Gene Regulatory Networks - Systems Biology Concepts and Insights - page 70

Node betweenness is another measure of information flow
through a node, and is also referred to as node centrality. This
metric is defined as the number of times a path between any
two nodes in the network passes through a node (i.e., not
necessarily directly connecting to the node) ( Figure 4.3 D).
Nodes with high between-ness are referred to as bottlenecks
and often connect different network modules (see below).
Metabolic networks exhibit a hierarchical organization
of modularity that is proposed to enable rapid reorganiza-
tion and integration with other cellular functions [126] .A
hierarchical organization has also been identified in GRNs,
using a variety of computational and mathematical
approaches. GRN hierarchy has so far been studied only in
unicellular organisms, such as bacteria (E. coli) and yeast
(S. cerevisiae), because sufficient amounts of high-quality
data are not yet available for higher organisms. These
unicellular GRNs were found to have a hierarchical orga-
nization with multiple layers of TFs that control the
expression of other TFs. While some studies suggest
a pyramidal structure for the network [127] ( Figure 4.3 C),
others have identified an overall feedforward structure
[128,129] . The different methods identified distinct
numbers of hierarchical levels that can be further grouped
into GRN layers. It has been proposed that a hierarchical
GRN structure can lead to signal amplification and can
confer robustness and adaptability [126,128] . Interestingly,
the TFs that reside in different GRN layers have distinct
attributes. For instance, TF hubs that regulate large
numbers of downstream targets are rarely found at the top
of the hierarchy. Similarly, bottleneck TFs are mostly found
in the middle, or 'core' layers. There are several conflicting
findings related to GRN hierarchy as well. Where some
studies find that the TFs at the bottom are more essential
[127] , others find that it is the TFs at the top that are
required most for viability [128] . Studies agree, however,
that the TFs occurring in the top layer are most highly
conserved. It will be interesting to determine whether the
overall networks of metazoans and plants are also hierar-
chical, and whether there are different structures in sub-
networks that control gene expression in the development
of different tissues or under particular physiological or
environmental conditions.
Network modules refer to 'neighborhoods' or clusters
of nodes that are highly interconnected ( Figure 4.3 D). This
is analogous to geographic locations such as countries in
a continent, or neighborhoods in a city, that usually are
inhabited by people with similar socioeconomic charac-
teristics. Hence, network modules have been used to
identify nodes with similar biological or biochemical
attributes. Network modules can be identified using
a variety of mathematical methods (see Chapter 9). A high
degree of modularity has been proposed to facilitate a rapid
response to external cues and has been observed in several
metabolic networks and GRNs. In GRNs, two types of

modularity can be studied: TF modules that are character-
ized by sharing target genes, and gene modules that share
interacting TFs [27,28,107] . Redundant TFs are expected to
reside in similar modules and may share target genes by
binding highly similar DNA sequences. Such redundancies
can be identified genetically, for instance in high-
throughput genetic network screens in yeast (see Chapter 6)
or by RNA interference in more complex metazoans (ref to
Perrimon chapter).
GRNs can be further broken down into different types
of small circuits or building blocks that provide different
types of information flow and hence the logic of gene
control ( Figure 4.3 E). A special type of such a building
block is called a network motif, which is defined as a type
of circuit that is overrepresented in real networks compared
to randomized networks [130] . This is analogous to TF-
binding motifs that can be enriched in the regulatory
regions of their target genes compared to background
sequence. Network motifs are identified by computation-
ally randomizing GRNs. This can be done in different
ways: by keeping the individual connectivity of each node
but swapping different edges; by changing the connectivity
of each node but keeping the overall degree distribution of
the GRN the same; or by completely randomizing the
GRN, which leads to a random degree distribution
[106,131] . Preserving network architecture is usually the
most informative, as most biological networks are scale
free, and therefore complete randomization would distort
the statistical significance because all nodes would have
approximately the same degree.
The simplest GRN building block is autoregulation and
involves only a single node: a TF that regulates its own
expression. There are two types of autoregulation: one in
which the TF represses its own expression (autorepression)
and one in which the TF activates its own gene (autoacti-
vation). Mathematical modeling has predicted possible
functionalities for autoregulation. Negative autoregulation
can confer an increase in the rate of reaching the steady state
of the TF, and results in less variability [132,133] . This can
be important to generate an appropriate response to a signal
that induces the initial expression of the TF. Depending on
the strength of its promoter and the input of other regulators,
the steady-state level can differ. Further, the rate increase
will depend on the strength of autorepression, which is
a combination of the affinity of the TF for its own cis-
regulatory regions and the mechanism by which it represses
transcription. Finally, transcriptional autorepression is
a more effective mechanism of gene regulation than protein
degradation, which could result in the same net effect, but is
more costly to the cell. Positive autoregulation can result in
a slowing down in reaching steady-state TF levels and an
increase in expression noise. Weak positive autoregulation,
for instance caused by low-affinity binding of the TF to its
own promoter, can result in large differences in TF