BIOINFORMATICS
ORIGINAL PAPER
Vol. 25 no. 15 2009, pages 1898–1904
doi:10.1093/bioinformatics/btp316
Systems biology
Hub genes with positive feedbacks function as master switches
in developmental gene regulatory networks
Chang H. Seo†,‡ , Jeong-Rae Kim† , Man-Sun Kim† and Kwang-Hyun Cho∗
Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon,
305-701, Republic of Korea
Received on December 11, 2008; revised and accepted on May 11, 2009
Advance Access publication May 13, 2009
Associate Editor: Limsoon Wong
ABSTRACT
Motivation: Spatio-temporal regulation of gene expression is an
indispensable characteristic in the development processes of all
animals. ‘Master switches’, a central set of regulatory genes
whose states (on/off or activated/deactivated) determine specific
developmental fate or cell-fate specification, play a pivotal role
for whole developmental processes. In this study on genome-wide
integrative network analysis the underlying design principles of
developmental gene regulatory networks are examined.
Results: We have found an intriguing design principle of
developmental networks: hub nodes, genes with high connectivity,
equipped with positive feedback loops are prone to function as
master switches. This raises the important question of why the
positive feedback loops are frequently found in these contexts. The
master switches with positive feedback make the developmental
signals more decisive and robust such that the overall developmental
processes become more stable. This finding provides a new
evolutionary insight: developmental networks might have been
gradually evolved such that the master switches generate digital-like
bistable signals by adopting neighboring positive feedback loops. We
therefore propose that the combined presence of positive feedback
loops and hub genes in regulatory networks can be used to predict
plausible master switches.
Contact: [email protected]
Supplementary information: Supplementary data are available at
Bioinformatics online.
1
INTRODUCTION
A key challenge in developmental biology is to unravel how
numerous genes are temporally activated within networks in
order to accomplish specific functions and regulate developmental
fate diversity (Boone et al., 2007). Developmental processes are
regulated via complex intracellular signaling networks (Arbeitman
et al., 2002; Freeman, 2000; Wolpert, 2007) whose constituent
genes have recently been identified using genome-sequencing and
other high-throughput measurements. In order to extract useful
∗ To
whom correspondence should be addressed.
† The authors wish it to be known that, in their opinion, the first three authors
should be regarded as joint First Authors.
‡ Present address: Department of Bioengineering, School of Engineering,
The University of Tokyo, Tokyo 113-8685, Japan.
1898
information from the large data sets that are generated, an overall
network map can be created by organizing the various inputs and
outputs between genes within a network. Subsequent analysis of the
network map provides testable predictions and new insight into the
complex developmental program (Oliveri et al., 2008; Stathopoulos
and Levine, 2005). These networks can be constructed from various
types of genome-wide data (such as expression, transcriptionfactor binding site and protein–protein interaction data) in order to
characterize particular developmental traits. Their analysis is one of
the key research areas in the post-genomic era (Yeger-Lotem et al.,
2004).
In order to further our understanding of Drosophila melanogaster
developmental processes, we have integrated gene regulatory
network and genetic interaction data. The rationale behind this
integration is that developmental processes are not regulated by a
single interaction independently, but by combinations of interactions
between genes encoding transcription and signaling factors (Oliveri
et al., 2008). While interpreting each interaction separately is likely
to prevent a comprehensive understanding of the regulation network,
integration of commonly accessible data types, such as protein
interaction networks, gene expression profiles and gene orthologues,
has previously provided subnetwork usages in various biological
processes (Boone et al., 2007; Han et al., 2004; Luscombe et al.,
2004).
One efficient way to analyze complex network structures is
to divide them into subnetwork components, frequently called
‘network motifs’. Each motif operates as an independent module
and has its own particular information-processing function. Analysis
of the motifs can often help develop insight regarding the principles
governing the architecture of complex developmental networks.
They have been observed in many regulatory networks (Alon, 2007;
Milo et al., 2002; Milo et al., 2004). Previous studies in numerous
biological systems have shown positive feedback loops to be key
regulatory modules that enhance a signaling system to have decisive,
all-or-none ‘digital’ output signal characteristics (Bagowski et al.,
2003; Brandman et al., 2005; Kim et al., 2008; Kwon and Cho,
2008; Shin et al., 2009). Moreover, they enhance robustness by
magnifying the effect of stimuli and producing bistability (i.e. the
ability of a system to exist in one of two stable steady states) so
that the activation level of a downstream pathway can be clearly
distinguished from non-stimulated states, and so that these states
can be maintained (Kim et al., 2008; Kitano, 2004). They are found
frequently in many biological regulatory processes such as the cell
© The Author 2009. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1898
1898–1904
Hub genes with positive feedbacks function as master switches
cycle, calcium signaling, mitotic trigger and eukaryotic chemotaxis
(Alon et al., 1999; Kim et al., 2007).
By searching complex developmental networks for motifs, we
detected characteristic network building blocks (structural and
functional motifs) and, in particular, uncovered that small structures
(subnetworks) equipped with positive feedback loops are the most
frequent and significant subnetworks in developmental networks.
We have compared gene ontology (GO) and function of highly
connected genes, both with and without positive feedback, and
have found a conclusive design pattern: hub nodes, genes with
high connectivity, equipped with positive feedback loops have
a high tendency to become master switches. These genes are
presumed to be central elements whose activation affects numerous
critical downstream genes that regulate specific developmental
processes. We note that this observation is consistent with a previous
report by Rothenberg (Rothenberg, 2007) which described how
developmental programs can be understood in terms of the impacts
of just a few master switches.
Why are positive feedback loops frequently found in contexts
where the genes are master switches? In development, evolution
may select for networks that have structurally stable forms, as these
configurations make the developmental process robust to intrinsic
and extrinsic variations. Master switches with positive feedback
make developmental signals more decisive and robust, so that the
overall developmental processes become more stable. This finding
provides the new evolutionary insight that developmental networks
have gradually evolved, such that the master switches generate
digital-like (on or off) signals by adopting positive feedback loops.
We also propose this approach as an applicable method to find key
elements of a biological network by providing the highly plausible
candidate genes as master switches.
2
2.1
METHODS
Data source and preparation
We utilized the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/
geo/) database to get microarray experimental data with the accession
number GSE6186. We used this data set as it contains the expression of
almost all genes (∼14 000) from the D. melanogaster 14k whole-genome
embryogenesis time-course.
interaction types (or options) to ‘expression’, ‘regulation’ and ‘genetic
interaction’. For instance, if there is confirmative evidence that two DEGs
interact with each other, an interaction edge is assigned between the two
genes (Chen et al., 2008). The considered integrated interactions have
been classified into two groups: (i) genetic interactions—using the genetic
interactions we identified hub genes [since hubs in the genetic interaction
network are considered to play important roles in development as they
are highly essential (Tong et al., 2004)]; and (ii) regulatory interactions—
using the regulatory interactions, we identified significant sub-modules and
explored their functional roles. From these processes, we obtained two kinds
of networks: the genetic interaction network and the regulatory network (see
Supplementary Fig. S1–S2).
2.4
2.5
2.2
Differentially expressed genes (DEGs)
The data set was normalized as follows (Hooper et al., 2007): the spot
intensities of the two channels were normalized by applying the Qspline
method with a log-normal distribution. From the normalized values of the
data set, we selected differentially expressed genes (DEGs) to obtain a group
of genes, which were significantly expressed at each time point. Selection
was determined using a 2-fold cutoff criterion, which is frequently used
in single-slide microarray experiments [we refer the interested reader to
(Rajagopalan, 2003; Yang et al., 2002) for further details]. In summary,
DEGs with expression levels greater than one (or smaller than − 1) were
regarded as significantly expressed.
Construction of interaction maps
Using Pathway Studio (Nikitin et al., 2003), a piece of software that
allows users to build and analyze molecular networks, network maps of
the differentially expressed genes were constructed at each time point.
Pathway Studio contains the Drosophila interaction database (Drosophila v.3)
from which we integrated the interactions between DEGs by limiting the
Gene Ontology (GO) classification and analysis
GO annotation provides an indication of the trait of a group of genes and is a
useful tool for both small- and large-scale analyses of genes. GO functional
annotations in this study were obtained from the GO database (Ashburner
et al., 2000). We evaluated the statistical significance of the overlap between
selected genes (co-expression or over-representation) through GO terms
enrichment using AmiGo (Carbon et al., 2009). FlyBase, a D. melanogaster
gene database (Tweedie et al., 2009), was used as a database filter in
the GO analysis. Test scores were assessed based on a hypergeometric
distribution and Bonferroni correction was employed for multiple tests in
the randomization procedure (Dennis et al., 2003; Huang et al., 2007).
3
2.3
Identification of network motifs
Large biological networks can be decomposed into smaller/simpler
subnetworks or network motifs, which are small architectures that are more
frequently observed in a given network than in randomized networks (Milo
et al., 2002). Once identified, the (networks of) motifs can be analyzed
in order to reveal structural or topological design principles of complex
networks (Alon, 2007; Milo et al., 2004; Milo et al., 2002). In order
to identify network motifs in the complex developmental network map
constructed in Section 2.3, we first examined small-scale interaction patterns
by investigating statistically significant network modules at each time point.
Motif analysis and visualization was performed using MAVisto (Schreiber
and Schwobbermeyer, 2005). Within the software, the statistical significance
of each sub-module was evaluated by calculating the probability (P-value)
that the frequency of the sub-module in distributions obtained from numerous
randomized networks is equal to or larger than the frequency of the submodule in the given network (Milo et al., 2002). Network motifs were
deemed to be statistically significant if their P-value was less than the
cut-off threshold 0.05. Note that this threshold is frequently used in motif
analysis (Taylor et al., 2007; Ward and Thornton, 2007; Zhang et al., 2005).
The randomized networks were generated through a random local rewiring
algorithm, which preserves the vertex degree (Maslov and Sneppen, 2002).
For instance, in the rewiring step, two edges (v1 , v2 ) and (v3 , v4 ) were rewired
such that v1 is connected to v4 and v3 to v2 , if the connection does not already
exist in the network. This rewiring step was repeated many times until a
properly randomized network was obtained. In this study, we considered
motifs of size three or four (which corresponds to the number of subnetwork
nodes) and randomizations of 1000 times in order to get reliable statistics.
For summary of the analysis results, we used U. Alon’s motif dictionary
(Milo et al., 2002).
RESULTS
3.1 Temporal developmental network maps and their
statistically significant subnetworks
In order to identify the developmental network properties of
D. melanogaster, we examined gene regulatory and genetic data
from the (Hooper et al., 2007) data set, which spans all the
1899
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1899
1898–1904
C.H.Seo et al.
Table 1. A summary of hub genes (the number in the parenthesis next to
each gene name denotes the connectivity of the gene)
Fig. 1. Statistically significant network motifs and their corresponding real
subnetworks in the developmental network maps. Predominant network
motifs of: (A) three nodes (ID14, ID74) and (B) four nodes (ID30, ID4370,
ID282). (C) An illustration of the four-node motif (ID282) as a combination
of two three-node motifs. (D) An illustration of the four-node motif in the
context of multiple interactions in the entire network. Grey circles denote
genes or proteins, blue squares denote master switches and red triangles
indicate co-regulators. Also, black arrows indicate directed interactions
(regulation of expression) and dashed lines denote genetic interactions.
stages of embryogenesis [as identified by Arbeitman et al. (2002)].
After identifying the differentially expressed genes (DEGs), we
constructed a developmental network map at each time point in the
data set and identified network motifs. The results are summarized
in Supplementary Table S1. We found that the two most significant
motifs are ID14 and ID74. Interestingly, both motifs contain a
positive feedback loop (Fig. 1A). In the case of four-node feedback
loops (see Supplementary Table S2), three significant motifs were
found: ID30, ID4370 and ID282 (Fig. 1B). The structures of these
motifs are similar in form to those of the three-node motifs, as each
four-node motif can be considered to be a combination of three-node
motifs (Fig. 1C). By mapping the motifs of three or four nodes to our
real developmental network at each time point, we have discovered
that three subnetworks (Fig. 1D) are statistically significant, and we
will now analyze their properties in further detail.
3.2
Hub nodes with positive feedback loops have a high
inclination to become a ‘master switch’ in
developmental networks
All the network motifs identified in the previous section contain
two-node positive feedback loops. Based on this observation,
we divided the hub genes in the genetic interaction network
into two groups according to the existence of two-node positive
feedback loops (Table 1), and then investigated the hallmark of
hub genes in groups. Hub genes were defined as the genes with
connectivity (degree) greater than 10 in the genetic interaction
network and, incidentally, are the top 10% genes of highest
connectivity (see Supplementary Fig. S3 and Supplementary Table
S3 for the distribution of connectivities of genes in the genetic
Time (h)
Hub genes with positive
feedback
Hub genes without positive
feedback
1–2
egfr(39), trx(21)
1.5–2.5
egfr(39), cyce(27), e2f (12)
2–3
egfr(31)
2.5–3.5
3–4
arm(16)
arm(17)
4–5
arm(18)
5–6
egfr(26), arm(23)
6–7
egfr(19)
10–11
arm(23), e2f (12), egfr(29)
11–12
arm(23), e2f (12), egfr(29)
17–18
19–20
23–24
egfr(12)
ras85d(57), w(39), h(31),
sev(31), hh(23),
DppIII(22), dpp(19),
rho(17), phl(17), brm(15),
e(z)(15), s(15), dsor1(14),
kr(14), ash1(14), pnt(13),
argos(13), Fz(12),
mam(12), mxc(12),
ato(12), Sty(12), ph-p(12),
vn(12), th(11), dsh(11),
puc(11)
ras85d(54), w(33), sev(32),
h(30), dl(23), hh(23),
phl(21), DppIII(20),
rho(17), argos(16),
dpp(16), rac1(15),
brm(14), dsh(14),
dsor1(14), Jra(14),
pnt(13), tkv(13), ubx(13),
dfd(12), s(12), vn(12),
sty(11), csw(11), ato(11),
mam(11), ash1(11),
ras85d(42), sev(26), hh(14),
rho(14), s(12), argos(11),
ato(11)
w(17), hh(11),
ras85d(34), w(18), sev(16),
rho(11)
w(16), sev(14), cyce(14),
dl(13)
ras85d(42), w(26), pnr(26),
sev(24), dl(16), cyce(16),
hh(15), dsh(14), brm(14),
fz(14), Jra(12), ato(11),
sgg(11), pnt(11), rho(11)
ras85d(26), egfr(19),
arm(15)
ras85d(42), egfr(29),
pnr(28), h(26), dl(20),
rho(16), Jra(15), fz(12),
dsh(11), argos(11)
ras85d(42), egfr(29),
pnr(28), h(26), dl(20),
rho(16), Jra(15), fz(12),
dsh(11), argos(11)
ras85d(19), pnr(13)
ras85d(21), pnr(15)
DppIII(16), pnr(16), dpp(14)
interaction network). Though we focus only on hubs and positive
feedback loops, negative feedback can play important roles in
development as negative feedback can stabilize responses (Kim
et al., 2008). However, compared to the number of positive feedback
motifs, there exist only a small number of negative feedback loop
motifs in the developmental network. The GO results on negative
feedback loops are summarized in Supplementary Table S4 where
1900
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1900
1898–1904
Hub genes with positive feedbacks function as master switches
we demonstrate that they are involved in ‘imaginal disc pattern
formation’.
As seen in Table 1, arm, egfr, e2f, cyce and trx are hub
genes with at least one positive feedback loop. Among these five
genes, egfr, e2f and arm are broadly expressed throughout the
developmental stages. Intriguingly, these three genes have been
shown to have the characteristics of master switches in previous
reports: (i) Drosophila epidermal growth factor (EGF) receptor
(DER or EGFR): EGFR/DER is a molecule of the EGFR/ErbB
family in the fly genome. egfr has been considered as one of the
key regulatory genes in development since it is repeatedly involved
in several developmental processes such as segmentation (Dubois
et al., 2001; Payre et al., 1999), germ band retraction (Raz and
Shilo, 1992), and viability of midline cells (Bergmann et al., 2002;
Klambt et al., 1991; Stemerdink and Jacobs, 1997). Tight and exact
regulation of duration, strength and range of activation is essential to
assure the correct developmental program sequences (Shilo, 2003).
(ii) e2f regulates cell cycle progression from G1 to S and its overexpression can force cells to transit from G2 to M by activating
the expression of stg, the homolog of yeast cdc25. Hence, E2f has
been regarded as a key regulator of cell-cycle progression (Asano
et al., 1996; Neufeld et al., 1998). In the Drosophila wing disc,
E2f is a critical player for both cell proliferation and cell survival
in the presumptive wing margin area (Delanoue et al., 2004).
(iii) Armadillo (Arm) is the homologue of β-catenin and is a key
mediator in the signaling pathway initiated by Wnt/Wingless (Wg)
proteins, regulating numerous embryonic developmental processes
such as segment polarity in drosophila and axis specification in
Xenopus (Boonchai et al., 2000; Miller and Moon, 1996; Siegfried
and Perrimon, 1994; Willert and Nusse, 1998). Furthermore, the
Wnt/β-catenin pathway is highly involved in tissue regeneration
and homeostasis. Activation of target genes through the Wnt/Wg
signal depends on the activator β-catenin/Arm which functions in
both cell adhesion and transcription. Correct switching between
these two functions is critical to continue normal development,
and any wrong decision can result in biological disorders such as
cancer. The challenge ahead is to identify how β-catenin interactions
are integrated and regulated during cell differentiation and tissue
formation (Harris and Peifer, 2005; Staal and Clevers, 2005).
In order to further analyze the relationship between hub genes
and positive feedback loops, we performed a chi-square test (cutoff
P-value 0.01). As seen in Table 2, 43% of the genes with positive
feedback are hub genes but only 8% of the genes without positive
feedback are hub genes. A chi-square test (P-value = 1.578e-09)
reveals a statistically significant difference between the number of
hub genes with and without positive feedback loop. Hence we can
conclude that genes with a positive feedback loop are more likely to
be hub genes than those without a positive feedback loop. Moreover,
Table 2. Summary of the numbers of genes with respect to hub and positive
feedback [the number in the parenthesis denotes the number of genes related
to developmental process (GO: 0030707, 0007424, 0035295)]
Genes with positive feedback
Genes without positive
feedback
Hub genes
Non-hub genes
12 (9)
46 (21)
16 (5)
519 (94)
75% (9 out of 12) of the hub genes with a positive feedback loop
are related to developmental processes (see Supplementary Table
S5–S8 for the detailed GO analysis results). Hence we suggest that
hub genes with positive feedback loop are likely to be involved in
the regulation of developmental process.
3.3
Computational test of a developmental switch
equipped with a positive feedback loop
The frequent presence of positive feedback loops with master
switches in developmental networks raises an important question
regarding their biological function. It has been considered that
developmental signaling networks have been evolved to obtain
highly efficient and stable architectures, whose components enable
the developmental process to be performed accurately (Becskei
et al., 2001; Cheng et al., 2008; Mitrophanov and Groisman,
2008). Therefore, the condition of developmental signals at master
switches must be robust and decisive enough to perform normal
developmental programs sequentially.
In order to test this hypothesis, we constructed two mathematical
models for developmental signaling switches with a positive
feedback, as shown in Figure 2A, and without a positive feedback,
as shown in Figure 2B (see Supplementary Mathematical Models for
details on the mathematical models). For both cases, we considered
input signals both with and without noise. The intensity of the noise
was relatively high compared to that of the developmental signal.
These models were simulated numerically and the resulting output
is shown in Figure 2C and D. We considered two types of stimuli as
an input developmental signal: a noise-free (Fig. 2C, bottom black
line) and a noisy stimulus (Fig. 2D, bottom black line). We assumed
that the normal developmental input signal is a step function with
magnitude 0.5 which is initiated at t = 0. The noise consists of
saw-toothed signals with a variation of amplitude ±0.1 and ±0.3
from the input signal. When a noise-free input signal was applied
to both hub genes (nodes with and without a positive feedback
loop), both turned on at approximately the same time. However,
the magnitude of the responses differs significantly. The response at
the hub with a positive feedback loop (Fig. 2C, top blue dotted line)
is approximately two times larger than that without (Fig. 2C, middle
red dashed line), although both responses demonstrate switch-like
behaviors. Through these simulations, we have demonstrated that a
positive feedback loop can amplify a weak input signal resulting in
a discrete output.
When we applied a noisy input signal, a critical difference
between both responses occurred. The response at a hub gene
without a positive feedback is very sensitive to noise, and the noise
induced fluctuations are high (Fig. 2D, middle red dashed line). A
similar result is obtained for stochastic simulations (Fig. 2E, see
Supplementary Stochastic Models for details). Therefore, if internal
(stochastic) or external noises are applied to this hub gene type,
it is not likely to respond in a reliable manner [see (Alon, 2007;
Brandman et al., 2005; Hallinan, 2005; Kwon and Cho, 2008; Wolf
and Arkin, 2003) for the dynamical properties of network motifs].
In contrast, the response at the gene with a positive feedback was
very robust to both internal and external noise (see the blue dotted
line in Fig. 2D for an external noise and the blue line in Fig. 2E
for an internal noise). Although a small fluctuation was generated
by the input noise, its degree was ignorable when compared with
the corresponding case without a feedback loop. Since there are
1901
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1901
1898–1904
C.H.Seo et al.
broad range of parameter values) than those without. We conclude
that the hub genes equipped with positive feedback loops can
perform their own switch-like functions with high robustness (see
Supplementary Fig. S5 for biological examples of fast switching).
This computational study assists our understanding of the robust
regulation of developmental processes by suggesting that the hub
genes in developmental networks have evolved to be equipped
with positive feedback loops such that resulting signals are reliable
developmental switches.
4
Fig. 2. Comparison of different responses of the switch equipped
with/without a positive feedback loop. A schematic diagram of the
developmental switch at a hub node equipped (A) with a positive feedback
loop or (B) without a positive feedback loop. Dashed lines (or arrows and
circles) indicate ‘not considered’ (e.g., G1, G2, G3 and their interactions
with the hub in B) or ‘do not exist’ (e.g. Y and its interactions with the hub
in B). (C–D) Comparison of responses (y-axis) of different types of hub
genes shown in (A) and (B) as a function of time (x-axis)—the response at a
hub with (blue line)/without (red line) a positive feedback loop for an input
developmental signal (black line) without (C)/with (D) noise. (E) Stochastic
simulations for a hub with (blue)/without (red) a positive feedback loop.
See Supplementary Material for detailed model descriptions and parameter
values.
numerous noise sources in many biological or cellular systems, this
positive feedback can help a key node, such as a master switch, to
perform their roles decisively, regardless of noise so that the master
node may activate downstream nodes accurately.
In order to ensure that the effects of positive feedback are generic,
rather than specific to the chosen parameters, we examined the
responses of the model for an extensive range of parameter values
and found similar results (see Supplementary Fig. S4). We note
that a hub node with a positive feedback in our model can be
involved in many different biological processes, such as cell cycle
regulation, signal transduction and cell differentiation, and therefore
that the parameter values can differ depending on contexts. The
extensive computational experiments performed over a wide range
of parameter values strongly support our conclusion.
In our simulations, the responses at the hub node with a positive
feedback loop were more stable and robust against noise (over a
DISCUSSION
Embryogenesis, the first developmental stage, demands the highly
delicate orchestration of complex events that are precisely tuned by
an elaborate set of signaling networks. These complex networks
not only determine the eventual developmental fates of specific
organs but also generate patterning of the embryo body plan. During
embryonic processes, cells and tissues undergo coordinated change
in gene expression in order to reach a highly regulated steady state.
What are the underlying mechanisms or principles by which the
developmental sequences are maintained robustly to accomplish the
development of embryo? To attempt to answer this question, we have
analyzed embryonic developmental networks that were constructed
using a genome-wide microarray data set, and found interesting
developmental network traits over the course of embryogenesis. We
note that these traits would be difficult to observe using individual
experiments.
A total of nine highly connected genes (Egfr, E2f, Arm, Cyca,
Cyce, Trx, Cdc42, Sc, Tkv) bound by positive feedback loops were
found. The first and second most significant classifications were
‘regulation of developmental process (GO: 0050793)’ and ‘ovarian
follicle cell development (GO: 0030707)’, respectively. The result
that hub genes with positive feedback loops are predominately
abundant in the category of ‘regulation of developmental processes’
is biologically significant upon consideration of the following: genes
involved in this category serve central roles in the regulation of
various developmental processes by modulating the frequency, rate
or extent of development and its specific condition, which results
in the developmental progression of various organisms over time.
Several developmental processes are controlled and regulated in an
appropriate manner by these genes. This implies that such genes
function as important developmental switches to determine correct
and adequate developmental fates.
For an accurate regulation of developmental processes, it is
necessary that the responsible genes (mainly called ‘switches’),
which serve as developmental switches, function correctly. The input
developmental signals at these master switches should be decisive
enough to correctly discern the input signaling conditions (on/off)
so that the input signal may induce a new developmental process
(Fig. 3). However, almost all intracellular signals are likely to have
both intrinsic and extrinsic noise, which can generate biological
disorders, such as errors in DNA replication leading to mutation
(Rao et al., 2002). To recognize the input developmental signals in
the presence of noise or perturbations, signaling processing of the
master switches must be robust enough to generate decisive signals
that activate downstream genes in a stable manner.
Robustness of signaling networks underlying animal development
is a well-known trait in biology. Recent studies have identified many
potential sources of robustness such as modularity, redundancy,
1902
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1902
1898–1904
Hub genes with positive feedbacks function as master switches
Fig. 3. A schematic diagram illustrating the operating mechanism of a
developmental master switch. When an input developmental signal arrives
at the master switch, a new developmental process is initiated. Black
circles denote the downstream genes involved in a new developmental
process and gray circles indicate the downstream genes for a default
developmental process. Dashed (solid blue) arrows denote interactions when
the master switch is turned off (turned on, respectively). The enlarged part
of ‘developmental switch’ shows that developmental switching at the master
switch acquires robustness through positive feedback.
network architecture and feedback control (Ciliberti et al., 2007;
Han et al., 2004; Kitano, 2004; Kwon and Cho, 2007; Kwon
and Cho, 2008; Maslov and Sneppen, 2002; Stelling et al., 2004;
Wagner, 2005; Zauner and Sommer, 2007). In this study, using a
computational approach, we demonstrated the robustness of a hub
node with positive feedback against even large perturbations in the
form of noise. We also demonstrated that the positive feedback loop
can function as an amplifier by transforming a weak input signal
into a nearly discrete output with switch-like behavior.
The broadly expressed hub genes with positive feedback loops
(Egfr, Arm, E2f ) are all involved in the category of regulation of
development. Intriguingly, these genes are well-known as master
switches. Egfr has been considered as one of the key regulatory
genes in development since it is commonly involved in several
developmental processes such as segmentation (Dubois et al., 2001;
Payre et al., 1999). In the Drosophilar wing disc, E2f is a critical
player for both cell proliferation and cell survival in the presumptive
wing margin area (Delanoue et al., 2004). Arm is a key mediator
in the signaling pathway initiated by Wnt/Wingless (Wg) proteins,
regulating numerous embryonic developmental processes such as
segment polarity in D. melanogaster and axis specification in
Xenopus (Boonchai et al., 2000; Miller and Moon, 1996; Siegfried
and Perrimon, 1994; Willert and Nusse, 1998).
The genes expressed over short time periods are of particular
biological interest. In D. melanogaster the main biological process
occurring over the time span 0–2.5h are nuclear divisions and
mitoses (Wolpert, 2007). Our study has shown that Cyca and Cyce
are highly connected genes which are only expressed at 0–2.5h time
span. In addition, the result of GO enrichment demonstrates that hub
nodes (Cyca, Cyce, E2f, Egfr, Trx) with positive feedbacks are highly
enriched in the cell cycle related categories: interphase of mitotic
cell, interphase and regulation of cell cycle. There was therefore an
important correlation between the GO enrichment analysis and the
corresponding actual biological process in this time span. This result
also supports our hypothesis that key elements (master switches) of
developmental processes in a specific time span are evolved to gain
robustness in accomplishing given developmental programs. This
finding rationalizes why positive feedback loops have been adopted
in this developmental regulatory network as a key network motif,
especially to the class of the genes which are dominantly involved
in the regulation of developmental process. Recent reports have
supported our finding. For example, G. Huang and colleagues have
shown that Wor1 (white-opaque regulator 1) is a master regulator of
white-opaque switching (Huang et al., 2006). The white and opaque
phases have different cell shapes and gene expression programs.
Once switched, each phase is stable for many cell divisions. They
have shown that WOR1 expression is regulated in an all-or-none
bistable manner and proposed that a feedback regulation of WOR1
expression supports the bistability of WOR1 expression and provides
a mechanism for white-opaque switching.
Most living organisms are remarkably stable in the face of
environmental and genetic perturbations and noises. This hallmark,
called robustness or developmental stability, reduces the amount of
phenotypic variation that is visible for selection (Ciliberti et al.,
2007; Wagner, 2005; Zauner and Sommer, 2007). However, all
organisms do evolve gradually, and the accumulation of mutations
results in modular components, functional subnetworks such as
network motifs, which eventually give rise to developmental
stability. Robustness itself can evolve as well. This is especially
evident in the case of robustness to developmental noise: mutations
resulting in higher robustness in a given environment might increase
their frequency by previously known evolutionary mechanisms
(Zauner and Sommer, 2007). Our finding provides a reasonable
clue to address how the developmental network or programs
evolve to attain their intrinsic robustness so that their sequential
developmental processes are accomplished accurately and stably.
Since we introduced time-varying network structures to
developmental network, our approach can be used to predict when
a master switch (candidate) turns on and which developmental
pathways it controls. We also suggest our approach as an applicable
method to predict a set of key factors of developmental networks or
other biological networks, which can be used to prioritize a set of
genes for experimental identification.
ACKNOWLEDGEMENTS
The authors would like to thank Philip Murray, Eileen Furlong and
Allen E. Hong for valuable comments on the manuscript.
Funding: Korea Science and Engineering Foundation grant funded
by the Korea Ministry of Education, Science & Technology through
the Systems Biology grant (M10503010001-07N030100112); the
Nuclear Research grant (M20708000001-07B0800-00110); the 21C
Frontier Microbial Genomics and Application Center Program
(Grant MG08-0205-4-0); and the World Class University grant
(R32-2008-000-10218-0).
Conflict of Interest: none declared.
REFERENCES
Alon,U. (2007) Network motifs: theory and experimental approaches. Nat. Rev. Genet.,
8, 450–461.
1903
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1903
1898–1904
C.H.Seo et al.
Alon,U. et al. (1999) Robustness in bacterial chemotaxis. Nature, 397, 168–171.
Arbeitman,M.N. et al. (2002) Gene expression during the life cycle of Drosophila
melanogaster. Science, 297, 2270–2275.
Asano,M. et al. (1996) Ectopic E2F expression induces S phase and apoptosis in
Drosophila imaginal discs. Genes Dev., 10, 1422–1432.
Ashburner,M. et al. (2000) Gene ontology: tool for the unification of biology. The Gene
Ontology Consortium. Nat. Genet., 25, 25–29.
Bagowski,C.P. et al. (2003) The JNK cascade as a biochemical switch in mammalian
cells: ultrasensitive and all-or-none responses. Curr. Biol., 13, 315–320.
Becskei,A. et al. (2001) Positive feedback in eukaryotic gene networks: cell
differentiation by graded to binary response conversion. EMBO J., 20, 2528–2535.
Bergmann,A. et al. (2002) Regulation of cell number by MAPK-dependent control of
apoptosis: a mechanism for trophic survival signaling. Dev. Cell, 2, 159–170.
Boonchai,W. et al. (2000) Expression of beta-catenin, a key mediator of the WNT
signaling pathway, in basal cell carcinoma. Arch. Dermatol., 136, 937–938.
Boone,C. et al. (2007) Exploring genetic interactions and networks with yeast. Nat.
Rev. Genet., 8, 437–449.
Brandman,O. et al. (2005) Interlinked fast and slow positive feedback loops drive
reliable cell decisions. Science, 310, 496–498.
Carbon,S. et al. (2009) AmiGO: online access to ontology and annotation data.
Bioinformatics, 25, 288–289.
Chen,G.R. et al. (2008) Rank-based edge reconstruction for scale-free genetic regulatory
networks. BMC Bioinformatics, 9, 75.
Cheng,Z. et al. (2008) Robustness analysis of cellular memory in an autoactivating
positive feedback system. FEBS Lett., 582, 3776–3782.
Ciliberti,S. et al. (2007) Robustness can evolve gradually in complex regulatory gene
networks with varying topology. Plos Comput. Biol., 3, 164–173.
Delanoue,R. et al. (2004) The Drosophila wing differentiation factor vestigial-scalloped
is required for cell proliferation and cell survival at the dorso-ventral boundary of
the wing imaginal disc. Cell Death Differ., 11, 110–122.
Dennis,G. et al. (2003) DAVID: Database for annotation, visualization, and integrated
discovery, Genome Biol., 4, P3.
Dubois,L. et al. (2001) Regulated endocytic routing modulates wingless signaling in
Drosophila embryos. Cell, 105, 613–624.
Freeman,M. (2000) Feedback control of intercellular signalling in development. Nature,
408, 313–319.
Hallinan,J.S. and Jackway,P.T. (2005) Network motifs, feedback loops and the dynamics
of genetic regulatory networks. In Proceedings of the 2005 IEEE Symposium on
Computational Intelligence in Bioinformatics and Computational Biology (CIBCB
2005), IEEE Press, Piscataway, NJ, pp. 90–96.
Han,J.D.J. et al. (2004) Evidence for dynamically organized modularity in the yeast
protein-protein interaction network. Nature, 430, 88–93.
Harris,T.J. and Peifer,M. (2005) Decisions, decisions: beta-catenin chooses between
adhesion and transcription. Trends Cell Biol., 15, 234–237.
Hooper,S.D. et al. (2007) Identification of tightly regulated groups of genes during
Drosophila melanogaster embryogenesis. Mol. Syst. Biol., 3, 72.
Huang,D.W. et al. (2007) DAVID Bioinformatics Resources: expanded annotation
database and novel algorithms to better extract biology from large gene lists. Nucleic
Acids Res., 35, W169–W175.
Huang,G. et al. (2006) Bistable expression of WOR1, a master regulator of white-opaque
switching in Candida albicans. Proc. Natl Acad. Sci. USA, 103, 12813–12818.
Kim,J.R. et al. (2007) Topological difference of core regulatory networks induces
different entrainment characteristics of plant and animal circadian clocks. Biophys.
J., 93, L01–L03.
Kim,J.R. et al. (2008) Coupled feedback loops form dynamic motifs of cellular
networks. Biophys. J., 94, 359–365.
Kitano,H. (2004) Biological robustness. Nat. Rev. Genet., 5, 826–837.
Klambt,C. et al. (1991) The midline of the Drosophila central nervous system: a model
for the genetic analysis of cell fate, cell migration, and growth cone guidance. Cell,
64, 801–815.
Kwon,Y.K. and Cho,K.H. (2007) Analysis of feedback loops and robustness in network
evolution based on Boolean models. BMC Bioinformatics, 8, 430.
Kwon,Y.K. and Cho,K.H. (2008) Coherent coupling of feedback loops: a design
principle of cell signaling networks. Bioinformatics, 24, 1926–1932.
Kwon,Y.K. and Cho,K.H. (2008) Quantitative analysis of robustness and fragility in
biological networks based on feedback dynamics. Bioinformatics, 24, 987–994.
Luscombe,N.M. et al. (2004) Genomic analysis of regulatory network dynamics reveals
large topological changes. Nature, 431, 308–312.
Maslov,S. and Sneppen,K. (2002) Specificity and stability in topology of protein
networks. Science, 296, 910–913.
Miller,J.R. and Moon,R.T. (1996) Signal transduction through beta-catenin
and specification of cell fate during embryogenesis. Genes Dev., 10,
2527–2539.
Milo,R. et al. (2004) Superfamilies of evolved and designed networks. Science, 303,
1538–1542.
Milo,R. et al. (2002) Network motifs: Simple building blocks of complex networks.
Science, 298, 824–827.
Mitrophanov,A.Y. and Groisman,E.A. (2008) Positive feedback in cellular control
systems. Bioessays, 30, 542–555.
Neufeld,T.P. et al. (1998) Coordination of growth and cell division in the Drosophila
wing. Cell, 93, 1183–1193.
Nikitin,A. et al. (2003) Pathway studio - the analysis and navigation of molecular
networks. Bioinformatics, 19, 2155–2157.
Oliveri,P. et al. (2008) Global regulatory logic for specification of an embryonic cell
lineage. Proc. Natl Acad. Sci. USA, 105, 5955–5962.
Payre,F. et al. (1999) ovo/svb integrates Wingless and DER pathways to control
epidermis differentiation. Nature, 400, 271–275.
Rajagopalan,D. (2003) A comparison of statistical methods for analysis of high density
oligonucleotide array data. Bioinformatics, 19, 1469–1476.
Rao,C.V. et al. (2002) Control, exploitation and tolerance of intracellular noise. Nature,
420, 231–237.
Raz,E. and Shilo,B.Z. (1992) Dissection of the faint little ball (flb) phenotype:
determination of the development of the Drosophila central nervous system by
early interactions in the ectoderm. Development, 114, 113–123.
Rothenberg, E.V. (2007) Cell lineage regulators in B and T cell development. Nat.
Immunol., 8, 441–444.
Schreiber,F. and Schwobbermeyer,H. (2005) MAVisto: a tool for the exploration of
network motifs. Bioinformatics, 21, 3572–3574.
Shilo,B.Z. (2003) Signaling by the Drosophila epidermal growth factor receptor
pathway during development. Exp. Cell Res., 284, 140–149.
Shin,S.Y. et al. (2009) Positive- and negative-feedback regulations coordinate the
dynamic behavior of the Ras-Raf-MEK-ERK signal transduction pathway. J. Cell
Sci., 122, 425–435.
Siegfried,E. and Perrimon,N. (1994) Drosophila wingless: a paradigm for the function
and mechanism of Wnt signaling. Bioessays, 16, 395–404.
Staal,F.J.T. and Clevers,H.C. (2005) WNT signalling and haematopoiesis: a WNT-WNT
situation. Nat. Rev. Immunol., 5, 21–30.
Stathopoulos,A. and Levine,M. (2005) Genomic regulatory networks and animal
development. Dev. Cell, 9, 449–462.
Stelling,J. et al. (2004) Robustness of cellular functions. Cell, 118, 675–685.
Stemerdink,C. and Jacobs,J.R. (1997) Argos and Spitz group genes function to
regulate midline glial cell number in Drosophila embryos. Development, 124,
3787–3796.
Taylor,R.J. et al. (2007) Network motif analysis of a multi-mode genetic-interaction
network. Genome Biol., 8, R160.
Tong,A.H. et al. (2004) Global mapping of the yeast genetic interaction network.
Science, 303, 808–813.
Tweedie,S. et al. (2009) FlyBase: enhancing Drosophila Gene Ontology annotations.
Nucleic Acids Res., 37, D555–D559.
Wagner,A. (2005) Circuit topology and the evolution of robustness in two-gene
circadian oscillators. In Proceedings of the National Academy of Sciences of the
United States of America, 102, 11775–11780.
Ward,J.J. and Thornton,J.M. (2007) Evolutionary models for formation of network
motifs and modularity in the Saccharomyces transcription factor network. PLoS
Comput. Biol., 3, 1993–2002.
Willert,K. and Nusse,R. (1998) Beta-catenin: a key mediator of Wnt signaling. Curr.
Opin. Genet. Dev., 8, 95–102.
Wolf,D.M. and Arkin,A.P. (2003) Motifs, modules and games in bacteria. Curr. Opin.
Microbiol., 6, 125–134.
Wolpert,L. (2007) Principles of Development. Oxford University Press, Oxford, New
York.
Yang,I.V. et al. (2002) Within the fold: assessing differential expression measures and
reproducibility in microarray assays. Genome Biol., 3, research0062.
Yeger-Lotem,E. et al. (2004) Network motifs in integrated cellular networks of
transcription-regulation and protein-protein interaction. Proc. Natl Acad. Sci. USA,
101, 5934–5939.
Zauner,H. and Sommer,R.J. (2007) Evolution of robustness in the signaling network of
Pristionchus vulva development. Proc. Natl Acad. Sci. USA, 104, 10086–10091.
Zhang,L.V. et al. (2005) Motifs, themes and thematic maps of an integrated
Saccharomyces cerevisiae interaction network. J. Biol., 4, 6.
1904
[13:15 3/7/2009 Bioinformatics-btp316.tex]
Page: 1904
1898–1904