Report

Atomic-scale origins of slowness in the cyanobacterial circadian clock

See allHide authors and affiliations

Science  17 Jul 2015:
Vol. 349, Issue 6245, pp. 312-316
DOI: 10.1126/science.1261040

Biochemical basis of a 24-hour clock

Circadian clocks keep organisms in synch with such daily cycles as illumination, activity, and food availability. The circadian clock in cyanobacteria has the necessary 24-hour period despite its three component proteins having biochemical activities that occur on a much faster time scale. Abe et al. focused on the cyanobacterial clock component KaiC, an adenosine triphosphatase (ATPase) that can autophosphorylate and autodephosphorylate. The slow ATPase activity of KaiC, which is linked to a peptide isomerisation, provided the slow kinetics that set the speed of the 24-hour clock. Chang et al. found that another clock component, KaiB, also has slow changes in its protein conformation that help to set the oscillation period of the clock and its signaling output.

Science, this issue pp. 312 and 324

Abstract

Circadian clocks generate slow and ordered cellular dynamics but consist of fast-moving bio-macromolecules; consequently, the origins of the overall slowness remain unclear. We identified the adenosine triphosphate (ATP) catalytic region [adenosine triphosphatase (ATPase)] in the amino-terminal half of the clock protein KaiC as the minimal pacemaker that controls the in vivo frequency of the cyanobacterial clock. Crystal structures of the ATPase revealed that the slowness of this ATPase arises from sequestration of a lytic water molecule in an unfavorable position and coupling of ATP hydrolysis to a peptide isomerization with high activation energy. The slow ATPase is coupled with another ATPase catalyzing autodephosphorylation in the carboxyl-terminal half of KaiC, yielding the circadian response frequency of intermolecular interactions with other clock-related proteins that influences the transcription and translation cycle.

Circadian clocks comprise suites of biological processes that oscillate with a 24-hour period (1). Clock genes and clock proteins are present in prokaryotes and eukaryotes (2, 3); together, they constitute feedback loops that effect transcriptional and translational oscillations (TTOs). The origin of the slow circadian time scale is thought to be the time delay between clock gene transcription and feedback signals that regulate it; however, the transcriptional and translational events can occur quickly (i.e., within minutes) (4). Posttranslational oscillations (PTOs) (57) in biochemical modifications of clock proteins occur even without transcriptional and translational regulation. Proteins generally exhibit dynamics within picoseconds or seconds (8), much faster than the circadian time scale. Thus, both TTO and PTO circadian systems are assembled from building blocks with intrinsically fast dynamics, raising questions about how and why the systems are so slow and stable overall (9).

The cyanobacterium Synechococcus elongatus PCC7942 is the simplest organism known to have both TTOs (10) and PTOs (5). The S. elongatus PTOs can be reconstructed in vitro by incubating a core clock protein, KaiC, with two other clock proteins, KaiA and KaiB, and adenosine triphosphate (ATP) (6). The rhythmic behaviors of the Kai oscillator have been confirmed in many functional and structural analyses, which have probed the ATP hydrolysis [adenosine triphosphatase (ATPase)] activity of KaiC (11, 12), autophosphorylation and autodephosphorylation activities of KaiC (6, 13, 14), conformational transitions of the proteins (12, 15, 16), and assembly or disassembly of Kai complexes (1720). Because the PTO period in S. elongatus is firmly correlated to the TTO period during the day (5, 6), the in vitro Kai oscillator should enable us to identify the mechanisms underlying the slowness of the circadian clock.

We searched the Kai oscillator for a minimal slow reaction whose efficiency correlated with in vivo TTO frequency (Fig. 1) (see supplementary materials and methods). The ATPase activity of full-length wild-type KaiC (KaiC-WT), consisting of the N-terminal C1 and C-terminal C2 domains, has been proposed as the basic timing cue (11, 12). We identified the steady-state ATPase activity of the C1 domain (C1-ATPase) as a suitable slow reaction. A truncated version of KaiC consisting solely of the N-terminal domain, KaiC1-WT, exhibited much slower ATP hydrolysis (11 ± 1 ATP per day per KaiC at 30°C) than well-known motor proteins (103 to 107 day−1) such as myosin, kinesin, and F1-ATPase (table S1). We confirmed the correlation of this activity with the in vivo TTO frequency using a series of period-modulating KaiC mutations in the C1 domain [S157→P157 (S157P), S157C, T42S, S48T] (21) (Fig. 1D), in which higher steady-state C1-ATPase in vitro resulted in higher-frequency TTOs in vivo. Thus, KaiC should experience a certain number of hydrolysis events per cycle in vivo, and the absolute rate of ATPase activity is connected to the cellular clock’s overall slowness (11, 22). Therefore, we examined the structural origin of the slow C1-ATPase and its coupling with TTOs via spatiotemporally distinct events, including the intramolecular KaiC ATPase and its phosphorylation cycles, as well as intermolecular interactions with KaiA and KaiB (Fig. 1).

Fig. 1 trans-hierarchical nature of the cyanobacterial circadian clock.

(A to J) Scatter-plot matrix showing all pairwise correlations among the frequency of the in vivo TTO cycle (10), frequency of the in vitro phosphorylation cycle of the Kai oscillator, undamped natural frequency (ωn) of KaiC alone, steady-state ATPase in the full-length KaiC-WT background, and steady-state C1-ATPase in the KaiC1-WT background. r, correlation coefficient.

To this end, we crystallized KaiC1-WT and five period-modulating variants in the pre- or posthydrolysis states, or both. All resultant crystals were in the P212121 space group (Fig. 2, A and B, fig. S1, and table S2). The prehydrolysis states (Fig. 2A, blue subunits) exhibited common features: assembly of six subunits into a hexamer and incorporation of one molecule of the slowly hydrolyzed ATP analog adenosine 5′-(γ-thiotriphosphate) (ATP-γ-S) into every subunit-subunit interface. ATP-γ-S existed in a complex with a Mg2+ ion (Mg-ATP-γ-S) in the ordinary octahedral-coordination geometry (fig. S2A). We observed the posthydrolysis state of the long-period variant KaiC1-S48T (Fig. 2B, orange and green subunits), which crystallized as an asymmetrically ATP- or adenosine diphosphate (ADP)–bound hexamer (fig. S2B). The ring-shaped hexamer was deformed asymmetrically due to steric constraints resulting from close juxtaposition of three types of subunits, creating both tight and loose intersubunit interfaces (Fig. 2B, fig. S3, and supplementary text).

Fig. 2 Reaction cycle of the C1-ATPase, a slow but stable ATPase that serves as the basic timing cue.

(A) Schematic overview of the crystal structure of ATP-γ-S–bound KaiC1-WT (P212121). The drawing in the box to the right of (B) summarizes the nomenclature used to describe the structures. For the seven other crystal structures, refer to details in fig. S1. (B) Schematic overview of the crystal structure of adenylyl imidodiphosphate (AMP-PNP), or ADP-bound KaiC1-S48T (P212121). (C) Zoomed-in views of the prehydrolysis A–B interface (blue dotted box) of Fig. 2A, posthydrolysis B–C interface (orange dotted box) of Fig. 2B, and posthydrolysis C–D interface (green dotted box) of Fig. 2B. The mesh indicates the FobsFcalc omit map of a potential lytic water molecule (W1) and another water molecule (W2), contoured at 4σ. (D) Schematic drawing of W1 positioning (filled circles) in KaiC1-WT, along with motor proteins that hydrolyze ATP efficiently (table S1). (E) Distances of a potential lytic water, the carbonyl oxygen atom of F199, and Nη of R226 from the putative near–in-line position during a 10-ns molecular dynamics simulation. The inset describes the crystal structure of the prehydrolysis state of KaiC1-WT, with van deer Waals radii depicting surfaces for the oxygen atom of W1, the carbonyl oxygen atom of F199, Nη of R226, and the γ-phosphate group. (F) Less- and more-unfavorable positioning of W1 confirmed in the P212121 (S146G, blue) and P3121 (WT, ocher) crystal structures, respectively. The N-terminal fray of the α7 helix at S157, followed by the repositioning of the α6 helix (Q153), caused the rearrangement of the E77-E78 pair, through which the γ-phosphate group was rotated by ~25° and W1 moved away from Pγ.

We identified two structural sources of slow ATPase activity. The first is the regulatory influence of the protein moiety on a lytic water molecule (W1) near the phosphorus atom (Pγ) of the γ-phosphate group of an ATP. In the prehydrolysis state (blue dotted box in Fig. 2C), W1 was sequestered (with a W1–Pγ distance of 3.8 to 3.9 Å and a W1–Pγ–O angle of 154° to 158°) by H-bonding to the F199 carbonyl oxygen, the nitrogen atom of R226 side chain (Nη), and another water molecule (W2). This position of W1 was much farther from a near–in-line configuration, with respect to the Pγ–O bond (3.0 Å, 180°), than analogous water molecules in other motor proteins that hydrolyze ATP efficiently—e.g., kinesin (3.3 Å, 164° to 167°) (23), myosin (2.3 Å, 170°) (24), and F1-ATPase (3.1 to 3.8 Å, 168° to 172°) (25) (Fig. 2D and table S1). Molecular dynamics simulation suggested that W1 was only marginally stabilized. Although a water molecule might occasionally migrate into and out of the W1 position, the near–in-line position was completely inaccessible due to steric hindrance from the F199 carbonyl oxygen and the R226 Nη (Fig. 2E and movies S1 to S3). Thus, the sequestration of W1 from the near–in-line position is one atomic-scale origin of slow hydrolysis. In another crystal form of KaiC1-WT (P3121 space group; ocher subunits in Fig. 2F and fig. S1A), W1 was even farther from the Pγ of the γ-phosphate group, which should be even more unfavorable to hydrolysis than the W1 position that we observed in the P212121 space group. In yet another crystal form, more or less unfavorable positioning of W1 occurred asymmetrically within the hexamer (fig. S1B). On the basis of nine structures (Fig. 2F, fig. S1, and supplementary text), we identified S157P and Q153A as mutations that indirectly stabilized the less unfavorable positioning of W1 (blue subunit), resulting in higher C1-ATPase activity (Fig. 1J). By contrast, S157C stabilized the more unfavorable positioning of W1 (ocher subunit) through N-terminal fray of α7 (fig. S4) and repositioning of α6, resulting in lower C1-ATPase activity (Fig. 1J).

The second origin of slowness involves coupling of ATP hydrolysis to slow cis-trans isomerization of the peptide chain. In the prehydrolysis state, the peptide bond between D145 and S146 (D145S146 peptide) mainly adopted the cis conformation (Fig. 3A). By contrast, in the posthydrolysis state, the D145S146 peptide was entirely in the trans conformation (Fig. 3A and fig. S5A). This trans selectivity was achieved through hydrolysis of ATP bound in the counterclockwise (CCW) interface (Fig. 2C, dotted boxes: blue → orange → green; see also supplementary text), along with repositioning of helices α6 and α7 on the clockwise (CW) side (Fig. 3A). These structural observations support the idea that hydrolysis of ATP bound in the CCW interface results in a conformational change that forces the D145S146 peptide to adopt the trans conformation.

Fig. 3 ATP hydrolysis coupled to the cis-trans isomerization of the D145S146 peptide.

(A) ATP hydrolysis in the CCW interface, followed by repositioning of the α6 and α7 helices, is coupled to cis-to-trans isomerization of the D145S146 peptide. The prehydrolysis states possessing the cis- D145S146 peptide (KaiC-S146P, dark blue) (fig. S5B) and the trans-D145S146 peptide (KaiC-S146G, light blue) (fig. S5C) are superimposed on the posthydrolysis state (S48T, orange). The location of the D145S146 peptide is highlighted by a dotted circle. The meshes represent the FobsFcalc omit maps contoured at 3σ. In the prehydrolysis state, the fraction of the cis-D145S146 peptide ranged from 0.5 to 0.8, depending on the subunit position and the crystal forms (Fig. 2 and fig. S1). (B) Potential of mean force (kilocalories per mole) for cis-trans isomerization of the D145S146 peptide as a function of the dihedral angle Cα(D145)–N(D145)–C(S146)–Cα(S146).

According to our computational results, the prehydrolysis state containing the cis-D145S146 peptide must overcome a barrier of 14 to 16 kcal mol−1 (Fig. 3B) for cis-trans isomerization and a barrier of 11 to 17 kcal mol−1 (table S1) to disrupt the Pγ–O bond. These two events are potentially related to each other through repositioning of helices α6 to α8 (fig. S5G and supplementary text). Given a frequency factor of ~1012 s−1, the rate of a reaction that must overcome a barrier of ~22 kcal mol−1 is ~0.5 hours−1 (~12 ATP day−1) at 30°C. The impact of cis-trans flipping was confirmed using the KaiC1-S146P mutant, in which the D145S146 peptide is forced to adopt the cis conformation (fig. S5B). Moreover, full-length KaiC-S146P exhibited a notable decrease in steady-state ATPase activity (4.8 ATP per day per KaiC), as well as a nearly 50% increase in the phosphorylation cycle period (34 hours). Thus, cis-trans flipping of the D145S146 peptide could impose a substantial energy barrier on the hydrolysis-coupled transition of the subunits.

In full-length KaiC, the slow C1-ATPase is integrated with another ATPase in the C2 domain (C2-ATPase), forming a coupled C1-C2-ATPase system. At steady state, the C1-ATPase predominated over the C2-ATPase (Fig. 1J). The pacemaking role of C1-ATPase became obvious during the approach to the steady state (pre–steady state). Immediately after the addition of excess ATP to full-length apo-KaiC protein (26), nascent KaiC-WT hydrolyzed 35 ATP per day per KaiC at 30°C (Fig. 4A and fig. S6A). The activity suddenly decreased to ~8 ATP per day per KaiC during the first 6 hours and then stably recovered to and remained at 13 ± 1 ATP per day per KaiC, the activity required for steady-state KaiC-WT (11). The biphasic-exponential phases reflected the pre–steady-state relaxation of the coupled C1-C2-ATPase (supplementary text). Minimal dependencies on ATP, ADP, and KaiC protein concentrations (fig. S6B) further supported the hydrolysis-related maturation of ATP- or ADP-bound KaiC-WT hexamer (supplementary text). The phosphorylation state of the C2 domain was only weakly associated with the slow process, as the phosphorylation-mimic KaiC-S431D/T432E exhibited detectable relaxation (fig. S6C).

Fig. 4 Circadian periodicity determined by KaiC ATPase activity.

(A) Damped oscillation observed for pre–steady-state relaxation of KaiC ATPase as the source of the circadian time scale at 30°C. In the presence of both KaiA and KaiB, the ATPase activity of KaiC-WT exhibits a stable oscillation around its steady-state ATPase activity. The bar at lower left indicates the maximal contribution from ATP synthesis during the first hour (27). (B) Dynamic responses of the ATPase activity observed for period mutants of full-length KaiC. Each curve is offset longitudinally for clarity of presentation. The values in parentheses represent the periods of in vitro phosphorylation rhythms. (C) Temperature dependency of pre–steady-state dynamics of ATPase activity in KaiC-WT. The insets represent Arrhenius plots. Errors in values are derived from a linear regression analysis. EA represents the apparent activation energy.

Response speed of the C1-C2-ATPase is controlled by the C1-ATPase. Long-period C1 mutants (T42S, S157C, S48T, and A251V) with suppressed C1-ATPase activity exhibited slower relaxations of the C1-C2-ATPase than KaiC-WT, whereas short-period C1 mutants (Q153A and S157P) with elevated C1-ATPase activity exhibited more rapid relaxation with a deeper undershoot (Fig. 4B). These observations suggested that steady-state C1-ATPase activity governs the speed of pre–steady-state relaxation of the C1-C2-ATPase and oscillation period in vitro (Fig. 1G) and in vivo (Fig. 1D). The undamped natural frequency (ωn) is another measure of the response speed of the coupled C1-C2-ATPase, as determined by fitting each transient curve of ATPase activity (Fig. 4B) to the initial-state response curve of the quasi–second-order system (supplementary text). Furthermore, the ωn value of KaiC-WT alone in vitro was 0.91 ± 0.01 day−1, matching the circadian oscillatory frequency. We confirmed fine correlations among steady-state C1-ATPase activity, ωn value, and in vivo frequency of TTOs for a series of the period-modulating C1 mutations (Fig. 1, B, D, and I). Thus, the slow C1-ATPase contributes to the circadian pacemaker for both the C1-C2-ATPase system and TTOs.

Slow relaxation of the C1-C2-ATPase was related to emergence of temperature compensation in full-length KaiC. Early ATPase activity (<2 hours in Fig. 4C) decayed in a temperature-dependent manner (thermal sensitivity Embedded Image) (inset of Fig. 4C), whereas steady-state activity after the later decay (>24 hours) was almost independent of temperature Embedded Image. Because the contribution of C2-ATPase was negligibly small at steady state (Fig. 1J), the activity that did not vary with temperature can be attributed to the C1-ATPase. Thus, steady-state C1-ATPase activity may be kept constant in each hexamer by adjusting the ratio of prehydrolysis blue and ocher subunits and by scrambling their spatial arrangement (Fig. 2A and fig. S1, A and B). In this respect, oscillation of ATPase activity (Fig. 4A) is realized by perturbing the intramolecular homeostasis of KaiC ATPase via intermolecular interactions with KaiA and KaiB.

The C1-C2-ATPase is coupled indirectly to interactions with KaiA and KaiB. Binding affinities of KaiA and KaiB for pre–steady-state KaiC-WT were modulated in a biphasic and parabolic manner on the same time scale as relaxation of the C1-C2-ATPase (fig. S7, A and B). However, such time evolution was lost in KaiC-S431D/T432E (fig. S7, A and B), which retained pre–steady-state relaxation of C1/C2-ATPase (fig. S6C). Thus, C1-C2-ATPase status is coupled to intermolecular interactions with KaiA and KaiB through a phosphorylation-dependent conformational change of KaiC. This idea is also supported by the fact that the C2-ATPase is essential for the complete autodephosphorylation (27). Thus, absolute slowness encoded in the C1-ATPase is transferred to and correlated with the TTO cycle (Fig. 1) through a pathway in which non–steady-state relaxation of C1-C2-ATPase leads to cycles of phosphorylation that alter intermolecular interactions with KaiA or KaiB or other clock-related proteins (28, 29).

In cyanobacteria, slow homeostatic regulation of the coupled C1-C2-ATPase is critical for circadian periodicity (Figs. 1 and 4A) and also has an important role in entrainment of individual KaiC molecules by external stimuli such as temperature changes (30). Our results suggest how ancient cyanobacteria have incorporated Earth’s rotation period into their molecular systems.

Supplementary Materials

www.sciencemag.org/content/349/6245/312/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S7

Tables S1 to S3

References (3151)

Movies S1 to S3

References and Notes

  1. Single-letter abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.
  2. Acknowledgments: Diffraction data were collected at BL44XU in the SPring-8 facility under the proposals 2009A6902, 2009B6902, 2010A6502, 2010B6502, 2011A6602, 2011B6602, 2012A6702, 2012B6702, and 2013B6700. MX225-HE was financially supported by Academia Sinica and National Synchrotron Radiation Research Center (Taiwan, China). This work was supported by Grants-in-Aid for Scientific Research (25291039, 22687010, and 26102544 to S.A.; 24000016 to T.K.; 26888019 to T.M.; and 25288011 to S.Saito); the Platform for Drug Discovery, Informatics, and Structural Life Science from the Ministry of Education, Culture, Sports, Science and Technology of Japan; and partly by the Okazaki ORION project. The calculations were in part performed at the Research Center for Computational Science in Okazaki, Japan. S.A. designed the experiments. A.M., J.A., J.W., and M.O. conducted ATPase experiments. S.Son screened initial crystallization conditions, and J.A. and T.B.H. further refined them. S.Son, J.A., T.B.H., and S.A. collected diffraction data and analyzed the structures with inputs from E.Y. T.M. and S.Saito conducted molecular dynamics simulations. S.A., J.A., T.B.H., A.M., T.M., S.Saito, and T.K. drafted the manuscript with input from all authors. We declare no conflicts of interest. The atomic coordinates and structure factors are deposited in the Protein Data Bank with accession codes 4TL8 (KaiC1-WT, P212121), 4TL9 (ATPγS-bound KaiC1-S48T), 4TLA (ADP-bound KaiC1-S48T), 4TLB (KaiC1-S146P), 4TLC (KaiC1-S146G), 4TLD (KaiC1-S157P), 4TLE (KaiC1-S157C), 4TL7 (KaiC1-WT, P3121), and 4TL6 (KaiC1-WT, C2).
View Abstract

Stay Connected to Science

Navigate This Article