Content uploaded by Shelley Lumba
Author content
All content in this area was uploaded by Shelley Lumba on Oct 13, 2017
Content may be subject to copyright.
Developmental Cell
Resource
A Mesoscale Abscisic Acid Hormone
Interactome Reveals a Dynamic
Signaling Landscape in Arabidopsis
Shelley Lumba,
1
Shigeo Toh,
1
Louis-Franc¸ ois Handfield,
2
Michael Swan,
1
Raymond Liu,
1
Ji-Young Youn,
3
Sean R. Cutler,
4
Rajagopal Subramaniam,
5
Nicholas Provart,
1
Alan Moses,
1,2
Darrell Desveaux,
1,
*and Peter McCourt
1,
*
1
Cell & Systems Biology, University of Toronto and the Centre for The Analysis of Genome Evolution and Function, University of Toronto,
Toronto, ON M5S 3B2, Canada
2
Department of Computer Science, University of Toronto, Toronto, ON M5S 2E4, Canada
3
Banting and Best Department of Medical Research, Terrence Donnelly Centre for Cellular and Biomolecular Research, University of Toronto,
Toronto, ON M5S 3E1, Canada
4
Botany and Plant Sciences, Chemistry Genomics Building, University of California, Riverside, Riverside, CA 92521, USA
5
Agriculture and Agrifood Canada, 960 Carling Avenue, Ottawa, ON K1A 06C, Canada
*Correspondence: darrell.desveaux@utoronto.ca (D.D.), peter.mccourt@utoronto.ca (P.M.)
http://dx.doi.org/10.1016/j.devcel.2014.04.004
SUMMARY
The sesquiterpenoid abscisic acid (ABA) mediates
an assortment of responses across a variety of king-
doms including both higher plants and animals. In
plants, where most is known, a linear core ABA
signaling pathway has been identified. However,
the complexity of ABA-dependent gene expression
suggests that ABA functions through an intricate
network. Here, using systems biology approaches
that focused on genes transcriptionally regulated
by ABA, we defined an ABA signaling network of
over 500 interactions among 138 proteins. This map
greatly expanded ABA core signaling but was still
manageable for systematic analysis. For example,
functional analysis was used to identify an ABA
module centered on two sucrose nonfermenting
(SNF)-like kinases. We also used coexpression anal-
ysis of interacting partners within the network to un-
cover dynamic subnetwork structures in response to
different abiotic stresses. This comprehensive ABA
resource allows for application of approaches to un-
derstanding ABA functions in higher plants.
INTRODUCTION
Abscisic acid (ABA) is a sesquiterpenoid-based compound that
has biological activity in a variety of organisms ranging from
sponges and human parasites to mammalian cells in addition
to plants (see Wasilewska et al., 2008 and Li et al., 2011 for re-
view). In animals, for example, ABA appears to stimulate immune
responses and insulin release from pancreatic cells. It has also
been implicated in heat and light stress responses in animals.
But the absence of a good model genetic system to study ABA
in animals makes it difficult to understand the mechanisms un-
derlying the role of ABA in mediating various responses. By
contrast, in higher plants and particularly in the model Arabidop-
sis thaliana, functional analysis has shown ABA to be an impor-
tant hormone in both embryonic and vegetative growth and
development. In vegetative tissues, ABA protects plants from a
variety of abiotic stresses such as drought, temperature, salt,
and oxidative stresses (Nakashima and Yamaguchi-Shinozaki,
2013; Cramer et al., 2011). For this reason, the synthesis and
signal transduction of ABA have been intensively studied not
only at a fundamental level but also for potential applications in
crop-based biotechnology (Ben-Ari, 2012; Wilkinson et al.,
2012).
In higher plants, ABA appears to regulate both ‘‘fast re-
sponses’’ relating to ion channel flux in the guard cell and
‘‘slow responses’’ related to gene expression within the nucleus
(see Hubbard et al., 2010 for review). Surprisingly genetic dissec-
tion of ABA signaling defines a relatively simple hierarchical
signaling pathway that appears to regulate both responses (Fig-
ure 1A) (Cutler et al., 2010; Hubbard et al., 2010; Lumba et al.,
2010). In the case of transcriptional responses, ABA binds to
the PYR/PYL/RCAR (PYL) family of receptors, thereby allowing
them to interact and inhibit a class of A-type protein phospha-
tase 2Cs (PP2Cs). PP2C inactivation permits a small set of class
3 sucrose nonfermenting-1-related protein kinase 2s (SNRK2s)
to phosphorylate a collection of basic leucine zipper transcrip-
tion factors (TFs) (ABA-responsive element binding factors
[ABFs]). The phosphorylation of ABFs activates downstream
gene expression through the cis-acting ABA response element
(ABRE). ABA core components also directly regulate S-type
anion channels SLAC1 and SLAH3 in the guard cell, demon-
strating the core ABA signaling pathway also has cytosolic
‘‘fast response’’ targets (Geiger et al., 2011; Brandt et al., 2012).
Although the core signaling pathway illustrates the importance
of ABFs in ABRE-dependent gene expression, the TFs that regu-
late the expression of the many other genes induced by ABA but
lacking ABREs in their promoters have not been clearly defined.
For example, up to 11 different TF families have been implicated
in ABA-related processes (Fujita et al., 2011). Moreover,
because of the role of ABA in protecting against various environ-
mental stresses, the core ABA signaling pathway must work in
coordination with signaling pathways involved in drought, heat,
salt, and cold stress (Hey et al., 2010; Huang et al., 2012). The
360 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
overlap between primary ABA signaling and stress response
pathways results in the formation of subnetworks that could
reveal points of crosstalk and potential ways of modulating the
ABA signal.
System-based approaches are beginning to be used to link
signaling pathways into larger complex networks in plant biology
(Van Leene et al., 2011; Arabidopsis Interactome Mapping Con-
sortium, 2011; Lalonde et al., 2010). As networks become larger,
however, interactions between components can quickly scale to
a point where defining the roles of any particular protein is chal-
lenging (Hartwell et al., 1999; Spirin and Mirny, 2003). With these
considerations, we used specific parameters to expand the
components involved in ABA signaling while keeping the
signaling network space experimentally manageable. We did
Figure 1. ABA Signaling Networks
(A) Model of core ABA signaling pathway as
defined by genetic analysis. An arrow indicates
positive genetic interaction, whereas a T-bar rep-
resents negative genetic interaction.
(B) A comparison of the expression of the 282set in
wild-type (WT) seedlings exposed to 50 mM ABA
for 6 hr versus the snrk2 (snrk2.2/2.3/2.6) triple
mutant defective in core ABA signaling. The red
dots are genes induced at least 2-fold in the aba2-
2mutant in response to ABA. Blue dots are genes
repressed at least 2-fold by ABA. Gene expression
below the purple diagonal line are dampened in
induction in the snrk2 triple, whereas genes above
the diagonal line have increased expression.
Graph axes are log-scale fold change.
(C) An edge-weighted force-directed representa-
tion of the TRAIN. Node color is designated by GO
Slim Functional Annotation. Proteins with similar
interaction profiles cluster more closely to each
other. Line thickness represents edge-weighted
confidence based on statistical analysis.
See also Figure S1 and Table S1.
this by first building a protein-protein
interaction (PPI) network or interactome
via yeast two-hybrid (Y2H) technologies
in which the input gene set was limited
to genes involved in the primary tran-
scriptional response to ABA. Because
PPIs occur more frequently within a
group of coregulated genes than a
randomly expressed gene set, this ABA-
regulated gene set should be enriched
for interacting protein partners (Ge et al.,
2001). In addition, much of the core ABA
signaling pathway is transcriptionally
regulated by ABA, thus these core
components provide a framework on
which to expand the signaling network.
Next, unlike many yeast-based interac-
tome networks that use a single reporter
system, we generated our ABA network
using multiple reporter outputs and then
integrated these outputs using machine
learning algorithms and imaging soft-
ware. This analysis allowed us to rank protein interactions based
on statistically derived confidence levels.
This multilayered approach generated the ‘‘transcriptionally
regulated ABA interactome network’’ (TRAIN) that encompassed
over 500 additional PPIs of high quality. The TRAIN can be inte-
grated with published protein interaction databases to build an
expanded network (eTRAIN) of over 1,000 interactions. Our
approach verified most of the published ABA-related inter-
actions and expanded core ABA signaling by more than 50 pro-
teins. The development of the TRAIN is a useful resource for
plant hormone researchers. For example, because the inter-
action space is experimentally tractable, systematic and
targeted functional analysis can be performed. As a proof of
principle, we identified a 20-member subnetwork centered
Developmental Cell
ABA Signaling Networks in Plants
Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc. 361
around two SNRK3s that appears to negatively modulate ABA
responsiveness. We also show the TRAIN can be used to
examine the dynamic structure of protein networks in response
to various abiotic stresses. By mapping stress-dependent coex-
pression of gene pairs over multiple time points and tissues, we
were able to visualize how the wiring of the TRAIN subnetworks
changed in response to specific abiotic stresses. We used this
information to discover an important TF involved in Arabidopsis
salt stress response.
RESULTS
The Transcriptionally Regulated ABA Interactome
Network
To identify a collection of genes that were rapidly regulated by
ABA, we performed whole-genome transcript profiling on an
Arabidopsis mutant deficient in ABA synthesis (aba2)(Le
´on-
Kloosterziel et al., 1996). The use of an auxotroph versus a
wild-type plant not only avoided complications involving the
plant’s response to exogenous and endogenous ABA pools
but also may sensitize the transcription of both major and minor
genes in response to ABA application. Because we were using
seedlings deficient in endogenous ABA, we could limit ABA
exposure to low concentrations (1 mM) for a short duration
(6 hr). In addition, both ABA-treated and untreated plants were
exposed to the translational inhibitor cycloheximide to enrich
for primary transcriptional events (Figure S1 available online).
Using these experimental conditions, we identified 282 genes
(282set) whose expression changed 2-fold or more in response
to ABA versus untreated seedlings (Table S1). Within the
282set, we found that 85.2% and 60.2% of genes that were
induced and repressed by ABA, respectively, are also identified
in other ABA transcriptome studies, suggesting the 282set is
a good representation of ABA-regulated genes (Table S1).
Moreover, the ABRE promoter element was significantly over-
represented (125 of 282 genes, p value of 3.98 310
18
), and
approximately 80% of the 282set genes were dampened in the
snrk2.2/2.3/2.6 triple mutant, which is defective in core ABA
signaling (Figure 1B) (Fujita et al., 2009). Interestingly, although
the 282set contained many well-characterized ABA-responsive
genes, approximately 40% of the gene set fell below the 2-fold
expression cutoff that is often the basis of hormonally based
gene expression experiments, but followed the same trend of
induction or repression by ABA, which suggests aba2-2 did
sensitize transcription to ABA treatment (Table S1). In summary,
the 282set represented a transcriptionally sensitized gene set
that is mostly regulated by the core ABA signaling pathway.
We next constructed a protein interaction map for 258 genes
of the 282set based on a binary or an ‘‘all by all’’ Y2H approach
(Table S1). Because ABA receptor-PP2C interactions are ABA
dependent in yeast, all Y2H assays were performed in both the
presence and absence of ABA to determine whether any of the
66,564 potential PPIs were dependent on ABA (Park et al.,
2009). Therefore, this study represented a comprehensive
analysis of ABA-dependent interactions. Larger scale Y2H
analysis is often performed in series where autoactivating pro-
teins are first identified using one reporter system, and then all
subsequent interactions are retested using a different reporter
system (Yu et al., 2008). We found, however, that protein autoac-
tivation frequently depended on the reporter assay; thus, per-
forming Y2H assays in series could result in a sampling bias
that eliminates potential interactions only found in one reporter
system. To reduce reporter bias, we performed Y2H assays
using two reporter outputs in parallel. Although it is relatively
easy to score interactions involving a colorimetric output
(X-gal), the quantification of outputs based on yeast growth
(growth in the absence of leucine) required the development of
an imaging algorithm (DataEater) that did not require high-
resolution images of yeast colonies (Supplemental Experimental
Procedures). DataEater automatically generated a table of pixel
intensities for each colony as a relative quantification of yeast
growth. Using these values in combination with the X-gal data,
we devised a simple generative model that assigned confidence
values for each interacting pair that led to a list of 512 statistically
significant PPIs involving 138 gene nodes (Supplemental Exper-
imental Procedures and Table S1). This set of interactions repre-
sents approximately 0.8% of the possible interaction space.
The TRAIN contains genes that have functions in diverse
processes ranging from metabolism and proteolysis to signaling
and transcription (Figure 1C). A force-directed representation of
the TRAIN in which proteins with similar interacting partners are
more proximal while proteins with less similar interactions are
positioned further apart revealed a dense cluster of protein inter-
actions (Figure 1C). Gene ontology (GO) annotation suggested
this cluster was enriched for proteins that localize in the nucleus,
many of which are TFs. Furthermore, this dense region also
contained a number of core ABA signaling components,
including PYL receptors, PP2Cs, and ABFs, all of which are
thought to have roles in the nucleus. Thus, many of the PPIs iden-
tified were not only ABA dependent for coexpression but also
encoded nuclear proteins. Enrichment in protein pairs annotated
with common GO is a common criterion for high-quality and bio-
logically significant interactome data sets (Stelzl et al., 2005). In
summary, the TRAIN fulfills many of the standard validation re-
quirements commonly used in interactome studies.
The eTRAIN
The TRAIN is obviously not an exhaustive network because it is
biased toward ABA-regulated gene expression. We therefore
expanded the interactome by querying TRAIN proteins against
three large protein interactome databases and a collection of
literature-curated interactions (Arabidopsis Interactome Map-
ping Consortium, 2011; Lalonde et al., 2010; Popescu et al.,
2009). This analysis expanded the TRAIN to a network of 573
proteins that encompassed 1,008 interactions and was desig-
nated the eTRAIN (Figure 2). The eTRAIN not only demonstrated
the complex relationship of ABA signaling with respect to diverse
biological processes but also suggested potential interplay be-
tween ABA and these other processes. For example, 31 genes
within the eTRAIN were GO annotated as having roles in other
hormone signaling pathways, and these nodes show approxi-
mately 100 interactions within the eTRAIN (Figure 2). These
hormone-based nodes may act as points of crosstalk that help
coordinate ABA with other hormonal responses.
Core ABA Signaling Network
Presently, core ABA signaling is composed of 13 PYL receptors,
9 PP2Cs, 3 SNRK2s, and 5 ABF transcriptional activators (Cutler
Developmental Cell
ABA Signaling Networks in Plants
362 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
et al., 2010; Hubbard et al., 2010). The TRAIN consists of three
PYL receptors (PYR1,PYL4,PYL8), four PP2Cs(ABI1,AHG3,
HAB1,HAI1), two ABFs(ABI5,ABF3), but not SNRK2 kinases,
because the latter are not induced by ABA addition (Fujita
et al., 2009; Fujii and Zhu, 2009). Of the 12 possible TRAIN-
related PYL-PP2C interactions, 11 were recapitulated and no
other interactions were found to be ABA dependent (Figure S2).
We also identified previously reported interactions (PYL8/
MYB77, AFP2, ABI5/AFP4), suggesting that our interaction con-
ditions and statistical filtering were of high quality (Arabidopsis
Interactome Mapping Consortium, 2011; Garcia et al., 2008).
In addition to these published interconnections, interactions
among core proteins were identified between PP2Cs (ABI1/
HAI1, AHG3/HAI1) and between PP2Cs and ABFs (ABI1/ABI5,
HAI1/ABI5) (Figure 3A). These interactions are consistent with
reports of interactions and direct dephosphorylation of select
ABFs by various PP2Cs (Lynch et al., 2012). We also uncovered
additional connections between the PYL8 receptor and the TFs
MYB49 and a basic-helix-loop-helix (bHLH) (At1g10585) that
we have designated as AIB1 for ABA-induced basic helix-loop-
helix (Figure 3A). In addition, PP2Cs were found to interact with
kinases (SNRK3.15, SNRK3.22, MAP3Kv4), a phosphatase
(SSP4), a large set of TFs, metabolic enzymes, and proteins of
unknown function (Figure 3A). The PP2C interactions with
SNRK3.15 and SNRK3.22 add to the large number of core
PP2C-SNRK3 family interactions that already exist in the litera-
ture and further support the interplay between these ABA-regu-
lated phosphatases and these calcium-regulated kinases
(Batisti
c et al., 2012; Coello et al., 2011). Finally, ABFs were
found to interact with the following: two kinases (MAP3Kv4,
WNK2), a phosphatase (SSP4), seven TFs (ABA-responsive
NAC [ANAC]19, ANAC32, ANAC72, ERF58, CIR1, MYB49,
AIB1), and a collection of proteins involved in metabolism and
unknown functions (Figure 3A).
Functional analysis using sensitivity to ABA as a phenotype
has been useful in classifying genes as positive or negative
regulators of the overall ABA response. Literature curation of
the components that touch core signaling proteins indicate eight
genes encode positive regulators and 14 are negative (Figure 3A).
To expand this functional analysis, we arbitrarily identified loss-
of-function mutations for 16 TRAIN genes that interacted with
core PP2Cs and determined their ABA sensitivity at the level of
germination and cotyledon expansion (Figure 3B). Among the
TRAIN genes, myb12,athb12,erf58,rap2.2, and map3kv4
all showed increased sensitivity to ABA, suggesting they
encode negative regulators of the ABA response. The ABA
Figure 2. eTRAIN: An Expanded Map of ABA Signaling PPIs
Core ABA signaling components are represented by larger nodes at the upper center of the map, whereas TRAIN components are represented by nodes in the
middle ellipse. Core and TRAIN node component colors are designated by GO Slim Functional Annotation. Literature-reported interactions are shown on the
outside as orange nodes. Orange edges indicate interactions with core components. Categories shown on the outside are designated by GO Slim Functional
Annotation. JA, jasmonic acid; GA, gibberellic acid; BR, brassinosteroid. The list of genes and interactions can be found in Table S1.
Developmental Cell
ABA Signaling Networks in Plants
Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc. 363
hypersensitivity of athb12 is consistent with other studies
(Valde
´s et al., 2012). In addition, the myb12 triple mutant contain-
ing another myb12 allele in combination with loss-of-function
mutations in its closest homologs (myb12 myb11 myb111)
resulted in an ABA-hypersensitive phenotype (Figure 3B).
MYB12 has been implicated as a positive transcriptional regu-
lator of flavonol synthesis (Stracke et al., 2010). This additional
ABA-related phenotype suggests this TF impinges on both
pigment and hormone signaling. Finally, we assayed the effects
of ERF058,RAP2.2, and MAP3Kv4 on ABA sensitivity by con-
structing transgenic lines that conditionally misexpressed these
genes. We found that dexamethasone (DEX)-inducible gain-of-
function lines for all three transgenics were less sensitive to
ABA in the presence of DEX (Figure 3B). These results suggest
these genes were both necessary and sufficient negative regula-
tors of ABA responsiveness.
Figure 3. Core ABA Signaling Network
(A) A network of core ABA signaling pathway associations based on the TRAIN and literature-curated interactions. A list of the genes, their partners, and their
annotations can be found in Table S1. Gray edges represent previously published core ABA signaling interactions. Pink edges represent interactions found in this
study. The ABA core signaling pathway is represented by larger black nodes and arrows. The list of phos phorylated proteins (orange balls) can be found in Table
S1. The designation of positive and negative regulators in based on genetic analysis performed in this study or the literature.
(B) Loss- and gain-of-function mutant analysis of select core component partners. Lines were tested on 0.7 mM ABA for hypersensitivity to ABA and 2.0 mM ABA
for insensitivity. Three to six independent experiments were performed on each line using 50–100 seeds, and similar results were obtained. The loss-of-function
ABA-hypersensitive mutant ahg3-1 and ABA-insensitive abi5-1 were used for comparison. For gain-of-function analysis, a strong DEX-inducible line over-
expressing HAI1 was used. Data are represented as mean ± SD.
Developmental Cell
ABA Signaling Networks in Plants
364 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
In summary, over 50 interactions to core components have
been added from the TRAIN to approximately 60 PPIs curated
from the literature. Interestingly, 20 of these interactions included
a member in one of six TF families (MYB,HB,DELLA,ANAC,
ERF,bHLH). By contrast, only the B3 domain TF ABI3 has
been reported to interact with an ABA core component outside
of the ABFs (Nakamura et al., 2001). Thus, the conditions used
in this study appeared to enrich for TF-based interactions that
are missed by other high- and low-throughput interaction
studies. Finally, no loss- or gain-of-function lines tested in this
study uncovered a positive regulator of ABA response. Previous
studies indicate that many genes that interact with core signaling
components encode negative regulators of the ABA response
(Figure 3A; Table S1). TRAIN-related signaling components, for
which an ABA-related phenotype could be identified, appear to
add to this list of negative regulators.
An SNRK3 Network That Impinges on ABA
Responsiveness
One of our goals in developing a mesoscale ABA signaling
network was to generate an experimentally tractable system in
order to identify signaling modules. To test this hypothesis, we
decided to focus on a subnetwork centered around two kinases,
SNRK3.15 and SNRK3.22, that formed two large and overlap-
ping hubs (Figure 4A). The SNRK3 family is of particular interest
because members of this group in Arabidopsis are implicated in
a myriad of plant metabolic and stress responses (Batisti
c et al.,
2012; Coello et al., 2011). Moreover, a number of related SNRK3
proteins interact with known core PP2Cs, ABI1, and ABI2, sug-
gesting that these kinases crosstalk to the core ABA signaling
pathway (Batisti
c et al., 2012; Ohta et al., 2003). Finally, higher
resolution interaction mapping (Supplemental Experimental Pro-
cedures) revealed that SNRK3 interactions were significantly
enriched for TFs compared to other types of partners (Fisher’s
exact test, p < 2.8 310
6
)(Figure 4A). Moreover, many of the
TF partners of SNRK3 interacted with each other, often resulting
in three or four node interaction motifs that are indicative of
signaling modules (Zhang et al., 2005).
We used bimolecular fluorescence complementation (BiFC)
assays in Nicotiana benthamiana to validate a large number of
SNRK3 interactions in planta (Walter et al., 2004). Based on 22
Y2H results that included both presence and absence of
SNRK3-dependent interactions, 17 (77%) were recapitulated in
planta (Figures 4B and S3A–S3D). Green and blue lines indicate
Y2H interactions that were recapitulated and not recapitulated
by BiFC analysis, respectively. We then decided to study a
number of SNRK3-TF interactions in more detail by monitoring
protein modification using 2D gel shift assays in yeast. We found
that the TFs ATHB6, MYB49, RGL3, ERF058, RAP2.2, and
ANAC018 showed shifts in charge and size consistent with pro-
tein phosphorylation when coexpressed with a SNRK3 in yeast,
as indicated by green lines (Figures 4C and S4A). Blue lines
denote the absence of a shift. Finally, both SNRK3s phosphory-
lated RAP2.2, ATHB6, and ANAC18 in vitro, suggesting that
these TFs are direct downstream targets of SNRK3s (Figures
4D and S4B). In summary, the level of reproducibility between
the yeast- and plant-based assays indicated that many
SNRK3-dependent interactions are of high quality. More impor-
tantly, interactions gleaned from the TRAIN could be used as a
guide in biochemically based experiments to identify direct tar-
gets of SNRK3.15 and/or SNRK3.22.
Our analysis of the SNRK3.15/SNRK3.22 subnetwork sug-
gested these kinases form a signaling module that may have
roles in the plant’s overall ABA response. A key tenet of a
signaling module specifies that perturbing modular components
often result in similar phenotypes (Hartwell et al., 1999).
SNRK3.22 is a key regulator of a plasma membrane H
+
-adeno-
sine triphosphatase (ATPase) function, but there is little sugges-
tion of a role for this kinase in ABA-mediated transcription
(Fuglsang et al., 2007). Functional analysis of SNRK3.15 sug-
gests this gene is a negative regulator of ABA response, but
paradoxically loss of SNRK3.15 decreases expression of com-
mon ABA response genes in the presence of ABA (Qin et al.,
2008). The TRAIN suggested these kinases should function in
related processes that should connect to ABA-based transcrip-
tion. To explore this possibility, we first analyzed both loss- and
gain-of function alleles of these genes with respect to ABA sensi-
tivity. We found that loss-of-function alleles in both kinases had
increased sensitivity to ABA versus wild-type (Figure 4E).
Consistent with this, DEX-inducible gain-of-function lines for
both of the SNRK3s were less sensitive to ABA in the presence
of DEX (Figure 4E). Therefore, SNRK3.15 and SNRK3.22 were
both necessary and sufficient negative regulators of ABA
responsiveness. Next, to clarify the role of SNRK3.15 in ABA-
mediated transcription, we activated SNRK3.15 in our DEX-
inducible SNRK3.15 seedlings by DEX addition and monitored
genome-wide transcription after ABA treatment (Figure S5). After
filtering the data for consistency of expression, approximately
200 genes showed a 2-fold change in expression upon ABA
addition in the absence of SNRK3.15 activation (Table S2). Of
these, 141 genes were misexpressed in the snrk2 triple mutant;
thus, 70% of the ABA-responsive genes in our data set are influ-
enced by core ABA signaling (Fujita et al., 2009). Upon activation
of SNRK3.15, transcription of many of these ABA-responsive
genes was attenuated (Figure 4F). This result is consistent with
our functional analysis and suggests the SNRK3.15/SNRK2.22
hub plays a role in dampening transcription dependent on core
ABA signaling.
Integrative Mapping of the TRAIN by Coexpression
Analysis Identifies Putative Abiotic Modules
Whole-genome expression analysis has been used extensively
to dissect the role of ABA in the overall response of plants to
various stresses such as drought, salt, and temperature (Kilian
et al., 2007; Cramer et al., 2011; Zeller et al., 2009). However, an-
imal studies have demonstrated that the performance of tran-
scriptome data as a diagnostic tool of particular processes is
greatly improved when expression data are mapped onto protein
interaction maps (Taylor et al., 2009; Chuang et al., 2007). This is
particularly true when data integration focuses on proteins with
many interacting partners (‘‘hubs’’) as hubs are critical to
network conductivity (Fraser, 2005; Han et al., 2004). We there-
fore decided to combine transcriptome data garnered from a
variety of abiotic stresses with the TRAIN to see whether
dynamic patterns of network wiring emerge in response to
particular stresses.
Rather than simply overlaying the magnitude of expression of
TRAIN genes during a specific time or stress condition onto
Developmental Cell
ABA Signaling Networks in Plants
Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc. 365
(legend on next page)
Developmental Cell
ABA Signaling Networks in Plants
366 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
TRAIN nodes, we calculated a Pearson correlation coefficient
(PCC) of coexpression for each pair of interacting proteins within
the TRAIN, across multiple time points for both shoots and roots
upon individual stresses. A PCC value above 0.75 reflected
high coexpression between the partners, whereas values below
0.75 meant the gene pair is anticorrelated in expression. For
example, coexpression analysis based on osmotic stress
expression data from combined root and shoot tissues over 12
time points from AtGenExpress generated average PCC values
between PYL4 and its PP2C partners of 0.9766 (HAB1),
0.849 (ABI1), 0.8833 (HAI1), and 0.8378 (AHG3) (Kilian
et al., 2007). These highly negative correlations are consistent
with the diametrically opposed expression of these genes in
the presence of ABA.
To evaluate TRAIN coexpression analysis versus standard
representations of gene expression globally, we first generated
heatmap representations of TRAIN gene expression over the
same various abiotic stresses and times (Figure 5A). Osmotic,
salt, and cold stress all showed related patterns of gene expres-
sion, with osmotic and salt stresses being the most similar. How-
ever, although osmotic and salt stresses have similar heatmap
expression patterns, their TRAIN coexpression maps showed
clear differences (Figure 5B). The HAI1 hub, for example, was
highly correlated with the expression of many of its partners in
both osmotic and salt stress data sets. But the hub protein
MAP3kv4 only correlated well with its partners under osmotic
stress, whereas AIB1 highly correlated with its partners in salt
stress (Figure 5B). Neither MAP3Kvnor AIB1 have been exten-
sively studied, but these results suggest these genes may be
associated with these particular stresses.
To explore this possibility in detail, we examined the AIB1
subnetwork. The AIB1 hub represents 55 partners, many of
which show highly correlated expression with AIB1 under salt
stress conditions (Figure 6A; Table S1). Many of the AIB1 part-
ners that show salt-dependent coexpression are annotated as
having roles within the nucleus. After limiting partner interactions
to TFs, an AIB1-centered complex comprised of ATHB12 and a
collection of ANAC and MYB TFs emerged (Figure 6A). By com-
parison, similar coexpression analysis based on osmotic expres-
sion data did not predict these AIB1 complexes (Figure 6A). By
contrast, coexpression analysis identified an osmotic-related
complex consisting of ABI5, AFP2, and ATHB12, which were
not supported under salt stress conditions (Figure 6A). This anal-
ysis suggested different nuclear complexes may coalesce in
response to different abiotic stresses and that AIB1 might be a
key component of a transient nuclear complex that functions in
salt homeostasis. Unfortunately, loss-of-function mutations in
the AIB1 gene are not publically available to test its function.
Thus, we constructed DEX-inducible gain-of-function AIB1 lines
and tested them for osmotic- and salt-related phenotypes. The
presence of DEX did not influence the growth response of
AIB1 transgenic seedlings to increasing osmotic stress (Fig-
ure 6B). By contrast, DEX induction of AIB1 in transgenic plants
did confer an increased sensitivity to salt that was not observed
in the absence of DEX, which suggested AIB1 has a negative role
in salt homeostasis (Figure 6B). These results suggest that
changes in modularity as monitored by changes in coexpression
of protein partners may improve the predictive value of interac-
tomes for designing function experiments.
DISCUSSION
We have developed a systems-based strategy to define a
signaling landscape for the plant hormone ABA. Our approach
was centered on the premise that proteins contributing to the
ABA signaling network will often be coexpressed and should
interact at some point to transduce signaling information. Unlike
many ABA-based transcriptome studies, we limited our condi-
tions to low levels of ABA and a short duration of exposure to
ABA in order to enrich for primary transcriptional events. This
resulted in only a few hundred genes whose expression ap-
peared to be regulated mostly by core ABA signaling. This rela-
tively small number of genes allowed meticulous experimental,
statistical, and annotation analysis of our protein interaction
space to give a high-quality ABA-regulated network of over
500 interactions. Our approach confirmed many core ABA
signaling interactions and linked core components to over 50
additional protein partners, many of which represent various
transcriptional factor families. The mesoscale nature of the
TRAIN also allowed systematic functional analysis of specific
signaling modules. And finally, mapping coexpression correla-
tion onto the TRAIN allowed prediction of genes and signaling
modules that may play important roles in specific ABA-related
abiotic stresses.
Although the TRAIN is an attempt to define a comprehensive
ABA signaling network, there are several constraints to our
Figure 4. The SNRK3 Network
(A) The SNRK3.15-SNRK3.22 interaction network. Red connecting lines represent direct SNRK3.15 and SNRK3. 22 partners, and gray lines are other protein
interactions. Line thickness indicates the edge-weighted confidence based on statistical analysis. Node size is proportional to the total number of interactions in
the TRAIN data set.
(B) Summary of BiFC analysis of SNRK3.15 and SNRK3.22 interactions in planta. Green lines are Y2H interactions recapitulated by BiFC; blue lines are Y2H
interactions that were not recapitulated by BiFC analysis. See also Figures S3A–S3D.
(C) Summary of 2D gel analysis of TF mobility shifts in the presence of either SNRK3 in yeast. Green lines are SNRK3-dependent shifts, whereas blue lines denote
the absence of a shift. See also Figure S4A.
(D) In vitro phosphorylation by SNRK3s of a select number of TF targets. The band observed in the no TF lane is consistent with SNRK3 autophosphorylation. See
also Figure S4B.
(E) Loss- and gain-of-function analyses of SNRK3.15 and SNRK3.22 lines. Conditions were similar to those presented in Figure 3B. Quantification of germination
and cotyledon expansion of loss- and gain-of-function lines germinated on ABA are shown in the graph. Data are represented as mean ± SD.
(F) Transcriptome analysis of a DEX-inducible SNRK3.15 line in the presence of ABA. Left lane: genes are filtered for those that showed at least a 2-fold increase
(red) or decrease (yellow) in the absence of DEX. See also Figure S5 and Table S2. Right lane: expression of the 2-fold gene set in the presence of SNRK3.15
activation. If the expression of an ABA-induced gene is dampened, it becomes more yellow, whereas decreased expression of a repressed gene results in
increased red color. Dotted gray line represents 0-fold baseline, whereas the solid line represents the fold change for each gene.
Developmental Cell
ABA Signaling Networks in Plants
Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc. 367
approach. First, it is expected that ABA does not transcription-
ally regulate many important genes that contribute to its
signaling network. For example, the core SNRK2s involved in
ABA signaling are not regulated by ABA addition (Fujita et al.,
2011). With this in mind, we merged the TRAIN with literature-
curated protein interactions (Arabidopsis Interactome Mapping
Consortium, 2011; Lalonde et al., 2010; Popescu et al., 2009).
This eTRAIN doubled the number of interactions and contained
proteins that belong to over 30 GO-annotated functions. Inter-
estingly, although literature curation doubled the size of the inter-
actome, one gene, encoding a LEA protein (At1g65690), actually
accounted for over 200 of these additional interactions. Removal
of these LEA interactions meant that only 384 interactions out of
the 30,000 confirmed PPIs reported for Arabidopsis were added
to the TRAIN. Moreover, there was little interaction space over-
lap between the TRAIN and those published for high-throughput
studies. These differences may reflect the experimental limita-
tions that are inherent to high-throughput systems, and analysis
of high-throughput plant interactomes suggests these ap-
proaches have likely captured only 10% of the possible inter-
actions (Chen et al., 2010; Arabidopsis Interactome Mapping
Consortium, 2011). By contrast, the medium size of the TRAIN
space allows for multiple testing of interactions with different
vectors and reporter output systems, which has been shown to
be an essential benchmark for quality control in Y2H systems
(Chen et al., 2010). The finding that 70% of our Y2H interactions
could be recapitulated in planta further supports the experi-
mental and statistical approaches used here to build our inter-
action network. Whatever the case, the focused approach of
this study on a particular process such as hormone signaling is
essential in filling out the larger scale interaction maps built by
high-throughput methods.
A second limitation of our approach was the use of a specific
duration and amount of ABA added, thus limiting our transcrip-
tome analysis to only one stage of development. A recent
comprehensive comparison of ABA-regulated transcriptome
experiments found few genes in common among 14 data sets
(Wang et al., 2011). Although this most likely reflects the differ-
ences in ABA concentrations, tissues, and developmental
stages used in these experiments, it also demonstrates the flex-
ibility of ABA-related transcriptional response during the life
cycle of a higher plant. Thus, a comprehensive ABA network
Figure 5. Dynamic Modularity of the TRAIN in Response to Different Abiotic Stresses
(A) Heatmap representation of TRAIN gene expression in the shoot and root under different times of abiotic stress. The lower panel is a magnification of AIB1
expression.
(B) Network graph of the TRAIN with interactions shown as edges that are colored according to the PCC of coexpression of partner proteins. Edge color indicates
level of correlation between partner proteins, with bluer edges indicating more correlated coexpression and redder edges indicating more anticorrelation.
Developmental Cell
ABA Signaling Networks in Plants
368 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
will likely require a number of spatially and temporally based
‘‘TRAIN-like’’ interactomes. The TRAIN could also be reduced
to the resolution of a single cell, based on the integration of
cell-type specific interaction and expression data. This micro-
scopic level of analysis was not possible prior to the TRAIN
and will likely lead to valuable insight into the mechanisms of
how different cell types respond to ABA.
It is notable that functional analysis of the TRAIN uncovered
only negative regulators of ABA-mediated signaling. This could
mean the transcriptome assay conditions used to identify TRAIN
genes enriched for negative regulators. However, we think this
is unlikely because two-thirds of the genes curated from the liter-
ature that link to the core ABA signaling pathway and show ABA-
related phenotypes also appear to attenuate the ABA response.
This attenuation of the core ABA signaling pathway could reflect
a negative feedback response to the initial activation of the core
pathway by ABA.
With respect to the ABA signaling network, a large number of
kinases annotated as being involved in calcium-mediated
signaling appeared to interact with core ABA signaling compo-
nents and PP2Cs in particular. Within the TRAIN, there were
three major kinase hubs: a MAP3Kv4 and two SNRK3s.
Regarding SNRK3.15 and SNRK3.22, our analysis indicated
these kinases act as negative regulators of the ABA response,
which is opposite to the SNRK2 kinases involved in core ABA
signaling (Fujita et al., 2009; Fujii and Zhu, 2009). Interestingly,
it appears that a number of TFs interacting with SNRK3(s),
including TFs shown to be phosphorylated by either SNRK3.15
or SNRK3.22, were also negative regulators of the ABA
response. This could mean these kinases and a number of their
Figure 6. The AIB1 Salt and Osmotic Subnetworks
(A) Left: coexpression analysis of AIB1 partners under salt or osmotic stress. Middle: model of the AIB1 interaction network derived from TF partners. Edge color
indicates level of correlation between partner proteins, with bluer edges indicating more correlated coexpression and redder edges indicating more anti-
correlation. Right: model of nuclear AIB1 complexes based on coexpression of partners.
(B) Effects of mannitol (1%–5%) and NaCl (25–300 mM) on DEX-inducible AIB1 transgenic seedlings. Representative pictures are of seedlings grown in the
absence (DMSO) or presence of mannitol (5%) or salt (50 mM NaCl) for 7 days. Scale bar, 1.0 mm.
Developmental Cell
ABA Signaling Networks in Plants
Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc. 369
TF targets may form a signaling module. At this time, it is not
clear how these SNRK3s coordinate with core SNRK2s to modu-
late the overall ABA response. SNRK3 transcription is dependent
on SNRK2 activation; thus, each kinase group is temporally
separated. Furthermore, the absence of SNRK3 interactions
with either ABI5 or ABF3 suggests a separation between the
TFs regulated by SNRK2s in the core pathway and TFs targeted
by SNRK3s. Interestingly, a number of TF partners of SNRK3
interacted with the central ABA signaling transcriptional regu-
lator ABI5 (Lopez-Molina et al., 2001, 2002). Possibly, ABI5
may act as a common point of crosstalk between SNRK2 and
SNRK3 signaling.
Although little is known about the targets of SNRK3.15
signaling, SNRK3.22 does negatively regulate the activity of
the AHA2 plasma membrane H
+
-ATPase through phosphoryla-
tion (Fuglsang et al., 2007). The control of plasma membrane
H
+
-ATPase by SNRK3.22 regulates intracellular pH homeostasis
in response to alkaline pH, which in turn modulates the plant’s
overall response to salt stress under alkaline conditions. Apart
from its role in membrane-based salt homeostasis, we found
that one target of SNRK3.22, the bHLH TF AIB1, is also con-
nected to a salt stress response. Notably, many SNRK3.15
(24/26) and SNRK3.22 (22/23) interactors also partner with
AIB1 (Figure S6). Thus, SNRK3.22 and perhaps SNRK3.15 may
act as central coordinators in response to salt stress at both
the plasma membrane and the nucleus.
In conclusion, it is believed that most signaling modules are
mesoscale, encompassing 25–100 proteins (Spirin and Mirny,
2003). The experimental conditions used in this study, which
result in an interaction space around approximately 100 pro-
teins, were ideal in capturing a network of this scale. Our network
was simple enough to identify modules that could be probed
functionally. More importantly, unlike other plant studies that
identify subnetworks through functional analysis or gene expres-
sion signatures alone, the overlaying of coexpression onto the
TRAIN led to the identification of dynamic subnetworks based
on changes in network wiring. These changes in subnetworks
can be used to determine functional roles for hubs and specific
signaling modules. We believe this approach of merging coex-
pression onto protein interaction networks will have important
predictive value in assigning gene function. On this note, the
approaches used here should be readily applicable to other
mesoscale signaling processes such as pathogen infection or
growth on different nutrient conditions. These approaches would
also be advantageous in dissecting nonmodel plant and animal
processes where genetic tools are not readily available.
EXPERIMENTAL PROCEDURES
Plant Material and Growth Conditions
All Arabidopsis strains used in this study were of the ecotype Columbia (Col-0).
See Supplemental Experimental Procedures for details.
Construction of the Transcriptionally Regulated ABA Interactome
Network
Sterile seeds were placed on 1/23Murashige-Skoog (MS) plates and imbibed
for 4 days at 4C. Plates were transferred to room temperature under contin-
uous white light for 11 days. Eleven-day-old seedlings were moved to 1/23MS
plates (10 mM cycloheximide ± 1 mM ABA) for 6 hr. Total RNA was processed
and hybridized on a GeneChip Arabidopsis ATH1 genome array. Duplicate
samples were analyzed for each condition. See Supplemental Experimental
Procedures for construction of the TRAIN.
Protein Purification and Kinase Assays
Kinase reactions were conducted in 20 ml of kinase buffer with 1 mCi of g-[
32
P]
ATP for 15 min at room temperature and terminated by adding Laemmli buffer.
Reactions were loaded onto 12% SDS-PAGE gels, and incorporated radio-
label was visualized by autoradiography. Kinase assays were run on SDS-
PAGE gels and visualized by Coomassie blue staining. See Supplemental
Experimental Procedures for details.
Yeast Expression and 2D Gel Analysis
Streaked haploid EGY48 yeast expressing the various TFs in the pJG4-5
vector, or diploid yeast (EGY48/RFY206) coexpressing individual TFs and
kinases in the pJG4-5 and pEG202 vectors, respectively, were resuspended
from yeast nitrogen base (YNB) plates to 0.1 optical density 600 (OD
600
)in
3 ml of appropriate YNB galactose drop-out media to induce protein expres-
sion. Cultures were grown overnight at 30C. Pellets were solubilized with
glass beads in 200 ml of the rehydration buffer (Bio-Rad), and 75 mg of each
protein sample was absorbed into 7 cm Immobilized pH gradient (IPG)
(pH 3–10) strips and separated in the first dimension by the Protean IEF Cell
per the manufacturer’s instructions (Bio-Rad). The IPG strips were proce ssed,
and the proteins were separated in the second dimension by 12% SDS-P AGE.
Following electrophoresis, the TFs were detected by immunoblot analysis
using hemagglutinin antibodies (Roche; 1/15,000).
Bimolecular Fluorescence Complementation Analysis
BiFC and split yellow fluorescent protein (YFP) fusions were transiently
expressed using Agrobacterium tumefaciens in N. benthamiana as described
previously (Walter et al., 2004). The A. tumefaciens strain GV2260 (final density
of 0.2 OD
600
) was used to syringe-infiltrate N. benthamiana leaves. YFP was
visualized using standard techniques on a Leica TCS SP5 confocal
microscope.
Microarray Analysis of DEX-Inducible SNRK3.15 Lines
Stratified seeds were transferred to room temperature under continuo us white
light for 4 days and then transferred to either a 0.1% DMSO or a 30 mM DEX
plate (24 hr). After this incubation, seeds were transferred to 0.1% DMSO or
2mM ABA ± 30 mM DEX for 24 hr. RNA was analyzed by hybridization on a
GeneChip Arabidopsis ATH1 genome array. See Supplemental Experimental
Procedures.
Gene Expression Analysis
Gene expression data were generated from the AtGenExpress global stress
expression data set downloaded from the Bio-Analytic Resource (http://dx.
doi.org/10.1111/j.1365-313X.2005.02437.x) or from Gene Expression
Omnibus (Toufighi et al., 2005). The expression platform used in all cases
was the ATH1 GeneChip from Affymetrix. The data sets and comparisons
were as follows: aba2-2 seedlings one comparison: 6 hr with 1 mM ABA versus
mock-treated plant (http://bar.utoronto.ca/affydb/cgi-bin/affy_db_proj_
viewer.cgi?proj=26&view=general). Heatmaps of abiotic stress transcription
were generated using GENE-E (http://www.broadinstitute.org/cancer/
software/GENE-E/index.html). PCCs between each protein pair within the
TRAIN were generated over each abiotic stress using AtGenExpress global
stress expression data set expression data from both roots and shoots over
six time periods.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
six figures, and two tables and can be found with this article online at http://
dx.doi.org/10.1016/j.devcel.2014.04.004.
ACKNOWLEDGMENTS
We thank Dr. E. Nambara for helpful discussion. We also thank the Ohio State
Stock Center and Salk Institute Genomic Analysis Laboratory (SIGnAL) for
Developmental Cell
ABA Signaling Networks in Plants
370 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
providing materials, and Dr. N. Nishimura for ahg3-1 seed and Dr. Bernd Weis-
shaar for myb12 and myb12 myb11 myb111 triple mutant seed.
Received: October 18, 2013
Revised: February 21, 2014
Accepted: April 1, 2014
Published: May 12, 2014
REFERENCES
Arabidopsis Interactome Mapping Consortium (2011). Evidence for network
evolution in an Arabidopsis interactome map. Science 333, 601–607.
Batisti
c, O., Rehers, M., Akerman, A., Schlu
¨cking, K., Steinhorst, L., Yalovsky,
S., and Kudla, J. (2012). S-acylation-dependent association of the calcium
sensor CBL2 with the vacuolar membrane is essential for proper abscisic
acid responses. Cell Res. 22, 1155–1168.
Ben-Ari, G. (2012). The ABA signal transduction mechanism in commercial
crops: learning from Arabidopsis. Plant Cell Rep. 31, 1357–1369.
Brandt, B., Brodsky, D.E., Xue, S., Negi, J., Iba, K., Kangasja
¨rvi, J.,
Ghassemian, M., Stephan, A.B., Hu, H., and Schroeder, J.I. (2012).
Reconstitution of abscisic acid activation of SLAC1 anion channel by CPK6
and OST1 kinases and branched ABI1 PP2C phosphatase action. Proc.
Natl. Acad. Sci. USA 109, 10593–10598.
Chen, Y.C., Rajagopala, S.V., Stellberger, T., and Uetz, P. (2010). Exhaustive
benchmarking of the yeast two-hybrid system. Nat. Methods 7, 667–668,
author reply 668.
Chuang, H.Y., Lee, E., Liu, Y.T., Lee, D., and Ideker, T. (2007). Network-based
classification of breast cancer metastasis. Mol. Syst. Biol. 3, 140.
Coello, P., Hey, S.J., and Halford, N.G. (2011). The sucrose non-fermenting-1-
related (SnRK) family of protein kinases: potential for manipulation to improve
stress tolerance and increase yield. J. Exp. Bot. 62, 883–893.
Cramer, G.R., Urano, K., Delrot, S., Pezzotti, M., and Shinozaki, K. (2011).
Effects of abiotic stress on plants: a systems biology perspective. BMC
Plant Biol. 11, 163.
Cutler, S.R., Rodriguez, P.L., Finkelstein, R.R., and Abrams, S.R. (2010).
Abscisic acid: emergence of a core signaling network. Annu. Rev. Plant Biol.
61, 651–679.
Fraser, H.B. (2005). Modularity and evolutionary constraint on proteins. Nat.
Genet. 37, 351–352.
Fuglsang, A.T., Guo, Y., Cuin, T.A., Qiu, Q., Song, C., Kristiansen, K.A., Bych,
K., Schulz, A., Shabala, S., Schumaker, K.S., et al. (2007). Arabidopsis protein
kinase PKS5 inhibits the plasma membrane H+-ATPase by preventing interac-
tion with 14-3-3 protein. Plant Cell 19, 1617–1634.
Fujii, H., and Zhu, J.-K. (2009). Arabidopsis mutant deficient in 3 abscisic acid-
activated protein kinases reveals critical roles in growth, reproduction, and
stress. Proc. Natl. Acad. Sci. USA 106, 8380–8385.
Fujita, Y., Nakashima, K., Yoshida, T., Katagiri, T., Kidokoro, S., Kanamori, N.,
Umezawa, T., Fujita, M., Maruyama, K., Ishiyama, K., et al. (2009). Three
SnRK2 protein kinases are the main positive regulators of abscisic acid
signaling in response to water stress in Arabidopsis. Plant Cell Physiol. 50,
2123–2132.
Fujita, Y., Fujita, M., Shinozaki, K., and Yamaguchi-Shinozaki, K. (2011). ABA-
mediated transcriptional regulation in response to osmotic stress in plants.
J. Plant Res. 124, 509–525.
Garcia, M.E., Lynch, T., Peeters, J., Snowden, C., and Finkelstein, R. (2008). A
small plant-specific protein family of ABI five binding proteins (AFPs) regulates
stress response in germinating Arabidopsis seeds and seedlings. Plant Mol.
Biol. 67, 643–658.
Ge, H., Liu, Z., Church, G.M., and Vidal, M. (2001). Correlation between tran-
scriptome and interactome mapping data from Saccharomyces cerevisiae.
Nat. Genet. 29, 482–486.
Geiger, D., Maierhofer, T., Al-Rasheid, K.A., Scherzer, S., Mumm, P., Liese, A.,
Ache, P., Wellmann, C., Marten, I., Grill, E., et al. (2011). Stomatal closure by
fast abscisic acid signaling is mediated by the guard cell anion channel
SLAH3 and the receptor RCAR1. Sci. Signal. 4, ra32.
Han, J.D., Bertin, N., Hao, T., Goldberg, D.S., Berriz, G.F., Zhang, L.V., Dupuy,
D., Walhout, A.J., Cusick, M.E., Roth, F.P., and Vidal, M. (2004). Evidence for
dynamically organized modularity in the yeast protein-protein interaction
network. Nature 430, 88–93.
Hartwell, L.H., Hopfield, J.J., Leibler, S., and Murray, A.W. (1999). From molec-
ular to modular cell biology. Nature 402 (6761, Suppl), C47–C52.
Hey, S.J., Byrne, E., and Halford, N.G. (2010). The interface between metabolic
and stress signalling. Ann. Bot. (Lond.) 105, 197–203.
Huang, G.T., Ma, S.L., Bai, L.P., Zhang, L., Ma, H., Jia, P., Liu, J., Zhong, M.,
and Guo, Z.F. (2012). Signal transduction during cold, salt, and drought
stresses in plants. Mol. Biol. Rep. 39, 969–987.
Hubbard, K.E., Nishimura, N., Hitomi, K., Getzoff, E.D., and Schroeder, J.I.
(2010). Early abscisic acid signal transduction mechanisms: newly discovered
components and newly emerging questions. Genes Dev. 24, 1695–1708.
Kilian, J., Whitehead, D., Horak, J., Wanke, D., Weinl, S., Batistic, O., D’Angelo,
C., Bornberg-Bauer, E., Kudla, J., and Harter, K. (2007). The AtGenExpress
global stress expression data set: protocols, evaluation and model data anal-
ysis of UV-B light, drought and cold stress responses. Plant J. 50, 347–363.
Lalonde, S., Sero, A., Pratelli, R., Pilot, G., Chen, J., Sardi, M.I., Parsa, S.A.,
Kim, D.Y., Acharya, B.R., Stein, E.V., et al. (2010). A membrane protein/
signaling protein interaction network for Arabidopsis version AMPv2. Front.
Physiol. 1,24.
Le
´on-Kloosterziel, K.M., Gil, M.A., Ruijs, G.J., Jacobsen, S.E., Olszewski, N.E.,
Schwartz, S.H., Zeevaart, J.A., and Koornneef, M. (1996). Isolation and char-
acterization of abscisic acid-deficient Arabidopsis mutants at two new loci.
Plant J. 10, 655–661.
Li, H.H., Hao, R.L., Wu, S.S., Guo, P.C., Chen, C.J., Pan, L.P., and Ni, H. (2011).
Occurrence, function and potential medicinal applications of the phytohor-
mone abscisic acid in animals and humans. Biochem. Pharmacol. 82,
701–712.
Lopez-Molina, L., Mongrand, S., and Chua, N.H. (2001). A postgermination
developmental arrest checkpoint is mediated by abscisic acid and requires
the ABI5 transcription factor in Arabidopsis. Proc. Natl. Acad. Sci. USA 98,
4782–4787.
Lopez-Molina, L., Mongrand, S., McLachlin, D.T., Chait, B.T., and Chua, N.H.
(2002). ABI5 acts downstream of ABI3 to execute an ABA-dependent growth
arrest during germination. Plant J. 32, 317–328.
Lumba, S., Cutler, S., and McCourt, P. (2010). Plant nuclear hormone recep-
tors: a role for small molecules in protein-protein interactions. Annu. Rev.
Cell Dev. Biol. 26, 445–469.
Lynch, T., Erickson, B.J., and Finkelstein, R.R. (2012). Direct interactions of
ABA-insensitive(ABI)-clade protein phosphatase(PP)2Cs with calcium-depen-
dent protein kinases and ABA response element-binding bZIPs may contribute
to turning off ABA response. Plant Mol. Biol. 80, 647–658.
Nakamura, S., Lynch, T.J., and Finkelstein, R.R. (2001). Physical interactions
between ABA response loci of Arabidopsis. Plant J. 26, 627–635.
Nakashima, K., and Yamaguchi-Shinozaki, K. (2013). ABA signaling in stress-
response and seed development. Plant Cell Rep. 32, 959–970.
Ohta, M., Guo, Y., Halfter, U., and Zhu, J.-K. (2003). A novel domain in the
protein kinase SOS2 mediates interaction with the protein phosphatase 2C
ABI2. Proc. Natl. Acad. Sci. USA 100, 11771–11776.
Park, S.Y., Fung, P., Nishimura, N., Jensen, D.R., Fujii, H., Zhao, Y., Lumba, S.,
Santiago, J., Rodrigues, A., Chow, T.F., et al. (2009). Abscisic acid inhibits type
2C protein phosphatases via the PYR/PYL family of START proteins. Science
324, 1068–1071.
Popescu, S.C., Popescu, G.V., Bachan, S., Zhang, Z., Gerstein, M., Snyder,
M., and Dinesh-Kumar, S.P. (2009). MAPK target networks in Arabidopsis
thaliana revealed using functional protein microarrays. Genes Dev. 23, 80–92.
Qin, Y., Li, X., Guo, M., Deng, K., Lin, J., Tang, D., Guo, X., and Liu, X. (2008).
Regulation of salt and ABA responses by CIPK14, a calcium sensor interacting
protein kinase in Arabidopsis. Sci. China C Life Sci. 51, 391–401.
Developmental Cell
ABA Signaling Networks in Plants
Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc. 371
Spirin, V., and Mirny, L.A. (2003). Protein complexes and functional modules in
molecular networks. Proc. Natl. Acad. Sci. USA 100, 12123–12128.
Stelzl, U., Worm, U., Lalowski, M., Haenig, C., Brembeck, F.H., Goehler, H.,
Stroedicke, M., Zenkner, M., Schoenherr, A., Koeppen, S., et al. (2005). A
human protein-protein interaction network: a resource for annotating the
proteome. Cell 122, 957–968.
Stracke, R., Jahns, O., Keck, M., Tohge, T., Niehaus, K., Fernie, A.R., and
Weisshaar, B. (2010). Analysis of production of flavonol glycosides-dependent
flavonol glycoside accumulation in Arabidopsis thaliana plants reveals MYB11-
, MYB12- and MYB111-independent flavonol glycoside accumulation. New
Phytol. 188, 985–1000.
Taylor, I.W., Linding, R., Warde-Farley, D., Liu, Y., Pesquita, C., Faria, D., Bull,
S., Pawson, T., Morris, Q., and Wrana, J.L. (2009). Dynamic modularity in
protein interaction networks predicts breast cancer outcome. Nat.
Biotechnol. 27, 199–204.
Toufighi, K., Brady, S.M., Austin, R., Ly, E., and Provart, N.J. (2005). The
Botany Array Resource: e-Northerns, Expression Angling, and promoter
analyses. Plant J. 43, 153–163.
Valde
´s, A.E., Overna
¨s, E., Johansson, H., Rada-Iglesias, A., and Engstro
¨m, P.
(2012). The homeodomain-leucine zipper (HD-Zip) class I transcription factors
ATHB7 and ATHB12 modulate abscisic acid signalling by regulating protein
phosphatase 2C and abscisic acid receptor gene activities. Plant Mol. Biol.
80, 405–418.
Van Leene, J., Boruc, J., De Jaeger, G., Russinova, E., and De Veylder, L.
(2011). A kaleidoscopic view of the Arabidopsis core cell cycle interactome.
Trends Plant Sci. 16, 141–150.
Walter, M., Chaban, C., Schu
¨tze, K., Batistic, O., Weckermann, K., Na
¨ke, C.,
Blazevic, D., Grefen, C., Schumacher, K., Oecking, C., et al. (2004).
Visualization of protein interactions in living plant cells using bimolecular fluo-
rescence complementation. Plant J. 40, 428–438.
Wang, R.S., Pandey, S., Li, S., Gookin, T.E., Zhao, Z., Albert, R., and Assmann,
S.M. (2011). Common and unique elements of the ABA-regulated transcrip-
tome of Arabidopsis guard cells. BMC Genomics 12, 216.
Wasilewska, A., Vlad, F., Sirichandra, C., Redko, Y., Jammes, F., Valon, C.,
Frei dit Frey, N., and Leung, J. (2008). An update on abscisic acid sig naling
in plants and more. Mol. Plant 1, 198–217.
Wilkinson, S., Kudoyarova, G.R., Veselov, D.S., Arkhipova, T.N., and Davies,
W.J. (2012). Plant hormone interactions: innovative targets for crop breeding
and management. J. Exp. Bot. 63, 3499–3509.
Yu, H., Braun, P., Yildirim, M.A., Lemmens, I., Venkatesan, K., Sahalie, J.,
Hirozane-Kishikawa, T., Gebreab, F., Li, N., Simonis, N., et al. (2008). High-
quality binary protein interaction map of the yeast interactome network.
Science 322, 104–110.
Zeller, G., Henz, S.R., Widmer, C.K., Sachsenberg, T., Ra
¨tsch, G., Weigel, D.,
and Laubinger, S. (2009). Stress-induced changes in the Arabidopsis thaliana
transcriptome analyzed using whole-genome tiling arrays. Plant J. 58, 1068–
1082.
Zhang, L.V., King, O.D., Wong, S.L., Goldberg, D.S., Tong, A.H., Lesage, G.,
Andrews, B., Bussey, H., Boone, C., and Roth, F.P. (2005). Motifs, themes
and thematic maps of an integrated Saccharomyces cerevisiae interaction
network. J. Biol. 4,6.
Developmental Cell
ABA Signaling Networks in Plants
372 Developmental Cell 29, 360–372, May 12, 2014 ª2014 Elsevier Inc.
