A Morning-Specific Phytohormone Gene Expression Program underlying Rhythmic Plant Growth

Based on reports that phytohormone abundance changes over the day and the observation that there is time-of-day–specific hypocotyl growth, we hypothesized that genes involved in the generation and action of phytohormones might also be regulated in a time-of-day fashion. To test this hypothesis, we first assembled a list of 182 “phytohormone genes” from the literature, that represent six phytohormone pathways: ABA, ACC, BR, CK, GA, and IAA (). Since our goal was to focus on the generation and action of phytohormones, we chose the phytohormone genes based on genetic and expression data implicating them in biosynthesis, catabolism, receptor, and signaling processes. The phytohormone gene list should be considered a tool rather than representing an exhaustive inventory of all known phytohormone genes. Table S1

Since it has been shown that growth is maximal at the dark-to-light transition (dawn) under short-day photocycles (8-h light/16-h dark) and light-to-dark transition (dusk) under circadian conditions of continuous light and temperature [15–18], we asked whether the phytohormone genes are overrepresented during these times of day. We developed a phase overepresentation graphing tool to help us determine the statistical significance of any observed enrichment at a specific time of day. This tool works by calculating the number of genes exhibiting peak expression at a particular time of day versus the number expected, and statistics are derived by permutation. We developed a Web interface called PHASER that enables any gene set to be searched for a time-of-day coexpression signature (http://phaser.cgrb.oregonstate.edu↗). Using our phytohormone gene list, we identified a highly significant overrepresentation (p < 0.00001) of phytohormone genes 1 h before (zeitgeber time 23; ZT23) and 1 h after dawn (ZT0) under short-day photocycles. Likewise, we found a similar overrepresentation (p < 0.00001) at subjective dusk under circadian conditions (circadian time, CT8 and 9; Figure 1A). The correlation between the time of maximum growth and overrepresentation of peak phytohormone transcript abundance suggested to us that there may be a connection between these two activities (Figure 1A, dotted lines) [15–18].

The shift from dawn to subjective dusk in maximum growth rate between short-day and circadian conditions [15] most likely reflects a resetting of the circadian clock associated with the release into continuous light after growth under light/dark cycles. When the plant experiences the first day in continuous light, the extended light period may be interpreted as a long day, and the phase of the circadian clock is reset, resulting in a “subjective” phase delay in growth rate and the observed phytohormone transcript abundance [19,22].

Under short-day photocycles, we noted a small group of genes overrepresented after dusk at ZT13, 14 (p < 0.01; Figure 1A and 1B). To find out whether the different clusters of phytohormone genes were due to a specific phytohormone group, we constructed phase overrepresentation plots for each individual class of phytohormone genes, and plotted their corresponding Z-scores (Figure 1B). When each phytohormone class was evaluated separately under short-day photocycles, CK and ACC genes were overrepresented during the dark period, whereas BR, IAA, GA, and ABA genes were overrepresented at or around dawn (Figure 1B). These results most likely reflect the distinct roles of these classes of phytohormone genes in transverse and longitudinal cell growth, respectively. We have focused on the phytohormone genes that are correlated with maximal elongation rate, which are the genes that display peak abundance at or around dawn.

While visually inspecting the expression patterns of the phytohormone genes, we identified two patterns of expression that correlated well with maximum growth over the day: genes that peaked directly at dawn, such as the biosynthesis (Figure 1C) and catabolism genes, and genes that increased during the dark period, such as the signaling (Figure 1D) and receptor genes. We summarized these two growth-associated patterns as “dawn-spike” and “dawn-box,” respectively, and chose two genes from each group as “models” to interrogate the 182 phytohormone genes for similar patterns of expression. Seventy-one genes had significant correlations with the growth-associated patterns (p < 0.01; 31 dawn-box and 40 dawn-spike), and we focused on these genes in all subsequent analyses (Figure 2 and Table S2).

It has been shown that some hormone-responsive genes also cycle over the day [23]. We reasoned that if phytohormone pathways are active during the growth phase of the day, then phytohormone mutants should preferentially affect genes that are normally expressed during the growth phase. To test this hypothesis, we analyzed the genes affected in selected phytohormone mutants by using publicly available Affymetrix microarray datasets in either the ArrayExpress or the Gene Expression Omnibus (GEO) databases (Materials and Methods). To identify the genes associated with the GA pathway, we looked at the DELLApenta (ga1-3 gai-t6 rga-t2 rgl1-1 rgl2-1) mutant [24]; for the IAA pathway, the auxin response factor (arf) arf6-2 arf8-3 mutant [25]; for the ACC pathway, the ein5-1 mutant [26]; for the BR pathway, the brx mutant [27], for the ABA pathway, the abscisic acid insensitive 1 (abi1-1) mutant; and for the CK pathway, the ckx1-ox mutant. We identified the genes that were differentially regulated for each mutant versus wild type (p < 0.01, Materials and Methods) and used this gene list to identify a time-of-day signature under short-day photocycles using PHASER (Figure 3). We found that the genes that were differentially regulated in these mutants were expressed at specific times of day under short-day photocycles. The phase of expression of the genes in the ABA, IAA, GA, and BR pathways are closely associated with the time of active growth under short day (Figure 3). These results are consistent with the ABA, IAA, GA, and BR pathways acting through genes that are expressed in the late night or early morning.

Our observations suggest that time-of-day–specific growth rate changes may be controlled in part through a coordination of phytohormone transcript abundance. This observation provides a testable hypothesis as to how plants organize their growth programs so that maximal growth rate is restricted to the correct time of day. In a previous study, we identified several cis-acting elements with a specific pattern of overrepresentation at different phases of the day. One of these (CACATG) was overrepresented in the promoters of cycling genes whose phase of expression was around dawn under short-day photocycles and subjective dusk under continuous light (Figure 4A) [19]. The CACATG consensus element was shown previously to be overrepresented in genes that respond to both BR and IAA treatment [28], and is related to the Ebox (CANNTG), which is the target of the BR transcription factors BES1 and BIM1 [29]. We named this element Hormone Up at Dawn (HUD).

We searched the list of 71 phytohormone gene promoters (500 bp) for overrepresented words, and found that the HUD was significantly enriched (Z-score = 5.2, corrected p = 9.8 × 10⁻⁰⁸; Table S3). Furthermore, the 54 phytohormone genes with the HUD in their promoter were overrepresented at dawn under short-day photocycles and dusk under continuous light (Figure 4B). To display the level of association between the phytohormone gene list and the presence of HUD element, we created a Venn diagram comparing the presence of the HUD in the 71 phytohormone genes associated with growth and the 111 phytohormone genes not associated with growth (Figure 4C).

To test the activity of the HUD element, we analyzed T3 transgenic plants carrying a promoter::luciferase fusion (3xHUD::LUC). A multimerized version of the HUD motif was sufficient to confer time-of-day activity to LUC in vivo under both short-day photocycles, as well as in continuous light (Figure 4D). The 3xHUD::LUC activity fit our prediction well in continuous light with peak LUC activity at dusk. However, the LUC activity deviated from our prediction under short-day photocycles, showing increased activity immediately after the plants experienced darkness. This pattern of expression is reminiscent of previous reports showing that circadian clock mutants grow immediately when moved into the dark, which was attributed to the circadian clock controlling or gating the rate at which light signaling is attenuated during the early evening [15]. This expression pattern in short days suggested to us that although the HUD is sufficient to confer circadian regulation, additional element(s) are required in the HUD-containing promoter context for light-regulated transcription in the early evening.

To directly test the hypothesis that transcriptional coordination of phytohormone transcript abundance affects hypocotyl growth, we conducted microarray time courses in mutants that have aberrant hypocotyl growth under light/dark cycles: two arrhythmic circadian mutants lux arrhythmo (lux-2) [30] and late elongated hypocotyl (lhy) [31], and one red-light photoreceptor mutant, phytochrome B (phyB-9) [32]. Table S1 provides information about each phytohormone gene examined, including the time of maximum transcript abundance for the cycling gene. In addition, the wild-type time courses confirmed growth-specific expression of the phytohormone genes.

To examine the behavior of the phytohormone genes in circadian and light signaling mutants with aberrant hypocotyl growth, we assembled two groups of genes, HUD-containing phytohormone genes associated with growth, and genes lacking the HUD that are not associated with growth (30 and 87 genes, respectively, Figure 4C). Because we found that the HUD element was sufficient to confer diurnal and circadian regulation, we reasoned that genes with the HUD element would be preferentially disrupted in the mutants. We compared the average fold change at each time point, between the mutant and its respective wild type, of the HUD-containing genes, with the average fold change between the mutant and its respective wild type in the genes lacking the HUD. As shown in Figure 5A and 5B, there is an increase in expression of the HUD-containing phytohormone genes during the dark phase in the two clock mutants, which is not found in genes lacking the HUD. It appears that phytohormone gene transcript levels increase during the dark period, similar to the pattern of growth in circadian clock mutants [15] and the 3xHUD::LUC reporter under short-day photocycles. This pattern of expression suggests that the circadian clock acts to control or gate the response of the HUD-containing phytohormone genes so that they are not expressed during the dark phase of the cycle.

In the phyB mutant, the HUD-containing phytohormone genes are not directly induced by darkness, but are constantly elevated as compared with the phytohormone genes without the HUD. This suggests that the response to the onset of darkness may be mediated via PHYB signaling pathways. To further test the association between darkness, growth, and the HUD-containing phytohormone genes, we analyzed their expression level in two other sets of time courses. In the first set, short day versus long day, the length of the dark period is twice that of the light period following a photoperiod difference. In the second set, seedlings entrained in the dark or light by temperature cycles were respectively released under continuous dark (DD) or continuous light (LL) conditions to monitor expression. For each of these comparable sets, the HUD-containing phytohormone gene expression is increased as compared to the genes lacking the HUD (Figure 5D and 5E). These results suggest that the expression of the HUD-containing phytohormone genes increases in the absence of light signaling (darkness or the light signaling mutant, phyB) and that, in part, the circadian clock specifically acts to abrogate this expression during the early part of the dark period. One explanation is that the circadian clock is required to maintain flux through downstream light signaling pathways in the early part of the night, so despite the plant experiencing darkness, HUD-containing phytohormone gene expression is moderate under long nights.

These results also suggest that the circadian clock acts upstream of light signaling in controlling expression during the dark period of the diurnal cycle. To resolve this, we looked more closely at expression in the lhy mutant. LHY is a single MYB transcription factor that acts in the core feedback loop of the circadian clock. It was previously shown in the lhy mutant that the peak expression of genes directly controlled by the circadian clock are shifted by 12 h (antiphase) compared to wild type (Figure S2) [30,33]. To quantify the timing change of phytohormone genes, we determined the phase of expression by fitting the dawn models to the lhy and parental (Landsberg erecta; Ler) time course. Of the 95 phytohormone genes that significantly matched the models in Ler (p < 0.01), 73 had the same phase in lhy (p < 0.01), nine were antiphasic to their Ler counterpart (negative correlation), and 13 were not significantly expressed. In the case in which the circadian clock would have directly controlled phytohormone expression, we would have expected a much higher proportion of genes with antiphasic expression like the core circadian clock genes. Instead, we found that most (77%) of the genes maintained phase with increased expression during the dark phase of the day, consistent with the circadian clock gating expression, and not directly controlling it. We noted that consistent with this interpretation, the expression of circadian clock genes was not affected in a phyB mutant, suggesting that PHYB (or light signaling) is downstream of the circadian clock (Figure S1). Taken together, these results support the notion that the circadian clock gates light signaling, which directly controls the phasing of HUD-containing phytohormone genes.

Our hypothesis that the light signaling pathway in the dark directly controls growth is further supported by the observation that the long hypocotyl phenotype seen under light/dark cycles in the circadian clock mutants early flowering 3 (elf3), elf4, and lux can to be rescued if plants are grown under continuous white light [30,34]. To confirm this result with other clock mutants, we showed that under continuous light, both lhy and lux-2 can also be rescued for hypocotyl growth (Figure 6A). Strikingly, we noticed that under short-day photoperiods, lhy mutants exhibit an elongated hypocotyl phenotype during its entire life cycle, but in continuous light, it grew essentially as wild type (Figures 6B and S3). In contrast, we could not rescue the long hypocotyl phenotype of phyB-9 regardless of the light condition (Figure 6A), suggesting that hypocotyl defects can be uncoupled from circadian dysfunction, but not from the loss of PHYB activity or a period of darkness such as in short-day photoperiod.

To test whether the rescue of hypocotyl growth in clock mutant and the long hypocotyl of phyB-9 could be associated with phytohormone transcript abundance, we measured the level of phytohormone gene expression in lux-2, lhy, and phyB-9 under continuous light using quantitative real-time PCR (qRT-PCR). Expression of IAA19, CKX5, and BR6ox2 were restored to wild-type levels in lux-2 and lhy, whereas they remained higher in phyB-9, which in all cases is in accordance with the observed hypocotyl growth phenotype (Figure 6C–6E). These results provide further evidence for a functional association between the transcription module that we identified and the control of hypocotyl growth rate. They also demonstrate that darkness, or the loss of light signaling (such as in phyB-9), is important in the control of phytohormone transcript abundance, whereas the circadian clock must play an indirect role, perhaps through modulation of light signaling. Finally, these results may explain why arrhythmic circadian clock mutants do not have hypocotyl defects under continuous white light, yet have very long hypocotyls under light/dark cycles [30,34].

Shade avoidance includes a plant's response to external lighting conditions when a plant experiences a change in the red (R) to far-red (FR) ratio (R/FR). The shade-avoidance response is characterized by increased stem elongation and changes in leaf morphology in mature plants that are similar to the etiolated responses in embryonic stems and leaves [35]. Similar to hypocotyl elongation, shade avoidance is gated by the circadian clock with maximal gating expression of PIL1 and growth response around dusk under continuous light [2], similar to the coordination of phytohormone transcripts under the same condition. To test whether there is a time-of-day signature in shade-responsive genes, we analyzed genes that are differentially expressed by 1 h of shade treatment (low R/FR) [36] using PHASER. Genes that are up-regulated by the low R/FR treatment (p < 0.01) are highly overrepresented around dawn (p < 0.0001) or dusk (p < 0.0001) under short-day and continuous conditions, respectively (Figure 7). Of the 62 low R/FR up-regulated genes, 16 are in our list of phytohormone genes, consistent with these genes playing a role in the plant's adjustment to shade conditions. These results suggest that phytohormone transcript coordination could be involved with stem growth (and, potentially, changes in leaf patterns) in mature plants.

To establish whether the regulation of phytohormone transcript abundance during the growth phase of the day affects growth, we made use of recently described lines in which the BR receptor (BRASSINOSTEROID INSENSITIVE 1; BRI1) is expressed using the Arabidopsis thaliana meristem layer 1 (AtML1) promoter in a strong bri1 mutant background, AtML1::BRI1; bri1-116 [37]. These lines, which contain an equivalent amount of BRI1 protein as the wild type [37], were shown to rescue the severe growth defects of bri1-116 under long-day photocycles (16-h light/ 8-h dark). Whereas the peak transcript abundance of both AtML1 and BRI1 transcripts occurs at dawn under long-day photocycles, consistent with rescue of bri1-116, under short-day photocycles, the peak of AtML1 transcript abundance shifts to midday, 12 h later than BRI1 (Figure 8A). Whereas under normal conditions BRI transcript abundance starts to increase in the dark with peak abundance near dawn, AtML1::BRI1 expression is highest during the beginning of the dark period. We reasoned that shifting the peak of BRI1 transcript abundance to the early evening with the AtML1 promoter should result in longer hypocotyls under short days due to the increased overlap between BRI transcript abundance and the dark period. Consistent with our hypothesis, we found that AtML1::BRI1; bri1-116 hypocotyls were slightly, but significantly, longer than wild type under short days (Figure 8B; p < 0.001). Hypocotyl length was the same as wild type under continuous light or dark, and long days as previously reported (Figure 8B) [37]. This result demonstrates that the timing of growth-associated phytohormone expression is important for wild-type growth. Furthermore, it suggests that BRI1 expression reflects BRI1 protein activity, and that more BRI1 activity during the dark increases hypocotyl growth. The modest increase in hypocotyl length may reflect the fact that there is only a limited set of components in the phytohormone growth pathway available at that time of day, suggesting that the internal coincidence with other phytohormones is also important for growth.

All 2-d (12 time point) time courses were described [19]. The short-day time course was in the Ler background. Plants were grown under short-day photocycles, 8-h light (180 μE/m²s)/16-h dark. The continuous-light time course utilized in this study was in the Columbia (Col) background. The seedlings for the continuous-light time course were grown under 12-h light (100 μE/m²s)/12-h dark for 7 d and then collected under continuous light and temperature over 2 d (12 time points). The phyB-9 [32] and lux-2 [30] mutants were in the Col background, and lhy [31] was in the Ler background. phyB-9 and lhy were grown under short-day conditions of 8-h light/16-h dark at 22 °C for 7 d and collected every 4 h over 1 d (six time points). The lux-2 plants were grown under intermediate-day conditions of 12-h light/12-hs dark at 22 °C, and collected every 4 h over 1 d (six time points).

The phytohormone gene list was assembled from the literature based on genetic or expression data implicating them in the biosynthesis, catabolism, signaling, or reception of phytohormones.

Phase overrepresentation was calculated as the number of genes with a given phase in a list, divided by the number of genes in that list, over the number of genes called rhythmic, divided by the total number of genes on the array. (Number of genes with phase X in a list/total number of genes in the list)/(Number of genes with phase X across on the array/total number of genes on the array). Phase overrepresentation is double plotted (one day of data plotted a second day) for visualization purposes. A Web-based implementation of phase overrepresentation plots called PHASER can be found at: http://phaser.cgrb.oregonstate.edu/↗.

All nonmutant microarray time courses were performed as described [19,22]. These time courses are available through our Web interface at DIURNAL: http://diurnal.cgrb.oregonstate.edu↗, DACARI: http://dacari.rutgers.edu/dacari/↗, and at ArrayExpress: E-MEXP-1304. Mutant microarray time course experiments were performed in the same manner. Briefly, tissue was collected and frozen in liquid nitrogen, pulverized, and then RNA was extracted using RNAeasy with DNAase on column treatment (QIAGEN), and labeled and hybridized to Affymetrix ATH1 GeneChip per Affymetrix protocol. Resulting CEL files were checked for array quality using standard tools implemented in the Bioconductor packages simpleaffy and affyPLM, and microarrays were normalized together using gcRMA [48]. Present/absent calls were made using the Affymetrix MAS5 program (Affymetrix).

Cycling gene calls and phase estimates were described for the short-day and continuous-light time courses [19]. Briefly, HAYSTACK, a model-based pattern-matching algorithm, was used to identify transcripts that change abundance over the day according to the hypothesis that if genes cycle- they will have a specific pattern that we can predefine (http://haystack.cgrb.oregonstate.edu↗) [22]. Transcripts were called cycling if their p-value for correlation to a specific pattern was less than 0.05. Phase estimates (in hours from dawn) were based on the time of peak transcript abundance [19]. For the 1-d time courses described in this study (phyB-9, lhy, lux-2, and associated parental genotypes Col, Ler, and Col, respectively), data were double plotted and a factor of noise (δ) was introduced in order to reduce autocorrelation. Cycling transcripts were called with a p < 0.05, greater than 20 unlogged gene-chip Robust Multiarray Averaging (gcRMA) expression, and greater than 1.5-fold change between minimum and maximum expression to control for false positives introduced from the noise factor (δ).

For mutant or condition comparisons, gcRMA normalized data were fit using a linear model in the R Bioconductor limma package with a p < 0.01 cutoff. Datasets were downloaded from the ArrayExpress or GEO Web site: AtGenExpress light treatments GSE5617↗ and tissue (7-d-old cotyledons, hypocotyls, and roots), E-TABM-17; shade avoidance (low R/FR), E-MEXP-443 [36]; ckx1-ox, E-MEXP-344; DELLApenta (ga1-3 gai-t6 rga-t2 rgl1-1 rgl2-1) mutant, E-MEXP-849 [24]; arf6-2 arf8-3, GSE2848↗ [25]; ein5-1, GPL198↗ [26]; abi1-1, GSE6151↗; brx, E-MEXP-635&7 [27]. Whole datasets were downloaded from the respective Web sites, and all CEL files from a given experiment were normalized together, regardless of whether all conditions in the experiment were used to determine genes disrupted in the indicated mutant. Resulting gene lists were evaluated for time-of-day signatures using PHASER.

Z-score profiles for all diurnal and circadian elements are described [19]. Z-score profiles represent the overrepresentation of a specified word (3–8 bp) in the promoter (500 bp) of all Arabidopsis genes on the ATH1 GeneChip for a given phase. Overrepresentation was determined using ELEMENT, which is an enumerative promoter searching algorithm (http://element.cgrb.oregonstate.edu↗) [19,22,49]. Z-score profiles are double plotted for visualization purposes.

Phytohormone genes associated with the growth phase of the day were identified using the HAYSTACK model matching algorithm [19]. Four growth-associated patterns were chosen from the 182 phytohormone genes that were used as the models to search for similar patterns. Two types of patterns were used: dawn-box and dawn-spike (Figure 2). The expression patterns of TIR1 and ERS2 were used for dawn-box, and AC8 and GAOX6 were used for dawn-spike models. The 182 phytohormone genes were searched using these four models, and 71 highly correlated genes (p < 0.01) were identified (Table S2). These 71 phytohormone genes were used for subsequent analysis.

qRT-PCR was previously described [50]. Plants were grown under either thermocycles (12-h 22 °C/12-h 12 °C) and continuous light or short days for 7 d, RNA was isolated every 4 h, first-strand cDNA synthesis was carried out with 5 μg of RNA, and all qRT-PCR reactions were run on a BioRAD myIQ system using SYBRgreen. qRT-PCR time courses are double plotted for visualization purposes. Data presented represent the results of two independent experiments. Primers will be supplied upon request.

Hypocotyl assays were performed as described [51]. For hypocotyl length measurements, roughly ten seeds were stratified on plates for 4 d at 4 °C in dark, and then transferred to specific growth conditions. Seven days later, plants were flattened and imaged on a flatbed scanner. Hypocotyl lengths were measured using NIH Image. Data presented represents the results of at least three independent experiments.

The 3xHUD::LUC construct was made by ligating two long oligos containing the HUD (CACATG) into a vector containing the −101/+4 fragment of the NOS minimal promoter and modified firefly luciferase (luc+). Plants transformed with the empty plasmids did not confer any cyclic pattern to luciferase (unpublished data). Plasmids were transformed into the Col-0 accession using the floral dip method [52]. Except where indicated, seedlings were grown on MS medium (Gibco BRL) with 0.8% agar and 3% sucrose. Seedlings of the T1 generation were selected on kanamycin and transferred to soil for bulking. T3 seedlings were grown under nonselecting conditions before imaging. Wild-type seedlings were identified after image collection and removed from the analysis. During the initial week of growth, seedlings were all grown under light/dark at continuous 22 °C (LDHH) conditions and then 2 or 3 d prior to imaging, transferred to the proper entrainment condition (short day or continuous light) on smaller plates without sucrose. Over the course of 5 d, images of seedlings were collected using a cooled charged-coupling device (CCD) camera for 25 min every 2.5 h using the Wasabi software (Hamamatsu Photonics) using the slice photon counting mode. The images were quantified using the MetaMorph software (Universal Imaging) and graphed using Microsoft Excel (Microsoft). For each condition tested, six seedlings from each independent T3 line were analyzed in triplicate. In order to compare all independent T3 lines, each time series was normalized based on the respective median value. The average of the six T3 seedlings was plotted for each treatment.

The Affymetrix ATH1 GeneChip data for the light and circadian mutants phyB-9, lhy, and lux-2 described in this paper have been deposited at ArrayExpress under accession number E-MEX-1299. The raw data and a Web interface are also available at DIURNAL: http://diurnal.cgrb.oregonstate.edu/↗ or DACARI: http://dacari.rutgers.edu/dacari/↗.