Structural basis of the day-night transition in a bacterial circadian clock

Abstract

Circadian clocks are ubiquitous timing systems that induce rhythms of biological activities in synchrony with night and day. In cyanobacteria, timing is generated by a posttranslational clock consisting of KaiA, KaiB, and KaiC proteins and a set of output signaling proteins, SasA and CikA, which transduce this rhythm to control gene expression. Here, we describe crystal and nuclear magnetic resonance structures of KaiB-KaiC, KaiA-KaiB-KaiC, and CikA-KaiB complexes. They reveal how the metamorphic properties of KaiB, a protein that adopts two distinct folds, and the post–adenosine triphosphate hydrolysis state of KaiC create a hub around which nighttime signaling events revolve, including inactivation of KaiA and reciprocal regulation of the mutually antagonistic signaling proteins, SasA and CikA.

Circadian clocks are an ancient evolutionary adaptation found in archaea (1, 2), bacteria (3), and eukaryotes (4, 5) that synchronize genetic, epigenetic, and metabolic activities to Earth’s rotation (6–8). In humans, chronic desynchronization of cellular clocks with the environmental light/dark cycle increases susceptibility to disease (9–11). In the cyanobacterium Synechococcus elongatus, the circadian clock orchestrates gene expression for the majority of its genes (12) to regulate daytime and nighttime metabolic processes that enhance fitness (13) and acutely affect survival of these obligate phototrophs at night (14, 15).

The cyanobacterial circadian clock is an ideal model to elucidate mechanisms of biological timekeeping because the oscillator consists of only three proteins—KaiA, KaiB, and KaiC (16)—that generate circadian rhythms of KaiC phosphorylation in the presence of adenosine triphosphate (ATP) (17). Two sensor histidine kinases SasA (18) and CikA (19) self-assemble into day- and night-specific complexes with Kai protein complexes, based on the phosphorylation state of KaiC (20) to regulate the activity of the response regulator and transcription factor RpaA (21) (Fig. 1). However, the structural basis by which the clock autonomously transitions from its daytime to nighttime signaling state has remained elusive.

Fig. 1. Model of circadian formation of cyanobacterial clock protein complexes.

During the day, KaiA binds to the C-terminal extensions of KaiC to enhance the autokinase activity of the CII domain of KaiC, leading to sequential autophosphorylation. Yellow and red circles represent phosphorylated S431 (pS) and T432 (pT), respectively. The sensor histidine kinase, SasA, is activated by KaiC. During the day, the hexameric CII ring is loose and unstacked from the hexameric CI ring. At dusk, phosphorylation of S431 tightens the CII ring, leading to withdrawal of the KaiA binding sites on CII. The tightened CII ring interacts and stabilizes the CI ring in its post-ATP hydrolysis state, which now has its KaiB binding sites exposed. The rare fold-switched state of KaiB, KaiB_fs, cooperatively binds CI, displacing SasA. KaiB_fs sequesters the autoinhibited conformation of KaiA, allowing autodephosphorylation of KaiC. KaiB_fs also binds and activates the phosphatase activity of CikA. Upon autodephosphorylation of S431, the CII ring loosens, causing unstacking of the rings. The CI ring returns to its pre-ATP hydrolysis state, letting go of KaiB, KaiA, and CikA to begin a new day.

KaiC uses multiple enzymatic activities to act as the central circadian pacemaker (22). It has two homologous domains (CI and CII) that belong to the AAA+ ATPase (adenosine triphosphatase) family (23), with each domain associating into a hexameric ring (Fig. 1). The CII domain possesses autokinase (24) and phosphotransferase activities (25, 26). Autophosphorylation of two residues on the CII domain, S431 and T432, tighten and loosen the CII ring (27), respectively, which in turn regulates accessibility of a KaiA binding site on CII (28) and a KaiB binding site on CI (29, 30). Positive and negative feedback by phosphoryl T432 and phosphoryl S431, respectively, governs an ordered phosphorylation cycle throughout the day: ST → S/pT → pS/pT → pS/T → S/T (S, serine; T, threonine; p, phosphorylated) (31, 32). KaiA is an obligate dimer with no known enzymatic activities (33, 34) that stimulates KaiC CII autophosphorylation during the day (35, 36) by maintaining the C-terminal A loops of CII in their exposed state (28, 37, 38) (Fig. 1). Over the course of the day, the sensor histidine kinase SasA binds to and is activated by the CI domain of KaiC to phosphorylate RpaA, which then activates transcription of class 1 genes that show peak expression at dusk (20, 28, 29, 39, 40) (Fig. 1).

The inhibition of KaiA at night permits autodephosphorylation of KaiC to reset the cycle. Little is known about how inhibition of KaiA is achieved by interaction with a KaiB-KaiC complex (35, 41, 42). KaiB belongs to a rare class of so-called metamorphic proteins, which reversibly switch between different folds under native conditions (43). KaiB transitions from a highly populated, inactive tetrameric ground-state fold (KaiB_gs) to a rare, active-state monomeric fold (KaiB_fs) that is captured by the CI ring of KaiC when CII is predominantly in the pS/pT or pS/T states in the evening (39) (Fig. 1). In addition, formation of the KaiB_fs-KaiC complex requires ATPase activity in the CI domain (44). However, the structural basis for KaiB_fs binding and the requirement of CI domain ATPase activity in regulating these interactions has been unclear.

The phosphorylation status of the hexameric CII ring is communicated to the CI ring via phosphoryl S431-dependent ring-stacking interactions (27, 29). When stacked with CII, the CI ring captures KaiB_fs to displace SasA from KaiC via competition for overlapping binding sites on CI (28, 39), thereby turning off further stimulation of RpaA-mediated expression of class 1 genes (Fig. 1). While bound by CI, KaiB_fs recruits and inhibits KaiA (39). The KaiB_fs-KaiC complex also binds CikA to stimulate its phosphatase activity toward RpaA (21, 39), repressing class 1 genes and activating class 2 genes (peak expression at dawn) (Fig. 1). Therefore, the KaiB_fs-KaiC complex forms a hub that regulates multiple interactions to disengage SasA, inhibit KaiA, and activate CikA.

Thermophilic clock proteins from Thermosynechococcus elongatus (fig. S1) are structurally and functionally similar to those from S. elongatus and more amenable for structural studies (45). Using mutations that stabilize the rare fold of KaiB_fs from T. elongatus (39), here we present high-resolution structures of the KaiB_fs-KaiC hub alone and in complex with domains of its interacting partners, KaiA and CikA. Together, these structures explain the requirement for CI ATPase activity and KaiB metamorphosis in the inactivation of KaiA and SasA and the recruitment of CikA.

Basis of SasA inactivation by KaiB

KaiB competes with SasA for an overlapping binding site on KaiC (46, 47) at the CI domain (28, 29) to regulate clock output signaling (39). Previously, we designed a monomeric form of the isolated CI domain, hereafter referred to as CI_mono (fig. S1), that forms a stable complex with KaiB (29). Here, we designed slightly truncated forms to favor crystallization, one on the N terminus of CI_mono, hereafter referred to as CIcryst (fig. S1), and another at the C terminus of a fold-switched KaiB mutant stabilized by G89A and D91R mutations (KaiB_fs), hereafter referred to as KaiB_fs-cryst (fig. S1). Y8A and Y94A substitutions were also incorporated to enhance the stability of KaiB_fs-cryst. Using these optimized constructs, we obtained a 1.8-Å resolution crystal structure of the KaiB_fs-cryst-CI_cryst complex (Fig. 2A).

Fig. 2. Fold-switched KaiB binds to the posthydrolysis state of the CI domain of KaiC.

(A) Crystal structure of the KaiB_fs-cryst-CI_cryst complex at 1.8 Å. Orange, KaiB_fs-cryst; sky-blue, CI_cryst. (B) Secondary structures of KaiB_gs (PDB 2QKE, subunit A) and KaiB_fs-cryst bound to CI_cryst. Residues of KaiB_fs-cryst that interact with CI_cryst are highlighted in green. (C) Zoomed-in view of the boxed region in (A). Representative interacting residues in (B) are shown in green. Dashed lines: electrostatic KaiB_fs-cryst-CI_cryst interactions. (D) Superposition of CI structures before ATP hydrolysis. Sky-blue, CI_cryst; dark green, pre-ATP hydrolysis state of CI^Se (PDB 4TLC, subunit C), missing density for the B loop. Residue numbers in CI^Se are offset from those of T. elongatus CI by −1. (E) Superposition of CI structures after ATP hydrolysis. Sky blue, CI_cryst; magenta, post-ATP hydrolysis state of CI^Se (PDB 4TLA, subunit E). Please see fig. S1 and table S1 for construct details.

The interface predominantly comprises residues from the fold-switched C-terminal half of KaiB (Fig. 2, B and C), explaining the requirement of fold-switching for KaiB-KaiC complex formation (39). The interface centers on the B loop of CI_cryst, consistent with findings that deletion of the B loop and alanyl substitutions of B-loop residues E117, V118, and F122, as well as R130, abolished or weakened KaiB-KaiC binding (28, 39). Because KaiB_fs has the same thioredoxin-like fold as the domain of SasA that binds KaiC (39), their modes of binding to the CI domain are likely to be similar. Indeed, SasA binding to CI_mono was shown to be weakened by the same set of mutations (28, 39). Thus, our KaiB_fs-cryst-CI_cryst structure provides insights into the competition between KaiB and SasA that explains the down-regulation of SasA activity at night.

The post-ATP hydrolysis state of KaiC CI domain captures KaiB_fs

Mutations that tune the ATPase activity of the CI domain modulate the circadian period, but it is not yet known how this enzymatic activity exerts its intrinsic role in timekeeping (22, 44, 48). ATP hydrolysis by the CI domain is necessary for formation of the KaiB-KaiC complex (44, 48), suggesting that the hydrolysis activity is poised to control timing of the assembly of nighttime signaling complexes (49). Comparison of our adenosine diphosphate (ADP)–bound KaiB_fs-cryst-CI_cryst complex with structures of the CI^Se domain before and after ATP hydrolysis (superscript “Se” denotes proteins from S. elongatus) (22) revealed the structural basis for this obligate ATP hydrolysis. Residues at the CI ATP binding site display large conformational differences before and after hydrolysis (22) (Fig. 2, D and E). In attaining the posthydrolysis state, residue F200 near the ATP binding site rotates downward by ~130°, accompanied by the downward movement of helices α6 and α7. Our structure shows that α6 residues Q154 and Y155 drop down after ATP hydrolysis to form part of the KaiB_fs-cryst-CI_cryst interface (Fig. 2, D and E). Thus, ATP hydrolysis by the CI domain powers a conformational switch that is necessary to bind KaiB_fs.

Previous reports have shown that KaiB forms a hexameric assembly on KaiC (50, 51). To study the KaiB-KaiC complex in its natural hexameric state, we used a KaiB fold-switch variant harboring an I88A substitution, hereafter referred to as KaiB_fs-cryst* (fig. S1). Upon incubating KaiB_fs-cryst* with a KaiC_S431E mutant that mimics phosphoryl-S431 KaiC (fig. S1), we observed formation of a monodisperse hexameric assembly in solution (fig. S2), similar to that observed by native mass spectrometry with wild-type KaiB (50). To understand the structural basis for this hexameric assembly, we solved a 3.87-Å resolution crystal structure of the KaiB_fs-cryst*-KaiC_S431E complex (Fig. 3A). All previous crystal structures of full-length KaiC, including those with the S431D phosphomimetic mutation, depicted the prehydrolysis state of the CI and CII domains because the proteins were crystallized with ATP-γ-S (52). Our structure of the KaiB_fs-cryst*-KaiC_S431E hexamer was crystallized in the presence of ATP but had density for ADP between each CI subunit within the hexamer (Fig. 3B), as well as the concomitant conformational changes at CI helices α6 and α7 that accompany ATP hydrolysis (22) (fig. S3A). Thus, the fold-switched state of KaiB is captured by the post-ATP hydrolysis state of the CI ring within the KaiC hexamer.

Fig. 3. KaiB assembles as a ring on the posthydrolysis state of KaiC.

(A) The hexameric KaiB_fs-cryst*-KaiC_S431E complex at 3.87 Å. KaiC_S431E includes CII hexamer (dark cyan) and CI hexamer (sky blue). Orange, KaiB_fs-cryst*. (B) Zoomed-in view of the boxed region in (A) showing bound ADP highlighted by mesh representing the (F_obs − F_calc) omit maps contoured at 2.5σ. F_obs is the observed structure-factor amplitude, and F_calc is the calculated structure-factor amplitude. (C) Zoomed-in view of the boxed region in (A) showing KaiB_fs-cryst* interfaces in the KaiB_fs-cryst*-KaiC_S431E hexamer. Interfacial residues are shown in nontransparent mode. (D) Columbic electrostatic surface map of the interfacial residues between two subunits of KaiB_fs-cryst*. See fig. S1 and table S1 for construct details.

How does ATP hydrolysis by the CI domain functionally integrate with the phosphorylation cycle of the KaiC CII domain? When CII residue S431 is not phosphorylated, the CII ring is loose (27), the CII and CI rings are unstacked (29), CI ATPase activity is elevated (48), and KaiC is unable to bind KaiB (29). Upon S431 phosphorylation, the CII ring tightens (27), CII and CI rings stack (29), CI ATPase activity is suppressed (48), and KaiC becomes receptive to binding KaiB (29). Our crystal structures of the KaiB_fs-KaiC complexes (Figs. 2 and 3) provide a plausible explanation for these observations: Based on the inherently low rate of ATP hydrolysis by the CI domain (~0.5 ATP h⁻¹CI⁻¹) (22), CI domains make infrequent and transient excursions from a nonreceptive prehydrolysis state to a KaiB-receptive posthydrolysis state, which is captured and stabilized by the CII ring tightened from S431 autophosphorylation (29). Therefore, the slow rate of ATP hydrolysis by the CI domain creates a time delay between CII ring tightening and CII-CI ring stacking, as well as encoding a conformational switch needed to bind KaiB. Indeed, mutations that decrease CI ATPase activity extend clock period to over 40 hours (22). However, because CI ATPase activity does not affect CII autophosphorylation and autodephosphorylation rates (44), increasing the hypothetical rate of ATP hydrolysis at the CI domain does not further shorten the period below ~22 hours (fig. S4). Hydrolyzing one ATP every ~2 hours, the CI ATPase tunes the period of the oscillator from a lower limit of ~22 hours to ~24 hours (~22 + ~2 hours). Therefore, ATP hydrolysis by the KaiC CI domain works in concert with the phosphorylation cycle at the KaiC CII domain to control KaiB binding and assembly of nighttime signaling complexes.

KaiB forms a hexameric ring on the CI domains of KaiC

The KaiB_fs-cryst*-KaiC_S431E CI interface observed in the full-length hexamer structure agrees well with that of the isolated KaiB_fs-cryst-CI_cryst heterodimer (fig. S3B). The hexameric complex is also consistent with a three-tiered envelope derived from a 16-Å-resolution cryogenic electron microscopy study (51) (fig. S3C). Monomers of KaiB_fs-cryst* assemble in a ringlike structure on the bottom face of the KaiC_S431E CI domain (Fig. 3), forming KaiB-KaiB interactions that are likely to promote the cooperative assembly observed in vitro (50). The fold-switched C-terminal α3 helix of one KaiB_fs-cryst* protomer interacts with the N-terminal α1 helix of its clockwise neighbor, increasing the interfacial area by 206 Ų over the average KaiB_fs-cryst*-KaiC_S431E CI interface area of 983 Ų, adding ~20% more binding surface (Fig. 3C). Our structure of the hexameric assembly shows that residue R23 is at the center of the interface between two adjacent KaiB_fs-cryst* subunits, contributing to the electrostatic complementarity of the KaiB-KaiB interface (Fig. 3, C and D). Consistent with a role in cooperative assembly of the KaiB-KaiC hexamer (50, 53), an R23A mutation in KaiB^As (superscript “As” denotes proteins from Anabaena sp.) reduced its binding affinity for KaiC^As (54). Cooperative assembly of the KaiB-KaiC hexamer may facilitate a robust transition from subjective day to night in the clock by helping to efficiently capture the rare fold-switched form of KaiB.

Structural basis of cooperative assembly of the KaiA-KaiB-KaiC CI complex

The CII domain of KaiC autophosphorylates under stimulation by KaiA during the day (35, 36), whereas it autodephosphorylates at night when KaiA is inhibited by KaiB (41), presumably sequestered in a KaiA-KaiB-KaiC complex (39, 55) (Fig. 1). Both full-length KaiA or KaiA_ΔN (fig. S1), a construct missing the N-terminal domain yet retaining the ability to stimulate KaiC autophosphorylation and be inhibited by KaiB (28), are capable of cooperatively forming a KaiA-KaiB-KaiC ternary complex (28) (fig. S5, A to C) and promoting disassembly of the SasA-KaiC output signaling complex (28).

To begin to examine molecular determinants of the KaiA_ΔN-KaiB_fs-KaiC CI ternary complex in solution, we collected the methyl-TROSY (transverse relaxation-optimized spectroscopy) nuclear magnetic resonance (NMR) spectrum of U-[¹⁵N, ²H]-Ile-δ-[¹³C, ¹H]–labeled KaiA_ΔN in complex with KaiB_fs-cryst and CI_mono. We found that it was virtually identical to the spectrum bound to wild-type KaiB and CI_mono (fig. S5B), demonstrating that the conformationally locked KaiB variant and natively captured KaiB both interact with KaiA_ΔN in a similar manner. We also used one additional mutation (C272S) in KaiA_ΔN, hereafter referred to as KaiA_cryst (fig. S1), to prevent intramolecular disulfide formation within the KaiA dimer during the crystallization process (56). We then solved a 2.6-Å resolution crystal structure of the ternary complex KaiA_cryst-KaiB_fs-cryst-CI_cryst (Fig. 4A), consistent with the stoichiometry of 2 KaiA:1 KaiB:1 CI that we previously observed in solution (28).

Fig. 4. The mechanism of KaiA autoinhibition is revealed by the KaiA_cryst-KaiB_fs-cryst-CI_cryst complex.

(A) The ternary KaiA_cryst-KaiB_fs-cryst-CI_cryst complex at 2.6 Å. Orange, KaiB_fs-cryst; sky blue, CI_cryst; orchid, KaiA_cryst. (B) Zoomed-in view of the boxed region in (A). Dashed lines: electrostatic interactions. (C) Conformational changes of dimeric KaiA. NTD, N-terminal domain. Prime symbols denote the other protomer within the dimer. (Left) Crystal structure (PDB 5C5E) of KaiA^Se (purple) bound to KaiC CII^Se peptides (dark cyan). (Right) The KaiA_cryst-KaiB_fs-cryst-CI_cryst complex, with same coloring scheme as in (A). (Middle) Superposition of KaiA^Se (left) and KaiA_cryst in complex (right); only α5 helices and b6 strands are shown. (D) Fluorescence anisotropy of 6IAF-labeled CII peptides (0.05 μM). Open circles, 0 μM KaiA, titration with CI_cryst and KaiB_fs-cryst; triangles, 10 μM KaiA, titration with CI_cryst; diamonds, 10 μM KaiA, titration with KaiB_fs-cryst; squares, 10 μM KaiA, titration with equal molar of KaiB_fs-cryst and CI_cryst. Error bars, SD from triplicates. (E) Size-exclusion chromatography of ternary complex formation. Wild-type KaiA (green, dashed); wild-type KaiB (red, dashed); CI_cryst (orange, dashed); KaiA +KaiB+CI_cryst (green); L155A-KaiA+KaiB+CI_cryst (cyan); K158A-KaiA+KaiB+CI_cryst (blue); D212A-KaiA+ KaiB+CI_cryst (purple); D266A-KaiA+KaiB+CI_cryst (black); and N212A-D266A-KaiA+ KaiB+CI_cryst (red). (F) Bioluminescence rhythms from strains of S. elongatus: wt-kaiA^Se (blue), complemented with kaiA^Se (green), L156A-kaiA^Se (purple), and kaiA^Se knockout (red). L156A of kaiA^Se is analogous to L155A in kaiA. See fig. S1 and table S1 for construct details.

The structure of this isolated ternary complex demonstrates the molecular basis for cooperative assembly of KaiA-KaiB_fs-KaiC complexes. First, conserved residue K137 of CI_cryst forms a hydrogen bond with Q52 of KaiB_fs-cryst and a salt bridge to E161 of KaiA_cryst (Fig. 4B). Second, a triple junction of interactions was also observed between K137 and Y133 of CI_cryst; E161, R162, and N212 of KaiA_cryst; and Q52 and E56 of KaiB_fs-cryst. Finally, the presence of KaiA_cryst increased the number of KaiB_fs-cryst-CI_cryst hydrogen bonds from 13 to 19 (Figs. 2 and 4), and the KaiB_fs-cryst-CI_cryst interfacial surface area from 1000 to 1075 Ų (Figs. 2 and 4), supporting previous observations that KaiA induces an increase in apparent affinity of KaiB for the CI domain (28). This cooperative assembly also promotes SasA displacement from KaiC (28), likely eliciting a switchlike behavior as the clock transitions from daytime to nighttime mode.

KaiA adopts an autoinhibited conformation in the KaiA-KaiB-KaiC complex

During the day, the KaiA homodimer uses two symmetrically related sites located on its α9 helices to bind C-terminal peptides emanating from the CII domain (28, 37, 38, 57) (Fig. 4C). At night, KaiA is somehow prevented from doing so upon sequestration into a KaiA-KaiB_fs complex on the KaiC CI domain (28, 39) (Fig. 1). Because asymmetric sequestration of one KaiC binding site on KaiA could in principle still allow KaiA to bind and stimulate a second KaiC particle, both KaiC binding sites on KaiA should somehow be inactivated to allow KaiC to fully autodephosphorylate. However, it was unclear how this potent inactivation of KaiA could be achieved by binding to only one KaiB_fs monomer docked onto the KaiC CI domain (28).

Our structure of the ternary complex reveals that the KaiA_cryst dimer undergoes large-scale conformational changes when it binds to the KaiB_fs-cryst-CI_cryst complex, prohibiting itself from interacting with the CII domain (Fig. 4C). The b6 strands from KaiA_cryst monomers rotate over 70° relative to their previous positions in KaiA^Se (fig. S6); a β6 strand from one of the KaiA_cryst monomers then docks onto the β2 strand of KaiB_fs-cryst to form an antiparallel β sheet (Fig. 4). This asymmetric binding interface with KaiB_fs-cryst is also accompanied by a downward rotation of both a5 helices of KaiA_cryst to pack onto the KaiC binding sites on a7 and a9 of the KaiA_cryst dimer interface, thereby mimicking CII peptide binding (57) (Fig. 4C). Thus, both KaiC binding sites on KaiA_cryst are effectively blocked by their own α5 helices in this autoinhibited conformation through a long-range allosteric mechanism. This structure, along with previous solution NMR studies (33, 35, 37, 38), supports a model in which free KaiA, like KaiB, is in a dynamic equilibrium between active and inactive conformations; in the case of KaiA, these dynamics tune the stimulatory effect of KaiA on KaiC autophosphorylation (35, 37). Phosphoryl S431-dependent hexameric CII ring tightening allows the CII ring to withdraw its A loops from KaiA (28), and the inactive conformation then becomes selectively bound by the KaiB_fs-KaiC complex at night.

In support of KaiB_fs-induced autoinhibition of KaiA, hydrogen-deuterium exchange mass spectroscopy (HDX-MS) revealed strong protection of the α5 helix of KaiA_ΔN and the β2 strand of KaiB_fs-cryst upon formation of the KaiA_ΔN-KaiB_fs-cryst-CI_cryst complex in solution, matching the interfaces observed in our crystal structure (Fig. 4 and fig. S7A). To rule out that the autoinhibitory conformation is not influenced by our truncation of the N-terminal domain in KaiA, we used a fluorescence polarization–based displacement assay to probe for direct competition with the CII peptide that binds KaiA. In this assay, 6-iodoacetamidofluorescein (6IAF)–labeled CII peptides remained bound to full-length KaiA in the presence of either KaiB_fs-cryst or the KaiC CI_cryst domain but was efficiently competed away with the KaiB_fs-cryst-CI_cryst complex (Fig. 4D). Therefore, binding to the KaiB_fs-KaiC CI complex is sufficient to stabilize the autoinhibitory conformation in native KaiA and block interaction with the KaiC CII domain.

We conducted structure-guided mutagenesis to test our model that the α7 and α9 helices of KaiA_cryst appear to be crucial for stabilizing the α5 helices in the autoinhibited conformation that binds to the KaiB_fs-KaiC CI complex (Fig. 4C). We identified four substitutions on the α5 helix of KaiA (L155A, K158A, L159A, and L163A) that weakened ternary complex formation as determined by size-exclusion chromatography (Fig. 4E and fig. S7B). Likewise, mutagenesis of α5-facing residues N212 and D266 on α7 and α9 of KaiA also weakened formation of the ternary complex (Fig. 4E), further demonstrating the importance of the KaiA conformational change in binding to the KaiB_fs-KaiC CI complex.

Mutagenesis of A41D and K43E on the β2 strand of KaiB_fs eliminated KaiA_ΔN binding to the KaiB_fs-CI_mono complex by disrupting formation of the antiparallel β sheet (fig. S7C). The structural integrity of KaiB_fs mutants remained intact because they were still able to form complexes with CI_mono (fig. S7C). Likewise, an ¹⁵N, ¹H-HSQC (heteronuclear single-quantum coherence) spectrum of L155A-KaiA_ΔN showed similarly dispersed peaks as compared with the spectrum of KaiA_ΔN, indicating that the L155A mutation did not appreciably affect the structure of KaiA_ΔN (fig. S8A). We found that analogous mutations of the S. elongatus proteins, L156A in kaiA^Se, and A40D and K42E in kaiB^Se, individually disrupted cellular circadian rhythms in vivo (Fig. 4F and fig. S8D). Moreover, the A40D and K42E substitutions in kaiB^Se also produced abnormally long cells, consistent with the kaiB knockout phenotype (58) (fig. S8, E and F). Finally, superimposing the KaiA_cryst-KaiB_fs-cryst-CI_cryst ternary complex onto our hexameric KaiB_fs-cryst*-KaiC_S431E assembly shows that KaiA_cryst likely protrudes out from the side of the KaiC_S431E hexamer (fig. S9), consistent with negative-stain electron microscopy images of a presumed complex of native KaiA-KaiB-KaiC (59).

CikA-KaiB interactions critical for CikA activation

At night, the phosphatase activity of CikA is stimulated upon binding the KaiB_fs-KaiC complex to dephosphorylate and thereby repress RpaA-dependent regulation of class 1 genes and activate transcription of class 2 genes (21, 39) (Fig. 1). Although the molecular basis of this interaction has been unclear, the pseudoreceiver domain of CikA (CikA_PsR) (fig. S1) is known to directly interact with the KaiB_fs-KaiC complex (39, 60). Overexpression of this domain alone shortened cellular circadian rhythms by as much as 4 hours (60), and addition of the isolated CikA_PsR domain to an in vitro cycling reaction with purified Kai proteins shortened the circadian period in a dose-dependent manner (39). These data suggest that the CikA_PsR domain can interfere somehow with formation of Kai protein interactions that contribute to normal circadian timekeeping.

To probe how the PsR domain of CikA can influence circadian rhythms, we first investigated how CikA_PsR interacts with KaiB_fs-CI_mono complexes by NMR (fig. S5, D to I). Methyl-TROSY spectra of U-[¹⁵N, ²H]-Ile-δ-[¹³C, ¹H]–labeled CikA_PsR showed that it interacts with a complex of CI_mono bound to the fold-switched conformation induced from native KaiB but not to either protein separately (fig. S5, D and E). In support of this finding, NMR spectra of CikA_PsR bound to complexes with the conformationally locked KaiB mutant, KaiB_fs-CI_mono, and those originating from wild-type KaiB-CI_mono were similar (fig. S5E). We noted that CikA_PsR adopts two conformations in the ternary complex with wild-type KaiB and CI_mono, but only one of these is shared with KaiB_fs and CI_mono (fig. S5E). In the absence of CI_mono, KaiB_fs interacted more weakly with CikA_PsR (fig. S5F), suggesting that cooperativity is also important in formation of a CikA_PsR-KaiB-KaiC CI ternary complex, as was observed with KaiA-KaiB-KaiC binding (28) (fig. S5, A to C).

During our investigation, we identified an N29A substitution on KaiB_fs, hereafter referred to as KaiB_fs-nmr (fig. S1), that promoted virtually complete binding to CikA_PsR even without CI_mono (fig. S5G). We solved a solution structure of a CikA_PsR-KaiB_fs-nmr complex (Fig. 5) as a model to understand how CikA might bind to the KaiB_fs-KaiC complex. This complex has a binding interface of 1230 Ų that includes β2 of CikA_PsR and β2 of KaiB_fs-nmr, which interact in parallel to build a nine-stranded β sheet that spans both proteins (Fig. 5A). Residue A29 on KaiB_fs-nmr was found to interact with hydrophobic residues I641 and L654 on CikA_PsR (Fig. 5A), suggesting that the observed enhancement in binding arising from the N29A substitution was likely due to an increased hydrophobic effect. Residue I641 of CikA_PsR sits at the center of the β2-β2 heterodimeric interface (Fig. 5A), consistent with our observations that it experienced the largest chemical shift perturbation (fig. S5I) and was protected against hydrogen-deuterium exchange upon binding KaiB_fs-nmr in solution (fig. S10A).

Fig. 5. Structure of the CikA_PsR-KaiB_fs-nmr complex reveals intermolecular contacts essential for output signaling.

(A) Top, ensemble of the 20 lowest-energy NMR structures of the CikA_PsR-KaiB_fs-nmr complex. Orange, KaiB_fs-nmr; khaki, CikA_PsR. Bottom, zoomed-in view of boxed regions above using average minimized structure from the ensemble. Dashed lines, intermolecular electrostatic interactions. (B) Bioluminescence rhythms from S. elongatus. wt-cikA^Se (blue), complemented with cikA^Se (green), C644R-cikA^Se (purple), or cikA^Se knockout (red). C644R of cikA^Se is analogous to C630R of cikA. (C) Representative cell micrographs. wt-cikA^Se, complemented with cikA^Se, C644R-cikA^Se, or lacking cikA^Se. Red, membrane autofluorescence. Scale bars, 2.5 μm. (D) Scatter plots of cell lengths. wt-cikA^Se (blue) (n = 87), complemented with cikA^Se (green) (n = 110), C644R-cikA^Se (purple) (n = 353), or lacking cikA^Se (red) (n = 48). One-way analysis of variance of log₁₀-transformed cell length data produced P < 0.0001. ****, Bonferroni-corrected P values (< 0.0001) for pairwise comparisons to wt-cikA^Se (α = 0.05). (E) CikA_PsR and KaiA_cryst compete for the β2 strand of KaiB_fs. (Left) Orange, KaiB_fs-nmr (top) and KaiB_fs-cryst (bottom); khaki, CikA_PsR; orchid, KaiA_cryst; black lines, backbone-backbone or sidechain-backbone hydrogen bonds. (Right) Selected region of methyl-TROSY spectra of labeled CikA_PsR free (black), or mixed with wild-type KaiB+CI_mono (blue), or with wild-type KaiB+CI_mono+wild-type KaiA (red). See fig. S5, E and H, for full spectra. Also see fig. S1 and table S1 for construct details.

To examine the relevance of the CikA_PsR-KaiB_fs-nmr interface in vivo, we mutated CikA_PsR residue C630, which interacts with A41 in KaiB_fs-nmr at the center of the interface (Fig. 5A). The C630R substitution eliminated complex formation (fig. S10B) while maintaining structural integrity of CikA_PsR (fig. S8, B and C). Cyanobacterial strains that harbored the analogous C644R mutation in cikA^Se exhibited defects in circadian rhythms and cell division (58) similar to those observed in cikA^Se knockout strains (Fig. 5, B to D). Thus, in vitro and in vivo experiments demonstrate that the structure of the CikA_PsR-KaiB_fs-nmr complex accurately reflects interactions between wild-type proteins.

A side-by-side comparison of the parallel β2-β2 CikA_PsR-KaiB_fs-nmr and antiparallel β6-β2 KaiA_cryst-KaiB_fs-cryst interfaces revealed that KaiB_fs uses the same β2 strand to bind both KaiA_cryst and CikA_PsR (Fig. 5E). A41D and K43E substitutions in β2 of KaiB_fs-cryst, which weakened KaiA_ΔN-KaiB_fs-cryst interactions (fig. S7C), similarly reduced binding of CikA_PsR to KaiB_fs-nmr (fig. S10C). Consistent with their overlapping binding sites, methyl-TROSY spectra showed that both full-length KaiA and KaiA_ΔN could displace CikA_PsR from a KaiB-CI_mono complex (Fig. 5E and fig. S5H). Conversely, elevated levels of CikA_PsR would be expected to similarly displace KaiA from KaiB_fs-KaiC complexes, likely explaining how increased levels of the CikA_PsR domain shorten the period of circadian rhythms in vivo (39) and in vitro (60). Notably, although KaiA_cryst and CikA_PsR bind predominantly to the N-terminal half of KaiB_fs, which is conformationally similar in both the ground state (KaiB_gs) and fold-switched conformations (KaiB_fs) (fig. S11), our studies collectively show that both preferentially bind the fold-switched conformation of KaiB that is stabilized upon binding to the KaiC CI domain. A closer examination of both complexes reveals that the β4 strand and β2-to- β3 loop of the ground-state conformation (KaiB_gs) would create putative charge repulsions and steric clashes with KaiA_cryst and CikA_PsR (Fig. 6). Therefore, by binding to the same β2 strand on KaiB_fs, both KaiA and CikA_PsR are regulated by the same fold-switch mechanism of KaiB.

Fig. 6. Ground-state KaiB prevents daytime KaiA and CikA_PsR recruitment.

(A) Superposition of the non–fold-switched region (residues 7 to 46) of KaiB_gs (blue) (PDB 2QKE, subunit A) with the ternary KaiA_cryst-KaiB_fs-cryst-CI_cryst structure. (B) and (C) are zoomed-in views of the boxed region in (A). Dashed lines, estimated distance between two atoms. (D) Superposition of the non–fold-switched region (residues 7 to 46) of KaiB_gs (blue) (PDB 2QKE, subunit A) with the CikA_PsR-KaiB_fs-nmr structure. (E) and (F) are zoomed-in views of the boxed region in (D). Dashed lines, estimated distance between two atoms.

Discussion

Here, we presented four structures assembled by the KaiB_fs-KaiC hub, all of which are obligately dependent on the rare, metamorphic fold-switching of KaiB and ATP hydrolysis by the KaiC CI domain. Together, they reveal how this posttranslational circadian clock transitions from day to night and provide the basis for a revised mechanistic model of cyanobacterial circadian rhythms (Fig. 1). Additionally, structural analysis revealed that KaiA exists in dynamic equilibrium between active and inactive states, the latter of which is stabilized by interaction with the KaiB_fs-KaiC hub. Because KaiA and CikA compete for overlapping binding sites, as do KaiB and SasA, regulation of the oscillator and output signal transduction are tightly coupled. Although no metamorphic proteins have been reported in the transcription-based vertebrate circadian clock, several clock protein interactions depend similarly on competition for mutually exclusive binding events (61), suggesting that some parallels in timekeeping strategies may exist in the network architectures of diverse circadian clocks.