Assembly and Disassembly Dynamics of the Cyanobacterial Periodosome

Transcription

1 Article Assembly and Disassembly Dynamics of the Cyanobacterial Periodosome Shuji Akiyama, 1,2, * Atsushi Nohara, 4 Kazuki Ito, 2 and Yuichiro Maéda 3,5 1 PRESTO, Japan Science and Technology Agency, Honcho, Kawaguchi, Saitama , Japan 2 RIKEN SPring-8 Center, Harima Institute 3 ERATO Actin Filament Dynamics Project, Japan Science and Technology Agency, c/o RIKEN, Harima SPring-8 Center Kouto, Sayo, Hyogo , Japan 4 Division of Biological Science, Graduate School of Science, Nagoya University, Furo-cho, Chikusa-ku, Nagoya , Japan 5 Structural Biology Research Center, Graduate School of Science, Nagoya University, Furo-cho, Chikusa-ku, Nagoya , Japan *Correspondence: akishu@spring8.or.jp DOI /j.molcel SUMMARY In vitro incubation of three Kai proteins, KaiA, KaiB, and KaiC, with ATP induces a KaiC phosphorylation cycle that is a potential circadian clock pacemaker in cyanobacterium Synechococcus elongatus PCC The Kai proteins assemble into large heteromultimeric complexes (periodosome) to effect a robust oscillation of KaiC phosphorylation. Here, we report real-time measurements of the assembly/disassembly dynamics of the Kai periodosome by using smallangle X-ray scattering and determination of the lowresolution shapes of the KaiA:KaiC and KaiB:KaiC complexes. Most previously identified period-affecting mutations could be mapped to the association interfaces of our complex models. Our results suggest that the assembly/disassembly processes are crucial for phase entrainment in the early synchronizing stage but are passively driven by the phosphorylation status of KaiC in the late oscillatory stage. The Kai periodosome is assembled in such a way that KaiA and KaiB are recruited to a C-terminal region of KaiC in a phosphorylation-dependent manner. INTRODUCTION Circadian clocks are endogenous timing systems that have evolved in a variety of living organisms, enabling them to adjust their metabolic activities to day/night-alternating environments (Pittendrigh, 1993). Cyanobacteria, the simplest organisms known to possess a circadian oscillator, exhibit daily regulation of nitrogen fixation, photosynthesis, and amino acid uptake (Golden et al., 1997). The essential elements of the cyanobacterial oscillator are the three clock proteins termed KaiA, KaiB, and KaiC, all of which are encoded in the kai locus of the cyanobacterium Synechococcus elongatus PCC 7942 (S. elongatus) (Ishiura et al., 1998). The previous observation that KaiA regulates kaibc transcription positively, whereas KaiC does so negatively, has suggested a feedback regulation of clock gene expressions as a central timing process (Ishiura et al., 1998). However, Tomita et al. (2005) demonstrated in vivo oscillations of the phosphorylation level of KaiC under continuous dark conditions with minimal activities of transcription and translation. Furthermore, Nakajima et al. (2005) succeeded in reconstructing the KaiC phosphorylation cycle in vitro solely by incubating KaiA, KaiB, and KaiC in the presence of ATP. These results suggest that circadian timing mechanisms in cyanobacteria persist even without the transcription/ translation feedbacks proposed for various eukaryotic systems (Bell-Pedersen et al., 2005) and that the phosphorylation cycle of KaiC is the central pacemaker of the cyanobacterial clock. Considerable progress has been made in understanding the molecular mechanisms bywhich KaiC is phosphorylated ina circadian manner. KaiA enhances the autophosphorylation of Ser431 and Thr432 in KaiC (Iwasaki et al., 2002; Nishiwaki et al., 2004), whereas KaiB attenuates the effects of KaiA (Kitayama et al., 2003; Xu et al., 2003). Yeast two-hybrid analysis suggested the associations of three Kai proteins in various combinations (Iwasaki et al., 1999). Gel filtration and pull-down analyses indicated that the Kai proteins are repeatedly assembled and disassembled to form heteromultimeric Kai complexes (periodosome) (Kageyama et al., 2003, 2006). These results point to a potential relationship between the assembly/disassembly dynamics of the Kai proteins and the circadian phosphorylation of KaiC (Golden, 2004). In order to elucidate the assembly/disassembly dynamics of the Kai periodosome, extensive investigations have focused on the structure of the Kai proteins. To date, the crystal structure of each Kai protein has been determined independently. S. elongatus KaiA forms a homodimer (KaiA 2 ), the subunit of which is composed of an N-terminal pseudo-receiver domain and a C-terminal KaiC-interacting domain (Ye et al., 2004). The C-terminal domains of KaiA 2 from Thermosynechococcus elongatus BP- 1(T. elongatus) were determined independently by X-ray crystallography (Uzumaki et al., 2004) and NMR (Vakonakis et al., 2004). Crystallographic studies reported the KaiB from Anabaena PCC7120 to exist in a homodimeric conformation (KaiB 2 )(Garces et al., 2004) and the KaiB from Synechocystis PCC6803 (Hitomi et al., 2005) and T. elongatus (Iwase et al., 2005) to exist as a homotetrameric form (KaiB 4 ). The hexagonal pot-shaped structure of T. elongatus KaiC was first determined by negative-stain electron microscopy (Hayashi et al., 2003). The high-resolution crystal structures of S. elongatus KaiC indicated the formation of a double-doughnut-shaped homohexamer (KaiC 6 ), incorporation Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc. 703

2 of 2 ATP molecules into every protomer-protomer interface (total of 12 ATPs), and dual phosphorylation sites in the C2 domain of KaiC (Pattanayek et al., 2004; Xu et al., 2004). Despite this progress in structural characterization, the relationship between the assembly/disassembly dynamics and the circadian phosphorylation of KaiC is still poorly understood, mainly because of the difficulty in unraveling the underlying mechanisms solely from the static molecular pictures of individual clock components. It is essential to characterize the Kai complexes along a reaction coordinate as challenged in several studies (Pattanayek et al., 2006; Taniguchi et al., 2001; Vakonakis and LiWang, 2004). However, a blind search for crystallization conditions of oscillatory transient complexes may not be a promising strategy. We thus followed the assembly/disassembly dynamics of the S. elongatus Kai periodosome in real time by using small-angle X-ray scattering (SAXS). Four major advances are described in this report. The first is a discovery of a p/2 phase delay in the assembly/disassembly dynamics relative to the KaiC phosphorylation cycle. The second is a structural visualization of the Kai complexes with SAXS titration experiments. The third is an oscillatory model refined against both the KaiC phosphorylation cycle and the averaged molecular weight cycle determined by SAXS. Finally, based on the current and previous observations, the fourth is that we address a question, which comes first, assembly/ disassembly dynamics or the KaiC 6 phosphorylation cycle? RESULTS AND DISCUSSION p/2 Phase Delay of Assembly/Disassembly Dynamics In order to elucidate the assembly/disassembly dynamics of Kai proteins, an in vitro phosphorylation cycle was reconstructed by incubating a ternary mixture containing KaiA 2 (0.15 mg/ml), KaiB 4 (0.15 mg/ml), and KaiC 6 (0.60 mg/ml) and was followed with realtime SAXS measurements (Supplemental Experimental Procedures available online). Figure 1A indicates the time-dependent change of the forward scattering intensity, I(0), which serves as a sensitive measure of weight-averaged molecular weight. Because the I(0) of free KaiA 2, KaiB 4, and KaiC 6 was 460, 236, and 2200, respectively, the apparent forward scattering, I(0) app, is 1580 if all the Kai components remain free: (0.15/ 0.90) (0.15/0.90) (0.60/0.90) = Upon mixing the Kai proteins, I(0) app rapidly increased to 1900 within the experimental dead time (2 min). Subsequently, I(0) app revealed a gradual increase from 1900 to 3100 and peaked at an incubation time (IT) of 9 hr. After a monotonous decrease in I(0) app (from IT 9 to 24 hr), we observed a robust oscillation of I(0) app with a period of 24.4 ± 0.2 hr. The oscillation of I(0) app was in phase with the radius of gyration, R g, which oscillated with a period of 24.2 ± 1.3 hr (Figure 1B). These results indicate the accumulations of large Kai complexes (>470 kda) at ITs 38 and 62 hr and of small complexes at ITs 26 and 50 hr. The temporal pattern common to I(0) app and R g app comprised two distinct stages: an early synchronizing stage (from IT 0 to 18 hr) and a stable oscillatory stage (from IT 18 hr). Interestingly, both I(0) app and R g app were phase delayed by 6.6 ± 0.9 hr relative to the phosphorylated KaiC 6 fraction (Phos app ), whereas the periods of the SAXS parameters were nearly identical to that of Phos app (23.8 ± 0.5 hr). As a consequence, the large Kai complexes accumulate during the dephosphorylation step from IT 30 to 42 hr (Figures 1A and 1C), and the small Kai complexes accumulate in the phosphorylation step from IT 42 to 54 hr. These real-time data clearly demonstrate that the assembly/disassembly dynamics of the Kai complexes are not under steady-state conditions but are evidently coupled to the phosphorylation/ dephosphorylation processes of KaiC 6 with a phase delay of approximately p/2: 2 p 3 6.6/24.4 = (1.08 ± 0.15) 3 p/2. The presence of the p/2 phase delay was further confirmed by using a short-period KaiC 6 mutant (KaiC 6 S157P ) carrying S157P substitution (Nakajima et al., 2005). As shown in Figure 1D, a ternary mixture containing KaiA 2, KaiB 4, and KaiC 6 S157P revealed the robust oscillations of both I(0) app and Phos app with a short period of 22 hr. Phos app of KaiC 6 S157P peaked at approximately IT 27 hr (3 hr earlier than that of wild-type KaiC 6 ), and at that very moment, the increasing rate of I(0) app of KaiC 6 S157P was maximized. This observation suggests that the p/2 phase delay between I(0) app and Phos app was retained also in KaiC 6 S157P.Itmustbe noted that the robustness of the Phos app oscillation decreased at approximately IT 48 hr, possibly because a higher activity of KaiC 6 S157P than wild-type KaiC 6 resulted in an exhaustion of ATP in the system. Consistently to this interpretation, the Phos app oscillation of KaiC 6 S157P was rescued by adding a small aliquot of a concentrated ATP solution to the system at IT 54 hr (black arrows in Figure 1D). In contrast to a quick response of Phos app to the supply of ATP, I(0) app of KaiC 6 S157P remained unaffected for a certain period (5 6 hr) and then regained a robust oscillation so that I(0) app was phase delayed by p/2 relative to Phos app. These results demonstrate that the p/2 phase delay between I(0) app and Phos app is an essential character of the in vitro Kai oscillator and that the circadian rhythm of Phos app is closely related to the time derivatives of I(0) app (a relationship between reactant and product). In order to investigate the origin of the phase delay, we next conducted the SAXS measurements on binary mixtures. In the binary mixture of KaiA 2 with KaiC 6, both I(0) app and R g app exhibited a vary fast increase (k vf app ) within the experimental dead time and reached a constant level of 2100 and 48 Å, respectively (Figures 1A and 1B). Phos app was initially 0.7 and exhibited a fast increase up to 0.9 (k f app = 2.7 ± 1.9 hr 1 ). These observations are qualitatively explained by a very fast complexation between KaiA 2 and KaiC 6, followed by a fast phosphorylation of the associated KaiC 6. On the other hand, the I(0) app for the binary mixture of KaiB 4 with KaiC 6 obeyed a biphasic relaxation comprised of a slow increase (k s app = 0.18 ± 0.06 hr 1 ) and a very slow decrease (k vs app = 0.13 ± 0.05 h 1 ). The gradual decrease in Phos app was monophasic (0.13 ± 0.03 hr 1 ) and coupled to the very slow phase of I(0) app. These results suggest a slow assembly of KaiB 4 and KaiC 6 followed by a very slow dephosphorylation coupled to a dissociation of KaiB 4 from KaiC 6. The phase delay in the synchronizing stage is explained by the above kinetic parameters and models for the binary mixtures. Upon preparing a ternary mixture, KaiA 2 first binds KaiC 6 in the very fast phase (1/k vf app < 0.03 hr, in Figure 1A). The phosphorylation of KaiC 6 is nearly completed within the fast phase (1/k f app = 0.4 hr, Figure 1C) before notable association of KaiB 4 with KaiC 6. KaiB 4 subsequently binds KaiC 6 in the slow phase (1/k s app = 5.6 hr, Figure 1A), and then the dephosphorylation of KaiC 6 advances in the very slow phase (1/k vs app = 7.7 hr, 704 Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc.

3 Figure 1. Time Courses of SAXS Parameters and the Phosphorylated Fraction of KaiC 6 Red circles, a ternary mixture of KaiA 2 (0.15 mg/ml), KaiB 4 (0.15 mg/ml), and KaiC 6 (0.60 mg/ml); blue squares, a binary mixture of KaiA 2 (0.15 mg/ml) with KaiC 6 (0.60 mg/ml); and green triangles, a binary mixture of KaiB 4 (0.15 mg/ml) with KaiC 6 (0.60 mg/ml). (A) Time-dependent changes of apparent forward scattering intensity, I(0) app. Red line, I(0) app = sin{2p (t 7.3) / 24.4}; green line, I(0) app = exp( t) exp( t). (B) Time-dependent changes of apparent radius of gyration, R app g. The red line represents a fitting with R app g = sin{2p (t 7.3) / 24.2}. (C) Time-dependent changes in the phosphorylated fraction of KaiC 6 (Phos app ). Filled green triangles represent the time course of Phos app in the absence of both KaiA 2 and KaiB 4. The red line, Phos app = sin{2p (t 0.67) / 23.8}; the blue line, Phos app = exp( t); and the green line, Phos app = exp( t). (D) Retention of the p/2 phase delay in a short-period KaiC 6 mutant (KaiC S157P 6 ) carrying S157P substitution. A small aliquot of a concentrated ATP solution was added to the system at IT 54 hr (black arrows). Purple circles and black squares represent I(0) app (left axis) and Phos app (right axis), respectively, of a ternary mixture containing KaiA 2 (0.15 mg/ml), KaiB 4 (0.15 mg/ml), and KaiC S157P 6 (0.60 mg/ml). For comparison, I(0) app and Phos app of the wild-type ternary mixture are presented as purple and black dashed lines, respectively. Error bars were generated from three separate experiments. Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc. 705

4 Figure 2. Titration Experiments of Binary Complexes with SAXS (A) Phosphorylated fraction of KaiC 6 (Phos app ) in the presence of KaiA 2 (squares) or KaiB 4 (circles). (B) Plots of raw forward scattering intensity, ci(0) app, against total concentration of KaiA 2 (squares) or KaiB 4 (circles). The solid lines correspond to the leastsquares fitting to the data using a 1:1 binding scheme (see details in text). The resulting parameters are as follows: Kð0Þ KaiA2 = 30: , Kð0Þ KaiB4 = 10: ,Kð0Þ KaiC6 = 767:5310 6,Kð0Þ KaiA2:KaiC6 = , and Kð0Þ KaiB4:KaiC6 = 935: Error bars were generated from five separate measurements. (C) SAXS curves of KaiA 2 (red dots), P KaiC 6 (blue dots), and the KaiA 2 : P KaiC 6 complex (green dots). The red, blue, and green lines represent theoretical curves of low-resolution models of KaiA 2, P KaiC 6, and the KaiA 2 : P KaiC 6 complex, respectively, shown in Figure 3A. (D) SAXS curves of KaiB 4 (red dots), NP KaiC 6 (blue dots), and the KaiB 4 : NP KaiC 6 complex (green dots). The red, blue, and green lines represent theoretical curves of the low-resolution models of KaiB 4, NP KaiC 6, and the KaiB 4 : NP KaiC 6 complex, respectively, shown in Figure 3B. The black line corresponds to a theoretical SAXS curve calculated from the crystal structure of KaiB 4 (Hitomi et al., 2005) with the program CRYSOL (Svergun et al., 1995). Figure 1C). Therefore, I(0) app in the synchronizing stage of the ternary mixture can be approximated to the summation of a sudden rise in flat baseline upon KaiA 2 /KaiC 6 interaction and the biphasic fluctuation peaking at approximately IT 8 hr upon KaiB 4 /KaiC 6 interaction (Figure 1A). At the same time, because the fast KaiC 6 phosphorylation (k f app ) by KaiA is kinetically uncoupled from the slow KaiC 6 dephosphorylation (k vs app ), Phos app in the synchronizing stage can be regarded as a summation of these two decays with opposing signs of amplitudes that initially peaks at approximately IT 1 hr (Figure 1C). The difference in the initial peaking time between the two parameters was 7 hr and roughly matched the experimental observations (6.6 ± 0.9 hr). The phase delay by p/2 suggests kinetically favored KaiA 2 /KaiC 6 interactions followed by thermodynamically preferred KaiB 4 /KaiC 6 interactions. A Branched Pathway: Ternary Complexes or Competitive Binding to KaiC 6 To evaluate the thermodynamic properties of binary interactions, a fixed amount of KaiC 6 was titrated with either KaiA 2 or KaiB 4. The KaiC 6 solutions titrated with KaiA 2 and KaiB 4 were fully equilibrated to achieve maximal phosphorylation and dephosphorylation, respectively (Figure 2A) and then subjected to the SAXS measurements. The changes in the raw forward scattering intensity, ci(0) app, are plotted in Figure 2B without normalizing the particle concentration (c, mg/ml). ci(0) app increased gradually with increasing KaiA 2 concentration and displayed smooth saturation. In contrast, a sharp break in the titration curve was confirmed at a KaiB 4 concentration of 4 mm, which was nearly matched to the one equivalent molar concentration of KaiC 6 (3.84 mm). Each titration curve can be fitted uniquely to a theoretical curve for a 1:1 binding scheme, in which one KaiA 2 (or KaiB 4 ) molecule binds one KaiC 6 molecule to form a KaiA 2 : P KaiC 6 (or KaiB 4 : NP KaiC 6 ) complex. The dissociation constant (K D ) of the KaiA 2 : P KaiC 6 complex was estimated to be 4.7 ± 0.7 mm and 40-fold larger than that of the KaiB 4 : NP KaiC 6 complex (0.12 ± 0.08 mm). The energetic preference for the KaiB 4 /KaiC 6 interactions further supports the occurrence of a p/2 phase delay during the synchronizing stage. 706 Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc.

5 In order to gain structural insights into the binary interactions, SAXS curves of the binary complexes were deconvoluted from the titration data sets (Figures 2C and 2D and Supplemental Experimental Procedure) and subjected to ab initio shape reconstructions (Svergun, 1999). Figure 3A illustrates a low-resolution model of the KaiA 2 : P KaiC 6 complex. The overall size of the KaiA 2 : P KaiC 6 complex measured nm, and its R g and maximum dimension (D max ) were 53.2 ± 0.1 and 165 Å, respectively. The envelope of P KaiC 6 occupied 82% of the total volume and was composed of a globular head portion and a short tail. As shown in Figure S1A, a cavity located inside the head portion is remarkably consistent with the pot-shaped structure of KaiC 6 observed under the electron microscope (Hayashi et al., 2003). By virtue of the tail protrusion from the head, a recently determined X-ray crystal structure of KaiC 6 (Pattanayek et al., 2006) was uniquely superimposed on the low-resolution shape of P KaiC 6. The remaining portion of the envelope occupying 18% of the total volume, into which the X-ray crystal structure of KaiA 2 could be superimposed, was clustered near the tail. The structure of the KaiA 2 : P KaiC 6 complex clearly demonstrates that KaiA 2 interacts with the C2 domain and the tail of P KaiC 6. Pattanayek et al. (2006) postulated two models for T. elongatus KaiA 2 /KaiC 6 interactions based on their results from negative-stain electron microscopy. One is the tethered model in which KaiA 2, located 35 Å away from the hexameric barrel, binds the tail portion of KaiC 6. The other is the engaged model in which KaiA 2 directly binds the C2 domain of the hexameric barrel. In terms of the relative orientation of KaiA 2 and KaiC 6, our model may be more similar to the tethered model than to the engaged model. However, the KaiA 2 in our structure locates much closer to the hexameric barrel than depicted in the tethered model, thereby achieving a tighter interaction through a broad association interface ranged from the C2 domain to the tail of KaiC 6. This broad interface for KaiA 2 /KaiC 6 interaction provides possible explanations for a number of rhythmic mutants exhibiting a wide range of period length and arrhythmia in vivo. As shown in Figures 3C and 3D, 17 out of the 29 previously identified period-modulating KaiA 2 mutations (Nishimura et al., 2002; Uzumaki et al., 2004) mapped to the association interface between KaiA 2 and P KaiC 6 in our model. Furthermore, the tail portion mainly constituted by the C-terminal 30 residues of KaiC 6, whose deletion abolishes circadian rhythmicity of bioluminescence (Pattanayek et al., 2006), was involved in extensive contacts with KaiA 2 through the interface (yellow ribbon, Figure 3C). A specific interaction between the C-terminal domain of T. elongates KaiA 2 and a peptide mimicking the C terminus of KaiC was also confirmed in a previous NMR study (Vakonakis and LiWang, 2004). Thus, the SAXS-based model for the KaiA 2 : P KaiC 6 complex is essentially consistent with the suggestions from previous biochemical and structural studies. The observed discrepancy between our model and the tethered model can be partly attributed to the difference in the sample sources between our S. elongatus and their T. elongatus and may partly arise from the difference in the phosphorylation status of KaiC 6. The restored shape of the KaiB 4 : NP KaiC 6 complex measured nm (Figure 3B) and was slightly more compact than the KaiA 2 : P KaiC 6 complex (R g = 49.1 ± 0.1 Å,D max = 150 Å). We confirmed the presence of a short tail protruding from a globular head containing an internal cavity (Figure S1B), onto which the X-ray crystal structure of KaiC 6 was superimposed. The residual portion occupying 13% of the total volume was similar in shape to the X-ray crystal structure of KaiB 4 (Hitomi et al., 2005; Iwase et al., 2005) (Figure 3B). The homotetrameric architecture of S. elongatus KaiB was confirmed by both analytical ultracentrifugation (data not shown) and SAXS measurements (Figure 2D). The superimposition of each high-resolution structure indicated a potential correlation between period-altering mutations in KaiB 4 and the KaiB 4 /KaiC 6 interaction. To date, six mutations in KaiB 4 have been demonstrated to modulate the in vivo bioluminescence rhythm (Ishiura et al., 1998; Iwase et al., 2005). Interestingly, four out of six critical residues were located in the association interface of the KaiB 4 : NP KaiC 6 model (Figures 3E and 3F). Thus, our results, together with the previous biochemical results, support the assumption of a specific interaction between KaiB 4 and the C2 domain of NP KaiC 6. The reconstructed models suggest two possible scenarios for the docking of KaiB 4 onto the KaiA 2 : P KaiC 6 complex (n = 1, Figure 4A). If the KaiC 6 portions of two binary complexes are superimposed on each other (Figure 3G), the orientation of the bound KaiA 2 relative to KaiC 6 is similar to that of bound KaiB 4.Thisstructural feature points to one scenario, in which KaiB 4 competes with KaiA 2 for a common docking site of P KaiC 6 (step a in Figure 4A). Considering the greater affinity of KaiB 4 for P KaiC 6 than for NP KaiC 6 (Figure 1A and ref Kageyama et al., 2006), the K D value should follow the rank order KaiA 2 : P KaiC 6 >KaiB 4 : NP KaiC 6 > KaiB 4 : P KaiC 6. Therefore, KaiB 4 can displace the KaiA 2 associated with the C2 domain of P KaiC 6 and thereby inhibit the phosphorylation activity of KaiA 2. A previous crystallographic study also raised the possibility of competition based on the presence of a molecular surface common to KaiA 2 and KaiB 4 (Garces et al., 2004). The other scenario assumes the formation of a ternary complex (step b in Figure 4A). The in vitro accumulation of the ternary complexes at approximately IT 32 hr has been confirmed by several biochemical techniques (Kageyama et al., 2006). Because our titration experiments revealed extremely weak interactions between KaiA 2 and KaiB 4, a ternary complex without any direct interactions between KaiB 4 and P KaiC 6 is unlikely. As hypothetically illustrated in Figures 3H and 3I, KaiB 4 is possibly recruited to the unoccupied C2 domains of the KaiA 2 : P KaiC 6 complex and is likely to bind P KaiC 6 in close contact with bound KaiA 2, thereby inhibiting its phosphorylation activity. Irrespective of the docking process, the reaction is then switched from the phosphorylation to the dephosphorylation step (steps c and d in, Figure 4A). Both scenarios indicate that KaiB 4 attacks KaiA 2 to inhibit the KaiA 2 activity rather than to promote the autodephosphorylation of KaiC 6 and are consistent with the fact that the autodephosphorylation rate of KaiC 6 is insensitive to the presence or absence of KaiB 4 (0.11 hr 1, Figure 1C). Our low-resolution models suggest a branching from the KaiA 2 : P KaiC 6 complex to the competition or ternary complex, providing a reasonable explanation for the earlier accumulation of the KaiB 4 : KaiC 6 complex than the ternary complex (Kageyama et al., 2006). Hybrid Simulation of Oscillatory Dynamics An effective approach to elucidate oscillatory mechanisms is to refine a feasible reaction scheme against multiple experimental Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc. 707

6 708 Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc.

7 parameters. Kageyama et al. (2006) conducted a two-step pulldown assay of the ternary mixture and estimated the oscillatory changes in the fractions of associated and free KaiC 6. Based on the concentration profiles they estimated, we attempted to construct a model reproducing the simultaneous circadian oscillations of Phos app and I(0) app. The simplest, model I, comprises three first-order complexes: KaiA 2 :KaiC 6, KaiB 4 :KaiC 6, and KaiA 2 :KaiB 4 :KaiC 6 (n = 1, Figure 4A). We first assumed that each KaiC 6 fraction of model I (upper panel in Figure 5A) oscillates in the same manner as reported in the two-step pull-down assay (Kageyama et al., 2006) and calculated the concentration profile of each Kai protein (lower panel in Figure 5A). Because the I(0) values of the respective species were known, the forward scattering intensity of model I, I(0) Model I, was calculated by summing the contribution from each component. At the same time, the phosphorylated KaiC 6 fraction of model I, Phos Model I, was calculated by assuming the phosphorylated fractions of KaiA 2 :KaiC 6 and KaiB 4 :KaiC 6 to be 0.90 and 0.05, respectively. Note that these two values are based on the experimental estimates from the binary mixtures at equilibrium (Figure 1C). We then refined the phosphorylated fraction of the KaiA 2 :KaiB 4 :KaiC 6 complex so as to minimize the discrepancy, {I(0) app I(0) Model I } 2 + {Phos app Phos Model I } 2. As shown in Figures 4B and 4C, however, it was not possible to reproduce the robust oscillations of Phos app and I(0) app simultaneously, irrespective of the phosphorylated fraction of the KaiA 2 :KaiB 4 :KaiC 6 complex. The I(0) Model I remarkably smaller than I(0) app suggests the underestimation of the accumulated amount and/or molecular weight of the complexes. We next examined the populations of associated KaiC 6 in model I by tuning the amplitude and baseline of each concentration profile while retaining respective original phases (model II, Figure 5B). In model II, the concentrations of the ternary and binary complexes were greatly enhanced up to nearly maximum levels at the expense of free KaiC 6. Even under this extreme condition, I(0) Model II was significantly smaller than I(0) app (Figure 4B), indicating an involvement of higher-order Kai complexes. We thus attempted to improve model I by increasing the association order (n = 2 4, Figure 4A) while retaining the original fractional profiles of model I. As shown in Figure 4B, the rhythmic change of I(0) app was reproduced solely by model III comprising the fourth-order complexes (KaiA 2 ) 4 :KaiC 6, (KaiB 4 ) 4 :KaiC 6, and (KaiA 2 ) 4 :(KaiB 4 ) 4 :KaiC 6. However, model III failed to reproduce the robustness of Phos app simultaneously, irrespective of the phosphorylated fractions of the fourth-order complexes (Figure 4C). This is mainly because the oscillation of each complex concentration in model III is as faint as that in model I, although the populations of free KaiA 2 and KaiB 4 in model III are much smaller than those in model I (Figures 5A and 5C). A lesson from these two examinations is the necessity of improving model I in terms of both the associated KaiC 6 fractions and association order of the complexes. Simultaneous improvements of model I resulted in model IV (Figure 5D), in which the Kai proteins are repeatedly assembled and disassembled to form the second-order Kai complexes (n = 2, Figure 4A) in a more robust manner. As shown in Figures 4B and 4C, model IV could reproduce the robust oscillations of p/2-delayed I(0) app and Phos app simultaneously. The phosphorylated fraction of the ternary complexes was estimated to be 0.4, and the optimized concentration profiles of model IV are displayed in Figure 5D. In agreement with a previous gel filtration experiment (Kageyama et al., 2006), the concentration of free KaiB 4 oscillated robustly so that its ratio at IT 36 hr to that at IT 24 hr was 0.5, and the ternary complexes accumulated at approximately IT 36 hr. The inclusion of the large associates is further supported by the transient formation of a higher-order complex between KaiB 4 and KaiC 6 at IT 8 hr (Figure 1A) and also by the previous gel filtration results demonstrating a higher-order complex between KaiA 2 and KaiC 6 at IT 24 hr (Kageyama et al., 2006). Although Phos app and/or I(0) app might also be explained by a model including much higher-order complexes, the formation of a ternary complex larger than (KaiA 2 ) 2 :(KaiB 4 ) 2 :KaiC 6 is unlikely due to steric crowding. According to the low-resolution models of the binary complexes (Figure 3), the C2 domains of KaiC 6 will be hindered upon the binding of two KaiA 2 and two KaiB 4 proteins, and there will remain no space for further recruitment of KaiA 2 or KaiB 4. Thus, the current observations for the ternary mixture are best described by model IV (Figure 5D), calling for a refinement of the previous theoretical models that assumed only first-order Kai complexes Figure 3. Low-Resolution Models of Binary Complexes Restored from SAXS Data Smooth envelopes for the bead models of KaiA 2 (orange), KaiB 4 (pink), P KaiC 6 (cyan), and NP KaiC 6 (lime) were calculated by using the SITUS package (Wriggers and Chacon, 2001). Red and purple ribbons correspond to superimposed crystal structures of KaiA 2 (Ye et al., 2004) and KaiB 4 (Hitomi et al., 2005; Iwase et al., 2005), respectively. The crystal structure of KaiC 6 (Pattanayek et al., 2006) superimposed on the envelopes of P KaiC 6 and NP KaiC 6 is presented as blue and green ribbons, respectively. (A) Orthogonal views of the low-resolution envelope of the KaiA 2 : P KaiC 6 complex. (B) Orthogonal views of the low-resolution envelope of the KaiB 4 : NP KaiC 6 complex. (C) A potential binding interface between KaiA 2 and P KaiC 6. For clarity, only the C2 domains (blue ribbon) and the C terminus 30 residues of KaiC 6 (yellow ribbon) are displayed. (D) Period-modulating KaiA 2 mutations mapped onto its association interface with P KaiC 6. Seventeen out of twenty-nine mutations (Nishimura et al., 2002; Uzumaki et al., 2004) are highlighted in the CPK mode. (E) A potential binding interface between KaiB 4 and NP KaiC 6. For clarity, only the C2 domains and tails of KaiC 6 are illustrated (green ribbon). (F) Period-altering KaiB 4 mutations mapped onto its association interface with NP KaiC 6. Four out of six critical mutations (Ishiura et al., 1998; Iwase et al., 2005) are emphasized by the CPK mode and residue labeling. (G) Competitive binding to KaiC 6. The molecular shape of the KaiA 2 : P KaiC 6 complex is superimposed on the KaiB 4 : NP KaiC 6 complex. KaiB 4 (purple) sterically collides with KaiA 2 (orange). (H) A hypothetical ternary complex assuming KaiB 4 docking in close contact with the N-terminal pseudo-receiver domain of bound KaiA 2. (I) A putative ternary complex assuming KaiB 4 docking in close contact with the C-terminal domain of bound KaiA 2. All graphical presentations were prepared with VMD (Humphrey et al., 1996). Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc. 709

8 Figure 4. In Vitro Assembly/Disassembly Dynamics Revealed by Hybrid Simulation of SAXS and Phosphorylation Parameters (A) Feasible reaction scheme for in vitro oscillation of the Kai proteins. Models I and II, n = 1; model III, n = 4; and model IV, n = 2 (see details in the text). (B and C) Hybrid simulation of raw forward scattering intensity, ci(0) app, and phosphorylated KaiC 6 fraction, Phos app. Error bars were generated from three separate experiments. Broken, dotted, thin, and thick lines correspond to least-squares fits using models I, II, III, and IV, respectively, shown in Figure 5. (Mehra et al., 2006; Miyoshi et al., 2007; Takigawa-Imamura and Mochizuki, 2006). It is important to note that a variety of phosphorylation patterns within a single KaiC hexamer (interhexamer diversity) are not considered in our hybrid simulation. Except for fully phosphorylated and dephosphorylated cases, there could be various KaiC 6 shaving an identical degree of phosphorylation per one KaiC 6, but each of which possesses a different spatial arrangement of the phosphorylation sites, i.e., diverse arrangements of the KaiC monomers and/or phosphorylated locations within the KaiC monomer. In fact, a difference between the equilibrium and dynamic processes is likely to support the interhexamer diversity of the phosphorylation. The hybrid simulation of the dynamic oscillatory process points to the significant accumulations of the second-order complexes in the ternary mixture (n = 2, Figure 4A). This is in contrast to the equilibrium titration experiments suggesting the first-order complexes under the maximally phosphorylated or dephosphorylated status (Figures 2A and 2B). This difference might be a sign of various KaiC 6 s, each of which binds KaiA 2 and/or KaiB 4 with a different affinity/stoichiometry dependent on the interhexamer diversity of the phosphorylation. Detailed experiments and simulations dealing with the interhexamer diversity will be essential to build a perfect model explaining both the equilibrium and dynamic SAXS data. Viscosity-Compensated Period and Viscosity-Dependent Phase The in vitro oscillation of KaiC 6 phosphorylation is closely related to that in vivo (Nakajima et al., 2005). This observation is interesting 710 Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc.

9 Figure 5. Four Different In Vitro Models Evaluated by Hybrid Simulation Upper and lower panels indicate fractional changes of KaiC 6 and concentration profiles of the Kai proteins, respectively (see details in the Supplemental Experimental Procedures). KaiA 2, KaiB 4, and KaiC 6 particles are schematically illustrated by using orange plums, blue diamonds, and green barrels, respectively. fp shown in parenthesis is a phosphorylation parameter ranging from zero (full dephosphorylation) to one (full phosphorylation) dependent on the degree of the KaiC 6 phosphorylation. (A) Model I. The fractional change of associated KaiC 6 was taken from the previous work (Kageyama et al., 2006). (B) Model II. The fractions of the associated KaiC 6 were optimized by tuning their amplitude and baseline while retaining the original phase in model I. (C) Model III. The association order of the Kai complexes was optimized by using the original amplitude, baseline, and phase in model I. (D) Model IV. Both the fractional change of associated KaiC 6 and the association order were optimized simultaneously while retaining the original phase in model I. Model IV best describes the robust circadian oscillations of Phos app and I(0) app (Figures 4B and 4C). Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc. 711

10 Figure 6. Rhythmic Changes of KaiC 6 Phosphorylation in Intracellular Environments Mimicked by Additions of Cosolvents (A) In vitro oscillation of phosphorylated KaiC 6 fraction (Phos app ) as a function of glycerol concentration. Black circles, 0%; red circles, 9%; blue circles, 19%; and green circles, 26%. (B) Dependence of in vitro rhythm of Phos app on Ficoll 70 concentration. Black circles, 0%; red circles, 9%; blue circles, 17%; and green circles, 23%. The solid lines represent sine-wave functions fitted to the experimental data. The estimated parameters are plotted in Figure 7 and Figure S2. (C and D) Rapid jumps in Ficoll 70 concentration from 0% to 19% (a jump in relative viscosity from 1.0 to 8.2). A ternary mixture containing KaiA 2 (0.15 mg/ml), KaiB 4 (0.15 mg/ml), and KaiC 6 (0.60 mg/ml) was incubated at 30 C (black circles). At IT 24 or 36 hr (black arrows), the solvent condition was changed by a 3-fold dilution of the ternary mixture with a buffer containing concentrated Ficoll 70 (blue circles) or with a Ficoll 70-free buffer (red circles). The solid and dashed lines represent sine-wave functions fitted to the experimental data. Red line, period (T cycle ) = 23.6 ± 0.17 hr and phase (f) = 0.42 ± 0.31 hr; blue line, T cycle = 23.7 ± 0.43 h and f = 0.90 ± 0.94 hr; red dashed line, T cycle = 23.5 ± 0.37 hr and f = 0.19 ± 0.91 hr; and blue dashed line, T cycle = 23.9 ± 0.38 hr and f = 1.71 ± 0.89 hr. Error bars were originated from three separate experiments. because assembly/disassembly processes in vivo are affected to some extent by cytoplasmic viscosity, osmotic pressure, and macromolecular crowding (Parsegian et al., 2000). In order to assess how the Kai oscillator overcomes these effects, in vivo assembly/disassembly dynamics were mimicked by incubating the ternary mixture in the presence of glycerol or the globular sucrose polymer Ficoll 70. As shown in Figure 6A, the amplitude of the Phos app oscillation was reduced as the glycerol concentration was increased. In contrast, a robust oscillation persisted even at high concentrations of Ficoll 70 (Figure 6B). To explore the physicochemical factors modulating the clock activities, the oscillatory parameters were plotted against the concentration of cosolvents (w/w), solvent viscosity, and osmolality (Figure S2). Interestingly, the period was best correlated to the osmotic strength (R = 0.910) and shortened linearly as the osmolality increased (Figure 7A). Because the biological reactions associated with dehydration are favored in high-osmotic environments, the correlation is indicative of a perturbation of the water molecules incorporated into protein-protein interfaces or cavities. However, the observed dependence on osmolality was so weak as to maintain the circadian period at the physiological cytoplasmic osmolality of Osm kg 1 reported for cyanobacterial species (Papageorgiou and Alygizaki-Zorba, 1997). Furthermore, the period was stable over a broad range of Ficoll 70 concentrations (Figure 6B) and around an effective viscosity of cp reported for prokaryotic and eukaryotic cytoplasms (Dix and Verkman, 1990; Mullineaux et al., 2006; Potma et al., 2001) (Figure 7B). These observations suggest osmolality and viscosity compensations for the assembly/ disassembly period of the Kai proteins. On the other hand, the solvent viscosity gave rise to a drastic modulation of the synchronizing stage. Although the Phos app of all the preparations initially peaked at approximately IT 3 hr (Figures 6A and 6B), the slope after the initial peak became less steep as the concentration of Ficoll 70 increased. This resulted in phase shifts of the subsequent oscillatory process (Figure 6B), as evidenced by a linear correlation between the solvent viscosity and the second peaking time (R = 0.832, Figure 7C). Interestingly, the viscosity dependency was characteristic of the synchronizing stage and was not observed evidently in the oscillatory stage. As 712 Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc.

11 Figure 7. Viscosity and Osmolality Dependencies of In Vitro Oscillatory Parameters Circles and squares correspond to the estimates in the presence of glycerol and Ficoll 70, respectively. The line represents linear fitting to overall data on glycerol and Ficoll 70. (A) Osmolality dependence of the oscillatory period (R = 0.910). The inset represents an expanded view around the physiological cytoplasmic osmolality. (B) Viscosity dependence of the oscillatory period (R = 0.313). (C) Viscosity dependence of the second peaking time correlated to the oscillatory phase (R = 0.832). (D) Dependence of the logarithm of oscillatory frequency on osmotic pressure (R = 0.878, 1 atm = 0.1 MPa). Error bars were originated from three separate experiments. shown in Figures 6C and 6D, both the phase and period in the oscillatory stage were little affected by the rapid jumps in the Ficoll 70 concentration from 0% to 19% at IT 24 or 36 hr (a jump in relative viscosity from 1.0 to 8.2), each of which was expected to cause a p/2 phase delay (6 hr) according to the relationship shown in Figure 7C. The current observations suggest that the initial assembly/disassembly before a stable limit cycle oscillation is highly susceptible to the solvent viscosity and is a determinant of the phase of the subsequent oscillatory stage. The viscosity-dependent phase points to the importance of diffusion processes during the synchronizing stage. According to a previous study (Kageyama et al., 2006), the dephosphorylation step from IT 4 to 16 hr is closely related to the exchange of monomeric KaiC subunits among the KaiC 6 hexamers. Because our SAXS experiments indicated no detectable accumulations of the monomeric KaiC subunit (data not shown), the viscositydependent dephosphorylation suggests a collision of the KaiC 6 hexamers as being coupled to rate-limiting steps. However, a simple collision between two KaiC 6 hexamers theoretically occurs considerably more frequently (>10 7 molecule hr 1 ) than the KaiC 6 autodephosphorylation (0.13 hr 1 ). The KaiC 6 dephosphorylation in the synchronizing stage should be activated by a less probable event, such as a simultaneous collision of more than two KaiC 6 s in a unique and precise spatial arrangement and/or a collision among KaiC 6 s with specific phosphorylation states. Interestingly, a recent study demonstrated that the KaiC shuffling to effect autonomous synchronization during the oscillatory stage is highly dependent on the phosphorylation state of KaiC 6 (Ito et al., 2007). Which Comes First, Assembly/Disassembly Dynamics or the KaiC 6 Phosphorylation Cycle? In vitro oscillation of the Kai proteins is realized by passing though the synchronizing stage characterized by the following three points. First, I(0) app is phase delayed by p/2 relative to Phos app, because kinetic KaiA 2 /KaiC 6 interactions are followed by thermodynamic KaiB 4 /KaiC 6 interactions. Second, the viscosity-dependent deceleration of dephosphorylation is coupled to KaiC 6 shuffling. Third, this synchronizing stage is a determinant of the phase of the subsequent oscillatory stage. These characteristics suggest that assembly/disassembly diffusion of the Kai proteins plays an important role as a dominant process of the synchronizing stage. Although a transition from the synchronizing to oscillatory stage is apparently smooth, there should be a switching of the Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc. 713

12 pacemaker during this transition. If the oscillatory stage was driven by the assembly/disassembly diffusion as in the synchronizing stage, not only the phase but also the period should be perturbed in a viscosity-dependent manner. In addition, this can be a risk factor for cyanobacterial species because the period of the Kai oscillator becomes asynchronous dependent on local differences in intracellular viscosity. The viscosity compensation of the period suggests that another factor entrained during the synchronizing stage becomes an alternative pacemaker of the oscillatory stage and then actively drives the assembly and disassembly of the Kai proteins as a subordinate. One of the possible factors so far reported is KaiC 6 ATPase, which is gradually inhibited and entrained in the synchronizing stage (Terauchi et al., 2007). After switching to the oscillatory stage, KaiC 6 ATPase robustly oscillated with a p phase advance relative to I(0) app (p/2 phase advance to Phos app ). Thus, the assembly and disassembly of the Kai proteins during the oscillatory stage are possibly under the control of the KaiC 6 -phosphorylation status (Nishiwaki et al., 2007) that may be further governed by the entrained KaiC 6 ATPase (Terauchi et al., 2007). Our interpretation is further supported by the low-resolution models of the first-order binary complexes, in which KaiA 2 or KaiB 4 is bound to the C2 domains of KaiC 6 (Figures 3A and 3B). The C2 domains of KaiC 6 include functionally important components (Pattanayek et al., 2004), such as an ATP bound in an interface between two adjacent KaiC monomers and dual phosphorylation sites near the bound ATP (Ser431 and Thr432). Furthermore, the C-terminal tails for KaiA 2 recruitment are anchored to a C2 region nearby the bound ATP and dual phosphorylation sites (20 Å from Ile489). Thus, the recruitment of KaiA 2 or KaiB 4 to the C-terminal region of KaiC 6 is a reasonable way to control their assembly/disassembly timing by phosphorylationdependent structural changes of the C2 domains and tails. Once recruited, KaiA 2 and KaiB 4 are required to communicate positively with the pacemaker in order to effect a periodic oscillation. The weak osmotic dependence of the period may be an indication of some positive roles of the recruited KaiA 2 and KaiB 4. Osmotic stresses shift the equilibrium between unbound proteins and their complexes in such a way that the net volume change of osmotically active water becomes negative (DV w < 0). Thus, the formation of protein complexes is often enhanced by the osmotic stress due to the dehydration of molecular surfaces upon collision (Parsegian et al., 2000) and is occasionally reduced because of the additional uptake of water molecules into the proteinprotein interface (Furukawa and Morishima, 2001). A positive slope was confirmed in a logarithmic plot of frequency against the osmotic pressure (R = 0.878, Figure 7D), suggesting a net acceleration of the Kai oscillator by the osmotic enhancement of complex formation (DV w app = 48 ± 11 cm 3 mol 1 ). This result indicates that a period-determining step of in vitro cycling is weakly coupled to KaiA 2 and/or KaiB 4 assembly. In summary, the current study demonstrated that the initial phase of the cyanobacterial oscillator is determined predominantly by the assembly/disassembly communication of the clock components and that the period is essentially resistant to intracellular noise such as collisions, cytoplasmic viscosity, and crowding. These resistances are achieved in the binary and ternary complexes by recruiting KaiA 2 and/or KaiB 4 to the C-terminal regions of the pacemaking KaiC 6 in a phosphorylation-dependent manner. For an ultimate visualization of the cyanobacterial clock, studies to constitute differential equations explaining I(S) app and Phos app simultaneously are in progress. EXPERIMENTAL PROCEDURES Expression and Purification of Kai Proteins In order to construct expression plasmids for Kai proteins, PCR fragments amplified from the template plasmids produced by Nishiwaki et al. (2004) were introduced into a pet3a vector (Novagen). The constructed plasmids for KaiA or KaiB were used to transform Escherichia coli BL21 cells, and protein expression was induced with 0.1 mm isopropyl-b-d-thiogalactopyranoside (IPTG) in LB medium. The E. coli BL21 cells expressing KaiC were cultured in TB medium in the presence of 0.1 mm IPTG. The supernatant from disrupted cell lysates was fractionated by using an ammonium sulfate precipitation method at 4 C. An appropriate fraction was dissolved and purified by several chromatographic steps. The samples with a purity of greater than 95%, as judged by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), were pooled and used for SAXS measurements. SAXS SAXS experiments were carried out at the RIKEN Structural Biology Beamline I (BL45XU) at SPring-8, Japan (Fujisawa et al., 2000). The X-ray wavelength and the sample-to-detector distance were set at 0.9 Å and 2138 mm, respectively. The scattering curves of protein samples and buffer were collected at 30 Cby using a cooled charge-coupled device equipped with an X-ray image intensifier (Fujisawa et al., 1999). After correcting the image for distortions and sensitivities (Ito et al., 2005), the scattering data were normalized by the X-ray intensity, the data collection time, and a particle concentration, c (mg/ml), as described previously (Akiyama et al., 2004). These data treatments resulted in scattering curves, I(S), where S = 2sinq / l, 2q is the scattering angle, and l is the wavelength of the X-ray. The innermost portion of I(S) was fitted under Guinier approximation (Guinier and Fournet, 1955) to the equation I(S) = I(0)exp{ 4p 2 R g 2 S 2 /3}, where I(0) and R g are the forward scattering intensity (S = 0) and the radius of gyration, respectively. The I(0) value is proportional to the weight-averaged molecular weight. Pair distribution functions, P(r), were calculated by indirect Fourier transformation using the GNOM package (Svergun, 1992). Shape Reconstructions Low-resolution shape reconstructions of the Kai complexes were conducted with an ab initio bead-modeling program, MONSA (Svergun, 1999). The shape of the KaiA 2 :KaiC 6 complex was refined against the experimental SAXS curves of KaiA 2, KaiC 6, and the KaiA 2 :KaiC 6 complex, so that the resultant model satisfied the three curves simultaneously (R f = 1.26 ± 0.01). The refinement of the KaiB 4 :KaiC 6 complex was performed in the same manner with the three SAXS profiles of KaiB 4, KaiC 6, and the KaiB 4 :KaiC 6 complex (R f = 1.38 ± 0.01). To confirm the reproducibility of the solution, multiple reconstructions without any symmetrical constraints were performed independently. The resulting models scored with the DAMAVER package (Volkov and Svergun, 2003) revealed a good similarity in shape (NSD = 0.65 ± 0.06 for the KaiA 2 :KaiC 6 complex, NSD = 0.56 ± 0.01 for the KaiB 4 :KaiC 6 complex), from which the most representative model was displayed as shown in Figures 3A and 3B. SUPPLEMENTAL DATA Supplemental Data include Supplemental Experimental Procedures, Supplemental References, and two figures and can be found with this article online at ACKNOWLEDGMENTS The authors thank Drs. T. Kondo, T. Oyama, Y. Kitayama, M. Nakajima, H. Kageyama, T. Nishiwaki, and K. Terauchi (Nagoya University) for sharing data prior to publication; Dr. T. Oda (RIKEN SPring-8 Center) for technical 714 Molecular Cell 29, , March 28, 2008 ª2008 Elsevier Inc.