A system biology approach highlights a hormonal enhancer effect on regulation of genes in a nitrate responsive "biomodule"
© Nero et al. 2009
Received: 26 March 2009
Accepted: 06 June 2009
Published: 06 June 2009
Skip to main content
© Nero et al. 2009
Received: 26 March 2009
Accepted: 06 June 2009
Published: 06 June 2009
Nitrate-induced reprogramming of the transcriptome has recently been shown to be highly context dependent. Herein, a systems biology approach was developed to identify the components and role of cross-talk between nitrate and hormone signals, likely to be involved in the conditional response of NO3 - signaling.
Biclustering was used to identify a set of genes that are N-responsive across a range of Nitrogen (N)-treatment backgrounds (i.e. nitrogen treatments under different growth conditions) using a meta-dataset of 76 Affymetrix ATH1 chips from 5 different laboratories. Twenty-one biclusters were found to be N-responsive across subsets of this meta-dataset. N-bicluster 9 (126 genes) was selected for further analysis, as it was shown to be reproducibly responsive to NO3 - as a signal, across a wide-variety of background conditions and datasets. N-bicluster 9 genes were then used as "seed" to identify putative cross-talk mechanisms between nitrate and hormone signaling. For this, the 126 nitrate-regulated genes in N-bicluster 9 were biclustered over a meta-dataset of 278 ATH1 chips spanning a variety of hormone treatments. This analysis divided the bicluster 9 genes into two classes: i) genes controlled by NO3 - only vs. ii) genes controlled by both NO3 - and hormones. The genes in the latter group showed a NO3 - response that is significantly enhanced, compared to the former. In silico analysis identified two Cis-Regulatory Elements candidates (CRE) (E2F, HSE) potentially involved the interplay between NO3 - and hormonal signals.
This systems analysis enabled us to derive a hypothesis in which hormone signals are proposed to enhance the nitrate response, providing a potential mechanistic explanation for the link between nitrate signaling and the control of plant development.
Higher plants acquire nitrogen mainly as NO3 -. The soil concentration of this mineral ion can fluctuate dramatically in the rhizosphere, often resulting in limited growth and yield . Thus, nitrate signaling constitutes a key point of plant adaptation to environment. This is why nitrate signaling has so far been intensively studied by transcriptomic assays, involving more than 75 ATH1 chips in various background conditions and treatments. Taken together these transcriptomic data showed that NO3 --responses are very context dependent [2, 3], suggesting that evolution probably built very adaptable and robust networks involved in the integration of NO3 - with other signals including light, sugar, and hormones. For instance, as sessile organisms, plants have developed a strong capacity to modulate growth according to nutrient availability. On a molecular scale, this coordination between nutrition and growth can be mediated by the co-control of metabolism and hormonal signaling. For instance, a recent work reports that molecular reprogramming induced by nutritional starvation treatments significantly involve hormone regulated genes . Moreover, it has also been shown that such cross-controls exist between NO3 - and: cytokinin (for review see ), auxin [2, 6, 7], and ABA . To date, molecular players underlying those events are still under investigation. One striking example of such coordination at a molecular level is presented by the role of the iso-pentenyl-transferase 3 (IPT3) involved in the critical step of cytokinin biosynthesis. Transcription of the NO3 - induced gene IPT3 has been shown to be involved in the production of NO3 - induced cytokinins, hypothesized to coordinate shoot growth in response to NO3 - provision [9–14].
Root architecture is also under the coordinated control of nutrient availability and hormone signaling . For instance, NO3 - controls root branching under various pathways (for review see [16, 17]). Hormones have been shown to play important roles in the adaptation of root development to NO3 - availability. Indeed, NO3 - triggers root colonization in NO3 - rich patch of the soil. Zhang et al  have shown that this adaptation could involve AXR4, a gene initially demonstrated to be involved in auxin signaling. Later, AXR4 was shown to be involved in targeting the auxin influx transporter AUX1 to the plasma membrane . Thus AXR4 may provide a molecular link between the NO3 - signal and auxin signaling through regulating auxin transport. Furthermore, the dual affinity (high and low affinity NO3 - uptake) NO3 - transporter NRT1.1/CHL1, hypothesized to be a part of the NO3 - sensing system [20–23], was previously shown to be regulated by auxin . This evidence uncovers one facet of how the NO3 - sensing system is likely tuned by a hormonal/growth signal.
The complexity of the NO3 - effect on root development is further complicated by the fact that high NO3 - concentrations (50 mM) trigger an almost complete repression of the lateral root development (LRD). Abscisic acid (ABA) seems to be required for this effect, since the NO3 - inhibitory effect on LRD is reduced by mutating either the ABI4 or ABI5 genes .
In a previous study, meta-analysis of transcriptomes of NO3 - treated plants revealed that gene responses to nitrogen were very context-dependent, and only a very small number of core genes are regulated by NO3 - in a context-independent manner . The underlying rules of such coordination/context dependence between signals had recently been proposed at a genome wide level concerning the interaction of carbon, nitrate, and light . Moreover, in light of the context dependent nature of the N-response, mono-dimension clustering algorithms will miss genes that are co-regulated by N across a subset of treatment conditions. By contrast, an approach also known for decades  called biclustering can be used to identify nitrate responsive genes that are co-regulated, as a group, in response to a subset of nitrogen treatments across a matrix of meta-data (Figure 1) , likely susceptible to tackle the context dependence response to NO3 -. Thus, detected biclusters are subsets of the studied genes exhibiting consistent patterns over a subset of N-treatment conditions. Such sets of genes would not be found using mono-dimension clustering approaches, which require that the genes in the cluster behave the same across all treatments. We used this biclustering method to analyze five microarray data sets from N-treatments of Arabidopsis generated by three different laboratory groups: the Crawford lab [26, 27] (16 Affymetrix chips including controls), the Stitt lab  (14 Affymetrix chips including controls) and the Coruzzi lab [29, 30] (46 Affymetrix chips including controls). This combined meta-data set resulting from N-treatments corresponded to a total of 76 microarray chips with controls [For details see Additional File 1].
Over-represented functional categories in N-bicluster 9
Observed Frequency %
Expected Frequency %
metabolism of energy reserves (e.g. glycogen, trehalose)
glycolysis and gluconeogenesis
C-compound and carbohydrate metabolism
amino acid metabolism
assimilation of ammonia, metabolism of the glutamate group
nitrogen and sulfur metabolism
TRANSPORTED COMPOUNDS (SUBSTRATES)
NO3 -/Hormone Response Interaction.
Description of Comparison
N/H-bicluster 1 vs. Exclusive +N/H-bicluster 6
N/H-Bicluster 16, 19, 20 vs. Exclusive + N/H-bicluster 6
Treated vs. Control
N/H-Bicluster 16, 19, 20: Treated
The analysis of the N/Hormone biclusters (biclusters 1, 6, 16, 19, 20) revealed that cytokinin and ABA are the main hormone treatments under which the NO3 - regulated genes from N-bicluster 9 are co-regulated (Figure 3 and 5). N/H-biclusters 1 and 19 are both mainly driven by ABA treatments, although their respective regulation is in opposing directions (induced vs. repressed). N/H-biclusters 6, 16, and 20 are comprised of genes almost exclusively regulated in response to cytokinin treatment (Figure 5). Together, our results suggest that the coordinated regulation of these genes to nitrate as well as cytokinin or ABA may be part of a regulatory network that mediates the responsiveness of these genes.
As all N/H-biclusters were derived from the 126 genes contained in N-bicluster 9, we used a modified version of BioMaps analysis to determine which if any of the five selected N/H-biclusters were enriched for specific MIPS functional categories (see Methods). This analysis demonstrated that N/H-biclusters 1, 16 and 19 had at least one over-represented MIPS category, when compared to N-bicluster 9 [see Additional File 1]. The most significantly over-represented categories from these N/H-biclusters are genes involved in metabolic pathways, suggesting that this NO3 -/Hormone "crosstalk" may be directed towards the coordinate regulation of genes in interconnected metabolic pathways (see Network View of N-bicluster 9). Further, genes from N/H-bicluster 1 have several additional categories over-represented including Energy, Pentose phosphate pathway and Photosynthesis.
To quantify and statistically validate the regulation and nitrate-responsiveness of the genes in the N/H biclusters in the NR double mutant data set (Figure 6), we modeled the expression of genes from these groups using the lm, summary.lm and ANOVA functions in [R] . In this analysis the gene-expression response variable was modeled as a function of 4 explanatory-variable factors: i) Treatment, with 2 levels (nitrate treatment and control); ii) Tissue, with 2 levels (roots and shoots); iii) Genotype, with 2 levels (mutant (NR double mutant) and wild-type); iv) N/H-bicluster, with 6 levels (N/H bicluster 1, 6, 16, 19, 20 and N-bicluster 9 exclusive). To avoid any ambiguity between factor levels, overlapping genes from H-biclusters were removed from the analysis. The response variable (signal values) were taken from the normalized MAS5 data from the Wang et al, dataset  (These data were used to build Figure 6). In our ANOVA analysis, we started with an initial model that included main effects for each of the factors and an interaction term for the Treatment and N/H-bicluster factors. We simplified the model systematically in a step-wise procedure as outlined in Crawley  and fully described in Additional File 2. Briefly, our results from ANOVA analysis showed that the main effects of Tissue and Genotype factors were not significant (p-values of 0.13 and 0.94, respectively). Further it revealed that the factor levels of N/H-biclusters were not all significantly different from each other. Specifically NH-bicluster 16 19, and 20, in one hand, and 6 and N-bicluster 9 exclusive, in the other hand, were not significantly different from each other so that these levels could be combined into a single compound level. The final result of our simplification procedure was a model with main effects of Treatment (with 2 levels), N/H bicluster (with 3 compound levels), and an interaction term between N/H bicluster and Treatment. The R code and output for this model simplification are fully available in Additional File 2.
Using this final model, we were able to show that the N/H bicluster 1 level and the compound level for N/H biclusters 16, 19 and 20 are both significantly different from the compound level of N/H bicluster 6 and N-bicluster 9 exclusive in both having a stronger baseline response and a stronger response to nitrate (see Figure 6). Taken together our combined bicluster analysis data from N-treatment and hormone-treatment meta-datasets, leads us to propose a new hypothesis that hormone signaling (specifically ABA, and/or cytokinins which are represented in N/H biclusters 1, 16, 19 and 20) may act as an enhancer of NO3 - signaling/induction.
Over-represented known CREs in N-bicluster 9.
Cis Regulatory Element
% Present in Whole Genome
% Present in N-bicluster 9
Bellringer/replumless/pennywise BS1 IN AG
ATB2/AtbZIP53/AtbZIP44/GBF5BS in ProDH
AtMYB2 BS in RD22
Over-represented CREs for 4 significant N/H-biclusters.
P-value (N/H-Bicluster vs. N-bicluster 9 Exclusive)
% Present in N/H-bicluster Genes
% Present in N-bicluster 9 Exclusive Genes
% Present in N-bicluster 9 Genes
In a previous meta-analysis of nitrate-regulated genes, we demonstrated that a very small number of genes are nitrate regulated across a variety of background conditions, while the vast number of nitrate-regulated genes are regulated in a context-dependant manner . This observation suggests that the NO3 - signaling pathway is also under the influence of other (as yet) unidentified controls. Taking this observation as a starting point, we decided to use biclustering technique: an approach that clusters both genes and treatments, as a tool to discover genes that are co-regulated by nitrate across a wide variety of background conditions corresponding to a subset of the meta-data analysis. This biclustering approach allowed us to uncover a "biomodule" of 126 NO3 - regulated genes that are related in expression pattern and in biological function (N-bicluster 9, Figures 2 and 3, Additional file 1, table s2). Indeed, as a group, the genes in N-bicluster 9 comprise a set of 52 metabolic genes including, for example, all steps in the pathway of nitrate uptake & reduction (NRT1.1, NRT2.1, NRT3.1, NIA1, NIA2, NIR), as well as genes involved in N-assimilation into organic form (GDH1, ASN2 and GLT1). In addition, the N-bicluster 9 also contains significant overrepresentation of genes involved in Energy, Nitrogen and Carbon metabolism (Table 2). This strong functional coherence of the genes in N-bicluster 9 is illustrated by the interactions between 52 genes in the metabolic/protein interactions shown in the subnetwork (Figure 3).
It is noteworthy that the concept of the 126 genes in bicluster 9 constituting a "biomodule" in our study is comparable to ideas that have been already developed by others in the field of systems biology. For instance, i) Baliga et al.  state that "a biomodule is a group of proteins that execute a particular function", and ii) Bonneau et al.  also used a biclustering approach (cMonkey) to define "biologically meaningful biclusters". The conjunction of both above definitions match our concept/definition of a "biomodule".
As an insight into potential TFs that regulate the genes in this network, it is noteworthy that N-regulated bicluster 9 contains 17 transcription factors (based on AGRIS transcription factor annotation) whose regulation is by definition correlated with targets in N-bicluster 9, as well as with genes from other functional and unknown categories. N-bicluster 9 also contains other regulatory genes potentially involved in signal transduction such as kinases or phosphatases (6% of the genes from this N-bicluster fall into this category) (Table 1).
To identify potential regulatory mechanisms for the NO3 - responses of these genes to by other stimuli, we examined the regulation of the 126 genes in N-bicluster 9 across a metadata set of hormone microarrays. This was done in order to try to understand whether these genes, or subsets of these genes, are coregulated by hormones as well. Hormone treatments have been previously shown to have strong interactions with nitrogen signaling . This analysis identified a subset of 77 genes in N-bicluster 9 that also cluster together across a subset of hormone treatments. The position of these 77 genes present in the N/Hormone biclusters are shown in the context of the metabolic/protein interaction network presented in Figure 3. This view demonstrates a strong potential effect of diverse hormonal controls on the level of response of NO3 --controlled metabolic processes [See Figure 3 color coding for nodes: Red squares = significant N/H-bicluster genes (genes from N/H-biclusters 1, 16, 19 and 20), Blue squares = genes controlled only by NO3 - and not co-regulated by hormone treatments, based on results of hormone biclustering, Grey squares = non-significant N/H-bicluster genes controlled by hormones (i.e. no reproducible hormone response in N/H biclusters, see Methods)]. This hormonal control of nitrate-regulated genes represents a potential mechanism to fine tune and co-ordinate response levels of genes in a biomodule so that metabolic processes (here N-assimilation, carbon metabolism, and signaling components) can be regulated according to the growth rate of the plant. This is consistent with the observations made for phosphate [52, 53], sugar , sulfate , and iron metabolism . In all of these previous studies, when the hormone receptor is mutated, the response of genes to the nutrient under investigation is maintained, but the hormone response of the same genes is abrogated. This implies that hormonal control of nutrition pathways has a broad effect and controls metabolism as a whole, and is distinct from nutrient signaling. Our current work supports this view and also goes a step further. Indeed, our systems approach has enabled us to derive the hypothesis that hormone signals can interact with NO3 - signals to enhance the responsiveness of genes, and we have performed and in silico test of this hypothesis. This hypothesis is based on the finding that genes controlled by NO3 - only, were shown to be less responsive to NO3 - than genes under the control of NO3 - and hormones (Figure 6, Table 2). This kind of interaction has to our knowledge never been reported, and is a particularly novel aspect of for the effect of hormones as they relate to NO3 - induction. Although the effect of external hormone supply on genes belonging to NO3 - assimilation pathway or sensing system has already been documented, our results propose a new dimension of interaction at the transcriptional level between hormonal and NO3 - signaling. The existence of specific links between different nutrient and hormonal signals reported herein is also of particular interest and deserves further investigation.
Our study has identified two putative regulatory elements that are over-represented in the four significant N/H-biclusters identified by ANOVA (Table 2). To identify the potential role of such elements in mediating the hormone enhancing effect on nitrate responsiveness, we performed an in silico analysis aiming at deciphering the potential effect of each candidate binding site. By removing all genes from N-bicluster 9 exclusive gene list that contained these CREs (E2F and HSE), we were able to "virtually" examine their respective role in the enhancement of the baseline and NO3 - response by comparing these genes to genes from significant N/H-biclusters 1, 6, 16 and 20. The analysis demonstrated that E2F and HSE CREs are potentially involved in the hormonal enhancing effect of expression of these NO3 - responsive genes (Table 4, 5, and 6). To date, the heat shock elements (HSE) were not shown to be involved in the control of N-regulated genes though their role in Arabidopsis in the transcriptional control of responses to heat stress has been extensively studied . However, a heat shock transcription factor HsfA9 has been shown to be under hormonal control in seeds (ABA through ABI3) . This observation leads to the tentative hypothesis that heat shock elements could potentially be involved in conveying a hormonal signal. Moreover, to further have insight into the HSF/hormonal connection we ran a Sungear  analysis to decipher if these factors are under any other hormonal controls. To do so, we queried gene annotation for HSF term. We found 21 HSF, and looked to see if they were found regulated by any hormone as reported by Nemhauser et al. . Out of the 21 HSF detected we found that 6 (28%) are regulated by ABA (2 of which are also regulated by methyl-jasmonate), and 1 gene is regulated by cytokinins. This kind of co-regulation might further support the potential connection between HSF and hormonal signals.
E2F binding elements and the role of their associated transcription factors are still poorly understood in plants. However, what is known in plants as well as in other organisms is that these factors (considered in animals as oncogenes) are involved in the control of the cell cycle [61, 62]. Remarkably the role of E2F in the control of gene expression related to N-assimilation has already been shown in Arabidopsis, providing an independent validation of our results. Vlieghe et al.  demonstrated that the over-expression of the E2Fa-DPa transcription factor leads to the induction of nitrate reductase (NIA2), glutamine synthetase (GS), glutamate synthase (GOGAT), and nitrite reductase (NIR) gene. It is noteworthy that all of these genes respond to both nitrate and hormonal signals in our analysis (Figure 3). Furthermore, several genes in the N/H biclusters that are involved in C-metabolism are also mis-regulated in plants over-expressing E2Fa-DPa. Interestingly, E2F CREs were also identified in the nitrate reductase promoter of the green algae, Chlorella vulgaris. The protein binding activity at this site was validated but was not dependent on nitrate in the media . This confirms the idea that E2F CREs are involved in the interaction of the NO3 - response with other signals such as hormones and may be mediating crosstalk between these signals. Finally, the cell cycle is known to be an important target of hormonal signaling. For instance the Arabidopsis E2FC-DPB transcription factor was demonstrated to be involved in the control of the cell cycle. Also, cell division (monitored by CYCB1-GUS) in plants over-expressing E2FC-DPB was found to be less sensitive to auxin than cell division in wild types plants. This supports the hypothesis that E2F transcription factors are involved in mediating hormonal control of cell division .
In conclusion, our results suggest and highlight a significant level of control of NO3 - signaling by hormones. This control may allow plants to modulate biomodules of genes spanning N and C metabolism according to growth-dependant hormone signals. The systems biology approach presented herein demonstrates the inference of relationships between signals a postriori using extensive microarray data sets (76 chips for Nitrogen + 278 chips for hormones) to uncover new hypotheses for mechanisms underlying the much studied but poorly understood interactions between nutrient and hormone signaling. This in silico approach opens the door toward unraveling new biological concepts by systems analysis of existing microarray and other genome scale data sets within the public domain.
Expression values for all genes within the Arabidopsis genome present on the Affymetrix chip were taken from published data on nitrogen treatments vs. controls for all the available experiments from the data sets published in:[26–28, 30]. All microarray data used in this analysis was processed and normalized using Affymetrix Suite 5.0 or MAS5 Software (as implemented in the R statistical package  the two normalization Methods gave equivalent results. For biclustering analysis (see below), signal values were converted to log base 2 ratios with the treatment condition compared to its relative baseline condition (control). Genes with raw signal values less 100 in their treatment or control conditions in either replicate had their signal log ratio values replaced with a non-numerical NA value which is ignored by the biclustering algorithm. Finally, Log 2 ratio data from the microarray data was analyzed to determine which genes in the genome were greater then 1.5-fold responsive in any pair of replicate experiments in this meta data set. The resulting list of 3,752 N-responsive genes was used for biclustering, as described below.
All hormone microarray data was taken from the MAS5-processed NASC Microarray database (Nottingham Arabidopsis Stock Center ) data. Data was chosen based on annotation and experimental conditions that referred to a hormone or hormone inhibitor treatment vs. a relative control, with only replicated data used for biclustering analysis. A total of 19 data sets comprising 278 microarray experiments were compiled based on these criteria. The full list of data sets and the contributing number of microarrays from each data set is provided in Additional File 1. All hormone data with signal values of < 100 had their signal values replaced with a non-numerical NA value. Further, all data was converted to log base 2 ratios prior to biclustering analysis. The 126 genes from N-bicluster 9 were biclustered over all hormone data as described (see below).
Biclustering was performed using the SAMBA algorithm as described by  and as implemented in the CLICK and EXPANDER program . Biclustering was performed using default parameters except as follows: amount of overlap allowed (50%), gene coverage (set to cover all genes) and the number of genes expected (set to maximum number of genes in the data set). Biclusters that were used for further analysis were chosen based on genes being ≥ 1.5 fold regulated across reproducible experiments and the presence of replicate experiments for ≥ 50% of the experiments.
BioMaps analysis of N-biclusters was performed as described in  as accessed via http://www.virtualplant.org. The program was run using the MIPS  annotation option for functional definitions. A 5% FDR (False Discovery Rate) cut-off was computed using the R statistical package  to determine significant p-values.
As N/H-biclusters were derived from N-bicluster 9, over-representation of MIPS  functional terms were determined using the Fisher's Exact Test as implemented in the R statistical package  to compare the proportions of genes from N/H-biclusters containing a MIPS term vs. the proportion of genes from N-bicluster 9 not in N/H biclusters containing that same term.
To understand the relationships among the 126 genes from N-bicluster 9, the Arabidopsis multinetwork analysis tool was used  as accessed by http://www.virtualplant.org. This network contains many validated connections for gene interactions in the Arabidopsis genome. Network interactions were visualized using Cytoscape .
An FDR control method was used to determine a significance cutoff for p-values. This value was computed using a script written for the R statistical package . This script was derived from the Storey and Tibshirani method  which determines a cut-off based on the expected proportion of false positives incurred when calling a feature significant.
In order to detect known CREs that may be over-represented in a group of genes (e.g. N-bicluster 9, N/H-biclusters), sequence analysis of the promoter regions of these genes was performed. We used Cis Regulatory Element (CRE) annotation from the AGRIS Database (Arabidopsis Gene Regulatory Information Server ) as well as our own literature search to identify biologically active CREs that have been validated by in vivo experimentation. CRE detection was performed using the DNA pattern search tool available from RSA Tools  upon 3,000 bp of parsed upstream promoter region (taken from the AGRIS database).
The test for over-representation of CREs was performed using Fisher's Exact Test. This test compared the proportion of promoters in which a particular CRE of interest appeared in one group with the proportion of these same CREs detected in another group. A p-value cutoff was computed using a 5% FDR cut-off for significance.
We thank Sandrine Ruffel, and Miriam Gifford for helpful discussion and critical reading of the manuscript. We thank Dennis Shasha for his help in determining the statistical significance of the connectivity of multinetworks. This work was funded by NIH NIGMS GRANT GM032877 to G.C., an NIH Pre-doctoral minority fellowship GM032877-S1 to D.N., and an NSF Arabidopsis 2010 Genome Grant (IOB 0519985) to G.C.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.