Dimer dissociation is a key energetic event in the fold-switch pathway of KaiB

Abstract

Cyanobacteria possesses the simplest circadian clock, composed of three proteins that act as a phosphorylation oscillator: KaiA, KaiB, and KaiC. The timing of this oscillator is determined by the fold-switch of KaiB, a structural rearrangement of its C-terminal half that is accompanied by a change in the oligomerization state. During the day, KaiB forms a stable tetramer (gsKaiB), whereas it adopts a monomeric thioredoxin-like fold during the night (fsKaiB). Although the structures and functions of both native states are well studied, little is known about the sequence and structure determinants that control their structural interconversion. Here, we used confinement molecular dynamics (CCR-MD) and folding simulations using structure-based models to show that the dissociation of the gsKaiB dimer is a key energetic event for the fold-switch. Hydrogen-deuterium exchange mass spectrometry (HDXMS) recapitulates the local stability of protein regions reported by CCR-MD, with both approaches consistently indicating that the energy and backbone flexibility changes are solely associated with the region that fold-switches between gsKaiB and fsKaiB and that the localized regions that differentially stabilize gsKaiB also involve regions outside the dimer interface. Moreover, two mutants (R23C and R75C) previously reported to be relevant for altering the rhythmicity of the Kai clock were also studied by HDXMS. Particularly, R75C populates dimeric and monomeric states with a deuterium incorporation profile comparable to the one observed for fsKaiB, emphasizing the importance of the oligomerization state of KaiB for the fold-switch. These findings suggest that the information necessary to control the rhythmicity of the cyanobacterial biological clock is, to a great extent, encoded within the KaiB sequence.

Significance

The cyanobacterial circadian clock is composed of three proteins that act as a phosphorylation oscillator: KaiA, KaiB, and KaiC. Its timing is determined by the fold-switch of KaiB, a rearrangement of its secondary and tertiary structures that is accompanied by a change in its oligomerization state to enable its interaction with KaiC. We used a combination of computational and experimental approaches to identify key local regions in KaiB responsible for its fold-switch and determined that the dissociation of KaiB into monomers is the rate-limiting step for its structural rearrangement. Experimental characterization of KaiB mutants that alter the periodicity of this circadian clock suggest that the clock timing is controlled by protein-protein interactions within the KaiB oligomer or with KaiC.

Introduction

Most protein structures and their folding mechanisms depend on their amino acid sequences, leading to the classic view of “one sequence, one structure, one function” (1). However, several proteins have defied this perspective, such as those containing chameleonic sequences (2, 3, 4, 5, 6, 7) and, more recently, metamorphic proteins (8, 9, 10, 11, 12), whose sequences do not encode a single protein fold. Metamorphic proteins undergo global secondary and tertiary (and in some cases, quaternary) structure transformations, allowing them to interconvert between more than one structural native state, each with its own biological role (8, 9, 10, 11, 12). This peculiarity of metamorphic proteins enables new regulation mechanisms based on triggering their fold-switch, as seen in the transcriptional control of the expression of virulence genes on E. coli (13,14), the conversion of chloride channels between membrane and soluble forms (13, 14, 15), or the control of the circadian rhythmicity in cyanobacteria (16), among others.

The cyanobacterial circadian clock is composed by the proteins KaiA, KaiB and KaiC (17,18), and its 24-h periodicity relies on the metamorphosis of KaiB, a single-domain protein of 108 residues. One peculiarity of the Kai clock is its insensitivity to light changes; instead, it is an ATP-dependent oscillator that relies on the ATP/ADP ratio in cyanobacteria (19,20). Here, the subjective day physiology is dictated through the autophosphorylation of KaiC, which is stimulated by the binding of KaiA to its CII domain (21, 22, 23, 24, 25, 26). The metamorphosis of KaiB regulates this cycle by transforming its structure from a homotetramer (ground state, gsKaiB) into a monomeric thioredoxin-like fold (fold-switched, fsKaiB), which is then able to bind to the N-terminal domain of the phosphorylated KaiC and to KaiA (16,25,26). In this scenario, KaiA binds to fsKaiB and promotes the autodephosphorylation of KaiC, thus leading to the subjective cyanobacterial night physiology (21,27,28).

Both states of KaiB have been extensively studied through structural approaches (29, 30, 31, 32, 33). The gsKaiB state forms a homotetramer where each monomer has a topology βαββααβ (PDB: 1VGL) (29,32). This tetramer is formed by two asymmetrical dimers (29,31) and stabilized by interactions between N-terminal residues 1–14 and C-terminal residues 88–108 that are located at the dimer-dimer interface (34). The role of the tetramer is not well understood, but it has been suggested that it could serve as a monomer reservoir allowing the availability of fsKaiB over a wide range of KaiB expression levels (31). In contrast, the structure of fsKaiB has been determined when bound to KaiC (PDB: 4KSO) (32) or upon stabilization by adding mutations (PDB: 5JYT) (30), and its fold corresponds with a thioredoxin-like monomer with a topology βαβαββα (33). Although both structures of KaiB, as well as its interactions with its binding partners, have been described, little is known about how the fold-switch is encoded within the KaiB sequence.

To decipher how the fold-switch of KaiB is encoded in its sequence and oligomerization state, we studied the thermodynamics of KaiB transformation using free-energy calculations derived from confine-convert-release (CCR) molecular dynamics (MD) simulations in implicit solvent (35, 36, 37), as well as its refolding and fold-switch mechanism via simulations using structure-based models (SBMs) (38,39) with a dual-basin approach (40). CCR-MD simulations determined that the structural transformation of KaiB is dominated by the stabilization of the gsKaiB tetramer and dimer over the monomeric forms, while SBM simulations show that the dissociation of gsKaiB dimer before the fold-switch of KaiB monomers constitutes the highest energy barrier of its transformation mechanism. Consistently, both computational approaches agree in that dimer dissociation is the rate-limiting step of the KaiB fold-switch.

Per-residue energy decomposition from CCR-MD shows that the differential local stability inside the region that is constant upon fold-switching (residues 1–50) was, on average, close to zero. In contrast, residues 51–100, which participate in the structural transformation, showed a differential stability pattern where, strikingly, most local regions that stabilize the gsKaiB fold are not involved in the dimer interface. These patterns of differential stability were overall well-correlated with data obtained from hydrogen-deuterium exchange mass spectrometry (HDXMS) experiments. However, some regions involved in binding of fsKaiB to KaiC showed lower stability than expected based on CCR-MD simulations.

Additionally, we used HDXMS and size exclusion chromatography (SEC) to characterize the KaiB mutants R23C and R75C, which have been previously reported to display periodicities of 26 and 22 h, respectively (41). Also, it has been observed that R23 is relevant for the association with KaiC (30,42). Our analyses demonstrate that the R75C mutant populated different oligomerization states and displayed a deuterium uptake comparable with fsKaiB. In contrast, the R23C mutant exhibits a gsKaiB-like deuterium uptake pattern and does not change its oligomerization state. Therefore, these results suggest that the periodicity of the Kai clock can be modified by either altering the affinity between KaiB and KaiC, as canonically proposed, or by destabilizing the quaternary structure of KaiB itself, as shown in this work.

Materials and methods

Protein expression and purification

The coding sequence for wild type KaiB from T. elongatus (residues 1–108, termed gsKaiB^1–108) and the stabilized fsKaiB mutant A8Y/A89G/R91D/A94Y (fsKaiB^1–108) were synthetized and cloned into a pET28a (+) expression vector downstream of a His-tagged SUMO sequence (Supporting Information; GenScript, Piscataway, NJ). Mutations that alter the circadian rhythmicity (R23C and R75C) were performed following QuikChange site-directed mutagenesis kit (Invitrogen, Waltham, MA) (Supporting Information).

The plasmids were transformed into BL21(DE3) E. coli strain. A colony was picked and grown in liquid LB broth for 16 h at 37°C with vigorous shaking at 200 rpm. The saturated culture was used to inoculate 1 L of LB broth (at 1%) and allowed to grow at 37°C and 200 rpm. Protein overexpression was induced upon reaching an OD₆₀₀ of 0.5 with 1 mM IPTG for 3 h at 37°C and 200 rpm. Cells were harvested by centrifugation at 4000×g for 20 min using a Sorvall LYNX 4000 (Thermo Fisher Scientific, Waltham, MA), resuspended in lysis buffer (500 mM NaCl, 50 mM Na₂HPO₄ pH 8, 30 mM imidazole, 1 mM PMSF) and lysed by sonication using a Model 705 Sonic Dismembrator (Thermo Fisher Scientific) for 5 min at 40% power, using cycles of 5 s on and 5 s of rest time. The lysate was clarified by centrifugation at 20,000×g for 20 min at 4°C in a Centrifuge 5430 R (Eppendorf, Hamburg, Germany), and the supernatant was then loaded onto 3 mL of HisPur Ni-NTA Resin (Thermo Fisher Scientific) previously equilibrated with equilibrium/washing buffer (500 mM NaCl, 50 mM Na₂HPO₄ pH 8, 30 mM imidazole). A wash step was performed with 10 column-volumes of equilibrium/washing buffer to remove impurities before eluting the protein with elution buffer (500 mM NaCl, 50 mM Na₂HPO₄ pH 8, 150 mM imidazole). The pure fractions were collected and incubated with His-tagged ULP1 protease in a 1:50 ratio for excision of the SUMO tag. To remove ULP1 from the mixture, a second Ni-NTA purification was performed. Finally, the protein was purified by SEC on an ÄKTA purifier (GE Healthcare, Chicago, IL) equipped with a Superdex S75 column in SEC buffer (20 mM Tris-HCl pH 7.0, 100 mM NaCl).

HDXMS experiments

HDXMS was performed at the Biomolecular and Proteomics Mass Spectrometry Facility of the University of California San Diego, using a Waters Synapt G2Si system with HDX technology (Waters Corporation, Milford, MA) according to previously described methods (43). All HDXMS experiments were performed over a short period of time (<1 month) in a room with controlled airflow and all HDXMS reactions took place in a closed temperature-controlled sample tray.

Proteins were collected from SEC at approximately 60 μM and passed through a 0.2-μm filter. The samples were then diluted to 15 μM and supplemented to reach the HDXMS buffer (20 mM Tris-HCl pH 7.0, 100 mM NaCl, 1 mM MgCl₂, 1 mM ATP, and 5 mM DTT). Deuterated HDXMS buffer for the exchange reactions was prepared by first lyophilizing a buffer solution prepared in ultrapure water and then redissolving the powder in the same volume of 99.96% D₂O (Cambridge Isotope Laboratories, Inc., Andover, MA) immediately before use.

Protein samples were manually loaded into HDXMS vials (4 μL) to avoid protein aggregation during the experiments and kept at 25°C for 5 min. The samples were then diluted 15 times with deuterated HDXMS buffer using the Leap HDX PAL autosampler to a final volume of 60 μL (Leap Technologies, Carrboro, NC) and deuterium exchange was allowed for four reaction times at 25°C: 30 s, 1 min, 2 min, and 5 min. A zero timepoint was acquired by diluting the samples in a nondeuterated buffer. The reactions were quenched for 1 min at 0.1°C by combining 50 μL of reaction with 50 μL of 3 M GdnHCl with a final pH(D) of 2.66, and then injected into a 100 μL sample loop for digestion on an in-line pepsin column (Immobilized Pepsin, Thermo Fisher Scientific) at 15°C. The resulting peptides were captured on a BEH C18 Vanguard precolumn, separated by analytical chromatography (Acquity UPLC BEH C18, 1.7 μm 1.0 × 50 mm, Waters Corporation) using a 7%–85% acetonitrile gradient in 0.1% formic acid over 7.5 min, and electrosprayed into the Waters Synapt G2Si quadrupole time-of-flight mass spectrometer. Peptide masses were identified from triplicate analyses and data were analyzed using the ProteinLynx global server version 3.0 (Waters Corporation) and HDXMS data was analyzed using DynamX 3.0 (Waters Corporation) following previously described criteria (43).

The deuterium uptake was corrected for back-exchange using a global back-exchange correction factor for the amide deuterium incorporation, taking as reference the peptide with residues 13–24 from fsKaiB by using DECA version 1.12 (44). The HDXMS data processed with DECA for all identified peptides are provided in Table S1 in the Supporting material.

SEC experiments

To determine the oligomerization state of the purified proteins, we performed SEC analyses on an LC-4000 HPLC (Jasco Inc., Easton, MD) equipped with a Superdex S75 10–300 GL column (GE Healthcare). To estimate the experimental molecular weight of the proteins by SEC, a calibration curve was performed using the Bio-Rad Gel Filtration Standard (Bio-Rad, Hercules, CA) (Fig. S1). Purified protein samples of gsKaiB^1–108, fsKaiB^1–108, R23C and R75C were prepared to a final concentration of 15 μM; supplemented with DTT, ATP, and MgCl₂ to reach the conditions of HDXMS buffer; and incubated at 4°C overnight. After incubation, 500 μL of protein was loaded into the column using SEC buffer as mobile phase. To match the protein concentrations in the HDXMS experiment, a second SEC series was performed where the samples were 15-fold diluted before the injection.

CCR-MD

Simulation systems of KaiB were prepared using the crystallographic structure of single-point mutant gsKaiB (PDB: 1VGL) (29) and the NMR structure of quintuple-point mutant fsKaiB (PDB: 5JYT) (30). Both PBDs were mutated back to the wild type sequences and all flexible regions were modeled and relaxed using the loop remodel tool in Rosetta3 (45). This includes residues 2–7 in fsKaiB, and 2–6 and 96–99 in gsKaiB. The structures were parameterized using the implicit solvent Hawking, Cramer, and Truhlar force field and then minimized via the Newton-Raphson minimization incorporated in NAB from Amber16 (46). The systems were prepared to calculate the free energy of conversion from gsKaiB dimer into two monomeric fsKaiB. Hence, the simulated systems were: (gsKaiB)₂ (gs2), 2(gsKaiB) (2gs), and 2(fsKaiB) (2fs). To build the monomeric systems, a protein molecule was duplicated and separated 14 Å apart using Chimera (47). To determine the transformation energy from the gsKaiB tetramer to the dimer, the unmodified structure from the biological assembly 1 of PDB 1VGL was used as the tetramer, while for the two dimers system we manually separated one dimer 14 Å from the other using Chimera (47).

The confinement simulations were performed as reported by Galaz-Davison et al. (37). Briefly, 25 implicit solvent CCR-MD simulations of 30 ns each were performed for all systems without periodic boundary conditions in Amber16 (46). These simulations were confined toward a deeply energy minimized structure by using atom-wise cartesian harmonic restraints, where the stiffness (k_R) of this potential doubled between simulations from 0.000025 to 419.2 kcal/mol/Ų.

The confinement energy was determined as previously reported (37), by integrating through the fluctuations observed in RMSD as follows:

$$\Delta G_{conf}=\sum _{ki}^{kf}\frac{\left({\chi }_{i+1}{k}_{i+1}-{\chi }_i{k}_i\right)}{\left(\frac{\text{ln}\left({\chi }_{i+1}\right)-\text{ln}\left({\chi }_i\right)}{\ln \left(k_{i+1}\right)-\ln \left(k_i\right)}\right)+1}$$ (1)

$$\chi =N\ast RMS{D}^2,$$ (2)

where k is the restraining constant value, RMSD is calculated with respect to the minimized structure, and N is the number of atoms in the system, being this sum calculated for each molecular system simulated.

Finally, to calculate the conversion step (i.e., the free-energy difference between confined states), we used the frequencies obtained from the confined states using normal mode analysis (NMA), corresponding to 3N the number of atoms of each system, from which the first six were negative or null. These values were used as input for the following expression:

$$A\rightleftharpoons B$$

$$\Delta G_{NMA}=\Delta E-k_{B}T\sum _{i=1}^{3N-6}ln\left(\frac{v_A}{v_B}\right),$$ (3)

where ΔE is the difference in the potential energy for B and A, k_BT is the product between the Boltzmann constant and temperature, and υ refers to the frequencies obtained through NMA for each system.

Considering that our system has the following fold-switching pathway for KaiB: gs₂→2gs→2fs, the global energy transformation for the system used here is:

$$\Delta G_{g{s}_{2\to 2fs}}^{trans}={[\Delta G_{g{s}_2}-\Delta G_{2fs}]}^{conf}+\Delta G_{g{s}_2^{\ast }\to 2g{s}^{\ast }}^{NMA}+\Delta G_{2g{s}^{\ast }\to 2f{s}^{\ast }}^{NMA}.$$ (4)

The per-residue transformation energy of KaiB was decomposed as described previously (36,37). Briefly, the confinement energy was determined by applying Eq. 1, but $\chi$ now is determined with respect to individual residue contributions as:

$$\chi =\sum _{i=1}^L{r}_i.$$ (5)

Here, $L$ corresponds with the protein length and $r$ is the RMSF for the ith residue with respect to the minimized structure (36,37). The transformation energy between the microstates (gs₂^∗→2(gs^∗)→2(fs^∗)) was determined by applying the MM-PBSA method to the most restricted MD (k_R = 419.2 kcal/mol/Ų) for each system. The per-residue confinement and MM-PBSA energies determined for each chain and were averaged. Finally, the global transformation energy at the single residue level was determined as:

$$\Delta G_r^{trans}=\Delta G_r^{\text{MMPBSA}}+\Delta G_r^{conf}.$$ (6)

Statistical analysis of errors in CCR-MD

Each simulation at $k_i$ is divided into $N_B=6$ contiguous blocks. The global free energy change is then calculated by replacing the entire $k_i$ simulation with each $N_B$ block. Then, the standard deviation is calculated as

$${\sigma }_{ki}=\sqrt{\frac{1}{N_b-1}\sum _{j=1}^{N_b}{(\Delta G_c^j-\Delta G_C)}^2},$$ (7)

with $\Delta G_c^j$ and $\Delta G_C$ being the confinement free energy for the system with the replaced $N_B$ block and the entire simulation respectively.

Once the error for each leg is obtained, they are combined at each $k_i$ by standard error propagation as:

$$\Delta \Delta G_{AB}=\sqrt{{(\Delta \Delta {\text{G}}_{AA\ast }^{conf})}^2+{(\Delta \Delta {\text{G}}_{BB\ast }^{conf})}^2}.$$ (8)

Preparation of coarse-grained SBMs

Simulations of the gs₂ ⇌ 2fs fold-switch were performed using coarse-grained SBMs (38), as in our previous works (40,48). In these models, each residue is represented by a single bead centered at the coordinates of its Cα atom with native bonds, angles and dihedrals maintained through harmonic potentials, whereas residue pairs in contact in the native state are given attractive interactions and all other nonlocal interactions are treated as repulsive (38,49). In contrast with the use of attractive Lennard-Jones potentials from our previous works (40,48), here we used Gaussian contact potentials, in which the attractive part is treated separately from the repulsive term that defines the excluded volume (49,50):

$$C_G\left(r_{ij},r_0^{ij}\right)=A_{ij}\left[\left(1+\frac{1}{A_{ij}}{\left(\frac{{\sigma }_{NC}}{r_{ij}}\right)}^{12}\right)\left(1+G\left(r_{ij},r_0^{ij}\right)\right)-1\right]$$ (9)

$$G\left(r_{ij},r_0^{ij}\right)=-\exp \left[-{\left(r_{ij}-r_0^{ij}\right)}^2/\left(2{\sigma }^2\right)\right],$$ (10)

where $r_0^{ij}$ is the contact distance for residue pair $i,j$ in the native state (with $|i-j|>3$), $r_{ij}$ is the distance throughout the MD simulation, ${\sigma }_{NC}$ is the excluded volume distance (= 4 Å), $\sigma$ is the width of the attractive Gaussian term (= 0.5 Å), and $A_{ij}$ (= 1 ε) is the depth of the Gaussian minimum, thus effectively fixing the contact minimum $C_G\left(r_{ij},r_0^{ij}\right)$ at $\left(r_0^{ij},-A_{ij}\right)$.

The initial fs and gs₂ structures for generation of SBMs were the same as the ones used for CCR-MD, with a few minor modifications: (i) chain B in gs₂ was replaced by chain A after structural superposition (RMSD = 0.615 Å) such that both monomers have identical native bond lengths, angles, dihedrals, contacts and contact distances; and (ii) an additional SBM was generated in which the fs monomer was duplicated and placed 50 Å away (i.e., 2fs).

Coarse-grained SBMs for fs, 2fs, and gs₂ with Gaussian contact potentials were generated through SMOG2 v2.3 (39), using default parameters for bonded and nonbonded terms as described in detail elsewhere (49). Also, intermolecular native contacts between residue pairs $i,j^{\prime }$ that were lacking for pairs $i^{\prime },j$ in gs₂ were manually added (three contacts), such that each monomer contributed the same number of contacts and interacting residue pairs to the dimer stability. The final SBMs contained 287 and 289 monomer contacts per subunit for fs and gs₂, respectively, and 76 dimer contacts for gs₂.

Last, the SBMs for gs₂ and 2fs were used as input to generate a dual-basin model, in which the native angles, dihedrals and contacts from each model are merged into a single potential energy function (40):

$$V_{angles}^{db}=\sum _{angles\in gs2}{\epsilon }_{\theta }{\left(\theta -{\theta }_0^{gs2}\right)}^2+\sum _{angles\in 2fs}{\epsilon }_{\theta }{\left(\theta -{\theta }_0^{2fs}\right)}^2$$ (11)

$$V_{dihed}^{db}=\sum _{dihed\in gs2}{\epsilon }_{\phi }{F}_D{\left(\phi -{\phi }_0^{gs2}\right)}^2+\sum _{dihed\in 2fs}{\epsilon }_{\phi }{F}_D{\left(\phi -{\phi }_0^{2fs}\right)}^2$$ (12)

$$V_G^{db}=\sum _{ij\in gs2}{C}_G^{gs2}\left(r_{ij},r_0^{ij}\right)+\sum _{ij\in 2fs}{C}_G^{2fs}\left(r_{ij},r_0^{ij}\right)+\sum _{ij\in (gs2\cap 2fs)}{C}_{DB}\left(r_{ij},r_{gs2}^{ij},r_{2fs}^{ij}\right).$$ (13)

To efficiently merge only essential elements of these terms into a single potential energy function, additional criteria were established: (i) gs₂ was considered as the ground state for these dual-basin models, and thus initial native bonds, angles $\left({\theta }_0^{gs2}\right)$, and dihedrals $\left({\phi }_0^{gs2}\right)$ were taken from this structure; (ii) additional angle terms were taken from 2fs and added into the dual-basin model if $|{\theta }_0^{gs2}-{\theta }_0^{2fs}|>10°$ (102 angles); (iii) additional dihedral terms from 2fs were taken only from the fold-switching region (comprising residues 51–100) and added into the dual-basin model; (iv) given that the number of native contacts in gs₂ is higher than in 2fs mainly owing to the presence of dimer contacts, the depth of the native contacts for 2fs was rescaled such that $A_{ij}^{2fs}=A_{ij}\times \left(N_{contacts}^{gs2}/N_{contacts}^{2fs}\right)\cong 1.13$; (v) native contacts unique to gs₂ (189 monomer contacts per subunit and 76 dimer contacts) and 2fs (187 contacts per monomer) were merged and treated with single-basin Gaussian potentials (Eq. 9); and (vi) native contacts between residue pairs present in both gs₂ and 2fs with $r_{gs2}^{ij}/r_{2fs}^{ij}\le 1.2$ (59 contacts) were taken from gs₂ and treated with single-basin Gaussian potentials, whereas those with $r_{gs2}^{ij}/r_{2fs}^{ij}>1.2$ (41 contacts) were treated with a dual-basin Gaussian contact potential (49,50):

$$C_{DB}\left(r_{ij},r_{gs2}^{ij},r_{2fs}^{ij}\right)=A_{ij}\left[\left(1+\frac{1}{A_{ij}}{\left(\frac{{\sigma }_{NC}}{r_{ij}}\right)}^{12}\right)\left(1+G\left(r_{ij},r_{gs2}^{ij}\right)\right)\left(1+G\left(r_{ij},r_{2fs}^{ij}\right)\right)-1\right],$$ (14)

which fixes the depth of both minima, $r_{gs2}^{ij}$ and $r_{2fs}^{ij}$, to $-A_{ij}$.

Fold-switch MD simulations using dual-basin SBMs

MD simulations were run for dual-basin SBMs, as well as control simulations for both single-basin SBMs of fs (Fig. S3) and gs₂ (Fig. S4) and for a dual-basin model without rescaling of the depth of native contacts in 2fs (Fig. S5), using a modified version of GROMACS 4.5.4 (51) containing Gaussian contact potentials (49) available at https://smog-server.org/extension/.

For single-basin fs and gs₂, multiple simulations at 21 different temperatures around the folding temperature (T_F), corresponding with the protein being always folded to always unfolded, were run for 1 × 10⁹ steps using a time step τ of 0.0005, whereas simulations using dual-basin SBMs for KaiB were run at 26 different temperatures around T_F for 5 × 10⁹ steps. For both gs₂ and dual-basin KaiB, we added a harmonic restraint between the centers of mass of each monomer to avoid them from drifting away and performed simulations at two different spring constants for this harmonic restraint ($k$ = 1.0 ε·nm⁻² and $k$ = 4.0 ε·nm⁻²).

Given that folding transitions were seldom observed for gs₂, we employed umbrella sampling simulations at T ∼ T_F using the fraction of native contacts ($Q$) as reaction coordinate:

$$V_{\text{umbrella}}=\frac{1}{2}{k}_{umb}\left(Q-Q_0\right),$$ (15)

where $k_{umb}$ is the strength of the umbrella potential (= 0.005 ε) and $Q_0$ is the reference fraction of native contacts in which the potential is centered. Multiple simulations were performed at different $Q_0$ values ranging from approximately 0.05 to approximately 0.95, varying $Q_0$ in steps of 0.06.

Free-energy profiles as a function of different reaction coordinates were obtained from these simulations using the weighted histogram analysis method (52) implemented as a java application in SMOG2 (39).

Data availability

The CCR and SBM MD simulations are available for download at the laboratory's simulation archive in the Open Science Framework (https://osf.io/bn6u3/).

Results

The gsKaiB tetramer and dimer are thermodynamically favored over the monomeric folds

CCR has been previously used to calculate the transformation energy for monomeric metamorphic proteins (35, 36, 37). However, KaiB forms a homotetramer in solution in its ground state structure (29), comprising two dimers that interact in the tetramer through residues 1–14 and 88–108 (34). It has been observed that when the C-terminal or the N-terminal residues are missing, KaiB forms a dimeric structure that is sufficient to enable circadian rhythmicity in vitro (31,34).

The current structures of KaiB from T. elongatus available in the PDB are 1VGL for gsKaiB and 5JYT for fsKaiB, which have secondary structure information for residues 7–101 and for 8–99, respectively. Given the lack of structural information for residues 99–108 in fsKaiB and the requirement of the same number and identity of atoms in CCR (37), we only considered residues 1–100 for the refinement of both structures. Reasoning that the C-terminal residues are important for the tetramer stabilization without affecting the function of the circadian clock in vitro (31,34) and that each monomer in the dimeric gsKaiB keeps the same structure as in the tetramer (53), the fold-switch of KaiB was decomposed in CCR using the pathway gs₂→2gs→2fs. Thus, the CCR-MD simulations were performed for the dimeric and monomeric forms of gsKaiB and for monomeric fsKaiB, and the transformation energy (ΔG^trans) was determined as shown in Fig. 1 A.

Figure 1.

CCR for the transformation of KaiB. A. Thermodynamic cycle for the transformation of gs₂→2fs. The inset shows a pictorial representation of the thermodynamic analysis. Here, ΔG^trans corresponds with the global transformation free energy, ΔG^conf with the confinement free energy, and ΔG^NMA with the transformation energy from the confined states. B. Transformation energy of gsKaiB to fsKaiB, the global energy is shown in black, the dissociation energy of the dimer is shown in light gray, and the transformation between monomeric states is shown in dark gray. The error bars correspond to the confinement standard deviation determined as shown in Materials and methods. To see this figure in color, go online.

The transformation of KaiB from gsKaiB dimers to fsKaiB monomers has a global energy cost of 29 ± 2 kcal/mol, which can be decomposed in each transformation step. This value agrees with fold-switching energies observed for other metamorphic proteins and proteins with chameleonic sequences (36,37). As can be seen in Fig. 1 B, the transformation of one gsKaiB subunit from the dimeric state to its monomeric state has an energy cost of 30 ± 3 kcal/mol. It is important to highlight that our simulations ignore the entropic gain for the system, i.e., the solvent and number of molecules in the system owing to dimer dissociation. Consequently, the ΔG^trans of gs₂→2gs is not directly comparable with a dissociation free energy.

The fold-switch was then evaluated from the monomeric state of gsKaiB to fsKaiB, a transition that is favored by a free energy of −1 ± 3 kcal/mol (Fig. 1 B). Additionally, the energy for the structural transformation from the tetramer to the dimeric form of gsKaiB was also computationally determined by CCR-MD, using the solved tetrameric structure of KaiB (PDB: 1VGL). Using this approach, we determined a transformation energy from the tetramer to the dimer of 4 ± 3 kcal/mol per monomer (Fig. S2). This transformation energy is significantly smaller than the energy gap between the native gsKaiB dimer and the monomeric gs and fs folds.

Altogether, these results suggest that dimer dissociation is the most thermodynamically costly step for KaiB fold-switch and that, once the dimeric gsKaiB dissociates, the fold-switch to fsKaiB occurs at almost equimolar concentrations.

Dimer dissociation is the limiting step for the transformation of KaiB

Based on the CCR-MD results and on the fact that the gsKaiB dimer is sufficient to enable normal oscillations of the Kai circadian clock in vitro (31,34), we decided to computationally determine the fold-switch mechanism of KaiB and establish whether the dimer dissociation constitutes its rate-limiting step. To achieve this, we performed simulations using coarse-grained SBMs in which all native contacts are treated with attractive Gaussian contact potentials (49,50).

Folding simulations using single-basin SBMs show that the fsKaiB monomer folds in a two-state manner, with the native and unfolded states separated by a free-energy barrier of 5 k_BT (Fig. S3). In contrast, gsKaiB is consistent with a three-state model in which the native dimer (i.e., the fraction of native monomer (${\text{Q}}_{\text{M}}$) and dimer (${\text{Q}}_{\text{D}}$) contacts are ∼1) and the unfolded monomers (${\text{Q}}_{\text{D}}=0$, ${\text{Q}}_{\text{M}}\sim 0.1$) are connected via a scarcely populated, natively folded monomer (${\text{Q}}_{\text{D}}=0$, ${\text{Q}}_{\text{M}}\sim 1$, Fig. S4). Also, the free-energy barrier for the dimer dissociation in gsKaiB is just slightly higher than the free-energy barrier for gs monomer unfolding, but more than 2-fold higher than the free-energy barrier for fs unfolding.

To simulate the refolding from gsKaiB dimer to fsKaiB monomers, we used a dual-basin SBM generated by merging specific angle, dihedral, and native contact terms from the single-basin SBMs for gsKaiB and fsKaiB (see Materials and methods), as previously used for simulating the fold-switch of the metamorphic protein RfaH (40). It is worth noting that, in these models, the depth of the native contact minima for contacts formed in the fs state was rescaled based on the ratio between the total number of contacts between gs₂ and 2fs.

The results from these simulations (Figs. 2 and S6) show that these dual-basin models for KaiB exhibit two peaks in heat capacity, thus representing two folding temperatures, T_F1 and T_F2 (Fig. 2 A). We first explored the folding events occurring at T_F1 by examining the changes in formation of native monomer contacts unique to fsKaiB and gsKaiB ($Q_{fs}^{uniq}$ and $Q_{gs}^{uniq}$) and how these changes are associated with dimer dissociation (Fig. 2 B and C).

Figure 2.

Fold-switch of KaiB explored using dual-basin SBMs. Simulations were performed using a harmonic constraint between the centers of mass of each monomer with a spring constant k = 1.0 ε·nm⁻². A. Heat capacity change for dual-basin KaiB, with T_F1 ∼ 1.06 and T_F2 ∼ 1.15 reduced units. B. Two-dimensional free-energy landscape of the fold-switch of KaiB at T_F1 as a function of the fraction of monomer native contacts unique to the fs ($Q_{fs}^{uniq}$) and the gs ($Q_{gs}^{uniq}$) folds. The free-energy barrier separating both folds is ∼7 k_BT. C. Two-dimensional free-energy landscape of KaiB fold-switch, using the difference in the fraction of unique native contacts formed for fs and gs monomers ($Q_{diff}=Q_{gs}^{uniq}-Q_{fs}^{uniq}$) and the fraction of dimer contacts (${\text{Q}}_{\text{D}}$) as reaction coordinates. If no intermonomer contacts are formed (${\text{Q}}_{\text{D}}=$ 0), then the change in distance between the centers of mass of each monomer with respect to the distance in the native dimer ($d=d_{A-B}-$ 2.63 nm) is plotted below ${\text{Q}}_{\text{D}}=$ 0. D. Population fractions for KaiB dimer (green), monomeric gs (orange) and fs (blue) and the unfolded ensemble (red) as a function of temperature. To see this figure in color, go online.

As shown in Fig. 2 B, KaiB switches between the fs and gs folds at T_F1 via a transition state of approximately 7 k_BT that is defined within the regions [0.3 ≤ $Q_{fs}^{uniq}$ ≤ 0.5, 0.3 ≤ $Q_{gs}^{uniq}$ ≤ 0.5]. Control simulations using a dual-basin SBM without rescaling of the native interactions in fs show that KaiB does not fold-switch (Fig. S5). The use of a higher spring constant for the harmonic restraint between monomers (Fig. S6) does not significantly change the observed free-energy landscape.

We then examined the change in the number of unique native contacts from both monomer folds (${\text{Q}}_{\text{diff}}=Q_{gs}^{uniq}-Q_{fs}^{uniq}$) as a function of the fraction of dimer contacts (${\text{Q}}_{\text{D}}$), demonstrating that the fold-switch from gsKaiB dimer (${\text{Q}}_{\text{diff}}$ ∼ 0.5, ${\text{Q}}_{\text{D}}$ ∼ 1) to fsKaiB monomer (${\text{Q}}_{\text{diff}}$ ∼ −0.5, Q_D = 0) is mediated by a scarcely populated gs monomer (${\text{Q}}_{\text{diff}}$ ∼ 0.5, ${\text{Q}}_{\text{D}}$ = 0, Fig. 2 C). Moreover, the free-energy barrier for the dissociation of the gsKaiB dimer into gs monomers (Fig. 2 C) is approximately 3 k_BT higher than the free-energy barrier for the fold-switch between the gs and fs monomer (Fig. 2 B). It is also worth noting that, while a significant fraction of dimer contacts can be formed in the fs fold (${\text{Q}}_{\text{diff}}$ ∼ −0.5, $Q_D$ ∼ 0.6), these protein-protein interactions are not sufficient to drive KaiB from the fs to the gs fold via oligomerization into a fs dimer.

Last, we determined the population fractions of all states observed in these simulations, namely, the dimer (defined as any configuration with ${\text{Q}}_{\text{D}}$ > 0.40), the gs ($Q_{gs}^{uniq}$ > 0.53) and fs monomers ($Q_{fs}^{uniq}$ > 0.40), and the unfolded state (Fig. 2 D). These results show that the dissociation of the gsKaiB dimer into its constituent monomers leads to the concurrent fold-switch into the fs monomer such that the accumulation of fsKaiB occurs on the detriment of gsKaiB, and that T_F2 corresponds with the T_F of the fsKaiB monomer.

Altogether, our dual-basin models determine that dimer dissociation is the rate limiting step of the fold-switch of KaiB, which follows the pathway gs₂→2gs→2fs used in our CCR-MD simulations.

The local stability of KaiB is consistent with backbone flexibility ascertained by HDXMS

From the CCR-MD simulations, the transformation energy difference at single residue level was determined as described in the Materials and methods. Following the transformation pathway as gs₂→2fs, if ΔG^trans > 0, the residue is more stable gsKaiB and if ΔG^trans < 0 is more stable in fsKaiB (Fig. 3 A). These differences can be observed in the structure in Figs. 3 B and S7 A. A comparison of the residue-level transformation energies shows that the region that remains structurally invariant in both KaiB folds (residues 1–50) exhibits ΔG^trans values close to zero, meaning that these regions do not prefer one specific fold.

Figure 3.

Per-residue energy decomposition for KaiB fold-switching. A. Per-residue energy decomposition obtained from CCR-MD, where residues that are stabilized in gsKaiB and fsKaiB are colored in orange and blue, respectively. Also, a smoothing of the values using an average window of five residues is shown in dark gray. B. Cartoon representation of the structure of fsKaiB. The structure is colored following the energy color scale described in A. C. Local structural context for residue Y40 in both KaiB folds. To see this figure in color, go online.

The few residues from this region that stabilize the protein in gsKaiB conformation with a |ΔG^trans| ≥ 2 kcal/mol are E35, Y40 and K49 (Fig. 3 A). Particularly, Y40 is oriented toward the hydrophobic core in both KaiB states but having different environments (Fig. 3 C). In fsKaiB, Y40 is surrounded by residues from helix H3 (I97 and Y94), which is unstable and loses secondary structure in MD simulations with low restriction constants (Fig. S8). This could affect the stability of the pocket where Y40 is embedded and lead to overestimation of its preferential stability toward gsKaiB.

For fsKaiB, the residues in this region that meet the free energy threshold of 2 kcal/mol are N17 and N29 (Table S2). Both residues are exposed to solvent in fsKaiB state, but oriented toward the protein core in gsKaiB. Particularly, the orientation of N29 in fsKaiB could be a result of reversing the mutation (N29A) present in the solved structure for the fold-switch stabilized mutant of KaiB (PDB: 5JYT) in this position. This could have led to finding an optimal configuration for N29 that is more stable in the fsKaiB state and that might cause overestimation of its energetic contribution to the stability of this fold.

Regarding the protein region that experiences the structural transformation upon fold-switch (residues 51–100), CCR simulations show that localized regions throughout its sequence differentially stabilize each fold (Fig. 3 A), as seen in other metamorphic proteins (36,37). Here, the regions that display a noticeable stabilizing effect for gsKaiB are between residues 53–60, 64–70, 78–81, and 93–98. Among them, residues 53–57 correspond with the only region that participates in the dimer interface. The highest free energies are exhibited by residues T64 and V68, with values greater than 3.5 kcal/mol. Both residues are part of helix H2 and are exposed to the solvent in gsKaiB, whereas their side chains are buried in polar surroundings in fsKaiB.

The region that preferentially stabilizes fsKaiB comprises residues 71–77, where P71 is stabilized by an energy of 3.4 kcal/mol. This residue is at the end of helix H3 in gsKaiB and is exposed to solvent in the loop that connects strands B2 and B3 on fsKaiB. Although P71 undergoes an isomerization going from trans to cis after fold-switch (54), the stabilization of this residue could be associated with the preference of the proline to form turns instead of being in a helical structure (55). Altogether, the observations from per-residue energy decomposition suggest that the surroundings of the residues define the differential stability behavior seen in CCR-MD.

The local stability determined by CCR-MD was compared with the local flexibility ascertained by HDXMS. While CCR-MD were performed using the dimeric KaiB, our experimental approach used the full-length wild type sequence of KaiB from T. elongatus, gsKaiB^1–108, which displays a tetrameric oligomerization state in solution. Also, we studied the fold-switch stabilized mutant fsKaiB^1–108 (Y8A/G89A/D91R/Y94A). From the experiments, 58 peptides were obtained for gsKaiB^1–108 and 55 peptides for fsKaiB^1–108, allowing us to analyze the entire sequence (Fig. S9).

The difference in deuterium uptake (Δ²H) was determined from a subset of contiguous and overlapping peptides for each structure at the longest exposure time (5 min), which showed that gsKaiB^1–108 on average exchanges less deuterium in comparison with fsKaiB^1–108 (purple and magenta lines in Figs. 4 A and Fig. S9). These results suggest that gsKaiB^1–108 is more rigid throughout the whole protein sequence under these experimental conditions, except for the C-terminus (residues 85–108), which exhibits less deuteron incorporation in fsKaiB.

Figure 4.

Backbone flexibility of KaiB in both folds. A. Comparison between Δ²H uptake and per-residue ΔG^trans. A set of peptides identified for both KaiB folds was used to calculate the difference of deuteron incorporation as Δ²H = ²H_fs-²H_gs. When Δ²H ≥ 0 peptides are less flexible in the gsKaiB state (purple) and when Δ²H ≤ 0 peptides that are less flexible in fsKaiB (magenta). Orange and blue lines correspond to a smoothing of five residues of the per-residue energy decomposition. B. Cartoon representation of fsKaiB colored by following the same color code from A. C. Deuteron incorporation kinetics for peptides that showed opposite behavior in comparison to the per-peptide energy showed in A, with gsKaiB and fsKaiB shown in purple and magenta, respectively. The light pink dot in peptides 41–47 was not considered for the exponential fit. To see this figure in color, go online.

We then compared Δ²H with the smoothed ΔG^trans (orange and blue lines in Fig. 4 A), exhibiting an overall good consistency between experiments and simulations. For example, backbone amide Δ²H for residues 1–50 show that the gsKaiB fold is slightly more protected against deuteration, which is like the near-zero ΔG^trans observed in CCR simulations. Also, the dimer interface region comprised by peptide 57–65 is shown to be mostly stabilizing toward gsKaiB both by CCR-MD and HDXMS.

The most significant discrepancies between experimental and computational data are observed for regions comprising residues 13–32 and 41–53 (Fig. 4 A). To evaluate whether these differences were affected by the dissociation of the monomers or by the transformation itself, the per-residue energy was decomposed at each step in the reaction pathway. In this way, we noticed that the discrepancies observed for regions 13–32 and 41–53 are related to the transformation of KaiB at the monomeric state (Fig. S10).

We also individually analyzed the deuteron incorporation kinetics of peptides that were different between CCR and HDXMS experiments (Fig. 4 C). Residues 13–29, 25–32, and 41–53 in fsKaiB^1–108 not only incorporate more deuterons, but also with faster exchange rates than gsKaiB^1–108 (Fig. 4 C). These results suggest that fsKaiB has higher local structural flexibility than gsKaiB, which could be countered in vivo when fsKaiB binds to its partner KaiC, stabilizing the thioredoxin-like fold. It is interesting to highlight that all of these regions are part of the interaction surface with KaiC (30), and that the comparison of our computational results against readily available HDXMS data on fsKaiB in the context of its interaction with KaiC (30) corrects these discrepancies (Fig. S11). These observations suggest that KaiC is crucial to stabilize the structure of fsKaiB in solution.

Structural evaluation of mutants R23C and R75C on KaiB from T. Elongatus

It has been reported that mutations R22C and R74C in S. elongatus KaiB alter the rhythmicity of the Kai clock in vivo and in vitro (41). In the context of the tetramer of gsKaiB, R75 from chain D establishes an electrostatic interaction in the dimer-dimer interface with E95 from chain A, whereas R23 is exposed to solvent in both the tetrameric and dimeric forms of gsKaiB (Fig. 5 A, left). Both residues are exposed to the solvent in fsKaiB (Fig. 5 A, right). To evaluate whether these mutations alter the structure of gsKaiB from T. elongatus, we experimentally generated the R23C and R75C mutants for further analysis by HDXMS.

Figure 5.

Deuterium uptake for the wild type KaiB and its mutants. A. (Left) Structure of chains A and D of gsKaiB (PDB: 1VGL). The R22 side chain is in dark cyan and the R75 in gray. R75 from chain D forms an electrostatic interaction with the E95 from chain A. (Right) Structure of the fold-switched stabilized structure of KaiB (PDB: 5JYT), R23 is illustrated in dark cyan and R75 is illustrated in gray. B. Deuterium incorporation at 5 min in 14 peptides for wild-type KaiB, R23C, and R75C and the fs-stabilized mutant. Chromatograms of gsKaiB^1–108, fsKaiB^1–108, R23C, and R75C at 15 μM (C) and 1 μM (D). The arrows mark the peaks observed where 1 corresponds to an MW of approximately 60 kDa; 2 to approximately 20 kDa, and 3 to approximately 35 kDa (Table S3 and Fig. S1). To see this figure in color, go online.

In HDXMS experiments, 55 peptides were detected for the R23C mutant and 49 peptides for the R75C mutant (Fig. S12). Using the same set of peptides as those analyzed in Fig. 4 A, we compared the deuteron uptake in these single-point mutants against gsKaiB^1–108 and the fold-switch stabilized fsKaiB^1–108. The deuterium uptake of both mutants is similar to gsKaiB^1–108 over almost the entire sequence, except for peptides comprising residues 66–80 and 81–89 in the R75C mutant (Fig. 5 B). Given that the region covered by residues 66–80 harbors the R75C mutation, the observed increase in deuterium uptake for this peptide could be partly explained by a local destabilization. In contrast, residues 81–89 cover the end of helix H3 and the beginning of strand B4 that participate in the dimer interface of gsKaiB and are interconverted into helix H3 in fsKaiB.

Reasoning that these local changes in deuterium uptake in the R75C mutant resemble fsKaiB and that KaiB fold-switching occurs through a change in the oligomerization state, we evaluated the latter by SEC under protein concentration conditions in which the HDXMS experiments were performed (Fig. 5 C and D). Protein samples were prepared at 15 μM, where gsKaiB^1–108 and mutants R23C and R75C were tetrameric with predicted molecular weights (MW) of approximately 60 kDa (peak 1 in Fig. 5 C; Table S3), while fsKaiB^1–108 showed an MW of approximately 20 kDa, previously related to a monomeric species in solution (peak 2 in Fig. 5 C; Table S3) (34). Since the protein is 15-fold diluted in deuterated buffer for the HDXMS experiments, we evaluated whether there is any change on the oligomerization state upon dilution. In Fig. 5 D, it can be observed that, at 1 μM, there is no change in the fsKaiB^1–108 retention volume. However, gsKaiB^1–108 and mutants R23C and R75C display peak tailing, suggesting a subtle level of instability of the tetramer under these conditions. Interestingly, the R75 mutant populates a different species (peak 3 in Fig. 5 D), which corresponds with an MW of 35 kDa, consistent with a dimer (34). This propensity of populating the dimeric state could be associated with the disruption of the intermolecular interaction in which R75 participates in the tetramer (Fig. 5 A). Considering that the R75C mutation in gsKaiB affects the tetramer stability as ascertained by SEC, and the deuterium uptake profile for peptides comprising residues 66–80 and 81–89 resembles what is observed for fsKaiB, we hypothesized that this mutant has a greater propensity to populate the fsKaiB state. These results are in good agreement with the observations from our computational results, regarding dimer dissociation as the key energetic step for KaiB fold-switch, and with the in vivo effect of this mutant in shortening the period of the circadian clock (41).

Discussion

Metamorphic proteins have been intensively studied both computationally and experimentally, but the thermodynamics and mechanisms of fold-switch have been less explored. Regarding the thermodynamics, a few years ago Cecchini et al. reported a new methodology that allows computation of the transformation energy using the CCR method via MD simulations (35). Using this approach, the thermodynamics and per-residue energetic contributions to the transformation of the natural RfaH and the engineered GB proteins have been recently reported (35, 36, 37). In terms of refolding, we reported the use of dual-basin SBMs to explore the fold-switch mechanism of RfaH (40). Here, we employed these strategies to study the fold-switch of the oligomeric metamorphic protein KaiB.

In solution, the ground state of KaiB forms an asymmetrical dimer of dimers, composed of four subunits (29,31). This protein undergoes a change in its oligomerization state and subsequent fold-switch, where each monomeric unit can now bind to KaiC and KaiA (16,25,26). While it has been suggested that the tetrameric form of gsKaiB could have a role in controlling the availability of the monomeric protein inside the cell, which could help on the regulation of the cyanobacterial clock (31), it has been observed that the stability of this oligomer is susceptible to high salt concentrations (34) and that the dimer is sufficient to sustain the biological role of KaiB in vitro (31,34). Because of the sufficiency of the gsKaiB dimer for biological function, the lack of structural information of the ends of KaiB sequence in the deposited structures available and the absence of differences between gsKaiB monomers within the dimer when compared with the monomers within the tetramer (53), we decided to perform our CCR-MD and SBM simulations on the dimeric structure of gsKaiB.

Both simulation strategies are consistent in showing that the tetramer and dimer are the most stable oligomeric states of KaiB. Indeed, dimer dissociation is both the most thermodynamically costly step and the highest kinetic barrier for the fold switch of KaiB. The per-residue energy decomposition from CCR-MD showed that residues in the N-terminal half of KaiB sequence do not change upon fold-switching and maintain similar energies after the structural rearrangement. In contrast, residues 51–100 that are subject to changes in secondary structure topology during fold-switch exhibit a differential stability pattern, showing a preference for either the gsKaiB or the fsKaiB fold. The regions that remain structurally invariant, with a ΔG^trans close to zero, are important to stabilize the core of this metamorphic protein, suggesting they act as an anchor for mutually exclusive interactions against the C-terminal half of KaiB in both folds. In this scenario, we argue that mutations in this region should equally destabilize both folds and primarily alter how the transforming regions interact with the protein core, thus mainly affecting the fold-switching kinetics.

CCR predictions have previously shown a good correlation with experimental data from HDXMS, NMR-based HDX (37), and with GB90/GA90 sequence composition (36). Despite an overall agreement when this comparison is now made against HDXMS data for KaiB, some discrepancies are observed. While in solution residues 1–80 in fsKaiB^1–108 are more flexible when compared with gsKaiB^1–108, CCR predicts that some regions are more stable in fsKaiB. The regions that present the most significant discrepancies between our experimental and computational data comprise residues 13–32 and 41–53. All the discrepancies are corrected when we consider fsKaiB bound to KaiC for the HDXMS data analysis, data available from Tseng et al. (30). These results suggest that KaiC is important to maintain structural rigidity when KaiB visits the fold-switched state and, therefore, binding to KaiC is crucial for fsKaiB stabilization.

It is well known that the Kai clock does not depend on the day/night light signal to maintain a robust 24-h oscillation (56, 57, 58, 59); however, these light changes alter the ATP/ADP ratio in cyanobacteria (60), which in turn shift the phosphorylation phases of KaiC (61). Interestingly, it has been observed that the interaction between KaiC and KaiB relies on the ATP hydrolysis of the CI domain of KaiC, which post-ATP-hydrolysis can bind to KaiB (62,63). Therefore, the interaction between KaiB and KaiC relies on the ATP/ADP ratio in cyanobacteria and therefore to stabilize the structure of fsKaiB it is important the posthydrolysis state on KaiC.

Last, we examined the local effect of two KaiB mutants, R22C and R74C (corresponding with R23C and R75C in T. elongatus), that alter the circadian rhythmicity of the Kai clock in S. elongatus. Residue R23 is located at the interface with KaiC (30,42) and contributes to electrostatic complementarity between adjacent monomers of KaiB (30). Mutants in R23 can cause a change in the clock's periodicity (41), probably owing to a reduction in the binding affinity with KaiC (53,64). The R75C mutant had deuterium uptake profiles for some regions that were similar to fsKaiB^1–108. Mutations in R75 have been shown to change the clock periodicity to 22 h (17,41). Our work provides an explanation of this observation; the R75C mutant shows both global and local structural consequences. Mutation of R75 disrupts an intermolecular interaction in the tetrameric state of gsKaiB, leading to a change in oligomerization state at low protein concentrations. In addition, this mutation induces changes in local flexibility in residues 66–80 and 81–89, as ascertained by HDXMS, suggesting that the R75C mutant makes KaiB more prone to adopt the fold-switched conformation without the presence of its binding partner KaiC.

Altogether, the computational and experimental approaches used in this work consistently indicate that the dissociation of gsKaiB dimer is the limiting step for KaiB fold-switch, and that the presence of KaiC could be necessary to properly stabilize fsKaiB. Thus, we hypothesize that the periodicity of the Kai clock can be modified by altering the affinity between KaiB and KaiC (R23C mutant) or by destabilizing the quaternary structure of KaiB itself (R75C mutant).

Author contributions

M.R., P.G.-D., E.A.K., and C.A.R.-S. designed the research. M. R., P.G.-D., I.R.-F., and C.A.R.-S. conducted and analyzed the computational work. M.R. conducted HDXMS experiments. M.R., P.G.-D., I.R.-F., E.A.K., and C.A.R.-S. analyzed the data. M.R., P.G.-D., E.A.K., and C.A.R.-S. wrote the manuscript.

Editor: Yuji Sugita.