NETWORKS IN BIOLOGY

SPECIAL SECTION

VIEWPOINT

Biological Networks: The Tinkerer as an Engineer

U. Alon

This viewpoint comments on recent advances in understanding the design principles connection can be added that reduces modu-
of biological networks. It highlights the surprising discovery of “good-engineering” larity and increases the fitness of the network.
principles in biochemical circuitry that evolved by random tinkering. This is the reason that NNs almost always
display a nonmodular design. A clue to the
François Jacob pictured evolution as a tink- networks. Here are three of the most impor- reason that modules evolve in biology can be
erer, not an engineer (1). Engineers and tink- tant shared principles, modularity, robustness found in engineering (16). Modules in engi-
erers arrive at their solutions by very different to component tolerances, and use of recurring neering convey an advantage in situations
routes. Rather than planning structures in ad- circuit elements. where the design specifications change from
vance and drawing up blueprints (as an engi- The first principle, modularity (10–12), is time to time. New devices or software can be
neer would), evolution as a tinkerer works an oft-mentioned property of biological net- easily constructed from existing, well-tested

Downloaded from https://www.science.org at Universidad de Malaga on November 29, 2023

with odds and ends, assembling interactions works. For example, proteins are known to modules. A nonmodular device, in which ev-
until they are good enough to work. It is work in slightly overlapping, coregulated ery component is optimally linked to every
therefore wondrous that the solutions found groups such as pathways and complexes. En- other component, is effectively frozen and
by evolution have much in common with gineered systems also use modules, such as cannot evolve to meet new optimization con-
good engineering design (2). This Viewpoint subroutines in software (13) and replaceable ditions. Similarly, modular biological net-
comments on recent advances in understand- parts in machines. The following working works may have an advantage over non-
ing biological networks using concepts from definition of a module is proposed based on modular networks in real-life ecologies,
engineering. comparison with engineering: A module in a which change over time: Modular networks
Biological networks are abstract represen- network is a set of nodes that have strong can be readily reconfigured to adapt to new
tations of biological systems, which capture interactions and a common function. A mod- conditions (16, 17 ).
many of their essential characteristics. In the ule has defined input nodes and output nodes The second common feature of engineer-
network, molecules are represented by nodes, that control the interactions with the rest of ing and biological networks is robustness to
and their interactions are represented by edg- the network. A module also has internal component tolerances. In both engineering
es (or arrows). The cell can be viewed as an nodes that do not significantly interact with and biology, the design must work under all
overlay of at least three types of networks, nodes outside the module. Modules in engi- plausible insults and interferences that come
which describes protein-protein, protein- neering, and presumably also in biology, with the inherent properties of the compo-
DNA, and protein-metabolite interactions. In- have special features that make them easily nents and the environment. Thus, Escherichia
herent in this description is suppression of embedded in almost any system. For exam- coli needs to be robust with respect to tem-
detail: many different mechanisms of tran- ple, output nodes should have “low imped- perature changes over a few tens of degrees,
scription regulation, for example, may be de- ance,” so that adding on additional down- and no circuit in the cell should depend on
scribed by a single type of arrow. Further- stream clients should not drain the output to having precisely 100 copies of protein X and
more, the interactions can be of different existing clients (up to some limit). not 103. This point has been made decades
strengths, so there should be numbers or Why does modularity exist in biological ago for developmental systems (17, 18) and
weights on each arrow (3). Whenever two or networks? It is important to realize that not metabolism (2, 19, 20). The fact that a gene
more arrows converge on a node, an input all networks that evolve by tinkering are circuit must be robust to such perturbations
function needs to be specified (for example, modular. A well-studied example is computer- imposes severe constraints on its design:
AND or OR gates) (4, 5). At present, many of science neural networks (NNs). NNs are a set Only a small percentage of the possible cir-
the connections, numbers and input functions of interconnected nodes, each of which has a cuits that perform a given function can per-
are not known. However, something can still state that depends on the integrated inputs form it robustly. Recently, there have been
be learned even from the very incomplete from other nodes (14). As do protein signal- detailed experimental-theoretical studies
networks currently available (6–8). First, the ing networks, NNs function to process infor- that demonstrate how particular gene cir-
network description allows application of mation between input and output nodes (15). cuits can be robust, for example, in bacte-
tools and concepts (9) developed in fields In a way analogous to biological networks, rial chemotaxis (21, 22) and in fruit-fly
such as graph theory, physics, and sociology NNs are optimized by an “evolutionary” tink- development (23).
that have dealt with network problems before ering process of adding and removing arrows The third feature common to engineering
(see D. Bray on pg. 1864 in this issue). and changing their weights until the NN per- and biological networks is the use of recurring
Second, biological systems viewed as net- forms a given computational goal (gives the circuit elements. An electronic device, for ex-
works can readily be compared with engi- “correct” output responses to input signals). ample, can include thousands of occurrences of
neering systems, which are traditionally de- Unlike biological networks, however, NNs circuit elements such as operational amplifiers
scribed by networks such as flow charts and are nonmodular. They typically have a highly and memory registers. Biology displays the
blueprints. Remarkably, when such a com- interconnected architecture in which each same principle, using key wiring patterns again
parison is made, biological networks are seen node participates in many tasks. Viewed in and again throughout a network. Metabolic net-
to share structural principles with engineered this perspective, the modularity of biological works use regulatory circuits such as feedback
networks is puzzling because modular struc- inhibition in many different pathways (24). The
tures can be argued to be less optimal than transcriptional network of E. coli has been
Department of Molecular Cell Biology and Depart-
ment of Physics of Complex Systems, Weizmann In-
NN-style, nonmodular structures. After all, shown to display a small set of recurring circuit
stitute of Science, Rehovot, Israel 76100. E-mail: modules greatly limit the number of possible elements termed “network motifs” (25). Each
urialon@weizmann.ac.il connections in the network, and usually a network motif can perform a specific informa-

1866 26 SEPTEMBER 2003 VOL 301 SCIENCE www.sciencemag.org

 NETWORKS IN BIOLOGY

SPECIAL SECTION
tion processing task such as filtering out spuri- engineering, because prototypes often do not 5. Y. Setty, A. E. Mayo, M. G. Surette, U. Alon, Proc. Natl.
ous input fluctuation (25), generating temporal work and need to be understood in order to Acad. Sci. U.S.A. 100, 7702 (2003).
6. D. Thieffry, A. M. Huerta, E. Perez-Rueda, J. Collado-
programs of expression (3, 25) or accelerating correct their design. The program of molecular Vides, Bioessays 20, 433 (1998).
the throughput of the network (2, 26). Recently, biology is reverse-engineering on a grand scale. 7. T. I. Lee et al., Science 298, 799 (2002).
the same network motifs were also found in the Reverse engineering a nonmodular network of 8. E. H. Davidson et al., Science 295, 1669 (2002).
transcription network of yeast (7, 27). It is im- a few thousand components and their nonlinear 9. S. H. Strogatz, Nature 410, 268 (2001).
portant to stress that the similarity in circuit interactions is impossible (exponentially hard 10. L. H. Hartwell, J. J. Hopfield, S. Leibler, A. W. Murray,
Nature 402, C47 (1999).
structure does not necessarily stem from circuit with the number of nodes). However, the spe-
11. J. Ihmels et al., Nature Genet. 31, 370 (Aug, 2002).
duplication. Evolution, by constant tinkering, cial features of biological networks discussed 12. E. Ravasz, A. L. Somera, D. A. Mongru, Z. N. Oltvai,
appears to converge again and again on these here give hope that biological networks are A.-L. Barabási, Science 297, 1551 (2002).
circuit patterns in different nonhomologous sys- structures that human beings can understand. 13. C. R. Myers, arXiv: cond-mat/0305575 (2003).
tems (25, 27, 28), presumably because they Modularity, for example, is at the root of the 14. J. J. Hopfield, Proc. Natl. Acad. Sci. U.S.A. 79, 2554
carry out key functions (see Perspective (29) success of gene functional assignment by ex- (1982).
15. D. Bray, J. Theor. Biol. 143, 215 (1990).
STKE). Network motifs can be detected by pression correlations (11, 34). Robustness to 16. H. Lipson, J. B. Pollack, N. P. Suh, Evolution 56, 1549
algorithms that compare the patterns found in component tolerances limits the range of pos- (2002).
the biological network to those found in suitably sible circuits that function on paper to only a 17. J. Gerhart, M. W. Kirschner, Cells, Embryos, and Evo-
randomized networks (25, 27). This is analo- few designs that can work in the cell. This can lution: Toward a Cellular and Developmental Under-

Downloaded from https://www.science.org at Universidad de Malaga on November 29, 2023

gous to detection of sequence motifs as help theorists to home in on the correct design standing of Phenotypic Variation and Evolutionary
Adaptability (Blackwell Science Inc, Oxford, 1997).
recurring sequences that are very rare in with limited data (21–23). Network motifs de- 18. C. H. Waddington, Nature 150, 563 (1942).
random sequences. fine the few basic patterns that recur in a net- 19. H. Kacser, J. A. Burns, Symp. Soc. Exp. Biol. 32, 65
Network motifs are likely to be also work and, in principle, can provide specific (1973).
found on the level of protein signaling net- experimental guidelines to determine whether 20. M. Savageau, Nature 229, 542 (1971).
works (30). Once a dictionary of network they exist in a given system (25). These con- 21. N. Barkai, S. Leibler, Nature 387, 913 (1997).
22. U. Alon, M. G. Surette, N. Barkai, S. Leibler, Nature
motifs and their functions is established, cepts, together with the current technological
397, 168 (1999).
one could envision researchers detecting revolution in biology, may eventually allow 23. A. Eldar et al., Nature 419, 304 (2002).
network motifs in new networks just as characterization and understanding of cell-wide 24. D. Fell, Understanding the Control of Metabolism
protein domains are currently detected in networks, with great benefit to medicine. The (Portland Press, London, 1997).
the sequences of new genes. Finding a se- similarity between the creations of tinkerer and 25. S. S. Shen-Orr, R. Milo, S. Mangan, U. Alon, Nature
Genet. 31, 64 (2002).
quence motif (e.g., a kinase domain) in a engineer also raises a fundamental scientific
26. N. Rosenfeld, M. B. Elowitz, U. Alon, J. Mol. Biol. 323,
new protein sheds light on its biochemical challenge: understanding the laws of nature that 785 (2002).
function; similarly, finding a network motif unite evolved and designed systems. 27. R. Milo et al., Science 298, 824 (2002).
in a new network may help explain what 28. G. C. Conant, A. Wagner, Nature Genet. 34, 264 (2003).
29. A. Wagner, Sci. STKE 2003, pe41 (2003).
systems-level function the network per- References and Notes 30. C. V. Rao, A. P. Arkin, Annu. Rev. Biomed. Eng. 3, 391
forms, and how it performs it. 1. F. Jacob, Science 196, 1161 (1977). (2001).
2. M. Savageau, Biochemical Systems Analysis: A Study
Will a complete description of the biological of Function and Design in Molecular Biology (Addi- 31. M. E. Csete, J. C. Doyle, Science 295, 1664 (2002).
networks of an entire cell ever be available? son-Wesley, Reading, MA, 1976). 32. A. Arkin, P. Shen, J. Ross, Science 277, 1275 (1997).
3. M. Ronen, R. Rosenberg, B. I. Shraiman, U. Alon, Proc. 33. T. S. Gardner, D. di Bernardo, D. Lorenz, J. J. Collins,
The task of mapping an unknown network is Science 301, 102 (2003).
Natl. Acad. Sci. U.S.A. 99, 10555 (2002).
known as reverse-engineering (3, 31–33). 4. C. H. Yuh, H. Bolouri, E. H. Davidson, Science 279, 34. M. B. Eisen, P. T. Spellman, P. O. Brown, D. Botstein,
Much of engineering is actually reverse- 1896 (1998). Proc. Natl. Acad. Sci. U.S.A. 95, 14863 (1998).

VIEWPOINT

Social Insect Networks

Jennifer H. Fewell

Social insect colonies have many of the properties of adaptive networks. The simple How can viewing insect societies as net-
rules governing how local interactions among individuals translate into group be- works shape our understanding of social orga-
haviors are found across social groups, giving social insects the potential to have a nization and evolution? First, they have become
profound impact on our understanding of the interplay between network dynamics one of the central model systems for exploring
and social evolution. self-organization: the process by which interac-
tions occurring locally between individuals
The formal exploration of social insect col- groups generally) have key network attributes produce group-level attributes. Self-organi-
onies as networks is in its infancy. Howev- that appear consistently in complex biological zation in a social insect colony produces
er, social insects such as wasps, ants, and systems, from molecules through ecosystems; emergent properties: social phenotypes that
honeybees provide a powerful system for these include nonrandom systems of connectiv- are greater than a simple summation of
examining how network dynamics contrib- ity and the self-organization of group-level individual worker behaviors (2). The basic
ute to the evolution of complex biological phenotypes (1–3). Colonies exhibit multi- rules generating these dynamics are broad-
systems. Social insect colonies (and social ple levels of organization, yet it is still ly applicable across taxa whose members
possible to track individuals, making these show social behavior, and they produce
School of Life Sciences, Arizona State University,
societies more accessible to experimen- ubiquitous patterns of social organization,
Tempe, AZ 85287–1501, USA. E-mail: j.fewell@asu. tal manipulation than many other com- including mass action responses, division
edu plex systems. of labor, and social hierarchies (2, 4 ).

www.sciencemag.org SCIENCE VOL 301 26 SEPTEMBER 2003 1867

 Biological Networks: The Tinkerer as an Engineer
U. Alon

Science 301 (5641), . DOI: 10.1126/science.1089072

Downloaded from https://www.science.org at Universidad de Malaga on November 29, 2023

View the article online
https://www.science.org/doi/10.1126/science.1089072
Permissions
https://www.science.org/help/reprints-and-permissions

Use of this article is subject to the Terms of service

Science (ISSN 1095-9203) is published by the American Association for the Advancement of Science. 1200 New York Avenue NW,
Washington, DC 20005. The title Science is a registered trademark of AAAS.

American Association for the Advancement of Science