Biology Reference
In-Depth Information
to perform ChIP for all TFs; instead, Y1H assays could be
used to screen a collection of TFs for binding to the
promoter of interest. In addition to being conceptually
complementary, the two approaches also have unique
advantages and limitations ( Box 4.1 ). For example, ChIP
assays provide data on TF-binding specificity in particular
cellular and environmental conditions, and can detect
binding of TF complexes and modified TFs in vivo. Y1H
assays can detect interactions with low-abundance TFs,
identify multiple TFs in a single experiment, and detect
direct interactions between DNA and TFs. PBM assays can
provide a comprehensive survey of all possible DNA-
binding site sequences recognized by TFs or TF complexes
of interest, and they can be used as a biochemical tool to
assess the impact of co-factors, such as interacting proteins,
on DNA-binding specificity.
As mentioned above, a key advantage of ChIP is that it
can be used to identify enhancers that are active in the cell
or tissue used. In other words, in such experiments the
identification of network edges can be used to annotate
GRN nodes, although it may not be clear which gene(s)
such putative enhancers regulate. For instance, genomic
regions bound by the transcriptional co-activator p300/CBP
and related histone acetyltransferases, sometimes together
with chromatin signatures comprising particular histone
modifications, can be used to accurately predict locations of
active enhancers [93,109 e 111] , even where cross-species
sequence conservation is weak [92] . Based on such data, it
has been estimated that there are of the order of 10 5
network inference has been applied successfully to
a variety of model systems, including yeast and mammals
[66,114] . A major disadvantage of these approaches is that
many of the interactions that are predicted are likely indi-
rect, and many relevant interactions are missed because the
TF that regulates a set of genes does not itself change
in expression in the same way or at all [44] (see also
Box 4.2 ).
GRNs: Visualization
When the interactions between TFs and their target genes
are connected, complex and highly intricate networks
emerge that are challenging to analyze by eye. Several tools
are available for the visualization and interrogation of
GRNs, including Cytoscape [115] , Osprey [116] , Nbrowse
[117] and Biotapestry [118] . It is important to bear in mind
that the graphs generated by these tools may reflect
a compilation of molecular events that occur throughout the
lifetime of the cell or organism, depending on the method
that was used to delineate the network, and that not all
interactions are necessarily direct or have a regulatory
consequence.
GRNs: Data Quality
The quality of GRNs depends on the proportion of real
interactions included and the number of included interactions
that are real. False negatives refer to real interactions that are
not detected with the experimental method used to delineate
the network, and false positives refer to detected interactions
that are not real. Issues related to false positives and negatives
in different types of genomic data sets have been hotly
debated over the past 15 years or so, particularly with regard
to yeast two-hybrid and Y1H assays [44,119,120] .
False negatives can arise because of limitations in each
of the assays used to build a GRN. For example, in
computationally inferred GRNs, false negatives occur
when the expression of the regulatory TF itself does not
change together with the gene module it regulates. Since
many TFs are activated at the protein level by an outside
signal that results in a post-translational modification (or
removal thereof), the number of interactions missed by
gene expression profiling is likely substantial. More tech-
nically, the degree of correlation between a TF and its target
genes, together with the threshold applied, will determine
which interactions are included and which are not.
The different types of physical interaction detection
assay also each have their limitations. Gene-centered Y1H
assays are tested in the heterologous environment of the
yeast nucleus. This provides distinct advantages, but also
results in limitations: Y1H assays are not effective at
detecting interactions of DNA with heterodimeric TFs or
TFs that need to be post-translationally modified prior to
e 10 6
enhancers in the human genome [110] .
Multiple experimental approaches have been developed
to map regulatory interactions between TFs and their target
genes, or the enhancers ascribed to these genes. In a series
of classic, one-gene-at-a-time experiments, Davidson and
colleagues delineated numerous regulatory relationships
involved in endomesoderm development in the sea urchin
(see Chapter 11). The compilation of this work has led to
one of the first GRNs described in any metazoan system
[112] . More recently, the development of a variety of high-
throughput experimental technologies ( Box 4.1 ) has
enabled the large-scale identification of regulatory inter-
actions, leading to the prospect of more complete GRNs for
a variety of model organisms and humans. A widely used
computational method is to infer GRNs from gene
expression data (sometimes called 'reverse engineering')
( Box 4.2 ). The underlying hypothesis is that sets of co-
expressed genes ('gene modules' or 'gene batteries') are
co-regulated by the same TF(s) (e.g., [64,65,113] ). TFs
are identified that exhibit a similar profile as a particular
gene battery and are subsequently inferred to be respon-
sible for, or involved in, generating the expression profile of
the gene set [65,66] . Opposite expression profiles of TFs
and target gene sets can also be informative because they
may point
to transcriptional
repression. Regulatory
Search WWH ::




Custom Search