NMR-derived Topology of a GFP-photoprotein Energy Transfer Complex*

Förster resonance energy transfer within a protein-protein complex has previously been invoked to explain emission spectral modulation observed in several bioluminescence systems. Here we present a spatial structure of a complex of the Ca2+-regulated photoprotein clytin with its green-fluorescent protein (cgGFP) from the jellyfish Clytia gregaria, and show that it accounts for the bioluminescence properties of this system in vitro. We adopted an indirect approach of combining x-ray crystallography determined structures of the separate proteins, NMR spectroscopy, computational docking, and mutagenesis. Heteronuclear NMR spectroscopy using variously 15N,13C,2H-enriched proteins enabled assignment of backbone resonances of more than 94% of the residues of both proteins. In a mixture of the two proteins at millimolar concentrations, complexation was inferred from perturbations of certain 1H-15N HSQC-resonances, which could be mapped to those residues involved at the interaction site. A docking computation using HADDOCK was employed constrained by the sites of interaction, to deduce an overall spatial structure of the complex. Contacts within the clytin-cgGFP complex and electrostatic complementarity of interaction surfaces argued for a weak protein-protein complex. A weak affinity was also observed by isothermal titration calorimetry (KD = 0.9 mm). Mutation of clytin residues located at the interaction site reduced the degree of protein-protein association concomitant with a loss of effectiveness of cgGFP in color-shifting the bioluminescence. It is suggested that this clytin-cgGFP structure corresponds to the transient complex previously postulated to account for the energy transfer effect of GFP in the bioluminescence of aequorin or Renilla luciferase.

The bioluminescence of many marine coelenterates, wellstudied examples being the jellyfish Aequorea and the seapansy Renilla, involves the interaction of two proteins, a Ca 2ϩ -regulated photoprotein in the jellyfish case, aequorin, and its cognate green-fluorescent protein, Aequorea GFP (1). Addition of Ca 2ϩ to the purified aequorin produces a blue bioluminescence. It was early recognized that, in the jellyfish itself, the in vivo bioluminescence was a green color and after further study, the origin of this green emission was identified as the GFP. A Förster-type resonance energy transfer (FRET) 3 mechanism was invoked to explain how this bioluminescence spectrum is shifted (2). However, the well-known Förster theory requires concentrations of the donor-acceptor partners in the millimolar range, whereas in some bioluminescence systems, e.g. from the sea pansy Renilla and also the jellyfish Clytia subject herein, the GFP effect on the in vitro bioluminescence is observed at micromolar concentrations (3). Clearly the bioluminescence interaction has to involve formation of a complex and, in the case of Renilla, the formation of a luciferase-GFP complex has been shown (3).
In this work, we have determined by x-ray crystallography the spatial structures of the recombinant Ca 2ϩ -regulated photoprotein clytin and Clytia GFP (cgGFP), which were cloned from a single specimen of Clytia gregaria (syn. Phialidium gregarium). Based on the structures, NMR titration experiments were employed to identify the interaction surfaces in a complex of both proteins. For a mixture of clytin and cgGFP at millimolar concentration, 1 H- 15 N HSQC experiments revealed perturbation of chemical shifts of the separate proteins, which could be mapped to particular residues being affected by complexation. The NMR experiments also indicated that the association was weak but from knowledge of the interaction surface, computational docking was employed to propose an overall three-dimensional structure of the clytin-cgGFP complex.

EXPERIMENTAL PROCEDURES
Molecular Biology-Cloning of the clytin and Clytia GFP genes from a single specimen of the jellyfish C. gregaria, expression, purification, and characterization of recombinant clytin and cgGFP have been published (4). Site-directed and truncation mutagenesis of clytin were done on the template p22-Cl3 E. coli expression plasmid carrying the apo-clytin gene of wild-type C. gregaria. Mutations resulting in the amino acid change: K11A, K13A, N15A, N109A, or N188A were carried out using the QuickChange site-directed mutagenesis kit (Stratagene). N-terminal-truncated clytin mutants 5A (sequence starts from Ala-5) and 10V were amplified by PCR. The plasmids harboring mutations were verified by DNA sequencing.
Crystallography-Crystals of clytin grew at 16°C within 1 week to a size of 50 ϫ 50 ϫ 300 m. The crystallization droplet was set up using Mosquito crystallization robot (TTP Labtech) and contained equal volumes of protein (15 mg/ml) and reservoir solution (20% PEG-3350, 0.2 M NaH 2 PO 4 , pH 8.8) derived from the Peg/Ion crystallization screen (Hampton Research). The crystal was flash-frozen in liquid nitrogen. Native diffraction data were indexed and scaled to 1.9 Å resolution using HKL2000. The space group of clytin was C222 1 with unit cell dimensions (Å), a ϭ 43.39, b ϭ 68.93, c ϭ 115.35. Phases were determined by molecular replacement with PHASER (5), using the structure of obelin (PDB code 1JFO) as a search model. The final models were refined with PHENIX (6). Manual adjustments to the model were done using COOT (7). The quality of the final model was validated with MOLPROBITY (8). The detailed data processing and refinement statistics are shown in Table 1.
Crystals of cgGFP grew at 4°C within 5 days to a size of 200 ϫ 200 ϫ 250 m. The crystallization droplet contained equal volumes of protein (9 mg/ml) and reservoir solution (2 M ammonium sulfate, 0.1 M sodium citrate, pH 5.5) derived from the Wizard I crystallization screen (Emerald Biosystems). The crystal was flash-frozen in liquid nitrogen. Native diffraction data were indexed and scaled to 1.55 Å resolution using HKL2000. The space group of cgGFP was I2 1 2 1 2 1 with unit cell dimensions (Å), a ϭ 53.09, b ϭ 91.45, c ϭ 110.61. Phases were determined by molecular replacement with MOLREP (9) using GFP from Aequorea victoria as a search model (PDB code 1EMA). Iterative model validation, rebuilding and refinement, were carried out with MOLPROBITY (8), XFIT (10), and REFMAC5 (11), respectively. The detailed data processing and refinement statistics on cgGFP are shown in Table 1.
Protein concentrations were determined by the dye-binding method of Bradford (12) using an assay kit (Bio-Rad) and bovine serum albumin as a standard. On this basis extinction coefficients for clytin (⑀ 280 ϭ 65,200 M Ϫ1 cm Ϫ1 ) and cgGFP (⑀ 485 ϭ 64,000 M Ϫ1 cm Ϫ1 ) were calculated and subsequently, protein concentrations were determined by absorbance.
NMR Sample Preparation-Uniformly 15 N-or 15 N-, 13 Clabeled clytin and cgGFP were obtained from the cells grown in M9 minimal medium containing 15 NH 4 Cl or additional [ 13 C]glucose. To acquire 15 N-, 13 C-, and 60% 2 H-labeled cgGFP, 99.8% 2 H 2 O-based M9 was used. To exchange amide group deuterium to hydrogen, triple-labeled cgGFP was subjected to reversible denaturation in 6 M guanidine-HCl for 10 min, followed by 20-fold droplet dilution in PBS at 25°C and overnight incubation at 4°C. Activity of cgGFP was restored upon refolding as indicated by absorbance and fluorescence spectra and bioluminescence color shift assay with clytin.
Chemical shift perturbation analyses were performed at 293 K for clytin and 298 K for cgGFP by monitoring the twodimensional 1 H-15 N HSQC spectra of titrated proteins. Unlabeled clytin WT or mutants K11A, K13A, N109A, N188A, and 5A (0.5 mM, 1 mM), were added to 15 N-or 15 N-, 2 H-labeled cgGFP (0.4 mM). Alternatively, unlabeled cgGFP (0.3 mM, 0.6 mM) was added to 15 N-labeled clytin (0.2 mM). The amide hydrogen and nitrogen chemical shift changes were calculated according to Equation 1, where ⌬␦ N and ⌬␦ H represent the changes in the amide nitrogen and proton chemical shifts (in parts per million), respectively.
Calculations of the Clytin-cgGFP Complex Structure-The computational structures of the clytin-cgGFP complex were generated with HADDOCK2.0 (17,18) in combination with CNS (19). Ambiguous interaction restraints (AIRs) were generated for both clytin and cgGFP based on chemical shift perturbation studies (Table 2) as described (17). The starting structures were the monomer cgGFP (PDB code 2HPW) and the clytin (PDB code 3KPX) with the manually added 2-8 N-terminal segment, which is absent in the deposited structure and was defined as fully flexible during docking. The standard HADDOCK protocol was used. For the rigid-body energy minimization, 1,000 structures were generated, with the 200 lowest energy solutions used for subsequent semiflexible simulated annealing and water refinement. Resulting structures were sorted according to intermolecular energy and clustered using a 7.5 Å cut-off criterion. Subsequent cluster analysis was performed within a 4.0 Å cut-off criterion. The 10 lowest energy solutions were taken to represent the structure of the complex (supplemental PDB files).
Bioluminescence Assay-Bioluminescence spectra of clytin and clytin with cgGFP, were measured with a Varioskan Flash Spectrofluorimeter (Thermo Scientific). All measurements were carried out at 25°C. Luminescence was triggered by injection of 7 l of 40 mM CaCl 2 into the wells containing 150 l of isolated clytin (final concentration is in the 0.4 -1.5 M range) or clytin mixed with cgGFP (final concentrations from 0 to 9.7 M) in 2 mM EDTA, 10 mM NaCl, 20 mM Tris-HCl, pH 7.0 buffer. Emission spectra were fully corrected for instrumental spectral sensitivity with the computer program supplied with the instrument, and also for bioluminescence intensity decay over the time for the spectral scan. All spectra were the average of three measurements. The energy transfer efficiency coefficient (K ET ) for clytin and clytin mutants was determined by plotting the I 500 /I 470 ratio versus total concentration of cgGFP, where I 500 and I 470 are bioluminescence intensities at 500 nm and 470 nm, respectively. The slope of the linear regression fitted data was taken as the K ET value.
Isothermal Titration Calorimetry-Isothermal titration calorimetry measurements were performed on an ITC 200 calorimeter (Microcal Inc., Northampton, MA). All experiments were carried out at 25°C in 20 mM Tris-HCl, 10 mM NaCl, 2 mM EDTA, pH 7.0. The reactant (0.1 mM clytin) was placed in the 200-l sample chamber and cgGFP (4.68 mM for monomer) in the syringe was added with 20 successive additions of 2 l for 4 s (with an initial injection of 0.5 l). The interval between each injection lasted 150 s. The peaks generated were corrected for cgGFP heat of dilution and integrated using the ORIGIN software (Microcal Inc) by plotting the values in microcalories against the ratio of total moles of injectant, monomer cgGFP, to reactant clytin, within the cell. Data were fit using a 1:1 clytin:cgGFP monomer binding model.

JOURNAL OF BIOLOGICAL CHEMISTRY 40893
no electron density for the Thr 2 -Ala 9 region implying significant structural flexibility of the N terminus of clytin. cgGFP forms the well-known barrel structure built of 11 ␤ strands (S1-S11) with the chromophore buried inside (24, 25) (Fig.  2B). Despite a low sequence identity (Fig. 1B), the structure of cgGFP highly resembles GFPs from Aequorea (26) (RMSD 1.04 Å, sequence identity 41%), and Renilla (27) (RMSD 1.84 Å, sequence identity 19%). The cgGFP homodimer can be generated from the cgGFP monomer in the crystallographic asymmetric unit by applying a 2-fold symmetry axis. The dimerized form of cgGFP is evident from a mass of 52 kDa determined by analytical ultracentrifugation, close to that of the natural cgGFP (57 kDa) (28,29), consistent with the 27 kDa monomer mass determined by SDS-PAGE. The cgGFP monomer buries 1,370 Å 2 (13% of the total surface area) in the dimer interface. The dashed lines represent the one standard deviation (gray) and two standard deviations (blue, magenta) cut-offs. Residues, whose cross-peak shifted more than one or two standard deviations above the average, are mapped on the spatial structures in gray and blue for clytin, and in gray and magenta for cgGFP, respectively. Residues of clytin (mostly N-terminal) with significant peak intensity perturbations, are also mapped in blue.
Mapping the Clytin-cgGFP Interface-The backbone NMR resonance assignments of both proteins, which are the basis for our chemical shift perturbation mapping, were obtained by heteronuclear NMR spectroscopy using 15 N, 13 C-labeled clytin and 15 N, 13 C, 2 H-labeled cgGFP. The backbone resonances of more than 94% residues were assigned for both proteins (supplemental Fig. S1). Lack of assignment for residues Lys 159 -Asn 165 located at the end of ␣-helix G and the EFhand IV loop of clytin might be due to the line broadening caused by the structural flexibility in this region.
A similar effect was shown in the NMR study of aequorin (30). Line broadening for the Thr 2 -Lys 7 residues of clytin is in agreement with x-ray crystallography verified structural flexibility of the N terminus. Residues of cgGFP lacking assignments mostly belong to the loop regions comprising the top and bottom of the GFP barrel.
We were unable to crystallize any clytin-cgGFP complex either from a mixture of proteins or for a covalently crosslinked complex under around 300 crystallization conditions available from commercial kits (Hampton Research, Emerald Biosystems). Instead evidence for protein-protein association was inferred from two 1 H-15 N HSQC titration experiments, first with 15 N-labeled clytin and unlabeled cgGFP, then vice versa. NMR titration could not be saturated because of the limited solubility of both proteins and the weak interaction between them, and therefore the K eq could not be derived from the NMR titration data (supplemental Fig. S5). Protein concentrations were 0.3-1 mM, which implies an equilibrium constant K eq in the mM range. Nevertheless, a number of peaks in the HSQC spectra showed chemical shift and/or peak intensity perturbations from which the binding surfaces of clytin and cgGFP were mapped (31-34) (Fig. 2). The perturbed peaks show concentration-dependent chemical shift changes, which indicates a fast exchange on the NMR chemical shift time scale (supplemental Fig. S5).
When mapping the perturbed residues on the structures of clytin and cgGFP, these residues are well clustered on the surfaces of both proteins, which implies specific interactions between the two proteins (Fig. 2). The remaining chemical shift changes upon the titration are relatively small (Fig. 2, C & F), with maximum chemical shift perturbation values for both proteins not exceeding 0.08 ppm, which indicates no large structural rearrangement during the complex formation. For clytin (Fig. 2, B & C), the perturbations are assigned to residues within three segments: 9 -17 at the N terminus, 100 -109 in the ␣-helix D, and at the C terminus, 180 -193, as well as some adjacent residues. For cgGFP (Fig. 2, E & F), the perturbations are identified as belonging to residues 55-65, mainly in segments of loop S3-S4 and the central ␣-helix, 132-149 of the longest loop S6 -S7 covering the GFP barrel from the top, and 209 -218 of the remarkably acidic loop S10 -S11, as well as some adjacent residues. For cgGFP, however, chemical shifts are also detected for residues buried inside the protein molecule. These are segment 60 -66, "connecting" the interaction surface with the chromophore, and Ser 146 and His 149 , which form hydrogen bonds with Tyr 69 of the chromophore through the water molecule. The cgGFP contact surface is less uniform and narrower than that of cly-tin, which may be explained by overlapping or lack of assignment for some of the resonances in the main interacting regions: excluding prolines these are Ser 133 , Asn 134 , Ile 137 , Arg 141 , Tyr 144 of the loop S6 -S7 and Asp 215 of the loop S10 -S11. On the clytin-cgGFP complex structure these residues are found buried in the protein-protein contact region. cgGFP forms a homodimer in solution, and there is no overlap between the cgGFP dimer interface and the clytin-binding patches.
We employed isothermal titration calorimetry (ITC) to obtain an independent assessment of the interaction of clytin with cgGFP. ITC has been shown to be capable of recovering weak binding constants although the accuracy of thermodynamic parameters is problematic compared with protein-protein affinities in the micromolar regime (57). The net heats of interaction of cgGFP as titrant added to clytin are shown in Fig. 7. The experiment and data analysis take into account the precautions suggested by Turnbull et al. (57). Above molar ratio about 3.0 there is large uncertainty due to the mixing signal being hardly different from the control dilution heat of cgGFP alone. The full line is an unweighted fit to a binding model with fixed stoichiometry (n ϭ 1.0) using a 1 clytin:1 cgGFP monomer binding model, and the derived affinity constant is K D ϭ 0.90 Ϯ 0.07 mM. A model with 1 clytin:1 cgGFP dimer yields almost the same result. This affinity is consistent with the millimolar range estimate for K eq from the NMR titration experiment. Interestingly, clytin and cgGFP could be separately concentrated up to 50 mg/ml and 160 mg/ml, respectively, while the mixture of proteins always showed precipitation at concentrations higher than 30 mg/ml under the same conditions. Because the clytin-cgGFP complex buries predominantly hydrophilic residues, (discussed below), the decreased solubility of the protein mixture is more evidence of complexation at these concentrations.
Computational Docking of the Clytin-cgGFP Complex-Because of weak interaction between clytin and cgGFP under NMR conditions accompanied by the large molecular size of the proteins, it was not possible to derive accurate spatial restraints from measuring intermolecular NOEs for the complex. Therefore we use a docking approach, named HADDOCK2.0 (17,18), which relies on ambiguous restraints originating from initial NMR chemical shift perturbation data ( Table 2) to derive an accurate model of a protein-protein complex (31,(35)(36)(37). A feature of HADDOCK to introduce backbone flexibility was applied to the Thr 2 -Ala 9 N-terminal region of clytin. Fig. 3 is the computational result for the cly- tin-cgGFP complex dimer, initially based on the identified interaction surfaces which were highly suitable for structure calculation using HADDOCK. The family of final structures had the lowest intermolecular energy (Ϫ382.81 kcal/mol) and the highest buried surface area (1,913 Å 2 ) (supplemental Table S1). The average pairwise RMSD in this cluster is 1.29 Ϯ 0.48 Å for backbone atoms. cgGFP forms a homodimer in solution, and the clytin-binding patch on each cgGFP monomer would be distant from each other, thus one cgGFP monomer could accommodate one clytin. As the HSQC chemical shift perturbations and the ITC data both indicate an interaction constant in the millimolar range, this is considered as very weak. The contact surface indeed reveals a relatively low number of hydrophobic contacts and hydrogen bonds, although in total the complex buries 1,913 Ϯ 87 Å 2 of surface area, which is average for a protein-protein complex (38).
Calculation of the electrostatic potential of clytin and cg-GFP reveals that the interfaces have remarkable charge complementarity, which might assist the complex stabilization (Fig. 4). The clytin ␣-helix D and the proximal N terminus carry the positive charge and occupy, also with good shape complementarity, the negatively charged gutter on the top of the cgGFP barrel formed by the S3-S4, the distal part of the S6 -S7, and the S10 -S11, loops. The S10 -S11 loop is strongly acidic and appears as the least structured region of the cgGFP molecule. This may enable it to adjust for best fit to the clytin interface. In the 10 lowest free energy structures (supplemental Fig. S2 and PDB files), contacts in this region (Fig. 4) are prevalently formed by lysine residues of clytin (Lys 11 , Leu 12 , Lys 13 , Thr 14 , Lys 100 , Lys 104 ) and aspartic and glutamic residues of cgGFP (Asp 55 , Lys 210 , Asp 211 , Pro 212 , Asp 213 , Asp 214 , Asp 215 , Glu 216 ). Also Gln 108 of clytin approaches Phe 210 of cgGFP, and Asn 109 of clytin lies adjacent to His 145 and Tyr 144 of cgGFP. As expected for electrostatic in-teractions, the clytin-cgGFP complex formation should be considerably sensitive to ionic strength (Fig. 6). These features of charge complementarity of interfaces together with a low binding affinity, are largely found among transient complexes of various proteins, the well studied examples being redox proteins and Ras or Rap with their signaling effectors (39 -41).
Another major interacting region is moderately polar and comes from overlapping of the surface accessible region of the S6 -S7 loop of cgGFP (residues Ser 133 , Asn 134 , Leu 138 , Gly 139 , Met 140 , Arg 141 ) and some adjacent residues (Met 173 , Met 174 , Gly 174 ) by the distal part of the N terminus of clytin (Thr 2 , Asp 3 , Thr 4 , Ala 5 , Ser 6 , Lys 7 , Tyr 8 , Ala 9 , Val 10 ). The exact contacts in this region are highly variable because of flexibility of the residues 2-9 of clytin, which together with mutagenesis data discussed below implies a less important role for the N terminus of clytin in binding cgGFP compared with the charge complementarity region (Fig. 4). Superimposition of the 10 best structures of the complex demonstrates flexibility of structural elements comprising the interface (RMSD for Thr 2 -Ala 9 region of clytin 4.51 Ϯ 1.57 Å, RMSD for the S10 -S11 loop of cgGFP 1.13 Ϯ 0.21 Å) (supplemental Fig. S2).
Clytin Mutants-Site-directed and N-terminal truncation mutagenesis of clytin was introduced to assess the contribution of the residues found at the interface to the degree of protein association and to determine if there was a correlation with the effectiveness of cgGFP in producing the well-known bioluminescence color shift in the reaction. These mutants of clytin were K11A, K13A, N15A, N109A, N188A, and two Nterminal truncations (5A, 10V). All substitutions had negligible effect on the bioluminescence properties of clytin, implying no significant rearrangements in spatial structures of clytin mutants. Fig. 5 shows the bioluminescence spectral titration with cgGFP for one of the point mutants compared with native clytin. The cgGFP effectiveness in producing a color shift was measured by an interaction constant, K ET , and shown to decrease for all the mutants (supplemental Fig. S4) and most significantly for the substitution K11A (Fig. 5A). The degree of cgGFP HSQC chemical shift perturbations, obtained upon titration with clytin mutants, was used to calculate a quantitative parameter of protein association, named K CSD . Values of chemical shift perturbation after subtracting the average perturbation plus one standard deviation were summed to give the K CSD value in ppm units. Peak tables of cgGFP upon titration with clytin mutants are shown in supplemental Fig. S4 and for K11A mutant in Fig. 4B. The apparent association constant K CSD correlates well with K ET , all mutants show a decrease in K ET with a lessening of the degree of association (K CSD ). Substitutions K11A, K13A, and the 10V truncation, had the strongest effect, reducing K ET up to 4-fold from clytin along with a strong reduction in binding affinity. These positions are of interest because of their contribution to the surface charge complementarity of the interacting proteins. On the other hand it indicates the important role of the  charge complementarity region compared with the flexible N-terminal segment, because the effects of K11A and K13A substitutions are comparable to that of deleting the first 9 N-terminal residues of clytin. It also implies a minimal role of any small structural rearrangement of clytin mutants in affecting binding to cgGFP. Substitutions N109A, N15A, and N188A had strong, moderate and the smallest effect, respectively.

DISCUSSION
Energy transfer or the bioluminescence color shift on the addition of GFP, has previously received detailed study for two bioluminescence systems, that of aequorin and of Renilla luciferase (3, 42, 44 -46). The mechanism has been proposed to be by FRET within a transient protein-protein complex. According to the well-known FRET equation, the probability or rate of energy transfer, from the excited donor to the acceptor, depends on several parameters, most critical being the donor-acceptor separation and the spectral overlap of donor fluorescence and acceptor absorption. A convenient measure is the "Förster separation" where the probability of the donor radiative S 1 3 S 0 transition equals the probability of energy transfer populating the acceptor S 1 state; in almost all cases this distance is less than 10 nm. This means that for the partners randomly distributed in free solution, they need to be in the millimolar concentration range. The bioluminescence color shifts however, are observed at micromolar protein concentrations so for FRET to be feasible the donor-acceptor separation must be constrained within a protein-protein complex.
For the Renilla luciferase bioluminescence in particular, the addition of Renilla GFP at micromolar concentrations, not only produced the green color shift but enhanced the bioluminescence quantum yield about three times. This is conclusive evidence for FRET indicating that efficient excited state coupling in the transient complex competes with both radiative and non-radiative deactivation pathways of the primary excited S 1 state formed by the reaction on the luciferase. A stable complex was not observed by direct methods, chromatography, fluorescence anisotropy, at these micromolar concentrations. However, using the Hummell-Dryer chromatographic method, Ward and Cormier (45) reported the presence of a Renilla luciferase-Renilla GFP complex. Further evidence that such a complex must be involved for the bioluminescence shift was that the energy transfer was specific for the type of GFP, it occurred with GFPs from other species of Renilla but not from GFPs of more distantly related organisms. Also, the shift effect was negated by amino acid modification in the GFP and by higher ionic strength in the buffer (Ͼ100 mM) (Fig. 6).
A complex has also been reported for the aequorin-Aequorea GFP bioluminescence using the Hummel-Dryer method (47). In that case no bioluminescence quantum yield increase accompanying the energy transfer was observed (42) as also the case here for the clytin bioluminescence in Fig. 5A. Morise et al. (42) however, demonstrated that energy transfer was significantly enhanced in a suspension of DEAE particles on which the aequorin and Aequorea GFP had been co-ad-sorbed, presumably bringing the two partners to proximity, but the color shift was also observed to an unrelated acceptor, FMN, meaning that it was nonspecific.
The observations on the clytin bioluminescence system reported here bear similarity to these earlier reports. The clytin bioluminescence spectrum is shifted to the fluorescence of cgGFP by only micromolar concentrations of cgGFP, the effect is diminished by modification of amino acid residues in the clytin, which otherwise affect no change in the clytin bioluminescence properties, the cgGFP shift is an order of magnitude less effective using the distantly related photoprotein obelin, even though with this pair, the spectral overlap is significantly higher (4), and the K ET is reduced but not eliminated at increased ionic strength.
Additional similarity to earlier reports was that no clytin-cgGFP interaction in the micromolar range could be detected by the methods of fluorescence anisotropy, analytical ultracentrifugation, or plasmon resonance (results not shown). However, in contrast to the cases of Renilla and aequorin just mentioned, Markova et al. (4) recently observed no complex by Hummel-Dryer chromatography using a starting concentration ten times higher than Ward and Cormier used for their Renilla experiment. Altogether, we estimate here a 0.9 mM value for the clytin-cgGFP affinity constant, consistent with the weak K eq in the mM range inferred from the NMR perturbations.
Although the computational model in Fig. 3 needs to be interpreted with appropriate reservation, we point out that the spatial arrangement of the donor and acceptor makes it attractive to consider this complex as the functional biolumi- nescence unit in vitro. There is a very favorable spectral overlap, 1.3 ϫ 10 Ϫ13 M Ϫ1 cm 3 , between the bioluminescence from clytin, maximum 470 nm, and the absorption of cgGFP, having a monomer extinction coefficient of 64,000 M Ϫ1 cm Ϫ1 at 485 nm (4). Combined with the 45 Å separation of the donor and acceptor in the structure of the complex (Fig. 3), and the fact that the cgGFP will be dimerized in the complex, the electronic transitions are very strongly coupled. The energy from the bioluminescence reaction of the clytin will be quantitatively deposited into the excited state of the acceptor, the cg-GFP. However, as the protein-protein complex is weak with a dissociation constant (K eq ) in the millimolar range according to the NMR chemical shift and ITC methods (Fig. 7), the mechanism by which added cgGFP at only micromolar concentrations is able to shift the bioluminescence toward the fluorescence of cgGFP, remains to be established.
The computational structure of the clytin-cgGFP complex resembles features of a weak protein-protein complex predominantly governed by electrostatic forces, with a low number of total intermolecular contacts (39 -41). For a weak protein interaction the relatively high value of the clytin-cgGFP buried surface area (1,913 Å) derives from the impact of the distal (flexible) part of the clytin N terminus interacting with the top cgGFP barrel loops which together account for 30% of the total buried surface. However, intermolecular contacts in this region are minimal and the position of the clytin N terminus itself is highly variable among the best batch of structures, which implies a less significant impact of the distal part of the clytin N terminus compared with its proximal part and the ␣-helix D carrying the positive charge. This conclusion is supported by mutagenesis of clytin where we observe that deletion of the flexible part of the N terminus has the same effect on complex affinity and cgGFP color shift efficiency as the single substitutions K11A and K13A. These substitutions evidently affect electrostatics similarly to the high ionic strength conditions.
The question arises as to the physiological relevance of this clytin-cgGFP computational structure in Fig. 3. The photocytes of the jellyfish Aequorea and Clytia can be assumed to be the same, contain concentrations of the bioluminescence proteins estimated to be in the millimolar range (42,47), similar to the concentrations required to form the complex detected by the NMR and ITC experiments. The in vivo bioluminescence spectra of several animals or their tissue samples, reveal nearly exact correspondence to the fluorescence of GFP, i.e. no contribution from the blue emission implying near 100% FRET efficiency (50,52,53). This demands that the origin of the emission is from a complex where the donor and acceptor have restricted separation and orientation. The inhibition of the GFP shift at increased salt concentration is consistent with electrostatic forces at the proteinprotein interface driving the clytin-cgGFP complexation. This would argue against this same spatial structure existing in vivo if within the photocytes, the ionic strength approaches that of sea-water, or is even as low as that characteristic of eukaryotic cells, 100 -150 mM because of potassium ions. On the other hand, several bioluminescent organisms are found to contain their bioluminescence systems within membrane enclosed vesicles, "lumisomes" in Renilla (54) and "scintillons" in the dinoflagellates (43). Such vesicles apparently modulate the intracellular environment for the benefit of the bioluminescence function (50). Because cgGFP itself is a tight dimer it is probable that in vivo the clytin-cgGFP complex is a heterotetramer. It should be noted that this supposition was advanced for in vivo aequorin-Aequorea GFP complex (51).
For a heterotetrameric complex of this size, Ͼ100 kDa, and weakly interacting, there is little prospect that further NMR experiments will yield unambiguous distance constraints for model refinement. Whether the spatial structure of the in vivo complex relates to that determined here at low ionic strength, hopefully will be proven by crystallography, although for a weak protein-protein complex this methodology presents its own set of impediments.