a framework of modular resistance. (World Scientific, Singapore, 2002), pp. 498 –509.
open-source software for systems biology re- Although systems biology is in its infan- 10. T. Ideker et al., Pacific Symposium on Biocomputing
(World Scientific, Singapore, 2000), pp. 302–313.
search. Both SBML and SBW are collective cy, its potential benefits are enormous in both 11. J. Edwards et al., Nature Biotechnol. 19, 125 (2001).
efforts of a number of research institutions shar- scientific and practical terms. A transition is 12. M. Borisuk et al., J. Theor. Biol. 195, 69 (1998).
ing the same vision (25). occurring in biology from the molecular level 13. K. Chen et al., Mol. Biol. Cell 11, 369 (2000).
14. M. E. Csete, J. C. Doyle, Science 295, 1664 (2002).
How does the idea of systems biology im- to the system level that promises to revolu- 15. N. Barkai et al., Nature 387, 913 (1997).
pact pharmaceutical industries and medical tionize our understanding of complex biolog- 16. U. Alon et al., Nature 397, 168 (1999).
practice? The most feasible application of sys- ical regulatory systems and to provide major 17. T.-M. Yi et al., Proc. Natl. Acad. Sci. U.S.A. 97, 4649
(2000).
tems biology research is to create a detailed new opportunities for practical application of 18. V. Dassaw et al., Nature 406, 188 (2000).
model of cell regulation, focused on particular such knowledge. 19. G. Weng et al., Science 284, 92 (1999).
signal-transduction cascades and molecules to 20. Alliance for Cellular Signaling (http://www.cellular-
signaling.org/).
provide system-level insights into mechanism- References and Notes 21. M. Burns et al., Science 282, 484 (1998).
based drug discovery (26–28). Such models 1. N. Weiner, Cybernetics or Control and Communica- 22. R. Anderson et al., Nucleic Acids Res. 28, e60 (2000).
tion in the Animal and the Machine (MIT Press, Cam-
may help to identify feedback mechanisms that bridge, MA, 1948).
23. P. Simpson et al., Proc. Natl. Acad. Sci. U.S.A. 95,
offset the effects of drugs and predict systemic 2256 (1998).
2. H. Kitano, Foundations of Systems Biology (MIT Press, 24. P. Gilles et al., Nature Biotechnol. 17, 365 (1999).
side effects. It may even be possible to use a Cambridge, MA, 2001). 25. Additional information can be obtained at http://
multiple drug system to guide the state of mal- 3. E. H. Davidson et al., Science 295, 1669 (2002).

Downloaded from www.sciencemag.org on June 3, 2012

www.cds.caltech.edu/erato/ or http://www.sbml.org.
4. Examples of such databases are: Signal Transduction 26. J. Gibbs, Science 287, 1969 (2000).
functioning cells to the desired state with min- Knowledge Environment (STKE http://www.stke.org/); 27. C. Sander, Science 287, 1977 (2000).
imal side effects. Such a systemic response KEGG (http://www.genome.ad.jp/); EcoCyc (http:// 28. D. Noble, Science 295, 1678 (2002).
cannot be rationally predicted without a model ecocyc.org/). 29. I thank J. Doyle, M. Simon, and members of ERATO
of intracellular biochemical and genetic inter- 5. M. Eisen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 14863 Kitano project for fruitful discussions. Supported by
(1998). the ERATO and BIRD program of the Japan Science
actions. It is not inconceivable that the U.S. 6. S. Chu et al., Science 282, 699 (2000). and Technology Corporation, and the Rice Genome
Food and Drug Administration may one day 7. S. Onami et al., in Foundations of Systems Biology, H. and Simulation Project of the Ministry of Agriculture,
mandate simulation-based screening of thera- Kitano, Ed. (MIT Press, Cambridge, MA, 2001), pp. 59–75. Japan.

REVIEW

Reverse Engineering of Biological

Complexity
Marie E. Csete1 and John C. Doyle2*

Advanced technologies and biology have extremely different physical ty in components or the environment (14).
implementations, but they are far more alike in systems-level organization Biologists and biophysicists new to study-
than is widely appreciated. Convergent evolution in both domains pro- ing complex networks often express surprise at
duces modular architectures that are composed of elaborate hierarchies of a biological network’s apparent robustness
protocols and layers of feedback regulation, are driven by demand for (15). They find that “perfect adaptation” and
robustness to uncertain environments, and use often imprecise compo- homeostatic regulation are robust properties of
nents. This complexity may be largely hidden in idealized laboratory networks (16, 17), despite “exploratory mech-
settings and in normal operation, becoming conspicuous only when con- anisms” that can seem gratuitously uncertain
tributing to rare cascading failures. These puzzling and paradoxical fea- (18 –20). Some even conclude that these mech-
tures are neither accidental nor artificial, but derive from a deep and anisms and their resulting features seem absent
necessary interplay between complexity and robustness, modularity, feed- in engineering (20, 21). However, ironically, it
back, and fragility. This review describes insights from engineering theory is in the nature of their robustness and complex-
and practice that can shed some light on biological complexity. ity that biology and advanced engineering are
most alike (22). Good design in both cases (e.g.,
The theory and practice of complex engineer- approaches in biology have a long history (1, 2) cells and bodies, cars and airplanes) means that
ing systems have progressed so radically that but are just now receiving renewed mainstream users are largely unaware of hidden complexi-
they often embody Arthur C. Clarke’s dictum, attention (3–13), whereas systems-level design ties, except through system failures. Further-
“Any sufficiently advanced technology is in- has consistently been at the core of modern more, the robustness and fragility features of
distinguishable from magic.” Systems-level engineering, motivating its most sophisticat- complex systems are both shared and neces-
ed theories in controls, information, and com- sary. Although the need for universal principles
1
Departments of Anesthesiology and Cell and Devel- putation. The hidden nature of complexity of complexity and corresponding mathematical
opmental Biology, University of Michigan Medical (“magic”) and discipline fragmentation with- tools is widely recognized (23), sharp differenc-
School, Ann Arbor, MI 48109, USA. 2Control and in engineering have been barriers to a dialog
Dynamical Systems, Electrical Engineering, and Bio-
es arise as to what is fundamental about com-
engineering, California Institute of Technology, Pasa- with biology. A key starting point in devel- plexity and what mathematics is needed (24).
dena, CA 91125, USA. oping a conceptual and theoretical bridge to This article sketches one possible view, using
*To whom correspondence should be addressed. E- biology is robustness, the preservation of experience and theoretical insights from engi-
mail: doyle@cds.caltech.edu particular characteristics despite uncertain- neering complexity that are relevant to biology.

1664 1 MARCH 2002 VOL 295 SCIENCE www.sciencemag.org