Senescence is the final stage of leaf development. Although it means the loss of vitality of leaf tissue, leaf senescence is tightly controlled by the development to increase the fitness of the whole plant. The molecular mechanisms regulating the induction and progression of leaf senescence are complex. We used a cDNA microarray, containing 11 500 Arabidopsis DNA elements, and the whole-genome Arabidopsis ATH1 Genome Array to examine global gene expression in dark-induced leaf senescence. By monitoring the gene expression patterns at carefully chosen time points, with three biological replicates each time, we identified thousands of up- or down-regulated genes involved in dark-induced senescence. These genes were clustered and categorized according to their expression patterns and responsiveness to dark treatment. Genes with different expression kinetics were classified according to different biological processes. Genes showing significant alteration of expression patterns in all available biochemical pathways were plotted to envision the molecular events occurring in the processes examined. With the expression data, we postulated an innovative biochemical pathway involving pyruvate orthophosphate dikinase in generating asparagine for nitrogen remobilization in dark-treated leaves. We also surveyed the alteration in expression of Arabidopsis transcription factor genes and established an apparent association of GRAS, bZIP, WRKY, NAC, and C2H2 transcription factor families with leaf senescence.
Unable to flee the habitat in which they grow, plants proceed with senescence in a highly coordinated manner (Buchanan-Wollaston et al., 2003; Gan, 2003; Lim et al., 2003; Quirino et al., 2000; Yoshida, 2003). Many biotic and abiotic stresses also trigger senescence. Processes associated with leaf senescence include the disorganization of chloroplasts, shrinkage of cytoplasmic volume and decrease in cellular metabolic activities. Significant metabolic changes are involved in the degeneration and remobilization of macromolecules that accumulate during growth and maturation.
Research efforts to reveal the underlying molecular mechanism of plant senescence began with the collecting and analyzing of senescence-associated genes (SAGs). Sequence and/or functional analyses revealed that SAG-encoded proteins include proteases, nucleases, lipid-, carbohydrate- and nitrogen-metabolizing enzymes, stress-responsive proteins, and transcriptional regulators (for review see Buchanan-Wollaston et al., 2003). Previous studies have shown that SAGs have common and distinct expression profiles in response to various senescence-inducing factors (He et al., 2001), which suggests that plant cells have a complex fine-tuning mechanism to cope with various signals in senescence.
The advances in the tools of molecular biology research have greatly helped the investigation of SAGs. Bhalerao et al. (2003) reported on the genes, preferentially expressed in autumn leaves, by sequencing expressed sequence tags (ESTs) of cDNA libraries made from field-grown aspen (Populus tremula). A DNA microarray with 13 490 aspen ESTs was later created and used to study the leaf transcriptome of aspen leaves during natural autumn senescence (Andersson et al., 2004). Large-scale identification of SAGs via suppression substractive hybridization has added 70 new members to the current SAG collection in Arabidopsis (Gepstein et al., 2003). More recently, transcriptome associated with leaf senescence was examined by large-scale single-pass sequencing of ESTs prepared from senescing Arabidopsis leaves (Gepstein, 2004; Guo et al., 2004). A custom microarray of approximately 100 genes related to programmed cell death was constructed to study gene regulation in cell death in suspension culture (Swidzinski et al., 2002). In addition, Buchanan-Wollaston et al. (2003) also compared gene expression in mature, green, early and mid-senescing leaves by hybridizing the Arabidopsis GeneChip of 8000 genes.
With the completion of genome sequencing and the availability of several research tools, we can now systematically survey the molecular events in senescence in Arabidopsis. Here, we describe our use of DNA microarray and whole-genome GeneChip hybridization to observe gene expression induced by dark treatment in Arabidopsis leaf senescence. In addition to expanding the inventory of SAGs, we attempted to associate the genes and their response behaviors with their corresponding biochemical events and pathways.
Results and discussion
Effects of dark treatment on intact Arabidopsis rosette leaves
Leaf senescence is a highly regulated process and is influenced by various internal and external factors, including dark treatment (Gan, 2003; Lim et al., 2003; Yoshida, 2003). Intact Arabidopsis plants at stage 1.10 (Boyes et al., 2001) underwent dark treatment for 6 h, 12 h, 1, 2, 3, 4, 5 and 6 days. The use of intact plants eliminated the potential effects of wounding or dehydration, which also trigger leaf senescence (He et al., 2001). As dark-induced senescence occurs slowly in intact plants, molecular regulators (e.g. transcriptional regulators) are better visualized in such plants. Photographs of representative dark-treated plants show yellowing phenotypes with prolonged dark incubation, whereas control plants remain green (Figure 1a); chlorophyll and protein content decreased gradually (Figure 1b; Weaver and Amasino, 2001). Thus, dark treatment initiates leaf senescence in Arabidopsis.
Gene expression profiling of dark-treated Arabidopsis rosette leaves
To elucidate the molecular events in dark-induced senescence, we surveyed global gene expression using cDNA microarray and Affymetrix ATH1 Genome Array hybridization. We pooled the samples from 12 plants at each time point for RNA isolation, labeling and hybridization of cDNA microarrays containing 11 500 DNA elements [equivalent to approximately 7500 unique genes; Arabidopsis Functional Genomic Consortium (AFGC); Wu et al., 2001; http://www.arabidopsis.org/info/2010_projects/comp_proj/AFGC/RevisedAFGC/site2Over1L.htm]. A total of 27 microarray data sets was generated with RNA samples from the nine time points, with three biological replicates for each time point. The complete data sets for all 27 experiments have been deposited with the Stanford Microarray Database, Academia Sinica Computing Centre (SMD, ASCC) at http://www.bitora.sinica.edu.tw:7777/smd/MicroArray/SMD/. The expression data from the three biological replicates are highly reproducible, with a coefficient of variation less than 0.35 in triplicate from at least one time point (data not shown). We used cluster analysis to organize the differentially expressed genes according to their expression in response to dark treatment as described in the Experimental procedures. The seven most representative expression patterns are depicted in Figure 2. ‘Early’ describes genes up- or down-regulated with dark treatment for 1 day or shorter. ‘Late’ describes genes up- or down-regulated with dark treatment for 2 days or longer. Using the terms of differential expression cut-offs described in the Experimental procedures, we selected 588 unique genes (equivalent to 890 ESTs) by their high expression correlation with these seven patterns. The identities and expression data of genes in each corresponding group are listed in Table S1.
As we used dark treatment to induce the senescing process, some genes regulated by diurnal changes or carbon starvation could have been included in our analyses. To clarify this, we first compared genes in the above seven patterns with diurnal genes discerned previously with the use of the same cDNA microarray system (Schaffer et al., 2001). The results indicated that 24.7% of the genes (145 of 588) reported in Figure 2 are also considered diurnal genes (highlighted in Table S1). These diurnal genes are distributed among all seven expression patterns shown in Figure 2 but with preferential presence in the ‘up and down’, ‘early up’ and ‘down and up’ groups (47, 42 and 28%, respectively; Table S2). We also compared genes, which our experiments revealed had altered expression, with those affected by carbon starvation (Thimm et al., 2004) using data obtained from the Arabidopsis whole-genome ATH1 GeneChip (see Experimental procedures and data shown below for ATH1 experiments conducted in this report; Figure S1 and Table S3). We found that 26% (17%) of the genes induced (repressed) by 1 day dark treatment and 19% (14%) of the genes induced (repressed) by 5-day dark treatment also responded to carbon starvation. In Figure S1, a Venn diagram shows genes with common and distinct patterns in response to carbon starvation, and 1- and 5-day dark treatments. Those patterns were classified into 14 expression groups: seven for upregulated and seven for downregulated genes with these treatments. We consider this classification to be an initial attempt to analyze the differential contribution of genes in carbon starvation and the progress of leaf senescence in Arabidopsis.
Molecular events in senescing Arabidopsis leaves
The gene expression data are usually composed of a massive amount of information but require intensive data mining to extrapolate biological contributions. To integrate the expression data into the biological context, we sought to superimpose such data onto biochemical pathways. One clear example is to visualize the expression kinetics of genes involved in the light and dark reaction of photosynthesis (Figure 3). These genes are significantly downregulated in dark-treated leaf samples (expression data listed in Table S4). Although the results were expected, the expression kinetics of these photosynthetic genes has rarely been comprehensively investigated in senescing Arabidopsis leaves.
To gain more insight into the molecular events in senescing Arabidopsis leaves, we performed Arabidopsis ATH1 Genome Array hybridization experiments with RNA samples from control and 1- and 5-day dark-treated leaf tissues. Samples from these two time points were chosen on the basis of the time-course experiments performed above that represent ‘early’ and ‘late’ responding genes.
Replicate ATH1 experiments were performed with RNA samples from both 1- and 5-day dark-treated tissues. Data extracted from these two expression data sets (Tables S5–S11) showed high reproducibility, with a regression correlation coefficient of 0.95 and 0.94 for 1- and 5-day dark treatments, respectively (see Figure S2). We also used the expression data from cDNA microarray and ATH1 hybridization analyses to cross-validate the quality of our expression data. Of the 588 differentially regulated loci (cDNA microarray data, Figure 2), 558 have probe sets represented on ATH1. Among them, 93.1% of the genes showed positive correlation in expression tendency between cDNA microarray and ATH1 hybridization results (Table S12). We thus believe that ATH1 hybridization gives quality results and used these results to investigate genes associated with biochemical events in the senescing process.
As illustrated in Figure 1(b), the chlorophyll content in senescing Arabidopsis leaves decreased to almost 30% after 6 days dark treatment. Chlorophyll degradation during senescence has been summarized in recent review articles (Hortensteiner and Feller, 2002; Takamiya et al., 2000). Although some of the genes in the chlorophyll catabolism pathway have been cloned, their expression profiles in senescing leaves are yet to be examined. Three key enzymes in this pathway are chlorophyllase (AtCLHs; Benedetti and Arruda, 2002; Tsuchiya et al., 1999), pheophorbide a oxygenase (AtPaOs; Pruzinska et al., 2003) and red chlorophyll catabolite reductase (RCCRs; Mach et al., 2001; Wuthrich et al., 2000). Table S5 shows the ATH1 genome array expression data for these genes, and Table 1 contains the summarized results. The expression of AtCLH2 remained constant during dark treatment, whereas that of AtCLH1/CORI1 was downregulated. RCCR was downregulated, but the results were not significant. For AtPaOs, however, 17 of the 21 predicted AtPaOs showed detectable hybridization signals. Five (At1g05750, At1g54320, At2g07370, At2g26920, and At3g44880) were clearly upregulated. The upregulation of the ACD1 homolog At4g25650 was marginal, which is consistent with other findings (Pruzinska et al., 2003). Taken together, these results suggest that AtPaOs may play a rate-determining role in chlorophyll degradation in senescing leaves.
Table 1. Summary of Arabidopsis genes showing up- or down-regulation in selected macromolecule degradation processes in senescence
Represented on ATH1c
Total number annotatedd
aGenes with ratio values equal to or greater than 2 and expression values equal to or greater than 200 in 1- or 5-day dark treatment.
bGenes with ratio values equal to or less than 0.5 and expression values equal to or greater than 200 in controls.
cNumber of genes represented on ATH1 Genome Array for each category specified.
dTotal number of genes annotated for each category specified.
eThe 21 putative AtPaOs are predicted on the basis of bioinformatic searches of the Arabidopsis genome by using the biochemical characteristics of PaOs as computing factors. In vitro PaO activity is observed for one of the selected AtPaO members, AtPaO/ACD1 (At3g44880) (Pruzinska et al., 2003).
Chloroplast protein degradation
Non-chloroplast protein degradation
E1s and E2s
Conserved proteins of ubiquitylation pathways
Protein degradation in chloroplasts
Chloroplast proteins account for more than 70% of the total leaf proteins. Although some vacuolar proteases were upregulated in senescing leaves, less than 40% of proteins were degraded after 6 days dark treatment (Figure 1b). It is thus unlikely that vacuolar proteases are primarily responsible for the degradation. Therefore, we first examined the chloroplast stroma-localized Clp protease family members [nomenclature adopted from Adam et al., 2001 and updated annotation obtained from TAIR (http://www.arabidopsis.org)]. Most of the genes in the Clp protease complex (ClpP1-3, ClpP5-6) were expressed constitutively in dark-treated Arabidopsis leaves, which corresponds in general to previous findings (Nakabayashi et al., 1999; Zheng et al., 2002). Other Clp protease subunits, ClpR3 and ClpR4, which have not been characterized previously, also showed constitutive expression patterns (see Table S6). This observation coincides with the prediction that Clp proteases are housekeeping proteases to turnover plastid proteins for maintaining appropriate stoichiometry and removing damaged or mistargeted proteins (Adam and Clarke, 2002). Only two Clp protease members were downregulated: ClpS1 (Peltier et al., 2001) and ClpP4; the downregulation of the latter gene has been reported previously (Nakabayashi et al., 1999). However, we also observed a strong dark-induction of ClpD/ERD1 (Nakabayashi et al., 1999; Nakashima et al., 1997). ClpC1 was upregulated in our study, which implies that ClpD/ERD1 and ClpC1 may play regulatory roles in the function of the Clp protease complex in senescing Arabidopsis leaves. Both ClpC1 and ClpD encode proteins with sequence similarity to HSP100 chaperones. The upregulation of these two transcripts might reflect the need for the recruitment of unfolded proteins for degradation by Clp proteases during senescence.
In contrast to the expression patterns of Clp proteases, that of thylakoid-associated protease family members, FtsH and DegP, are poorly characterized. FtsH protease, a metalloprotease, is responsible for protein quality control of photosynthetic proteins and the biogenesis of chloroplasts (Chen et al., 2000; Haussuhl et al., 2001; Takechi et al., 2000). DegP1, 5 and 8 are localized in the lumen side of the thylakoid membrane and thus are considered to be good candidates for the degradation of lumen proteins (Adam and Clarke, 2002; Adam et al., 2001). Another DegP protease, DegP2, is localized on the stroma side of the thylakoid membrane and is responsible for the initial cleavage of damaged D1 before the complete degradation of D1 by FtsH (Haussuhl et al., 2001). Surprisingly, in our study, prolonged dark treatment seemed to downregulate all four members of DegP proteases and two of the eight chloroplast FtsH proteases (see Table S6). We also analyzed another class of plastid proteases, the Lon proteases. One of the Lon proteases was marginally upregulated under dark treatment. For the membrane-associated proteases to participate in degrading chloroplast proteins of senescing leaves, it seems that the regulation of FtsH, DegP and Lon may occur at the translational and/or post-translational level(s).
Non-chloroplast protein degradation
The activation of vacuolar proteases in senescence has been well documented (for review see Buchanan-Wollaston et al., 2003). Increasing evidence has revealed the importance of the ubiquitin-26S proteasome pathway for targeted protein degradation in normal development, including senescence, and in response to environmental cues (review see Sullivan et al., 2003; Vierstra, 2003). The increased expression of SEN3, a polyubiquitin gene, was observed in senescing leaves of Arabidopsis (Park et al., 1998). In addition, delayed senescence was observed with the mutation for Ore9 in Arabidopsis, which encodes an F-box protein and interacts with ASK1 (Woo et al., 2001). To examine the involvement of the ubiquitin-26S proteasome pathway in Arabidopsis senescing leaves, we systemically surveyed the expression profiles of all annotated genes for ubiquitins, ubiquitin-activating/conjugating enzymes (E1, E2), UBC-domain proteins without the conserved catalytic Cys residues (ubiquitin-enzyme variants, UEVs), F-box proteins and subunits of 26S proteasome (nomenclature and annotation adopted from Bachmair et al., 2001; Gagne et al., 2002; Yang et al., 2003). The expression data for these genes are presented in Table S7. UBQ3 and UBQ4 were the predominant polyubiquitin genes showing upregulation in dark-induced senescing leaves. However, the expression of SEN3/UBQ10 remained constant.
A total of 13 E1 and E2 genes, four UEV genes and five conserved component genes of the ubiquitylation pathway were upregulated. Of genes encoding 26S proteasome subunits, RPN11, a lid subunit and a metalloprotease for disassembling poly-Ub chains (Verma et al., 2002), was upregulated. Of the approximately 1200 potential E3 components, we examined approximately 700 genes encoding F-box proteins in senescing leaves. Interestingly, 71 of the 377 F-box genes represented on ATH1 were upregulated. The substrate specificities of SCF E3 complexes are basically conferred by F-box proteins. These F-box proteins genes are potential candidates for the search of proteins targeted for degradation in leaf senescence.
Lipid degradation and nitrogen remobilization
Some molecular events associated with lipid degradation during senescence have been reviewed previously (Thompson et al., 1998). Transgenic Arabidopsis with altered membrane lipase expression shows a phenotype with delayed leaf senescence (Thompson et al., 2000). Similarly, with altered expression of an acyl hydrolase gene, leaf senescence is perturbed (He and Gan, 2002). These enzymes are hypothesized to trigger the lipid degradation pathway in senescing tissues and release free fatty acids. To monitor the subsequent actions occurring in the senescing tissues, we examined genes involved in fatty acid catabolism (nomenclature and annotation adopted from Graham and Eastmond, 2002). Among the 35 genes in this category and represented on the ATH1 Genome Array, 21 were greatly upregulated; most of these involved both alpha and beta-oxidation of fatty acids (Table S8).
The above results imply an active membrane breakdown process in senescing leaves. Previous research of senescing cucumber tissues revealed increased gene expression for isocitrate lyase and malate synthase, two key enzymes in the glyoxylate cycle (Graham et al., 1992; McLaughlin and Smith, 1994). The upregulation of pyruvate orthophosphate dikinase (PPDK) has also been observed in senescing maize leaves (Smart et al., 1995). Thus, the carbon salvaged from the lipid degradation might be converted to sucrose by the gluconeogenesis pathway via the activity of PPDK. PPDK is characterized mostly in C4 plants and functions in the regeneration of the primary CO2 acceptor PEP (Hatch, 1987). PPDK also ubiquitously exists in many C3 plants, but its exact metabolic functions are not clear (Matsuoka et al., 1993; Moons et al., 1998; Rosche et al., 1994). One single-copy gene encodes the cytosolic form of PPDK in Arabidopsis. Results from both our triplicate time-course cDNA microarray and the Affymetrix ATH1 Genome Array experiments showed PPDK to be significantly upregulated.
As PPDK has a possible role in carbon salvage after lipid degradation, we examined the expression patterns of genes in the pathways hypothesized to be involved (i.e. glyoxylate cycle, citric acid cycle and the gluconeogenesis pathways). Surprisingly, only a few genes in the citric acid cycle were upregulated, whereas those in the glyoxylate cycle and gluconeogenesis pathway showed either low expression or were downregulated. The enzymes in these pathways might be regulated at the translational and/or post-translational level to account for their needs in salvaging carbon. Moreover, upregulated PPDK in dark-adapted C3 plants might have a specific role of their own.
By cross-examining the enzymatic product of PPDK and its putative derivatives, we postulate another pathway for the function of PPDK in senescing Arabidopsis leaves (Figure 4a). Genes corresponding to each intermediate enzyme were clearly upregulated (expression data for each gene in this pathway are in Table S9), which suggests that PPDK might contribute to the metabolic precursors for the synthesis of asparagine, believed to be the nitrogen carrier for nitrogen remobilization during senescence (Hayashi and Chino, 1990). Indeed, the amount of free asparagines increased significantly (Figure 4b).
Other biochemical pathways
The above observations led us to investigate which other biochemical pathways were altered in senescing Arabidopsis leaves. Toward this purpose, genes showing differential expression patterns from whole-genome ATH1 Genome Array experiments were extracted and superimposed on biochemical pathways where they belong, as described in the Experimental procedures. Results for genes participating in 97 biochemical pathways in Arabidopsis are shown in Table 2. The expression data corresponding to that in Table 2 are in Table S10. To better visualize how dark treatment influences the biochemical pathways, we integrated the expression kinetics with a weighted representation of genes in each pathway. The proportion of genes showing differential expression patterns in each biochemical pathway was calculated and color coded. Warm and cold colors were chosen for pathways showing up- and down-regulation behaviors, respectively. This presentation demonstrates the degree (proportion) and direction (up or down) of the involvement for each corresponding pathway in senescing Arabidopsis leaves. Table 2 shows that, under darkness, Arabidopsis leaves quickly adjust their cellular metabolite composition by altering more than 90% of the biochemical pathways, even under 1 day dark treatment (early genes). Genes in pathways such as photosynthesis light reaction, carbon fixation, porphyrin/chlorophyll biosynthesis, ribosome composition, amino acid biosynthesis, glycolysis/gluconeogenesis, flavonoid biosynthesis and many others exhibit early downregulation. In contrast, genes for lipid metabolism, amino acid metabolism, and carbohydrate metabolism, for example, respond positively to dark treatment. We observed neither phenotypic changes nor chlorophyll/protein alterations with only 1 day dark treatment (Figure 1). Yet, our expression data showed profound immediate responses of genes in most of the biochemical pathways inspected. It is a common belief that plant cells are highly coordinated in coping with environmental changes. Our analyses provide direct molecular evidence as to how efficient and how broad the modulations could be for plants undergoing senescence.
Genes classified as ‘early up’ are marked as ‘present’, with expression values equal to or greater than 200 and ratios equal to or more than 2 at 1-day dark treatment. Those classified as ‘late up’ should fulfill the above criteria at 5-day treatment. Genes classified as ‘early down’ are marked as ‘present’, with expression values equal to or greater than 200 in control (0-d) data and ratio values equal to or less than 0.5 at 1-day treatment. The ‘late down’ genes should fulfill the above description but with a 0.5-fold or lower expression ratio at 5-day treatment.
Moreover, genes for 40% of the pathways were both up- and down-regulated in dark-treated Arabidopsis leaves. It would be interesting to pursue why contradictory regulation is needed for genes encoding enzymes in the same biochemical pathway. Many pathways of this kind are dedicated to the metabolism or biosynthesis of secondary metabolites in plants. It is conceivable that the specific up- or down-regulation of enzymes in such pathways will lead to either accumulation of preferential compounds or divergence of the metabolic routes as needed.
The transcription factor AtWRKY6 (At1g62300) has been shown to upregulate senescence- and pathogen defense-associated genes, possibly via recognition and binding to W-box (Robatzek and Somssich, 2001, 2002). The expression of WRKY53 (At4g23810), another WRKY family member, has been shown to be associated with leaf senescence (Hinderhofer and Zentgraf, 2001). In tobacco, two bZIP transcription factors are expressed in senescing tissues (Yang et al., 2001). Previous studies involving the 8K Arabidopsis Affymetrix Array have shown that 43 transcription factors are induced in Arabidopsis senescence triggered by pathogen and environmental stresses (Chen et al., 2002). In addition, the pathway analysis above revealed a clear upregulation of basal transcription factors and RNA polymerases (Table 2). Together with our results showing significant alteration in gene expression (Figure 2; Table 2), it is apparent that gene expression is highly orchestrated in senescing leaves. We therefore analyzed the expression profiles of 1201 transcription factors that belong to 35 superfamilies (nomenclature and annotations adopted from AGRIS; http://www.arabidopsis.med.ohio-state.edu/). The expression data are listed in Table S11. In summary, 303 and 81 transcription factors were up- and down-regulated, respectively, in dark-treated Arabidopsis leaves.
The overall effects of dark treatment on Arabidopsis transcription factors are illustrated in Figure 5. When we cross-examined both the up- and down-regulation behaviors of each transcription factor family, preferential or only upregulation was observed for only a few families, including TUB (recently re-annotated as TLP; Lai et al., 2004), GRAS, bZIP, WRKY, NAC, and C2H2. Interestingly, results of recent EST sequencing analyses also revealed high representation of genes from these transcription factor families in senescing Arabidopsis leaves (Guo et al., 2004). All TUB family members contain an F-box domain (Gagne et al., 2002; Lai et al., 2004), which may function in the SCF E3 ligase complex for the ubiquitin-mediated proteolysis pathway, as described above. For GRAS family members, research results mostly suggest their connection to root development and the GA responses (Fu et al., 2001; Ikeda et al., 2001; Pysh et al., 1999; Wen and Chang, 2002). However, Bolle et al. (2000) uncovered PAT1, another GRAS family member, in the phyA-mediated light signaling pathway, although the expression of this GRAS was not altered in our results. It would be interesting to illustrate whether the 11 dark-induced upregulated GRAS transcription factors are light dependent. Knowledge of NAC transcription factors is limited to their function in administrating meristemic cell differentiation and transducing auxin signals (Takada et al., 2001; Vroemen et al., 2003; Xie et al., 2000). More than one-quarter of NAC proteins were upregulated in dark-treated Arabidopsis leaves. Future research on the biological roles of NAC proteins in leaf senescence will shed light on the functions of NAC proteins in Arabidopsis.
It is also interesting to note that almost half of the bZIP transcription factors in Arabidopsis were induced in dark-treated leaves. Paradoxically, a few bZIP transcription factors have been shown to contribute to the light induction of gene expression by directly binding to the light-responsive element G-box. The best characterized cases are HY5 (Chattopadhyay et al., 1998; Osterlund et al., 2000), CPRFs (Kircher et al., 1998; Weisshaar et al., 1991), and GBFs (Schindler et al., 1992). The increased upregulation of bZIP transcription factors in darkness might reflect the negative regulatory role of bZIP proteins to ‘switch off’ genes expressed in light. Alternatively, in the absence of light, this subgroup of bZIP proteins may coordinate the upregulation of genes repressed by light. This suggestion coincides with our observation that the G-box element was predominantly present in dark-induced genes in our current study (data not shown). The upregulated bZIP proteins in our results also include those involved in ABA responses (ABF3, AREB2, AREB3, and ABI5; see Table S11 for details). Whether the upregulation of bZIP transcription factors reflects the ABA effect or light fluctuation, or both, is an intriguing topic.
Our study, involving whole-genome ATH1 Genome Array, clearly expands the scope of our current understanding of senescence-associated transcription factors. The results are especially valuable for future comparisons and clarifications of transcription factors of unique response to certain internal/external factors, especially because the senescence process is governed by multiple elements.
In general, the protein products of genes are well associated with their corresponding transcripts in eukaryotic cells (Ghaemmaghami et al., 2003). To date, the methodology and sensitivity for measuring transcripts are far more established than that for measuring their protein cognates. Thus, the genome-wide gene expression data could well be considered as reflecting the molecular events occurring in plants.
Here, we present the gene expression profiles of nine dark-induced senescence stages in three biological replicates of Arabidopsis leaves (Figures 1 and 2). The examination of gene expression data allows a global view of molecular events in the dark-treated leaves. When genes with annotated roles in biological processes were categorized according to their responses to the dark, those involved in photosynthesis and macromolecule biosynthetic pathways were generally downregulated, whereas those involved in the degradation pathway were upregulated (Figure 3; Table 2). We discuss the genes yielding protein products in degrading chlorophyll, proteins, and lipids (Table 1). The in-depth analyses of the expression data led to our postulation of an innovative pathway involving PPDK, which generates asparagine, presumably for nitrogen salvage in dark-treated leaves (Figure 4). To better envision the global metabolite modulation in senescing leaves, subsequent experiments could be performed in conjunction with metabolonomic measurement. For the coordination of the thousands of SAGs that are expressed, we report families of transcription factors that may take leading roles in regulating senescence (Figure 5). The wealth of resources and tools for Arabidopsis research can reveal the actual contribution of each participating transcription factor.
One of the prominent advantages of large-scale gene expression analyses is the possibility of global observation. Such observation is by no means conclusive but, rather, is more equivalent to an initiation of testable hypotheses. Massive information is buried in the genome-scale expression data. Extensive data mining and the integration of molecular events at large will certainly offer knowledge beyond what a gene list can do. The main theme of this current report is to provide another option for expression data analyses. By classifying genes according to their ontological and functional categories, one could extrapolate the gene expression data into information-driven experimental designs.
Plant material and treatment
Arabidopsis thaliana ecotype Col-0 was grown at 22°C under short-day conditions (8 h L/16 h D) with 100 μmol m−2 sec−1 white light (light switched on at 9.00 am every day). For dark treatments, Arabidopsis plants at stage 1.10 (10 true leaves; approximately 40–50 days after germination) were placed in the same growth chamber (22°C) without light (light switched off at dusk of the day prior to treatment). The plant materials for control (0 day), 1, 2, 3, 4, 5, 6, and control 6 days were harvested at 9.00 am at the days indicated. The 6 and 12-h samples were harvested at 3.00 and 9.00 pm, respectively, on day 1. Time point C6D represented samples from Arabidopsis plants continuing to grow for six more days without dark treatment, which were used as controls to verify the differential gene expression specific to dark treatment rather than to developmental alteration after 6 days of dark treatment. At each time point, 12 plants were harvested for RNA isolation and subsequent microarray hybridization. Samples from three independent biological replicates were used to generate the global gene expression profile.
Chlorophyll and total protein quantitation
Plant tissues were frozen and ground in liquid nitrogen. The powder was re-suspended in 96% (v/v) ethanol (3 mg tissue/1 ml ethanol) and incubated at room temperature in darkness for 30 min. After clarification by centrifugation, the chlorophyll content of the supernatant was quantified spectrophotometrically as described previously (Wintermans and DeMots, 1965). The remaining pellet was then rinsed once with 96% (v/v) ethanol, allowed to air dry, and resuspended in 60 μl of 1% (w/v) SDS, 1% (v/v) NP40, and 25 mm Tris–HCl, pH 7.5, by vortexing and heating for 30 min at 70°C. Protein was then quantified by use of the Bio-Rad DC Protein Assay Kit (Bio-Rad Laboratories, Hercules, CA, USA) (Weaver and Amasino, 2001).
Amino acid analyses
Plant tissues were harvested, frozen, and ground immediately in liquid nitrogen. The samples (about 25 mg) were resuspended in 175 μl of protein grinding buffer [0.1 m Tris–HCl, pH 8.0, 0.5% (v/v) 2-mecaptoethanol] supplemented with 300 nmol of nor-Leu as the internal standard for amino acid analyses. A total of 165 μl of each sample extract was mixed with 510 μl of methanol:chloroform (6:2.5, v/v) and then vortexed and incubated on ice for 30 min. After 450 μl of water and 300 μl of chloroform were added, the samples were vortexed again and spun for 30 s in a microcentrifuge. The upper layer of each sample was collected and dried in a speed vacuum system. The dried pellets were resuspended in 400–500 μl of lithium buffer as suggested by the manufacturer and filtered through 0.45-μm nylon filters before being analyzed with amino acid analyzers (System 6300, Beckman, Fullerton, CA USA; Lam et al., 2003).
Arabidopsis total RNA samples were isolated as described previously (Chang et al., 1993). In short, 1 g of plant tissues was frozen and ground in liquid nitrogen and resuspended with 8 ml extraction buffer (2% hexadecyltrimethylammonium bromide, 2% polyvinylpyrrolidone K 30, 100 mm Tris–HCl pH 8.0, 25 mm EDTA, 2.0 m NaCl, 0.5 g l−1 spermidine, 2%β-mercaptoethanol) pre-warmed at 65°C by vortexing. The homogenate was then extracted twice with an equal volume of chloroform:isoamyl alcohol (24:1, v/v) by vortexing and centrifugation for 10 min at 12 000 × g. One-quarter volume of 10 m LiCl was then added to the aqueous phase for selective precipitation of RNA molecules. After overnight incubation at 4°C, the RNA pellet was harvested by centrifugation at 12 000 × g for 20 min and dissolved in 200 μl RNase-free water. The population of mRNA was then isolated from total RNA with use of the Oligotex mRNA kit (Qiagen, Valencia, CA, USA).
DNA microarray fabrication and hybridization
The cDNA microarrays used in this study were printed on CMT-GAPS2-coated glass slides (Corning, New York, NY, USA) with use of the OmniGrid 100 microarrayer (GeneMachines, San Carlos, CA, USA) according to the manufacturer's instructions. After printing, slides were baked at 80°C for 6 h and blocked before hybridization. The slides were plunged up and down quickly and vigorously in the blocking reagent (6 g succinic anhydride dissolved in 335 ml 1-methly-2-pyrrolidinone and 15 ml 1-Na-borate, pH 8.0) for 20 s, and incubated in the blocking reagent for 20 min. To denature the printed DNA, slides were transferred into 95°C water for 2 min. After a brief rinse in 95% EtOH for 1 min, the slides were dried with speedvac (Savant SC210A, Savant Instruments, Inc. Holbrook, NY, USA). Methods for preparing the fluorescent probe preparation and hybridization were as described (http://www.botany.sinica.edu.tw/microarray/protocols.htm). The hybridization signals for each DNA element were acquired with the use of Axon GenePix 4000B and analyzed with use of GenePix 4.0 (Axon Instruments, Inc., Union City, CA, USA).
Affymetrix ATH1 Genome Array hybridization
ATH1 Genome Array hybridization was performed with the use of the Arabidopsis ATH1 Genome Array (Affymetrix Inc., San Jose, CA, USA), which contains more than 22 500 probe sets representing approximately 24 000 gene sequences on a single array. Total RNA from controls, 1 day, and 5 day dark-treated Arabidopsis rosette leaves were isolated as described above. All starting total RNA samples were quality assessed with use of the Agilent Bioanalyzer 2100 (Agilent Technologies, Palo Alto, CA, USA). The fluorescent cRNA was labeled and used to hybridize the ATH1 Genome Array as suggested by the manufacturer. The results were quantified and analyzed with use of the MicroArray Suite 5.0 software (Affymetrix Inc.).
Data analyses and organization
For DNA microarray data analysis, data files were imported into SMD, ASCC (http://www.bitora.sinica.edu.tw:7777/smd/MicroArray/SMD/) and GeneSpring 6.1 (Silicon Genetics, Redwood, CA, USA) for further analyses. ‘Default Computed Normalization’ and ‘LOWESS Normalization’ were applied for data normalization for SMD and GeneSpring, respectively. Expression data sets must pass all the following quality control categories before they are used for cluster analyses. First, the DNA element used for generating the microarray might be present as a single band when analyzed on agarose gel electrophoresis. Second, the hybridization results were not flagged as bad. Third, net intensities of both channels were equal to or greater than 500. Fourth, statistical analyses were applied to the triplicate data for each spot. Only the spots with a coefficient of variation less than 0.35 in at least one time point were used for further analyses.
For expression data shown in Figure 2, ratio values equal to or greater than three were considered upregulated. Ratio values equal to or less than 0.25 were considered downregulated.
One of the two Affymetrix replicate data sets was used for further analyses because of the high correlation of the two replicate datasets as described in the Results and discussion. ‘Set measurements less than 0.0 to 0.0’, ‘Per Chip: Normalize to 50th percentile’ and ‘Per Gene: Normalize to specific samples’ were applied for data normalization when Affymetrix data files were imported into GeneSpring 6.1 (Silicon Genetics) for further analyses. The average intensity of all probe sets of each chip was scaled to 500 so that the hybridization intensity of all chips was equivalent. Genes marked as ‘absent’ in all the chips analyzed were removed from further data analyses. Genes considered upregulated had to have ratio values equal to or greater than two, with intensity values greater than 200 and marked as ‘present’ in the experimental data (1 or 5 days). Genes considered downregulated had to have ratio values equal to or less than 0.5, with intensity values greater than 200 and marked as ‘present’ in the control data (0 day).
Hierarchical clustering in SMD and gene tree clustering in GeneSpring were used to organize the expression data into groups. For pathway organization shown in Figure 3 and Table 2, biochemical pathways in Arabidopsis were downloaded from the Kyoto Encyclopedia of Genes and Genomes (KEGG) (http://www.genome.ad.jp/kegg/) and imported to GeneSpring. Expression values of genes were superimposed on the corresponding pathway on the basis of their annotation with the ‘pathway’ function. Three specific pathways not included in the current analyses are ath00601 (blood group glycolipid biosynthesis pathway–lactoseries), ath00602 (blood group glycolipid biosynthesis pathway–lactoseries), and ath00521 (streptomycin biosynthesis). For transcription factor analyses shown in Figure 5, the lists for 1411 transcription factor genes were obtained from AGRIS (http://www.arabidopsis.med.ohio-state.edu/AtTFDB/index.jsp). A total of 1201 are represented on the Arabidopsis ATH1 Genome Array.
We thank Drs Tuan-hua Ho, Honyong Fu and Erh-Min Lai for helpful discussions and Drs Kuo-Chen Yeh and Ms Wei-Ning Huang for carefully reading the manuscript. We also thank Ms. Shu-Jen Chou, microarray facility, Institute of Botany, Academia Sinica, and Mr SJ Wang, Academia Sinica Computing Centre, for technical support with the gene expression profiling experiments. This research is supported by research grants to S-H Wu from Academia Sinica and National Science Council (NSC91-3112-P001-012-Y), Taiwan.
Figure S1. Venn diagram of genes with alteration in expression in response to dark treatment for 1 day (1D DT), 5 days (5D DT), or carbon starvation (CS). (a) Upregulated genes. Genes with ratios equal to or greater than 2, marked as ‘present’ and with intensity of experimental sample equals to or greater than 200 were considered upregulated in DT. For CS data, genes with ratios equal to or greater than 2 and marked as ‘present’ in both pgm/EN and 6/0 h were considered upregulated. (b) Downregulated genes. Genes with ratios equal to or less than 0.5, marked as ‘present’ and with intensity of control sample equals to or greater than 200 were considered downregulated in DT. For CS data, genes with ratios equal to or less than 0.5 and marked as ‘present’ in both control samples (EN and 0 h) were considered downregulated. Genes for all 14 patterns (Up_1 to UP_7 and Down_1 to Down_7) are listed in supplemental Table S3. Number of genes in each pattern is indicated in each corresponding group.
Figure S2. Scatter plot to show ATH1 hybridization reproducibility. Ratios of 1255 genes listed in supplemental Table S5–11 from two independent ATH1 experiments. (a) 1-day dark treatment. (b) 5-day dark treatment.
Table S1 Control: net intensity of control channel (from 0-day_control sample) normalized against each corresponding experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Criteria used for data analyses were described in Experimental procedures. The values within the parenthesis represent the data ranges from three biological replicates. Diurnal- or circadian-regulated genes are color-coded in blue or purple, respectively. Noted that all circadian-regulated genes are also diurnal-regulated. Genes showing inconsistent results between cDNA microarray and Affymetrix ATH1 are shaded in light blue.
Table S2 Percentage of diurnal and circadian-regulated genes in the seven expression patterns shown in Figure 2.
Table S3 Control: net intensity of control channel normalized against each corresponding experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. DT, dark treatment; CS, carbon starvation; EN, end of the night. Criteria used for DT data analyses are described in Experimental procedures.
Table S4 Control: net intensity of control channel (from 0-day control sample) normalized against each corresponding experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Criteria used for data analyses are described in Experimental procedures. Values within parenthesis represent the data range from three biological replicates.
Table S5 Control for 1 or 5 days: net intensity of control channel (from 0-day control sample) normalized against 1- or 5-day experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. Criteria used for data analyses are described in Experimental procedures. The genes up- or down-regulated are color-coded in red or green, respectively.
Table S6 Control for 1 or 5 days: net intensity of control channel (from 0-day control sample) normalized against 1- or 5-day experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. Criteria used for data analyses are described in Experimental procedures. The genes up- or down-regulated are color-coded in red or green, respectively.
Table S7 Control for 1 or 5 days: net intensity of control channel (from 0-day control sample) normalized against 1- or 5-day experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. Criteria used for data analyses are described in Experimental procedures. The genes up- or down-regulated are color-coded in red or green, respectively.
Table S8 Control for 1 or 5 days: net intensity of control channel (from 0-day control sample) normalized against 1- or 5-day experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. Criteria used for data analyses are described in Experimental procedures. The genes up- or down-regulated are color-coded in red or green, respectively.
Table S9 cDNA microarray data control: net intensity of control channel normalized against each corresponding experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Criteria used for data analyses are described in Experimental procedures. Values within parenthesis represent the data range from three biological replicates.
Table S10 Control for 1 or 5 days: net intensity of control channel (from 0-day control sample) normalized against 1- or 5-day experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. Criteria used for data analyses are described in Experimental procedures.
Table S11 Control for 1 or 5 days: net intensity of control channel (from 0-day control sample) normalized against 1- or 5-day experimental channel. Experimental: net intensity of experimental channel. Ratio: experimental/control, and color-coded in yellow. Flag: gene transcripts are present (P), marginal (M), or absent (A) in Affymetrix data acquisition. Criteria used for data analyses are described in Experimental procedures. The genes up- or down-regulated are color-coded in red or green, respectively.
Table S12 Percentage of correlation between cDNA microarray and ATH1 hybridization.