OPEN ACCESS ARTICLE

This Article Free via Open Access: OA OA Abstract Full Text (PDF) Supplemental Data OA All Versions of this Article: 146/4/1834    most recent pp.107.115592v1 Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via CrossRef Citing Articles via ISI Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by Usadel, B. Articles by Stitt, M. Search for Related Content PubMed PubMed Citation Articles by Usadel, B. Articles by Stitt, M. Agricola Articles by Usadel, B. Articles by Stitt, M.

SYSTEMS BIOLOGY, MOLECULAR BIOLOGY, AND GENE REGULATION

Global Transcript Levels Respond to Small Changes of the Carbon Status during Progressive Exhaustion of Carbohydrates in Arabidopsis Rosettes^1,[W],[OA]

	ABSTRACT

TOP ABSTRACT RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

 
The balance between the supply and utilization of carbon (C) changes continually. It has been proposed that plants respond in an acclimatory manner, modifying C utilization to minimize harmful periods of C depletion. This hypothesis predicts that signaling events are initiated by small changes in C status. We analyzed the global transcriptional response to a gradual depletion of C during the night and an extension of the night, where C becomes severely limiting from 4 h onward. The response was interpreted using published datasets for sugar, light, and circadian responses. Hundreds of C-responsive genes respond during the night and others very early in the extended night. Pathway analysis reveals that biosynthesis and cellular growth genes are repressed during the night and genes involved in catabolism are induced during the first hours of the extended night. The C response is amplified by an antagonistic interaction with the clock. Light signaling is attenuated during the 24-h light/dark cycle. A model was developed that uses the response of 22K genes during a circadian cycle and their responses to C and light to predict global transcriptional responses during diurnal cycles of wild-type and starchless pgm mutant plants and an extended night in wild-type plants. By identifying sets of genes that respond at different speeds and times during C depletion, our extended dataset and model aid the analysis of candidates for C signaling. This is illustrated for AKIN10 and four bZIP transcription factors, and sets of genes involved in trehalose signaling, protein turnover, and starch breakdown.

The diurnal cycle provides a defined experimental system to investigate how the supply and utilization of C is coordinated (Smith and Stitt, 2007; Stitt et al., 2007). Plants switch each day between a surplus of C in the light and a negative C balance at night. These diurnal changes are buffered by retaining some photosynthate in the leaves as starch in the light and remobilizing it at night to support Suc synthesis and export (Geiger and Servaites, 1994; Smith et al., 1997, 2005; Stitt et al., 2007). The importance of starch turnover is demonstrated by studies of mutants that are defective in starch synthesis or mobilization. They grow at the same rate as wild-type plants in continuous light or very long days, but show a large inhibition of growth in short days (Caspar et al., 1985, 1989; Lin et al., 1988; Schulze et al., 1991; Huber and Hanson, 1992; Zeeman et al., 1998; Zeeman and ap Rees, 1999; Gibon et al., 2004b). This inhibition is due to the transient depletion of C during the night (Gibon et al., 2004b).

Starch turnover is regulated in wild-type plants to avoid a shortfall of C at the end of the night. It is typically degraded in a near-linear manner, leaving only a small amount at the end of the night (Fondy and Geiger, 1985; Geiger and Servaites, 1994; Matt et al., 1998; Smith et al., 2004). When the C supply is decreased, for example, in short days or low irradiance, the rate of starch synthesis is increased and the rate of degradation is decreased. As a result, C reserves still last until the end of the night (Stitt et al., 1978; Chatterton and Silvius, 1979, 1980, 1981). Starch turnover starts to adjust on the first day after transfer from a long- to a short-day regime and, within 2 to 3 d, the levels of starch and sugars at the end of the night resemble those in long days (Gibon et al., 2004b). These observations raise fundamental questions about how plants sense changes in C availability over time and use this information to adjust and coordinate starch synthesis and breakdown with the use of C for growth.

Recently, Stitt et al. (2007) and Smith and Stitt (2007) proposed that plants respond to decreasing C in an acclimatory manner (i.e. where signaling triggers changes in storage, allocation, and growth before C falls so far that it exerts an acute limitation on metabolism or growth). This hypothesis provides a framework to understand how starch turnover, metabolism, and growth are coordinated to avoid C starvation. It makes several predictions. One is that sugars should inhibit starch synthesis. This prediction is counterintuitive because increased availability of C might be expected to stimulate starch synthesis. It was recently shown that starch synthesis is decreased when small amounts of Suc are included in the rooting medium of Arabidopsis (Arabidopsis thaliana) seedlings provided they are grown in short-day conditions where C is limiting (Gibon et al., 2004b; Stitt et al., 2007). Second, and crucially, changes in signaling should be initiated by relatively small changes in the C status before the onset of acute C starvation.

Transcript profiles provide the most comprehensive readout of signaling pathways that is currently available. It has been known for >20 years that sugars regulate gene expression in plants (Yu, 1999; Koch, 2004; Gibson, 2005; Rolland et al., 2006). Koch (1996) developed the concept of feast and famine genes; high sugar induces feast genes that are required for biosynthesis and growth, whereas low sugar induces famine genes that are required for C assimilation or the catabolism of alternative C sources. Prolonged C starvation or readdition of sugars to starved seedlings leads to changes of transcript levels for hundreds of genes that are involved in metabolism, signaling, and growth (Contento et al., 2004; Price et al., 2004; Thimm et al., 2004; Thum et al., 2004; Li et al., 2006; Osuna et al., 2007). However, these treatments are too drastic to reveal whether expression responds to small changes of C (Osuna et al., 2007). Thousands of genes show significant changes of their transcripts during diurnal light/dark cycles (Smith et al., 2004; Bläsing et al., 2005). Two lines of evidence indicate that C contributes to these changes. First, there is qualitative agreement between the diurnal changes of sugars and transcripts for many C-responsive genes (Bläsing et al., 2005). Second, the accentuated diurnal changes of sugars in pgm are accompanied by larger and more widespread changes of transcripts (Gibon et al., 2004b; Thimm et al., 2004; Bläsing et al., 2005). This is mainly due to depletion of C at night (Bläsing et al., 2005).

This article presents a more comprehensive analysis of the global response of transcripts to progressive depletion of C. Arabidopsis plants were darkened at the end of the night and metabolite profiling was performed to identify when C becomes severely limiting. This information was used to define times at which samples were taken for expression profiling during a transition to acute C deprivation. This dataset was combined with published datasets for diurnal changes of transcripts and analyzed to provide an overview of the global response of transcript levels during a gradual transition from high to acutely limiting C and to identify candidates for upstream components of the transcriptional response to small changes in the C status. It was also used to formulate and test a simple linear model, which predicts the global responses of gene expression in diurnal cycles from three inputs—the clock, light, and C. This allows us to validate the role of C in the diurnal regulation of a large number of genes and provides a framework in which the role of candidate genes can be quantitatively evaluated.

	RESULTS

TOP ABSTRACT RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

 

Changes of Metabolites in an Extended Night

Ecotype Columbia of Arabidopsis (Col-0) was grown for 5 weeks at 20°C in a 14-h light/10-h dark cycle at a moderate light intensity in well-fertilized soil. Whole rosettes were harvested at the end of the night and at different times after transfer into continuous darkness. Figure 1A shows the response of the metabolite profile and Figure 1, B to H, the responses of selected metabolites in another experiment with more time points.

View larger version (38K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 1. Changes of metabolites measured in an extended night. Arabidopsis Col-0 was grown for 5 weeks at moderate light intensity (130 µmol m–2 s–1) in a 14-h light/10-h dark cycle at 20°C in well-fertilized soil. After 5 weeks, plants were subjected to an extended night by not illuminating them at the start of the light period and whole rosettes harvested at the end of the night and at different times into the treatment. A, Changes of metabolites summarized as a heat map; all data are normalized on the values at the end of the night, with blue and red indicating an increase and decrease, respectively (see legend for scale). Gray indicates missing values. Some of the data are recalculated from Gibon et al. (2006). B to J, Starch (B), Suc (C), Glc-6-P (D), Fru-6-P (E), ATP (F), total amino acids (G), Asn (H), Phe (I), protein (J). The results in B to I are the mean and SD of five independent samples.

Organic acids like fumarate and malate are present at high levels in Arabidopsis (Chia et al., 2000). They show a gradual decrease when the night is extended, which becomes more marked from 48 h onward. Other C-containing metabolites also decrease, including raffinose, inositol, and several fatty acids, especially palmitolenoate (16:2) with less marked decreases of palmitate (16:0), stearate (18:0) and linoleate (18:2), eicosenoate (20:1), and hexacosanoate (26:0). Amino acids start to rise in the first hours of the extended night (Fig. 1, A and G). It has been proposed that amino acids like Asn (Fig. 1H) and Arg increase in C starvation because they have a high nitrogen (N)-C ratio (Lam et al., 1996, 1998). However, there is an equally large increase of low-N amino acids, including aliphatic (Leu, Ile, and Val; Fig. 1A) and aromatic amino acids (Phe, Tyr, Trp; Fig. 1, I and A), and smaller increases of Thr, Ala, and Glu (Fig. 1A). The only amino acids that do not increase are Asp and Gln. This general accumulation is unlikely to be due to increased synthesis; indeed, there is a marked decrease of shikimate (Fig. 1A), an intermediate in the aromatic amino acid synthesis pathway. The amino acids are likely to be released by protein degradation (Fig. 1J).

Expression Profiling during an Extended Night

Three separate experiments provided triplicate transcript profiles for the end of the night, duplicate samples 4, 8, 24, and 48 h into an extended night, and single samples 2 and 6 h into the extended night. Robust multiarray analysis (RMA) expression measures (Bolstad et al., 2003) were calculated on all arrays used. There was close agreement between the experiments. Pairwise scatter plots between the end of the night samples yielded R² values of 0.977, 0.975, and 0.979 (Supplemental Fig. S1a). Principal component analysis (PCA) showed samples for each time group together (Supplemental Fig. S1c). The original data are provided in Supplemental Table S1.

To identify genes that show significant changes of their transcript levels early in the extended night, a linear model was fitted with limma (Smyth, 2004) and corrected for multiple testing using a false discovery rate of 0.05 (Benjamini and Hochberg, 1995). Significantly changed genes are listed in Supplemental Table S2. After 4 h, 1,885 genes were induced and 2,147 were repressed, rising to 2,626 and 3,032 genes after 8 h, 3,135 and 3,388 genes after 24 h, and 3,929 and 3,888 genes after 48 h. Figure 2 shows the temporal responses of the 200 most strongly induced and the 200 most strongly repressed genes. Most show a rapid response that is near-saturated after 6 to 8 h. A smaller group shows a delayed response, with marked changes after 24 to 48 h.

View larger version (68K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 2. Temporal changes of global gene expression during an extended night. The plots show the changes of transcript levels of the 200 most strongly induced and the 200 most strongly repressed genes. Line shading is used to distinguish genes that respond early and later in the treatment.

We have published replicated datasets for global changes of transcripts (1) during diurnal cycles in wild-type Col-0 (Bläsing et al., 2005); (2) after illumination of Col-0 for 4 h at the end of the night at 50 and 350 ppm [CO₂] (the former prevents CO₂ fixation and carbohydrate synthesis (Bläsing et al., 2005); and (3) during the diurnal cycle in the starchless pgm mutant (where there are exaggerated changes of sugars; Gibon et al., 2004b; Bläsing et al., 2005). Related treatments grouped together when the extended night dataset was combined with these datasets and subjected to hierarchical clustering (Supplemental Fig. S2) or PCA (Fig. 3 ).

View larger version (8K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 3. PCA of an extended night treatment in Col-0 (x) combined with datasets for a diurnal cycle in wild-type Col-0 (d) and pgm (p). A cluster analysis of the same datasets is shown in Supplemental Figure S2. White symbols and black symbols represent day and (extended) night time points, respectively. The star, circles (), rectangles (), and triangles () represent time points –2, 4, 8, and 12 h into the day or night. Further, all extended night treatments are also marked with a closed triangle (). Treatments with ambient and compensation point [CO2] are represent by diamonds. White diamonds () represent the light treatment at ambient [CO2], black diamonds () the dark treatment at compensation point [CO2], and the half-black diamond represents the light treatment at compensation point [CO2].

The response to an extended night was investigated in plants growing in a 14-h light/10-h dark cycle, whereas the diurnal cycle and response to [CO₂] were analyzed in a 12-h light/12-h dark cycle (Bläsing et al., 2005). We checked whether this small difference in the photoperiod precludes joint analysis of the datasets. Hierarchical cluster analysis (Supplemental Fig. S2) and PCA (Fig. 3) showed that samples collected at the end of the night group together, regardless of whether they are from plants in a 12/12 (diurnal cycle) or a 14/10 cycle (extended night experiment). The dark controls from the low [CO₂] experiments (12/12 cycle) group with samples collected 4 h into the extended night (14/10 cycle). Samples collected at the end of the night in a 12/12 cycle were separated from samples collected 2 h after the end of the night in a 14-h light/10-h dark cycle, even though the plants had been in the dark for 12 h in both treatments. We calculated, for all >22K genes on the ATH1 array, the log₂-fold change 2, 4, 6, 8, 24, and 48 h into the extended night relative to the average level in the three end-of-the-night samples from the extended night experiments and the average level in the three end-of-the-night samples in the diurnal cycle dataset. Pairwise scatter plots of the ratios after 2, 4, 6, 8, 24, and 48 h yielded correlation coefficients of 0.88, 0.96, 0.97, 0.98, 0.98, and 0.99, respectively, which is comparable to or better than the agreement for biological replicates.

These results show that plants in 12/12 and a 14/10 light/dark cycle have similar transcript profiles at the end of the night, and show a similar response to an extension of the night. This conclusion was checked by fitting a linear model to the whole dataset and statistically analyzing the end-of-the-night samples from the 12/12 and 14/10 treatments with limma (Smyth, 2004). Only 917 genes had a P value <0.05, which is the proportion expected by chance (5% of approximately 22,000), only eight genes showed a >2-fold change, and no genes showed significant changes after correcting for multiple testing (Benjamini and Hochberg, 1995).

Contribution of Sugars to the Changes of Transcript Levels

To provide qualitative evidence that changes of C make a major contribution to the global changes of transcription in this large dataset, we compared the weighting of genes in PC1 (see Supplemental Table S3) with the changes of transcript levels in five published datasets where different treatments were used to alter C status. These were adding 15 mM Suc to C-starved seedlings for 30 min or 3 h (Osuna et al., 2007), adding 100 mM Glc to C-starved seedlings for 3 h (Bläsing et al., 2005), comparing seedlings in full nutrient medium with C-starved seedlings (Osuna et al., 2007), and illuminating rosettes for 4 h at 350 or 50 ppm [CO₂] (Bläsing et al., 2005). Regression plots yielded highly significant correlation coefficients (r) of –0.44, –0.5, –0.43, –0.47, and –0.6. With 1,000 shuffled datasets, most values of r were between 0.1 and –0.1, and none were >0.25 or <–0.25 (Supplemental Fig. S3).

Temporal Kinetics of the Responses of C-Responsive Genes during Diurnal Cycles and an Extended Night

A test set for C-induced genes was generated by combining the 200 most strongly induced genes 3 h after adding 15 mM Suc to C-depleted seedlings, the 200 most strongly induced genes 3 h after adding 100 mM Glc to C-starved seedlings, and the 200 most strongly induced genes after illuminating 5-week-old plants for 4 h in 350 compared to 50 ppm [CO₂] (Bläsing et al., 2005; Osuna et al., 2007). Because there was overlap, the test set contained 484 genes (Supplemental Table S4). An analogous procedure generated a test set of 383 C-repressed genes. Undirected and directed approaches were applied to test whether the temporal responses are consistent with these genes being regulated by changes of C during the diurnal cycle and the first hours of the extended night.

In the undirected approach, the C-repressed and C-induced genes were each separated into seven groups by K-means clustering (Fig. 4 ; genes are identified in Supplemental Table S4). Over 40% are in the clusters whose response is consistent with C regulating their expression during the night or early in the extended night. This includes cluster 8 (transcripts for 25 C-induced genes rise in the light and relax by the end of the night), clusters 9 to 11 (transcripts for 263 C-induced genes rise during the light period and relax during the night and first hours of the extended night), and cluster 1 (transcripts for 73 C-repressed genes decrease in the light and recover during the night and first 4 to 8 h of the extended night). Clusters 2 to 3 and 11 (representing 14% of the genes) are stable during the diurnal cycle, but respond strongly at the start of the extended night. A few genes did not respond until 24 to 48 h into the extended night (cluster 13). About 30% are in clusters that showed mostly <2-fold changes (clusters 4 and 12), and 10% showed qualitatively inconsistent responses.

View larger version (36K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 4. K-means clustering of the responses of sugar-responsive genes during the diurnal cycle and an extended night. The 200 genes that are induced and the 200 genes that are repressed most strongly 3 h after adding 15 mM Suc to C-depleted seedlings were downloaded from Osuna et al. (2007) and corresponding sets of genes for the responses 3 h after adding 100 mM Glc to C-starved seedlings and after illuminating 5-week-old plants for 4 h in 350 or 50 ppm [CO2] were downloaded from Bläsing et al. (2005). The genes were combined to provide a test set of 484 C-induced genes and 383 C-repressed genes. The responses of these genes in the diurnal cycle (Bläsing et al., 2005) and the extended night were extracted from Supplemental Table S4. All transcript levels were normalized on the level at the end of the night in the same experiment and K-means clustered. Clusters for C-repressed (clusters 1–7) and C-induced (clusters 8–14) genes are shown. The numbers of genes in each cluster are indicated.

The test sets used in the last section were compiled from treatments that lasted 3 to 4 h. The different temporal responses during the diurnal cycle and extended night might reflect differences in the timing of the signal to which they are responding or differences in the speed of their response. We therefore repeated this analysis with genes that respond rapidly to added Suc.

Osuna et al. (2007) short listed >150 genes that show marked changes (>1.41, log₂ scale) of their transcript levels within 30 min of adding 15 mM Suc to C-depleted seedlings. These were manually separated into groups based on the kinetics of their response during the diurnal cycle and extended night (Fig. 5A ; Table II ). Some show marked changes at the start of the light period that are reversed early in the night (groups 1 and 6), some show a slightly slower response that is almost completed by the end of the night (groups 2 and 7), some do not show large changes until the start of the extended night (groups 3 and 8), and some (group 4) do not respond until 24 to 48 h into the extended night. Because their transcripts respond 30 min after adding Suc to C-starved seedlings, the differing temporal kinetics during the night and extended night indicate that they respond to signals that are generated at different phases as C is depleted. The diurnal changes in pgm are summarized in the right-hand images. Responses that do not occur until the extended night in wild-type plants occur in the normal night in pgm, providing independent support that the responses in the wild type are due to the C status. A small group of genes is induced in the light in pgm, but not in wild-type plants (group 9), indicating they only respond to very high sugar.

View larger version (56K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 5. Temporal responses during a diurnal cycle and an extended night of a set of 158 genes, which respond rapidly to Suc addition. A, Clustering of rapidly responding genes. Osuna et al. (2007) short listed 158 genes as showing marked changes (>1.41, log2 scale) of their transcript levels 30 min after adding 15 mM Suc to C-depleted seedlings. The responses of these genes in the diurnal cycle (Bläsing et al., 2005) and the extended night in wild-type Col-0 and a diurnal cycle in pgm (Bläsing et al., 2005) were extracted from Supplemental Table S4. The genes were manually sorted into different response groups. The panels show groups of C-repressed (groups 1–5) and C-induced (groups 6–10) genes. The plots show absolute transcript levels. B, Heat map showing correlations between rapidly responding genes and other more slowly responding C-responsive genes. The genes in the rapid- and slow-responding groups were preordered based on internal clustering. Positive correlations are shown as pink and negative as mauve, with a cutoff filter of R2 > 0.7. The inset shows a frequency diagram of pair-pair correlations between members of a rapidly responding gene set and the other genes.

We investigated whether subsets of genes could be identified that might lie downstream of these early responders. To do this, we first clustered the rapid responding genes and all other C-responsive genes separately, and then correlated them with each other (Fig. 5B). Using a fairly stringent filter of R² > 0.7, about one-half of the rapidly responding genes were correlated with each other. This set of tightly clustered genes correlated with large sets of the more slowly responding genes.

Data Condensation to Identify Biological Processes

We next investigated which metabolic and cellular processes are subject to transcriptional regulation. The dataset comprising the global transcript profiles during the diurnal cycle and extended night was analyzed with the Pageman application (Usadel et al., 2006), which queries whether the response of the genes in a functional category differs from the response of other genes on the array. It uses an extensive plant-specific ontology developed in MAPMAN (Thimm et al., 2004; Usadel et al., 2005; http://gabi.rzpd.de/projects/Map-Man; TAIR6_MappingFile_ath_affy_tair6.m02), which contains >1,000 hierarchical and mainly nonredundant categories. This condenses the >22,000 features on an ATH1 array into about 1,000 features, many of which represent a defined biological process. Transcript levels were normalized on the values at the end of the night in the same experiment, the ratios averaged across the biological replicates, and imported into PAGEMAN to calculate the average change for the genes in a given category. To provide statistical support, we calculated Wilcoxon P values, which give the probability that the response of the genes in a given category is significantly different from the response of all other genes on the ATH1 array. An increasingly deep color indicates an increasingly large change of the average value or an increasingly significant P value. Blue and red distinguish between categories where expression increases or decreases, respectively. It should be noted that when a category contains a large number of genes, P values can be significant even though the average change is very small. When a category contains few genes, a large average change may not be significant. The full analysis is provided in Supplemental Table S5. Figure 6A provides a schematic representation of different types of response during the diurnal cycle and extended night.

View larger version (63K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 6. Ontology-based overview of the global responses of transcript levels during a diurnal cycle and an extended night. Transcript levels were expressed as a log2 ratio compared to the value at the end of the night in the same experiment. The response was averaged across all data points available for that time point. This procedure was repeated for each time point during the diurnal cycle and each time point during the extended night treatment. The left-hand image shows the average change of the transcript level (on a log2 scale) for all of the genes in a given BIN or subBIN. An increasingly deep blue or deep red color indicates an increasingly large average increase or decrease, respectively, of all the transcripts in a given BIN or subBIN. The right-hand image shows the P value (calculated using Wilcoxon's test) that the changes of transcript levels of genes in a given BIN or subBIN are significantly different from the response of all the other genes represented on the array. An increasingly deep blue or red color indicates an increasingly significant P value that transcript levels for genes in a given category increase or decrease compared to the level at the end of the night. The major BINS and subBINs are indicated on the left-hand side of the diagram. Selected BINS or subBINS that showed highly significant changes are indicated on the right-hand side of the diagram together with the P value. A, Schematic illustration of selected biologically relevant responses. B, Responses for selected functional categories related to metabolism and cellular growth. C, Responses for selected functional categories related to metabolism and cellular growth.

C depletion is accompanied by transcriptional repression of biosynthetic pathways and cellular growth processes that use C and induction of processes that mobilize C from alternative sources (Fig. 6B). Some of the responses are initiated in the night and others early in the extended night.

Categories showing a coordinated increase of transcripts in the light and a decrease that starts during the night include starch synthesis, glycolysis, amino acid synthesis, nucleotide synthesis, deoxynucleotide metabolism, many aspects of protein synthesis (amino acid activation, cytosolic ribosomal proteins, translation initiation, and elongation), protein folding, cell cycle, cell division, and histone synthesis. Categories showing a coordinated decrease of transcripts in the light, a small increase during the night, and a further increase early in the extended night include invertases, autophagy, and ubiquitin-regulated protein degradation. Slightly delayed responses are shown by genes in categories related to Suc synthesis, glycolipid synthesis, and plastid and mitochondrial metabolite transport, which are repressed after extension of the night, and genes related to gluconeogenesis, lipid degradation, and amino acid degradation, which do not show marked changes during the diurnal cycle but are induced early in the extended night. As already noted, metabolite profiling reveals that catabolism of alternative sources of C, including protein, commences early in the extended night (Fig. 1).

Other functional categories show more complex responses. For example, genes in categories related to photosynthesis (light reactions, Calvin cycle, photorespiration), chlorophyll synthesis, chloroplast biogenesis (plastid ribosomal proteins), pigment synthesis (isoprenoids, flavanols, isoflavonols), and nitrate and sulfate assimilation decrease during the light period and recover during the night, but decrease during the extended night (see Supplemental Fig. S4).

Coordinated Responses of Regulatory Genes

Some gene categories involved in signaling also show marked changes during the night and early extended night (Fig. 6C). Particularly striking responses are seen for TPS-like genes and a small BTB/POZ family. These genes respond within 30 min of adding Suc to C-depleted seedlings (Osuna et al., 2007; Table III ), indicating they are early components in transcriptional responses. Many genes involved in hormone metabolism, sensing, and signaling, several families of transcription factors and E3 ubiquitin ligases, and receptor kinases are also progressively induced or repressed, with the changes starting in the normal night and being completed a few hours into the extended night (a more expanded display of the response of these genes is provided in Supplemental Fig. S4).

Baena-Gonzalez et al. (2007) reported that overexpression of AKIN10 leads to changes of transcripts of >1,000 genes. This included widespread induction of genes that are involved in catabolism and repression of genes that are involved in biosynthesis and cellular growth, as well as repression of TPS8 to TPS11 and APG8e. This resembles the responses that we have seen during the night and early extended night (see above). We therefore investigated how all approximately 1,000 AKIN10-responsive genes behave in our dataset. K-means clustering (Supplemental Fig. S5) and Boolean analysis (Table I) illustrated that many of them show marked changes during the night or early in the extended night. Very few respond to extreme C starvation. This provides evidence that AKIN10 contributes to the regulation of expression in response to small changes in the C status.

The diurnal cycle and extended night are complex multifactorial responses in which many other inputs, including the clock and light, may be affecting gene expression. To deepen the analysis, we investigated how C interacts with these inputs.

Interaction between C and the Clock

Using a stringent filter, we extracted 604 genes with an unambiguous timing of a circadian peak from a dataset of constant light-grown plants (free-running cycle) by Edwards et al. (2006). This is smaller than the list of 3,503 genes in Edwards et al. (2006) because our stringent filter excludes clock-regulated genes whose response for various technical or biological reasons is less clearly defined. Table III summarizes the overlap between the clock- and C-regulated genes and uncovers an unexpected antagonistic interaction. Most overlapping genes fall into two categories; some show a circadian peak in the subjective light period (i.e. 0–12 h, when the light would be on in a light/dark cycle), but are repressed by C, and the others show a circadian peak toward the end of the night, but are induced by C.

This antagonistic interaction is explored in Figures 7 to 9 (see also http://mapman.mpimp-golm.mpg.de/supplement/xn). The left-hand images show, from left to right, the response of shoots after illuminating dark-grown seedlings with weak white light for 4 h (AtGenExpress), the response after illuminating rosettes for 4 h at 50 ppm [CO₂] compared to 4-h darkness (two criteria for light responsiveness), the response after illumining rosettes for 4 h at 350 ppm compared to 50 ppm [CO₂], and the response 30 min or 3 h after adding Suc to C-starved seedlings (two independent criteria for C responsiveness). They also show the response in a free-running cycle, a diurnal cycle, and extended night in Col-0, and a diurnal cycle in pgm.

View larger version (63K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 7. Antagonistic interaction between C and the clock. The images show, from left to right, the response of shoots after illumination of dark-grown seedlings with weak white light for 4 h (AtGenExpress), the response after illumination of rosettes for 4 h at 50 ppm [CO2] compared to 4-h darkness (two independent inputs for light responsiveness), the response after illumination of rosettes for 4 h at 350 ppm compared to 50 ppm [CO2], and the response 30 min or 3 h after adding Suc to C-starved seedlings (two independent inputs for the C responsiveness), the response in a free-running cycle over 48 h (input for the clock) and the measured changes during a diurnal cycle and extended night in Col-0, and a diurnal cycle in the starchless pgm mutant. The original data and identities of the genes are given in Supplemental Table S4. Transcript levels in the wild-type diurnal cycle, the pgm diurnal cycle are normalized on the levels at the end of the night in wild-type plants. Treatments with light and 50 ppm or 350 ppm [CO2] are normalized on the respective dark or 50 ppm [CO2]. Transcript levels in the extended night treatment are normalized on the levels at the start of the extended night treatment (i.e. the end of the normal night in these experiments). Transcript levels in a free-running cycle are normalized on an estimated level at the end of the 24-h period. Transcript levels after illuminating etiolated seedlings are normalized on the levels in nonilluminated controls. Transcript levels in seedlings in full nutrition and 30 min and 3 h after adding 15 mM Suc to C-starved seedlings are normalized on the levels in the C-starved seedlings. The plots show genes that are regulated by C and the clock (Table III) and peak 6 h (A) and 10 h (B) into the subjective day in a free-running cycle.

View larger version (59K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 8. Changes of transcripts for clock genes. The display is as described in the legend of Figure 7. A, Central clock genes. B to D, Genes that entrain the clock.

View larger version (60K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 9. Changes of transcripts for selected genes that are antagonistically regulated by the clock output pathways and C. The left-hand display shows, from left to right, the input s for light, C, and the clock (see legend of Fig. 7 for details), and the observed responses in a diurnal cycle and extended night treatment in Col-0, and a diurnal cycle in pgm. Transcript levels were normalized as in Figure 8. The right-hand images show the response predicted by our linear model (for details, see the legend to Table IV; also "Materials and Methods" and "Results"). A, Genes involved in Tre-6-P signaling: TPS5, TPS8, TPS9, TPS10, and TPS11. B, Genes involved in the regulation of protein synthesis and degradation: EF1B -subunit 1 (At5g12110), nucleolin organizer (NUC-L1, At1g48920), and autophagy protein 8e (ATG8e, At2g41570). C, Genes involved in starch degradation (AMY3, PHS1, ISA3, DPE2, PHS2, DPE2, SEX4, GWD, PWD, MEX1, BMY9, LDA1). D, bZIP1 (Atg5g49450), which is implicated in AKIN10 signaling (Baena-Gonzalez et al., 2007).

Responses of Clock Genes

We next investigated the responses of central clock genes (Fig. 8A) and genes that entrain the clock (Fig. 8, B–D). The central clock genes show a strong circadian response in a free-running cycle, with TOC1 changing reciprocally to CCA1 and LHY, as typically seen (Salome and McClung, 2005; Gardner et al., 2006). The phase is unaltered and the amplitude is only slightly increased in a light/dark cycle in Col-0 or pgm or during the first 8 h of extended darkness. The clock is disrupted after a 24- to 48-h extension of the night when the transcripts resemble those at the end of the subjective light period, although these times correspond to the start of the subjective light period. A range of responses was found for genes that entrain the clock. FKF1, PCL1, and PRR3 transcripts change in parallel in a free-running cycle. These genes are not regulated by light or C and their response is not unaffected in a light/dark cycle. It should be noted that cross-hybridization of the ATH1 probe sets is to be expected between FKF1 and PCL1. GI is strongly induced by light, but comparatively unaffected by C. GI transcript rises steeply in the subjective light period in a free-running cycle. This response is retained in a light/dark cycle and is not modified in the first 8 h of an extended night. PRR5, PRR7, and PRR9 are induced by light and repressed by C. They show slightly larger amplitudes in a light/dark cycle. Potential cross-reactivity might lead to artefacts for PRR9; however, given the similar response of PRR7 and PRR5, the data probably do reflect sugar and light regulation. ELF3, ADO2, and ZTL are repressed by C. Their free-running response is modified in a light/dark cycle. ELF3 shows a stronger minimum at the start of the light period, especially in pgm. ADO2 and ZTL do not show circadian changes in a free-running cycle, but show a small diurnal change in wild-type plants with a minimum at the end of the day and an increase in the night, which is amplified in pgm. Thus, whereas C status does not modify the central clock pacemaker, it does affect some of the entraining genes.

Responses of Genes Involved in Trehalose-6-P Signaling, Protein Turnover, and Starch Degradation

There is mounting evidence that trehalose-6-P (Tre-6-P) acts as a sugar signal in plants (see "Discussion"). Arabidopsis contains a small family of genes annotated as TPS (Lunn, 2007). In a free-running cycle, TPS8 to TPS11 show a peak and TPS5 a minimum during the subjective light period (Fig. 9A). TPS8 to TPS11 are strongly repressed and TPS5 is induced by sugars. This antagonizes the circadian response, resulting in a minimum of TPS8 to TPS11 and a maximum of TPS5 during the day, and a peak of TPS8 to TPS11 and a minimum of TPS5 at the end of the night. In pgm, the diurnal changes are larger and occur even more rapidly. The clock and C act in concert in an extended night; in this case, the clock leads to an increase of TPS8 to TPS11 and decrease of TPS5 transcripts, which is reinforced by the simultaneous depletion of C.

Figure 9B shows the response of three genes that might be involved in the regulation of protein turnover. AtNUCL1 plays a key role in nucleolus organization and is required for rRNA synthesis (Kojima et al., 2007; Pontvianne et al., 2007). AtNUC-L1 transcript is subject to weak circadian regulation with a peak in the light period and is strongly C induced. This additive combination leads to a marked diurnal change with a maximum in the light and a decrease at night in Col-0, which is accentuated in the pgm. Elongation factor (EF)--subunit 1 is circadian regulated with a minimum during the subjective light period, but is induced by C. In this case, C antagonizes the clock. ATG8e plays a key role in the regulation of autophagy (Doelling et al., 2002; Downes and Vierstra, 2005). ATG8e shows weak circadian regulation, peaking in the subjective light period, but is repressed by C. ATG8e transcript shows a strong decrease in the light, a recovery in the night, and a further increase early in the extended night.

Figure 9C shows the response of genes involved in starch degradation (Smith et al., 2005; Zeeman et al., 2007a, 2007b). Most show a marked circadian cycle with a peak toward the end of the subjective light period and a minimum at the end of the subjective dark period (Harmer et al., 2000; Lu et al., 2005), which is retained in a light/dark cycle (see also Smith et al., 2004; Bläsing et al., 2005). Almost all show a modified response in the first hours of extended night. The circadian increase is delayed and weakened for PHS1, ISA3, and DPE1 and almost abolished for PHS2, DPE2, GWD, PWD, and SEX4. The decrease is prevented by light alone (GWD) or a combination of light and ambient [CO₂]. MEX1, BMY9, and LDA1 do not show marked changes in a free-running cycle or a light/dark cycle in Col-0, but show a pronounced decrease in an extended night, and a small diurnal change in pgm. These results point to a potential mechanism to decrease the rate of starch breakdown when C is low at the beginning of the day (see "Discussion"). Surprisingly, the absolute transcript levels (data not shown) and the diurnal responses are very similar in Col-0 and pgm. This implies that expression of these genes is not regulated by events related to starch synthesis, starch accumulation, or starch degradation.

Light-Regulated Genes

Bläsing et al. (2005) concluded that light does not play a major role in the diurnal regulation of gene expression in Arabidopsis growing in a regular light/dark cycle. This unexpected conclusion was based on the diurnal response of a test set of approximately 400 light-regulated genes, identified from a treatment in AtGenExpress in which dark-grown seedlings were exposed to weak white light for 4 h. We used K-means clustering (Fig. 10 ) and directed grouping (Table I) to reanalyze the response of this test set in the enlarged dataset with the extended night treatment. Lists of the genes in the different groups are provided in Supplemental Table S4.

View larger version (36K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 10. Temporal responses of light-regulated genes during a diurnal cycle and an extended night. Test sets of approximately 200 light-induced and 200 light-repressed genes were abstracted from a treatment in AtGenExpress in which dark-grown seedlings were exposed to weak white light for 4 h. To investigate how these genes respond during a more prolonged exposure to darkness, the responses of this test set were reanalyzed in a dataset combining the diurnal cycle (Bläsing et al., 2005) and an extended night treatment. The genes in the test sets are listed and datasets are provided in Supplemental Table S4. Clusters 1 to 7 show light-induced genes and clusters 8 to 14 show light-repressed genes. The pie chart inserts show the proportion of the genes in a given cluster that are present in the sets of C-induced (blue) or C-repressed (red) genes (white, no overlap). Lists of genes in the various clusters are provided in Supplemental Table S4.

This raises the question of whether light signaling is attenuated by an interaction with the clock or C. About 13% of the light-repressed genes are present in the clock-regulated gene set. Almost all of these shared genes peak in the subjective light period (Table III), indicating that an antagonistic interaction with the clock contributes to the attenuated response of light-repressed genes. About 20% of the light-induced genes were present in the circadian set, but, in this case, the peak was at the end of the subjective night or the start of the subjective day (Table III). The pie diagrams in Figure 10 indicate the proportion of C-induced, C-repressed, and nonoverlapping genes in each cluster (blue, red, and white sectors, respectively). Many C-repressed genes in the clusters of light-repressed genes show little (9) or no (10, 11) change during the light/dark cycle, but decrease at the start of the extended night. This is at first sight surprising because additive action of light and C might be expected to generate exaggerated, rather than damped, changes during the light/dark cycle. One possible explanation is that C is still high enough during the night to repress these genes and override the effect of darkness and that a further decrease of C combined with darkness leads to strong induction when the night is extended. These clusters are enriched for genes involved in amino acid degradation (Supplemental Table S4). The situation is less clear for the light-induced genes. For example, there is no evidence that C contributes to the damped response during the light/dark cycle in the large cluster 3, which contains many (approximately 35%) genes related to photosynthesis and chlorophyll synthesis (Supplemental Table S4).

Modeling the Interaction between C, Light, and the Clock

The data analysis presented so far involved qualitative comparison of simplified experimental situations in which one input plays a dominant role and complex situations where there will be a multifactorial interaction between several inputs. It would clearly be advantageous to make the analysis more rigorous by performing it in the framework of a quantitative model. We investigated whether global transcriptional responses during the diurnal cycle in Col-0, the extended night in Col-0, and the diurnal cycle in pgm (Table IV ) can be predicted by a simple linear model with three inputs: the clock, light, and C.

In this model, the response of each individual gene to C is defined by the dataset from comparing rosettes after illuminating them for 4 h in the presence of 50 or 350 ppm [CO₂] (Bläsing et al., 2005). The response to light is defined by the dataset from comparing rosettes after 4-h additional darkness or 4-h illumination at 50 ppm [CO₂] in Bläsing et al. (2005). We use the dataset from Edwards et al. (2006) to define the response of each individual gene to the circadian clock six times during the 24-h cycle. The model attempts to predict the response of >22K genes from a linear combination of the response to sugar plus the response to light plus the response at a given time in the circadian cycle, each multiplied by a separate weighting factor. The weighting factor is varied to optimize the fit, but must be the same for all genes for a given input and time point (see "Materials and Methods" for details). For each time point in Col-0, signals for each gene were normalized on the signal at the end of the night in the same treatment. For pgm, the signals were normalized on the corresponding signals at the end of the night in Col-0 (as in Figs. 8–10).

Table IV summarizes the values for the weighting factors, the fit (R²) generated by the model, and, for comparison, the fit found by regression against each individual input dataset. The model performs reasonably well. First, it provides a much better fit to the measured global response than the individual inputs. The fit was especially good in an extended night and during the night in pgm, which highlights the importance of the low sugar signal during the night in pgm (Bläsing et al., 2005; Osuna et al., 2007). Second, the weighting factors chosen by the model for C and light were qualitatively correct (i.e. daytime points were more positive and the nights lower or, in the extended night or in pgm in the dark, strongly negative). In particular, the weightings for C show very good quantitative agreement with the sugar content. When the sugar levels were extracted from Figure 1, normalized to the levels at end of the night, logged, and plotted against the weighting factors, a correlation of >0.8 was obtained for Suc (Fig. 11A ) and reducing sugars (data not shown). This provides strong evidence that global transcriptional responses can be predicted from endogenous sugar levels.

View larger version (15K): [in this window] [in a new window] [as a PowerPoint slide]   Figure 11. Modeled contribution of C. A, Comparison of the weighting factors for sugar extracted from the models and the level of Suc in the tissue at the corresponding time point. The Suc levels are normalized on the reference (wild type, end of the night) and expressed as a log ratio of Suc. B, Estimated relative contribution of the clock, light, and C to the global transcript level. For each treatment and time point, the weighting factors generated by the model were calculated by scaling the variance of the corresponding input dataset to 1 to estimate their relative contribution. The variances of the input datasets before scaling were 0.01, 0.04, 0.11, 0.08, 0.08, 0.05, and 0.04 for CC at the 2-, 4-, 6-, 8-, 12-, 16-, and 20-h time points, 0.16 for light, and 0.41 for sugar.

Gene-Specific Assessment of the Quality of the Model

We assessed the quality of the prediction for each individual gene by plotting the measured and predicted values at all 17 modeled time points against each other and calculating the correlation coefficient (r). The average correlation coefficient was 0.41. The individual values are given in Supplemental Table S6. They allow us to investigate which sets of genes or individual genes are particularly well or badly predicted.

We first investigated model performance for genes that respond rapidly to Suc (>1.41 log₂ scale within 30 min of adding 15 mM Suc to C-starved seedlings; Osuna et al., 2007; see Fig. 5). The model performed very well (r = 0.72). However, some interesting exceptions were found. GPT2, a CONSTANS-like transcription factor, several heat-responsive genes, and MBF1C (At3g24500, a transcriptional coactivator) performed particularly badly. Interestingly, MBF1C responded reciprocally to the CONSTANS-like transcription factor (Supplemental Table S6). Noise can be excluded as an explanation for the discrepancies because the signals and replicates were good (data not shown). Possible explanations include strong nonlinear interactions or that these genes are regulated by an input that is not included by our model.

The model performed well with the test set of light-regulated genes that was identified from AtGenexpress (see Fig. 10; Supplemental Table S4; r = 0.62). This is interesting because most of these genes do not respond to light during the 24-h light/dark cycle (Fig. 10). The good performance of our model confirms that this is at least partly due to an interaction between light signaling and circadian and C signaling. We also tested whether clock-regulated genes performed qualitatively well in the model and whether this could partially be attributed to the presence of cis-elements