Mechanism and reconstitution of circadian transcription in cyanobacteria

TL;DR

The paper explains how a bacterial circadian clock controls gene activity and demonstrates circadian transcription in a test tube using purified proteins.

Contribution

The study reveals the mechanism of clock-regulated transcription and reconstitutes it in vitro for sustained circadian gene expression.

Findings

RpaA acts as an activator or repressor of RNA polymerase depending on its binding position.

A heterologous in vitro system using T7 RNA polymerase sustains circadian transcription for multiple days.

The findings define minimal components for circadian clock function.

Abstract

Circadian biological clocks evolved across kingdoms of life as an adaptation to predictable cycles of sunrise and sunset. In the cyanobacterium Synechococcus
elongatus, a protein-based clock precisely controls when different genes are turned on and off during the 24-h day but the phasing mechanism remains unclear. Here we show the molecular basis of this regulation and reconstitute clock-controlled transcription in vitro using purified components. Biochemical and structural analyses revealed that the clock-regulated transcription factor RpaA can function as either an activator or a repressor of cyanobacterial RNA polymerase, depending on its binding position relative to core promoter elements. Leveraging the repressor mechanism, we developed a heterologous in vitro system driven by bacteriophage T7 RNA polymerase that sustains circadian transcription for multiple days. These findings explain how a single clock output generates opposite phases of gene expression and define the minimal components for circadian clock function, enabling synthetic or biotechnological applications.

Fang et al. reveal how a bacterial circadian clock turns genes on and off at the right times of day and use the purified proteins to drive circadian gene transcription in a test tube for days.

Figures

RpaA controls opposite phases of gene expression.a, Schematic of opposite phases of dusk-peaking and dawn-peaking genes in constant-light conditions. b, Bioluminescence reporter assay results showing the expression of P*kaiBC* peaks at dusk (strain AMC541) and P*purF* peaks at dawn (strain AMC601). Bioluminescence magnitudes were normalized to focus on the phases of gene expression (mean, *n* = 5 biological replicates). c, In vitro runoff transcription from equimolar mixture of P*kaiBC* and P*purF* promoters. Increasing RpaA~P concentration (0–2.5 μM) redirects transcription from P*purF* to P*

RpaA recruits Syn7942 RNAP-σ^A^ holoenzyme to P*kaiBC* by interacting with the alpha and sigma subunits.a, Cryo-EM map and overall structure of the RpaA TAC. b, P*kaiBC* sequence used in the TAC formation. Nucleotide sequences highlighted in red are RpaA recognition sites. The green region is occupied by αCTD. c, Structural model featuring the protein–DNA interactions in the TAC. The RpaA dimer recognizes two sites (red) on the promoter, stabilizing the αCTD–promoter interaction (green). In the TAC, σ^A^ occupies the −10 (magenta) but not the −35 (purple) element.

Interaction between RpaA and αCTD σ4 is required for gene activation both in vitro and in vivo*.*a, Structural model of RpaA dimer (violet and light brown) on promoter DNA showing its interaction with the CTD of alpha subunit (salmon) and region 4 of sigma factor (yellow). Key residues involved in the interaction between RpaA and αCTD and σ4 are highlighted. b, Fold activation (mean ± s.d., *n* = 3 independent experiments) from P*kaiBC* by Syn7942 RNAP holoenzyme with the addition of RpaA variants in in vitro transcription reactions. One-way analysis of variance with Dunnett’s multiple-compari

In vitro circadian gene transcription.a, Schematic of clock-controlled rhythmic DNA binding by RpaA. During the subjective day, CikA acts as a kinase and phosphorylates RpaA. CikA binds to the clock complex that forms during the subjective night and is activated to serve as an RpaA phosphatase. Binding of RpaA on DNA prevents T7 RNAP from transcribing Broccoli RNA. b, Change in fluorescence (Δ*F*) over time from in vitro transcription of Broccoli by T7 RNAP without (blue) or with (green) clock components. RpaA-binding elements were introduced downstream of the T7 promoter to block transcriptio

RpaA controls opposite phases of gene expression.a, Schematic of opposite phases of dusk-peaking and dawn-peaking genes in constant-light conditions. b, Bioluminescence reporter assay results showing the expression of P*kaiBC* peaks at dusk (strain AMC541) and P*purF* peaks at dawn (strain AMC601). Bioluminescence magnitudes were normalized to focus on the phases of gene expression (mean, *n* = 5 biological replicates). c, In vitro runoff transcription from equimolar mixture of P*kaiBC* and P*purF* promoters. Increasing RpaA~P concentration (0–2.5 μM) redirects transcription from P*purF* to P*

RpaA recruits Syn7942 RNAP-σ^A^ holoenzyme to P*kaiBC* by interacting with the alpha and sigma subunits.a, Cryo-EM map and overall structure of the RpaA TAC. b, P*kaiBC* sequence used in the TAC formation. Nucleotide sequences highlighted in red are RpaA recognition sites. The green region is occupied by αCTD. c, Structural model featuring the protein–DNA interactions in the TAC. The RpaA dimer recognizes two sites (red) on the promoter, stabilizing the αCTD–promoter interaction (green). In the TAC, σ^A^ occupies the −10 (magenta) but not the −35 (purple) element.

Interaction between RpaA and αCTD σ4 is required for gene activation both in vitro and in vivo*.*a, Structural model of RpaA dimer (violet and light brown) on promoter DNA showing its interaction with the CTD of alpha subunit (salmon) and region 4 of sigma factor (yellow). Key residues involved in the interaction between RpaA and αCTD and σ4 are highlighted. b, Fold activation (mean ± s.d., *n* = 3 independent experiments) from P*kaiBC* by Syn7942 RNAP holoenzyme with the addition of RpaA variants in in vitro transcription reactions. One-way analysis of variance with Dunnett’s multiple-compari

In vitro circadian gene transcription.a, Schematic of clock-controlled rhythmic DNA binding by RpaA. During the subjective day, CikA acts as a kinase and phosphorylates RpaA. CikA binds to the clock complex that forms during the subjective night and is activated to serve as an RpaA phosphatase. Binding of RpaA on DNA prevents T7 RNAP from transcribing Broccoli RNA. b, Change in fluorescence (Δ*F*) over time from in vitro transcription of Broccoli by T7 RNAP without (blue) or with (green) clock components. RpaA-binding elements were introduced downstream of the T7 promoter to block transcriptio

Full text

Main

Circadian clocks are internal biological timing systems found throughout diverse kingdoms of life, including animals, plants, fungi and bacteria^1,2^. They regulate various physiological, metabolic and behavioral processes on a 24-h cycle, aligning biological activities with environmental day–night cycles. The cyanobacterium Synechococcus elongatus PCC 7942 has a circadian clock that directs cell metabolism and gates cell division^3–6^. The core of the cyanobacterial clock is a three-protein oscillator based on interactions among the proteins KaiA, KaiB and KaiC. KaiA stimulates autophosphorylation of KaiC, whereas KaiB acts antagonistically to oppose KaiA’s stimulation^7^. The result of these interactions is a ~24-h cycle of phosphorylation and dephosphorylation of KaiC, which can be reconstituted in vitro^4^. Intricate structural changes in the core oscillator as a result of KaiC phosphorylation control the engagement at different times of day of two output proteins, SasA and CikA, stimulating their opposing activities to determine the phosphorylation state of the master circadian transcription factor, RpaA^8^. Rhythmic phosphorylation regulates RpaA’s DNA-binding activity, a process we recently recapitulated in a real-time assay using purified components^9^.

To coordinate cellular physiology, the circadian clock guides the expression of various genes to peak precisely at specific times^6^. In S. elongatus PCC 7942, the free-running rhythmic expression of genes manifests in different temporal patterns, with variations in wave forms (symmetric or asymmetric) and phases (relative timing of peaks)^10^. The two main peaks of transcription are observed at subjective ‘dusk’ or ‘dawn’ phases^10^. Gene expression relies on promoter recognition by RNA polymerase (RNAP), followed by transcription initiation^11^. This reliance suggests that the circadian transcription program is regulated by altering the set of promoters that RNAP uses at different circadian phases^12^. Transcriptomic data show that all rhythmic gene expression is dependent on RpaA^5^, posing the intriguing question of how one transcription factor controls multiphasic transcript profiles. In this work, we focus on two promoters that exhibit opposite phases of gene expression: the dusk-peaking kaiBC promoter (PkaiBC) and the dawn-peaking purF promoter (PpurF) (Fig. 1a,b). Using S. elongatus PCC 7942 (Syn7942) RNAP and the housekeeping sigma factor σ^A^, we applied biochemical assays and structural analysis to elucidate the mechanism of RpaA-mediated transcription regulation. We further generated an in vitro bacteriophage T7 RNAP-based transcription system under the control of the circadian clock. This work begins to explain how multiphasic gene expression is generated from a single direct temporal output of the circadian clock in cyanobacteria. Moreover, it introduces a strategy for orthogonal circadian control that may be applicable in heterologous systems.Fig. 1. RpaA controls opposite phases of gene expression.a, Schematic of opposite phases of dusk-peaking and dawn-peaking genes in constant-light conditions. b, Bioluminescence reporter assay results showing the expression of PkaiBC peaks at dusk (strain AMC541) and PpurF peaks at dawn (strain AMC601). Bioluminescence magnitudes were normalized to focus on the phases of gene expression (mean, n = 5 biological replicates). c, In vitro runoff transcription from equimolar mixture of PkaiBC and PpurF promoters. Increasing RpaAP concentration (0–2.5 μM) redirects transcription from PpurF to PkaiBC. The experiment was repeated once with similar results. d, Syn7942 RNAP holoenzyme was used to drive in vitro transcription of the aptamer Broccoli from PkaiBC or PpurF (mean, n = 3 independent experiments). Fluorescence signal accumulates over time as Broccoli is transcribed. The addition of phosphorylated RpaA (pink traces) has opposite effects on the in vitro transcription reactions from PkaiBC or PpurF (blue traces). e, DNase I footprint assay showing that RpaA protects sites in both PkaiBC and PpurF DNA from DNase I digestion when phosphorylated. Syn7942 RNAP holoenzyme leaves a footprint on PpurF in the absence of phosphorylated RpaA but requires RpaAP for recruitment to PkaiBC. Pink boxes highlight the footprint of RpaA and blue boxes highlight that of RNAP. The experiment was repeated twice with similar results.Source data

Results

Dual roles of RpaA

First, we demonstrated that we could mimic in vivo gene regulation by phosphorylated RpaA in an in vitro transcription system based on incorporation of the ^32^P-labeled nucleoside triphosphate (NTP) substrates into an RNA. We supplemented Syn7942 RNAP-σ^A^ holoenzyme with an equimolar mixture of DNA fragments containing PkaiBC or PpurF promoters, NTPs and an increasing concentration of phosphorylated RpaA (RpaAP). We observed that increasing RpaAP concentration caused an increase in PkaiBC transcription and a simultaneous decrease in PpurF transcription, supporting a hypothesis that RpaA~P activates dusk-peaking promoters and represses dawn-peaking promoters (Fig. 1c).

Next, we developed a long-running in vitro transcription assay based on synthesis of the fluorescent Broccoli aptamer^13^. Syn7942 RNAP-σ^A^ holoenzyme was used to drive transcription from DNA templates that carry the Broccoli-encoding sequence downstream of PkaiBC or PpurF, with RpaAP added. In this real-time system, fluorescence signal increases over time as the aptamer molecules are synthesized^14^. In this system, addition of RpaAP increased transcription about fivefold from PkaiBC and moderately repressed PpurF expression (Fig. 1d).

We next performed DNase I footprinting assays to investigate how RpaAP binding at the promoter affects recognition of PkaiBC and PpurF by Syn7942 RNAP. We confirmed that RpaA binds to both PkaiBC and PpurF only when phosphorylated^5^ (Fig. 1e). RpaA binds from positions −58 to −36 relative to the start site on PkaiBC and −16 to +2 on PpurF (Extended Data Fig. 1a). However, the footprint of Syn7942 RNAP-σ^A^ holoenzyme could be seen on PkaiBC only when RpaAP was added, whereas the footprint of Syn7942 RNAP-σ^A^ on PpurF was independent of and excluded by addition of RpaAP (Fig. 1e). On the basis of the DNase I footprinting assays, we propose that, upon phosphorylation, RpaA binds to PkaiBC at around position −35 and recruits Syn7942 RNAP-σ^A^ holoenzyme to activate transcription, whereas it binds to PpurF at the −10 promoter element to sterically exclude Syn7942 RNAP and thereby repress transcription. These results are further supported by KMnO_4_ footprinting, which revealed the formation of open promoter complex by Syn7942 RNAP-σ^A^ holoenzyme on PkaiBC and the disappearance of the open complex on PpurF in the presence of RpaAP, respectively (Extended Data Fig. 2).

To further determine how the binding of RpaA affects gene expression in vivo, we generated strains that carry a luciferase reporter gene driven by varied PkaiBC promoter sequences (Extended Data Fig. 1a). When the sequence of the RpaA-binding site was disrupted in PkaiBC (PkaiBCv), the magnitude of gene expression decreased to about 15%, showing the importance of RpaA binding to activate PkaiBC (Extended Data Fig. 1c). We also observed a 3-h phase advance in peak bioluminescence compared to expression from the wild-type (WT) promoter (Extended Data Fig. 1c). DNase I footprinting assays confirmed that RpaAP is unable to bind to PkaiBCv alone; however, with the addition of Syn7942 RNAP, a footprint could still be seen for both, accounting for the residual rhythmic expression (Extended Data Fig. 1b). These results suggest that RpaA~P and RNAP bind synergistically on PkaiBC to affect both the level and phase of transcription in vivo.

Mechanism of RpaA-dependent gene activation

To elucidate precise molecular details of the RpaA activation mechanism, we determined a cryo-electron microscopy (cryo-EM) structure of the transcription activation complex (TAC), which contained both RpaA and Syn7942 RNAP bound to PkaiBC (Table 1 and Extended Data Fig. 3). Because RpaA binds DNA only when phosphorylated, we first induced the formation of the RpaA–promoter complex by phosphorylating RpaA with its cognate kinase, CikA, in the presence of adenosine triphosphate (ATP) and PkaiBC promoter DNA. To assemble the TAC, we incubated the RpaA–promoter complex with Syn7942 RNAP-σ^A^ holoenzyme, using a buffer that lacks ATP to suppress transcription initiation.Table 1. Cryo-EM data collection, refinement and validation statistics of S. elongatus RpaA bound to the PkaiBC DNA together with RNAPRNAP(EMD-47221)(PDB 9DVS)RpaA + PkaiBC DNA(EMD-47222)(PDB 9DVT)RpaA + RNAP + PkaiBC DNA(EMD-47223)(PDB 9DVU)Data collection and processingMagnification×130,000Voltage (kV)300Electron exposure (e^−^ per Å^2^)50Defocus range (μm)−0.8 to −2.0Pixel size (Å)0.935Symmetry imposedC1C1C_1_Initial particle images (no.)892,198892,198892,198Final particle images (no.)221,55281,19281,192Map resolution (Å; FSC^1^ threshold 0.143)2.53.743.7Map resolution range (Å; 25–75%)2.51–5.464.36–9.67N/ARefinementInitial model used (PDB code)8SYIModel resolution (Å; FSC threshold 0.143)2.53.72.2Model resolution range (Å)N/AN/AN/AMap sharpening B factor (Å^2^)000Model composition Nonhydrogen atoms28,6346,36935,249 Protein residues3,5856074,192 Nucleic acid residues2972113 Ligands303B factors (Å^2^) Protein117.66133.97113.52 Nucleic Acid161.40164.75156.34 Ligand117.38N/A101.26Root-mean-square deviations Bond lengths (Å)0.0040.0030.006 Bond angles (°)0.6170.6630.754Validation MolProbity score2.172.132.28 Clashscore8.548.8211.13 Poor rotamers (%)4.203.574.13Ramachandran plot Favored (%)96.3996.4396.30 Allowed (%)3.612.553.53 Disallowed (%)01.020.17^1^FSC, Fourier shell correlation.

The RpaA TAC structure revealed that the Syn7942 RNAP-σ^A^ adopts an overall conformation like that of the Synechocystis sp. PCC 6803 (Syn6803) RNAP-σ^A^ transcription initiation complex^15^ (Fig. 2a). Additionally, the C-terminal domain (CTD) of the alpha subunit (αCTD) of RNAP that was missing from the previous Syn6803 structure was resolved in our structure because of its interactions with the DNA-binding domain (DBD) of RpaA (Fig. 2b). The structure revealed that activated RpaA recruits Syn7942 RNAP-σ^A^ holoenzyme by interacting with both αCTD and region 4 of σ^A^ (σ4) (Fig. 2c). The interaction between RpaA and the RNAP-σ^A^ holoenzyme induces DNA bending, leading to the hypersensitive site seen in the DNase I footprinting assay (marked with asterisk in Fig. 1e). RpaA locates on promoter DNA as an asymmetric dimer, with the two subunits’ DBDs binding in a head-to-tail orientation on a tandem repeat upstream of the −35 site (TAAA-N_7_-TTAA) and the receiver domains (RDs) forming a two-fold symmetric dimer; this results in a conformation with one compact and one extended RpaA protomer. The DBD of the extended protomer interacts with αCTD and that of the compact protomer interacts with σ4 (Extended Data Fig. 4). The compaction of one subunit is caused by interdomain interaction between the RD and the DBD, as seen in other OmpR/PhoB-family response regulators^16,17^; however, the overall structure of the DNA-bound RpaA and its mode of interaction with RNAP are different from other response regulators in the same family (Extended Data Fig. 4), which contact either αCTD or sigma factor, not both at once^18,19^.Fig. 2. RpaA recruits Syn7942 RNAP-σ^A^ holoenzyme to PkaiBC by interacting with the alpha and sigma subunits.a, Cryo-EM map and overall structure of the RpaA TAC. b, PkaiBC sequence used in the TAC formation. Nucleotide sequences highlighted in red are RpaA recognition sites. The green region is occupied by αCTD. c, Structural model featuring the protein–DNA interactions in the TAC. The RpaA dimer recognizes two sites (red) on the promoter, stabilizing the αCTD–promoter interaction (green). In the TAC, σ^A^ occupies the −10 (magenta) but not the −35 (purple) element.

Our structure of the RpaA TAC shows the Syn7942 RNAP-σ^A^ holoenzyme poised for transcription at a different site than the one identified previously^20^. By comparing the position of RpaA and its binding sequence TAAA-N_7_-TTAA^5^ with that of the −10 element binding σ2 and its binding sequence in the structure, we identified a new −10 and a putative −35 element on PkaiBC (Fig. 2b). The new −10 element and transcription start site (TSS) are both 7 bp downstream of those identified in a previous study, in which the elements we identified were labeled as a secondary TSS^20^. S. elongatus has multiple alternative sigma factors and the previously identified promoter and TSS might be recognized by other sigma factors^21^. Our KMnO_4_ footprinting assay also supports the assignment of the −10 and TSS sites as observed in our structure (Extended Data Fig. 2).

To assess the roles of the interactions between RpaA and αCTD/σ4 in transcriptional control, we substituted selected RpaA residues at the observed interaction interfaces and tested whether these RpaA variants can still activate transcription (Fig. 3a and Extended Data Fig. 5). In vitro transcription assays showed that substitutions of E134, E146, E165 and E186 at the RpaA DBD/αCTD interface and Y190 and D194 at the RpaA DBD/σ4 interface reduce transcript production (Fig. 3b, Extended Data Fig. 6b and Supplementary Table 1). Like other OmpR/PhoB-family response regulators^18,19^, transcription activation requires RpaA to dimerize, bind to the promoter and recruit RNAP. A mutation that compromises any of these steps will result in a loss of transcriptional activation. We used a previously developed in vitro clock (IVC) reaction to distinguish the latter two processes^9^. In the IVC reactions, the fluorescence anisotropy value of a fluorescently labeled DNA segment is monitored, with greater anisotropy indicating higher mass as a result of protein binding. The changes in fluorescence anisotropy reflect the rhythmic binding of RpaA to the promoter DNA as RpaA becomes phosphorylated and dephosphorylated when its kinase CikA disengages from or engages with the KaiABC core oscillator. The IVC results (Extended Data Fig. 6a) confirmed that these RpaA mutants still bind promoter DNA rhythmically with one exception (D194A). We next combined binding-competent mutations at both RpaA DBD/αCTD and DBD/σ4 interfaces. The resulting variants E165A;Y190A and E165A;D194K both showed a greater defect in activating transcription in vitro than single mutants, while still binding DNA rhythmically in IVC reactions (Fig. 3b,d). The reciprocal deletion of RNAP αCTD resulted in ~80% loss of transcription activation (Fig. 3c).Fig. 3. Interaction between RpaA and αCTD σ4 is required for gene activation both in vitro and in vivo*.a, Structural model of RpaA dimer (violet and light brown) on promoter DNA showing its interaction with the CTD of alpha subunit (salmon) and region 4 of sigma factor (yellow). Key residues involved in the interaction between RpaA and αCTD and σ4 are highlighted. b, Fold activation (mean ± s.d., n = 3 independent experiments) from PkaiBC* by Syn7942 RNAP holoenzyme with the addition of RpaA variants in in vitro transcription reactions. One-way analysis of variance with Dunnett’s multiple-comparisons test was used for comparison of transcription activation levels by RpaA mutants. Detailed results are available in Supplementary Table 1. c, Fold activation (mean ± s.d., n = 3 independent experiments) from PkaiBC by WT Syn7942 RNAP holoenzyme and ΔαCTD RNAP upon the addition of RpaA in in vitro transcription reactions. Statistical significance was evaluated using a two-tailed Welch’s t-test. d, IVC reactions with WT RpaA and double mutants (mean, n = 3 independent experiments). e, WT RpaA (green) restored rhythmic gene expression from PkaiBC::luc in an rpaA-null background in vivo (black; AMC2650). RpaA-E165A;Y190A (red) and RpaA-E165A;D194K (orange) showed a severely reduced activation of gene expression (mean, n = 6 biological replicates). cps, counts per second.Source data

We further verified the importance of these interactions in vivo using luciferase reporter strains. We expressed WT RpaA, single-mutant variants E165A, Y190A and D194K and double-mutant variants E165A;Y190A and E165A;D194K from ectopic alleles in an rpaA-deletion strain that carries a PkaiBC luciferase reporter (AMC2650). Complementation with a WT rpaA gene restored rhythmic gene expression from PkaiBC, whereas none of the strains that express mutant rpaA alleles showed rhythmic gene expression (Extended Data Fig. 7a and Supplementary Table 2). It is known that the expression of the kaiBC genes is under control of a feedback loop; phosphorylated KaiC promotes phosphorylated RpaA and RpaA~P in turn activates kaiBC expression to maintain KaiC protein at a level needed to support rhythmic KaiC phosphorylation^5,9,22,23^. To rule out the possibility that the arrhythmicity comes from insufficient KaiC levels, we released kaiBC expression from the feedback loop by introducing another copy of the kaiBC genes in the chromosome under the control of Ptrc, a strong promoter insensitive to RpaA activation. With the addition of IPTG, the levels of KaiB and KaiC expressed from Ptrc are high enough to maintain a functional clock regardless of the expression from the native kaiBC promoter^5^. Under this condition, we found that the RpaA single mutants still activated robust rhythmic expression from PkaiBC (Extended Data Fig. 7b and Supplementary Table 2), whereas the double-mutant strains had barely detectable rhythmic expression (Fig. 3e). Thus, we verified the structural model and mechanism of dusk-peaking gene activation. In addition, we identified RpaA mutants that separate their activity of DNA binding and transcription repression from that of transcription activation.

Generating circadian rhythms of transcription in vitro

The most striking feature of the cyanobacterial circadian clock is that it is a discrete nanomachine that can be reconstituted in vitro using purified components. The core oscillator and the output pathway have been reconstituted previously^4,9^. In a manual experimental protocol in which clock reaction aliquots were added to the transcription reaction mix every 2 h over a 20-h period, we could detect the trend of antiphase sinusoid-like transcription signal from a mixture of PkaiBC and PpurF (Extended Data Fig. 8). Therefore, we reasoned that we could achieve long-running in vitro circadian gene expression by establishing a real-time in vitro Broccoli transcription assay. However, we were unable to find reaction conditions that could sustain the multisubunit Syn7942 RNAP-based in vitro transcription reactions for the days-long assays required to observe the effect of the clock.

Thus, we turned to the single-subunit RNAP from T7 bacteriophage to establish a long-running in vitro transcription assay. Lacking the required interactions, RpaA cannot recruit T7 RNAP to the promoter as an activator. However, it can still serve as a repressor through phosphorylation-dependent DNA binding. We designed a DNA template with a T7 promoter sequence and Broccoli transcript sequence^24^ separated by RpaA-binding sites^5^ so that, in the presence of clock proteins, RpaA can rhythmically bind DNA, as a function of its phosphorylation, to repress Broccoli transcription by T7 RNAP (Fig. 4a).Fig. 4. In vitro circadian gene transcription.a, Schematic of clock-controlled rhythmic DNA binding by RpaA. During the subjective day, CikA acts as a kinase and phosphorylates RpaA. CikA binds to the clock complex that forms during the subjective night and is activated to serve as an RpaA phosphatase. Binding of RpaA on DNA prevents T7 RNAP from transcribing Broccoli RNA. b, Change in fluorescence (ΔF) over time from in vitro transcription of Broccoli by T7 RNAP without (blue) or with (green) clock components. RpaA-binding elements were introduced downstream of the T7 promoter to block transcription as RpaA is rhythmically phosphorylated by the clock complex. Bottom, rates of the reactions at each time point were estimated as the change in fluorescence over 4 h (Methods). Each plot represents an average of three replicates. c, Results of IVC reaction showing the rhythmic change in the fluorescence anisotropy (FA) value of labeled DNA that is bound by phosphorylated RpaA (mean, n = 3 independent experiments) with in vitro transcription of Broccoli by T7 RNAP (Methods). d, Normalized and detrended rate (mean ± s.d., n = 3 independent experiments) from in vitro transcription reactions with (red) or without (green) resetting by ADP treatment. e, Period lengths (mean ± s.d., n = 6 independent experiments) of the rhythmic change in transcription rate at different temperatures.Source data

To couple the IVC and in vitro transcription reactions, we performed buffer optimization to find a condition that supports both (Extended Data Fig. 9a–c). Both reactions require Mg^2+^ and ATP^4,25^; however, in vitro transcription is favored by low salt, reducing agent and spermidine, which differ from the IVC conditions^25^. With an optimized buffer, we could run the T7 in vitro transcription reaction for longer than 2 days, greatly extending the hours-long reactions used in routine Broccoli aptamer-based real-time in vitro fluorescent reporter assays^24,25^ (Fig. 4b). Comparing fluorescence signals from the in vitro transcription assays with or without clock proteins, we detected inflections in the curves from the reactions with clock proteins, implying clock-dependent changes in transcription (Fig. 4b and Extended Data Fig. 9d). We converted the fluorescence signals to relative transcription rates and could see clear peaks and troughs that are in accordance with RpaA binding as revealed by IVC reactions (Fig. 4b,c). The elapsed time between two peaks is roughly 24 h, supporting the premise that rhythmic changes in transcription rates reflect a circadian rhythm. We further tested whether the timing of peaks could be reset by altering the ratio of adenine diphosphate (ADP) to ATP, as previously was shown for IVC reactions^26,27^. Upon adding ADP during the time known to reset the phase of the IVC, the transcription rhythm could also be reset (Fig. 4d). In addition, we ran the reactions at different temperatures and calculated the periods (Fig. 4e and Extended Data Fig. 10). The results showed that the periods of the reactions are temperature compensated as is characteristic of circadian rhythms^28^ (23.7 h at 25 °C versus 20.2 h at 35 °C; Q10 = 1.17). Thus, the system fulfills the criteria agreed upon for circadian phenomena in diverse systems: self-sustained rhythm with a period of ~24 h, phase resetting by an external signal and a stable period over a physiological range of temperatures^28^.

Discussion

Here, we demonstrated the simplicity and potential universality of the prokaryotic circadian transcription mechanism by reconstituting it in vitro. Just six proteins in total are required to keep time, transmit the clock signal and produce rhythmic gene transcription. Among these, RpaA has a central role in coordinating the circadian program by directly and oppositely regulating the RNAP recognition of dusk-peaking and dawn-peaking protomers.

Upon phosphorylation, RpaA binds promoter DNA and acts as either a direct activator or a repressor of cyanobacterial RNAP, depending on its binding position relative to the −10 and −35 consensus promoter elements. In vitro transcription assays with mixed dusk and dawn promoters showed that RpaA~P simultaneously activates transcription from a dusk promoter while repressing the dawn one. This action results in gene expression patterns with opposite peak times for the two promoters^5,29^. Given that at least 70% of the genome is expressed rhythmically^5,10^, it is expected that some promoters will not follow these mechanisms in future studies. Our cryo-EM structure reveals that RpaA recruits RNAP to the dusk promoter through interactions with both αCTD and σ4, which is uncommon for OmpR/PhoB-family response regulators^11^. The structural determinants of RpaA required for transcription activation differ from those needed for DNA binding and repression; thus, the two parts of the mechanism do not interfere with each other. It was previously proposed that RpaB, another transcription factor in S. elongatus, has a central role in circadian gene regulation while RpaA works as an antirepressor of RpaB to activate dusk genes^30^. Our results demonstrate directly that RpaA is the major transcription factor needed to generate opposite phases of circadian gene expression and that it can directly activate dusk gene transcription by recruiting RNAP to a target promoter. RpaB, whose phosphorylation and DNA-binding activity change rapidly in response to light intensity^31^, is unlikely to be primarily a circadian regulator. Instead, RpaB, which binds to some promoters (including kaiBC) at a different site than does RpaA, likely helps to integrate environmental signals to fine-tune circadian gene expression and cellular physiology^31^.

We engineered an orthologous in vitro circadian transcription system driven by single-subunit bacteriophage T7 RNAP, unrelated to the multisubunit RNAP of cyanobacteria, by applying the repressor part of the RpaA mechanism. This orthologous system nevertheless fulfills all the criteria for a circadian rhythm^28^. Our results represent a step forward in reconstructing the cyanobacterial circadian system: from KaiC phosphorylation rhythms^4^ to RpaA DNA-binding rhythms^9^ and now to rhythmic transcription. These results also demonstrate unequivocally that the KaiABC posttranslational oscillator can drive circadian gene expression without the need for a genetic feedback loop. Recently, a slow ATPase pacemaker mechanism as used by the cyanobacterial clock was discovered to function also in eukaryotic clocks^2^. Understanding how the slow ATPase activity of KaiC controls circadian gene expression provides conceptual insights into other ATPase-based timing mechanisms. Furthermore, this simplified repression-based system has the potential to be transplanted into a heterologous organism to generate in vivo circadian gene expression without the need to match a transcription activator and the RNAP with which it evolved. Alternatively, by fusing Syn7942 αCTD and σ4 to host RNAP subunits, the activator mechanism could also be incorporated into other bacteria.

While central mechanisms of circadian transcription in cyanobacteria are simple, the generation of robust multiphasic gene expression in vivo is likely more complex, involving combinatorial regulation, gene expression cascades and feedback regulation^31–34^. The high-throughput in vitro transcription assays we applied here form a foundation to untangle the complicated regulatory network that results in remarkably precise circadian orchestration, potentially yielding a quantitative understanding of multiphasic gene expression.

Methods

Strains and plasmids

Escherichia coli and cyanobacterial strains were cultured with appropriate antibiotic supplementation as described^35^. E. coli strains DH5α and BL21(DE3) were used for cloning and protein overexpression, respectively. E. coli cultures were routinely maintained at 37 °C in Luria–Bertani (LB) liquid medium with shaking or on LB agar plates. S. elongatus strains were cultured in liquid BG-11 medium with shaking or on BG-11 agar plates at 30 °C under continuous illumination provided by cool white fluorescent bulbs at an intensity of 100–150 μmol photons m^−2^ s^−1^.

The pET28b vector was used for overexpression and purification of all proteins except for Syn7942 RNAP (pET28a) in this study. For KaiA, KaiB, KaiC and RpaA, the 6×His–SUMO-tagged S. elongatus constructs were used as reported previously^9^. Similarly, the primary σ70 sigma factor, SigA (σ^A^), was also overexpressed as a 6×His–SUMO-tagged fusion protein with the tag removed during the purification step. Gibson assembly-based site-directed mutagenesis was used to generate plasmids for 6×His–SUMO-tagged RpaA mutants. Mutations in RpoA in the pET28a_SelRNAP gene were generated by site-directed mutagenesis kit (New England Biolabs). S. elongatus CikA was purified as a Strep-tagged (WSHPQFEKSA) protein.

The pUC19 vector was used for routine cloning and assembly of DNA fragments. Different promoter sequences were fused with a firefly luciferase gene^36^ or Broccoli aptamer sequence^37^ using Gibson assembly in pUC19 (ref. ^38^). These plasmids served as PCR templates for generating double-stranded DNA (dsDNA) fragments used in DNase I footprinting and in vitro transcription assays.

The pAM1303 neutral site 1 (NSI) targeting vector was used to insert luciferase reporter genes with various promoter sequences and rpaA mutant alleles into different S. elongatus genetic backgrounds (Supplementary Table 3). To isolate the role of RpaA, reporter strains with shorter promoter sequences than those used in our standard reporters were generated. The luciferase gene was fused downstream of PkaiBC and PpurF sequences that spanned −75 bp from the TSS to 24 bp after the start codon, in frame. Variants of PkaiBC were generated using a Gibson assembly-based strategy. To complement an rpaA-null reporter strain (AMC2650), a segment including ±300 bp upstream and downstream of the rpaA coding sequence was amplified from the S. elongatus genome and inserted in the BamHI site in pAM1303. Site-directed mutagenesis was used to introduce mutations in the rpaA gene to generate reporter strains with various rpaA alleles (Supplementary Table 3).

The pAM5553 NSIII targeting vector was used to introduce the kaiBC gene under control of the tunable E. coli promoter Ptrc in rpaA-mutant backgrounds. A lacI-Ptrc::kaiBC fragment was cloned from the NSI plasmid HS1 (gift from E. O’Shea lab)^5^ and inserted in the SwaI site of pAM5553.

Recombinant S. elongatus strains were generated by a standard natural transformation protocol^39^. E. coli strain AM1359 was used to introduce genes into the nontransformable rpaA-null and rpaA-impaired mutant strains by biparental mating as described previously^40^.

Monitoring S. elongatus circadian rhythms using bioluminescence as readout

Bioluminescence was monitored using firefly luciferase fusion reporters as previously reported^9^. In brief, cyanobacterial cultures were diluted to an optical density at 750 nm (OD_750_) of 0.2 and distributed onto BG-11 agar pads containing luciferin in 96-well plates for measuring bioluminescence every 2 h with a Spark 10 M (Tecan) plate reader. To induce the expression of kaiBC from Ptrc, 10 μM IPTG was added to the wells containing NSIII-Ptrc::kaiBC strains during plate setup. The plates were entrained for two 12-h light–dark cycles (80 μmol photons m^−2^ s^−1^) at 30 °C to synchronize phases before releasing to continuous light (30 μmol photons m^−2^ s^−1^) for data collection. Because of light–dark sensitivity of RpaA-defective mutants, all the plates used in the experiments containing RpaA mutants were entrained for only one 12-h light–dark cycle with lowered light intensity (40 μmol photons m^−2^ s^−1^), determined to be permissive for growth in the absence of RpaA.

Protein overexpression and purification

Overexpression and purification of 6×His–SUMO-tagged KaiA, KaiB, KaiC, RpaA and SigA, were performed as previously reported^27^. A 20-ml overnight E. coli starter culture in LB medium was transferred to 1 L of LB medium supplemented with 25 μg ml^−1^ kanamycin and grown to OD_600_ of ~0.5 at 37 °C before induction with 0.2 mM IPTG overnight at 22 °C. The 6×His–SUMO-tagged proteins were first purified as tagged proteins by affinity chromatography on Ni-NTA resin before cleavage of the 6×His–SUMO tag by ULP1 protease. After the removal of the 6×His tag by a second Ni-NTA column, proteins were further purified by size-exclusion chromatography. Tris-buffered solutions were used throughout purification steps instead of phosphate-buffered solutions as previously reported^27^. In brief, For KaiA, KaiB and KaiC, clock reaction buffer (20 mM Tris pH 8.0, 150 mM NaCl, 5 mM MgCl_2_, 0.5 mM EDTA and 1 mM ATP) was used as elution buffer for the gel-filtration chromatography. For RpaA and SigA, a buffer containing 20 mM Tris pH 8.0, 150 mM NaCl and 5% glycerol was used as elution buffer during the gel-filtration chromatography step.

For Strep-tagged CikA, steps before affinity purification (induction, cell lysis and clarification of cell lysate) were the same as described above except that the cells were resuspended in Strep-Tactin wash buffer (50 mM Tris pH 8.0, 150 mM NaCl, 5% glycerol and 1 mM DTT) before cell lysis. Clarified cell lysate was loaded onto a Strep-Tactin XT Superflow (IBA Lifesciences) gravity column equilibrated with Strep-Tactin wash buffer. The column was washed with Strep-Tactin wash buffer, followed by protein elution with the same buffer containing 50 mM biotin. The eluents were concentrated using spin concentrators and loaded onto a Superdex 200 column equilibrated in a buffer containing 20 mM Tris pH 8.0, 150 mM NaCl and 5% glycerol for size-exclusion chromatography.

For preparation of Syn7942 recombinant double 6×His–Strep-tagged RNAP, T7 express (New England Biolabs) cells were transformed with pET28a_SelRNAP^15^. After the culture in 1.8 L of LB supplemented with 50 μg ml^−1^ kanamycin was grown to an OD_600_ = 0.5–0.6, it was moved to room temperature with shaking and, after a further 30 min, was supplemented with 1 mM IPTG and moved to a temperature-controlled shaker (New Brunswick) at 18 °C overnight. The cells were harvested by centrifugation for 20 min at 4 °C at 6,000g. The cell pellet was resuspended in lysis buffer (50 mM Tris-HCl pH 8.0, 250 mM NaCl, 10% v/v glycerol, 20 mM imidazole, 1 mM β-mercaptoethanol and 1× protease inhibitor cocktail (Sigma-Aldrich)). Cell were lysed by sonication and centrifuged 20 min at 45,000g at 4 °C. The collected supernatant was filtered through a 0.45-μm filter. Using an AKTA-Pure system equipped with a Cytiva HisTrap HP 5-ml column, the filtered supernatant was loaded onto the column after it was washed with five column volumes (CVs) of HisTrap buffer A (50 mM Tris-HCl, 250 mM NaCl, 10% glycerol, 20 mM imidazole and 1 mM β-mercaptoethanol) at the recommended flow rate. The column was then washed with another five CVs of Buffer A, before eluting with five CVs of 50% HisTrap buffer B (buffer A but with 500 mM imidazole). The eluted fractions were verified by SDS–PAGE and RNAP-containing fractions pooled and loaded onto a StrepTrap XT 5-ml column washed with five CVs of StrepTrap buffer A (100 mM HEPES–KOH pH 8.0, 150 mM NaCl and 1 mM EDTA). The RNAP was eluted with StrepTrap buffer B (buffer A + 50 mM biotin). Fractions were verified by SDS–PAGE, pooled, diluted tenfold with heparin buffer A (50 mM HEPES–KOH pH 8.0 and 50 mM NaCl) and loaded onto a Cytiva HiTrap heparin HP 5-ml column washed with five CVs of buffer A. After adding 5% buffer B (50 mM HEPES–KOH pH 8.0 and 1 M NaCl), the RNAP was eluted with 45% buffer B. Fractions were verified by SDS–PAGE, pooled, dialyzed overnight against storage buffer (40 mM HEPES pH 8.0, 200 mM KCl, 50% glycerol, 1 mM DTT and 1 mM EDTA) and stored at −20 °C.

dsDNA fragment preparation

dsDNA fragments were used in the DNase I footprinting assay, IVC reactions, TAC formation and in vitro transcription reactions. For the DNase I footprinting assay, fragments were amplified from pUC19 templates by PCR, with the forward primer labeled with 5′-FAM (Integrated DNA Technologies). dsDNA fragments used in in vitro transcription reactions were also amplified by PCR from pUC19 templates and column-purified.

For dsDNA used in IVC, Cy3-labeled single-stranded DNA (ssDNA) and its reverse complementary ssDNA were ordered from Integrated DNA Technologies. dsDNA used in TAC was also ordered from Integrated DNA Technologies as ssDNA oligonucleotides (Supplementary Table 4). To prepare the dsDNA fragments from ssDNA, equimolar samples of the two complimentary oligonucleotides were mixed before denaturing at 95 °C for 5 min followed by slowly cooling to room temperature.

DNase I footprinting assay

First, ~350-bp dsDNA fragments covering kaiBC and purF promoter regions were prepared as described above. Promoter DNA fragments (20 nM) were incubated with a combination of RpaA (5 μM) and/or Syn7942 RNAP holoenzyme (0.5 μM) in 20 μl of binding buffer (25 mM Tris-HCl pH 8.0, 150 mM NaCl, 2 mM MgCl_2_, 10% glycerol and 0.6 mg ml^−1^ heparin) for 10 min at 30 °C followed by a 15-min DNase I digestion at 22 °C that was started by addition of 5 μl of 1:128 diluted DNase I (New England Biolabs). The reaction was stopped by adding the same volume of 0.5 M EDTA (pH 8.0). CikA (0.5 μM) and ATP (0.5 mM) were added when RpaA phosphorylation was needed. DNA fragments were column-purified and sent to Eton Bioscience for fragment analysis on a 3730 DNA analyzer (Applied Biosystem). The results were further analyzed using Peak Scanner software version 1.0 (Applied Biosystem).

Monitoring in vitro transcription reactions using fluorescence as readout

Two kinds of RNAP were used in this study to run in vitro transcription reactions: S. elongatus RNAP-σ^A^ holoenzyme and T7 RNAP (New England Biolabs). For S. elongatus RNAP-σ^A^ holoenzyme, the in vitro transcription reactions were carried out in a buffer containing 20 mM Tris pH 8.0, 20 mM KCl, 10 mM MgCl_2_, 1 mM TCEP and 1 mM ATP. Additionally, RNAP-σ^A^ holoenzyme (150 nM), DNA template (10 nM), NTPs (0.4 mM total) and (Z)-4-(3,5-difluoro-4-hydroxybenzylidene)-2-methyl-1-(2,2,2-trifluoroethyl)-1H-imidazol-5(4H)-one) (DFHBI-1T, 0.1 mM) were added. RpaA variants (2.5 μM) and CikA (0.7 μM) were added as needed. Fold activation was calculated by comparing the change in fluorescence after 4 h with the control reaction in which no RpaA was added. T7 RNAP used a buffer detailed below.

In vitro transcription reactions were set up on ice and aliquots (20 μl) of reactions were pipetted into a 384-well plate for monitoring fluorescence. Here, 485 ± 20 nm and 535 ± 25 nm were chosen for excitation and emission filters, respectively, to monitor the fluorescence signal from Broccoli–DFHBI-1T. Reactions were routinely run at 30 °C except in temperature compensation experiments, in which 25 °C and 35 °C were used for in vitro transcription reactions.

In vitro runoff transcription using radiolabeled substrates

To form promoter complex, 0.05 μM linear dsDNA fragments containing promoter were incubated with 0.25 μM core RNAP, 0.75 μM σ^A^ and phosphorylated RpaA (concentrations indicated in the respective figures) in transcription buffer (20 mM HEPES pH 8.0 and 50 mM NaCl) for 5 min at 30 °C. The transcription was started with addition of 10 mM MgCl_2_, 500 μM ATP, guanosine triphosphate, cytidine triphosphate and 50 μM α^32^P-labeled uridine triphosphate (12.5 Ci mmol^−1^; Hartmann Analytic). Reactions were incubated for 15 min at 30 °C and stopped with formamide-containing buffer. Products were resolved by denaturing 23% PAGE (8 M Urea), revealed by PhosphorImaging (Cytiva) and visualized using ImageQuant (Cytiva) software.

Cryo-EM sample preparation

To induce the formation of RpaA-dependent TAC, an RpaA–promoter complex and Syn7942 RNAP holoenzyme were first assembled separately. RpaA (25 μM) was phosphorylated by CikA (2.5 μM) in a buffer containing 20 mM Tris pH 8.0, 150 mM NaCl, 5 mM MgCl_2_ and 1 mM ATP and incubated with 5 μM promoter DNA at 30 °C for 2 h. The RpaA–promoter complex was then buffer-exchanged into the TAC buffer containing 20 mM Tris pH 8.0, 50 mM NaCl, 5 mM MgCl_2_ and 5% glycerol at 4 °C using a centrifugal concentrator. At the same time, equimolar Syn7942 RNAP and SigA were incubated in TAC buffer at 4 °C. The TAC was formed by adding the RNAP holoenzyme in the buffer-exchanged RpaA–promoter complex so that the final concentrations for each component were 20 μM RpaA, 4 μM RNAP holoenzyme and 4 μM promoter DNA. After incubating on ice for 30 min, the complex was immediately loaded onto the grids for freezing. A 3.5–4-µl protein sample was applied to glow-discharged grids (Quantifoil Cu 300-mesh R2/1) within the environmental chamber adjusted to 4 °C temperature and approximately 95% humidity in a Vitrobot Mark IV (Thermo Fisher Scientific). After a brief incubation, grids were blotted with Vitrobot’s standard setting of blot force of 3 for 4 s; the sample was then plunge-frozen into liquid-nitrogen-cooled liquid ethane.

Cryo-EM data collection and processing

Grids were clipped into standard AutoGrids (Thermo Fisher Scientific) for imaging on Autoloader-equipped EM instruments. Videos were collected at a magnification of ×130,000 on a Krios G4 system equipped with a Falcon 4 camera and Selectris X energy filter (Thermo Fisher Scientific) using a 10-eV slit, with a pixel size of 0.935 Å and a total dose of 50 e^−^ per Å^2^. A defocus range of −0.5 to −2.2 μm was used during the data collection. Automated data acquisition was performed using EPU (Thermo Fisher Scientific). In total, two datasets were collected, with a total of 14,499 (7,145 + 7,354) videos used in the final data processing (Table 1).

All data processing was performed using cryoSPARC (version 4)^41^. After cryoSPARC live session with blob picking (100–200 Å), 1,311,494 particles were subjected to multiple rounds of two-dimensional (2D) classification, a single heterogeneous refinement with three models and a three-dimensional (3D) classification with ten models, yielding a single 3D class with 60,866 particles representing the fully assembled TAC, which was then used for template generation for template-based particle picking. Two datasets were processed in parallel following template picking. Multiple rounds of 2D classification and a final heterogeneous refinement with three ab initio models (one model with RNAP alone, one model with RNAP plus density around RpaA and a noise model), yielding 550,561 particles and 938,691 particles from the two datasets that contained RNAP (with or without RpaA). These two sets of particles were combined and subjected to another round of 2D classification, which yielded 892,198 particles (Extended Data Fig. 3).

The combined particle set was cleaned up further with an extra heterogeneous refinement with the above three-model setting and another round of 2D classification, yielding a final particle set of 789,278 particles that contain RNAP (with or without RpaA). To isolate particles with ordered RpaA, a 3D variability analysis using the full complex mask was performed with five modes and resolved with ten intermediate frames. In total, 330,169 particles were pooled from frames containing some density around the RpaA region. Then, a further 3D classification with a focused mask on the RpaA dimer and bound DNA resulted in a small particle set of 83,966 particles that contained strong RpaA occupancy. Finally, these particles were reextracted with a box of 600 pixels (binned to 400 pixels, with a pixel size of 1.4025 Å) and subjected to a new round of 3D variability analysis with an RpaA-focused mask, which helped to remove some bad particles. In the end, 81,192 particles were used for local refinement to reconstruct the RNAP–RpaA dimer–DNA complex, with a final resolution of 3.7 Å (Table 1).

To determine the RNAP structure (without bound RpaA), we took a similar approach from the combined particle set of 789,278 particles. During the data processing, we noticed that the RpoC2 arm region is quite flexible, which contributes to the relatively low resolution in that region. A 3D variability analysis with a focused mask on the RpoC2 arm region isolated 221,552 particles that showed a relatively rigid conformation of the RpoC2 arm. This particle set was reextracted with a full box size of 512 pixels (pixel size of 0.935 Å) and then subjected to a nonuniform refinement, which yielded a final resolution of 2.5 Å for the RNAP-σ^A^ complex lacking RpaA (Table 1).

To reconstruct the full complex map of RpaA bound to PkaiBC DNA and Syn7942 RNAP-σ^A^ holoenzyme, we fitted refined models from the above two structures into a consensus map generated from the whole-body refinement of 81,192 particles that contained strong RpaA dimer density and then combined them in ChimeraX with the ‘vop max’ command.

Model building and refinement

For model building, a prior structure of the S. elongatus RNAP elongation complex (PDB 8SYI)^42^ was used as an initial model for the RpoA (UniProt Q31L30; chains A and B), RpoB (UniProt Q31N17; chain C), RpoC1 (UniProt P42079; chain D), RpoZ (UniProt Q31MH9; chain E) and RpoC2 (UniProt Q31N15, chain F, residues 4–431 and 993–1237) subunits. A structure of the Synechocystis sp. PCC 6803 RNAP initiation complex (PDB ID 8GZG)^15^ was used as an initial model for SigA (Uniprot P38023; chain G) and bound DNA (chains 1 and 2). AlphaFold-predicted structures (https://alphafold.ebi.ac.uk)^43^ were used as initial models for the RpoA CTD (chain A, residues 239–300), RpoC2 (residues 440–987) and RpaA (UniProt Q31S42). For RpaA, models of the N-terminal domain (residues 1–124) and CTD (residues 140–231) were fitted separately. Residue D53 of RpaA was modeled as a phosphorylated aspartate.

Initial models for RpoA, RpoB, RpoC1, RpoZ and RpoC2 (residues 4–431 and 993–1237 plus bound zinc ion) were manually rebuilt in Coot^44^ and refined using phenix.refine^45^ into the 2.5-Å resolution cryo-EM map of the RNAP–RpaA–DNA complex. For RpoC2 (residues 440–987), individual domains of the initial AlphaFold model were rigid-body fitted and then refined using phenix.refine into a cryo-EM map generated by focused refinement on this region of RpoC2. The RpoA CTD and RpaA were rigid-body fitted into a map generated by focused refinement on RpaA and bound DNA. DNA was manually built into a composite of the original map and the two focused refinement maps and refined with strict base-pairing and base-stacking restraints in phenix.refine^45^. The final model was assembled and refined against the composite map in phenix.refine with input model restraints, secondary-structure restraints, and DNA base-pairing and base-stacking restraints. As expected, the CikA kinase, present at only 1:10 to RpaA and carrying out a phosphorylation event that typically involves only a transient interaction, was not present in the complex.

Monitoring IVC reactions using fluorescence anisotropy as readout

The IVC reactions used in this study contained KaiA (1.2 μM), KaiB (3.5 μM), KaiC (3.5 μM), CikA (0.7 μM), RpaA (2.5 μM) and a promoter-bearing Cy3-labeled dsDNA fragment (50 nM). The standard clock buffer contained 20 mM Tris pH 8.0, 150 mM NaCl, 5 mM MgCl_2_, 0.5 mM EDTA and 1 mM ATP. For maintaining longer IVC reactions, 4 mM phosphoenolpyruvate pH 8.0 and 10 U per ml Bacillus stearothermophilus pyruvate kinase (Sigma-Aldrich) were added to provide ATP regeneration^26^. WT RpaA and mutant variants were added in the reactions to test whether they are still active in binding DNA rhythmically.

Reactions were set up on ice and aliquots (20 μl) of reactions were pipetted into a 384-well plate before sealing with transparent film (MicroAmp, Applied Biosystems). Fluorescence anisotropy data were collected every 15 min in a Spark 10 M (Tecan) plate reader at 30 °C as previously reported^9^. The excitation filter was set to 520 ± 10 nm and the emission filter was set to 580 ± 10 nm for Cy3-labeled promoter DNA.

Coupling IVC and in vitro transcription reactions

By comparing the buffer composition used for T7 RNAP in vitro transcription reactions (40 mM Tris pH 7.9, 6 mM MgCl_2_, 1 mM DTT and 2 mM spermidine; New England Biolabs) and that for the IVC reaction, we found differences in NaCl concentration, reducing agent and spermidine. These factors were varied and tested in IVC reactions in a base buffer containing 20 mM Tris pH 8.0, 5 mM MgCl_2_ and 1 mM ATP. After selecting a NaCl concentration (40 mM), this concentration of NaCl was added to the base buffer to test for additional components including DTT, TCEP and spermidine. The final IVC-compatible buffer contained 20 mM Tris pH 8.0, 5 mM MgCl_2_ 40 mM NaCl, 0.5 mM TCEP, 0.25 mM spermidine and 1 mM ATP. To set up the reactions, clock components (1.2 μM KaiA, 3.5 μM KaiB, 3.5 μM KaiC, 0.7 μM CikA and 2.5 μM RpaA), T7 RNAP (5 μM), DNA template (10 nM), NTP (0.5 mM total), DFHBI (25 μM) and ATP regeneration system (4 mM phosphoenolpyruvate pH 8.0, 10 U per ml pyruvate kinase) were mixed on ice in the buffer.

A phase shift of the peaks of in vitro circadian gene expression was induced 26 h after the measurement started in the plate. ADP was added to the reaction mix at 26 h, chosen as the time point at which the strongest resetting could be observed in the IVC^27^, to lower the ATP-to-ADP ratio to 1:1. Pyruvate kinase was excluded from the reaction set up and added 4 h later to convert ADP back to ATP to restore the ATP-to-ADP ratio. For control reactions, ADP and pyruvate kinase were added at the same time.

To compare the rhythms of clock-controlled in vitro transcription reactions with those of standard IVC reactions, the latter were performed separately under the same buffer conditions and in the presence of Broccoli transcription. The only difference was that DFHBI was omitted from the reaction and fluorescently labeled promoter DNA was added instead.

Calculation of rate, period and phase

For time point i (i ≥ 2), Ri = (Fi + 2 − Fi − 2)/4 was used to estimate the rate of transcription (where Fi is the fluorescence reading at time i). To calculate the relative rate of transcription reactions, the time-series fluorescence readings from in vitro transcription reactions were normalized to (0, 1), with the lowest reading as 0 and the highest reading as 1, before using the same equation for rate calculation.

Bioluminescence and fluorescence time-series data were analyzed with the BioDare2 online service (https://biodare2.ed.ac.uk/)^46^ to obtain period and phase information. Stable rhythmic data with at least three cycles of oscillation (for in vitro transcription reactions, at least two cycles were used) were linearly detrended to extract period and phase (by averaged peaks) using the fast Fourier transform nonlinear least squares method.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41594-025-01740-0.

Supplementary information

Reporting Summary Supplementary Tables 1–4Supplementary Tables 1–4.

Source data

Source Data Fig. 1. Statistical source data. Source Data Fig. 1cUnprocessed gel image. Source Data Fig. 3. Statistical source data. Source Data Fig. 4. Statistical source data. Source Data Extended Data Fig. 1. Statistical source data. Source Data Extended Data Fig. 2. Unprocessed gel image. Source Data Extended Data Fig. 6. Statistical source data. Source Data Extended Data Fig. 7. Statistical source data. Source Data Extended Data Fig. 8. Statistical source data. Source Data Extended Data Fig. 9. Statistical source data. Source Data Extended Data Fig. 10. Statistical source data.