Crystal structure of circadian clock protein KaiA from Synechococcus elongatus.

The circadian clock found in Synechococcus elongatus, the most ancient circadian clock, is regulated by the interaction of three proteins, KaiA, KaiB, and KaiC. While the precise function of these proteins remains unclear, KaiA has been shown to be a positive regulator of the expression of KaiB and KaiC. The 2.0-A structure of KaiA of S. elongatus reported here shows that the protein is composed of two independently folded domains connected by a linker. The NH(2)-terminal pseudo-receiver domain has a similar fold with that of bacterial response regulators, whereas the COOH-terminal four-helix bundle domain is novel and forms the interface of the 2-fold-related homodimer. The COOH-terminal four-helix bundle domain has been shown to contain the KaiC binding site. The structure suggests that the KaiB binding site is covered in the dimer interface of the KaiA "closed" conformation, observed in the crystal structure, which suggests an allosteric regulation mechanism.

Circadian timekeeping systems are present in virtually all eukaryotic organisms and modulate diverse physiological processes ranging from leaf movement to transcription regulation (1, 2) with a ϳ24-h periodicity. Important mechanistic similarities, such as positive and negative transcription-translation feedback loops, are shared between all circadian clocks studied thus far (3). The evolutionary importance of an endogenous circadian oscillator is further emphasized by its presence in ancient organisms such as cyanobacteria (4), the subject of this work. Indeed, it has been shown that a circadian oscillator tuned to the light-dark cycle period increases the fitness of the organism (5). The cyanobacterial clock is the simplest and the oldest (6) identified thus far, composed of at least three interacting clock proteins, KaiA, KaiB and KaiC (4,7), that form heteromultimeric complexes in vivo of sizes that oscillate in a circadian manner (8).
Genomic analysis indicates that KaiC is a member of the bacterial RecA/DnaB family (9). The KaiC phosphorylation state has been shown to change in vivo with a circadian pattern (10,11), and it is known that ATP binding by KaiC promotes formation of a hexameric KaiC particle (12,13), which is presumed to be the functional form of the protein. Experimental evidence suggests that KaiC binds forked DNA when in the ATP-dependent hexameric form (13). KaiA, the positive element of the cyanobacterial circadian oscillator, activates KaiC autophosphorylation and/or inhibits KaiC dephosphorylation, which is believed to be important for its circadian clock function (10,14). KaiB alone does not affect KaiC autophosphorylation in vitro, but it antagonizes the action of KaiA both in vivo and in vitro (11,14,15).
Iwasaki et al. (7) demonstrated that Kai proteins directly associate in all possible combinations with data from a twohybrid system, in vitro, and in cyanobacterial cells. The association between KaiA and KaiB is weak but can be dramatically enhanced by KaiC (7). A KaiA long period mutation allele (E103K, kaiA1) was reported to dramatically enhance the KaiA-KaiB interaction in vitro but without any effect on KaiA-KaiC interaction (7). More interestingly, KaiC failed to enhance the kaiA1-KaiB interaction (7).
KaiA is composed of two domains, both solution structures of which were recently solved (14,16). The structure of the NH 2terminal domain of KaiA is that of a pseudo-receiver domain (14), similar to those found in bacterial response regulators (17). Although the fold is that of a canonical receiver domain, the primary sequence is dissimilar, and it lacks the conserved aspartate residue necessary for phosphorylation. The solution structure of KaiA COOH-terminal domain showed a novel all ␣-helical homodimeric fold (16). Our earlier results show that the COOH-terminal domain of KaiA possesses a groove, which likely binds to the linker region between the CI and CII domains of KaiC. This interaction between KaiA and KaiC enhances the autokinase activity of KaiC.
Here, we present the structure of full-length KaiA, which was determined by x-ray crystallography to a resolution of 2.0 Å. The full-length structure provides insight into the mechanism of KaiA regulation.

EXPERIMENTAL PROCEDURES
Expression, Purification, and Crystallization-The gene corresponding to Synechococcus elongatus KaiA was subcloned in a pET32aϩ (Novagen) expression vector, and protein was overexpressed in Escherichia coli BL21(DE3) by making the solution 1 mM in isopropyl-␤-Dthiogalactopyranoside. Cell lysates were prepared by French press in a buffer containing 200 mM potassium chloride, 20 mM Tris-HCl, pH 7.0. The protein was purified by metal affinity chromatography, cleaved by * This work was supported by the Robert A. Welch Foundation (to J. C. S.) and National Institutes of Health Grants GM064576 (to A. C. L.) and GM62410 (to J. C. S.). Use of the Argonne National Laboratory Structural Biology Center beamlines at the Advanced Photon Source was supported by the United States Department of Energy Office of Energy Research under contract number W-31-109-ENG-38. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The enterokinase (Novagen), and further purified by metal affinity chromatography, two steps of ion exchange chromatography and gel filtration. During the last gel filtration step KaiA was exchanged to the final buffer of 20 mM sodium chloride, 20 mM HEPES, pH 7.0. The protein was concentrated to 10 -15 mg/ml by stirred ultrafiltration. Protein concentration was determined by absorbance at 280 nm, using a corrected extinction coefficient as suggested by Pace et al. (18). KaiA, as prepared, has the following leading NH 2 -terminal sequence AMADIV followed by the wild-type sequence starting on residue Leu 2 from our overexpression system.
The protein was crystallized in Micro Batch plates at 20°C under Al's oil (Hampton Research) with a 4-l sitting droplet consisting of a 1:1 mixture of the stock protein solution and a solution containing 130 -180 mM ammonium sulfate, 85 mM sodium cacodylate, pH 6.5, 19 -22% polyethylene glycol 8000, and 15% v/v glycerol. Thick plate shaped crystals appeared in about 2 days and grew to about 0.5 mm in the longest dimension and varying width and thickness. Our attempts to obtain crystals of Se-Met incorporated KaiA were unsuccessful.
X-ray Data Collection and Structure Determination-For x-ray data collection, crystals were flash-frozen at 100 K using paratone as cryoprotectant. KaiA crystals belong to space group P2 1 with unit cell dimensions of a ϭ 47.1 Å, b ϭ 125.8 Å, c ϭ 56.8 Å and ␤ ϭ 114.9°, two monomers per asymmetric unit, and a solvent content of 45.6%. Data were collected at Advanced Photon Source (APS) beamlines and processed with HKL2000 programs (19). Several native data sets were collected, the best one diffracted to 2.0 Å. For heavy atom derivatization, the native crystals were soaked in varying concentrations of heavy atom solutions made in synthetic mother liquor. Extensive screening of a large number of heavy atom soaked crystals resulted in one useful 2.8-Å Nd derivative. This derivative was obtained by 24-h soaking of crystals grown under the aforementioned conditions in a 1:1 mixture of stock solution (20 mM sodium chloride, 20 mM HEPES, pH 7.0) and a solution containing 170 mM ammonium acetate, 85 mM sodium cacodylate, pH 6.5, 25% polyethylene glycol 8000, 15% v/v glycerol, and 10 mM neodymium acetate.
Molecular replacement trial with the KaiA135N model (14) using the program AMoRe (20) yielded a weak solution for the first molecule with correlation coefficient of 28.8%, but failed in searching the second KaiA135N molecule. A highly redundant data set to 2.8 Å collected at an in-house x-ray generator lacked isomorphism with the existing native data sets. Experimental phases were derived by SAD 1 methods using only the Nd derivative. The substructure of Nd in the asymmetric unit was solved with the program SHELXD (21) with only one Nd site. The Nd site was refined using the program SHARP (22), and initial protein phases were calculated. The phases resulting from Nd derivative were improved by solvent flattening using the program DM (20). The resulted phases were too poor to be used for chain tracing and model building. Real space molecular replacement search was performed with the KaiA135N model on the molecular replacement server of New York Structural Genomics Research Consortium (NYSGRC) (russel.bioc.aecom.yu.edu/cgi-bin/inhouse/rotptf/rotptf.cgi). Encouragingly, the top solution from the real space molecular replacement search matched the solution from AMoRe. Therefore, the top two solutions were combined to be a dimer model. The dimer model looked reasonable judged by their packing in the lattice. A molecular replacement search with the dimer model yielded a solution with a correlation coefficient of 33.8%. The experimental phases were further improved by solvent flattening and 2-fold non-crystallographic symmetry averaging averaging using the program DM (20) by assuming the model was correct. Since CD spectra of the KaiA COOH-terminal domain revealed primarily ␣-helical structures (14), a real space molecular replacement search was performed again with several ␣-helices with different lengths. The results were conformed upon electron density map, and six ␣-helices were added to the model (two of the six ␣-helices, corresponding to ␣9 of the two molecules, turned out to be correct).
Model Building and Refinement-Iterative cycles of model building with XtalView (23) using the ARP/wARP improved phases figure of merit weighted 2F o Ϫ F c maps (24) and refinement with REFMAC (20) on the 2.0-Å resolution data set were performed for improving the quality of the model. Water molecules were added in the difference electron density maps at positions corresponding to peaks (Ͼ3.0 ) and with appropriate hydrogen bonding geometry. The details of the final refinement parameter are presented in Table I. The figures were pre-pared using Molscript (25), Bobscript (26), and GLR 2 and rendered by POV-Ray (www.povray.org). The atomic coordinates of KaiA have been deposited in the Protein Data Bank with the ID code 1R8J.

RESULTS AND DISCUSSION
Structure Determination and Overall Structure-KaiA with NH 2 -terminal sequence AMADIV followed by the wild-type sequence starting from residue Leu 2 was overexpressed in E. coli BL21 (DE3). The crystals belong to space group P2 1 with unit cell dimensions of a ϭ 47.1 Å, b ϭ 125.8 Å, c ϭ 56.8 Å, ␤ ϭ 114.9°. The asymmetric unit consists of two chains forming a homodimer. The structure was determined by a combination of SAD phasing, real space molecular replacement, ARP/wARP phase improvement and non-crystallographic symmetry averaging. Five NH 2 -terminal non-KaiA residues, two COOH-terminal residues, and two loop regions (residues 85-92 and 137-146) are highly flexible and were omitted from the refinement ( Table I). The current model, refined at 2.0-Å resolution, has a crystallographic R value of 21.0% and an R free of 27.1%. The model has good geometry (Table I), and 100% of the non-glycine backbone dihedral angles is in the most favored or allowed regions.
KaiA exists as a dimer in the crystal (Fig. 1A), in agreement with the data from gel filtration and ultracentrifugation equilibrium (16) and previous reports that KaiA exists as homodimer or heteromultimeric protein complexes with other clock components in vivo in a circadian fashion (8). One subunit of KaiA contains two independently folded domains connected by a canonical linker (NH 2 -terminal pseudo-receiver domain, residues 1-164, canonical linker, residues 165-173, COOHterminal four-helix bundle domain, residues 174 -284). One unique feature of the KaiA dimer is domain swapping (27), in that the domains are swapped so that the NH 2 -terminal domain of one chain pairs with the COOH-terminal domain of the other. There are no interdomain interactions within a subunit, unlike most other multidomain proteins. The two subunits twist around each other forming a dimer, related by a noncrystallographic 2-fold axis. The dimer interface between the two monomers buries a total of 3440 Å 2 accessible surface areas per subunit, about 21% of the total surface of a single subunit, which is on the high side of the range of buried surface area observed in other dimeric proteins (28).
The NH 2 -terminal portion (residues 1-135) of this domain is structurally similar to the response regulator receiver family, which is involved in a wide range of regulatory processes (29), discussed in detail below. This domain consists of a central five-stranded (␤1 to ␤5) parallel ␤-sheet flanked by two groups of ␣-helices (␣1, ␣4 and ␣2, ␣3) packed on either side of the ␤-sheet and an additional ␣-helix (␣5) lying near the amino terminus of the central ␤-strand (Fig. 1A). The COOH-terminal domain contains two parallel helix-hairpin-helix motifs that form a four-helix bundle, which represents a new protein folding motif (16).
Structure and Function of the NH 2 -terminal Pseudo-receiver Domain-The portion of the NH 2 -terminal domain that is structurally similar to the response regulator receiver family was previously studied by NMR (14). The backbone root mean square deviation (r.m.s.d.) between the KaiA crystal structure and KaiA135N solution structure (average minimized structure, Protein Data Bank ID code 1M2E) is 1.31 Ϯ 0.04 Å for residues 4 -83 and 98 -135. The sequence alignment (Fig. 1B) shows gaps/insertions in two loop regions of the NH 2 -terminal domain among the four known homologs. Most of the gaps/ insertions are in disordered regions of the protein except for those that happen to fall in crystal lattice contacts. The first disordered loop (residues 84 -97) was also observed to be highly dynamic in the solution structure (14). In addition, a cis-peptide (Ala 129 -Pro 130 ) was observed in the structure and this proline is well conserved in KaiA sequences (Fig. 1B).
The structural similarity between the KaiA NH 2 -terminal domain and the response regulator receiver family was unanticipated since no amino acid sequence homology was detected prior to solving the structure. All receiver domains contain four highly invariant residues necessary for a common Mg 2ϩ -dependent phosphoryl-transfer mechanism in the carboxyl-terminal crevice of the central ␤-strand, which correspond to Asp 12 , Asp 13 , Asp 57 (phosphate-accepting residue) and Lys 109 in E. coli. CheY (17). We aligned the amino acid sequence of KaiA NH 2 -terminal domain against the response regulator receiver family based on a Hidden Markov model (HMM), downloaded from the PFAM web site (pfam.wustl.edu), using the HMMER software (30). However, the resulting sequence alignment (Fig.  2) shows that none of the four highly invariant residues involved in phosphorylation is conserved in the KaiA sequence, indicating that the KaiA NH 2 -terminal domain lacks the phosphorylation-dependent regulatory function. The very weak relationship of KaiA to the response regulator receiver family suggests migration of function for this domain relationship. An ␣-helix (␣5) that lies near the amino terminus of the central ␤-sheet is also not found in the receiver protein structure. Three well conserved leucine residues (Leu 156 , Leu 160 , and Leu 164 ) (Fig. 1B) anchor the ␣-helix to a hydrophobic pocket on the KaiA135N surface. The residues forming the hydrophobic pocket, including Ile 5 , Val 77 , Pro 78 , Leu 122 , Phe 125 , and Leu 126 , are either conserved or chemically similar (Fig. 1B). This hydrophobic feature is not found in the response regulator receiver family (Fig. 2).
Thirty-four KaiA alleles involving mutations to 28 different residues have been reported to alter the periodicity of clock-regulated gene expression rhythms in S. elongatus (4,31). The alleles are mapped onto the KaiA structure (Fig. 1, B and C) and are distributed throughout the two domains and the canonical linker, with 44.1% of the alleles mapping to the pseudoreceiver domain, 52.9% to the four-helix bundle domain and 3.0% to the canonical linker. The alleles map to residues with structural roles, to residues with roles in dimer interactions, or to residues with potential roles in KaiC interactions, indicating that both domains and the linker are important for the circadian clock function of KaiA.
The alleles mapping to the NH 2 -terminal pseudo-receiver domain involve mutations of twelve residues. Eight of the twelve residues (Ile 9 , Ile 16 , Leu 31 , Ser 36 , Cys 53 , Val 76 , Glu 103 , and Val 131 ) are buried residues. Mutation of these residues resulted in unfolded domains or aggregation (14), reflecting their structural roles. Three residues (Gln 113 , Gln 117 , and Asp 119 ) mapped to the ␣4 helix are involved in the dimer interface, as discussed later. The role of Asp 136 is unclear, since it is in a flexible surface loop.
The COOH-terminal Four-helix Bundle Domain and Its Involvement in Dimer Structure-The solution structure of the KaiA COOH-terminal domain from the Thermosynechococcus elongatus (ThKaiA180C) was recently solved independently of this study (16). The backbone r.m.s.d. is 1.28 Ϯ 0.07 Å between the KaiA crystal structure and the ThKaiA180C solution structure (average minimized structure, Protein Data Bank ID code 1Q6A) for residues 184 -282. The main conformational change between KaiA and ThKaiA180C occurs in the loop prior to ␣-helix 6 (␣6), which is highly dynamic and in the NH 2 terminus of the ThKaiA180C solution structure (16). This loop, from Asp 174 to Pro 183 , makes a type-I reverse turn from Phe 178 to Asn 181 and a 3 10 -helix from Asp 174 to Arg 177 that snake across the four-helix bundle to the canonical linker, interacting with the loop between ␣7 and ␣8, and ␣4 from the other monomer in ͘ is the root mean square heavy atom structure factor, and E is the residual lack of closure error. R cullis is the mean residual lack of closure error divided by the dispersive or anomalous difference. R-factor ϭ ⌺͉F obs Ϫ F calc ͉/⌺͉F obs ͉, where F obs and F calc are the observed and calculated structure factors, respectively. R free ϭ R-factor calculated by using a subset ‫)%5ف(‬ of reflection data chosen ramdomly and omitted throughout refinement. r.m.s.d., root mean square deviations from ideal geometry. the dimer (Fig. 3B). Three alleles were mapped to Phe 178 and Arg 180 in this loop, which are well conserved (Fig. 1, B and C). Arg 180 forms a salt bridge with conserved residue Asp 227 , and the side chain of Phe 178 stacks with Pro 175 and forms hydrophobic interactions with the C␤ of Asp 227 and C␤ of Arg 173 (Fig.  3B). Hence, the role of Phe 178 and Arg 180 might be structural. The two subunits within a KaiA dimer in the crystal interact with each other throughout the dimer interface. We divide the dimer interactions to three regions (Fig. 1A). The first region (dimer interface I) is between the four-helix bundle domains of the two monomers. The two domains pack head-to-head in a 2-fold symmetric manner (Fig. 3A). The interactions are dominated by the coiled-coil interactions along the long helices (␣9) involving residues 260 -280 from each chain. Additional interactions involve the connecting loop between ␣7 and ␣8 from 227-230. The interface is primarily hydrophobic, containing hydrophobic clusters and aromatic stacking interactions including residue Pro 229 , Val 230 , Leu 263 , Leu 265 , Ile 266 , Ile 269 , Ala 270 , Leu 272 , Cys 273 , Tyr 276 and Ile 280 from both molecules. The hydrophobic interactions appear to be the main force driving dimer formation in this interface. The packing is further stabilized by three salt bridges, including Asp A260 -Arg B262 , Asp A227 -Arg B277 and Asp B227 -Arg A277 , and four hydrogen bonds, including that between O⑀2 of the carboxylate side chain of E274 from one molecule and the backbone amide of Val 230 from another molecule, and the carbonyl of Asp 227 from one molecule and the NH1 of Arg 277 from another molecule. Interestingly, a water molecule (Wat 168 ) forms a mediate contact between Asp B260 and Arg A262 , asymmetric to the salt bridge between Asp A260 and Arg B262 (Fig. 3A). The dimer interface I buries a total of about 990 Å 2 surface areas per subunit. Three alleles were mapped to Ile 266 , Cys 273 , and Glu 274 that are involved in this dimer interface (Figs. 1, B and C, and 3A).
The second interface (dimer interface II) forms between the strictly conserved (Fig. 1B) linkers of the two subunits (Fig.  1A). The two linkers adopt an atypical anti-parallel ␤-strand conformation, zipping with each other through eight main chain hydrogen bonds and two additional hydrogen bonds formed between the N of Lys 172 from one molecule and the carbonyl of Arg 163 from another molecule. Two hydrogen bonds also form between the N of Lys 172 from one molecule and the carbonyl of Val Ϫ1 from another molecule, which is the last NH 2 -terminal non-KaiA residue replacing Met 1 position (Fig.  3B). Four hydrophobic residues (Leu A167 , Leu B167 , Val A169 , and Val B169 ) cluster together and are exposed to solvent, forming a hydrophobic surface patch. The COOH terminus conformation is stabilized by three hydrogen bonds forming in same molecule, including two between the side chain hydroxyl of Tyr 166 and the carbonyl of Pro 281 , and one between the side chain hydroxyl of Tyr B170 and the O␥ of Ser B279 . Asymmetrically, the side chain of Tyr A170 stacks between Pro A281 and Pro B281 , occupying the non-crystallographic 2-fold axis. Only one allele was mapped to Tyr 166 in the canonical linker (Fig. 1, B and C). Mutation of this residue (Y166C) causes arrhythmia (31). The interface between the NH 2 -terminal pseudo-receiver domain of one molecule and the COOH-terminal four-helix bundle domain of the other molecule in the dimer forms the third interface (Fig. 1A). Each interface buries a total of about 860 Å 2 of surface area per monomer. ␤-Strand 5 (␤5) and ␣-helix 4 (␣4) from the pseudo-receiver domain, along with ␣-helix 7 (␣7) and the NH 2  droxyl of Tyr 171 . Phe 225 mainly forms hydrophobic interactions with the main chain atoms and side chain C␤ of atoms Asp 119 and Ala 120 . Even conservative changes in these residues (Q113R, Q117L, D119E, and D119G) extend the period of the circadian rhythm. Phe 225 is well conserved in KaiA sequence (Fig. 1B). Mutation of Phe 225 (F225S) causes arrhythmia (31).
Although the COOH-terminal domain of the KaiA crystal structure and ThKaiA180C solution structure are similar, their dimerization modes are slightly different. Superposition of molecule A of both structures showed that molecule B is rotated ϳ20°relative to A (Fig. 4). The pivot is at Cys 273 , which forms disulfide bond in the ThKaiA180C solution structure but is reduced in the KaiA crystal structure. Cys 273 is well conserved in all but one KaiAs (Fig. 1B), and substitution of that residue alters the circadian phenotype (Fig. 1C) (31). Experiments in E. coli showed that cytoplasmic proteins do not generally contain structural disulfide bonds in vivo because of the reducing environment (32). It is still unclear whether there are any physiological or genetic relevances of the disulfide bond to circadian mechanism. Salt bridges between Asp 227 and Arg 277 (Asp 227 -Arg 277 ) and the four hydrogen bonds at the dimer interface I surrounding Cys 273 are also observed in the ThKaiA180C solution structure. The side chains of Pro A281 and Pro B281 directly interact with each other in ThKaiA180C solution structure, whereas the side chain of Tyr A170 stacks between Pro A281 and Pro B281 in KaiA crystal structure, leveraging the dimer formation through the pivot.
Potential KaiC Binding Site in the Four-helix Bundle Domain-Interactions among clock proteins have been proposed as a crucial step for the feedback loop within the circadian period (3). In cyanobacteria, interactions between Kai proteins in various combinations have been reported to be important for circadian timekeeping (7). KaiA stimulates KaiC phosphorylation both in vivo and in vitro (10,14), and the function was further mapped to the KaiA COOH-terminal domain (14). A KaiA allele (R249H, kaiA2) with a mutation at the KaiA COOH-terminal domain was reported with reduced effect of KaiA stimulated KaiC autokinase activity, and this effect can be restored by a KaiC allele (A422T, kaiC15) (10), which has been mapped to one of two KaiA binding domains (C KABD2 ) of KaiC (33). This suggests that Arg 249 is involved in the direct interaction with KaiC. Residue Arg 249 , together with the other six residues (Glu 239 , Met 241 , Asp 242 , Glu 243 , Phe 244 , and Ala 245 ), cluster on the apical portion of ␣8 in the four-helix bundle domain. Mutations of these residues exhibit only long period phenotypes (31). Multiple sequence alignment shows that they are either conserved or chemically similar (Fig. 1B). Structural analysis shows that Met 241 and Phe 244 are buried residues, whereas Glu 239 , Asp 242 , Glu 243 , Ala 245 , and Arg 249 are well exposed to the solvent (Fig. 1C). The surface formed by these residues is likely involved in the direct association with KaiC.
Although the function of KaiC in the clock apparatus is still unclear, it appears to involve interaction with DNA. Its forked-DNA substrate binding activity was detected by electrophoretic mobility-shift assay (13). The ATP-induced hexameric KaiC ring structure was revealed by electron microscopy (12,13,16). Each KaiC subunit in the electron micrographs showed a dumbbell-shaped structure with two domains corresponding to duplicated CI (the first half) and CII (the second half) domains and connected by an ϳ10-Å narrow linker. Based on electron microscopy data, a KaiA-KaiC interaction model has been proposed (16) (14). KaiB causes a 50% reduction in the KaiAstimulated KaiC autophosphorylation when the KaiB:KaiA molecular ratio is 40:1 (15) and a 30% reduction when the ratio is 1:1 (14). Interestingly, KaiB caused almost 100% reduction in the KaiA COOH-terminal domain (KaiA180C)-stimulated KaiC autophosphorylation when the KaiB:KaiA180C molecular ratio is 1:1 (14). These data suggest that KaiB directly interacts with KaiA and that the NH 2 -terminal pseudo-receiver domain of KaiA antagonizes the KaiB effect.
Analysis of the three dimer interfaces reveals distinctive features. First, the dimer interface I is primarily hydrophobic and the interactions are strong. Without the NH 2 -terminal pseudo-receiver domain and the canonical linker, the COOHterminal four-helix bundle domains still strongly associate to form a dimer, as observed in the ThKaiA180C solution structure (16). Second, the main driving force in the dimer interface II is eight main chain hydrogen bonds, which are also strong. Third, the dimer interface III is a mixture of hydrophobic, hydrogen bonding, and salt bridge interactions but is primarily hydrophilic and weak, indicating that the NH 2 -terminal pseudo-receiver domain in one mononer and the COOH-terminal four-helix bundle domain in another monomer might be able to dissociate, exposing the surface between the domains.
We therefore propose an allosteric regulation mechanism, in which there exist two alternative conformations of KaiA: an open and a closed form. What we observed in the crystal structure represents the "closed" conformation. We propose that the KaiB binding site is in the KaiA COOH-terminal domain and is covered by the NH 2 -terminal pseudo-receiver domain from the other KaiA monomer in the closed conformation, forming part of the dimer interface III. The NH 2 -terminal pseudo-receiver domain allosterically regulates the KaiB binding activity. This would be similar to how the receptor binding site on the insulin surface is covered by the extended B-chain COOH-terminal peptides, which have great mobility (34). Also, an allosteric regulation mechanism has been reported for integrin (35). Furthermore, domain-swapped dimer has been indicated to confer some advantages to allosteric properties, as in bovine seminal RNase (36).
The proposed allosteric regulation mechanism may explain several previous experimental results. KaiA exists as closed conformation in regular physical condition, in which the KaiB binding site is buried in the dimer interface III. This explains the weak KaiA-KaiB interaction. KaiA activates KaiC autophosphorylation through KaiA-KaiC interaction. We propose that the KaiC-phosphorylated form might allosterically induce KaiA to change from "closed" to "open" conformation, exposing the KaiB binding site, thus dramatically enhancing the KaiA-KaiB association. One possible explanation of the enhanced interaction of kaiA1 (E103K)-KaiB in vitro is based on the location of Glu 103 in the crystal structure. Residue Glu 103 is strictly conserved in KaiA sequences (Fig. 1B), and it is buried in the structure of the dimer. The O⑀1 and O⑀2 of the carboxylate side chain of Glu 103 form strong hydrogen bonds with the backbone amides of Leu 98 , Tyr 99 , and His 100 and the side chain N⑀2 of His 105 (not shown). Mutation (E103K) on this residue will obviously break the hydrogen bonding network. Actually, this mutation (E103K) probably resulted in an unfolded KaiA135N domain, as judged by a lack of chemical shift dispersion in its spectrum (14), thus reducing the association of the dimer interface III, and thus possibly induces KaiA closed to open conformation change. This may explain the dramatically enhanced kaiA1-KaiB interaction in vitro and that KaiC failed to enhance the kaiA1-KaiB interaction. The KaiB function as an attenuator of KaiC phosphorylation requires the presence of KaiA, which suggests a heteromultimeric KaiA-KaiB-KaiC complex. Based on the KaiA-KaiC interaction model (16), a space allowing KaiB to interact with both KaiC and KaiA will be created when KaiA is in its open conformation. Without the allosteric regulation from the pseudo-receiver domain in Kai180C, KaiB 100% reduces the KaiA180C-stimulated KaiC autophosphorylation.
Cyanobacterial clock proteins, including Kai proteins and SasA, form heteromultimeric protein complexes dynamically in a circadian fashion, and the time-specific formation of the clock protein complexes is a critical process in the generation of circadian rhythm in cyanobacteria (8). It was observed that KaiC rhythmically associated with KaiA prior to binding with KaiB, suggesting that KaiA binding with KaiC enhances KaiB-KaiC interaction (15). Our proposed allosteric regulation mechanism suggested that KaiA first binds to KaiC in its closed conformation and activates KaiC autophosphorylation. The KaiC-phosphorylated form induces KaiA conformational change to the open conformation, which enhances KaiB binding to KaiA-KaiB complex. KaiB then attenuates KaiC autophosphorylation, possibly through activating dephosphorylation. It was proposed (14) that the NH 2 -terminal pseudo-receiver domain is a timing input device, which receives environmental cues, such as the signals from CikA, and regulates KaiA stimulation of KaiC autophosphorylation. Based on the KaiA crystal structure, we propose that the pseudo-receiver domain changes the circadian clock through changing the affinity in the dimer interface III.
Conclusions-In conclusion, the KaiA structure reported here provides the first complete three-dimensional view of this circadian clock protein. The KaiA monomer is composed of two domains. The NH 2 -terminal pseudo-receiver domain is structurally similar to the bacterial response regulator receiver domain, whereas the COOH-terminal four-helix bundle domain is a novel fold. It further shows the formation of a homodimer through extensive dimer interface forming between both domains and the canonical linker. Analysis of the dimer interface suggests that there exist two alternative conformations. The conformation in the crystal structure represents the closed conformation. We have proposed an allosteric regulation mechanism for KaiA-KaiB interaction and time-specific clock protein complex formation. The structure of KaiA reported here is the first step in efforts to use structural studies for elucidation of circadian clock function. It serves as a framework for further studies of the KaiABC system and other clock-related proteins.