190
Xenopus
12.2.4.2. Feedforward Loop Subcircuits
To understand the function of feedforward loops (FFLs), it
is necessary to understand how the actions of gene A and B
are integrated to regulate the activity of gene C. Two common input functions are an AND gate, in which the presence
of both products A and B is required, and an OR gate, in
which binding of either A or B is suffcient to activate gene
C (Peter and Davidson, 2015). An AND gate can be benef -
cial in the tight control of factor C expression, as factor C is
only activated when both factor A and B are expressed. On
the other hand, an OR gate enables the sustained expression
of factor C despite the loss of the initial factor A. Much of
the essential behavior of FFLs appears to use either an AND
or an OR gate system.
We investigated the types of FFLs that are utilized by
the Xenopus mesendodermal GRN. Among 89 three-gene
FFLs, 63 were type I coherent FFLs (Charney et al., 2017a).
The coherent FFL network structure assumes product A
activates product B, and the activation of product C requires
positive inputs from both products A and B. In general, the
coherent FFL allows a cell to respond rapidly to stimuli
in both positive (ON) or negative (OFF) directions and/or
regulate the temporal onset of gene expression (Mangan
et al., 2003). In the majority of cases examined during
early mesendoderm development (Charney et al., 2017a),
the initial activators are frequently maternal TFs such as
Ctnnb1, Foxh1, Smad2/3, or Vegt (gene A). These maternal factors positively stimulate the expression of early- and
mid-blastula zygotic genes such as wnt8a, sia1, sia2, mix1,
gsc, and the nodal genes (gene B), which, in turn, activate
the expression a larger number of later expressed mesendodermal genes (gene C) together with sustained inputs
from Ctnnb1, Foxh1, Smad2/3, or Vegt (gene A). Thus,
in this case, the direct, primary, activated zygotic targets
of maternal TFs can function to maintain the expression
of later, secondary, activated genes. Coherent FFLs offers
an effective way to robustly regulate the temporal onset of
genes so that sequential activation of genes can be attained.
The prominence of the coherent FFL during mesendoderm
formation suggests that temporal regulation of the cascade
may be extremely important during early embryogenesis,
where events are changing rapidly. Additionally, this subcircuit can provide lineage memory to the cell: Only when
cells frst express gene A, they attain a specifc cell fate
upon expression of gene B.
12.2.4.3. Spatial Exclusion Subcircuits
Repression of alternative cell fates is an important and
general feature of development (Figure 12.2C ). As cells
differentiate, it is common for gene expression domains to
initially have broad overlap, but then their expression patterns are gradually segregated into distinct spatial domains,
representing different cell specifcation states. This is
accomplished by silencing the expression of alternative
GRN subcircuits. This subcircuit requires the expression of
a gene encoding a repressor that specifcally targets a key
molecule in an alternative GRN (Peter and Davidson, 2015).
This can also be accomplished by two genes encoding
transcription factors that mutually and directly repress one
another. Typically, exclusion subcircuits are revealed experimentally when expression of the repressor is disrupted,
leading to ectopic activation of the alternative regulatory
state. Examples of spatial exclusion subcircuits include the
interaction between Gsc-Hhex in specifcation of prechordal
plate mesoderm and endoderm formation (Brickman et al.,
2000), and Gsc-Ventx2 (Yasuoka et al., 2014) and Gsc-Tbx2
(Artinger et al., 1997), some of which are mutual exclusion
interactions in dorsal and ventral mesoderm formation.
12.3. PRESENT STATUS OF THE FIELD
Despite past advances, all GRNs are still far from complete.
This is in part because the past network connections provide only limited previews of the selected interactions that
were chosen with a priori knowledge—large-scale genomic
data were not fully integrated into the network analysis.
Even the most intensely studied vertebrate mesendoderm
GRN will inevitably miss the involvement of TFs and many
important interactions. However, with the accumulation of
large “omics” data sets, generation of a GRN based on a
combination of computational and genomic methods is feasible. Mechanistic GRN building requires two key pieces
of information—native TF binding data to CRMs and the
expression output of the gene as a consequence of TF binding to the CRM. By integrating high-throughput omics data
detecting these two events, it is feasible to build a genomescale GRN. Specifcally, RNA-seq transcriptome prof ling
studies can reveal the timing and scale of gene activation,
and expression changes in gain- and loss-of-function experiments provide a wealth of potential regulatory connections
between TFs and CRMs of these target genes. ChIP-seq
identifes physical sites of TF binding, and ATAC-seq and
DNase-seq datasets can provide information about the
accessibility of CRMs, thus providing evidence for direct
physical interactions.
Recently, a new approach was reported capable of building a Xenopus tropicalis mesendoderm GRN after integrating over 150 transcriptomic RNA-seq and ChIP/ATAC-seq
data sets (Jansen et al., 2022). The integration of large
genomic datasets derived from different data types has
generally been diffcult in the past. The frst challenge is
to obtain the data themselves, which is labor intensive. The
second is a signif cant diffculty of integrating different data
types—the RNA-seq datasets, which report expression levels of transcripts/genes, and the DNA datasets which report
genomic regions outside of the transcription units that are
bound by TFs or are contained in open chromatin. The third
is to fnd ways to interrogate the integrated data in ways that
permit building GRNs. In order to integrate different types
of large data sets, a type of machine learning, called a selforganizing map (SOM), was applied, which is a type of unsupervised neural network that also permits the visualization
of high-dimensional data (Kohonen, 2001). A SOM assesses
Xenopus
12.2.4.2. Feedforward Loop Subcircuits
To understand the function of feedforward loops (FFLs), it
is necessary to understand how the actions of gene A and B
are integrated to regulate the activity of gene C. Two common input functions are an AND gate, in which the presence
of both products A and B is required, and an OR gate, in
which binding of either A or B is suffcient to activate gene
C (Peter and Davidson, 2015). An AND gate can be benef -
cial in the tight control of factor C expression, as factor C is
only activated when both factor A and B are expressed. On
the other hand, an OR gate enables the sustained expression
of factor C despite the loss of the initial factor A. Much of
the essential behavior of FFLs appears to use either an AND
or an OR gate system.
We investigated the types of FFLs that are utilized by
the Xenopus mesendodermal GRN. Among 89 three-gene
FFLs, 63 were type I coherent FFLs (Charney et al., 2017a).
The coherent FFL network structure assumes product A
activates product B, and the activation of product C requires
positive inputs from both products A and B. In general, the
coherent FFL allows a cell to respond rapidly to stimuli
in both positive (ON) or negative (OFF) directions and/or
regulate the temporal onset of gene expression (Mangan
et al., 2003). In the majority of cases examined during
early mesendoderm development (Charney et al., 2017a),
the initial activators are frequently maternal TFs such as
Ctnnb1, Foxh1, Smad2/3, or Vegt (gene A). These maternal factors positively stimulate the expression of early- and
mid-blastula zygotic genes such as wnt8a, sia1, sia2, mix1,
gsc, and the nodal genes (gene B), which, in turn, activate
the expression a larger number of later expressed mesendodermal genes (gene C) together with sustained inputs
from Ctnnb1, Foxh1, Smad2/3, or Vegt (gene A). Thus,
in this case, the direct, primary, activated zygotic targets
of maternal TFs can function to maintain the expression
of later, secondary, activated genes. Coherent FFLs offers
an effective way to robustly regulate the temporal onset of
genes so that sequential activation of genes can be attained.
The prominence of the coherent FFL during mesendoderm
formation suggests that temporal regulation of the cascade
may be extremely important during early embryogenesis,
where events are changing rapidly. Additionally, this subcircuit can provide lineage memory to the cell: Only when
cells frst express gene A, they attain a specifc cell fate
upon expression of gene B.
12.2.4.3. Spatial Exclusion Subcircuits
Repression of alternative cell fates is an important and
general feature of development (Figure 12.2C ). As cells
differentiate, it is common for gene expression domains to
initially have broad overlap, but then their expression patterns are gradually segregated into distinct spatial domains,
representing different cell specifcation states. This is
accomplished by silencing the expression of alternative
GRN subcircuits. This subcircuit requires the expression of
a gene encoding a repressor that specifcally targets a key
molecule in an alternative GRN (Peter and Davidson, 2015).
This can also be accomplished by two genes encoding
transcription factors that mutually and directly repress one
another. Typically, exclusion subcircuits are revealed experimentally when expression of the repressor is disrupted,
leading to ectopic activation of the alternative regulatory
state. Examples of spatial exclusion subcircuits include the
interaction between Gsc-Hhex in specifcation of prechordal
plate mesoderm and endoderm formation (Brickman et al.,
2000), and Gsc-Ventx2 (Yasuoka et al., 2014) and Gsc-Tbx2
(Artinger et al., 1997), some of which are mutual exclusion
interactions in dorsal and ventral mesoderm formation.
12.3. PRESENT STATUS OF THE FIELD
Despite past advances, all GRNs are still far from complete.
This is in part because the past network connections provide only limited previews of the selected interactions that
were chosen with a priori knowledge—large-scale genomic
data were not fully integrated into the network analysis.
Even the most intensely studied vertebrate mesendoderm
GRN will inevitably miss the involvement of TFs and many
important interactions. However, with the accumulation of
large “omics” data sets, generation of a GRN based on a
combination of computational and genomic methods is feasible. Mechanistic GRN building requires two key pieces
of information—native TF binding data to CRMs and the
expression output of the gene as a consequence of TF binding to the CRM. By integrating high-throughput omics data
detecting these two events, it is feasible to build a genomescale GRN. Specifcally, RNA-seq transcriptome prof ling
studies can reveal the timing and scale of gene activation,
and expression changes in gain- and loss-of-function experiments provide a wealth of potential regulatory connections
between TFs and CRMs of these target genes. ChIP-seq
identifes physical sites of TF binding, and ATAC-seq and
DNase-seq datasets can provide information about the
accessibility of CRMs, thus providing evidence for direct
physical interactions.
Recently, a new approach was reported capable of building a Xenopus tropicalis mesendoderm GRN after integrating over 150 transcriptomic RNA-seq and ChIP/ATAC-seq
data sets (Jansen et al., 2022). The integration of large
genomic datasets derived from different data types has
generally been diffcult in the past. The frst challenge is
to obtain the data themselves, which is labor intensive. The
second is a signif cant diffculty of integrating different data
types—the RNA-seq datasets, which report expression levels of transcripts/genes, and the DNA datasets which report
genomic regions outside of the transcription units that are
bound by TFs or are contained in open chromatin. The third
is to fnd ways to interrogate the integrated data in ways that
permit building GRNs. In order to integrate different types
of large data sets, a type of machine learning, called a selforganizing map (SOM), was applied, which is a type of unsupervised neural network that also permits the visualization
of high-dimensional data (Kohonen, 2001). A SOM assesses