The 1.5-Å Resolution Crystal Structure of Bacterial Luciferase in Low Salt Conditions*

Bacterial luciferase is a flavin monooxygenase that catalyzes the oxidation of a long-chain aldehyde and releases energy in the form of visible light. A new crystal form of luciferase cloned from Vibrio harveyi has been grown under low-salt concentrations, which diffract x- rays beyond 1.5-Å resolution. The x-ray structure of bacterial luciferase has been refined to a conventional R - factor of 18.2% for all recorded synchrotron data between 30.0 and 1.50-Å resolution. Bacterial luciferase is an (cid:97) - (cid:98) heterodimer, and the individual subunits fold into a single domain ( (cid:98) / (cid:97) ) 8 barrel. The high resolution structure reveals a non-prolyl cis peptide bond that forms between Ala 74 and Ala 75 in the (cid:97) subunit near the putative active site. This cis peptide bond may have functional significance for creating a cavity at the active site. Bacterial luciferase employs reduced flavin as a substrate rather than a cofactor. The structure presented was determined in the absence of substrates. A comparison of the structural similarities between luciferase and a nonfluorescent flavoprotein, which is ex- pressed in the lux operon of one genus of bioluminescent bacteria, suggests that the two proteins originated from a common ancestor. However, the flavin binding sites of the nonfluorescent protein are likely not repre- sentative of the flavin binding site on luciferase. The structure presented here will furnish a detailed molec- ular model for all bacterial luciferases.

Bacterial luciferase is a flavin monooxygenase that catalyzes the oxidation of a long-chain aldehyde and releases energy in the form of visible light. A new crystal form of luciferase cloned from Vibrio harveyi has been grown under low-salt concentrations, which diffract xrays beyond 1.5-Å resolution. The x-ray structure of bacterial luciferase has been refined to a conventional Rfactor of 18.2% for all recorded synchrotron data between 30.0 and 1.50-Å resolution. Bacterial luciferase is an ␣-␤ heterodimer, and the individual subunits fold into a single domain (␤/␣) 8 barrel. The high resolution structure reveals a non-prolyl cis peptide bond that forms between Ala 74 and Ala 75 in the ␣ subunit near the putative active site. This cis peptide bond may have functional significance for creating a cavity at the active site. Bacterial luciferase employs reduced flavin as a substrate rather than a cofactor. The structure presented was determined in the absence of substrates. A comparison of the structural similarities between luciferase and a nonfluorescent flavoprotein, which is expressed in the lux operon of one genus of bioluminescent bacteria, suggests that the two proteins originated from a common ancestor. However, the flavin binding sites of the nonfluorescent protein are likely not representative of the flavin binding site on luciferase. The structure presented here will furnish a detailed molecular model for all bacterial luciferases.
Living organisms that radiate light have been captivating people throughout the ages. Bioluminescent organisms such as fireflies, glowworms, mushrooms, fish, or bacteria represent a diverse range of species, which are widely dispersed in nature (1,2). The enzymes that catalyze the bioluminescence reactions are named luciferases, and in most cases, their substrates are designated luciferins. These enzymes comprise a large evolutionarily diverse group, and the chemistry they catalyze is quite varied. Indeed, the only common factors of these enzymes is the requirement of O 2 , which was first established by Robert Boyle (3) more than 3 centuries ago. Today, it is known that all luciferase reactions are oxidative processes that convert a substrate to an electronically excited intermediate. Light emission occurs when the excited-state intermediate reverts back to the ground state resulting in the final product.
Luminous bacteria are the most abundant and widely distributed of all bioluminescent organisms and are found in marine, freshwater, and terrestrial environments. Bacterial luciferase has been studied extensively and is the best understood of all luciferases. The luciferase of luminous bacteria is a flavin monooxygenase. Bacterial luciferase is an uncommon flavoprotein in that it employs reduced flavin as a substrate rather than a tightly bound cofactor. The enzyme catalyzes the reaction of FMNH 2 , O 2 , and a long-chain aliphatic aldehyde to yield FMN, the aliphatic carboxylic acid and blue-green light. All bacterial luciferases studied so far appear to be homologous, and all catalyze the same reaction: FMNH 2 ϩ O 2 ϩ RCHO 3 FMN ϩ RCOOH ϩ H 2 O ϩ h͑ max ϭ 490 nm͒ REACTION 1 The reaction proceeds through a series of intermediates leading to the formation of a C4a hydroxyflavin (for review see Ref. 4). Light emission apparently occurs from this hydroxyflavin, which dehydrates to yield FMN. Bacterial luciferase is a heterodimeric enzyme of 77 kDa, composed of ␣ and ␤ subunits with molecular masses of 40 and 37 kDa, and in the case of Vibrio harveyi, 355 and 324 residues, respectively. The two polypeptides, encoded on adjacent genes, luxA and luxB in the lux operon, display sequence homology and appear to have arisen by gene duplication (4). There is a single active center in the luciferase heterodimer that resides on the ␣ subunit (5) and binds one reduced flavin molecule (6,7). The role of the ␤ subunit is not clear at this time but is essential for a high quantum yield reaction (8). Amino acid sequence alignment between the two subunits reveals that they share 32% sequence identity. The ␣ subunit contains 29 additional amino acid residues inserted between residues 258 and 259 of the ␤ subunit (9,10). This region of the ␣ subunit is known to be sensitive to proteolytic digestion in the absence of substrates (11,12). A single proteolytic cleavage in the region of residues 274 -291 in the ␣ subunit inactivates the enzyme (11,13). The protease labile region of the ␣ subunit appears to move during the catalytic cycle and becomes protected from protease treatment. Binding of FMNH 2 to the ␣ subunit reduces vulnerability to proteolytic inactivation (11,14,15). The ␤ subunit is insensitive to proteases, and the quaternary structure of the ␣-␤ complex is not altered by proteolytic cleavage (13).
Last year, we reported the crystal structure of bacterial luciferase from V. harveyi at 2.4-Å resolution (16). That structure was determined by multiple isomorphous replacement from crystals grown in 1.4 M ammonium sulfate, 0.2 M phosphate. Each subunit folds into a single domain (␤/␣) 8 barrel motif. Dimerization is mediated through a parallel four-helix bundle centered on a pseudo 2-fold axis that relates the structurally similar subunits. Recently, Conti and co-workers (17) ascertained the crystal structure of firefly luciferase. Firefly luciferase is a 62-kDa monomer that folds into a structure different from bacterial luciferase. This was anticipated since the protein sequence and the chemistry catalyzed by these two luciferases are considerably different. We report here a new crystal form of bacterial luciferase grown in low-salt conditions that diffract x-rays to significantly higher resolution than the previous crystals grown in high-salt concentrations. The structure was determined at 1.50-Å resolution from a single crystal of bacterial luciferase grown in methyl ether polyethylene glycol, which was frozen at Ϫ160°C. This higher resolution structure has revealed many new features of luciferase including the solvent structure and the observation of a non-prolyl cis peptide bond between residues Ala 74 and Ala 75 of the ␣ subunit, which may have functional significance. The higher resolution structure categorically defined the geometry of all the residues of the 77-kDa bioluminescent enzyme, with the exception of the protease-sensitive loop, and provides a molecular framework for all bacterial luciferases.

EXPERIMENTAL PROCEDURES
Crystallization and Data Collection-Luciferase, cloned from V. harveyi, was expressed in Escherichia coli and purified as described earlier (18). A new crystal form of bacterial luciferase was grown under conditions different than previously reported (16,19). Crystals were grown by micro batch method at 4°C in 17% methyl ether polyethylene glycol (ME-PEG) 1 (M r 5000), 250 mM MgCl 2 , buffered at pH 6.5 with 100 mM MES. The final protein concentration was 7.5 mg/ml. Crystallization was induced by introduction of micro-or macroseeds obtained from preliminary hanging drop experiments. Crystals utilized for seeding were prepared by soaking in 14% ME-PEG for 5 min to dissolve any additional nucleation points. The new crystal form of bacterial luciferase grown in methyl ether polyethylene glycol grew to the size of 0.7 ϫ 0.3 ϫ 0.1 mm in 7 days and diffracts x-rays to better than 1.5-Å resolution at a synchrotron radiation source. A crystal large enough for data collection was transferred directly into a cryoprotectant consisting of 20% ethylene glycol, 22% ME-PEG, 300 mM MgCl 2 , 50 mM MES, pH 6.5.
After transferring to the cryoprotectant, the crystal was immediately mounted in a loop (20) constructed of 20-m thick surgical suture and was frozen in the stream of nitrogen (Ϫ160°C) directly on the rotation camera at the Stanford synchrotron radiation laboratory, beam line 7-1 ( ϭ 1.08 Å). Strong diffraction maxima were observed beyond Bragg spacings of 1.5 Å. Data were collected by oscillation photography. Two scans of a single frozen crystal were employed for data acquisition strategy. The first scan consisted of 1°oscillations for long doses to increase the intensity of the high resolution reflections but resulted in overloading the low resolution data. The exposure time depended on beam energy and varied during the ring fill to equalize the number of photons per exposure. Diffraction data were collected on a MAR image plate system with a crystal to detector distance of 160 mm. After the high resolution data were collected, the crystal to detector distance was increased to 210 mm, and 2.5°oscillation photographs were taken for lower photon counts to record the low resolution data. Diffraction intensities were measured and scaled together with the programs DENZO and SCALEPACK (21,22). Partial reflections recorded on adjacent images were added together to approximate full reflections.
The crystals belong to the monoclinic space group C2 with unit cell parameters: a ϭ 150.5 Å, b ϭ 59.0 Å, c ϭ 76.5 Å, ␤ ϭ 93.86°. There is one ␣-␤ heterodimer in each asymmetric unit (V m ϭ 2.20 Å 3 /Da, solvent content Ϸ44%). A 99% complete data set to 1.5-Å resolution was collected on a MAR image plate system from a single crystal that was frozen to Ϫ160°C. The overall R merge is 4.1% for all data to 1.5-Å resolution. Table I gives the data collection statistics.
Structure Determination and Refinement-The structure of luciferase grown in ME-PEG was solved by the molecular replacement method (23). The 2.4-Å resolution ammonium sulfate structure was used as a search model in the molecular replacement program AMORE (24). Data between 10.0-and 4.0-Å resolution were used in the rotation search that resulted in a peak of 14.6 , the highest false peak was 7.3 . The rotated model was applied in a translation search resulting in a single large peak of 47.6 and an R-factor of 37.9%. The luciferase structure was then refined against all recorded data to 2.0-Å resolution by the conjugate direction algorithm implemented in TNT (25), lowering the R-factor to 30.3%. An initial electron density map was computed at 2.0-Å resolution employing SIGMAA coefficients to suppress model bias (26). The resulting map was of excellent quality, and manual adjustments were made with the program O (27). The ensuing model was then subjected to another round of TNT refinement, which lowered the R-factor to 24.1% for all data to 2.0-Å resolution. Subsequent refinement against all recorded data to 1.50-Å resolution resulted in an R-factor of 26.0%, which was reduced to 21.1% with minor manual adjustments of the model and adding 302 water molecules with the programs PEKPIK in TNT (25) and WATPEAK in the CCP4 program suite (28). The final model of the ␣ subunit consists of residues 1-261 and 291-355. The 29 residues (262-290) for which there is no electron density corresponds to the protease-sensitive loop that is also disordered in the ammonium sulfate structure. SDS gel analysis of luciferase crystals indicates that both subunits are intact (data not shown). All but the last four amino acids of the ␤ subunit were traced in the electron density map. The final model has an R-factor of 18.2% for all recorded data (͉F͉ Ͼ 0) to 1.50-Å resolution where the root mean square (r.m.s.) deviation from ideal bond lengths, angles, and planes are 0.015 Å, 2.40°, and 0.007 Å, respectively. Table I presents the final refinement statistics including number of atoms and average B-values. A plot of the average main-chain temperature factors and correlation coefficient are shown in Fig. 1, a and b. The correlation coefficient is a measure of how well the atoms fit the electron density as calculated in the molecular graphics program O (27). The mean main-chain temperature factor and correlation coefficient for the ␣ subunit are 16.9 Å 2 and 0.948, respectively, and for the ␤ subunit 16.6 Å 2 and 0.945, respectively.

RESULTS AND DISCUSSION
Structure Description-A Ramachandran plot (29) of the main-chain conformation angles indicates that 99% of the nonglycine residues lie in the allowed regions as defined by the program PROCHECK (30). The average coordinate error in the final model, as estimated from a Luzzati plot (31), is between 0.125 and 0.15 Å. Figs. 7a and 9 show a region of representative electron density at 1.50-Å resolution computed with the coefficients of 2͉F o ͉ Ϫ ͉F c ͉ and phases calculated from the final model.
Hydrophobic residues pack in the ␤-barrel inner core of both subunits. However, NH 2 -terminal residues of the ␤-strands are hydrophilic and exposed to solvent. Part of the ␤-barrel's Cterminal end is hydrophobic and shielded from solvent by two ␣-helices. The ␣-helices (␣7a and ␣7b) emerge from the ␤7-␣7 loop. This feature is observed in both ␣ and ␤ subunits. In the ␤ subunit, helix ␣7a extends along the top of the barrel, followed by a tight turn then helix ␣7b, which runs antiparallel to helix ␣7a. In the ␣ subunit, helix ␣7a stretches toward the subunit interface. The loop that connects helices ␣7a to ␣7b is disordered in the electron density map. Residues 262-290 of the ␣ subunit are not seen in the electron density map. The disordered loop in the ␣ subunit corresponds to the 29-residue insert when compared with the ␤ subunit (residues 258 -286) and is the loop that is readily cleaved by proteases in the absence of substrates (11,12,14). In the ␣ subunit, helix ␣7b is short, consisting of 5 residues, although its true length may be obscured by the flexibility in the preceding loop. After helix ␣7b, both subunits contain a 3-residue ␤-strand (␤7a) that runs parallel to and augments ␤7, which extends past the other ␤-strands of the ␤-barrel.
The only other deviations from the (␤/␣) 8 topology is a small helix (␣4a) that is positioned at the C-terminal end of the ␤-barrel of each subunit near the subunit interface. There is also a hairpin loop structure in both subunits that runs along the periphery of the subunit interface and embraces the parallel four-helix bundle at the dimer interface. This hairpin loop contains internal main-chain hydrogen bonds, but the mainchain torsion angles are inconsistent with ␤ structure required to designate the loop an antiparallel ␤-hairpin. Pro 154 , conserved in both subunits, disrupts the possible ␤-strand (Fig. 4,  a and b). Furthermore, Pro 146 disrupts the other strand in the ␤ subunit opposite Pro 154 (Fig. 4b). The reverse turn at the apex of the hairpin loop structure closely resembles a ␤ type III turn, but the carbonyl oxygen of residue i does not hydrogen bond with the main-chain amide nitrogen of residue i ϩ 3. This structure is observed in the hairpin loops of both subunits. Also in both subunits, the residue at position i of the turn is Asn 148 , which favors ␤ reverse turns because the O␦-1 atom hydrogen bonds to the main-chain amide nitrogen of residue i ϩ 2 as is observed in both luciferase subunits. The hairpin loops in both subunits terminate with Pro 160 whose peptide bond adopts the cis configuration in both subunits. Pro 160 is conserved among all luciferase ␣ and ␤ subunits suggesting the importance of a cis peptide bond conformation at this position.
Dimerization is mediated through a parallel four-helix bun- dle, which is centered on a pseudo 2-fold axis that relates the ␣ and ␤ subunits (Fig. 2). Each subunit contributes helices ␣2 and ␣3 to form the four-helix bundle. Helix ␣2 lies very close to the pseudo 2-fold axis resulting in an unusually close packing of the ␣2 helices from each subunit. At one point, the main chain atoms from one helix reside within 3.2 Å from the main chain atoms in the pseudo 2-fold-related helix in the other subunit. In this region, glycines and alanines shape the surface of the helix allowing for the close contact. In particular, Gly 64 is totally conserved in all luciferase ␣ and ␤ subunits permitting this intimate contact.
There are a considerable number of intersubunit interactions arising from the dimer interface. Most of these contacts occur in the parallel four-helix bundle. The majority of intersubunit contacts established in the four-helix bundle are van der Waals interactions. 2150 Å 2 of accessible surface area is buried upon dimer formation based on a search probe radius of 1.4 Å (33). This value falls in the expected range given the size of the luciferase subunits (34). Twenty-two intersubunit hydrogen bonds help tether the two subunits together (Table II). An interesting hydrogen bond occurs between residues His 45 and Glu 88 . These two residues are conserved among the ␣ and ␤ subunits creating similar intersubunit hydrogen bonds related by pseudo 2-fold symmetry (His ␣45 H-bonds to Glu ␤88 and His ␤45 to Glu ␣88 ). Both of these residues are conserved among all bacterial luciferase ␣ and ␤ subunits, and mutating His 45 in the ␣ subunit of V. harveyi luciferase results in a substantial decrease of bioluminescence activity (35). Another intriguing interaction occurs between Arg ␣85 and Thr ␤80 . The side chain of Arg ␣85 extends across the subunit interface, and the guanido group is in hydrogen bonding distance with both the O␥-1 and the carbonyl oxygen of Thr ␤80 . These two residues are also totally conserved among all luciferase subunits resulting in a similar interaction between the pseudo 2-fold-related residues. Arg 85 also points toward Glu 43 , which again is conserved among all luciferase subunits, but is 3.7 and 4.4 Å away from the carboxylate group in the ␣ and ␤ subunits, respectively. These residues line a large surface accessible cavity that is formed at the dimer interface (Fig. 5). Many ordered water molecules dwell in this pocket. There are also 45 water-mediated intersubunit hydrogen bonds where an atom from each subunit is within hydrogen bonding distance to the same water molecule.
In addition to the polar contacts discussed above, there are also a substantial number of hydrophobic interactions involved in dimer formation. Hydrophobic intersubunit contacts are found between the hairpin loop structure in one subunit and the C-terminal end of helix ␣4a plus a region of the ␤2-␣2 loop of the other subunit. The C-terminal region of helix ␣3 establishes hydrophobic contacts with the N-terminal region of helix ␣1 in the other subunit. Most of the hydrophobic interactions are conserved among the subunits resulting in pseudo symmetric interactions.
A great deal of sequence conservation exists between the two luciferase subunits. Many of the conserved residues are also preserved in the luciferases from other bioluminescent bacteria. Thirty residues are totally conserved among all bacterial luciferase ␣ and ␤ subunits whose sequences are known presently (8,36). A majority of the conserved residues dwell near the luciferase ␣-␤ dimer interface (Fig. 6). This demonstrates that the pseudo 2-fold axis, which relates the ␣ and ␤ subunits, is also evident at the level of the primary structure. Furthermore, conservation of the 2-fold symmetry at the interface suggests its significance for dimerization and enzyme function. This is confirmed by the mutational and structural studies described above. Similar intersubunit interactions are also observed between these residues in the crystal structure of the LuxB homodimer. 2 Non-prolyl Cis Peptide Bond-In the ␣ subunit, ␤-strand 3 terminates with a bulge that protrudes into the core of the ␤-barrel. This bulge contains a cis peptide bond between residues Ala 74 and Ala 75 . Fig. 7a illustrates the conformation of ␤3 with the bulge and the cis peptide displayed with the electron density map. The 1.5-Å resolution electron density map unequivocally demonstrates the cis conformation of the peptide bond between residues Ala 74 and Ala 75 in the ␣ subunit (Fig.  7a). In the 2.4-Å resolution ammonium sulfate structure, the bulge did not fit the density extremely well, but the map was not high enough resolution to confidently build a cis peptide bond. A similar bulge terminates ␤3 in the ␤ subunit, but the density clearly indicates a trans peptide bond between Leu 74 and Asn 75 . Fig. 7b illustrates the similarity of the bulge and overall shape of ␤3 between the two subunits.
Non-prolyl cis peptide bonds are rare (37, 38) but have been observed in a few other crystal structures, and almost all play significant roles in positioning crucial residues to carry out ligand binding and/or catalysis (39). In luciferase, the cis peptide bond occurs in a bulge at the end of ␤3 positioning it at the C-terminal end of the barrel, where all (␤/␣) 8 barrels exhibit their active sites (40). Ala 74 and Ala 75 form the bottom floor at the entrance of a small cavity projecting off the larger and deeper pocket in the center of the ␤-barrel of the ␣ subunit (Fig.  8). The walls of this smaller cavity include His ␣44 on one side and Cys ␣106 on the other. Mutation of His ␣44 to Ala or Asp results in inactivation of the enzyme (35). Cys 106 of the ␣ subunit is a highly reactive thiol whose chemical modification resulted in inactivation of the enzyme (41,42). However, sitedirected mutagenesis experiments have clearly demonstrated that the reactive thiol is not involved in the bioluminescence reaction (18). Binding of either FMN or FMNH 2 in the presence of O 2 to luciferase protects the reactive thiol from modification (41), and modification of the Cys ␣106 thiol substantially decreases the affinity of the protein for FMNH 2 (43). However, modification of the reactive thiol has little effect on the binding of FMN (44). These observations suggest that there is not a direct interaction of the flavin with the thiol that affects protection but rather a conformational change resulting from flavin binding.
These data demonstrate the importance of the small cavity projecting off the central large pocket and could justify the reason for the cis peptide bond, because a trans conformation would decrease the size of the opening. Residue 75 of the ␣ subunit is either an alanine or glycine residue in all luciferases. Proline, which is more energetically favorable in cis peptides, would introduce a larger side chain and reduce the size of the opening. Additionally, the main-chain dihedral angles for Ala ␣75 ( ϭ Ϫ153.1, ϭ 164.5) are unfavorable for proline residues, which prefer to reside around Ϸ Ϫ60.
As seen in Fig. 8, two residues from the ␤ subunit also play a role in the small cavity. Glu 88 from the ␤ subunit hydrogen bonds to His ␣45 , which forms part of the cavity sidewall. This intersubunit interaction, as pointed out above, is conserved in all luciferase subunits. In addition, the guanido group of Arg ␤85 forms the back of the cavity. Arg 85 is also mentioned above for its conserved interactions between subunits. These two residues might suggest a possible role for the ␤ subunit during the bioluminescent reaction if this cavity, which extends off the larger pocket at the C-terminal end of the barrel, is part of the active site. The cavity in the ␤-subunit is more confined because of a trans peptide bond between positions 74 and 75, and larger residues line the cavity's entrance; Asn replaces Ala at position 75 and Tyr substitutes for Leu at position 42.
Magnesium Binding and Crystal Packing-Crystallization of luciferase in methyl ether polyethylene glycol requires the presence of magnesium. Omission of magnesium results in no crystal growth. Removal of magnesium from crystals, by addition of EDTA, results in cracking. During the first manual rebuilding, it became evident why magnesium was required for crystallization. A total of three magnesium ions were observed in the crystal structure (Fig. 6). Two magnesium ions are involved in crystal contacts between symmetry related dimers and the third ion binds to the ␣ subunit but does not have any functional or structural capacities.
One magnesium ion (Mg 2ϩ 2002) involved in crystal packing is coordinated by O⑀-1 of Glu ␤237 and O␦-1 of Asp ␣Ј346 in a symmetry related subunit (Fig. 9) (prime Greek letters represent crystallographic symmetry related subunits). Four ordered water molecules complete the octahedral geometry. The Mg 2ϩ -oxygen coordination distances range from 2.03 to 2.33 Å.
O⑀-1 of Glu ␣19 coordinates the other magnesium ion involved in crystal packing (Mg 2ϩ 2001), and the five remaining Mg 2ϩ ligands are ordered water molecules. The carboxylate group of Glu ␤Ј130 from a symmetry related subunit hydrogen bonds to two of the five Mg 2ϩ water ligands. Water molecule 3101 is 2.75 Å away from O⑀-1 of Glu ␤Ј130 , and the distance between O⑀-2 of Glu ␤Ј130 and water 3177 is 2.76 Å.
The third magnesium ion seen in the crystal structure (Mg 2ϩ 2003) is not involved in any crystal contacts but binds to the periphery of the ␣ subunit. No protein atoms directly ligate the Mg 2ϩ . Six ordered water molecules coordinate the ion with octahedral geometry. This magnesium ion binds near the N- terminal opening of the ␤-barrel and interacts with residues in the loops preceding ␤ strands ␤7 and ␤8 of the ␣ subunit (Fig.  6). Six protein atoms hydrogen bond to five of the water ligands. The O␦-2 atom of Asp ␣223 and the O␦-2 of Asp ␣321 hydrogen bond to the same water ligand (3418), 2.75 and 2.70 Å, respectively. O␦-1 of Asp ␣321 hydrogen bonds to water 3335. The main-chain carbonyl oxygen of Lys ␣221 is 2.91 Å away from water ligand 3352, and the carbonyl oxygen of Ile ␣222 is 3.18 Å away from water ligand 3380. The amine group of Lys ␣2 hydrogen bonds to Mg 2ϩ ligating waters 3418 and 3486. The binding site of this magnesium ion was unexpected but probably does not have any functional significance since no protein atoms directly ligate the Mg 2ϩ . Furthermore, the B-factors of the third Mg 2ϩ and its coordinated waters are approximately 15 Å 2 higher than for the other two Mg 2ϩ sites involved in crystal contacts, suggesting this binding might be nonspecific and a result of the high MgCl 2 concentration used in crystallization.
In addition to the magnesium ions binding to the luciferase structure, five well ordered ethylene glycol molecules are apparent in the solvent structure. Ethylene glycol was used as a cryo-protectant to preserve the crystal during freezing at Ϫ160°C. All five ethylene glycols bind at the protein surface, three of them at the ␣-␤ subunit interface. One of the ethylene glycol molecules binds in a small cavity that is formed between helices ␣1 and ␣2 of the ␣ subunit. Another ethylene glycol molecule mediates an intersubunit contact. His 82 from the ␣ subunit hydrogen bonds to a glycol hydroxyl oxygen, which in turn hydrogen bonds to the carbonyl oxygen of Phe 226 in the ␤ subunit. In all five cases, one or both ethylene glycol hydroxyl oxygens hydrogen bonds to the protein.
Structural Similarities-There is extensive structural conservation between the ␣ and ␤ subunits confirming their common origin (4). The topology of the ␣ and ␤ subunits is identical, and the secondary structural elements align exactly with the sequence (Fig. 10). The two luciferase subunits superimpose FIG. 8. Active site cavity. Stereo balland-stick rendering of the small cavity that projects off the main active site pocket at the ␣ subunit's C-terminal portion of the ␤-barrel. The small cavity is lined by the cis peptide bond on the bottom, Cys 106 whose modification occludes substrate binding, and His 44 , that when mutated (35) inactivates the enzyme. Also illustrated, with red bonds, is Arg 85 from the ␤ subunit that forms the back wall of the cavity and Glu 88 from the ␤ subunit that hydrogen bonds to His 45 . with a root mean square deviation of 1.99 Å for 300 equivalent ␣-carbons (Fig. 11a). The structures of the ␤-barrels are very similar with only a 0.61-Å r.m.s. deviation in the superposition of the barrel's 39 ␣-carbons. Most of the differences in the ␣-␤ superposition occur in the exterior ␣-helices, which are slightly displaced relative to their pseudo 2-fold-related subunit. The largest displacement appears in the short helix ␣4a near the C-terminal end of the ␤-barrel. Helix ␣4a in the ␣ subunit shifts approximately 3 Å along the helix axis away from the barrel's center permitting a larger opening to the active site. The regions involved with dimerization, helices ␣2 and ␣3 and the hairpin loop structure, are exceptionally similar in structure.
Structural similarities were also observed between the luciferase subunits and the nonfluorescent flavoprotein (NFP) (45,46) from Photobacterium leiognathi. Bioluminescent bacteria belonging to the genus Photobacterium contain an additional gene located between luxB and luxE in the lux operon. This gene now known as luxF, was originally designated luxG (47), and independently as luxN (48). The luxF gene encodes a 24-kDa nonfluorescent flavoprotein whose function is unknown at present but binds two molecules of an unusual flavin mononucleotide adduct (45,46,49). Myristic acid is covalently linked to C-6 of the isoalloxazine ring of the flavin mononucleotide. Interestingly, both myristic acid and FMN are end products of the luciferase bioluminescence reaction. However, the connec-tion between the nonfluorescent flavoprotein and bioluminescence remains unclear. NFP displays sequence similarity to both luciferase subunits and is 22.4 and 33.3% identical in amino acid sequence to the luciferase ␣ and ␤ subunits, respectively (Fig. 10).
The crystal structure of NFP has been recently determined (49) and refined to high resolution (50). The crystal structure revealed that NFP forms a homodimer, and each monomer folds into a novel seven-stranded ␤-barrel surrounded by seven ␣-helices. Given the NFP structure and sequence alignment of luxF to luxA and luxB, Moore and James (51) correctly predicted the structure of the luciferase monomer to have a (␤/␣) 8 fold. The structure of NFP superimposes surprisingly well with the individual luciferase ␣ and ␤ subunits, 2.48 and 1.55 Å r.m.s. deviation, respectively (Fig. 11b). The NFP ␤-barrel is mostly parallel (strands ␤3 and ␤4 form an antiparallel hairpin) and contains a considerable gap between strands ␤2 and ␤3. The seven NFP ␤-strands align well structurally with seven of the eight luciferase strands. In the superposition, strand ␤3 of luciferase (␣ and ␤ subunits) resides in the gap between strands ␤2 and ␤3 of NFP that would complete an eightstranded ␤-barrel (Fig. 11b). In the NFP structure, this gap is filled with ordered water molecules that fasten the two ends of the barrel together with a hydrogen bonding network to strands ␤2 and ␤3 (50). It is interesting to point out that the FIG. 10. Sequence alignment. Protein sequence alignment of luciferase ␣ and ␤ subunits from V. harveyi (LuxA and LuxB) and nonfluorescent protein from P. leiognathi (LuxF). Alignment was based on sequence and secondary structural elements from the crystal structures. The luciferase ␣ and ␤ subunits share 31.9% sequence identity, whereas NFP is 22.4 and 33.3% identical to the luciferase ␣ and ␤ subunits, respectively. The secondary structural elements, as observed in the crystal structures, are displayed above the sequence and by colored boxes.
The asterisks label residues that are conserved in all three proteins. strand missing in the NFP structure is the same strand that terminates with a non-prolyl cis peptide bond in the luciferase ␣ subunit. Strand ␤3 of NFP structurally aligns with ␤4 of luciferase in the superposition but runs in the reverse direction.
The seven helices of NFP align with helices ␣1, ␣5, ␣6, ␣7a, ␣7b, ␣7, and ␣8 of the luciferase ␣ and ␤ subunits (Figs. 10 and 11b). NFP does not contain residues or secondary structural elements corresponding to the helices and the hairpin loop involved in the luciferase dimerization. Sequence alignment of NFP to luciferase ␣ and ␤ subunits reveals a gap in the NFP sequence corresponding to helices ␣2, ␣3, ␤-strand ␤3, and the hairpin loop (Fig. 10). However, homo-dimerization of NFP still occurs along the same relative region of the molecule, but the intersubunit interactions occur between ␤-strands (49,50) and not helices as observed in luciferase. Additionally, the ␤7-␣7 loop in the luciferase ␤ subunit, which contains helices ␣7a and ␣7b and the short strand ␤7a that augments ␤7, is also seen in NFP and superimposes with an r.m.s. deviation of 0.79 Å.
This evidence suggests that luxF may have arisen from gene duplication of luxB (luciferase ␤ subunit) and subsequently lost its ability to associate with the luxA gene product by deletion of the residues involved in dimerization. Yet LuxF still maintained (or developed) its ability to form homodimers. The function of LuxF, which is found in only one genus of bioluminescent bacteria, is unknown, but is not required for bioluminescence (36).
The locations of the two unique flavin adducts that bind to NFP are shown in Fig. 11b. Both flavin cofactors bind on the side of the ␤-barrel between the surface helices. These binding sites probably do not reveal the flavin active site in luciferase, because helices ␣4 and ␣8 in luciferase extend over the equivalent flavin binding sites and would occlude FMN binding. Moreover, both sites are distant from the C-terminal end of the FIG. 11. Luciferase structural comparison. a, superposition of the luciferase ␣ and ␤ subunits drawn in stereo. The ␣-carbon trace is drawn in blue and red to delineate the ␣ and ␤ subunits, respectively. The two subunits superimpose with an r.m.s. deviation of 1.99 Å for 300 equivalent ␣-carbons. b, stereo superposition of LuxF NFP (black lines) onto the luciferase ␤ subunit (red lines). NFP superimposes onto the ␤ subunit with an r.m.s. deviation of 1.55 Å for 132 equivalent ␣-carbons. The myristoylated flavin adduct bound to NFP is drawn in green bonds.
␣ subunit's ␤-barrel, which is the location of the active site in enzymes that exhibit the (␤/␣) 8 motif (40). The phosphate moiety of the flavin molecule that binds near the N terminus of helix ␣5 in NFP is 3.2 Å away from the phosphate binding site seen in the luciferase structure previously determined in ammonium sulfate (16). This region was hypothesized to bind the phosphate moiety of reduced flavin in luciferase (16). By anchoring the phosphate moiety of FMNH 2 at this site, the flavin can be modeled extending across the C-terminal portion of the ␣ subunit ␤-barrel. This positions the isoalloxazine ring next to Trp 194 and Trp 250 , which have been implicated to interact with the flavin ring as measured by fluorescence spectroscopy and circular dichroism spectroscopy. 3 A phosphate ion was not observed in the ME-PEG structure because phosphate was not included in the crystallization conditions.
The regions of high temperature factors in the ␣ subunit correspond to loops that have been proposed to bind flavin (16). The peaks in the ␣ subunit temperature factor plot (Fig. 1a) around residues 109, 121, and 175 all map to loops in the vicinity of the phosphate binding site that was observed in the structure of luciferase derived from crystals grown in ammonium sulfate. The high thermal parameters in the phosphatefree structure suggest flexibility in these areas, which would become stationary upon binding reduced flavin.
There appears to be no significant differences between the structure of luciferase solved in ME-PEG and the original structure solved in ammonium sulfate. Superposition of 630 equivalent ␣-carbons results in an r.m.s. deviation of 0.59 Å between the two structures. This value falls in the range observed for other proteins whose structures have been determined from different space groups (52,53). Furthermore, in addition to the differences described above, more of the ␣-subunit's ␣7a-␣7b loop is disordered in the ME-PEG structure, which contains 13 fewer ordered residues.
Active Site Pocket-The active sites of all (␤/␣) 8 barrel enzymes reside at the C-terminal end of the ␤-barrel (40). In most cases, residues in the loops that connect the ␤-strand to the subsequent ␣-helix fabricate the active site. Many flavoenzymes employ the TIM barrel motif to bind flavin (40). Glycolate oxidase (54), flavocytochrome b 2 (55), trimethylamine dehydrogenase (56), and old yellow enzyme (57) are all (␤/␣) 8 barrels that tightly bind flavin mononucleotide as a coenzyme. In these enzymes the phosphate moiety of FMN binds between the ␤7-␣7 loop and the NH 2 terminus of an additional small helix in the ␤8-␣8 loop. Similar interactions are also observed in other TIM barrels that secure phosphate components in their substrates (58). Bacterial luciferase does not contain a small helix in the ␤8-␣8 loop. Its absence could explain why luciferase utilizes FMNH 2 as a substrate and not as a prosthetic group as in other (␤/␣) 8 flavoenzymes.
The structure of luciferase reveals a large deep pocket entering the C-terminal end of the ␣ subunit's ␤-barrel (Fig. 12). Projecting off this large central pocket is a smaller accessible cavity formed by the non-prolyl cis peptide bond highlighted above (Fig. 8). These pockets are sufficiently large enough to accommodate FMNH 2 , O 2 , and a long-chain aldehyde. Furthermore, the pocket is expected to exclude access water from the C4a hydroperoxyflavin intermediate and the excited flavin that is formed following the decay of the tetrahedral intermediate (8). The disordered loop is likely to achieve this task by blocking the entrance to the pocket after substrate binding, thus protecting itself from proteolysis (11,14,15). The current 1.5-Å resolution structure contains a few ordered water molecules in the pocket. Even though the structure of luciferase was determine in the absence of substrates, we feel confident that the active site resides within this large internal cavity of the ␣ subunit. It should be noted that every amino acid implicated as an active center residue, either by mutagenesis or chemical modification, contacts this internal cavity. Unfortunately, attempts to soak in both oxidized and reduced flavin with and without additional substrates into the crystal have proven unsuccessful.
Folding and Assembly of Luciferase-Protein unfolding, refolding, and assembly of bacterial luciferase has been extensively studied (59 -62). It has been demonstrated that separate ␣ and ␤ subunits, purified from recombinant E. coli independently bearing the luxA or luxB genes, carry out a bioluminescent reaction, but at a quantum efficiency 6 orders of magnitude below that of the heterodimer (63). Moreover, the active dimer fails to assemble when the purified folded ␣ and ␤ subunits are combined (64,65). It has recently been demonstrated that purified luciferase ␤ subunit forms a very stable ␤ 2 homodimer that is trapped in a heterodimerization-incompetent complex and is unable to form functional heterodimers due to kinetic partitioning of the folding pathway (62). Functional dimers can assemble upon renaturation of the unfolded individual subunits (60,62). Equilibrium unfolding studies of the luciferase heterodimer have shown that the enzyme unfolds through a well-populated non-native intermediate (59,61). Conversion from the non-native heterodimeric intermediate to a functional enzyme is independent of protein concentration.
Some of the protein folding and assembly observations might be explained in part by the presence of the protein's two prolyl and one non-prolyl cis peptide bonds. It has been substantiated that cis/trans isomerization of the peptide bond preceding proline residues can limit the rate at which a protein can fold into its native conformation (66 -68). The cis-Pro 160 , which is found in both luciferase subunits, is located at the end of the hairpin loop structure that forms extensive intersubunit contacts. The trans isomer of this peptide bond would cause minor perturbations in the loop that would affect dimerization contacts. In both subunits the residue preceding the cis-proline is an aspar-  Stereo view of the molecular surface illustrating the size and depth of the active site pocket in the luciferase ␣ subunit. The molecular surface was drawn with the program GRASP and color-coded by electrostatic potential. The small cavity bordered by the cis peptide bond is projecting down out of view in the pocket. Residues that flank the disordered loop and helices ␣5 and ␣7a of the ␣ subunit are labeled for orientation.
agine that contributes to dimerization by forming intersubunit hydrogen bonds (Table II).
Non-prolyl cis peptide bonds are scarce in nature but can be generated by mutating cis-proline residues in proteins (69,70). Energy calculations predict that non-prolyl cis peptide bonds should destabilize a folded protein by 10 -20 kJ/mol (37). Protein thermal stability measurements yield results that compare with the calculated values (70,71). The presence of a non-prolyl cis peptide bond in the ␣ subunit alone could explain why the heterodimer is apparently less stable than the ␤ 2 homodimer (59 -62).