Circadian Autodephosphorylation of Cyanobacterial Clock Protein KaiC Occurs via Formation of ATP as Intermediate*

Background: The autodephosphorylation mechanism of the KaiC autokinase/autophosphatase is currently unknown. Results: ATP was transiently formed during autodephosphorylation prior to the formation of Pi. Conclusion: Autodephosphorylation occurs through the reversal of autophosphorylation, followed by hydrolysis of an ATP intermediate. Significance: This mechanism is completely different from that of Ser/Thr and Tyr-specific protein phosphatases. The cyanobacterial circadian oscillator can be reconstituted in vitro; mixing three clock proteins (KaiA, KaiB, and KaiC) with ATP results in an oscillation of KaiC phosphorylation with a periodicity of ∼24 h. The hexameric ATPase KaiC hydrolyzes ATP bound at subunit interfaces. KaiC also exhibits autokinase and autophosphatase activities, the latter of which is particularly noteworthy because KaiC is phylogenetically distinct from typical protein phosphatases. To examine this activity, we performed autodephosphorylation assays using 32P-labeled KaiC. The residual radioactive ATP bound to subunit interfaces was removed using a newly established method, which included the dissociation of KaiC hexamers into monomers and the reconstitution of KaiC hexamers with nonradioactive ATP. This approach ensured that only the signals derived from 32P-labeled KaiC were examined. We detected the transient formation of [32P]ATP preceding the accumulation of 32Pi. Together with kinetic analyses, our data demonstrate that KaiC undergoes dephosphorylation via a mechanism that differs from those of conventional protein phosphatases. A phosphate group at a phosphorylation site is first transferred to KaiC-bound ADP to form ATP as an intermediate, which can be regarded as a reversal of the autophosphorylation reaction. Subsequently, the ATP molecule is hydrolyzed to form Pi. We propose that the ATPase active site mediates not only ATP hydrolysis but also the bidirectional transfer of the phosphate between phosphorylation sites and the KaiC-bound nucleotide. On the basis of these findings, we can now dissect the dynamics of the KaiC phosphorylation cycle relative to ATPase activity.

The circadian clock is an endogenous timing mechanism in living organisms that coordinates various biological activities with daily environmental changes. The circadian clock of the cyanobacterium Synechococcus elongatus PCC 7942 is the sole example that can be reconstituted in vitro; the phosphorylation state of KaiC shows robust circadian rhythms by mixing KaiA, KaiB, KaiC, and ATP in a test tube (1). Eukaryotic circadian clocks have long been understood as transcriptional-translational feedback loops in which clock gene products repress their own transcription (2). Recently, however, protein-based circadian oscillations have been found in human red blood cells (3) and in microalgae (4), suggesting that protein-based circadian clocks exist ubiquitously in living organisms. Among the many experimental systems for studying circadian oscillations, the cyanobacterial in vitro system is the simplest one and is expected to be the best model to understand the principle underlying the protein-based oscillations.
KaiC belongs to the P-loop NTPase superfamily, which includes both NTPases and kinases characterized by a strongly conserved nucleotide-binding motif, the P-loop (5). KaiC is a double-domain P-loop ATPase consisting of the N-terminal CI and C-terminal CII domains (6). KaiC forms a double doughnut-like homohexamer with an ATP molecule at each subunit interface in the two rings (7,8). KaiC exhibits ATPase and autokinase activities (9,10). KaiA activates both of these activities, whereas KaiB inhibits the effect of KaiA (9,11,12). The active sites for ATP hydrolysis reside at the subunit interfaces of CI and CII (9), whereas the autokinase activity is located only in CII (13). The autophosphorylation sites, Ser-431 and Thr-432 (14), face the ATP molecule on the neighboring protomer (8). In addition to these activities, KaiC exhibits autophosphatase activity (15), despite a lack of homology with the Ser/Thr and Tyr-specific protein phosphatases (16,17). The reaction mechanism for KaiC autodephosphorylation has not yet been elucidated.
To address this issue, we performed an autodephosphorylation assay using KaiC labeled with 32 P. Prior to the assay, we removed the residual radioactive ATP bound to subunit interfaces using a newly established method, which included the dissociation of KaiC hexamers into monomers and the reconstitution of KaiC hexamers with nonradioactive ATP. This approach ensured that only the signals derived from 32 P-labeled KaiC were examined. Surprisingly, we detected a transient elevation of [ 32 P]ATP preceding the formation of 32 P i , suggesting that KaiC dephosphorylates itself through a novel sequential mechanism, i.e. KaiC generates ATP as a reaction intermediate, followed by its hydrolysis. The first step of the reaction is regarded as a reversal of autophosphorylation. On the basis of our observations, we demonstrate that the dephosphorylation of KaiC is catalyzed by the active site of ATPase through a completely different mechanism from those of typical Ser/Thr and Tyr phosphatases.

EXPERIMENTAL PROCEDURES
Bacterial Strains-We used Escherichia coli BL21 cells as the host strain for expressing recombinant Kai proteins.
Proteins-Recombinant KaiA was expressed as described previously (6) and purified according to a previously reported method (14). Recombinant KaiB was expressed and purified as described previously (6,11,18). Full-length or CII-truncated KaiC hexamers were expressed and purified as described previously (6,9,14,18). Protein concentrations were determined using the Bradford method and bovine serum albumin as a standard.
Reconstitution of Circadian Oscillation in Vitro-Unless stated otherwise, the phosphorylation rhythm of KaiC was reconstituted and examined as reported previously (1). Briefly, 3.5 M KaiC was incubated at 30°C in the presence of 1.2 M KaiA, 3.5 M KaiB, and 1 mM ATP in 20 mM Tris (pH 8.0), 150 mM NaCl, 5 mM MgCl 2 , and 1 mM DTT. Aliquots of the reaction mixture were taken every 4 h and subjected to SDS-PAGE (gels with 11% T and 0.67% C) (18). After electrophoresis, gels were subjected to Coomassie Brilliant Blue staining using Quick CBB Plus (Wako Pure Chemical Industries, Ltd.) and scanned using an ImageScanner III system (GE Healthcare). Densitometry of the gels was performed using ImageJ 1.41 software (National Institutes of Health). The levels of phosphorylated KaiC relative to those of total KaiC were plotted against time.
Preparation of 32 P-Labeled KaiC Monomers-To remove [␥-32 P]ATP bound to the subunit interfaces of KaiC hexamers, 32 P-labeled KaiC hexamers were dissociated into monomers. The reaction mixture was passed twice through Micro Bio-Spin P-30 columns (Bio-Rad) equilibrated with buffer A containing 0.1 mM ADP, and samples were incubated on ice for 24 h. KaiC monomer fractions were then obtained via gel filtration chromatography using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated with buffer A containing 0.1 mM ADP.
Autodephosphorylation Assay-The autodephosphorylation reaction was initiated by replacing 0.1 mM ADP with 1 mM ATP using Micro Bio-Spin P-30 columns and incubating the resulting solution, which contained ϳ1 mg/ml KaiC, at 30°C. Aliquots of the reaction mixture were collected at predetermined time points, and reactions were stopped by adding an equal volume of Laemmli sample buffer (62.5 mM Tris (pH 6.8), 2% SDS, 25% glycerol, and 0.01% bromphenol blue; Bio-Rad) supplemented with 5% (v/v) 2-mercaptoethanol. The resulting samples were subjected to either SDS-PAGE or thin-layer chromatography on PEI-cellulose F plates (Merck) using 0.75 M KH 2 PO 4 as the mobile phase (20). 32 P-Derived signals were detected via autoradiography using BAS IP MS 2040E imaging plates (GE Healthcare) and a Typhoon 9400 image analyzer (GE Healthcare). Signals were quantified using the Toolbox module of ImageQuant TL 7.0 software (GE Healthcare). Background signals were subtracted using the local median method performed by the software. The standard curves used to calculate concentrations of 32 P-labeled molecules from autoradiograms were obtained using serial dilutions of 1 mM [␥-32 P]ATP solution (12 GBq/mmol). Radioactivity was detected at the positions corresponding to ATP and P i markers as well as at the origin on the chromatogram (supplemental Fig. S1). We calculated the concentrations of the molecules that stayed at the thin-layer chromatography origin and of phosphorylated KaiC separated using SDS-PAGE and plotted the data against time. We concluded that the radioactivity at the origin was derived from 32 P-labeled KaiC because the two traces almost completely overlapped (supplemental Fig. S2). Therefore, in this experiment, we analyzed all three 32 P-labeled species on a single chromatogram. The kinetic analysis was performed using Global Kinetic Explorer 2.5 software (KinTek Corp.) (21). Errors for each parameter were estimated using the FitSpace Explorer module in the software (22) by calculating the lower and upper boundaries. The calculated boundaries reflect fits within minimum 2 multiplied by 1.2.
Detection of KaiC-bound Nucleotides-Nucleotides bound to either full-length or CII-truncated KaiC were analyzed by reconstituting KaiC hexamers from monomers in the presence of 1 mM [␣-32 P]ATP. The specific activity of ATP in the reaction mixture was 6.0 GBq/mmol. The resulting solution, which contained ϳ1 mg/ml protein, was incubated at 30°C. Aliquots were collected at predetermined time points and applied twice to Micro Bio-Spin P-30 columns to remove unbound nucleotides. The eluate was mixed with an equal volume of Laemmli sample buffer containing 5% (v/v) 2-mercaptoethanol to release KaiC-bound nucleotides from the proteins. Samples were spotted onto PEI-cellulose plates and subjected to thin-layer chromatography to separate ATP and ADP using KH 2 PO 4 as the mobile phase.

RESULTS
Preparation of 32 P-Labeled KaiC without KaiA-To monitor the autodephosphorylation of KaiC, we prepared 32 P-labeled KaiC phosphorylated at Ser-431 and Thr-432. At 30°C, KaiC is markedly phosphorylated in the presence of KaiA (11) and dephosphorylated in its absence (15). For autodephosphorylation assays, however, it is desirable to phosphorylate KaiC without KaiA because residual KaiA might interfere with autodephosphorylation. Ito et al. (19) showed that the ratio of phosphorylated KaiC to total KaiC reached ϳ0.7 after 30 h of incubation at 4°C, suggesting that incubation at low temperature promotes autophosphorylation without KaiA. We incubated KaiC in the presence of [␥-32 P]ATP on ice and monitored the incorporation of radioactive phosphate for 24 h. After 24 h of incubation, ϳ30% of KaiC was newly phosphorylated based on Coomassie Brilliant Blue staining (Fig. 1A, upper panel). KaiC-specific radioactive signals increased together with the increasing intensity of phosphorylated KaiC bands (Fig. 1A,  lower panel). Based on the autoradiogram, ϳ0.3 mol of phosphate was incorporated into 1 mol of KaiC monomers.
Dissociation of KaiC Hexamers into Monomers-The 32 P-labeled KaiC hexamer obtained as described above contained residual [␥-32 P]ATP at the subunit interface. Next, we eliminated KaiC-bound [␥-32 P]ATP to ensure that only signals derived from 32 P-labeled KaiC were examined during the autodephosphorylation assay. We expected that [␥-32 P]ATP bound to the subunit interface would be released when 32 Plabeled KaiC hexamers were dissociated into monomers. To date, the procedure for the monomerization of Synechococcus KaiC hexamers has not been established. Hayashi et al. (23) showed that Thermosynechococcus KaiC hexamers dissociated into monomers in the absence of ATP. Therefore, we removed ATP from the reaction mixture and incubated the resulting solution to prepare Synechococcus KaiC monomers. Because KaiC formed aggregates in the absence of ATP at temperatures higher than ϳ10°C, we incubated the reaction mixture on ice. We added 0.1 mM ADP to stabilize the KaiC monomers during the incubation period. After 24 h of incubation, the KaiC monomers were separated using gel filtration chromatography (Fig.  1B). To verify that the KaiC monomers were functional, we used them to recreate the KaiC phosphorylation rhythm. The KaiC monomers were allowed to re-form hexamers by replacing 0.1 mM ADP with 1 mM ATP, and the resulting KaiC hexamers were incubated at 30°C in the presence of KaiA and KaiB. It took Ͻ5 min for KaiC hexamers to form from monomers (supplemental Fig. S3). As shown in Fig. 1C, this approach did not affect the KaiC phosphorylation rhythm. As expected, the radioactive ATP that had been bound to KaiC was almost completely removed with this procedure (ATP-derived signals at time 0 in Fig. 2A).
Assay of Autodephosphorylation-Autodephosphorylation was initiated by adding 1 mM nonradioactive ATP to 32 P-labeled KaiC monomers and incubating the resulting 32 P-labeled KaiC hexamers at 30°C. At the indicated time points, an aliquot of the reaction mixture was withdrawn and subjected to analysis. Surprisingly, we observed a transient elevation in the ATP concentration prior to an increase in the P i levels ( Fig. 2A). The sum of the concentrations of these molecules was assumed to be constant throughout the reaction (Fig. 2A). These results suggest that KaiC autodephosphorylation occurs through a previously unknown two-step mechanism. First, KaiC forms ATP as an intermediate during autodephosphorylation using ADP as a phosphate acceptor. This reaction can be regarded as the reversal of the autophosphorylation reaction. Subsequently, [ 32 P]ATP is hydrolyzed by KaiC ATPase to form 32 P i . We normalized the total concentration of these molecules to 1 and plotted the relative concentration of each molecule over time (Fig. 2B) The small standard deviations observed in four independent experiments indicate that the autodephosphorylation process was highly reproducible.
Characterization and Quantification of KaiC-bound Nucleotides-Although the reversal of kinase reactions requires ADP as a phosphate acceptor, no reports have previously shown that KaiC binds ADP. To address this issue, we examined the time-dependent changes in the levels of adenine nucle- otides bound to KaiC. We reconstituted KaiC hexamers from monomers in the presence of 1 mM [␣-32 P]ATP and incubated the samples at 30°C. The number of adenine nucleotides bound to each KaiC protomer remained nearly constant at two (supplemental Fig. S4), indicating that the nucleotide-binding sites of both CI and CII were occupied. The ratio of ADP to the total bound nucleotides increased, reaching ϳ0.7 after 4 h of incubation (Fig. 2C, closed circles). These results suggest that KaiC hexamers contain both ADP and ATP. We also examined the nucleotides bound to CII-truncated KaiC (Fig. 2C, open circles). In contrast to full-length KaiC, Ͼ95% of the nucleotides bound to CII-truncated KaiC were ATP. It is possible that the CII domain is responsible for ADP binding to KaiC.
Kinetic Analysis of KaiC Autodephosphorylation-To test our hypothesis of the reaction mechanism, we modeled the autodephosphorylation process (see Equations 1-3 below) and fit our data to the rate equations derived from Equations 1-3. Autodephosphorylation is an intersubunit reaction that occurs at CII subunit interfaces. To simplify the model, we assumed that the hypothetical reaction unit consisted of two subunits facing each other; one contained a nucleotide-binding site, and the other contained a phosphorylation site. We defined KaiC a as the subunit containing the nucleotide-binding site, and KaiC b as the subunit containing the 32 P-labeled phosphorylation site (Fig. 3).
KaiC-bound ADP should be a product of the hydrolysis of bound ATP and should not be incorporated from the reaction mixture because the reaction mixture contained little ADP. We first modeled the hydrolysis of nonradioactive ATP bound to KaiC a , which provides the phosphate acceptor ADP. k 1 is the rate constant of ATP hydrolysis, and k Ϫ1 is the overall rate constant of the opposing reaction, including the reversal of ATP hydrolysis, the release of products, and the incorporation of nonradioactive ATP (Equation 1).
We modeled the transfer of phosphate from 32 P-labeled KaiC b to ADP on KaiC a . We expected that the amount of both 32 Plabeled KaiC b and KaiC a -bound ADP would affect the rate of this process (Equation 2).
[ 32 P]ATP is hydrolyzed to form 32 P i (Equation 3). The forward and reverse rate constants of ATP hydrolysis are k 3 and k Ϫ3 , respectively. The reincorporation of [ 32 P]ATP into the KaiC FIGURE 2. Kinetics of KaiC autodephosphorylation and nucleotide binding. A, the autodephosphorylation reaction was performed by incubating 32 P-labeled KaiC in the presence of 1 mM nonradioactive ATP at 30°C. At each time point, an aliquot of the reaction mixture was collected and subjected to thin-layer chromatography, followed by autoradiography. The absolute concentrations of phosphorylated KaiC (PKaiC; closed black circles), ATP (closed red circles), and P i (open circles) were calculated from the autoradiogram using serial dilutions of 1 mM [␥-32 P]ATP as a standard. The data were then plotted against time. The sum of the concentrations of these molecules (Total; closed blue circles) was assumed to be constant throughout the reaction. A representative result from four independent experiments is shown. In this experiment, the concentration of KaiC was ϳ0.9 mg/ml (15 M KaiC subunits). Each sample was subjected to thin-layer chromatography three times, and values are shown as means Ϯ S.D. from three measurements. B, the sum of the concentrations of phosphorylated KaiC, ATP, and P i at each time point was normalized to 1, and the relative concentrations of phosphorylated KaiC (closed black circles), ATP (closed red circles), and P i (open circles) were plotted against time. The data are presented as means Ϯ S.D. from four independent experiments. The lines were computed via numeric integration based on the best global fit ( 2 value is 76.4 with 91 degrees of freedom). C, full-length and CII-truncated KaiC monomers formed hexamers with [␣-32 P]ATP and were incubated at 30°C. At the indicated time points, aliquots of the reaction mixtures were collected, and unbound nucleotides were removed by spin desalting columns. The aliquots were subjected to thin-layer chromatography, and KaiC-bound nucleotides were detected by autoradiography. The ratio of ADP bound to full-length KaiC (closed circles) or CII-truncated KaiC (open circles) relative to the total bound nucleotides was plotted against time. Data represent means Ϯ S.D. from three or four independent experiments. Although we could not discriminate between CII-bound ADP and CI-bound ADP in this experiment, the CII domain was shown to be responsible for ADP binding.
hexamer was not considered because residual [ 32 P]ATP was removed prior to autodephosphorylation (Equation 3).
We performed analyses using Global Kinetic Explorer (21), a program that allows for the numeric integration of rate equations and the fitting of experimental data based on nonlinear regression analysis. We set the initial values of 32 P-labeled KaiC b and the sum of nonradioactive KaiC a -bound nucleotides to 1 and assumed that the initial values of [ 32 P]ATP and 32 Pi were 0 (Fig. 2B). We estimated that the ratio of KaiC a -bound ADP at time 0 was 0.10 by subtracting the amount of ADP bound to CII-truncated KaiC (Fig. 2C, open circles) from the amount bound to full-length KaiC (Fig. 2C, closed circles). We allowed k 1 and k 3 to float together, whereas k Ϫ1 and k Ϫ3 varied independently. The solid lines in Fig. 2B represent the best fit to the data of Equations 1-3, the 2 value of which is 76.4 with 91 degrees of freedom. The rate constants and associated errors for the parameters calculated using FitSpace Explorer (22) are shown in Table 1. The forward rate constants are well constrained by the experimental data. Reverse rate constants of k Ϫ1 and k Ϫ2 cannot be accurately resolved on the time scale of the experiment, and only the upper limits are defined.

DISCUSSION
We have shown that KaiC undergoes autodephosphorylation via a previously unknown mechanism that differs from the reactions catalyzed by standard Ser/Thr and Tyr phosphatases (16,17). This mechanism includes generation of an ATP intermediate, followed by hydrolysis of this molecule (Fig. 3). The former step is regarded as the reversal of autophosphorylation. Previously, we (18) and another group (24) proposed that the phosphorylation cycle of KaiC was understood as the sequential transition among the four phosphorylation states derived from Ser-431 and Thr-432. This model was insufficient to explain the phosphorylation cycle because two independent autokinase and autophosphatase activities were assumed to exist in KaiC. In this study, we have demonstrated that both the phosphorylation and dephosphorylation of KaiC are mediated by a single phosphotransfer reaction catalyzed by the active site of ATP hydrolysis in CII. On the basis of our findings, we propose a revision of the previous model: the KaiC phosphorylation rhythm is sustained by a periodical shift in the equilibrium of the intersubunit phosphate transfer reaction between KaiCbound nucleotides and phosphorylation sites (Equation 2 and Fig. 3).
It is possible that the direction in which the intersubunit phosphate transfer reaction proceeds is determined by KaiA, which functions as an autokinase activator. The reaction favors ATP formation in the absence of KaiA at 30°C. In the presence of KaiA, the reaction proceeds to phosphorylated KaiC formation (11). The local structure of the active site may be affected by the presence or absence of KaiA, which determines the direction of phosphate transfer. In agreement with this idea, Kim et al. (25) suggested that KaiA regulated the orientation of ATP bound at the active site through a structural change in a looped segment near the C terminus. KaiB inactivates KaiA by forming a ternary complex with KaiA and Ser-431-phosphorylated KaiC (18), which should result in a circadian change in the direction of phosphate transfer.
We previously proposed that the phosphorylation cycle was coupled with the ATPase activity of KaiC (9). The precise relationship was not elucidated because the autodephosphorylation mechanism was not understood correctly. How, then, is the KaiC phosphorylation cycle regulated by the ATPase? The ATPase activity of CII has two critical roles: one is the degradation of the ATP intermediate of autodephosphorylation and the other is providing ADP, a substrate for the reversal of the autokinase reaction. We believe that ADP bound to the active site was not incorporated from the reaction mixture but was a product of the hydrolysis of bound ATP because the reaction mixture contained little ADP, if any. In agreement with this idea, Rust et al. (26) reported that increasing the ADP/ATP ratio in the reaction mixture did not affect autodephosphorylation. It is possible that ADP is not released immediately after ATP hydrolysis, which suggests that ADP release is the rate- KaiC autodephosphorylation occurs at subunit interfaces between CII domains in KaiC hexamers. We assumed that a hypothetical reaction unit consists of two subunits, KaiC a and KaiC b , which contain a nucleotide-binding site and a phosphorylation site, respectively. Because we assayed autodephosphorylation using 32 P-labeled KaiC, the products of autodephosphorylation were detected as radioactive signals. Autodephosphorylation occurred as a sequential reaction, and we have expressed each step as an equation (Equations 1-3; see "Results" for details). KaiC undergoes autodephosphorylation via the reversal of autophosphorylation (Equation 2), followed by the hydrolysis of the ATP intermediate (Equation 3). The phosphate acceptor ADP, which is required for the reversal of autophosphorylation, is generated by the hydrolysis of nonradioactive ATP prebound to KaiC (Equation 1). limiting step in the CII ATPase reaction. The ATPase activity of CII should regulate the phosphorylation cycle as a partial reaction of autodephosphorylation, whereas the ATPase activity of CI may function as a pacemaker of circadian oscillation (9). We anticipate that other as-yet unknown proteins that catalyze both forward and reverse kinase reactions exist to regulate reversible or cyclic physiological processes. One candidate is E. coli AceK, which exhibits kinase, phosphatase, and ATPase activities (27). Although direct evidence has not yet been reported, the phosphatase activity of AceK has been postulated to result from a reversal of the kinase reaction (27). AceK phosphorylates and dephosphorylates isocitrate dehydrogenase, reversibly controlling the flux of metabolites at branching points in metabolic pathways (27).