LUX ARRHYTHMO Encodes a Night Time Repressor of Circadian Gene Expression in the Arabidopsis Core Clock

Summary

Circadian clocks provide an adaptive advantage by allowing organisms to anticipate daily and seasonal environmental changes [1, 2]. Eukaryotic oscillators rely on complex hierarchical networks composed of transcriptional and post-translational regulatory circuits [3]. In Arabidopsis, current representations of the circadian clock consist of three or four interlocked transcriptional feedback loops [3, 4]. Although molecular components contributing to different domains of these circuits have been described, how the loops are connected at the molecular level is not fully understood. Genetic screens previously identified LUX ARRHYTHMO (LUX) [5], also known as PHYTOCLOCK1 (PCL1) [6], an evening-expressed putative transcription factor essential for circadian rhythmicity. We determined the in vitro DNA-binding specificity for LUX by using universal protein binding microarrays; we then demonstrated that LUX directly regulates the expression of PSEUDO RESPONSE REGULATOR9 (PRR9), a major component of the morning transcriptional feedback circuit, through association with the newly discovered DNA-binding site. We also show that LUX binds to its own promoter, defining a new negative autoregulatory feedback loop within the core clock. These novel connections between the archetypal loops of the Arabidopsis clock represent a significant advance towards defining the molecular dynamics underlying the circadian network in plants and provide the first mechanistic insight into the molecular function of the previously orphan clock factor LUX.

Results and Discussion

LUX selectively binds DNA

To determine whether LUX can bind to DNA and to identify potential target sequences, we made use of universal protein binding microarrays (PBMs). PBM technology provides a rapid means of comprehensively characterizing the in vitro DNA-binding specificities of transcription factors, regardless of structural class or species of origin [7]. We used custom-designed DNA microarrays consisting of ~44,000 60-mer oligonucleotides that collectively represent all possible 10-bp DNA sequences. Each 8-mer is represented at least 16 times on the array, providing a comprehensive and quantitative in vitro assessment of binding preferences for a given protein. We performed triplicate PBM experiments on two different “all 10-mer” designs, for a total of 6 replicates. Full-length GST-tagged LUX protein was produced in Escherichia coli, purified, and applied to the DNA microarray; LUX binding at each DNA spot was detected and quantified using fluorescence-conjugated anti-GST antibody. To determine LUX binding preferences over all 8-mers, the 60-mer probe data were used to calculate enrichment scores (E-scores). E-scores reflect the relative preference of the protein for binding each 8-mer; E>0.45 is indicative of strongly preferred binding sequences [8]. The highest ranked 8-mer bound by LUX was AGATACGC (E= 0.487) (Figure 1B; Tables S1 and S2). Variations in the first or last position in this 8-mer did not greatly affect the E-score, while changes at the third or fourth position drastically decreased it, bringing it to negative values (Figure 1B). The overall LUX binding site (LBS) motif can be represented as GATWCG (where W indicates A or T) (Figure 1A; Table S3). To confirm binding to the identified LBS sequence, we constructed synthetic multimers carrying four copies of the LBS or four copies of mutant versions of the LBS with single or multiple mutations. These multimers were cloned upstream of a yeast minimal promoter:: β-galactosidase (lacZ) transcriptional fusion, and the generated reporters were used in a yeast one-hybrid system to test binding of LUX fused to a GAL4 activation domain (LUX-GAL4AD). LUX bound to the multimerized LBS, but not to the mutated sequences, confirming that LUX selectively binds the sequence GATWCG (Figure 1C). We conclude that LUX is a sequence-specific DNA-binding protein and that we have determined its DNA binding specificity.

Figure 1. LUX is a sequence-specific DNA-binding protein.

(A) LUX DNA binding site motif determined by universal protein binding microarray (PBM) experiments. (B) Effect of point mutations on DNA binding affinity. The scatter plot shows enrichment scores (E-scores) for two “all 10-mer” microarrays of different design; for each design, the E-scores from 3 replicates were averaged. The E-score correlates with the binding affinity of LUX for the sequence, and is measured on a scale of −0.5 (worst) to 0.5 (best). Spots containing the 6-mers GATACG and GATTCG are marked in red and orange, respectively; spots containing the variants GGTACG and GATGCG are marked in green and blue, respectively. Below the scatter plot, E-scores are shown for variants at each position of the most preferred 8-mer AGATACGC. (C) LUX binding to synthetic multimers of the binding motif in a yeast one-hybrid system. Perfect match or mutant versions of the binding motif were multimerized and cloned upstream of a minimal promoter::LacZ transcriptional fusion. Bars represent the fold of induction in β-galactosidase activity in the presence of LUX-GAL4AD over control plasmid (means ± SEM, n=6 independent experiments). The selected mutations were predicted to abolish binding, based on PBM E-Scores.

See also Tables S1–S3 and Figure S1.

In most systems, distinct transcription factor (TF) families participate in clock networks. Additionally, it is common that several members of a particular family are involved. CLOCK/CYCLE and CLOCK/BMAL1 are basic helix-loop-helix (bHLH) PAS domain TFs that interact to form an activator complex in Drosophila [9, 10] and mammals [11], respectively. In Arabidopsis, the morning Myb-like transcription factors CIRCADIAN CLOCK-ASSOCIATED1 (CCA1) and LATE ELONGATED HYPOCOTYL (LHY) homo- and hetero-dimerize to function as a repressor complex [12, 13]. LUX belongs to a small family of five proteins with a single DNA-binding domain unique to plants, part of the larger group of Myb-like GARP transcription factors [5, 6]. Among these, a protein encoded by a gene on chromosome 5, At5g59570, is most similar (72 %) to LUX. In particular, their DNA-binding domains share 97 % identity (Figure S1A). In addition, its expression pattern, like that of LUX, is circadian regulated with a peak in the evening (Figure S1B, C); we therefore called this gene NOX (from the Latin word for “night”). As T-DNA insertion lines for NOX are not available, we generated RNAi lines to observe the effects of reduced NOX levels. Contrary to lux mutants, they displayed robust circadian rhythms (Figure S1D–F), suggesting that NOX is not fully redundant with LUX. However, NOX was able to bind to the multimerized LBS in the yeast one-hybrid system but not to multimers of LBS variants (Figure S1D).

TF families are greatly expanded in plants relative to other organisms [14]. However, the DNA-binding specificities for most plant TFs are still unknown due to lack of high-throughput studies like those conducted in yeast or mouse [15, 16]. We show that LUX is a transcription factor and determined its DNA-binding specificity. The DNA-binding domain in the LUX family is also found in genes similar to two-component signal transduction systems: the B-type Arabidopsis response regulators (ARRs), GOLDEN2-LIKEs (GLKs), and PSEUDO RESPONSE REGULATOR2 (PRR2). It is distantly related to the authentic Myb R1R2R3 repeat originally identified in the mammalian c-Myb oncoprotein [17]. A previous report proposed AGATT, which shares 4 nucleotides (GATT) with the LBS, as the best target sequence for the B-type ARRs ARR1 and ARR2, based on in vitro studies using their Myb-like motif fused to GST [18]. Subsequently, a study on ARR10 showed through several in vitro techniques that its optimal recognition sequence is also AGATT [19]. The high degree of primary sequence conservation among the DNA binding domains of these GARP proteins and LUX (Figure S1A) is consistent with the ability to bind the same core DNA sequence. Furthermore, 9 of the 11 amino acid residues shown by Hosoda et al. [19] to be critical to the ARR10-DNA interaction are conserved in LUX. The two amino acids substitutions (L187->V148 and A237->Y198) are also present in NOX, which we found to also bind the LBS sequence in yeast one-hybrid assays. It is interesting to note that, while LUX and NOX can bind to the same DNA sequence, they are not functionally redundant, as shown by the arrhythmic phenotype of the single lux mutant. The actual overlap between the overall DNA-binding specificities of LUX and NOX has not been investigated yet and might give insight into how these two closely related TFs achieve distinct functions. The functional discrepancy might be due to differences in expression patterns, differences in DNA-binding preferences, or interaction with different proteins that modulate binding activity or transcriptional activity, giving them separate sets of targets. Similarly, the REVEILLE1 (RVE1) gene, belonging to the same single Myb-domain subfamily as CCA1 and LHY, is clock regulated with a morning peak and binds to the evening element, like CCA1. RVE1, however, has a distinct function and was shown to be primarily a clock output [20]. Using the universal PBMs, we showed that the LUX binding site motif is longer than the core recognition sequence shared with ARR1, ARR2, and ARR10. Comparing the DNA-binding profile of LUX with other DNA-binding sites that might be identified in the future for other family members will significantly help to refine the molecular basis for DNA recognition by the GARP family of TFs, which has over 50 members in Arabidopsis [14].

LUX associates with PRR9 and LUX promoters in vivo

We have initially focused on the role of LUX in regulating genes in the central oscillator. The LBS was found in the promoters of several clock genes, including the morning genes PRR9 and PRR7 and the evening genes LUX and EARLY FLOWERING4 (ELF4). In particular, PRR9 contains a perfect match LBS (GATTCG) 166 bp upstream of the transcriptional start site. We generated strains for yeast one-hybrid, harboring different PRR9 promoter fragments designed around the LBS (Figure 2A). LUX bound to the fragment containing the LBS, but not to the shorter fragment (consisting of the 5’-UTR only) lacking this site. Additionally, mutating the LBS abolished binding (Figure 2B), confirming that this sequence is responsible for specific binding of LUX to the PRR9 promoter. To investigate LUX binding in vivo, we generated transgenic plants expressing a C-terminal fusion of LUX to GFP under control of either a constitutive promoter (35S::LUX-YFP) or the native LUX promoter (LUX::LUX-GFP) in the lux-4 mutant background. The lux-4 allele [5], also characterized as pcl1-1 [6], carries a nonsense mutation. Thus, in lux-4 plants a full length transcript is generated but only encodes a 149 amino acids protein, truncated at the beginning of the DNA binding domain. Both lux-4 35S::LUX-YFP and lux-4 LUX::LUX-GFP lines had restored circadian rhythms (Figure S2A, B), showing that the LUX-GFP fusion is functional. Although LUX overexpression was reported to cause arrhythmicity after several days in constant conditions [6], our lux-4 35S::LUX-YFP lines maintained robust rhythms (Figure S2A). Previously described overexpression lines were in a wild-type background, which might explain the discrepancy with our observations. In the lux-4 LUX::LUX-GFP lines, hypocotyl growth was restored (Figure S2C), and LUX-GFP transgene expression followed the expression profile of LUX in wild-type plants (Figure S2D–G), indicating that the LUX-GFP fusion is a good proxy for the native LUX protein. We therefore used these lines for chromatin immunoprecipitation (ChIP) experiments to determine whether LUX is bound to the PRR9 promoter in vivo. Samples were collected at Zeitgeber Time 14 (ZT14) during the first day in LL, when LUX protein levels are at a maximum (E. Hamilton and S. Kay, personal communication) and processed for ChIP using an anti-GFP antibody. We analyzed by quantitative PCR (qPCR) different target amplicons from the PRR9 promoter, as shown in Figure 2A. Amplicons located in the PRR9 coding sequence and in the promoter of the UBIQUITIN (UBQ) gene served as negative controls for binding. Several amplicons showed significant enrichment in lux-4 LUX::LUX-GFP over wild-type control plants (Figure 2C). Amplicon 0, centered on the LBS, showed the greatest enrichment. Amplicons upstream (−150, −500) and downstream (+200) of the LBS also showed enrichment, with lower values as the distance from the LBS increased. This enrichment is likely due to size variation in DNA fragments produced during sonication (on average 500bp). As LUX binds to the LBS, adjacent regions are also expected to be pulled down with LUX. As expected, enrichment was not observed for more distal amplicons in the PRR9 coding sequence or in the UBQ promoter. These results show that, as observed in vitro, LUX binds to the PRR9 promoter in vivo.

Figure 2. LUX binds to PRR9 and LUX promoters in vivo.

(A) Schematic of the PRR9 promoter (+1 is the transcriptional start site). The black arrowhead indicates the LBS; white arrowheads indicate degenerate binding sites (GATWKG or GATWCY where K indicates C or T, and Y indicates G or T). White rectangles represent the promoter fragments used for yeast one-hybrid assays in (B), with numbers relative to the transcriptional start site. Grey rectangles show the amplicons used in the ChIP experiments (C) and are numbered using the LBS as a reference (positions relative to the transcriptional start site are detailed in Table S4). (B) Binding of LUX to PRR9 promoter in yeast. Bars represent the fold of induction in β-galactosidase activity in the presence of LUX-GAL4AD over control plasmid (n=4 independent experiments). −243/+225(LBSm) is the −243/+225 fragment with a mutated LBS (GATTCG ->TCGGAT). (C) Binding of LUX to the PRR9 promoter in vivo. ChIP assays were performed with wild-type CAB2::LUC (wt) or lux-4 LUX::LUX-GFP (LUX::LUX-GFP) seedlings. Plants were grown under 12 h light/12 h dark (LD) cycles and transferred to continuous light (LL). Samples were collected from two independent lines (#43 and 49) at Zeitgeber Time 14 (ZT14) during the first day in LL, and processed for ChIP using an anti-GFP antibody. The immunoprecipitated DNA was quantified using real-time polymerase chain reaction (RTPCR) with primers specific for the amplicons represented in (A). UBQ, UBIQUITIN; CS coding sequence. Results were normalized to the input DNA (n=3 independent experiments). (D) Schematic of the LUX promoter. Black and white arrowheads indicate the LBS and degenerate LBS sequences, respectively, as described in (A). The grey rectangle shows the amplicon centered on the LBS used for ChIP assays. (E) Binding of LUX to its own promoter in vivo. The ChIP assays were performed as described in (C), with regions of the UBQ promoter or LUX CS as negative controls. Values represent means ± SEM in (B), (C), and (E). All primer sequences are detailed in Table S4.

See also Figure S2.

Interestingly, although the promoter of PRR7 has a perfect match LBS and PRR7 has a similar expression pattern to PRR9, LUX did not bind this promoter region in yeast one-hybrid or ChIP assays (data not shown). PRR9 and PRR7 are often considered at least partly redundant. For instance, they are wired in the morning loop as negative regulators of CCA1 and LHY, which in turn activate both PRR9 and PRR7 expression [21, 22]. PRR9 and PRR7 also both participate in temperature entrainment of the clock [23]. However, it has also been shown that PRR9 and PRR7 have overlapping but distinct roles in the circadian clock. The single mutant phenotypes have different light quality dependencies while an additive phenotype is observed in the double mutant [21, 24]; they display distinct overexpression phenotypes as well [25–27]. Although PRR9 and PRR7 have been incorporated as a single component in some mathematical models of the Arabidopsis clock [28], they have been separated in others [4, 29]. Additionally, both mathematical modeling [4] and experimental studies [22] indicate that the sequential expression of PRR9, PRR7, and PRR5 as a “wave of inhibitors” is required for proper repression of CCA1 and LHY expression from morning until mid-night. Here we provide additional evidence of the distinction between the morning genes PRR9 and PRR7 at the level of transcriptional regulation.

Clock oscillator genes in several organisms are known to control their own expression level by negative feedback [30–32]. It has been previously reported that overexpression of LUX represses endogenous LUX expression and disrupts its circadian expression [6], suggesting that it might be part of an autoregulatory feedback loop. We found that the perfect match LBS present in LUX promoter is bound by LUX in vivo, as shown by ChIP assays using the lux-4 LUX::LUX-GFP lines (Figure 2D, E). The region surrounding the LBS in the promoter was specifically enriched, whereas regions used as negative controls (UBQ promoter and LUX coding sequence) were not. This result combined with the observation that LUX transcript levels are constitutively high under constant light conditions in the lux mutant [6] suggests that LUX defines a new negative autoregulatory feedback circuit within the core clock mechanism. In plants, as in other organisms, self-regulation is a widespread mechanism used to achieve rapid and tight control, and is used in cell cycle regulation (reviewed in [33]) and ethylene signaling (reviewed in [34]). It is also a common feature in clock transcriptional feedback loops, although it appears mostly indirect, i.e., involving the activation of a repressor or the repression of an activator rather than direct self-repression. In Drosophila, the CLOCK/CYCLE (CLK/CYC) heterodimer binds to E-boxes to activate the key clock genes PERIOD (PER) and TIMELESS (TIM). PER and TIM proteins then interact and inhibit CLK/CYC activity [9, 10]. PER-mediated transcriptional repression is associated with the rhythmical binding of PER to circadian promoters, in particular PER and TIM promoters [35]. In the Arabidopsis circadian system, CCA1 has been proposed to regulate itself, based on repression of the endogenous transcript in CCA1 overexpressing lines [32]. Additionally, the CCA1 promoter contains a CCA1 binding site [36], a motif shown to be bound by CCA1 in vitro and in vivo [21, 37], although the binding of CCA1 to this element in its own promoter has yet to be confirmed. Similarly, increased expression of LHY caused the endogenous gene to lose rhythmic expression, suggesting that LHY may also be part of a feedback circuit that regulates its own expression [31]. There is, however, no evidence for this type of self-regulation for TOC1 [38]. We provide here the first in vivo evidence in Arabidopsis of direct self-regulation of a clock transcription factor through binding to its own promoter, indicating that fine-tuning of LUX levels may be important for proper clock function.

LUX function is lost when fused to a strong activator domain

Because PRR9 and LUX are antiphasic (Figure 3A, B) and LUX binds to the PRR9 promoter, we reasoned that LUX likely acts as a transcriptional repressor. Therefore, we measured PRR9 expression in the lux-4 mutant, starting at the beginning of the night, when LUX levels are peaking, and extending through the first day in constant light, since the mutant becomes arrhythmic after release from driven to constant conditions [5]. We found PRR9 expression to be higher than wild type levels throughout the entire time course (Figure 3C), consistent with a repressor activity for LUX. To investigate further, we generated transgenic lines overexpressing LUX fused to either the VP64 activation domain [39] (35S::LUX-VP64) or the CRES repressor domain [40] (35S::LUX-CRES) in the lux-4 mutant background and monitored circadian rhythms. Interestingly, we found that the lux-4 35S::LUX-CRES lines had robust rhythms, showing that the repressor fusion restored rhythmicity (Figure 3D, E). On the contrary, expression of the LUX-VP64 activator fusion did not complement the lux-4 arrhythmic phenotype (Figure 3D, F). Moreover, while the LUX-CRES construct complemented the hypocotyl growth defect of lux-4, the lux-4 LUX-VP64 lines had an enhanced phenotype with longer hypocotyls than the mutant (Figure 3G). Measurement of LUX expression in these lines showed that they have comparable transcript levels (Figure S3), demonstrating that adding the activator or repressor domain did not prevent expression of the transgenes. Noteworthy, LUX endogenous levels in lux-4 LUX-CRES lines, but not in lux-4 LUX-VP64 lines, showed a similar pattern to wild-type with higher expression in the evening (Circadian Time (CT) 12 in LL) than in the morning (CT0), consistent with restored rhythms (Figure S3B). Taken together, these results are consistent with LUX acting as a transcriptional repressor in vivo.

Figure 3. LUX functions as a repressor.

(A and B) LUX and PRR9 expression in wild-type plants grown in 12 h light/12 h dark (LD) cycles (A) or constant light (LL) (B). (C) PRR9 expression in wild-type CAB2::LUC (wt) and lux-4 mutant in LD released to LL. (A–C) Seedlings were entrained in LD for 10 days before release to LL. mRNA levels were normalized to IPP2 expression (mean values ± SEM, n=3 independent experiments). (D–G) Effect of the overexpression of LUX fused to either a repression domain (CRES) or an activation domain (VP64) in the lux-4 mutant. (D–F) Bioluminescence assays in wild-type CAB2::LUC (wt), lux-4 mutant, lux-4 35S::LUX-CRES (lux-4 LUX-CRES), and lux-4 35S::LUX-VP64 (lux-4 LUX-VP64) plants. (D) Period length and relative amplitude error (RAE) were calculated using fast Fourier transform nonlinear least-squares analysis (FFT-NLLS). Only plants for which the algorithm retrieves period length and RAE values can be represented on the plot (wt: 8 out of 8; lux-4: 6 out of 8; lux-4 LUX-CRES: 14 out of 14; lux-4 LUX-VP64: 3 out of 16). Individuals with an RAE lower than 0.6 are considered rhythmic. (E, F) Luciferase activity in wt, lux-4, lux-4 LUX-CRES (E), and lux-4 LUX-VP64 (F) lines. Third generation (T3) homozygous plants were entrained in LD for 8 days, then released to LL and imaged every 2.5h for 5 days. Values represent means ± SEM (n=8 for wt and lux-4; n=14 for lux-4 LUX-CRES; n=16 for lux-4 LUXVP64). The experiment was repeated three times with similar results, using two different transgenic lines (data shown for one representative line) selected from an initial screen of 48 primary transformants for each construct (data not shown). (G) Mean hypocotyl lengths of wt, lux-4, lux-4 LUXCRES, and lux-4 LUX-VP64 plants. Seedlings were grown in LD for 10 days before measuring the hypocotyl lengths (means ± SEM, n=20 plants).

See also Figure S3.

Perspectives

Circadian clock networks in most organisms consist of multiple interlocked feedback loops, with complex dynamics. Multiple components contribute to each circuit within the overall network. In plants, previous models with morning, evening, and central feedback loops were based mainly on genetic networks derived from mutant analyses, with few mechanistic predictions [31, 32, 41]. However, recent in vivo studies have been contributing more insight into the direct molecular connections within the clock circuit [22, 37]. Our study shows that LUX represses PRR9 through direct binding to its promoter, adding a novel connection to the previously described circadian network (Figure 4). The lux mutation abolishes rhythms entirely in free-running conditions, which cannot be fully explained by the PRR9 misregulation. Indeed, although PRR9 overexpressing lines are early flowering like lux mutants, they have a short period phenotype [25]. PRR9, like PRR7 and PRR5, associates with the CCA1 promoter to act as a transcriptional repressor [22]. This could explain the low expression level of CCA1 in the lux mutant [5, 6], where loss of LUX-mediated down-regulation of a CCA1 repressor results in indirect repression. Future studies involving genome-wide identification of LUX direct targets using ChIP-Seq will unravel new elements of critical transcriptional networks which are perturbed in the lux mutant, as well as help elucidate output pathways downstream of the clock that are regulated through LUX, making the identification of direct LUX targets other than PRR9 and LUX itself an exciting future challenge.

Figure 4. Model for the proposed role of LUX in the Arabidopsis clock.

LUX is responsible for the down-regulation of PRR9 and LUX transcription during late night. Some components of the network were omitted to simplify the model. Morning-expressed genes/proteins are represented in white, while evening-expressed genes/proteins are represented in black.

Few GARP TFs have been characterized in terms of DNA-binding specificity and target genes. Two members of the family, GLK1 and GLK2, have been implicated in the regulation of chloroplast development and were shown to upregulate similar sets of genes, primarily involved in photosynthetic function [42]. Attempts at experimentally defining the in vitro GLK1 binding site failed, possibly indicating that GLK1 alone may not bind DNA in a sequence-specific manner. However, a 6-bp motif was over-represented in the promoters of the proposed target genes. Since GLK1 and GLK2 were previously shown to interact with G-box binding factors in yeast, it has been proposed that GLK proteins act in concert with partners to attain specificity in DNA binding [42]. We are interested in identifying potential partners that modulate LUX molecular function, in particular other clock-regulated proteins that would participate in the same protein complex. Such partners should provide further molecular explanations for the dramatic circadian defect seen in lux mutants.

Experimental procedures

Protein binding microarrays, yeast one-hybrid, hypocotyl growth assays, luciferase imaging, qPCR assays, and chromatin immunoprecipitation are described in supporting information (Text S1).

Plant materials and growth conditions

Arabidopsis thaliana ecotype Columbia (Col-0) was used. CAB2::LUC reporter lines [43] and lux-4 mutants [5] were previously described.

For 35S::LUX-YFP lines, we constructed pENTR::LUX by cloning LUX coding sequence (forward primer 5’-caccATGGGAGAGGAAGTACAAA-3' and reverse primer 5’-ATTCTCATTTGCGCTTCCACC-3’) in pENTR/D-TOPO (Invitrogen). This construct was recombined using Gateway LR clonase II (Invitrogen) into pEarleyGate101 [44], generating 35S::LUX-YFP-HA. For LUX::LUX-GFP lines, we cloned a 1870 bp fragment comprising the promoter region up to the previous annotated gene (At3g46630), the 5’UTR and the coding sequence of LUX into pENTR/D-TOPO (forward primer 5’-caccCGACCACAATCAAGGAGTAAT-3’, reverse primer 5’-ATTCTCATTTGCGCTTCCACC-3’). This construct was recombined using LR Clonase II into pMDC107 [45], generating LUX::LUX-GFP. For 35S::LUX-CRES and 35S::LUX-VP64 lines, pENTR::LUX was recombined into the destination vectors pB7WG2-CRES and pB7WG2-VP64, whose construction is detailed in supporting information (text S1). For NOX RNAi lines, an 87 bp fragment of the coding region (forward primer 5’-caccGGCTTATGGTTTATTTTCCCACT, reverse primer 5’-TCGAAATCATTTTCTATACAAAGGC) was cloned in pENTR/D-TOPO. This fragment is absent from LUX and does not show strong homology to any other coding sequence in the genome. The construct was then recombined using LR Clonase II into pB7GWIWG2(II) [46] which allows for overexpression of the hairpin construct targeting NOX. All binary constructs were transformed into the appropriate background using Agrobacterium (strain GV3101) infiltration [47].

Seeds were gas sterilized and plated on 1x Murashige and Skoog basal salt medium with 1.5% agar and 3% (w/v) sucrose (MS plates). After stratification for 3 days, plates were transferred to a Percival incubator (Percival-scientific.com) set to a constant temperature of 22°C. Light entrainment was 12 h light / 12 h dark (LD) cycles, with light supplied at 80 μmol.m-2.sec-1.