Cite this article: Zhang Q, Bhattacharya S,
Andersen ME. 2013 Ultrasensitive response
motifs: basic amplifiers in molecular signalling
networks. Open Biol 3: 130031.
Received: 15 February 2013
Accepted: 2 April 2013
systems biology/molecular biology/
ultrasensitivity, sigmoid, response coefficient,
Hill coefficient, motif
Author for correspondence:
Electronic supplementary material is available
Ultrasensitive response motifs:
basic amplifiers in molecular
Qiang Zhang, Sudin Bhattacharya and Melvin E. Andersen
Center for Dose Response Modeling, Institute for Chemical Safety Sciences,
The Hamner Institutes for Health Sciences, Research Triangle Park, NC 27709, USA
Multi-component signal transduction pathways and gene regulatory circuits
underpin integrated cellular responses to perturbations. A recurring set of net-
work motifs serve as the basic building blocks of these molecular signalling
networks. This review focuses on ultrasensitive response motifs (URMs) that
amplify small percentage changes in the input signal into larger percentage
changes in the output response. URMs generally possess a sigmoid input–
output relationship that is steeper than the Michaelis–Menten type of response
and is often approximated by the Hill function. Six types of URMs can be com-
monly found in intracellular molecular networks and each has a distinct kinetic
mechanism for signal amplification. These URMs are: (i) positive cooperative
binding, (ii) homo-multimerization, (iii) multistep signalling, (iv) molecular
titration, (v) zero-order covalent modification cycle and (vi) positive feedback.
Multiple URMs can be combined to generate highly switch-like responses. Ser-
ving as basic signal amplifiers, these URMs are essential for molecular circuits
to produce complex nonlinear dynamics, including multistability, robust adap-
tation and oscillation. These dynamic properties are in turn responsible for
higher-level cellular behaviours, such as cell fate determination, homeostasis
and biological rhythm.
Cells constantly sense changes in their surrounding environment and elicit appro-
priate responses. These responses require information about the surroundings to
be conveyed into and then processed by intracellular biochemical networks.
Although cellular responses can sometimes be proportional to the environmental
cues, biological signals often propagate in a nonlinear fashion, resulting in altered
amplitude, duration and phase [1–5]. Ultrasensitivity is a form of nonlinear signal
processing where a small fractional change in the input signal is amplified, produ-
cing a larger fractional change in the output response [6–8]. As a result, the output
is not a proportional function of the input, and viewed on a double-linear scale the
input–output (I/O) curve of an ultrasensitive response generally has a sigmoid
The term ‘amplification’ can cause confusion in biology, at times referring to
qualitatively different concepts. In some cases, the term is used to refer to absol-
ute concentration amplification, where a chemical species operating in a low
molar concentration range controls another species existing in a high molar con-
centration range. This form of amplification is necessary for control of actuator
molecules in the cellular machinery, where high abundance is needed to carry
out functions on a scale significant to the cell or tissue. A primary example is
the coagulation enzyme cascade that may start with a few molecules of factor
XII and culminate in the activation of millions of times more fibrin molecules.
&2013 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution
License http://creativecommons.org/licenses/by/3.0/, which permits unrestricted use, provided the original
author and source are credited.
In most other contexts (and also here in this review), the term
‘amplification’ means relative concentration amplification
or sensitivity amplification. Weber’s law, which states that
sensation of the environment by an organism works by
recognition of the relative change in the perceived signal,
also appears to operate at the molecular signalling level
. The magnitude of relative (or fold) change in protein
signalling in response to extracellular stimuli can be more
robustly retained than the absolute change among a popu-
lation of isogenic but otherwise heterogeneous cells , or
in the presence of perturbations that cause absolute protein
level changes . More importantly, phenotypic outcomes
such as embryonic development respond more consistently
to signals that retain the same fold, rather than absolute,
change , suggesting that cells choose to interpret relative
changes in the level of signalling molecules as the bona fide
instructing signals. The role of ultrasensitivity is to amplify
these relative changes at appropriate locations in molecular
Signal amplification through basic circuit units—referred to
here as ‘ultrasensitive response motifs’ (URMs)—is essential for
enabling multiple cellular dynamics. In the absence of URMs, a
signalling cascade is not even likely to output a linear response
owing to saturation of binding. Amplification via URMs can
make up for the amplitude loss and help maintain the dynami-
cal range of the original signal. A highly ultrasensitive motif can
function as a switch, transforming a continuous signal into an
all-or-none response. The functional importance of signal
amplification, as engendered by URMs, can be best understood
by studying complex nonlinear dynamics, such as bistability,
adaptation and oscillation. These dynamics are fundamental
to a multitude of integrated cellular functions, including pro-
liferation, differentiation, homeostasis and biological rhythm
[13–15]. URMs confer the nonlinearity necessary for these
dynamical properties to be rendered by properly structured
molecular networks. In this sense, URMs are the biochemical
equivalents of current- or voltage-amplifying transistors, the
fundamental building component of modern analogue and
digital electronic devices .
We begin the review by first introducing response coeffi-
cient as the measure of ultrasensitivity. We discuss how it
is related to the Hill function that is often invoked to appro-
ximate sigmoid responses. We then extensively cover six
distinct types of URMs. For each URM, we provide an intui-
tive explanation of the signal-amplifying mechanism as well
as a simple mathematical model to quantitatively illustrate
the chemical kinetics underlying amplification. Numerous
biological examples are covered to demonstrate the ubiquity
of ultrasensitivity in molecular signalling networks. In §5,
we illustrate, with feedback circuits capable of bistability,
adaptation and oscillation, the critical role of ultrasensitivity
in enabling complex dynamical behaviours. Mathematical
models discussed in the review are available in SBML
format as electronic supplementary material.
3.1. Response coefficient, ultrasensitivity and
The sensitivity of the steady-state stimulus–response func-
tion of a target molecular species Y that is directly or
indirectly regulated by a signalling molecular species X can
be quantified by the ratio of the fractional changes in Y and X:
R ¼ lim
R is known as response coefficient in metabolic control
analysis [17,18] and as logarithmic gain (‘gain’ for short)
in biochemical systems theory [19,20]. When R ¼ 1, the
response is proportionally linear. When R . 1, a small
percentage increase/decrease in X results in a larger percen-
tage increase/decrease in Y, indicating a response more
sensitive than the linear case. Ultrasensitivity is thus defined
as a response that has a response coefficient significantly
greater than 1. Conversely, when 0 , R , 1, a small percen-
tage increase/decrease in X results in an even smaller
percentage increase/decrease in Y, which is referred to as a
subsensitive response. When X inhibits Y, R has a negative
value, and the conditions jRj . 1 and 0 , jRj , 1 define
ranges of ultrasensitivity and subsensitivity, respectively.
If R remains constant as X varies, the steady-state
relationship between Y and X is described by the equation
lnY ¼ RlnX þ lnk;
where k is a constant. Transformed to a linear scale, it
Y ¼ kXR:
For R . 1 (i.e. an ultrasensitive response), the Y versus X
stimulus–response curve is concave upward; for 0 , R , 1
(i.e. a subsensitive response), the curve is concave downward
as X varies, the shape of the stimulus–response curve would
remain upward concave. Although ultrasensitivity is a form of
nonlinear amplification, as far as relative (percentage) change
is concerned, the amplification can be regarded as ‘linear’ as
long as R remains constant, as shown on the log–log scale
(figure 1a). However, the response coefficient of a signalling
cascade rarely stays constant with respect to the input signal.
An important feature of biochemical signalling is saturation
(i.e. when the input signal is sufficiently strong, the response
tends to level off). Thus, for an ultrasensitive motif that is
saturable, the response coefficient would decrease from R . 1
to R , 1 and to R ? 0 as the input signal intensifies.
Correspondingly, the upward concave curve would gradually
grow less steep as it moves first into a downward concave
(figure 1c). Therefore, ultrasensitivity is typically characterized
by a full-range steady-state response that is sigmoidally
lar hyperbola characterizing the Michaelis–Menten type of
3.2. Hill function
An ultrasensitive response is often empirically approximated
by the Hill function, which was initially derived from the
study of oxygen binding to haemoglobin :
Ymaxis the maximum activity of Y, K is the level of X produ-
cing a response of 50 per cent of Ymax and n is the Hill
Open Biol 3: 130031
coefficient determining the steepness of the curve. A measure
of the Hill coefficient n is provided by
where X0.1and X0.9are the levels of X associated with, respect-
ively, 10 per cent and 90 per cent of the maximum response
(figure 1d). The Hill coefficient n thus numerically quantifies
the steepness of a sigmoid curve relative to the hyperbolic
Michaelis–Menten curve, where X0.9/X0.1¼ 81 and n ¼ 1.
A higher n-value means a shorter distance between X0.1and
X0.9, and hence a steeper sigmoid curve. Unlike the response
coefficient, which defines the local ultrasensitivity (steepness)
of the stimulus–response curve, the Hill coefficient provides
a global measurement of the overall steepness of the curve.
The response coefficient of the Hill function is given by
Goldbeter & Koshland  and Sauro :
When X is very small compared with K, R approximately
equals n. Thus, the Hill coefficient, a metric describing the
global steepness of the Hill function, is equivalent to the
response coefficient at low input levels.
In many signalling cascades, the output may already have
some basal activities even in the absence of the input signal.
This situation can be described by the equation
where Y0is the basal activity of Y. The presence of the basal
activity desensitizes the response, particularly for low input
levels . After all, the sensitivity of a response, as measu-
red by the response coefficient, is related to the percentage
rather than absolute change. When Y0is sufficiently large,
ultrasensitivity can disappear completely even though the
response curve still remains fairly sigmoid (figure 2a–c).
Thus, response coefficient is always a more reliable measure
for ultrasensitivity than Hill coefficient, especially for stimu-
lus–response curves that cannot be easily fitted with Hill
functions. A simple way to visually gauge the degree of ultra-
sensitivity is to compare the slopes of the response curve with
straight lines of slope of unity in a log–log plot (figure 2d–f ).
Figure 1. Response coefficient, shape of ultrasensitive response curve and Hill function. (a) On a log–log scale, if response coefficient R remains constant, pro-
portional, ultrasensitive or subsensitive responses are straight lines of slope of 1, greater than 1 or less than 1, respectively. (b) On a linear scale, if response
coefficient R remains constant, a proportional response (R ¼ 1) is a straight line; an ultrasensitive response (R . 1) appears as a curve concave upward and
a subsensitive response (0 , R , 1) appears as a curve concave downward. (c) A typical saturable ultrasensitive stimulus–response has a sigmoid appearance
(blue curve, left y-axis). Not all regions of the sigmoid curve are ultrasensitive (i.e. capable of percentage amplification). The actual ultrasensitive region corresponds
to the range of X where the local response coefficient R (red curve, right y-axis) is greater than 1. (d) Hill function (blue curve) is frequently used to represent an
ultrasensitive response. The global steepness of the Hill curve is defined by the Hill coefficient n (see equation 3.5), which quantifies the relative fold change in the
level of X that produces from 10 to 90 per cent of the maximum response. The Michaelis–Menten response is plotted as a reference (grey curve).
Open Biol 3: 130031
4. Ultrasensitive response motifs
The empirical description of ultrasensitive response by the
Hill function does not provide necessary mechanistic insight,
and is sometimes inadequate to delineate the exact shape of
an actual stimulus–response curve. Ultrasensitivity must
arise from the kinetics of specific biochemical interactions.
Based on knowledge of known interactions, both theoretical
and experimental studies in the past several decades have
uncovered a number of URMs, which can be grouped by
and large into six common categories: (i) positive cooperative
binding, (ii) homo-multimerization, (iii) multistep signalling,
(iv) molecular titration, (v) covalent modification cycle (zero-
order ultrasensitivity) and (vi) positive feedback. Although
the topic of ultrasensitivity was previously reviewed by
others and us [8,9,24,25], here we attempt to provide a
much more comprehensive and up-to-date coverage of
these motifs, by elucidating their specific ultrasensitive
mechanisms and including relevant biological examples.
4.1. Positive cooperative binding
Many receptor proteins exist as multimeric complexes,
comprising multiple identical or similarly structured subunits.
Each subunit contains one binding site for one molecule
of the cognate ligand. According to the common Adair/
Koshland–Nemethy–Filmer model developed for oxygen
binding of haemoglobin [26–28], positive cooperative binding
occurs when the receptor subunits already occupied by ligand
molecules through early binding events can facilitate sub-
sequent binding of the remaining unoccupied subunits by the
ligand (figure 3a). This sequential increase in binding affinity
can result from allosteric interactions among the subunits of
the receptor. In the extreme case where the affinities of the
late binding events are enormously greater than those of the
early binding events, the receptor tends to reside in one of
two states: either free of any ligand molecules, or fully occupied
by ligand molecules. This is because once one subunit is occu-
pied first, binding to the remaining subunits will follow suit
quickly owing to the enhanced affinity. Such binding kinetics
tend to give rise to a sigmoid response in terms of percentage
receptor occupancy, with the degree of ultrasensitivity (or coop-
erativity) dependent on the total number of binding sites
per receptor molecule and the extent of increment in binding
affinity for sequential binding events. In the electronic supple-
mentary material, a mathematical model of ligand–receptor
interaction is provided to illustrate the ultrasensitive mechanism
of positive cooperative binding (motif 1).
Positive cooperative binding can provide ultrasensitive
signalling for second messengers. Activation of protein
kinase A (PKA) requires binding of cyclic adenosine mono-
phosphate (cAMP) to its regulatory subunit, which contains
two cAMP-binding sites. It was demonstrated that cAMP
binding to the regulatory subunit proceeds with positive
cooperativity, giving rise to a sigmoid PKA activation curve
with a Hill coefficient of 1.4–1.6 [29,30]. Ca2þas a second
messenger is also capable of positive cooperative binding.
Signalling information carried by Ca2þis routinely relayed
through Ca2þ-binding proteins such as calmodulin, cytosolic
phospholipase A2 and calretinin, which contain multiple
Ca2þ-binding sites. In vitro evidence indicates that occupancy
of these sites by Ca2þoften exhibits positive cooperativity of
various degrees [31–34]. Ultrasensitivity arising from positive
Figure 2 . Effect of basal activity of output on ultrasensitivity. (a–c) Solid blue curves describe the Y/Ymax(left y-axis) versus X stimulus–response as represented
by equation 3.7, and red curves are the corresponding response coefficient R (right y-axis). As the basal activity of Y (Y0) increases (from (a) through (c)), the
maximum response coefficient decreases. The actual ultrasensitive regions are marked by the shaded areas, which have response coefficients of more than 1. The
sigmoid response curve in (c) loses ultrasensitivity completely. (d–f ) Blue stimulus–response curves in (a–c) re-plotted on a log–log scale, respectively. The
degree of ultrasensitivity can be visually assessed by comparing the slopes of the stimulus–response curve with a series of straight lines of slope of unity (grey
lines). Ultrasensitivity is indicated when a section of the curve is steeper than the straight lines.
Open Biol 3: 130031
R RR RR R
Pro Pro Pro
Figure 3. Illustrations of ultrasensitive response motifs. (a) Positive cooperative binding between ligand L and multimeric (two subunits illustrated) receptor R. The
sequential increase in binding affinity is indicated by changes in the thickness of transition arrows. The overall activity of R is proportional to its percentage occu-
pancy by L. (b) Positive cooperative binding between TF and multiple response elements in gene promoters. The transcriptional activity of the promoter is
proportional to its percentage occupancy by TF. (c) Homo-multimerization of TFs to transcriptionally active multimers. Illustrated are TFs activated by ligand binding
to form homo-dimers, which gain affinity for DNA promoter. (d) Many inducible enzymes catalysing xenobiotic detoxification or metabolic reactions function as
homo-multimers. Here, inducible enzyme monomers E associate with one another to form homo-tetramers, which are fully enzymatically active to convert substrate
S to product P. (e) Synergistic multistep signalling where a TF directly increases the abundance of the target protein (Pro) through transcriptional induction, and
indirectly increases the activity of Pro (dashed line) through processes such as induction of a kinase (not shown) that phosphorylates and thus activates Pro. (f ) ATF
may increase the abundance of the target protein Pro through direct transcriptional induction, and indirectly by inhibiting degradation of Pro (dashed line) by
inducing factors (not shown) that stabilize Pro. (g) Multisite phosphorylation of protein substrate Pro by the same kinase in a non-processive manner is a
common multistep signalling ultrasensitive motif. (h) Molecular titration with decoy or dominant-negative receptor D competing with wild-type receptor R for
ligand L. (i) Molecular titration with transcriptional repressor R competing with activator protein A for transcription factor T. (j) Molecule I competitively inhibits
enzyme E, preventing it from binding to substrate S and catalysing the reaction. (k) Zero-order ultrasensitivity by covalent modification cycle. Protein substrate Pro
can be reversibly modified and de-modified by modifier enzyme (ME) and de-modifier enzyme (DE). (l) Positive gene auto-regulation where ligand L activates
receptor R, which transcriptionally upregulates its own abundance, thus forming a positive feedback loop. (m) Auto-catalysis where an activator, such as a kinase,
phosphorylates a substrate protein (Pro). Then phosphorylated Pro can also function as a kinase to catalyse its own phosphorylation. Solid arrow head, chemical
conversion or flux; empty arrow head, positive regulation; blunted arrow head, negative regulation.
Open Biol 3: 130031
cooperative binding in the second messenger system may
serve as an amplifying mechanism to ensure unattenuated
Cooperative binding can also occur between transcription
factors (TFs) and cis-regulatory response elements in gene
promoters (figure 3b). In the invertebrate and vertebrate
genomes, multiple response elements for a particular TF are
frequently clustered together [35–37], making cooperative
binding possible through allosteric interactions between adja-
cent elements. The cooperativity can also be facilitated by
protein–protein interaction between free and DNA-bound
TF molecules [38,39]. Bicoid, a morphogenic TF, forms a con-
centration gradient along the anterior–posterior (A–P) axis
in the early Drosophila embryo . Bicoid can bind to multi-
ple copies of a cis-acting consensus DNA sequence in a highly
cooperative manner, contributing to a sharp, nearly step-like
expression distribution along the A–P axis of some target
genes, such as hunchback, which direct embryonic pattern for-
mation [41–43]. Heat shock factors (HSFs), activated by rise in
temperature to induce heat shock and chaperon proteins, also
appear to interact with its target gene promoters in a highly
cooperative manner [38,44,45].
Many proteins function in the form of homo-multimers.
In the process of protein homo-multimerization (or homo-
oligomerization), identical monomers reversibly associate with
one another to form higher-order multimers that usually pos-
sess full functional activity. According to mass action kinetics,
the formation rate of the multimer varies as a power function
of the monomer concentration, with the exponent equal to
the order of multimerization (see motif 2 in the electronic
supplementary material). As a result, a linear increase in the
concentration of the free monomer would drive an ultrasensi-
tive increase in the steady-state concentration of the multimer.
In theory, the response coefficient can be as high as 2 for
homo-dimerization, 3 for homo-trimerization, and so on.
Protein homo-multimerization is a common step in
signal transduction, gene regulatory and metabolic networks.
Examples are formation of dimeric, trimeric or tetrameric
receptors, TFs and holoenzymes. Activation of cell membrane
receptors belonging to the receptor tyrosine kinase (RTK)
derived growth factor (PDGF) receptor, demonstrated that
phosphorylation of the receptor, when stimulated by PDGF
ligands, exhibited a sigmoid response with Hill coefficient of
1.55. Using mathematical modelling, they suggested dimeriza-
tion between two monomeric ligand–receptor complexes as a
possible mechanism behind the observed sigmoid response. By
contrast, the epidermal growth factor (EGF) receptor showed
negative cooperativity with its cognate ligand, a phenomenon
resulting from sequential ligand-binding kinetics in which the
affinity of the second EGF ligand-binding event (to singly
liganded receptor dimers) is allosterically weakened .
As an essential step towards their genomic action, steroid
hormone receptors associate into homo-dimers upon ligand
binding (figure 3c) to gain high affinity for the hormone
response elements in target genes [48–50]. In vitro binding
assays with oestradiol, progesterone and their cognate recep-
tors demonstrated that the monomer–homodimer kinetics
can lead to ultrasensitive responses in the formation of
ligand-bound receptors [51,52]. Many TFs activated by mech-
anisms other than ligand binding also function as high-order
homo-multimers. Although remaining to be validated exper-
imentally, activation of multimeric TFs is expected to exhibit
ultrasensitivity if the activating signal ultimately drives more
monomers to associate into multimers as opposed to just
modifying constitutively expressed, pre-existing multimers.
Examples of homodimeric TFs are tonicity-responsive enhan-
cer-binding protein (TonEBP or NFAT5) mediating osmotic
stress response , members of the signal transduction
and activator of transcription family , immediate early
gene products such as c-Jun  and the myogenic determi-
nation factor involved in muscle lineage development .
Active HSF is an example of a homo-trimer ; OxyR,
activated by oxidative stress in bacteria , and p53,
activated by DNA damage, are homo-tetramers .
Many metabolic enzymes induced in cellular stress response
act as homo-multimers (figure 3d). For instance, a suite of anti-
oxidant proteins induced by oxidative stress are homo-dimers
or homo-tetramers . In particular, glutathione peroxidase
and catalase, the two enzymes catalysing reactions to detoxify
hydrogen peroxide, function as homo-tetramers [61,62]. Anti-
stress proteins that function as homo-dimers also include metal-
lothioneins induced by heavy metal stress to chelate metal
molecules , and the growth arrest and DNA damage protein
(GADD45) induced by genotoxic stress to repair damaged
DNA and control cell growth [64,65]. Transcriptional induction
of monomeric proteins and their subsequent multimerization
into active enzymatic complexes play a crucial ultrasensitive
role in feedback networks that cope with cellular stresses to
maintain robust homeostasis [14,60]. In Escherichia coli, acti-
vation of glutamine synthetase (GS) by glutamine is a bicyclic
cascade process involving an intermediate protein PII; it was
found that PII functioning as a homo-trimer is a necessary
step towards rendering the activation of GS by glutamine
ultrasensitive, with a Hill coefficient as high as 6.5 in vitro .
4.3. Multistep signalling
Multistep signalling describes a signalling scheme where a
common input signal simultaneously regulates two or more
biochemical processes that synergistically activate an output
response. For instance, (i) a regulatory protein may increase
both the abundance and activity of a target protein, respect-
ively, through transcriptional induction and post-translational
modification; (ii) a TF may simultaneously induce the transcrip-
tion of a target gene and also indirectly inhibit the degradation
of its protein product; and (iii) a kinase may activate a target
protein through non-processive multisite phosphorylation
(figure 3e–g). In each of these signalling schemes, synergy
between parallel processes is manifested ultimately as multipli-
cative terms in the mathematical description of the output
response. As a result, ultrasensitivity would arise even if the
input signal regulates each individual process linearly. In the
electronic supplementary material, a mathematical model of
dual regulation is provided to illustrate the multistep signalling
effect (motif 3).
Non-processive (distributive) multisite protein phosphoryl-
ation by a single kinase is a common multistep signalling motif
[67,68]. In this situation, only fully phosphorylated or fully
dephosphorylated proteins are assumed to have the maximal
activity. Ultrasensitivity arising from multisite phosphorylation
can be understood by using mitogen-activated protein kinase
Open Biol 3: 130031
(MAPK) as an example. Dual phosphorylation of MAPK is
achieved through two separate reactive collisions (rather than a
single collision) between the MAPK kinase (MAPKK, as
enzyme) and MAPK (as substrate) molecules [69,70], during
ate product and then re-associates with MAPKK as a substrate.
ing amount of single-phosphorylated MAPK asthe substrate for
the second phosphorylation reaction, and (ii) an increasing
amount of kinase to catalyse the second phosphorylation reac-
tion. As a result, the production rate of dual-phosphorylated
MAPK can vary ideally as a square of MAPKK concentration,
contributing to MAPK ultrasensitivity . Were multisite phos-
phorylation achieved processively in a single collision, no
ultrasensitivity would arise. MAPK ultrasensitivity is also con-
tributed to by multisite dephosphorylation, and in this regard,
it has been shown that MAPK phosphatase-3 dephosphorylates
ERK2 in a non-processive manner .
In budding yeast, stoichiometric inhibitor of cyclin-depen-
dent kinase 1 (Cdk1; Sic1), a protein inhibiting G1/S phase
transition in the cell cycle, has to be phosphorylated on at
least six sites by the Cdk in order to be ubiquitinated for degra-
dation . The multisite phosphorylation process is believed
to occur in a distributive fashion (although this was recently
challenged ) and generate a potentially ultrasensitive
response that contributes to the bistable switch underlying
G1 to S phase transition [75,76]. Activation of transcription
factor NFAT1 requires dephosphorylation of 13 serine residues
by calmodulin-dependent phosphatase calcineurin; removal of
the multiple phosphate groups masks the nuclear export
sequence and exposes the nuclear import sequence, allowing
NFAT1 to translocate into the nucleus and become transcrip-
tionally active . Mathematical modelling predicted that if
some of the dephosphorylation steps proceed distributively
(i.e. with multiple association/dissociation events between cal-
cineurin and the intermediate substrates), an ultrasensitive
response with high Hill coefficient would arise . This may
partially explain the nonlinear induction of NFAT1 target
genes observed experimentally . Recently, Trunnell et al.
 demonstrated that activation of Cdc25C by Cdk1, two key
components involved in a bistable switch circuit responsible
for entry into mitosis, exhibits a highly ultrasensitive response
in Xenopus oocyte extracts. The ultrasensitive mechanism is
attributed to multisite phosphorylation of Cdc25C by Cdk1.
While increasing the number of phosphorylation sites
generally enhances the degree of ultrasensitivity, theore-
tical work has predicted that non-processive multisite
phosphorylation alone tends to create a response with a
threshold followed by a more graded change, rather than
an abrupt switch . To generate a switch-like response,
additional mechanisms are needed, including cooperativity
associated with sequential phosphorylation, competition for
kinase between intermediate substrates in variously phos-
phorylated states, substrate sequestration, sequential rather
than random phosphate processing and local kinase satur-
ation owing to anchorage of substrates to cell membranes
[67,68, 78,81–83]. Another situation that may complicate sig-
nalling through multisite phosphorylation is the existence of
scaffoldproteins, such asthose required for theMAPKcascade.
Computational studies have shown that scaffold proteins can
modulate MAPK activation in terms of magnitude, timing
and degree of ultrasensitivity [84,85]. By physically bringing
the kinases and their next-level protein substrates into close
proximity, scaffold proteins can increase signalling strength
and specificity. However, their existence may also diminish
MAPK signalling by (i) the ‘prozone effect’, wherein exces-
sive scaffold molecules may hold the kinases and substrates in
separate non-functional complexes, and (ii) the tendency of
on-scaffold kinases to be phosphorylated processively rather
Multistep signalling is also involved in the regulation of
protein activity by small signalling molecules. For example,
an increase in the AMP/ATP ratio as a result of energy
depletion activates AMP-activated protein kinase (AMPK)
through four distinct mechanisms simultaneously. These
multistep regulations include: (i) AMP allosterically activates
AMPK kinase (AMPKK), which phosphorylates (i.e. activates)
AMPK ; (ii) by binding to unphosphorylated AMPK, AMP
enhances the rate of phosphorylation of AMPK by AMPKK
; by binding to phosphorylated AMPK; (iii) AMP reduces
the rate of dephosphorylation of AMPK by phosphatases
; and (iv) AMP allosterically enhances the activity of
AMPK as a kinase . Together with some degree of
zero-order ultrasensitivity, these multiple signalling steps con-
tribute to a sigmoid activation of AMPK . In a similar
manner, Ca2þ/calmodulin-dependent kinase I is also activated
by Ca2þ/calmodulin through multistep regulations .
Many TFs involved in cellular stress response are activated
by stress signals in multiple ways. Under hypoxia, as O2
level decreases, proline hydroxylation of hypoxia inducible
factor-1a (HIF-1a) is diminished, which stabilizes HIF-1a,
leading to its accumulation [91,92]. Second, lower O2level
also decreases the hydroxylation of an asparagine residue
of HIF-1a, leading to enhanced transcriptional activity of
HIF-1a [93,94]. Together with a potential molecular titration
mechanism , this dual regulation by O2partial pressure
may lead to an ultrasensitive activation of HIF-1a under
hypoxia, which in turn contributes to an exponential or
switch-like induction of anti-hypoxic genes such as erythro-
poietin [96–98]. Another example of stress activation of TF
via multistep signalling is nuclear factor E2-related factor 2
(Nrf2). The cellular redox state regulates Nrf2 in at least
three ways. (i) An oxidative environment in the cell tends to
stabilize Nrf2 protein by inhibiting its redox-sensitive negative
regulator Kelch-like ECH-associated protein 1, which is an
adaptor protein for E3 ubiquitin ligase targeting Nrf2 for
proteasomal degradation [99,100]. (ii) The 50untranslated
region of Nrf2 mRNA contains an internal ribosomal entry
site, which can enhance the translation of Nrf2 protein in a
redox-sensitive manner . (iii) Nrf2 protein itself also con-
tains a redox-sensitive nuclear export signal that is inhibited by
an oxidative intracellular environment . Thus, under
oxidative stress, Nrf2 could accumulate in the nucleus under
three synergistic forces: (i) increased protein stabilization,
(ii) enhanced translation and (iii) increased nuclear retention.
4.4. Molecular titration
Many stoichiometric inhibitors exist in cells to scavenge
signalling molecules into inactive complexes, titrating them
away from downstream target molecules. Generic exam-
ples of titration (also termed as protein sequestration when
titration occurs between protein molecules) are: (i) wild-
type and decoy/dominant-negative receptors competing for
a common ligand [103–106]; (ii) TFs dimerizing with either
partner proteins to form a transcriptionally active complex
Open Biol 3: 130031
or with repressor proteins to form a transcriptionally inactive
complex [107–110]; and (iii) competitive enzyme inhibition
(figure 3h–j). Hidden in this seemingly trivial inhibition
scheme is an ultrasensitive response, which occurs when
the inhibitor (the total amount of which is I) exists in a
large quantity and the signalling molecule (the total
amount of which is S) has a higher binding affinity for the
inhibitor than for the target molecule [24,111,112]. Ultrasensi-
tivity arises near the point where nearly all of the inhibitor
molecules are ‘used up’ by forming inactive complexes with
the signalling molecules. At that point, any additional small
increase (DS) in the amount of the signalling molecules in
the system will be almost entirely available for binding to
its target molecules, thus producing a sharp increase in the
formation of the active complex. Mathematically, it is
straightforward to note that once the inhibitor is saturated,
the fractional increase in the available signalling molecule,
which is roughly equal to DS/(S 2 I), is always greater
than the fractional increase in the total signal molecule,
which is DS/S. Thus the ratio of the two fractions (i.e. the
response coefficient) will be greater than unity, and ultrasen-
sitivity is indicated. It is also obvious that a larger I denotes a
larger DS/(S 2 I) and hence a higher response coefficient. In
the electronic supplementary material, motif 4 illustrates this
A number of synthetic biology studies have provided con-
vincing experimental evidence for ultrasensitivity generated
via molecular titration. By introducing transcription factor
CCAAT/enhancer-binding protein a (CEBPa) and an engin-
eered high-affinity dominant-negative inhibitor into yeast
cells, Buchler & Cross  demonstrated a nearly switch-like
gene expression response that is consistent with ultrasensitivity
predicted by molecular titration. Many long stretches of non-
to act as repressive decoy TF-binding sites that can sequestrate
freeTFs . Recently, by introducingplasmid arrays contain-
ing a couple of hundred of non-functional tet operators into
budding yeasts, Lee & Maheshri  demonstrated that
gene expression driven by tet-transcriptional activators can be
converted into a sharp sigmoid response in the presence of
these repressive binding sites. Likewise, in vitro occupancy of
target DNA sequence by TATA-binding protein, as detected
by optical DNA sensors, also exhibited switch-like responses
in the presence of competing sequences .
Many signalling enzymes can act on two or more sub-
strates, and sometimes competitive inhibitors also exist. In
Xenopus oocyte extract, phosphorylation of Wee1 by Cdk1,
required for interphase to mitosis transition, is ultrasensitive,
a response partially arising from intermolecular competition
for Cdk1 between Wee1 and some unidentified substrates
. Ultrasensitivity arising from intramolecular titration was
recently observed in the spindle orientation signalling pathway
in Drosophila neuroblasts, which contains (i) heterotrimeric
G-protein a-subunit Gai, (ii) Partner of nscuteable (Pins) and
(iii) mushroom body defect (Mud) . Containing three bind-
ing domains for Gai (GL1, 2 and 3), a Pins molecule becomes
activated only when GL3 is occupied by Gai. Activated Pins
in turn recruits Mud to guide spindle alignment in preparation
for cell division. It was recently demonstrated that the non-func-
tional GL1 and GL2 domains actually serve as decoy binding
sites to sequester Gai away from GL3, leading to ultrasensitive
activation of Pins by Gai . In the anti-hypoxic stress path-
way, factor inhibiting HIF (FIH), which hydroxylates HIF-1a
at an asparagine residue, also has a broad range of ankyrin-
repeat domain (ARD)-containing proteins as substrate .
The competition for FIH between HIF-1a and ARD-containing
proteins was predicted to generate switch-like activation of
HIF-1a under hypoxia .
A variant of substrate competition is the branching point
in a metabolic pathway in which two different enzymes com-
pete for the same substrate with vastly different affinities
(Michaelis–Menten constants) and metabolize the substrate
into two different products. When the fraction of the meta-
bolic flux through the high-affinity enzyme branch is high,
the flux through the low-affinity enzyme branch can be ultra-
sensitive with respect to the substrate supply rate near the
point where the high-affinity enzyme is saturated .
This is similar to the idea that the stoichiometric inhibitor
needs to exist in high abundance for molecular titration to
display ultrasensitivity. Recently, this idea of ultrasensitivity
arising from flux competition has been extended to trans-
lational networks where mRNA molecules belonging to
different genes compete for access to a limited pool of ribo-
somes . Metabolic flux through a futile (substrate)
cycle is another example of ultrasensitivity that may be par-
tially explained in the spirit of substrate competition under
certain conditions. In this motif, the net flux flowing out of
the cycle is sensitive to changes in the forward flux of the
cycle when (i) the backward flux of the cycle (which takes
mass away from the net flux) is at a level close to the forward
flux and (ii) the enzyme catalysing the backward reaction is
The quantitative signalling properties of small non-coding
RNAs, which repress gene expression by promoting degra-
dation or inhibiting translation of mRNAs, have recently
been intensively investigated. Mathematical and experimental
studies indicate that small RNAs may regulate gene expression
in an ultrasensitive manner by titrating target mRNAs
[123–126]. In mammalian cells, the amount of protein trans-
lated by the target mRNA exhibited threshold-like response in
the presence of a specific microRNA, consistent with a model
of molecular titration . Ultrasensitivity through inhibitory
titration is also possible with expressed pseudogenes, which
functional proteins, or (ii) antisense RNAsthat bind to mRNAs
and inhibit translation .
4.5. Covalent modification cycle (zero-order
One of the most prevalent means by which protein activity is
regulated is post-translational covalent modification, such as
phosphorylation, acetylation and methylation. These covalent
modifications affect theaffinityof theprotein substrateforinter-
acting with other proteins, DNAs or small molecules, and can
thus effectively switch the activity of the protein substrate on
or off . The modification is usually reversible, involving
two opposing processes catalysed by specific enzyme pairs,
such as kinase/phosphatase, acetyltransferase/deacetylase
and methyltransferase/demethylase. By varying the active/
inactive ratio of the protein substrate, the modifier enzyme can
regulate its overall activity (figure 3k).
Theoretical studies by Goldbeter & Koshland [21,130] three
decades ago predicted that when the kinase and phosphatase
operate under conditions near saturation by their protein
Open Biol 3: 130031
substrates, an ultrasensitive response in substrate phosphoryl-
ation can be expected. Known as ‘zero-order ultrasensitivity’ in
covalent modification cycle, this prediction was later validated
experimentally with a number of signalling proteins and
enzymes. The initial evidence came fromisocitrate dehydrogen-
ase, an enzyme involved in the Krebs cycle and inhibited by
phosphorylation. In an in vitro assay system purified from
E. coli, phosphorylation of isocitrate dehydrogenase exhibited a
sigmoid steady-state response that could be partially attributed
to zero-order ultrasensitivity . Glycogen phosphorylase,
the glycogenolytic enzyme converting glycogen into glucose-1-
by dephosphorylation. Studies of an in vitro system containing
phosphorylase, phosphorylase kinases and protein phospha-
tase-1 purified from rabbit skeletal muscles showed that the
steady-state level of phosphorylated phosphorylase increased
ultrasensitively with a fitted Hill coefficient of 2.35 as the
kinase/phosphatase ratio was varied [132,133]. Mathematical
modelling strongly suggested that the ultrasensitivity exper-
imentally observed with the activation of AMPK by AMP in
INS-1 cells could stem from both multistep signalling as
described above and from a zero-order effect owing to possible
saturation of AMPKK by AMPK . Similarly, some degree
of zero-order ultrasensitivity through covalent modification
cycle also plays a role in the activation of MAPK by upstream
kinases [24,71]. In addition, MAPK appears to phosphorylate
some of its protein substrates with zero-order ultrasensitivity.
For instance, Yan, a transcriptional repressor involved in
patterning the Drosophila embryo, is regulated by MAPK. Phos-
phorylation of Yan by MAPK promotes its degradation,
resulting in a sharp, step-like expression pattern of Yan along
the medial–lateral axis in the ventral ectoderm . Zero-
conversion of small-molecule substrates, such as NAD and
NADH by the enzyme pair formate dehydrogenase and lactate
dehydrogenase, can also be switch-like .
Zero-order ultrasensitivity associated with covalent modifi-
cation cycles can be intuitively understood as follows. Consider
a system that is at the mid-point steady state with 50 per cent of
the protein substrate phosphorylated and the other 50 per cent
dephosphorylated. The total amount of the protein is large, far
exceeding the Michaelis–Menten constants of the kinase and
phosphatase. Both enzymes are thus near being saturated by
their respective substrates at the initial steady state. Owing to
saturation, both the phosphorylation and dephosphorylation
reaction rates are insensitive to changes in substrate concen-
trations (i.e. zero-order). Because the system is at steady state,
the rates of the two opposing reactions are in balance. Now, if
the kinase concentration increases slightly, the phosphorylation
rate would instantaneously exceed the dephosphorylation rate,
causing more protein molecules to become phosphorylated. As
the phosphatase is already saturated, any additional increase
in the concentration of its substrate (i.e. the phosphorylated
protein) would not increase the dephosphorylation rate to coun-
teract the increased phosphorylation rate. Similarly, because the
kinase is also saturated, any fractional decrease in the concen-
tration of its substrate (i.e. the dephosphorylated protein) has
little effect on the phosphorylation rate. As a result, the net
phosphorylation flux would continue until the dephosphory-
lated protein markedly decreases to a level where the kinase
is less saturated, and the phosphorylation rate and the depho-
sphorylation rate become equal again. A similar but opposite
response can be expected when the kinase concentration
decreases from the initial mid-point steady state. In either
case, a large swing in the concentrations of the phosphory-
lated and dephosphorylated protein will result, producing a
steep response. In the electronic supplementary material, by
overlaying the phosphorylation and dephosphorylation rate
illustrated graphically (motif 5).
4.6. Positive feedback
A signalling protein can activate itself through positive feed-
back, which can be in the form of gene auto-regulation,
auto-catalysis or through a feedback loop involving intermedi-
ate signalling molecules (figure 3l–m). A positive feedback
loop can behave as a bistable switch if one arm of the loop con-
tains an ultrasensitive motif locally (discussed in §5.1).
However, when there is no ultrasensitivity embedded within
any arm, the entire loop may function as a monostable URM
in response to a stimulatory signal external to the loop. In
this case, each arm of the feedback loop may only transfer
signal linearly, but ultrasensitivity arises because the signal
through the feedback can further activate the molecular
species that is directly stimulated by the external signal, thus
reinforcing the initial activation. Ultrasensitivity is expected
when the external signal and feedback signal impinge on sep-
arate but synergistic processes that regulate the common target
molecules. In the electronic supplementary material, a math-
ematical model of auto-catalysis is provided to illustrate how
ultrasensitivity can arise with positive feedback (motif 6).
As a common positive feedback motif, gene auto-regulation
allows a TF to induce its own transcription. This motif can be
frequently found in gene regulatory networks involved in
binary lineage specification during development, where attrac-
tor states representing different cell types need to be established
[136–141]. In theory, auto-regulatory motifs that are not them-
selves bistable could provide necessary ultrasensitivity for a
system containing coherently coupled feedback loops to pro-
duce robust bistability [142–145]. Positive feedback regulation
in nucleosome modification provides another mechanism for
switch-like gene induction by transcriptional activators .
In this framework of gene regulation, a TF, after binding to
a gene promoter, recruits histone-modifying enzymes. The
enzymes modify the chromosomal structure of the local
nucleosome to a configuration favouring transcription. The
structurally altered nucleosome is also able to recruit additional
histone-modifying enzymes to modify nearby nucleosomes to
a similar transcription-favouring state. This positive feedback
loop, operating locally between histone-modifying enzymes
and nucleosomes, has the potential to produce a highly
ultrasensitive response in gene activation , as well as
bistability that allows epigenetic memory .
Each of the six URMs described here has its own unique bio-
chemical, and therefore kinetic, basis for ultrasensitivity. To
some extent, the underlying mathematics for positive coop-
erative binding, multimerization and multistep signalling is
similar. The input signal of these three motifs would
appear somehow as a power function in the mathematical
terms describing the activation process, with the exponent
by and large reflecting the number of available binding
sites, order of homo-multimers or number of synergistic
Open Biol 3: 130031
signalling steps. Molecular titration and zero-order ultrasensi-
tivity take advantage of the dramatic kinetic changes in the
molecular binding process near saturation to achieve abrupt
responses. For positive feedback loops, the self-reinforcing
nature of signalling amplifies the initial activation many more
times to produce ultrasensitivity. Biological examples of these
ultrasensitive motifs are summarized in table 1.
The term ‘cooperativity’ is commonly used in biochemis-
try to describe synergistic binding events involving multiple
molecular subunits, but it is also loosely used in the literature
to refer to biochemical processes that generate sigmoid
responses through other ultrasensitive mechanisms. Notably,
none of the six motifs discussed here seems to produce
responses that can be fitted exactly with the Hill function.
Moreover, the estimated Hill coefficients of the sigmoid
responses are not necessarily true indicators of the maxi-
mal local response coefficients. In some cases, the maximal
response coefficient can be quite high even though the Hill
coefficient is only slightly greater than one (motif 6 in the
electronic supplementary material).
Although an individual URM may output steep sigmoid
responses, the degree of ultrasensitivity is limited by its kin-
etic mechanism and the cellular condition under which the
motif operates. In the case of positive cooperative binding,
multimerization and multistep signalling, the maximal
response coefficient depends, respectively, on the number
of available binding sites, order of homo-multimers and
number of synergistic signalling steps. A steeply sigmoid
response through cooperative binding requires multiple bind-
ing sites and highly allosteric interactions among these sites,
which is structurally challenging. For optimal ultrasensitivity
through homo-multimerization where the protein level is
transcriptionally regulated, it is preferable for the multimer
to be more stable than the monomer . Molecular
titration and zero-order ultrasensitivity are in theory capable
of producing nearly switch-like responses under appropriate
conditions (motifs 4 and 5 in the electronic supplementary
material). However, the specific state of the cell in vivo may
limit their capability. For covalent modification cycles, subopti-
mal conditions, such as random fluctuation owing to a limited
amount of converting enzymes , substantial seques-
tration of protein substrates by the converting enzymes or
downstream target proteins [153–155], spatial separation of
the opposing converting enzyme pairs , and existence
of converting enzymes that are not strongly irreversible ,
may compromise the degree of zero-order ultrasensitivity.
Consequently, it is common for multiple URMs to be
arranged together in signalling networks to achieve robustly
steep responses. The effect is similar to volume amplification
by connecting preamplifiers and power amplifiers in an audio
system. The maximal response coefficient of the combined
provide functional robustness , as the loss or weakening
of ultrasensitivity in one motif owing to gene mutation may
only partially compromise the overall ultrasensitivity .
A canonical example of motif combination is the MAPK signal-
ling cascade, which exhibits increasing response sigmoidicity
moving down the cascade [71,160]. The MAPK cascade ultra-
sensitivity can be attributed primarily to three mechanisms:
(i) multistep signalling through dual phosphorylation; (ii) zero-
order ultrasensitivity; and (iii) three-tiered structure of the
cascade to multiply the ultrasensitivity achieved at each tier.
Recently, it was also suggested that kinase cascading itself
might be another possible source of increased sigmoidicity
when enzyme distribution among intermediate complexes
is explicitly considered . Combination of various URMs is
Ca2þ/calmodulin-dependent protein kinase II (CaMKII) that
underlies long-term potentiation and memory formation in the
5. Ultrasensitivity and complex network
Complex dynamics of molecular signalling networks arise col-
lectively from interactions among individual components.
Multistability, adaptation, oscillation and chaos are common
examples of network dynamics. Higher-level cellular functions,
such as proliferation, differentiation, homeostasis, mobility,
metabolism and rhythmic behaviours, require proper inte-
gration of these dynamical properties across a multitude of
intricate biochemical networks. Using bistability, adaptation
and oscillation, we illustrate below the importance of signal
amplification conferred by ultrasensitivity in rendering these
dynamics from properly structured networks.
Many cellular-level responses, including proliferation, differ-
entiation, lineage specification and apoptosis, are all-or-none,
in which cells choose between two discrete outcomes. Once
cells commit to one fate over the other, the state transition is
usually irreversible under physiological conditions. Gene and
protein networks capable of bistability underpin the dis-
creteness and irreversibility of many of these all-or-none
responses [13,162]. Bistability generally requires two con-
ditions: (i) the network topology must be positive and/or
double-negative feedback loops; and (ii) at least one arm of
the feedback loop must embed motifs that can transfer signal
The ultrasensitivity requirement can be illustrated graphi-
cally with a simple two-variable system in which genes X and
Y activate each other transcriptionally, with linear degra-
dation of each gene product (figure 4a). Y activates X in a
simple Michaelis–Menten fashion, whereas X activates Y
ultrasensitively, as described by the Hill function:
Kxþ Y? k2X
yþ Xn? k4Y:
The possible steady states of this system appear as intersec-
tion points of the X and Y nullclines, which are obtained
by setting dX/dt and dY/dt to zero (figure 4b–d):
Open Biol 3: 130031
Table 1. Ultrasensitive regulations in molecular signalling networks. nH, Hill coefficient; n.a., not available.
activation of PKA by cAMP
Ca2þbinding to calmodulin
Ca2þbinding to cPLA2
Ca2þbinding to calretinin
activation of PDGFR by PDGF
activation of Mek-1 by Mos
(þ) cooperative binding
(þ) cooperative binding
(þ) cooperative binding
(þ) cooperative binding
(þ) cooperative binding
activation of p42 by Mos4.9
dissociation of Fus3 from ste5 stimulated by a-factor6
activation of CaMKI by Ca2þ/calmodulin
activation of CaMKII by Ca2þ
activation of Pins by Gai
regulation of transcription factors
activation of ER by oestradiol
activation of PR by progesterone
dephosphorylation of NFAT1 by calcineurin
activation of HIF-1 by low O2
activation of Nrf2 by ROSn.a. 
phosphorylation and degradation of Yan by Erk
transcriptional and translational regulation
bicoid promoter binding and induction of Hunchback
HSF promoter binding and induction of heat
gene induction by CEBPa in the presence of
stoichiometric protein inhibitor
gene induction by tet activators in the presence of
decoy DNA binding sites
binding of TATA-binding protein to target sequence
in the presence of depleting hairpin DNAs
nucleosome modification and recruitment of
translation of target mRNA in the presence of
regulation of metabolic enzymes and flux
adenylylation of glutamine synthetase activated
(þ) cooperative binding
(þ) cooperative binding
4.3molecular titration 
n.a.molecular titration 
Open Biol 3: 130031
For the system to be bistable, the two nullclines must intersect
each other three times, corresponding to two stable steady
states and one unstable steady state in between . Given
that the X nullcline bends upward or is at best a straight
line, the Y nullcline has to be sufficiently ‘twisted’ in a certain
way in order to cross the X nullcline back and forth multiple
times. This behaviour can be readily achieved when the Y
nullcline is sigmoid (i.e. when Y responds to X in a typical
ultrasensitive manner; figure 4c–d). The higher the degree
of sigmoidicity of the Y nullcline (achieved by increasing
the Hill coefficient n here), the more robust the bistability.
Conversely, if the Y nullcline is not ultrasensitive, the
system only has a single stable steady state (figure 4b).
On the other hand, if the X nullcline is sigmoid, the ultrasen-
sitivity requirement for the Y nullcline could be relaxed, still
permitting three intersection points. Thus, a certain degree of
ultrasensitivity in either of the two arms of a positive feed-
back loop is essential for bistability to arise. In addition to
the graphical argument, the requirement of ultrasensitivity
for bistability can be captured more formally by examining
the eigenvalues of the Jacobian matrix of the feedback
system, or, for a metabolic pathway, by examining the ratio
of the feedback elasticity and the degradation elasticity of
the product exerting the feedback [164,165].
5.2. Adaptation and cellular homeostasis
To survive, biological organisms must be able to adapt to a
fluctuating environment and maintain a relatively stable
internal milieu in both tissues and cells. At the cellular
level, many physical and chemical variables (such as cell
size, ions and oxygen) are maintained at a relatively constant
molecules such as reactive oxygen species (ROS), mis-
folded proteins and toxic metals have to be kept within
certain healthy ranges.
Although a feed-forward mechanism can be useful, nega-
tive feedback regulation is primarily responsible for robust
cellular adaptation and homeostasis [14,166]. Figure 4e illus-
trates a general homeostatic control scheme against cellular
stress. Cellular stressor S (S ¼ background/internal stressor
Sbkgþ external stressor Sext) increases the level of controlled
variable Y. Changes in the level of Y are sensed either directly
or indirectly by transcription factor T, which in turn induces
anti-stress gene G that functions to counteract changes in Y. In
a simple form such a feedback system can be described by the
following ordinary differential equations:
dt¼ k1S ? k2GY;
dt¼ k3Y ? k4T
Knþ Tn? k6G:
Such a model system can simulate a typical adaptive
response. At the onset of the stress, Y first spikes up and
then it gradually returns to a steady-state level close to the
baseline in the continuous presence of the stress (figure 4f).
Adaptation occurs because the anti-stress gene G is slowly
upregulated by T during the process (figure 4g). In the
absence of feedback control, Y is assumed to increase linearly
with S. To understand how ultrasensitivity modulates
the steady-state Y versus S response, we need to calculate
the systems-level response coefficient RY
S; which is
1 þ jr1r2r3j;
Table 1. (Continued.)
activation of AMPK by AMP 2.5multistep signalling
dephosphorylation of isocitrate dehydrogenase by
phosphorylation of phosphorylase
conversion between NAD and NADH by FDH and LDH
metabolism of isocitrate by lyase in the presence of
cell cycle control
degradation of Sic1 owing to phosphorylation by Cln-
molecular titration (multisite
phosphorylation of Cdc25c by Cdk1 2.3
phosphorylation of Wee1 by Cdk13.5
Open Biol 3: 130031
according to Kholodenko et al.  and Zhang & Andersen
. Here, r1, r2and r3are local response coefficients (gain)
for the regulation of T by Y, G by T and Y by G, respectively.
Because the feedback loop gain Rloop¼ jr1r2r3j ? 0, and thus
sensitive, appearing concave downward (figure 4h). Strong
S? 1; the steady-state Y versus S response is mostly sub-
homeostatic performance requires a small RY
requires a high loop gain. In the specific example here, a
high loop gain can be achieved by increasing Hill coefficient
n in the term describing the ultrasensitive induction of G by
T. This way, the percentage increase in the adapted steady-
state level of Y becomes much smaller than the increase in
S, which in turn
Figure 4. (Caption overleaf.)
Open Biol 3: 130031
stressor S. Multiple URMs with large r1, r2and r3can accom-
plish a high loop gain, thus ultrasensitivity helps negative
feedback loops to achieve robust adaptation and homeostasis
against external perturbations. Similarly, feedback control
through allosteric inhibition of enzymes by downstream pro-
ducts can also be found in metabolic pathways to maintain
flux or metabolite homeostasis [167,168].
For the negative feedback circuit in figure 4e, it is worth
noting that the steady-state expression level of anti-stress
gene G with respect to stressor S is governed by the following
systems-level response coefficient:
1 þ jr1r2r3j:
If the loop gain jr1r2r3j ? 1 and is primarily due to large r1
and/or r2, while r3?21, then RG
suggesting a linear relationship between G and S. Thus a nega-
tive feedback circuit embedding highly ultrasensitive motifs
may improve, counter-intuitively, the linearity of cell signal-
ling. Indeed, in the signal transduction pathway involving
the ultrasensitive MAPK cascade, which is encompassed in
various negative feedback loops , phosphorylation of
ERK (in NIH 3T3 cells) exhibited a nearly linear relationship
with the extracellular stimuli , a result that was predicted
by an earlier computational study . Similarly, engineered
negative gene auto-regulation circuits harbouring high-
degree cooperativity within the feedback loop in yeast cells
have been demonstrated to output linearized expression of
reporter genes with respect to the inducer concentration .
Stends to approach unity,
Many biological rhythms originate at the cellular level, with
oscillating periods ranging widely from seconds to days.
Examples include spontaneous action potential in cardiac
pacemaker cells, pulsatile hormone secretion from endocrine
cells and the circadian clock in the suprachiasmatic nucleus
neurons. Cells also exhibit oscillatory dynamics in response
to external perturbations, such as cytosolic Ca2þspikes stimu-
lated by activation of G-protein-coupled receptors, sustained
p53 pulses triggered by double strand DNA breaks, damped
NF-kB oscillation stimulated by lipopolysaccharide and
damped oscillatory response to iron stress in E. coli
[172–175]. While oscillation may arise from positive feedback
with substrate depletion , most cellular oscillatory beha-
viours require negative feedback as the primary network
structure. For sustained oscillation, the negative feedback top-
ology has to be complemented with two additional conditions:
(i) time delay in signalling and (ii) ultrasensitivity [177,178].
A simple two-variable negative feedback system with
time delay (figure 4i), described by the following two ordin-
ary differential equations, is used here to illustrate the role of
dt¼ k1? k2X ? Yðt ? tÞð5:10Þ
Knþ Xn? k4Y;
where X activates Y ultrasensitively as represented by the Hill
function, and Y inhibits X by promoting its degradation in a
linear fashion but with a time delay t. The role of ultrasensitiv-
ity and time delay for oscillation can be intuitively understood
as follows. A key kinetic property of biochemical processes,
which invariably involve molecular binding and unbinding,
is that steady states are usually approached asymptotically in
time. Therefore, for a linear negative feedback system (where
X activates Y linearly and Y inhibits X linearly), as X rises
and falls, Y would never be able to rise and fall by exactly
the same amplitude as X (same fold change, to be exact,
from peak to trough), even given sufficient time. This would
result in a pulse of Y of smaller amplitude than the preceding
X pulse. By the same token, the smaller Y pulse would in turn
lead to an even smaller X pulse, and so on. Thus, a linear nega-
tive feedback system can at best exhibit damped oscillation. In
a nonlinear feedback system, where X can activate Y ultrasen-
sitively, a pulse of X can result in a pulse of Y of larger
amplitude owing to signal amplification. This larger Y pulse
is then likely to promote a subsequent X pulse of equal or
even higher amplitude than the previous X pulse even if Y
only regulates X linearly. The non-diminishing X pulse
allows the process to repeat itself, resulting in sustained
Figure 4. (Overleaf.) Illustration of the roles of ultrasensitivity for complex network dynamics. (a–d) Ultrasensitivity is required for bistability. (a) Gene X and Y form a
double-positive feedback loop, where X activates Y in an ultrasensitive manner, and Y activates X in a Michaelis–Menten manner. The system is described by equations
(5.1) and (5.2), and the parameters are k1¼ 3, k2¼ 1, k3¼ 1, k4¼ 1, Kx¼ 2, Ky¼ 0.5 and n ¼ 1, 3 or 5. (b–d) Stability analysis using nullclines with different
n-values. The intersection points between X (red) and Y (blue) nullclines indicate the steady states of the feedback system (solid dot, stable steady state; empty dot,
unstable steadystate). The system is bistable when there are three intersection points: two stable steadystates and one unstable steadystate in between (c) and (d). The
Y nullclines in (c) and (d) show increasing degree of ultrasensitivity, making bistability arise easily. Reducing ultrasensitivity makes the X and Y nullclines difficult to
intersect three times, leading to monostability, as illustrated in (b). (e–h) Ultrasensitivity helps negative feedback loops to achieve robust cellular adaptation and
homeostasis. (e) A generic negative feedback circuit underlying cellular adaptation and homeostasis against stress. S represents the total stress level containing back-
ground/internal stress (Sbkg) and external stress (Sext), thus S ¼ Sbkgþ Sext. The system is described by equations (5.5)–(5.7), and the default parameters are k1¼ 1,
k2¼ 1, k3¼ 0.1, k4¼ 0.1, k5¼ 1.01, k6¼ 0.01, Sbkg¼ 1 and n ¼ 2. (f,g) Adaptive response of controlled variable Y and underlying induction of anti-stress gene
G under persistent external stress at various levels (Sext¼ 1, 2 and 3). Dashed lines are baseline levels of Y and G in the absence of Sext. (h) Adapted steady-state levels
of Y with respect to various levels of Sext. In the open-loop case (Rloop¼ 0), the response is linear (grey line). As Rloopincreases by setting Hill coefficient n ¼ 1, 2 and
3, the respective response (red, green and blue curves) becomes increasingly subsensitive, indicating improved adaptation and more robust homeostasis. To maintain the
same basal level of G, k5¼ 0.02, 0.11, 1.01 and 10.01 for n ¼ 0, 1, 2 and 3, respectively. (i–l) Ultrasensitivity is required for a negative feedback loop to generate
sustained oscillation. (i) Genes X (red) and Y (blue) form a negative feedback loop, where X activates Y in an ultrasensitive manner, and Y inhibits X linearly with a time
delay. The system is described by equations (5.10) and (5.11), and the parameters are k1¼ 1, k2¼ 1, k3¼ 1, k4¼ 1, K ¼ 3, t ¼ 5 and n ¼ 1, 2 or 3. t denotes
the time delay from Y to X. Initial X ¼ 3 and Y ¼ 0.5. ( j–l) As the Hill coefficient n increases from 1 to 3, the feedback system tends to oscillate better. Small n-values
only give rise to damped oscillation, whereas large n-values lead to sustained oscillation.
Open Biol 3: 130031
oscillation. Thus, ultrasensitivity compensates for the inherent
loss of pulse amplitude occurring in a linear system (figure 4j–
l). Increasing the time delay by increasing the number of inter-
mediate steps in the feedback loop generally relaxes the
requirement for the degree of ultrasensitivity and vice versa
[20,179,180]. It was long predicted that the intrinsically ultra-
sensitive MAPK cascade, when operating in a negative
feedback loop, may bring about sustained oscillations .
More recently, Shankaran et al.  indeed observed that
phosphorylation of ERK in the nucleus and cytoplasm of
human mammary epithelial cells stimulated by EGF is
robustly oscillatory, with pulse frequencies comparable with
those predicted by the earlier MAPK oscillation model. Finally,
ultrasensitivity is also required in the so-called relaxation oscil-
lator, which contains essentially a negative feedback loop and
a nested positive feedback loop. In this circuit, the positive
feedback loop with embedded ultrasensitivity provides a
reversible bistable switch, whereas the negative feedback
loop functions to drive the switch on and off periodically .
6. Concluding remarks
In the new millennium, as the connection details of large
molecular signalling networks are increasingly mapped out,
understanding their dynamical behaviours has become the
new challenge in biological research. Similar to the way engin-
eers learn how electrical circuits function, biologists need to
first discover and understand small network motifs and recur-
ring sub-networks before undertaking the task of making
sense of more complex biological networks. URMs, character-
ized by a sigmoid I/O relationship, can arise through a variety
of kinetic mechanisms. As central to cellular processes as
transistors are to modern electronics, URMs are the basic bio-
chemical signal amplifiers necessary for complex molecular
networks to generate bistability, adaptation/homeostasis,
oscillation and other nonlinear dynamics. The discovery and
characterization of these network motifs will continue to
help bring systems-level perspectives to the quantitative
investigation of existing and newly discovered biochemical
pathways and their attendant cellular outcomes.
We would like to thank the financial support from NIEHS-
P42ES04911, NIEHS-R01ES016005, NIEHS-R01ES020750, Dow
Chemical Company, ExxonMobil Foundation and the Long-
Range Research Initiative of the American Chemistry Council
for supporting this work. The authors have declared that there
are no conflicts of interest.
1.Tyson JJ, Chen KC, Novak B. 2003 Sniffers, buzzers,
toggles and blinkers: dynamics of regulatory and
signaling pathways in the cell. Curr. Opin. Cell Biol.
15, 221–231. (doi:10.1016/S0955-0674(03)00017-6)
Alon U. 2007 Network motifs: theory and
experimental approaches. Nat. Rev. Genet. 8,
Brandman O, Meyer T. 2008 Feedback loops shape
cellular signals in space and time. Science 322,
Sneppen K, Krishna S, Semsey S. 2010 Simplified
models of biological networks. Annu. Rev. Biophys. 39,
Tyson JJ, Novak B. 2010 Functional motifs in
biochemical reaction networks. Annu. Rev. Phys.
Chem. 61, 219–240. (doi:10.1146/annurev.
Kholodenko BN, Hoek JB, Westerhoff HV, Brown GC.
1997 Quantification of information transfer via
cellular signal transduction pathways. FEBS Lett.
414, 430–434. (doi:10.1016/S0014-
Legewie S, Bluthgen N, Herzel H. 2005 Quantitative
analysis of ultrasensitive responses. FEBS J. 272,
Goldbeter A, Koshland Jr DE. 1982 Sensitivity
amplification in biochemical systems. Q. Rev.
Biophys. 15, 555–591. (doi:10.1017/
Koshland Jr DE, Goldbeter A, Stock JB. 1982
Amplification and adaptation in regulatory and
sensory systems. Science 217, 220–225. (doi:10.
10. Ferrell Jr JE. 2009 Signaling motifs and Weber’s law.
Mol. Cell. 36, 724–727. (doi:10.1016/j.molcel.
11. Cohen-Saidon C, Cohen AA, Sigal A, Liron Y, Alon U.
2009 Dynamics and variability of ERK2 response
to EGF in individual living cells. Mol. Cell. 36,
12. Goentoro L, Kirschner MW. 2009 Evidence that fold-
change, and not absolute level, of b-catenin
dictates Wnt signaling. Mol. Cell. 36, 872–884.
13. Ferrell Jr JE. 2002 Self-perpetuating states in signal
transduction: positive feedback, double-negative
feedback and bistability. Curr. Opin. Cell. Biol. 14,
14. Zhang Q, Andersen ME. 2007 Dose response
relationship in anti-stress gene regulatory networks.
PLoS Comput. Biol. 3, e24. (doi:10.1371/journal.
15. Goldbeter A. 2002 Computational approaches to
cellular rhythms. Nature 420, 238–245. (doi:10.
16. Sauro HM, Kholodenko BN. 2004 Quantitative analysis
of signaling networks. Prog. Biophys. Mol. Biol. 86,
17. Fell DA. 1997 Understanding the control of
metabolism, p. 301. London, UK: Portland Press.
18. Heinrich R, Schuster S. 1996 The regulation of
cellular systems. New York, NY: Chapman & Hall.
19. Savageau MA. 1971 Concepts relating the behavior
of biochemical systems to their underlying
molecular properties. Arch. Biochem. Biophys. 145,
20. Savageau MA. 1976 Biochemical systems analysis: a
study of function and design in molecular biology.
Reading, MA: Addison-Wesley.
21. Goldbeter A, Koshland Jr DE. 1981 An amplified
sensitivity arising from covalent modification in
biological systems. Proc. Natl Acad. Sci. USA 78,
22. Hill AV. 1913 The combinations of haemoglobin
with oxygen and with carbon monoxide. I. Biochem.
J. 7, 471–480.
23. Sauro HM. 2011 Enzyme kinetics for systems biology.
Seattle, WA: Ambrosius Publishing.
24. Ferrell Jr JE. 1996 Tripping the switch fantastic: how
a protein kinase cascade can convert graded inputs
into switch-like outputs. Trends Biochem. Sci. 21,
25. Zhang Q, Bhattacharya S, Woods CG, Andersen ME.
2010 Ultrasensitive response motifs in biochemical
networks. In Quantitative modeling in toxicology
(eds K Krishnan, ME Andersen), pp. 199–217.
Hoboken, NJ: John Wiley & Sons.
26. Ferrell Jr JE. 2009 Q&A: cooperativity. J. Biol. 8, 53.
of experimental binding data and theoretical models
in proteins containing subunits. Biochemistry 5,
Open Biol 3: 130031
28. Adair GS. 1925 The hemoglobin system. VI. The
oxygen dissociation curve of hemoglobin. J. Biol.
Chem. 63, 529–545.
29. Robinson-Steiner AM, Corbin JD. 1983 Probable
involvement of both intrachain cAMP binding sites
in activation of protein kinase. J. Biol. Chem. 258,
30. Bubis J, Saraswat LD, Taylor SS. 1988 Tyrosine-371
contributes to the positive cooperativity between
the two cAMP binding sites in the regulatory
subunit of cAMP-dependent protein kinase I.
Biochemistry 27, 1570–1576. (doi:10.1021/
31. Kuznicki J, Strauss KI, Jacobowitz DM. 1995
Conformational changes and calcium binding by
calretinin and its recombinant fragments containing
different sets of EF hand motifs. Biochemistry 34,
15 389–15 394. (doi:10.1021/bi00047a001)
32. Malmberg NJ, Varma S, Jakobsson E, Falke JJ. 2004
Ca2þactivation of the cPLA2C2 domain: ordered
binding of two Ca22þions with positive
cooperativity. Biochemistry 43, 16320–16 328.
33. Crouch TH, Klee CB. 1980 Positive cooperative
binding of calcium to bovine brain calmodulin.
Biochemistry 19, 3692–3698. (doi:10.1021/
34. Valeyev NV, Bates DG, Heslop-Harrison P,
Postlethwaite I, Kotov NV. 2008 Elucidating the
mechanisms of cooperative calcium–calmodulin
interactions: a structural systems biology approach.
BMC Syst. Biol. 2, 48. (doi:10.1186/1752-0509-2-48)
35. Berman BP, Nibu Y, Pfeiffer BD, Tomancak P,
Celniker SE, Levine M, Rubin GM, Eisen MB. 2002
Exploiting transcription factor binding site clustering
to identify cis-regulatory modules involved in
pattern formation in the Drosophila genome. Proc.
Natl Acad. Sci. USA 99, 757–762. (doi:10.1073/
36. Segal E, Raveh-Sadka T, Schroeder M, Unnerstall U,
Gaul U. 2008 Predicting expression patterns from
regulatory sequence in Drosophila segmentation.
Nature 451, 535–540. (doi:10.1038/nature06496)
LA, Ovcharenko I. 2010 Homotypic clusters of
human promoters and enhancers. Genome Res. 20,
38. Erkine AM, Magrogan SF, Sekinger EA, Gross DS.
1999 Cooperative binding of heat shock factor to
the yeast HSP82 promoter in vivo and in vitro.
Mol. Cell. Biol. 19, 1627–1639.
39. Johnson AD, Meyer BJ, Ptashne M. 1979 Interactions
between DNA-bound repressors govern regulation by
the lambda phage repressor. Proc. Natl Acad. Sci. USA
76, 5061–5065. (doi:10.1073/pnas.76.10.5061)
40. Driever W, Nusslein-Volhard C. 1988 A gradient
of bicoid protein in Drosophila embryos. Cell 54,
41. Driever W, Nusslein-Volhard C. 1988 The bicoid
protein determines position in the Drosophila
embryo in a concentration-dependent manner. Cell
54, 95–104. (doi:10.1016/0092-8674(88)90183-3)
42. Gregor T, Tank DW, Wieschaus EF, Bialek W. 2007
Probing the limits to positional information. Cell
130, 153–164. (doi:10.1016/j.cell.2007.05.025)
43. Burz DS, Rivera-Pomar R, Jackle H, Hanes SD. 1998
Cooperative DNA-binding by Bicoid provides a
mechanism for threshold-dependent gene activation
in the Drosophila embryo. EMBO J. 17, 5998–6009.
44. Amin J, Fernandez M, Ananthan J, Lis JT, Voellmy R.
1994 Cooperative binding of heat shock
transcription factor to the Hsp70 promoter in vivo
and in vitro. J. Biol. Chem. 269, 4804–4811.
45. Xiao H, Perisic O, Lis JT. 1991 Cooperative binding of
Drosophila heat shock factor to arrays of a conserved
5 bp unit. Cell 64, 585–593. (doi:10.1016/0092-
46. Park CS, Schneider IC, Haugh JM. 2003 Kinetic
analysis of platelet-derived growth factor receptor/
phosphoinositide 3-kinase/Akt signaling in
fibroblasts. J. Biol. Chem. 278, 37064–37 072.
47. Alvarado D, Klein DE, Lemmon MA. 2010 Structural
basis for negative cooperativity in growth factor
binding to an EGF receptor. Cell 142, 568–579.
48. Wrange O, Eriksson P, Perlmann T. 1989 The
purified activated glucocorticoid receptor is a
homodimer. J. Biol. Chem. 264, 5253–5259.
49. Kumar V, Chambon P. 1988 The estrogen receptor
binds tightly to its responsive element as a ligand-
induced homodimer. Cell 55, 145–156. (doi:10.
50. Kuntz MA, Shapiro DJ. 1997 Dimerizing the estrogen
receptor DNA binding domain enhances binding to
estrogen response elements. J. Biol. Chem. 272,
51. Carbajo P, Christensen K, Edwards DP, Skafar DF.
1996 Binding of [3H]progesterone to the human
progesterone receptor: differences between
individual and mixed isoforms. Endocrinology 137,
52. Notides AC, Lerner N, Hamilton DE. 1981 Positive
cooperativity of the estrogen receptor. Proc. Natl
Acad. Sci. USA 78, 4926–4930. (doi:10.1073/pnas.
53. Lee SD, Woo SK, Kwon HM. 2002 Dimerization is
required for phosphorylation and DNA binding of
TonEBP/NFAT5. Biochem. Biophys. Res. Commun.
294, 968–975. (doi:10.1016/S0006-
54. Shuai K, Horvath CM, Huang LH, Qureshi SA,
Cowburn D, Darnell Jr JE. 1994 Interferon activation
of the transcription factor Stat91 involves
dimerization through SH2-phosphotyrosyl peptide
interactions. Cell 76, 821–828. (doi:10.1016/0092-
55. Halazonetis TD, Georgopoulos K, Greenberg ME,
Leder P. 1988 c-Jun dimerizes with itself and with
c-Fos, forming complexes of different DNA binding
affinities. Cell 55, 917–924. (doi:10.1016/0092-
56. Maleki SJ, Royer CA, Hurlburt BK. 1997 MyoD-E12
heterodimers and MyoD–MyoD homodimers are
equally stable. Biochemistry 36, 6762–6767.
57. Liu PC, Thiele DJ. 1999 Modulation of human heat
shock factor trimerization by the linker domain.
J. Biol. Chem. 274, 17 219–17225. (doi:10.1074/
58. Knapp GS, Tsai JW, Hu JC. 2009 The oligomerization
of OxyR in Escherichia coli. Protein Sci. 18,
59. Okorokov AL, Sherman MB, Plisson C, Grinkevich V,
Sigmundsson K, Selivanova G, Milner J, Orlova EV.
2006 The structure of p53 tumour suppressor
protein reveals the basis for its functional plasticity.
EMBO J. 25, 5191–5200. (doi:10.1038/sj.
60. Zhang Q, Pi J, Woods CG, Andersen ME. 2010 A
systems biology perspective on Nrf2-mediated
antioxidant response. Toxicol. Appl. Pharmacol. 244,
61. Asayama K, Yokota S, Dobashi K, Hayashibe H,
Kawaoi A, Nakazawa S. 1994 Purification and
immunoelectron microscopic localization of cellular
glutathione peroxidase in rat hepatocytes:
quantitative analysis by postembedding method.
Histochemistry 102, 213–219. (doi:10.1007/
62. Kirkman HN, Gaetani GF. 1984 Catalase: a tetrameric
enzyme with four tightly bound molecules of
NADPH. Proc. Natl Acad. Sci. USA 81, 4343–4347.
63. Suzuki KT, Yamamura M. 1980 Isolation and
characterization of metallothionein dimers. Biochem.
Pharmacol. 29, 689–692. (doi:10.1016/0006-
64. Kovalsky O, Lung FD, Roller PP, Fornace Jr AJ. 2001
Oligomerization of human Gadd45a protein. J. Biol.
Chem. 276, 39 330–39 339. (doi:10.1074/jbc.
65. Schrag JD, Jiralerspong S, Banville M, Jaramillo ML,
O’Connor-McCourt MD. 2008 The crystal structure
and dimerization interface of GADD45g. Proc. Natl
Acad. Sci. USA 105, 6566–6571. (doi:10.1073/pnas.
66. Jiang P, Ninfa AJ. 2011 A source of ultrasensitivity
in the glutamine response of the bicyclic cascade
system controlling glutamine synthetase
adenylylation state and activity in Escherichia coli.
Biochemistry 50, 10 929–10 940. (doi:10.1021/
67. Salazar C, Hofer T. 2009 Multisite protein
phosphorylation—from molecular mechanisms to
kinetic models. FEBS J. 276, 3177–3198. (doi:10.
68. Gunawardena J. 2005 Multisite protein
phosphorylation makes a good threshold but can be
a poor switch. Proc. Natl Acad. Sci. USA 102,
14617–14 622. (doi:10.1073/pnas.0507322102)
69. Ferrell Jr JE, Bhatt RR. 1997 Mechanistic studies of
the dual phosphorylation of mitogen-activated
protein kinase. J. Biol. Chem. 272, 19 008–19 016.
70. Burack WR, Sturgill TW. 1997 The activating dual
phosphorylation of MAPK by MEK is nonprocessive.
Open Biol 3: 130031
Biochemistry 36, 5929–5933. (doi:10.1021/
71. Huang CY, Ferrell Jr JE. 1996 Ultrasensitivity in the
mitogen-activated protein kinase cascade. Proc. Natl
Acad. Sci. USA 93, 10078–10 083. (doi:10.1073/
72. Zhao Y, Zhang ZY. 2001 The mechanism of
dephosphorylation of extracellular signal-regulated
kinase 2 by mitogen-activated protein kinase
phosphatase 3. J. Biol. Chem. 276, 32 382–32 391.
73. Nash P, Tang X, Orlicky S, Chen Q, Gertler FB,
Mendenhall MD, Sicheri F, Pawson T, Tyers M. 2001
Multisite phosphorylation of a CDK inhibitor sets a
threshold for the onset of DNA replication. Nature
414, 514–521. (doi:10.1038/35107009)
74. Koivomagi M, Valk E, Venta R, Iofik A, Lepiku M,
Balog ER, Rubin SM, Morgan DO, Loog M. 2011
Cascades of multisite phosphorylation control Sic1
destruction at the onset of S phase. Nature 480,
75. Harper JW. 2002 A phosphorylation-driven
ubiquitination switch for cell-cycle control. Trends
Cell. Biol. 12, 104–107. (doi:10.1016/S0962-
76. Deshaies RJ, Ferrell Jr JE. 2001 Multisite
phosphorylation and the countdown to S phase. Cell
107, 819–822. (doi:10.1016/S0092-
77. Okamura H et al. 2000 Concerted dephosphorylation
of the transcription factor NFAT1 induces a
conformational switch that regulates transcriptional
activity. Mol. Cell 6, 539–550. (doi:10.1016/S1097-
78. Salazar C, Hofer T. 2003 Allosteric regulation of the
transcription factor NFAT1 by multiple
phosphorylation sites: a mathematical analysis.
J. Mol. Biol. 327, 31–45. (doi:10.1016/S0022-
79. Dolmetsch RE, Xu K, Lewis RS. 1998 Calcium
oscillations increase the efficiency and specificity of
gene expression. Nature 392, 933–936. (doi:10.
80. Trunnell NB, Poon AC, Kim SY, Ferrell Jr JE. 2011
Ultrasensitivity in the regulation of Cdc25C by Cdk1.
Mol. Cell 41, 263–274. (doi:10.1016/j.molcel.2011.
81. Liu X, Bardwell L, Nie Q. 2010 A combination of
multisite phosphorylation and substrate
sequestration produces switchlike responses.
Biophys. J. 98, 1396–1407. (doi:10.1016/
82. Dushek O, van der Merwe PA, Shahrezaei V. 2011
Ultrasensitivity in multisite phosphorylation of
membrane-anchored proteins. Biophys. J. 100,
83. Salazar C, Hofer T. 2007 Versatile regulation of
multisite protein phosphorylation by the order of
phosphate processing and protein–protein
interactions. FEBS J. 274, 1046–1061. (doi:10.1111/
84. Kocieniewski P, Faeder JR, Lipniacki T. 2012 The
interplay of double phosphorylation and scaffolding
in MAPK pathways. J. Theor. Biol. 295, 116–124.
85. Levchenko A, Bruck J, Sternberg PW. 2000 Scaffold
proteins may biphasically affect the levels of
mitogen-activated protein kinase signaling and
reduce its threshold properties. Proc. Natl Acad.
Sci. USA 97, 5818–5823. (doi:10.1073/pnas.
86. Hawley SA, Selbert MA, Goldstein EG, Edelman AM,
Carling D, Hardie DG. 1995 50-AMP activates the
AMP-activated protein kinase cascade, and Ca2þ/
calmodulin activates the calmodulin-dependent
protein kinase I cascade, via three independent
mechanisms. J. Biol. Chem. 270, 27 186–27 191.
87. Moore F, Weekes J, Hardie DG. 1991 Evidence that
AMP triggers phosphorylation as well as direct
allosteric activation of rat liver AMP-activated protein
kinase: a sensitive mechanism to protect the cell
against ATP depletion. Eur. J. Biochem. 199,
88. Davies SP, Helps NR, Cohen PT, Hardie DG. 1995
50-AMP inhibits dephosphorylation, as well as
promoting phosphorylation, of the AMP-activated
protein kinase: studies using bacterially expressed
human protein phosphatase-2Ca and native bovine
protein phosphatase-2AC. FEBS Lett. 377, 421–425.
89. Carling D, Clarke PR, Zammit VA, Hardie DG. 1989
Purification and characterization of the AMP-
activated protein kinase: copurification of acetyl-CoA
carboxylase kinase and 3-hydroxy-3-methylglutaryl-
CoA reductase kinase activities. Eur. J. Biochem.
186, 129–136. (doi:10.1111/j.1432-1033.1989.
90. Hardie DG, Salt IP, Hawley SA, Davies SP. 1999
AMP-activated protein kinase: an ultrasensitive
system for monitoring cellular energy charge.
Biochem. J. 338, 717–722. (doi:10.1042/0264-
91. Kallio PJ, Wilson WJ, O’Brien S, Makino Y, Poellinger
L. 1999 Regulation of the hypoxia-inducible
transcription factor 1a by the ubiquitin-proteasome
pathway. J. Biol. Chem. 274, 6519–6525. (doi:10.
92. Huang LE, Gu J, Schau M, Bunn HF. 1998 Regulation
of hypoxia-inducible factor 1a is mediated by an
O2-dependent degradation domain via the
ubiquitin-proteasome pathway. Proc. Natl Acad.
Sci. USA 95, 7987–7992. (doi:10.1073/pnas.
93. Lando D, Peet DJ, Gorman JJ, Whelan DA, Whitelaw
ML, Bruick RK. 2002 FIH-1 is an asparaginyl
hydroxylase enzyme that regulates the
transcriptional activity of hypoxia-inducible factor.
Genes Dev. 16, 1466–1471. (doi:10.1101/
94. Lando D, Peet DJ, Whelan DA, Gorman JJ, Whitelaw
ML. 2002 Asparagine hydroxylation of the HIF
transactivation domain a hypoxic switch. Science
295, 858–861. (doi:10.1126/science.1068592)
95. Schmierer B, Novak B, Schofield CJ. 2010 Hypoxia-
dependent sequestration of an oxygen sensor by a
widespread structural motif can shape the hypoxic
response—a predictive kinetic model. BMC Syst.
Biol. 4, 139. (doi:10.1186/1752-0509-4-139)
96. Jiang BH, Semenza GL, Bauer C, Marti HH. 1996
Hypoxia-inducible factor 1 levels vary exponentially
over a physiologically relevant range of O2tension.
Am. J. Physiol. 271, C1172–C1180.
97. Koury ST, Koury MJ, Bondurant MC, Caro J, Graber
SE. 1989 Quantitation of erythropoietin-producing
cells in kidneys of mice by in situ hybridization:
correlation with hematocrit, renal erythropoietin
mRNA, and serum erythropoietin concentration.
Blood 74, 645–651.
98. Fandrey J, Bunn HF. 1993 In vivo and in vitro
regulation of erythropoietin mRNA: measurement
by competitive polymerase chain reaction. Blood 81,
99. Hong F, Sekhar KR, Freeman ML, Liebler DC. 2005
Specific patterns of electrophile adduction trigger
Keap1 ubiquitination and Nrf2 activation. J. Biol.
Chem. 280, 31 768–31 775. (doi:10.1074/jbc.
100. Kobayashi A, Kang MI, Watai Y, Tong KI, Shibata T,
Uchida K, Yamamoto M. 2006 Oxidative and
electrophilic stresses activate Nrf2 through Inhibition
of ubiquitination activity of Keap1. Mol. Cell.
Biol. 26, 221–229. (doi:10.1128/MCB.26.1.
101. Li W, Thakor N, Xu EY, Huang Y, Chen C, Yu R,
Holcik M, Kong AN. 2010 An internal ribosomal
entry site mediates redox-sensitive translation of
Nrf2. Nucleic Acids Res. 38, 778–788. (doi:10.1093/
102. Li W, Yu SW, Kong AN. 2006 Nrf2 possesses a
redox-sensitive nuclear exporting signal in the Neh5
transactivation domain. J. Biol. Chem. 281,
27251–27 263. (doi:10.1074/jbc.M602746200)
103. Sheridan JP et al. 1997 Control of TRAIL-induced
apoptosis by a family of signaling and decoy
receptors. Science 277, 818–821. (doi:10.1126/
104. Colotta F et al. 1993 Interleukin-1 type II receptor: a
decoy target for IL-1 that is regulated by IL-4.
Science 261, 472–475. (doi:10.1126/science.
105. Klein DE, Stayrook SE, Shi F, Narayan K, Lemmon
MA. 2008 Structural basis for EGFR ligand
sequestration by Argos. Nature 453, 1271–1275.
106. Klein DE, Nappi VM, Reeves GT, Shvartsman SY,
Lemmon MA. 2004 Argos inhibits epidermal growth
factor receptor signalling by ligand sequestration.
Nature 430, 1040–1044. (doi:10.1038/
107. Riechmann V, van Cruchten I, Sablitzky F. 1994 The
expression pattern of Id4, a novel dominant
negative helix-loop-helix protein, is distinct from
Id1, Id2 and Id3. Nucleic Acids Res. 22, 749–755.
108. Benezra R, Davis RL, Lockshon D, Turner DL, Weintraub
H. 1990 The protein Id: a negative regulator of helix-
loop-helix DNA binding proteins. Cell 61, 49–59.
Open Biol 3: 130031
109. Campuzano S. 2001 Emc, a negative HLH regulator
with multiple functions in Drosophila development.
Oncogene 20, 8299–8307. (doi:10.1038/sj.
110. Hughes KT, Mathee K. 1998 The anti-sigma factors.
Annu. Rev. Microbiol. 52, 231–286. (doi:10.1146/
111. Buchler NE, Cross FR. 2009 Protein sequestration
generates a flexible ultrasensitive response in a
genetic network. Mol. Syst. Biol. 5, 272. (doi:10.
112. Buchler NE, Louis M. 2008 Molecular titration and
ultrasensitivity in regulatory networks. J. Mol. Biol.
384, 1106–1119. (doi:10.1016/j.jmb.2008.09.079)
113. Liu X, Wu B, Szary J, Kofoed EM, Schaufele F. 2007
Functional sequestration of transcription factor
activity by repetitive DNA. J. Biol. Chem. 282,
20 868–20 876. (doi:10.1074/jbc.M702547200)
114. Lee TH, Maheshri N. 2012 A regulatory role for
repeated decoy transcription factor binding sites in
target gene expression. Mol. Syst. Biol. 8, 576.
in vitro validation of the sequestration mechanism for
generating ultrasensitive dose-response curves in
regulatory networks. PLoS Comput. Biol. 7, e1002171.
116. Kim SY, Ferrell Jr JE. 2007 Substrate competition as
a source of ultrasensitivity in the inactivation of
Wee1. Cell 128, 1133–1145. (doi:10.1016/j.cell.
117. Smith NR, Prehoda KE. 2011 Robust spindle
alignment in Drosophila neuroblasts by
ultrasensitive activation of pins. Mol. Cell. 43,
118. Cockman ME, Webb JD, Kramer HB, Kessler BM,
Ratcliffe PJ. 2009 Proteomics-based identification of
novel factor inhibiting hypoxia-inducible factor (FIH)
substrates indicates widespread asparaginyl
hydroxylation of ankyrin repeat domain-containing
proteins. Mol. Cell. Proteomics 8, 535–546. (doi:10.
119. LaPorte DC, Walsh K, Koshland Jr DE. 1984 The
branch point effect: ultrasensitivity and
subsensitivity to metabolic control. J. Biol. Chem.
259, 14 068–14 075.
120. De Vos D, Bruggeman FJ, Westerhoff HV, Bakker
BM. 2011 How molecular competition influences
fluxes in gene expression networks. PLoS ONE 6,
121. Crabtree B. 1976 Theoretical considerations of the
sensitivity conferred by substrate cycles in vivo.
Biochem. Soc. Trans. 4, 999–1002.
122. Fell DA, Sauro HM. 1985 Metabolic control and its
analysis. Additional relationships between
elasticities and control coefficients. Eur. J. Biochem.
148, 555–561. (doi:10.1111/j.1432-1033.1985.
123. Lenz DH, Mok KC, Lilley BN, Kulkarni RV, Wingreen
NS, Bassler BL. 2004 The small RNA chaperone Hfq
and multiple small RNAs control quorum sensing in
Vibrio harveyi and Vibrio cholerae. Cell 118, 69–82.
124. Legewie S, Dienst D, Wilde A, Herzel H, Axmann IM.
2008 Small RNAs establish delays and temporal
thresholds in gene expression. Biophys. J. 95,
125. Mehta P, Goyal S, Wingreen NS. 2008 A quantitative
comparison of sRNA-based and protein-based gene
regulation. Mol. Syst. Biol. 4, 221. (doi:10.1038/msb.
126. Levine E, Zhang Z, Kuhlman T, Hwa T. 2007
Quantitative characteristics of gene regulation by
small RNA. PLoS Biol. 5, e229. (doi:10.1371/journal.
127. Mukherji S, Ebert MS, Zheng GX, Tsang JS, Sharp
PA, van Oudenaarden A. 2011 MicroRNAs can
generate thresholds in target gene expression.
Nat. Genet. 43, 854–859. (doi:10.1038/ng.905)
128. McCarrey JR, Riggs AD. 1986 Determinator–
inhibitor pairs as a mechanism for threshold setting
in development: a possible function for
pseudogenes. Proc. Natl Acad. Sci. USA 83,
129. Uy R, Wold F. 1977 Posttranslational covalent
modification of proteins. Science 198, 890–896.
130. Goldbeter A, Koshland Jr DE. 1984 Ultrasensitivity in
biochemical systems controlled by covalent
modification: interplay between zero-order
and multistep effects. J. Biol. Chem. 259,
131. LaPorte DC, Koshland Jr DE. 1983 Phosphorylation
of isocitrate dehydrogenase as a demonstration of
enhanced sensitivity in covalent regulation. Nature
305, 286–290. (doi:10.1038/305286a0)
132. Meinke MH, Bishop JS, Edstrom RD. 1986 Zero-order
ultrasensitivity in the regulation of glycogen
phosphorylase. Proc. Natl Acad. Sci. USA 83,
133. Meinke MH, Edstrom RD. 1991 Muscle
glycogenolysis: regulation of the cyclic
interconversion of phosphorylase a and
phosphorylase b. J. Biol. Chem. 266, 2259–2266.
134. Melen GJ, Levy S, Barkai N, Shilo BZ. 2005
Threshold responses to morphogen gradients by
zero-order ultrasensitivity. Mol. Syst. Biol. 1, 0028.
135. Cimino A, Hervagault JF. 1987 Experimental evidence
for a zero-order ultrasensitivity in a simple substrate
cycle. Biochem. Biophys. Res. Commun. 149,
136. Chickarmane V, Troein C, Nuber UA, Sauro HM,
Peterson C. 2006 Transcriptional dynamics of the
embryonic stem cell switch. PLoS Comput. Biol. 2,
137. Boyer LA et al. 2005 Core transcriptional regulatory
circuitry in human embryonic stem cells. Cell 122,
138. Odom DT et al. 2006 Core transcriptional regulatory
circuitry in human hepatocytes. Mol. Syst. Biol. 2,
139. Zingg JM, Pedraza-Alva G, Jost JP. 1994 MyoD1
promoter autoregulation is mediated by two
proximal E-boxes. Nucleic Acids Res. 22,
140. Okuno Y et al. 2005 Potential autoregulation of
transcription factor PU.1 by an upstream regulatory
element. Mol. Cell. Biol. 25, 2832–2845. (doi:10.
141. Nishimura S, Takahashi S, Kuroha T, Suwabe N,
Nagasawa T, Trainor C, Yamamoto M. 2000 A GATA
box in the GATA-1 gene hematopoietic enhancer is
a critical element in the network of GATA factors
and sites that regulate this gene. Mol. Cell. Biol. 20,
142. Tiwari A, Balazsi G, Gennaro ML, Igoshin OA. 2010
The interplay of multiple feedback loops with post-
translational kinetics results in bistability of
mycobacterial stress response. Phys. Biol. 7, 036005.
143. Palani S, Sarkar CA. 2008 Positive receptor feedback
during lineage commitment can generate
ultrasensitivity to ligand and confer robustness to a
bistable switch. Biophys. J. 95, 1575–1589. (doi:10.
144. Bhattacharya S, Conolly RB, Kaminski NE, Thomas
RS, Andersen ME, Zhang Q. 2010 A bistable switch
underlying B-cell differentiation and its disruption
by the environmental contaminant 2,3,7,8-
tetrachlorodibenzo-p-dioxin. Toxicol. Sci. 115,
145. Siegal-Gaskins D, Mejia-Guerra MK, Smith GD,
Grotewold E. 2011 Emergence of switch-like
behavior in a large family of simple biochemical
networks. PLoS Comput. Biol. 7, e1002039. (doi:10.
146. Sneppen K, Micheelsen MA, Dodd IB. 2008
Ultrasensitive gene regulation by positive feedback
loops in nucleosome modification. Mol. Syst. Biol. 4,
147. Dodd IB, Micheelsen MA, Sneppen K, Thon G. 2007
Theoretical analysis of epigenetic cell memory by
nucleosome modification. Cell 129, 813–822.
148. Malleshaiah MK, Shahrezaei V, Swain PS, Michnick
SW. 2010 The scaffold protein Ste5 directly controls
a switch-like mating decision in yeast. Nature 465,
149. Bradshaw JM, Kubota Y, Meyer T, Schulman H. 2003
An ultrasensitive Ca2þ/calmodulin-dependent
protein kinase II-protein phosphatase 1
switch facilitates specificity in postsynaptic
calcium signaling. Proc. Natl Acad. Sci. USA
100, 10512–10517. (doi:10.1073/pnas.
150. Semenza GL. 2004 Hydroxylation of HIF-1: oxygen
sensing at the molecular level. Physiology
(Bethesda) 19, 176–182. (doi:10.1152/physiol.
151. Buchler NE, Gerland U, Hwa T. 2005 Nonlinear
protein degradation and the function of genetic
circuits. Proc. Natl Acad. Sci. USA 102, 9559–9564.
152. Berg OG, Paulsson J, Ehrenberg M. 2000
Fluctuations and quality of control in biological
cells: zero-order ultrasensitivity reinvestigated.
Biophys. J. 79, 1228–1236. (doi:10.1016/
Open Biol 3: 130031
153. Bluthgen N, Bruggeman FJ, Legewie S, Herzel H, Download full-text
Westerhoff HV, Kholodenko BN. 2006 Effects of
sequestration on signal transduction cascades.
FEBS J. 273, 895–906. (doi:10.1111/j.1742-4658.
154. Pedersen MG, Bersani AM. 2010 Introducing total
substrates simplifies theoretical analysis at non-
negligible enzyme concentrations: pseudo first-order
kinetics and the loss of zero-order ultrasensitivity.
J. Math. Biol. 60, 267–283. (doi:10.1007/s00285-
155. Ventura AC, Jiang P, Van Wassenhove L, Del Vecchio
D, Merajver SD, Ninfa AJ. 2010 Signaling properties
of a covalent modification cycle are altered by a
downstream target. Proc. Natl Acad. Sci. USA 107,
10 032–10 037. (doi:10.1073/pnas.0913815107)
156. van Albada SB, ten Wolde PR. 2007 Enzyme
localization can drastically affect signal amplification
in signal transduction pathways. PLoS Comput. Biol.
3, 1925–1934. (doi:10.1371/journal.pcbi.0030195)
157. Xu Y, Gunawardena J. 2012 Realistic enzymology for
post-translational modification: zero-order
ultrasensitivity revisited. J. Theor. Biol. 311,
158. Kitano H. 2004 Biological robustness. Nat. Rev.
Genet. 5, 826–837. (doi:10.1038/nrg1471)
159. Shah NA, Sarkar CA. 2011 Robust network
topologies for generating switch-like cellular
responses. PLoS Comput. Biol. 7, e1002085.
160. Bagowski CP, Besser J, Frey CR, Ferrell Jr JE. 2003
The JNK cascade as a biochemical switch in
mammalian cells: ultrasensitive and all-or-none
responses. Curr. Biol. 13, 315–320. (doi:10.1016/
161. O’Shaughnessy EC, Palani S, Collins JJ, Sarkar CA.
2011 Tunable signal processing in synthetic MAP
kinase cascades. Cell 144, 119–131. (doi:10.1016/j.
162. Ferrell JE, Xiong W. 2001 Bistability in cell signaling:
how to make continuous processes discontinuous,
and reversible processes irreversible. Chaos 11,
163. Angeli D, Ferrell Jr JE, Sontag ED. 2004 Detection of
multistability, bifurcations, and hysteresis in a large
class of biological positive-feedback systems. Proc.
Natl Acad. Sci. USA 101, 1822–1827. (doi:10.1073/
164. Sauro HM. 2009 Network dynamics. Methods
Mol. Biol. 541, 269. (doi:10.1007/978-1-59745-
165. Iglesias PA, Ingalls BP. 2009 Control theory and
systems biology. Cambridge, MA: The MIT Press.
166. Simmons SO, Fan CY, Ramabhadran R. 2009 Cellular
stress response pathway system as a sentinel
ensemble in toxicological screening. Toxicol. Sci.
111, 202–225. (doi:10.1093/toxsci/kfp140)
167. Kacser H, Burns JA. 1973 The control of flux.
Symp. Soc. Exp. Biol. 27, 65–104.
168. Savageau MA. 1974 Optimal design of feedback
control by inhibition: steady state considerations.
J. Mol. Evol. 4, 139–156. (doi:10.1007/
169. Cirit M, Wang CC, Haugh JM. 2010 Systematic
quantification of negative feedback mechanisms in
the extracellular signal-regulated kinase (ERK)
signaling network. J. Biol. Chem. 285, 36 736–
36 744. (doi:10.1074/jbc.M110.148759)
170. Sturm OE et al. 2010 The mammalian MAPK/ERK
pathway exhibits properties of a negative feedback
amplifier. Sci. Signal. 3, ra90. (doi:10.1126/
171. Nevozhay D, Adams RM, Murphy KF, Josic K, Balazsi
G. 2009 Negative autoregulation linearizes the
dose-response and suppresses the heterogeneity
of gene expression. Proc. Natl Acad. Sci.
USA 106, 5123–5128. (doi:10.1073/pnas.
172. Woods NM, Cuthbertson KS, Cobbold PH. 1986
Repetitive transient rises in cytoplasmic free calcium
in hormone-stimulated hepatocytes. Nature 319,
173. Tay S, Hughey JJ, Lee TK, Lipniacki T, Quake SR,
Covert MW. 2010 Single-cell NF-kappaB dynamics
reveal digital activation and analogue information
processing. Nature 466, 267–271. (doi:10.1038/
174. Geva-Zatorsky N et al. 2006 Oscillations and
variability in the p53 system. Mol. Syst. Biol. 2,
175. Amir A, Meshner S, Beatus T, Stavans J. 2010
Damped oscillations in the adaptive response of the
iron homeostasis network of E. coli. Mol. Microbiol.
76, 428–436. (doi:10.1111/j.1365-2958.2010.
176. Yang JH, Yang L, Qu Z, Weiss JN. 2008 Glycolytic
oscillations in isolated rabbit ventricular myocytes.
J. Biol. Chem. 283, 36 321–36327. (doi:10.1074/
177. Ferrell Jr JE, Tsai TY, Yang Q. 2011 Modeling the cell
cycle: why do certain circuits oscillate? Cell 144,
178. Novak B, Tyson JJ. 2008 Design principles of
biochemical oscillators. Nat. Rev. Mol. Cell Biol. 9,
179. Nguyen LK, Kulasiri D. 2009 On the functional
diversity of dynamical behaviour in genetic and
metabolic feedback systems. BMC Syst. Biol. 3, 51.
180. Savageau MA. 1975 Optimal design of feedback
control by inhibition: dynamic considerations.
J. Mol. Evol. 5, 199–222. (doi:10.1007/
181. Kholodenko BN. 2000 Negative feedback and
ultrasensitivity can bring about oscillations in the
mitogen-activated protein kinase cascades.
Eur. J. Biochem. 267, 1583–1588. (doi:10.1046/j.
182. Shankaran H, Ippolito DL, Chrisler WB, Resat H,
Bollinger N, Opresko LK, Wiley HS. 2009 Rapid and
sustained nuclear-cytoplasmic ERK oscillations
induced by epidermal growth factor. Mol. Syst. Biol.
5, 332. (doi:10.1038/msb.2009.90)
Open Biol 3: 130031