Reaction mechanism of the bioluminescent protein mnemiopsin1 revealed by X-ray crystallography and QM/MM simulations

Bioluminescence of a variety of marine organisms, mostly cnidarians and ctenophores, is carried out by Ca2+-dependent photoproteins. The mechanism of light emission operates via the same reaction in both animal families. Despite numerous studies on the ctenophore photoprotein family, the detailed catalytic mechanism and arrangement of amino acid residues surrounding the chromophore in this family are a mystery. Here, we report the crystal structure of Cd2+-loaded apo-mnemiopsin1, a member of the ctenophore family, at 2.15 Å resolution and used quantum mechanics/molecular mechanics (QM/MM) to investigate its reaction mechanism. The simulations suggested that an Asp-156–Arg-39–Tyr-202 triad creates a hydrogen-bonded network to facilitate the transfer of a proton from the 2-hydroperoxy group of the chromophore coelenterazine to bulk solvent. We identified a water molecule in the coelenteramide-binding cavity that forms a hydrogen bond with the amide nitrogen atom of coelenteramide, which, in turn, is hydrogen-bonded via another water molecule to Tyr-131. This observation supports the hypothesis that the function of the coelenteramide-bound water molecule is to catalyze the 2-hydroperoxycoelenterazine decarboxylation reaction by protonation of a dioxetanone anion, thereby triggering the bioluminescence reaction in the ctenophore photoprotein family.

Ca 2ϩ -binding photoproteins are considered as valuable biological tools in a variety of biochemical fields such as gene expression analysis (1)(2)(3), drug discovery (4,5), and protein dynamics investigations (6). These proteins consist of an apoprotein and a noncovalently bound chromophore, called peroxy-coelenterazine, located in a hydrophobic cavity. Ca 2ϩ binding causes an intermolecular reaction where peroxycoelenterazine is oxidized to coelenteramide, resulting in CO 2 release and light emission (7)(8)(9). To regenerate the active complex, the coelenteramide dissociates from the active site, together with the loss of calcium ions, to form the apoprotein. The charged photoprotein is then regenerated by the addition of peroxy-coelenterazine in the presence of molecular oxygen and the binding of calcium ions to the protein.
Structural investigations of a number of cnidarian photoproteins including aequorin, obelin, and clytin in apo (10), coelenterazine-bound (11)(12)(13), and coelenteramide-bound (14 -17) states indicate that some intermediates are engaged in the oxidative decarboxylation of coelenterazine. Regeneration of the peroxy anion occurs in the presence of Ca 2ϩ and with the help of a His-Tyr-Trp catalytic triad. In this mechanism, the binding of Ca 2ϩ causes a conformational change that makes the hydrogen bond between His-175 and Tyr-190 (obelin numbering) stronger. As a result, His-175 and Tyr-190 residues become protonated and nucleophilic, respectively, with the tyrosine being protonated by the hydroperoxy group of coelenterazine leading to the production of a peroxy anion (8).
The only published crystal structures of a photoprotein from the ctenophore family is apo-berovin, bound to calcium (18) and magnesium ions (19). Despite the similar bioluminescence reaction and a high degree of structural similarity between the cnidarian and ctenophore photoproteins, there is a low degree of sequence identity between these families; for instance, the sequence identity between berovin and aequorin is 29.4%. Indeed, attempts to identify the catalytic triad in the cnidarian family based on structural and mutagenesis studies of the ctenophore family have not successful (18, 20 -24). In our recent QM/MM study on the holo-berovin, we proposed that Asp-158 -Arg-41-Tyr-204 could form a hydrogen-bonded network around the peroxy group that could shuttle a proton from the peroxide group to bulk solvent (25).
The ctenophore family member mnemiopsin1 was originally isolated from luminous Mnemiopsis leidyi from the Caspian Sea in Iran and has been well-studied since (20 -22, 24 -27). In the present study, we report the crystal structure of apo-mne-miopsin1 at 2.15 Å resolution. The structure reveals that mne-miopsin1 has a two-domain fold with four ␣-helices in each domain. We then applied QM/MM 4 calculations on the structure to identify the catalytic triad of residues in the cnidarian family and to gain a detailed understanding of the reaction mechanism.

Overall structure
Like other photoproteins, apo-mnemiopsin1 has a two-domain fold with four ␣-helices A-D and E-H in the N-and C-terminal domains, respectively (Fig. 1). The 78% sequence identity between mnemiopsin1 and berovin is reflected in the RMSD of 0.8 Å over 179 C␣ atoms. Similar to apo-berovin, the C-terminal residues (residues 194 -206) of apo-mnemiopsin1 are not observed in the electron density map and are assumed to be disordered. However, the N-terminal loop (residues [27][28][29][30][31][32][33][34][35], missing in apo-berovin, is present in the electron density map of apo-mnemiopsin1 (Fig. S1). The EF-hand interhelical hydrogen-bond network for apo-mnemiopsin1 is shown in Fig. 2. The loops of EF-hand I are paired with EF-hand II by main-chain/main-chain hydrogen bonds between Met-51 and Leu-96, Met-53 and Asp-91, and Gly-49 and Arg-98. Side-chain/side-chain hydrogen bonds occur between Asp-41 and Arg-98. The loops of EF-hand III and IV interact with each other through hydrogen bonds between main-chain atoms of Val-143 with Leu-177. Side-chain/ main-chain hydrogen bonds occur between Arg-179 and Asn-129, Try-132, and Leu-177/Arg-179. Finally, sidechain/side-chain interactions are found between Asp-133 and Arg-179, Lys-176 and Thr-142/Asp-140, and Tyr-163 and Lys-149. The hydrogen bonds from Arg-35 to Lys-188 and Glu-192 together with hydrogen bonds from Arg-39 to Phe-189 cap the hydroperoxycoelenterazine cavity, resulting in a solvent-inaccessible environment.

Calcium-binding loops
The presence of three Cd 2ϩ loaded EF-hand Ca 2ϩ -binding loops is observed in the apo-mnemiopsin1 structure (Fig. 3). Each loop (I, III, and IV) displays a canonic pentagonal bipyramidal bound cadmium ion which is coordinated by seven oxygen atoms. The cadmium ion located at loop I is coordinated equatorially to Asp-48, Asp-46, Glu-55, and main-chain carbonyl of Lys-50. Axial coordination is achieved by Asp-44 and a water molecule. Loop III equatorial coordination of Cd 2ϩ occurs via the main-chain carbonyls of Thr-142, Asp-140 and Glu-147. Axial interactions are via a water molecule and Asp-138. For loop IV, equatorial coordination to the metal occurs via Asp-170, Thr-174, Glu-178, and Glu-181. The side chain of Asp-172 and the main-chain carbonyl of Lys-176 provide the axial coordination. The average distance between the cadmium and the pairing oxygen atoms are 2.38 Å for site I, 2.46 Å for site III, and 2.53 Å for site IV. These distances are similar to EFhands bound to Ca 2ϩ ions rather than Mg 2ϩ ions (19,28). Like other Ca 2ϩ -binding photoproteins, EF-hand II is not capable of binding divalent metal ions. A fourth nonfunctional cadmium ion is located between symmetry-related molecules, probably aiding crystallization.

QM/MM studies on the coelenterazine bound mnemiopsin1
To gain insights into how mnemiopsin1 binds coelenterazine, a complete model was calculated by adding residues 194 -206 that were not observed in the electron density map of apo-mnemiopsin1. The missing residues were added using the mitrocomin (PDB code 4NQG) and berovin (PDB code 5BPJ) structures as templates by the I-Tasser server. The best model with C-score of Ϫ1.35, TM score of 0.55 Ϯ 0.15, and RMSD of 8.4 Ϯ 4.5 Å was refined by molecular dynamics (MD) simulation.
The binding orientation of coelenterazine in mnemiopsin1 was predicted by molecular docking using AutoDock (29 -32). The results showed that coelenterazine docks into the mnemi-opsin1 and berovin structures with a similar binding mode. To investigate the dynamic stability of the resultant docked structure, a 30-ns MD simulation was performed. The RMSD was monitored for backbone C␣ atoms of mnemiopsin1 relative to the starting structure during the whole MD run (Fig. 4). The backbone trajectories of the structure were stable, reaching equilibrium after the first 10 ns of the simulation. The final model was used for subsequent computational analyses. A comparison of the holo-mnemiopsin1 and holo-berovin structures reveals identical amino acid residues in the binding cavity. Asn-105, Lys-88, and Trp-101 residues interact with the oxygen of the 6--hydroxybenzyl group of coelenterazine. The hydroxyl group of Tyr-202 has a predicted strong hydrogen bond to the 2-hydroperoxy group of coelenterazine. Arg-39, Asp-156, and Trp-190 residues were also observed close to the 2-hydroperoxy group. Furthermore, the 2--hydroxybenzyl group of coelenterazine is stabilized through hydrogen bonding with Ser-128 side chain.

Reaction mechanism of the bioluminescent protein mnemiopsin1
QM/MM calculations on mnemiopsin1 were performed similar to that described in the recent study on berovin (25). In the previous study, we only observed the presence of Asp-158 -Arg-41-Tyr-204 triad around the 2-hydroperoxy group (25). As shown in Fig. 5, additional insight comes from consideration of the QM/MM calculations on the mnemiopsin1 structure in which transfer of a proton from the hydroperoxide group of coelenterazine to Tyr-202 is observed. A nucleophilic attack of the peroxy anion of coelenterazine onto its C 3 position results in a dioxetane intermediate, so that the distance of C 3 -O 34 decreases from 3.64 to 1.57 Å, whereas the distance of N 4 -C 3 is increased from 1.41 to 1.55 Å (Table 1).

QM/MM studies on the coelenteramide-bound mnemiopsin1
The coelenteramide was oriented into the mnemiopsin1 structure in the same orientation as we previously docked coelenterazine. The resultant complex was refined by MD simulation and used for QM/MM analyses. The hydroxyl group of Tyr-131 is 7.5 Å from N1 of 2-hydroperoxycoelenterazine. In addition, in the apo-mnemiopsin1 structure the Tyr-131 hydroxyl group forms a hydrogen bond to a water molecule (Fig. S2), both of which are far from the N1 atom (Fig. 6A). Based on the bioluminescence mechanism in the cnidarian family, the amide anion would be protonated before it has a chance for radiative emission. Thus, in the QM/MM calculations of Ca 2ϩ -discharged mnemiopsin1, this water molecule was manually protonated to form a hydronium ion. In an optimized structure, this water molecule forms a strong hydrogen bond with the amide N atom of coelenteramide (Fig. 6B). As shown in Fig. 6C, at the end of calculations, the proton was

Reaction mechanism of the bioluminescent protein mnemiopsin1
transferred from of water molecule to the amide N atom of coelenteramide.

Discussion
The structure of mnemiopsin1 bound to cadmium ions reveals a similar topology to that of berovin complexed with calcium (18) and magnesium ions (19). Based on the structure of aequorin bound to coelenterazine, the proposed substratebinding site of mnemiopsin1 is located in the core of the protein, surrounded by four EF-hand loops that provide a fairly hydrophobic pocket that contains mostly hydrophobic side chains but some hydrophilic aromatic residues.
The bioluminescence reaction in both photoprotein families of cnidarians and ctenophores appear to be similar. However, the mechanism of reaction initiation and light emission of the ctenophore family was largely unknown until our recent QM/MM studies on berovin, in which we proposed a catalytic mechanism predicting that Asp-158 -Arg-41-Tyr-204 triad, equivalent to the His-175-Tyr-190 -Trp-179 triad (obelin numbering) in the cnidarian family, could shuttle a proton from the 2-hydroperoxy group of coelenterazine to bulk solvent (25). The QM/MM studies on the coelenterazine bound mnemiop-sin1 suggest a break of the N 4 -C 3 bond and decarboxylation of coelenterazine (30) triggers the bioluminescence reaction in the ctenophore family (Fig. 5).
Studies on the bioluminescence mechanism of cnidarian photoproteins show that some intermediates are involved in the oxidative decarboxylation of coelenterazine (8). They suggest that after the release of CO 2 , the primary excited product was an amide anion, which would be expected to be rapidly protonated before it has a chance for radiative emission (8).
There are a number of proton sources for this reaction in the binding cavity of mnemiopsin1. One candidate is Tyr-138, which is already hydrogen-bonded to the N 1 of peroxycoelenterazine, assuming that it does not shift its position when the excited amide anion is formed (8). Although the position of this amino acid in structures of aequorin and obelin is almost iden-tical, its mutation is associated with different outcomes for each protein, probably because of slight differences in the mechanism. For instance, Y132F mutation resulted in the loss of aequorin bioluminescence activity, whereas mutation of its corresponding residue in obelin (Tyr-138) retains 60% of its activity. Under the proposed mechanism, it was assumed that the proton transfer was from Tyr-138 to a dioxetanone intermediate. If this is the case, we would expect the Y138F mutant to be similarly inactive (16). In addition, the crystal structure of obelin after the bioluminescence discharge showed that Tyr-138 is replaced by a water molecule. Thus, it was proposed that the critical function of this water molecule is to catalyze the decarboxylation reaction by protonating the dioxetanone anion (34). This was supported by the structure determination of the Y138F mutant of obelin in coelenterazine and coelenteramidebound states by Natashin et al. (16). In the native obelin structure, the coelenterazine N 1 atom is located at a distance of 2.7 Å from the hydroxyl group of Tyr-138. This residue in turn interacts with the nitrogen atom of His-64 via a water molecule. However, the crystallographic results of the Y138F mutant revealed the lack of the aforementioned water molecule and subsequent hydrogen bond network around the N 1 of 2-hydroperoxycoelenterazine. Additionally, the investigation into the hydrogen-bonding network of the coelenteramide-bound Y138F mutant indicates a water molecule within the cavity that connects N 1 of coelenteramide to His-64 (16). These results are consistent with the hypothesis that proton transfer from His-64 to the nitrogen atom of coelenteramide occurs via a water molecule.
To investigate whether a similar mechanism could occur in the ctenophore family, we examined the coelenteramidebound mnemiopsin1 using QM/MM calculations. The most remarkable finding from these results is the presence a water molecule as a hydrogen bond donor near the amide N of coelenteramide that hydrogen bonds to Tyr-131. Similar to the cnidarian family, a possible function from this water molecule is to catalyze the decarboxylation reaction by protonating the dioxetanone anion, which then yields the neutral coelenteramide as the primary excited product in bioluminescence (Fig. 6). Additional support for this suggestion is found in the mutation study of the Tyr-131 in mnemiopsin1. Structural alignment shows that Tyr-131 of mnemiopsin1 is equivalent to Tyr-132 and Tyr-138 in aequorin and obelin, respectively. The replacement of

Reaction mechanism of the bioluminescent protein mnemiopsin1
Tyr-131 to Phe leads to the loss of activity compared with the WT protein, which could be related to suppression of protonation of the dioxetanone anion through a water molecule (25). Finally, it has been proposed that the triggering step in bioluminescence of the ctenophore photoproteins involves transfer of a proton from the 2-hydroperoxy group of coelenterazine to Tyr-202 and deprotonation of hydroperoxide (Scheme 1, step 1), followed by formation of the dioxetanone anion (Scheme 1, step 2). Our computational studies showed that dioxetanone anion intermediate is unstable in this step, resulting in the release of CO 2 (Scheme 1, step 3). The structural investigations showed that there is a water molecule near amide N 1 atom of coelenteramide. In addition, the protonation of the anion from the water molecule occurs and produces the neutral coelenteramide as the primary excited product in the bioluminescence (Scheme 1, step 4).
In conclusion, we have determined the three-dimensional crystal structure of apo-mnemiopsin1 from the ctenophore family in the cadmium-loaded state. The structure revealed that mnemiopsin1 retains the structural fold characteristic of cnidarian photoproteins despite there being low sequence identity. Our recent QM/MM study on berovin suggested that the Asp-158 -Arg-41-Tyr-204 triad is equivalent to the His-175-Tyr-190 -Trp-179 triad in obelin and could play an important role in triggering bioluminescence. In the present study, a QM/MM study on mnemiopsin1 suggested that an Asp-156 -Arg-39 -Tyr-202 triad creates a hydrogen-bonded network so that the proton is transferred from the 2-hydroperoxy group of coelenterazine to bulk solvent. A QM/MM study on the coelenteramide-bound mnemiopsin1 indicated the presence of a water molecule in close distance to the amide N atom of coelenteramide. Thus, it appears that despite the low sequence identity between two these families, the function of the water molecule is to catalyze the decarboxylation reaction by protonating the dioxetanone anion, as previously suggested for the cnidarian family. Confirmation of this mechanism awaits the determination of a 3D atomic structure of a ctenophore in its holo-form state.

Bacterial strain and growth condition
Mnemiopsin1-encoding gene (GenBank TM accession no. GQ231544) (26) was cloned into pProEX-HTb between the EheI and HindIII restriction sites. Transformed Escherichia coli BL21(DE3)Star cells (Novogen) were grown in Luria-Bertani (LB) medium containing ampicillin to an A 600 of ϳ0.6, and protein expression was induced by the addition of isopropyl-␤-D-thiogalactoside (final concentration, 0.5 mM). The cells were cultured at 27°C for 5 h before being harvested. The cells were centrifuged at 3,000 ϫ g for 30 min at 4°C. Cell pellets were stored at Ϫ20°C.

Purification of mnemiopsin1
Bacteria pellets were resuspended in 500 mM NaCl, 10 mM imidazole, 100 mM Tris, pH 7.5, and were lysed using an EmulsiFlex-C 5 homogenizer (Avestin, Ottawa, Canada) with insoluble cell debris being removed by centrifugation at 38,000 ϫ g at 4°C for 30 min. The soluble fraction was applied to a 5-ml HiTrap HP Sepharose column (GE Healthcare) equilibrated with buffer A (500 mM NaCl, 100 mM Tris, pH 7.5). The column was washed with buffer A for 100 ml. Bound proteins were eluted in a 100 ml of 0 -100% gradient of buffer A to B (500 mM NaCl, 100 mM Tris, 1 M imidazole, pH 7.5).
The buffer was exchanged with 25 mM Hepes, pH 7.7, 100 mM NaCl, 1 mM TCEP using a HiPrep 26/10 desalting column (GE Healthcare) and then concentrated to 4 ml using 10-kDa molecular mass cutoff Ultra-4 centrifugal filters (Merck Millipore). TEV protease was added in a (w/w) ratio of 1 TEV : 100 Mnemiopsin1 and then incubated at 4°C overnight. The protein solution was passed down a second HiTrap HP Sepharose column (pre-equilibrated with buffer A) to remove the cleaved hexahistidine tag from uncleaved protein, as well as the TEV

Reaction mechanism of the bioluminescent protein mnemiopsin1
protease. Cleaved mnemiopsin1 protein appeared in the flowthrough. To ensure the EF-hands of mnemiopsin1 were fully occupied with cadmium ions, concentrated mnemiopsin1 was dialyzed overnight into Cd 2ϩ charging buffer composed of 20 mM Tris, pH 8.0, 150 mM NaCl, and 5 mM CdCl 2 . The protein sample was then applied to a High-load Superdex 75 16/600 gel filtration (GE Healthcare) column, pre-equilibrated with 10 mM Tris, pH 7.5, and 20 mM NaCl. Mnemiopsin1 eluted at a molecular mass corresponding to a monomeric species of 24,500 Da. Fractions were collected and analyzed by SDS-PAGE. Fractions containing mnemiopsin1 were pooled and concentrated to the desired protein concentration.

Crystallization
Initial crystallization trials were carried out using the PEG suite (Qiagen) in 96-well sitting drop vapor diffusion plates (Rigaku). Screens were made by adding 30 l of crystallant to the well and then adding 0.2 l of crystallant to 0.2 l of protein (at 20.5 mg/ml) to form the sitting drops, using a Gryphon robot (Art Robbins Instruments). A single crystal measuring 150 m ϫ 150 m ϫ 100 m was observed after 11 months at 21°C using the crystallant 20% PEG 3350 and 200 mM KCN. The crystal grew as a hexagonal bipyramid that fluoresced under UV light when viewed using a Rigaku Minstrel DT imager. The crystal was flash-cooled prior to data collection by passing it briefly through Paratone-N (Hampton Research) before being plunged directly into liquid nitrogen.

X-ray diffraction, structure determination, and refinement
The crystal formed in the hexagonal space group P6 5 , with unit cell dimensions a ϭ b ϭ 72.61 Å, c ϭ 81.43 Å. One molecule in the asymmetric unit gave a predicted Matthews coefficient of 2.68 Å 3 Da Ϫ1 and a corresponding solvent content of 54.1%. All diffraction data were collected at the Australian Synchrotron, Beamline 3BM2 at 100 K using the program Blue-Ice (35). The data were integrated using XDS (36) and scaled using SCALA (37). Initial phase information was generated by the molecular replacement method with berovin (PDB code 5BPJ) as the search model using the program MolRep (38). ARP/ wARP (39) was used to automatically build an initial model. Successive rounds of model building using COOT (40) and refinement using the program PHENIX (41) were carried out using data between 36.30 and 2.15 Å resolution. Residues 196 -206 were not visible in the electron density and thus were not modeled. The final model has a R factor of 16.1% and R free of 20.3%. Secondary structural elements were calculated using DSSP (42). A summary of the crystallographic data collection and the refinement statistics is given in Table 2. Structure figures were generated using PyMOL (43).

Initial structure preparation and docking studies
The crystal structure of mnemiopsin1 was used as input for the computational studies. Using the I-Tasser server (44 -46), missing residues 196 -206 were added (I-TASSER threads the protein sequence through the PDB structure library and searches for the possible alignments using ab initio modeling for unaligned regions). The resultant model was refined by MD simulation using the AMBER12 software package (47) with ff99SB force field (48). The binding orientation of coelenterazine and coelenteramide were determined using the docking program AutoDock (29 -33). The structures apo-aequorin and holo-aequorin show a displacement in the C-terminal helix (23). This movement of the C-terminal helix in apo-mnemiop-sin1 was also carried out, thus allowing the coelenterazine to penetrate into the cavity.

Molecular dynamics simulation
The final best-ranked complexes were minimized and used for MD simulations. The AMBER12 package (47) was used for preparation of two complexes. The Antechamber module was applied to calculate the AM1-BCC charges of coelenterazine and coelenteramide. Then ff99SB (48,49) and GAFF force fields (50) were used to model the complexes with the addition of sodium ions for neutralizing the systems. The resulting complexes were solvated with TIP3P water molecules in a truncated octahedron periodic box with an 8 Å radius buffer zone of water molecules around the complexes. The systems were neutralized by adding the corresponding number of counterions (Na ϩ ) using the LEaP module. Then the systems were energy minimized for 50,000 cycles, using the steepest descent algorithm together with the conjugate gradient method to remove any bad contacts between atoms. The minimized systems were first gradually heated to 300 K over 100 ps using a Langevin thermostat (51) in a constant condition (NVT). The systems were equilibrated for 100 ps at constant pressure (1 atm) with a 2-fs time step to adjust the periodic box size in a constant condition (NPT). Finally, 30 ns unrestrained production simulation was conducted for each system using an integration time step of 2.0 fs. During the production stage, after every 500 time steps, coordinates were saved and used for energy and structure calculation. The MD simulations were done employing periodic

Reaction mechanism of the bioluminescent protein mnemiopsin1
boundary conditions with a 10 Å cutoff for nonbonded interactions, and long-range electrostatics interactions were carried out adhering to the particle mesh Ewald method (52). The SHAKE algorithm (53) was used to fix all covalent bonds involving hydrogen atoms.

QM/MM simulations
The two QM/MM simulations were performed using the AMBER12 program (47). The QM/MM simulations were carried out using the PM6 method (54 -56) implemented in AMBER12. In the QM/MM simulations, final complex configurations resulting from the classical MD simulations were used as the initial structure. In the first QM/MM, QM part consisted of the coelenterazine and side chain of Tyr-202 so that tyrosine residue was manually deprotonated to form a tyrosinate anion. The QM region of the second QM/MM includes the coelenteramide and H 3 O ϩ . The rest of the protein, water molecules and (Na ϩ , Cl Ϫ ) ions were considered as the MM region. The MM part was defined by AMBER ff99SB force fields (48,49).