Arabidopsis Transcriptome Reveals Control Circuits Regulating Redox Homeostasis and the Role of an AP2 Transcription Factor^,,

Abstract

Sensors and regulatory circuits that maintain redox homeostasis play a central role in adjusting plant metabolism and development to changing environmental conditions. We report here control networks in Arabidopsis (Arabidopsis thaliana) that respond to photosynthetic stress. We independently subjected Arabidopsis leaves to two commonly used photosystem II inhibitors: high light (HL) and 3-(3,4-dichlorophenyl)-1,1-dimethylurea (DCMU). Microarray analysis of expression patterns during the period of redox adjustment to these inhibitors reveals that 20% and 8% of the transcriptome are under HL and DCMU regulation, respectively. Approximately 6% comprise a subset of genes common to both perturbations, the redox responsive genes (RRGs). A redox network was generated in an attempt to identify genes whose expression is tightly coordinated during adjustment to homeostasis, using expression of these RRGs under HL conditions. Ten subnetworks were identified from the network. Hierarchal subclustering of subnetworks responding to the DCMU stress identified novel groups of genes that were tightly controlled while adjusting to homeostasis. Upstream analysis of the promoters of the genes in these clusters revealed different motifs for each subnetwork, including motifs that were previously identified with responses to other stresses, such as light, dehydration, or abscisic acid. Functional categorization of RRGs demonstrated involvement of genes in many metabolic pathways, including several families of transcription factors, especially those in the AP2 family. Using a T-DNA insertion in one AP2 transcription factor (redox-responsive transcription factor 1 [RRTF1]) from the RRGs, we showed that the genes predicted to be within the subnetwork containing RRTF1 were changed in this insertion line (Δrrtf1). Furthermore, Δrrtf1 showed greater sensitivity to photosynthetic stress compared to the wild type.

Redox (oxidation-reduction) reactions are basic to all cellular processes. The redox environment of the cell governs the activity of metabolic processes by regulating protein function that in turn regulates key processes in growth and development (Sakuma et al., 2002; Wu et al., 2007). Buffering capacity of the cellular redox environment is thus very critical, and, as a result, elaborate mechanisms are in place in plants to keep control of redox levels (e.g. by the action of redox pairs [ferredoxin, thioredoxin, etc.], antioxidant systems [superoxide dismutase, glutathione, etc.], and other secondary metabolites [flavonoids, alkaloids, and carotenoids]; Dietz, 2003).

In plants, the major redox determinant is the photosynthetic electron transport chain. During the day, plants utilize light energy to assimilate carbon using reducing power generated by the photosynthetic electron transport chain and concomitantly generate oxidizing power in the form of molecular oxygen. The enzymes in the chloroplast cycle between their oxidized and reduced forms to regulate metabolic processes during the day (Scheibe, 1991), but by night these enzymes are in their oxidized form. Hence, chloroplast metabolism is tightly linked to the redox state of the cell.

In response to external stimuli, such as abiotic/biotic stresses, plants modify their normal metabolic responses and alter their physiological and developmental programs. The nature and extent of modification is highly dependent on the nature of the stimulus itself, the dose, and exposure time to the tissue in question. The cross talk between responses to different stresses may involve common intermediates, as has been suggested by identifying common genes (Seki et al., 2002). For example, the phytohormone abscisic acid (ABA) plays a crucial role in abiotic stress responses, but also interacts with downstream light signaling. Furthermore, ABA has been shown to regulate stomatal opening/closure in response to water loss (Mishra et al., 2006), which is also linked to redox status because closure of stomata in the presence of ABA limits uptake of CO₂ leading to a decrease in photosynthesis. Recent reports have also demonstrated that ABA interacts with salicylic acid (SA) and jasmonic acid (JA) pathways, both components of biotic stress/defense in plants (Karpinski et al., 2003; Mateo et al., 2006). In addition to ABA, reactive oxygen species are known to play a role as a signaling molecule during stress, as is hydrogen peroxide and its interaction with ABA, SA, and JA. Hence, the complex interactions identified pose challenging questions and require sophisticated approaches to dissect the core regulatory networks that govern these responses that maintain redox homeostasis.

Previous studies have focused on identification and characterization of individual redox sensors and modifiers. This includes the retrograde signaling pathways between chloroplast and nucleus (Ankele et al., 2007; Koussevitzky et al., 2007). Similarly, mitochondrial retrograde regulation has recently been highlighted (Rhoads and Subbaiah, 2007) and also shown to play a key role in maintaining cellular homeostasis (Noctor et al., 2007). Most of the information has been obtained by studying mutants defective in maintaining homeostasis, which is primarily due to a lack of a functional antioxidant enzyme (Karpinski et al., 1997; Vandenabeele et al., 2004). However, with the recent availability of complete genome sequences, we can now follow the changes in gene expression levels and identify all the genes that respond to change in redox status, as well as those that are expressed to maintain redox homeostasis. Utilizing this information to delineate signaling cascades and cross talk between different organelles/pathways under different stresses, one has the opportunity to identify the relevant gene networks, as well as new candidate genes that can be further validated for a role in maintaining redox homeostasis.

In our study, we used Arabidopsis (Arabidopsis thaliana) leaves and perturbed the cellular redox status by targeting a photosynthetic reaction center (PSII) using high light (HL) stress and the inhibitor 3-(3,4-dichlorophenyl)-1,1-dimethylurea (DCMU). We generated gene expression networks from control and treated leaves to identify genes that were highly connected to specific networks and possibly to phenotypes. We validated our findings in one subnetwork by demonstrating that when one novel transcription factor (redox-responsive transcription factor 1 [RRTF1]) was made nonfunctional, not only did the mutant plant alter the expression of genes associated in the network, but it lacked the ability to adjust to redox changes.

RESULTS

Two commonly used inhibitors of photosynthetic electron transport, HL and DCMU, were used to study redox regulation in Arabidopsis. To follow the redox state of control and treated Arabidopsis leaves, chlorophyll fluorescence was measured as an indicator of photosynthetic efficiency. Leaves that received HL showed a decline in PSII efficiency for the first 1.5 h of HL exposure, and maintained a PSII efficiency of approximately 70% for the next 4.5 h. (Fig. 1). DCMU had a more severe effect on PSII efficiency; a continuous decrease in PSII efficiency was observed for the first 3 h of treatment, reaching a steady state of only approximately 20% PSII efficiency by 6 h (Fig. 1). There was no detectable change in the PSII efficiency when leaves received regular light or in the absence of DCMU.

Figure 1.

Efficiency of PSII under different stresses. Chlorophyll fluorescence of Arabidopsis leaves was measured after exposure to HL of 750 μmol/m² intensity (top) or 5 μm DCMU (bottom) and represented as PSII efficiency. The graph represents the mean of three independent experiments, each of which was performed in triplicate. Samples selected for microarray analysis are from the time points ht1 to ht4 (HL) and dt1 to dt3 (DCMU).

RNA was extracted at each of the time points for both treatments (Fig. 1) and analyzed by microarray analysis. The total number of genes altered by the perturbations are given in Table I (for gene lists, see Supplemental Tables S1–S7), whereas Table II shows the number of genes that are differentially expressed at any one time point or at all time points for HL (htx) and DCMU (dtx) treatments (for gene lists, see Supplemental Tables S8 and S9). From all of the genes that are differentially expressed at all time points, about 20% of the Arabidopsis transcriptome is altered under HL, whereas only 8% of the transcriptome is affected by DCMU treatment.

Table I.

Number of genes that are either up- or down-regulated under different PSII stresses

Arabidopsis leaves were subjected to HL or DCMU at different time points. The number of transcripts that show a differential response of >2× when compared with untreated leaves are shown. Different times points for the HL stress correspond to 45, 90, 180, and 360 min (ht1–ht4, respectively), whereas for DCMU the times are 90, 180, and 360 min (dt1–dt3, respectively).

	HL				DCMU
	ht1	ht2	ht3	ht4	dt1	dt2	dt3
Up	1,335	1,408	1,603	897	782	429	505
Down	1,227	1,175	1,028	775	842	982	1,038
Total	2,562	2,583	2,631	1,672	1,624	1,411	1,543

Table II.

Number of genes that are either up- or down-regulated under different PSII stresses

Shown are results for any time point or for all time points under HL stress (htx) and for any one time point or for all time points under DCMU treatment (dtx).

	HL		DCMU
	Any htx	All htx	Any dtx	All dtx
Up	2,543	323	1,197	135
Down	2,111	274	1,520	474
Total	4,654	597	2,717	609

The set of differentially expressed genes in common to both HL- and DCMU-treated tissue includes the redox responsive genes (RRGs; Fig. 2). A redox network in response to HL for the RRG1 set was generated (Fig. 3). In the network of 1,201 genes in RRG1 (200 genes did not meet the statistical criteria for coexpression), more than 8,000 connections were demonstrated, indicating that they are coexpressed as a set. Based on their expression profile at different time points under HL, RRG1 were segregated into 10 subnetworks. Genes in subnetworks 1 to 5 have lower transcript expression following HL exposure, whereas genes in subnetworks 6 to 10 have higher expression of transcripts. Each subnetwork has transcripts representing genes from various metabolic pathways, but some subnetworks have a significant number of transcripts whose gene products belong predominantly to one physiological function (Supplemental Table S10). For example, ribosomal gene transcripts are preferentially increased in subnetworks 6 and 7, whereas transcripts from genes that function in energy metabolism are decreased in subnetwork 3. Subnetwork 1 is the largest and has 299 transcripts representing genes involved in several metabolic pathways and whose products are located in different subcellular compartments.

Figure 2.

Identification of RRGs. The gene set common between HL and DCMU is denoted as the RRG1. This gene set was derived from the microarray analysis and has a P value ≤0.01 and a fold change ≥±2 in at least one time point.

Figure 3.

Redox regulation network. Network diagram of RRG1 perturbed under HL treatment was generated in Cytoscape 2.3 using organic layout. Each point (node) represents a gene and a line (edge) is drawn between two nodes. Subnetworks with more than 15 genes are encircled and arbitrarily assigned a number (1–10).

In each of the 10 subnetworks generated under HL, we further investigated how these groups of genes behaved under HL and DCMU treatment. Hierarchal subclustering of genes in each of the 10 subnetworks identified under HL using expression data from the DCMU experiment was performed. The sets of genes that had similar expression profiles under HL and DCMU were obtained (Fig. 4). Hence, these new subgroups represent redox subnetworks represented in both HL and DCMU treatments.

Figure 4.

Hierarchal clustering of subnetworks. Differential expression of all transcripts from the subnetworks (numbered 1–10) under HL and DCMU perturbations is depicted as heat maps. Green represents decreased expression and red represents increased expression of a given transcript. Each column under HL and DCMU represents indicated time points (see Fig. 1) and each row represents a gene. The number of genes in a given subnetwork varies from 19 (subnetwork 6) to 297 (subnetwork 1). Motifs in a given subnetwork were identified using 500-bp upstream sequences. Assigned P value represents the chance of obtaining the given motif in a random set of sequences.

To further investigate whether there were common motifs in promoters of each redox subnetwork, as well as whether each redox subnetwork had a defining motif, we examined sequences within 500 bp upstream from the Met start site to identify the most common six-base element within the promoters in each cluster (Fig. 4). It is clear from the P value of the motifs identified for each cluster that they are significant and different from each other, except perhaps clusters 4 and 5, which have the (C)CACGT(G) motif associated with the ABA response element (Zhang et al., 2005).

In an attempt to reduce further the core RRGs, the list of genes in RRG1 was narrowed by applying a more stringent criterion. We identified those genes that were differentially expressed (increased or decreased with a fold change ≥2) in both treatments and at all the time points, instead of only one time point in RRG1. This set corresponds to 141 genes (RRG2; Fig. 5A).

Figure 5.

Characterization of RRGs. A, Venn diagram demonstrating 141 RRG2 transcripts that have a P value ≤0.01 and a fold change ≥±2 over all time points under both HL and DCMU perturbations. B, Functional categorization of RRG2 based on GO annotations.

Functional categorization of RRG2 based on the Gene> Ontology (GO) annotation (Fig. 5B) revealed that about one-half (72) of RRG2 were novel with no previously identified function. Because we are interested in defining the redox regulatory network, we observed that RRG2 contained 16 transcription factors, including overrepresentation from AP2 domain-containing transcription factors (Table III). To validate our prediction that any one of these transcription factors are controlling a set of genes that relate to the stress response, one of the AP2 domain-containing transcription factors was selected for further analysis. We chose At4g34410 (RRTF1), a member of subnetwork 1, whose previous link to the control of redox homeostasis and any other phenotype was unknown.

Table III.

Number of transcription factors and their binding motifs in different families of RRG2

Transcription Factor Family	No. in RRGs	Motif
AP2 domain containing	6	(G)CCGA(C)
Myb related	4	YAAC(G/T)G
Zinc finger type	3	GGGCGG
WRKY type	2	(TT)TGAC(C/T)
Heat shock	1

A T-DNA insertion line of RRTF1 (Δrrtf1) was obtained and confirmed that the corresponding transcript was absent (data not shown). When we compared the set of differentially expressed genes between wild type and Δrrtf1 after a 1.5-h exposure to HL, 30 of 297 genes present in subnetwork 1 were differentially expressed in the absence of RRTF1. These 30 genes are represented as a coexpression network (within subnetwork 1; Fig. 6) that is affected by the absence of RRTF1. These included transcripts whose genes are involved in signaling pathways and whose subcellular localizations were located in different compartments, including the nucleus, chloroplast, mitochondrion, and cell wall (Supplemental Fig. S1). These data validate the coexpression predictions based on the network analysis.

Figure 6.

Core redox network of subnetwork 1. This subnetwork contains RRTF1, demonstrating its associations with other genes that were altered in Δrrtf1 compared to wild type. White circles are genes that code for an unknown protein.

To further validate whether these 30 genes are under the control of RRTF1, HL, or both, semiquantitative reverse transcription (RT)-PCR analysis was performed (Fig. 7). Compared to normal light, all 30 genes are down-regulated under HL (wild type-HL) in wild type and are up-regulated in Δrrtf1 (Fig. 7). Compared to wild type-HL, 11 of 30 genes in the subnetwork were not changed significantly in Δrrtf1-HL (i.e. they showed the same expression response [up-regulated] as they did in Δrrtf1). This appears to indicate that these genes are normally under the control of RRTF1 during HL stress. However, the remaining 19 genes did change their expression pattern when compared to Δrrtf1 kept at normal light and are apparently not dependent on RRTF1 in response to HL.

Figure 7.

Semiquantitative RT-PCR on the subnetwork of genes shown in Figure 6. Transcript levels of 30 genes shown in Figure 6 were compared under different light conditions in wild type and Δrrtf1. The At identification reference (left) and common name (right) are given for each as well as their differential expression level: wild type compared to wild type-HL (left column, wild type-HL); wild type-HL compared to Δrrtf1 (middle column, Δrrtf1); and wild type-HL compared to Δrrtf1-HL (right column, Δrrtf1-HL).

The RRTF1 knockout plants did not exhibit any phenotypic differences under normal growth conditions (data not shown). However, when exposed to HL, we observed a greater sensitivity to this stress as evidenced by bleaching of leaves compared to controls (Fig. 8), a phenotype consistent with the inability to reach homeostasis under HL.

Figure 8.

Physiological response of Δrrtf1. Leaves from wild type and Δrrtf1 floating in water are exposed to HL and photographed after 48 h.

DISCUSSION

Our purpose was to identify global redox response networks by utilizing the photosynthesis perturbants HL and DCMU. Photosynthesis plays a crucial role in determining redox status of the cell as evidenced by the production of a strong oxidant (molecular oxygen at a redox potential of 0.8 V) and a strong reductant, for example, NADPH (redox potential of −0.34 V). We measured chlorophyll fluorescence, a fast, simple, and noninvasive way of detecting changes in PSII efficiency as a measure of redox changes. Following exposure to HL under our experimental condition (750 μE/mol⁻¹), Arabidopsis leaves immediately experience a redox imbalance, but by 90 min reach homeostasis at about 70% of controls. Because DCMU binds strongly to the Q_b site of the PSII complex, resulting in the inhibition of electron flow (Russell et al., 1995), this treatment also leads to a drastic change in the reducing environment, but by 3 h leaves reach homeostasis at about 20% of controls. Although Arabidopsis leaves attain a new redox homeostasis following HL or DCMU perturbation of PSII (Fig. 1), the time required to reach steady state was different for each stress, as was the magnitude of the decrease in photosynthetic efficiency.

This comprehensive study was used to gain insight into the global response of wild-type Arabidopsis to achieve redox homeostasis following exposure to photosynthetic perturbations using a systems approach. Our microarray analysis revealed that a major part of the transcriptome (approximately 20%) is responsive to HL. In addition to a large set of novel genes, most of the genes previously characterized to be redox responsive (for review, see Oelze et al., 2008; Pfannschmidt et al., 2008) are also included in our networks acting as internal controls for the study. The significant overlap between genes that are modulated under various signaling, e.g. HL and DCMU (this study), defense (Bartsch et al., 2006), reactive oxygen species (Gadjev et al., 2006), and singlet oxygen (op den Camp et al., 2003), highlights overlapping signaling networks containing common core elements given such diverse receptors.

During the period of redox adjustment during the first 3 h of either treatment, we were interested in the gene regulatory networks operating during these PSII stresses. By using two different inhibitors of PSII, we reasoned that the genes differentially expressed by both perturbations would more likely be directly identified with the network regulating PSII efficiency. As a first step, we identified 1,401 genes that showed a change of expression (>2×) during either treatment at any one of the seven time points. This set was called RRGs. It remains to be established for all RRGs that they indeed respond to redox status and not specifically to the treatment. However, a comparison between RRG1 and the genes observed to be responsive to the redox status of the plastoquinone (PQ) pool by Adamiec et al. (2008) reveals about 80% of the responding genes are in common with those reported in this article. A difference, however, is that Adamiec et al. (2008) reported a list of 50 genes that showed differential expression following exposure to HL, but reverted their expression under DCMU treatment. Adamiec et al. (2008) reasoned that the oxidized and reduced status of the PQ pool transmits the signal using a similar set of genes, but with an opposite expression pattern (based on the redox state of the PQ pool). However, the change in redox status of the cell (and not the direction) probably determines the effective differences in the transcriptome, and hence the response. Fey et al. (2005) used PSI, PSII, specific light, and DCMU to alter redox status. They identified 286 RRGs, of which 76 known genes with diverse functions are listed. However, significant overlap was not observed between this list of 76 genes and RRG1. Similarly, 1,590 nuclear-encoded chloroplast gene sets identified by Biehl et al. (2005) to be responsive to diverse environmental and genetic conditions did not show significant overlap with RRG1: Only 115 genes belonging to different regulons were part of RRG1. We observed that these 115 genes were scattered across different subnetworks. However, comparison between 1,206 genes that respond to singlet oxygen, superoxide, or hydrogen peroxide (op den Camp et al., 2003) and RRG1 shows about 27% conservation (377 genes). It is important to note that technical variations in different studies limit reproduction of gene lists. Fey et al. (2005) and Biehl et al. (2005) employ 3,292 gene sequence tag arrays that mainly represent nuclear-encoded chloroplast genes (approximately 10% of the whole genome).

We focused our efforts on the RRG1 set of genes that were altered in their expression during HL stress to generate a coexpression network. This analysis resulted in a network containing 1,201 genes with 8,000 connections and clearly revealed 10 distinct subnetworks, indicative of the complexity of the regulatory networks involved in the adjustment of homeostasis.

We further focused our attention on the largest network (i.e. subnetwork 1), which contained 299 genes, and attempted to identify a key regulator based on the network analysis. Within subnetwork 1, about 80 genes were up- or down-regulated by ≥2-fold (RRG2). A prediction would be that if we removed a key regulatory component within RRG2, genes linked to this putative regulator would change expression, as would the ability of leaves to reach homeostasis under HL stress.

Because removal of a transcription factor within this network would be the most efficient manner to remove a set of linked regulatory transcripts, we noted that there was an overrepresentation of AP2 domain-containing transcription factors in RRG2. There are 145 DREB/ERF family proteins in Arabidopsis, classified into 12 groups (Nakano et al., 2006), and of these, 122 are single AP2 domain-containing proteins. This classification was based on several criteria, such as gene structure, phylogeny, chromosome locations, and conserved motifs (Sakuma et al., 2002). One of the six AP2 proteins in RRG2, At4g34410 (RRTF1), was associated with two different groups and is unique among AP2 domain proteins in that it lacks an intron, an extra CMX-1 motif, and no reported annotation to a particular process or pathway.

In the redox subnetwork 1, RRTF1 has 13 direct connections, which are regulated at all time points in both treatments. When the transcriptome analysis after 1.5-h exposure to HL between wild-type and a T-DNA insertion into At4g34410 (Δrrtf1) was compared, we identified 293 differentially expressed genes, which were represented in RRG1, and 30 genes that were part of RRG2. Without the product of RRTF1, it appears that about 20% of the genes in RRG1 and 37% of the genes in RRG2 are affected.

Network analysis revealed that 30 genes that are differentially expressed in rrtf1 are in the vicinity of the RRTF1in subnetwork 1. This smaller subnetwork of 30 genes was defined as a core redox network (Fig. 6). In the core redox network, 70% of genes are well documented in literature to have a role during stress responses. If this core network is vital for maintenance of redox homeostasis, perturbation to this regulatory network will lead to a susceptible state.

This led us to the prediction that Δrrtf1 might also have a different response to redox imbalance if these genes were truly involved in redox homeostasis. Phenotypic characterization under normal growth conditions did not exhibit any differences between wild type and ΔRRTF1. However, exposure to HL for 24 h bleached the leaves of Δrrtf1, whereas wild-type plants did not exhibit any phenotypic differences. Hence, the regulatory network containing RRTF1 appears to play a major role in the adjustment of Arabidopsis leaves to reach homeostasis after HL stress.

MATERIALS AND METHODS

Plant Material and Growth Condition

A Salk T-DNA insertion line (SALK_150614; containing insertion in exon of At4g34410) was obtained from the Arabidopsis Biological Resource Center (ABRC). For SALK_150614, the gene-specific primers LP (5′-CGCGATGCTTTGTAGGAGTAG-3′) and RP (5′-GATCTCAGGGGAAAACGAAAC-3′) were used in conjunction with the T-DNA left-border primer LBb1 (5′-GCGTGGACCGCTTGCTGCAACT-3′; Alonso and Stepanova, 2003; Alonso et al., 2003) to identify the nature of insertion (absent, hemizygous, homozygous). These primers were created using the SIGnAL iSECT tool (http://signal.salk.edu/isects.html). The homozygous line for T-DNA insertion in At4g34410 (ΔRRTF1) was obtained and absence of transcript corresponding to At4g34410 was demonstrated.

Arabidopsis (Arabidopsis thaliana ecotype Columbia [Col]) wild-type and ΔRRTF1 seeds were germinated on one-half-strength Murashige and Skoog medium (Murashige and Skoog, 1962) containing phytagel (1.5 g/L) but lacking Suc. Approximately 30 plants were grown in 90-mm² petri plates at 25°C, under a16-h-day/8-h-night cycle with a light intensity of 75 μmol/m².

Redox Perturbation

Two redox perturbations were chosen that affect redox status of the PQ pool, namely, HL and DCMU. For HL treatment, light intensity of 750 μmol/m² and 5 μm DCMU were used. Leaves from 4-week-old plants were floated in water, adaxial side up, and used for microarray analysis after 0.75, 1.5, 3, and 6 h of HL or 1.5, 3, and 6 h of DCMU treatment. For each time point of a given treatment, three petri plates with floating leaves were used. Three such independent sets were pooled as one sample and RNA was extracted. At each time point of the treatment, RNA extracted from three samples was in turn pooled and hybridized to the reference RNA pool prepared in a similar manner from control leaves. Supplemental Figure S2 shows the flow chart of sample preparation and design for microarray experiments.

RNA Extraction

Leaves from the control sets and treatments were frozen in liquid nitrogen at indicated time points. RNA was extracted using Agilent's mini plant RNA isolation kit as per the manufacturer's instructions. A Nanodrop ND-1000 spectrophotometer was used to measure total RNA concentration. Quality of RNA was determined using RNA 6000 Nano Assay (2100 Bioanalyzer; Agilent Technologies). Good-quality RNA samples were pooled for microarray analysis (Supplemental Fig. S2).

Microarray Analysis

RNA samples were given to MOgene for microarray analysis (You et al., 2006). Agilent's technology was used for microarray and each treated sample was hybridized with control and a dye swap was performed. Microarray data were analyzed using Agilent's error model and replicates were combined using Luminator software that performs an ANOVA to identify genes with significantly different expression than control and also accounts for dye bias, if any, in the samples.

Network Analysis

Euclidean distance is calculated between each pair of genes. Low value of Euclidean distance indicates that the two genes have a high degree of correlation and are very close in their expression profile over time. An edge is drawn between two genes if Euclidean distance between them is below the arbitrary cutoff value 0.15. The results are then visualized using Cytoscape 2.3. A connectivity graph was generated using an organic model.

Motif Analysis

Subclusters obtained using gene expression value similarity are candidates for coregulated genes. Such coregulated genes usually share common binding-site motifs in their upstream regions. A widely used motif identification algorithm known as CONSENSUS developed by the Gary Stormo group at Washington University was used to identify the possible conserved sequences within subclusters. The first 500 bp of the upstream regions of genes belonging to each subcluster were obtained from The Arabidopsis Information Resource (TAIR; http://www.arabidopsis.org) database. Patterns with the length of six were searched without considering the complements of the sequences because the sequences obtained from TAIR had correct orientation of the genes. Each sequence contributed exactly one motif for the final results.

RT-PCR Analysis

RNA samples were isolated from leaves of wild type and Δrrtf1 floating in water exposed to either regular light or HL for 1.5 h. cDNA was synthesized using 1 μg of total RNA using SuperScriptIII reverse transcriptase (Invitrogen) as per the manufacturer's instructions. PCR reactions were performed in 25-μL volume using 2 μL of diluted cDNA (1:20), 0.1 μm each of forward and reverse primer (Table IV), and Promega's Green GO Taq master mix. PCR conditions were 94°C for 3 min, different cycles of 94°C for 30 s, 55°C for 30 s, and 72°C for 45 s, and a final extension cycle at 72°C for 7 min. At the end of reaction, 4 μL of PCR reaction was separated on a 1% agarose gel. Ethidium bromide-stained gel was documented using BIO-RAD gel documentation and intensities of bands were measured using ImageJ.

Table IV.

Primers and number of cycles used for RT-PCR amplification

Gene Name	Forward Primer (5′→3′)	Reverse Primer (5′→3′)	Cycle No.
At1g02400	AACTTCCCGGTGATCGATTTC	CGGTGCTGGTGGATAGTGATT	29
At1g07000	GACGAGAATCTTTACGCCGC	AGTGATTGGACGAGCAACGAT	29
At1g12610	TGCGGGAAGGAGAGTGTTTA	TGTCTTCCATATACGATCGCG	26
At1g18740	AGTAGCCACCACGAGCCAAG	ACCAATCGAACGGTGGTGAC	27
At1g19770	TCCATAGCGAAACTCCTCGAA	AGCAGCCCATGAGCCTCTAG	27
At1g24140	AACTCCTCAGCCATTCCGC	CACGCTCTTGACCTCTTCGG	29
At1g72520	ACCCTTTTGATCCGAGACCT	GGAAGTGAACAGGGCCACAT	26
At2g01180	GACCAGGAGGCGCAGAGAG	GTGGCCTTCCTTGACCTCAG	27
At2g18050	GAAAACCACCACTCATCCTCC	TGATAGACTTAGGCTGTCTCGC	26
At2g26530	CTGGCTGCTTCTTAAGCGC	GCGAAAGTGATCGAGCAACTC	22
At2g30020	GCTCCGTCGCCGTATGTAAT	TCGCCTGTTTACGATCTCCAT	29
At2g37940	TCGTCGTGAATCTTCCAAGCT	CGCATCCATACATCACCCC	29
At2g38470	AGAACAATGGAGCCAAACCG	GCTTCAGGTTCACTCCCACAA	22
At2g40140	CCACAGGAAGATCCGATGTGA	TACGAAACTCGGGACAAGGC	27
At3g48090	CACAAGGAAGAAGCAGGAGCA	CTTTCGAGCAAGCATAATCCG	29
At3g52400	CTCTCCGGCTCGTTTAAAACC	GCACATTCTCCCAACCGTCT	22
At3g55980	TACTCCGCTTCACTGTGCTGT	CCGTGAGCATACTCGCAAGA	22
At4g01360	TGAGCCAACAACGTACACGC	GATCCATCGATTCAACCAGC	26
At4g17615	CGGTTGTTGATGATGGTCTGA	GCAATCTCATCGACCTCCGA	27
At4g22690	TCGTCGTCTCTTCCCCTGAC	TCACATGGCCTTTCATCTGCT	27
At4g34390	GCGCTGTGTTCTCTCTGCC	TCCTGGGAAGTGTATCGCG	29
At4g34410	CCGTGTCAGGGTTTTTCCAG	GCTTGCACTTGCCTTTGCT	22
At4g36500	GAAGCCGTCGTGGAGGATAC	GGCATCCCACATACCACCTT	22
At4g39890	GTCGGAAAAACCAGCATCAT	CATGTCATCTGATTTTGTTGCC	26
At5g22250	GAAACCTCCCAGATCTCGGC	TCCCTCATCCTCTGAAACGC	29
At5g25930	GGATTGTTGCCACGTCTGC	CGAGCTTTCGACGAAAATCTT	26
At5g58430	CGAGATGGCTAAGCGAATGC	CCTCTTATCGCCTCTCCCAA	22
At5g59550	ATCACGGCTTCGTATTGGTGT	AATCGCCGATTTAGACGCC	29
At5g62520	AGTATGGGTTTAGCGAGCCG	CCGGTTATGCTTCTGACTCGT	26
At5g66210	CCTACGTCGCCATCCACAG	GCCCTGATCTGCGCTTTAAT	27

Supplemental Data

The following materials are available in the online version of this article.

• Supplemental Figure S1. Cellular localization of the 30 genes of the core redox network of subnetwork 1.

• Supplemental Figure S2. Microarray experiment design.

• Supplemental Table S1. Differentially expressed genes at ht1.

• Supplemental Table S2. Differentially expressed genes at ht2.

• Supplemental Table S3. Differentially expressed genes at ht3.

• Supplemental Table S4. Differentially expressed genes at ht4.

• Supplemental Table S5. Differentially expressed genes at dt1.

• Supplemental Table S6. Differentially expressed genes at dt2.

• Supplemental Table S7. Differentially expressed genes at dt3.

• Supplemental Table S8. Differentially expressed genes at htx.

• Supplemental Table S9. Differentially expressed genes at dtx.

• Supplemental Table S10. Pathways overrepresented in redox regulation network.