Crystal Structure of Quinohemoprotein Amine Dehydrogenase from Pseudomonas putida IDENTIFICATION OF A NOVEL QUINONE COFACTOR ENCAGED BY MULTIPLE THIOETHER CROSS-BRIDGES*

From the ‡Department of Chemistry, Graduate School of Science, Osaka City University, Osaka 558-8585, Japan, §Department of Structural Molecular Biology, Institute of Scientific and Industrial Research, Osaka University, Ibaraki, Osaka 567-0047, Japan, ¶Laboratory of Protein Biochemistry and Protein Engineering, University of Ghent, 9000 Ghent, Belgium, Department of Microbiology and Enzymology, Delft University of Technology, 2628 BC Delft, The Netherlands, and **Department of Biological Chemistry, Faculty of Agriculture, Yamaguchi University, Yamaguchi 753-8515, Japan

Recently, a number of modified amino acids have been identified in proteins that are generated by post-translational oxidation or non-oxidation processes (1,2). Such a controlled modification of a specific amino acid residue forming part of the active site provides catalytic power to the protein. In the case of a certain class of amine-oxidizing enzymes, depending on the enzyme concerned, oxidation of a specific tyrosine or tryptophan residue leads to the generation of a redox-active quinone cofactor: 2,4,5-trihydroxyphenylalanine quinone (topaquinone) (3), lysine tyrosylquinone (4), or tryptophan tryptophylquinone (TTQ) 1 (5). Together with several enzymes containing the first identified, non-proteinous quinone cofactor, pyrroloquinoline quinone (PQQ), the enzymes containing those cofactors constitute a quinoprotein family of enzymes (6).
Quinohemoprotein amine dehydrogenases (QH-AmDH) from Gram-negative bacteria represent a new type in the quinoprotein class of amine-oxidizing enzymes because they contain not only a quinone but also one or two hemes as a redox active group (7,8) providing an opportunity for intramolecular electron transfer. Intermolecular electron transfer from QH-AmDH has been demonstrated with the natural electron acceptors azurin for the enzyme from Pseudomonas putida (7) and cytochrome c-550 for the enzyme from Paracoccus denitrificans (9). The structure of the presumed quinone cofactor in QH-AmDH remain to be settled, although biochemical and spectroscopic analyses have suggested the presence of a quinone group similar to, but not identical, with TTQ (7,8).
To identify the quinone cofactor and its position in the protein, we (10) have recently determined the primary structure of the quinone-containing small subunit of QH-AmDH from P. putida using a combination of automated Edman degradation and mass spectrometry, and we have also cloned the genes coding for the three subunits of the heterotrimeric enzyme. Although we initially encountered difficulties in the interpretation of the chemical and mass data, the progress in elucidating the crystal structure of the enzyme enabled us to unequivocally identify a novel quinone cofactor, cysteine tryptophylquinone (CTQ) (naming is analogous with that for TTQ). Besides this novel "in situ-generated" quinone cofactor, the covalent structure reported in the preceding article (10) and the 1.9-Å crystal structure reported in the present article have also revealed a unique feature of the catalytic ␥-subunit in which CTQ is encaged by three intra-chain cross-links. These links are unprecedented as they are thioether bonds between a cysteine sulfur atom and a methylene carbon atom of an aspartic or a glutamic acid residue. drop vapor-diffusion method. Drops were prepared by mixing 5 l of 12 mg ml Ϫ1 protein solution with 5 l of a reservoir solution of 18% (w/v) polyethylene glycol 2000 monomethylether and 50 mM NiCl 2 . Crystals grew in a space group C2 (unit cell: a ϭ 167.2, b ϭ 91.5, c ϭ 78.9 Å, and ␤ ϭ 113°) with one heterotrimeric molecule in the asymmetric unit. To prepare the complex with p-nitrophenylhydrazine (pNPH), the native crystals were soaked in the solutions supplemented with 1 mM pNPH. The heavy atom derivatives were obtained by soaking the crystals in solutions containing heavy atom reagents. The data sets for the native crystals, the pNPH-complexed crystals, and the crystals for heavy atom derivatives were collected at 287 K on Beam Line 18B at Photon Factory (Tsukuba, Japan) (see Table I). The intensity data were integrated and scaled with DPS/MOSFLM (11) and SCALA/CCP4 (12).
Structure Determination and Refinement-The structure was solved by multiple isomorphous replacement and anomalous scattering. The location of the initial heavy atom sites was determined by the difference Patterson method. The experimental phases to 3.0 Å were calculated using SHARP (13) with a figure of merit of 0.74 including all derivatives followed by solvent flattening by SOLOMON (14) with a figure of merit of 0.94. The resulting phases were improved and extended to 2.6 Å. The model of QH-AmDH was built stepwise into the 2.6-Å map with the program O (15), and a series of refinements was performed with the program XPLOR (16). In this stage, the ␣and ␤-subunits were modeled nicely on the electron density map, but the ␥-subunit could be modeled only partially because of its anomalous structure containing very few ␣-helices and no ␤-strands (see "Results and Discussion"). Therefore, the phases were further improved based on this temporal model, and subsequently the 2F o Ϫ F c electron density map was calculated at 1.9-Å resolution. The resultant map was of sufficient quality for tracing the whole polypeptide chain of the ␥-subunit with the side chains referring to the primary structure. After several rounds of refinement and manual rebuilding, R factor and R free were reduced to 29.8 and 26.7%, respectively. Water molecules were picked up from the difference map on the basis of the peak heights and distance criteria except for those whose thermal factors after refinement were above 58 Å 2 (corresponding to the maximum thermal factor of the main chain atoms). Further model building and refinement cycles resulted in an R factor value of 21.1% and an R free value of 24.5% calculated for 87,307 reflections (F o Ͼ 2(F o )) observed in a 10.0 -1.9-Å resolution range (see Table II). During the last step of the refinement, unambiguous water molecules were added including those with a temperature factor higher than 58 Å 2 . The maximum temperature factor of the water molecules was 78 Å 2 .
The same refinement procedure was applied to the pNPH-complexed enzyme but using the coordinates of the native QH-AmDH as an initial model. When the R factor value reached below 29%, the difference Fourier map clearly exhibited the residual electron density corresponding to the bound pNPH. Further model building and refinement cycles gave R factor and R free values of 21.1 and 25.0%, respectively, calculated for 74,658 Table II). The maximum temperature factor of the assigned water molecules was 75 Å 2 .
Quality of the Structure-The final models of the native enzyme and the pNPH complex both contained 909 amino acid residues and two heme c groups with 457 water molecules for the native enzyme and 359 water molecules plus one pNPH molecule (as the hydrazone) for the pNPH complex. No interpretable electron density was observed for one N-terminal residue of the ␣-subunit, three N-terminal residues and residues 220 -226 of the ␤-subunit, and two N-terminal residues of the ␥-subunit. Both models had a good quality with 98.8% of residues falling in the most favorable and additionally allowed region and only 0.3% in the disallowed region, when the stereochemistry was assessed by PROCHECK (17) (see Table II). Structure diagrams were drawn with programs MOLSCRIPT (18), Raster3D (19), and BOBSCRIPT (20).

RESULTS AND DISCUSSION
Overall Structure-The crystal structure of the native QH-AmDH was determined at 1.9-Å resolution (Tables I and II), utilizing the DNA-based protein sequence (10). Consistent with the previous biochemical analysis (7), the enzyme is composed of three non-identical subunits: a 494-residue ␣-subunit (ϳ60 kDa) carrying two heme c groups, a 349-residue ␤-subunit (ϳ40 kDa), and a 79-residue ␥-subunit (ϳ9 kDa) bearing the quinone cofactor. The ␥-subunit is embedded in the crevice of the ␣-subunit with an extensive intersubunit interface with a contacting surface area of 4,640 Å 2 , making the ␣␥ pair look like a single subunit (Fig. 1A). When viewed in cross-section, the convex "bottom" of the heterodimer lies on the concave "top" surface of the ␤-subunit, with a contacting surface area of 4,913 Å 2 to build up the overall heterotrimeric structure. As a result, only 34% of the surface area of the ␥-subunit is exposed to solvent.
The ␣-subunit comprises four distinct domains (I-IV). Domain I has a predominant helical structure and can be subdivided into two subdomains, each containing a heme c group covalently attached to a Cys residue through a thioether linkage, as in class I cytochromes c (21) (Fig. 1A). The overall structure of domain I is similar to that of a di-heme cytochrome c (22)(23)(24), in which the two cytochrome c-like domains are related by a pseudo 2-fold axis, suggesting that the ␣-subunit harbors a di-heme cytochrome c forming an intramolecular electron transfer system. One of the two heme c groups (heme I) is encapsulated within the protein interior, facing the interface with the ␥-subunit, whereas the other heme c (heme II) is situated at the interface between the ␣and ␤-subunits, exposing one of its sides toward the solvent. The two heme planes are tilted by about 51°with each other and separated in an Fe-Fe distance of 15.8 Å. The two iron atoms of the heme groups have different axial ligands, His-106␣ and His-128␣ in heme I and His-18␣ and Met-46␣ in heme II, suggesting that they have different redox potentials. Domains II-IV of the ␣-subunit consist mainly of ␤-structures with two short ␣-helices located on the surface side of domain IV. Domain II has a typical ␤-barrel structure with eight up-and-down antiparallel ␤-strands. Domain III is a pseudo barrel formed by three and four antiparallel ␤-strands. Domain IV is also a pseudo barrel of seven ␤-strands with an additional ␤-sheet of three strands on the N-terminal side and the C-terminal loop interacting with the ␥-subunit. It is note-worthy that the barrel folds of domains III and IV resemble those of the first and third domains, respectively, of the monomeric galactose oxidase, which contains another type of posttranslationally generated redox cofactor, the Cys-Tyr adduct, in which Cys and Tyr are covalently coupled to each other by a thioether bond between the sulfur atom and the phenyl ring (25,26).
The ␤-subunit is folded into seven motifs of four up-anddown antiparallel ␤-strands, which are arranged around a pseudo seven-fold axis, giving a seven-bladed propeller. The Nand C-terminal ␤-strands of each blade are located on the inner and outer side, respectively, of the ␤-propeller, and the Cterminal strand is connected to the N-terminal strand of the neighboring blade. The ␤-propeller fold has also been observed in galactose oxidase (25) and in quinoproteins such as TTQcontaining methylamine dehydrogenase (27) and PQQ-dependent glucose (28), methanol (29) and ethanol (30) dehydrogenases. A structure similarity search (31) showed that the large ␤-propeller subunit of methylamine dehydrogenase is quite similar to the ␤-subunit of QH-AmDH.
Unique Structure of Catalytic Subunit-The most surprising finding revealed by this work concerns the structure of the small ␥-subunit. As deduced from the DNA sequence (10), this 79-residue subunit contains a total of four Cys and five Trp residues. However, neither free SH groups nor S-S bridges were detected in it by chemical analysis (10). Furthermore, during the initial stage of the structure determination by x-ray crystallography, modeling of the ␥-subunit puzzled us with its main chain branching off at many points in the electron density map. Complete modeling of the ␥-subunit became possible only  after the 2F o Ϫ F c electron density map was calculated at 1.9-Å resolution using the phases improved on the basis of the models of the ␣and ␤-subunits. The quality of the omit electron density map reached the point at which the side chains were identified from the shape of the electron densities contoured at the 1-level. In this stage, the side chains of all the Cys residues in the ␥-subunit were found to have unusually close contacts with a Trp, Asp, or Glu residue. For example, the S␥ atom of Cys-37␥ was within the covalent bond distance with the indolyl C4 atom of Trp-43␥. Also, extra electron densities were found to protrude from the C6 and C7 positions of the indolyl group of Trp-43␥, which were assigned to oxygen atoms based on the previous finding that the ␥-subunit contains a quinone group similar to, but not identical with, TTQ (7). Consequently, the crystal structure has revealed that one (Cys-37␥) of the four Cys residues is covalently cross-linked to the side chain of Trp-43␥ (at C4 of the indole ring), which is modified to an ortho-quinone, resulting in the novel quinone cofactor CTQ (Figs. 1B and 2A). Furthermore, the other three Cys residues are involved in thioether cross-bridges with an Asp or Glu residue, all in a hitherto unknown bonding with a methylene carbon atom of acidic amino acid side chains: Cys-7␥-Glu-16␥ (C ␥ ) (Fig. 2B), Cys-27␥-Asp-33␥ (C ␤ ) ( Fig. 2A), and Cys-41␥-Asp-49␥ (C ␤ ) (Fig. 2C). Crystallographic identification of these cross-linked structures greatly enabled us to interpret the results obtained by the chemical and mass spectrometric analyses of the ␥-subunit, as reported recently (10). Besides the novel CTQ cofactor encaged by multiple internal cross-bridges, another intriguing feature is that the ␥-subunit scarcely contains regular secondary structures as it has only two short ␣-helices with many turns and bends (Fig. 1B). A schematically drawn structure of the ␥-subunit (Fig. 2D) reminds us of peptide antibiotics with an internal loop rather than of a protein. Indeed, in a structure similarity search using the DALI calculation (31), no structure with Z scores of higher than 2.0 was found, indicating that the polypeptide fold of the ␥-subunit is unique. It thus appears that the multiple cross-links play at least a structural role in maintaining the globular structure of the small ␥-subunit polypeptide.
The ␥-subunit occupies the space formed between the interface of the ␣and ␤-subunits with 11 N-terminal residues (Ala-3␥, Cys-7␥-Thr-11␥, Val-17␥-Gly-21␥) and 29 C-terminal residues (Met-51␥-Lys-79␥) exposed to the solvent on both sides of the molecule. The three cross-bridges of Cys to Asp or Glu contained in 48 N-terminal residues (Ala-3␥-Met-50␥), together with numerous internal hydrogen bonds, lead to a compact ␥-subunit and an encaged CTQ. As compared with the C-terminal part, this region is rich in hydrophobic and acidic residues. CTQ resides in the vicinity of the interface between the ␤and ␥-subunits and near the pseudo seven-fold axis of the ␤-subunit, directing its C6 carbonyl group toward the interface. The topology of CTQ, heme I, and heme II (Fig. 1B) suggests that intramolecular electron transfer occurs from sub- strate-reduced CTQ to heme II via heme I followed by intermolecular electron transfer from heme II to azurin (7). The fact that the quinone ring of CTQ is separated by about 7.8 Å from the edge of heme I with a dihedral angle of 19°with respect to each plane supports this view. Active Site Cavity and Presumed Catalytic Base-On close inspection of the ␥-subunit structure, the active site cavity seems to be located at the interface between the ␤and ␥-subunits, surrounded by CTQ 43␥, the side chains of Pro-13␥, Asp-33␥, Pro-40␥, and Trp-42␥ from the ␥-subunit, and the side chains of Leu-198␤, Phe-200␤, Trp-201␤, Phe-258␤, Tyr-298␤, and Thr-341␤ from the ␤-subunit (Fig. 3A). Inside the cavity, two water molecules (W1 and W2) are found. A hydrogenbonding network exists between the C6 carbonyl group of CTQ, the carboxylate group of Asp-33␥, W1, and W2, and the hydroxyl group of Tyr-298␤. The N1 nitrogen and C7 carbonyl oxygen of CTQ are hydrogen-bonded to the main chain carbonyl oxygen of Thr-10␥ and amide nitrogen atoms of Asp-12␥ and Gly-14␥, respectively. To determine the catalytic site of CTQ where substrate amines react (either the C6 or the C7 carbonyl), the structure of enzyme inhibited by pNPH (7) was also determined at a 2.0-Å resolution (Tables I and II). The difference Fourier map clearly indicated a positive electron density peak in the active site cavity, which could be modeled nicely by the hydrazone of CTQ with pNPH bound to the C6 position and the phenyl ring accommodated in the hydrophobic cavity, displacing the two water molecules (Fig. 3B). Thus, in the catalytic process, the substrate amine presumably reacts with the C6 carbonyl group of CTQ, forming a Schiff base intermediate similar to the proposed reaction mechanism of TTQ-dependent methylamine dehydrogenase (32). Because Asp-33␥ is the sole residue close enough to the C6 carbonyl oxygen of CTQ and the Schiff base imine nitrogen, it could act as a catalytic base abstracting the ␣-proton from the CTQ substrate Schiff base. The hydrophobic environment around Asp-33␥ could be helpful in raising the pK a value of the carboxylate group, thereby enhancing its catalytic function. It should also be noted that the conformational flexibility of the carboxylate group of Asp-33␥ is largely restricted by the cross-link from Cys-27␥ to this residue.
Implication for Cofactor and Cofactor Subunit Biogenesis-The unique active site structure of QH-AmDH raises a number of questions: Why, when, where, and how are the CTQ cofactor and the multiple internal cross-links in the ␥-subunit generated? In the absence of further information, these questions cannot be answered yet. However, sequencing of the gene cluster encoding QH-AmDH of P. putida has already provided a clue for how the biogenesis could take place (10). In the stretch containing the structural genes for the three subunits, a hypothetical 53-kDa protein was found whose sequence is homologous to that of "sulfatase-activating enzyme," containing an iron/sulfur cluster and participating in the oxidative formation of the formylglycine cofactor in the active site of sulfatase (33). By analogy, the hypothetical 53-kDa protein could play a role in the post-translational formation of CTQ and/or the thioether bonds in the proenzyme of QH-AmDH.