Mathematical Biosciences 215 (2008) 193–195
Contents lists available at ScienceDirect
Mathematical Biosciences
journal homepage: www.elsevier.com/locate/mbs
Book Review
Uri Alon, An Introduction to Systems Biology: Design Principles of
Biological Circuits, Chapman & Hall/CRC, London, ISBN
1584886420, GBP 30.99, 2007 (320 pp.).
Over a dozen books carrying the term ‘‘systems biology” in the
title have been published recently, riding the recent rise of this
field to scientific prominence. Most of these books [e.g., [1–5]1
emphasize the analysis of ‘‘omic” experimental data and/or
large-scale mathematical modeling so as to describe how biological systems work at the molecular level. While this is doubtlessly
a very important problem in biology, the object of this review focuses on a no less important problem. Namely, that of discovering
design principles in biological circuits and understanding why nature adheres to those principles.
From decades of research in molecular biology it is emerging
that nature has converged time and again on similar molecular circuits. Such extensive convergent evolution owes to a combination
of the following four factors. First, various physical–chemical constraints limit what molecular circuits are feasible. Second, not all
feasible circuits are equally likely to be created form an existing
biological network by random mutation. Third, only circuits that
function very well and whose performance is robust to fluctuations
enable the carrying organisms to survive natural selection. Fourth,
similar functional properties (e.g., fast responses, robustness to
perturbations, high gains of outputs with respect to inputs) are required quite often in different contexts. Because most feasible circuits that could evolve from existing biological networks through a
small number of frequent genetic events (mostly point mutations,
gene duplications and gene deletions) are rarely found in living
organisms, the last two factors are critical for explaining the observed convergence. This convergence and its underlying causes
have deep implications. They mean that molecular biology might
one day be structured around a number of simple laws or principles whose understanding hinges largely on engineering considerations similar to those applying to human-designed circuits. The
major breakthroughs in the exact sciences occurred when the main
regularities (laws) were discovered and then explained. From this
process ensued the predictive power that earned these sciences the
qualifier ‘‘exact”, which still sets them apart from biology. If a similar process is nowadays taking place in molecular biology this is
largely through the discovery and explanation of design principles.
The focus on discovering and explaining engineering-derived
design principles of biological networks was pioneered by Michael
Savageau in the early 1970’s. And indeed it is to Michael Savageau’s 1976 classic ‘‘Biochemical Systems Analysis: A Study of
Function and Design in Molecular Biology” [6] that Uri Alon’s book
most directly compares. Both books focus on design principles of
biological networks, both are designed to serve as textbooks for
graduate or advanced undergraduate students with a standard
1
This is just a list of books I happened to have the occasion to consult recently. It is
neither the result of a systematic survey of the literature on systems biology or an
implicit endorsement of these books.
doi:10.1016/j.mbs.2008.07.002
background in either biological or exact sciences/engineering, both
are highly relevant and thought-provoking reading for advanced
researchers as well, both draw extensively on the respective
author’s research, both are rich in bright insights on why biological
circuits are as they are. Unfortunately, Michael Savageau’s book
has long been out of print, and used volumes are currently selling
for US$300. However, the lucky reader that manages to get hold of
both books will find that despite the above-mentioned similarities
these books cover complementary topics, use different approaches
and offer complementary perspectives on topics they have in common. Thus, whereas Savageau’s book devotes considerable space to
discussing mathematical modeling approaches and trades conciseness for deeper and more nuanced discussions of the various topics, Alon’s book discusses just the strictly necessary essentials of
modeling and emphasizes conciseness and simplicity. The latter
book also benefits greatly from the hindsight gained from 30 intervening years of biological research and, in particular, from the
author’s direct experimental tests of key concepts. It includes
chapters on bacterial chemotaxis, robust patterning in development, kinetic proofreading and optimal gene expression levels,
topics that were less extensively or not at all covered in Savageau’s
book. On the other hand, it gives little attention to metabolic networks, which deserved substantial attention in Savageau’s book.
The field of systems biology often appears daunting to biology
and exact sciences/engineering students alike. To the former because of the mathematics involved, to the latter because of the
amount of biological information they perceive as having to learn.
Uri Alon’s book contributes substantially to attenuate these fears.
It provides very concise and clear explanations of the basic biology
involved and then goes on to develop an intuitive understanding of
the issues based on very simple ‘‘toy models”. More-detailed discussions of technical aspects are deferred to appendices so that
they do not interfere with the flow of the main text. The required
mathematical skills are, in general, within the reach of advanced
undergraduate biology students. Each chapter is complemented
with a superb set of exercises. Each of these sets starts with a
few problems whose solution is explained step by step and then
presents several problems whose solutions are not provided. The
exercises are not only designed to train the student’s analytical
skills, but each addresses also a relevant and stimulating biological
issue or elaborates on issues discussed in the respective chapter.
They are neither trivial nor too daunting. Additionally, each chapter contains a ‘‘Further reading” section that provides a short selection of the most relevant references about each topic. (A more
extensive list of cited and ‘‘interesting” references is provided at
the end of the book.)
The printing I reviewed contains a fair number of typos. However, these are just a minor nuisance and do not hinder understanding of the text.
Most of the book is structured around the notion of network
motifs [7,8]. It is thus surprising, and what I find the book’s main
drawback, that it contains little critical discussion of the implications and caveats of this notion. Because the biological meaning
194
Book Review / Mathematical Biosciences 215 (2008) 193–195
and implications of the network motifs are prone to misinterpretation, it is worth discussing these issues here. Network motifs are
defined as patterns of interactions that occur in a regulatory network significantly more often than in randomized networks that
preserve the same number of edges (i.e. interactions) and nodes
(e.g., molecular species) [7,8]. Alon and co-workers hypothesize
that patterns of interactions that qualify as network motifs must
have been selected based on some advantage they give to the
organisms. Because molecular interactions are easily lost and created by mutation, they argue, mutational drift would otherwise
quickly have randomized those patterns. The idea of looking for
patterns that are significantly over-represented over what is expected in absence of selection has been quite successful in biological sequence analysis as a way to pinpoint binding sites for
transcription factors and other very recurrent sequence features.
However, its practical application to biological networks runs into
some difficulties. First, the expected distribution of interaction patterns in absence of selection may depart substantially from the
randomized networks envisaged by Alon and co-workers. This
happens because topological and physical–chemical constraints
prevent some interactions and favor others, and because the process whereby new genes are created—often gene duplication, with
the daughter gene initially inheriting the interactions of the parent
gene—further biases the distribution of interaction patterns. For instance, in metabolic networks each metabolite’s intrinsic chemical
properties severely limit the reactions in which it can participate,
and in protein interaction networks co-localization favors the creation of specific interactions [10]. Some patterns may thus qualify
as network motifs (in the sense defined above) as consequence of
these constraints and bias rather than as consequence of selection
[9]. This problem of evolutionary interpretation becomes acute
where, as in Chapter 6, profiles of motifs are compared among different types of networks that are subject to distinct constraints and
bias. The problem might be addressed by evaluating under/overrepresentation of the patterns with respect to more-realistic null
models that take all the constraints and bias above into account.
However, how such null models could be constructed remains unclear. A second concern is that statistics of interaction patterns
may be biased by the presently incomplete and fairly unreliable
knowledge about the structure of some biological networks.
Suppose now that we have comprehensive accurate information about a biological network, used the correct null model and
found some network motifs with respect to this null model. These
network motifs would then truly represent patterns that occur
more often than they would in absence of natural selection. Why
would these patterns become motifs? The prime necessary condition for a pattern of interactions to become a network motif in the
sense defined in this paragraph is that it can provide function(s)
that are required in many instances in the network. Furthermore,
in order to be selected over other patterns of interactions that
can provide similar functions the motifs should perform better
than these alternatives. However, non-motifs do not necessarily
perform poorly where they occur, nor are they necessarily under
weaker selection than motifs. They may just provide more unique
but no less essential or reliable functions.
Uri Alon partly deflects the concerns above by presenting network motifs simply as ‘‘a way to detect building-block patterns
in complex networks” (p. 27). Indeed—and irrespective of the
added value of statistical comparisons with respect to a questionable null model—the results and discussions in the book convincingly show that transcription and other biological networks are
mostly composed of a small number of highly recurrent patterns
of interactions, and that the frequent patterns differ among different types of networks in a way that can be rationalized with reference to the networks’ functions. It also shows that in the right
conditions the highly prevalent patterns can perform functions
X
Y
Z
Fig. 1. The incoherent type 2 feed-forward-loop. The –j symbols denote repression.
whose biological usefulness in the context where the patterns occur is plausible. In some cases it goes on to show that other patterns of interaction that are less prevalent tend perform those
functions less effectively or less robustly.
In face of the results above, it is tempting to think that a motif’s function could be univocally inferred from its structure or
from minimal information about the nature of the interactions
involved, and that a universal set of design principles would apply to all realizations of each motif. Unfortunately, these expectations may be over-optimistic. Many network motifs can
perform several different functions, depending on context, on
the properties of the edges, on how the inputs from different
edges relate to each other, etc. This is illustrated by the incoherent type 2 feed-forward-loop (FFL, Fig. 1), which is one of eight
possible types of FFL in transcriptional networks. Focusing just
in the case where the same signal modulates both transcription
factors and exploring a broad range of parameter values and the
three qualitatively different ways how the signal might influence
the action of each transcription factor, Wall et al. [11] find an
extensive repertoire of input–output patterns. This repertoire
of behaviors allows these circuits to alternatively provide various
different functions. Compounding the problem, transcription networks do not represent a totally autonomous layer of regulation,
but are instead enmeshed with metabolic and protein interaction networks. Thus, many ‘‘edges” in transcription networks
may actually represent complex processes, inputs to the network
motifs may be correlated, and there may be external (i.e. nontranscriptional) regulatory loops connecting outputs to inputs.
This suggests that one cannot in general reliably ascribe a function to a specific realization of a network motif without close
examination of its parameterization and biological context.
Perhaps nature will again reveal itself simpler than it could
be, and it will turn out that most instances of the same network
motif provide the same function and adhere to similar design
principles. This can only be ascertained after studying many
more biological circuits in detail. However, it has been a constant in the history of science that the laws of nature were only
discovered through painstaking work in identifying and eliminating potential confounding factors in a deliberate search for
regularities. Uri Alon’s book is an excellent contribution to motivate and prepare minds for such an endeavor in molecular biology. It is thus a must-read for any student entering the field of
systems biology and a very stimulating and enjoyable reading
for any researcher in the field.
References
[1] L. Alberghina, H.V. Westerhoff (Eds.), Systems Biology: Definitions and
Perspectives, Springer, Berlin, 2005.
[2] E. Klipp, R. Herwig, A. Kowald, C. Wierling, Systems Biology in Practice:
Concepts, Implementation and Application, Wiley-VCH, Weinheim, 2005.
[3] A. Kriete, R. Eils (Eds.), Computational Systems Biology, Academic Press, San
Diego, 2006.
[4] B.O. Palsson, Systems Biology: Properties of Reconstructed Networks,
Cambridge University Press, New York, 2006.
Book Review / Mathematical Biosciences 215 (2008) 193–195
[5] I. Rigoutsos, G. Stephanopoulos (Eds.), Systems Biology, Oxford University
Press, New York, 2007.
[6] M.A. Savageau, Biochemical Systems Analysis: A Study of Function and Design
in Molecular Biology, Addison-Wesley, Reading, Mass, 1976.
[7] S.S. Shen-Orr, R. Milo, S. Mangan, U. Alon, Network motifs in the
transcriptional regulation network of Escherichia coli, Nature Genetics 31
(2002) 64–68.
[8] R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii, U. Alon, Network
motifs: simple building blocks of complex networks, Science 298 (2002) 824–
827.
[9] Y. Artzy-Randrup, S.J. Fleishman, N. Ben Tal, L. Stone, Comment on network
motifs: simple building blocks of complex networks and superfamilies of
evolved and designed networks, Science 305 (2004) 1107c.
[10] J. Kuriyan, D. Eisenberg, The origin of protein interactions and allostery in
colocalization, Nature 450 (2007) 983–990.
195
[11] M.E. Wall, M.J. Dunlop, W.S. Hlavacek, Multiple functions of a feed-forwardloop gene circuit, Journal of Molecular Biology 349 (2005) 501–514.
Armindo Salvador
Molecular Systems Biology Group,Centre for Neuroscience and Cell
Biology, The University of Coimbra, Largo D. Diniz,
3004-535 Coimbra, Portugal
Biological Chemistry Group, Chemistry Department,
The University of Coimbra, Largo D. Diniz, 3004-535 Coimbra,
Portugal
Tel.: +351 91 9619593; fax: +351 239 827703
E-mail address: salvador@cnc.uc.pt