P. M. H. Kroneck, Fachbereich Biologie, Mathematisch-Naturwissenschaftliche Sektion, Universität Konstanz, 78457 Konstanz, Germany Fax: +49 7531 88 2966 Tel: +49 7531 88 4112 E-mail: firstname.lastname@example.org U. Ermler, Max-Planck-Institut für Biophysik, Max-von-Laue-Str. 3, 60438 Frankfurt, Germany Fax: +49 69 6303 1002 Tel: +49 69 6303 1054 E-mail: email@example.com
The thiamine diphosphate (ThDP) dependent flavoenzyme cyclohexane-1,2-dione hydrolase (CDH) (EC 188.8.131.52) catalyses a key step of a novel anaerobic degradation pathway for alicyclic alcohols by converting cyclohexane-1,2-dione (CDO) to 6-oxohexanoate and further to adipate using NAD+ as electron acceptor. To gain insights into the molecular basis of these reactions CDH from denitrifying anaerobe Azoarcus sp. strain 22Lin was structurally characterized at 1.26 Å resolution. Notably, the active site funnel is rearranged in an unprecedented manner providing the structural basis for the specific binding and cleavage of an alicyclic compound. Crucial features include a decreased and displaced funnel entrance, a semi-circularly shaped loop segment preceding the C-terminal arm and the attachment of the C-terminal arm to other subunits of the CDH tetramer. Its structural scaffold and the ThDP activation is related to that observed for other members of the ThDP enzyme family. The selective binding of the competitive inhibitor 2-methyl-2,4-pentane-diol (MPD) to the open funnel of CDH reveals an asymmetry of the two active sites found also in the dimer of several other ThDP dependent enzymes. The substrate binding site is characterized by polar and non-polar moieties reflected in the structures of MPD and CDO and by three prominent histidine residues (His28, His31 and His76) that most probably play a crucial role in substrate activation. The NAD+ dependent oxidation of 6-oxohexanoate remains enigmatic as the redox-active cofactor FAD seems not to participate in catalysis, and no obvious NAD+ binding site is found. Based on the structural data both reactions are discussed.
Database Structural data are available in the Protein Data Bank database under accession number 2PGO
Thiamine diphosphate (ThDP) dependent enzymes use common structural and mechanistic principles to catalyse a broad range of C–C, C–N, C–S and C–O formation/cleavage reactions [1,2]. Detailed structural information is available for several members of this family, such as transketolase , pyruvate oxidase  and pyruvate decarboxylase , which were its first structurally characterized representatives. All of them are organized as homodimers or homotetramers (dimer of dimers) whereby each monomer is composed of three domains of the α/β type. ThDP, the biologically active derivative of vitamin B1, binds in a V-shaped conformation into a cleft between two subunits which defines the homodimer as essential for catalytic activity . The catalytic power of most ThDP dependent enzymes is based on an inversion of the polarity of the substrate carbonyl group from a partially positive carbon atom into a potentially electron-rich centre capable of donating electrons. This process, called ‘Umpolung’ is realized by covalent coupling between the deprotonated/activated C2 atom of the thiazolium ring of ThDP and a carbonyl carbon of the substrate. The required ThDP-ylide is formed after abstraction of the C2 proton by the neighbouring 4′-imino group of the pyrimidine ring of ThDP . The energetically unfavoured 4′-imino state is stabilized relative to the 4′-amino state by a proton relay chain between the 4′-N and the 1′-N sites supported by a conserved glutamate residue that abstracts the additional 1′-N proton of the 4′-imino tautomer [7,8]. Solely in glyoxylate carboligase this glutamate is replaced by valine . Many ThDP dependent enzymes exhibit hysteretic activation of ThDP and negative cooperativity in binding their substrates  which presuppose a communication between the two spatially separated active sites of the dimer. Several coupling mechanisms have been postulated for enzymes using a ping-pong type mechanism in which the two half reactions proceed alternately in the two active sites like a two-stroke engine .
Despite their relative structural and mechanistic uniformity ThDP dependent enzymes participate in many different biosynthetic pathways in all kingdoms of life by various lyase and ligase reactions ; some of them open promising perspectives for biotechnological applications . In addition, some members of this enzyme family couple the ThDP dependent reaction with a redox reaction and therefore bind close to the ThDP molecule a redox cofactor that is frequently FAD but in special cases also lipoic acid or an Fe/S cluster . However, in a few ThDP/FAD dependent enzymes, such as acetohydroxy acid synthase and glyoxylate carboligase, FAD appears to solely maintain the structural integrity of the protein, without direct participation in catalysis [15,16].
Our studies are focused on cyclohexane-1,2-dione hydrolase (CDH) isolated from Azoarcus strain 22Lin, which is the first ThDP dependent enzyme characterized to date that cleaves an alicyclic ring compound [17,18]. A similar cleavage of a cyclic compound, the conversion of 3D-(3,5/4)-trihydroxycyclohexane-1,2-dione to 5-deoxy-d-glucuronic acid, was described as a reaction step of the myo-inositol catabolism in Bacillus subtilis . The enzyme encoded by iolD showed the highest relationship to the ThDP dependent enzymes acetolactate synthase from Spirulina platensis and Synechococcus sp. (26.4% and 26.0% sequence identity) . However, this enzyme has not yet been purified and characterized. In the CDH reaction cyclohexane-1,2-dione (CDO) is hydrolysed to 6-oxohexanoate and the latter is subsequently oxidized to adipate at the expense of NAD+ (Fig. 1). CDH follows Michaelis–Menten kinetics (kcat,app = 1.6 ± 0.1 s−1, KM,app = 13.3 ± 0.1 μm, and kcat,app/KM,app = 1.2 ± 0.1 × 105 s−1·m−1 for CDO, and kcat,app = 1.5 ± 0.1 s−1, KM,app = 164 ± 31 μm, kcat,app/KM,app = 0.91 × 104 s−1·m−1 for NAD+) and contains one ThDP, one Mg2+ and one FAD molecule per monomer  in agreement with the primary structure fingerprints for the ThDP and FAD binding sites [21,22]. As reported for the acetohydroxy acid synthase and glyoxylate carboligase [15,16] the FAD cofactor resists reduction under normal physiological conditions but can be partly reduced with the deazaflavin/oxalate system (E°′ = −0.65 V). Sequence comparison studies revealed acetohydroxy acid synthases, glyoxylate carboxylase and cytochrome pyruvate dehydrogenases as closest relatives of CDH, the sequence identities being in the range of approximately 25%. CDH is the key enzyme of a novel anaerobic degradation pathway [17,23] of alicyclic alcohols as the conventional oxidation and hydrolysis pathway is not feasible due to the 1,2 position of two hydroxyl groups . Cyclohexane-1,2-diol is a degradation product of steroids, terpenes and other cyclic compounds produced in large amounts by, for example, plants as secondary metabolites. The diol group is oxidized in a first step to the corresponding diketone by two NAD+ dependent oxidation reactions forming CDO, the substrate of CDH.
In the present work we describe the structure, the cofactor binding and the architecture of the active site funnel of CDH in comparison with other ThDP dependent enzymes. The binding characteristics of the competitive inhibitors 2-methyl-2,4-pentane-diol (MPD) and chloride anion mimicking CDO binding allow the identification of catalytic relevant residues and the postulation of an enzymatic mechanism. The binding of MPD exclusively to one substrate binding funnel of the CDH dimer and the NAD+ dependent oxidation of 6-oxohexanoate to adipate are discussed.
Results and Discussion
The crystal structure of CDH from Azoarcus sp. strain 22Lin has been determined at 1.26 Å resolution by the multiple isomorphous replacement with anomalous scattering (MIRAS) method and refined to Rcryst and Rfree factors of 10.0% and 12.2%, respectively (Table 1). Accordingly, CDH is a homotetrameric enzyme that is composed of a dimer of two dimers arranged in a pseudo 222 symmetric form. The two monomers of the functional dimer forming two complete active sites are referred to as A and B. Each monomer is built up of the domains α (residues 2–182), β (183–372) and γ (373–588); all three are folded as an open α/β structure with six β-strands flanked by α-helices (Fig. 2). This structural organization is characteristic for ThDP-containing enzymes [3,25] which vary, however, in their quaternary structures. Some members are present as dimers and others as tetramers, the latter with variable dimer–dimer orientations. For example, the dimer–dimer orientation deviates by about 15° between CDH and Lactobacillus plantarum pyruvate oxidase .
Table 1. Data collection and refinement statistics. ‘Pip’, di-μ-iodobis(ethyldiamine)-di-platinum nitrate; ‘terpi’, platinum(II) (2,2′:6′,2″-terpyridine) chloride. Rsym = Σ ∣Ii − 〈l〉 ∣/ΣIi, where Ii is the observed intensity and 〈l〉 is the averaged intensity obtained from multiple observations of symmetry-related reflections. Phasing power (iso) is Σn∣FH∣/Σn∣E∣where ∣FH∣ is the calculated structure factor amplitude of the heavy atom structure and E is the lack of closure error; phasing power (ano) is 2Σn∣FH″∣/Σn∣E∣where ∣FH″∣is the anomalous contribution amplitude. Rcryst = Σhkl(∣Fobs∣ − ∣Fcalc∣)/Σhkl∣Fobs∣. Rfree = Rcryst with 5% of the observed reflections selected randomly.
ID14-4 High res.
ID14-4 Low res.
ID14-4 HgAc2 peak
ID14-4 K2PtCl4 peak
ID29 cubic crystals
Soaking conditions (mm; h)
Cell constants (Å)
Solvent content (%)
I/σI (last shell)
Phasing power (iso) (centrics/acentrics)
Phasing power (ano) (centrics/acentrics)
No. of reflections
No. of protein atoms
Average B-factor (Å2), protein, ThDP, FAD
9.8, 4.7, 4.9
Bond length rms (Å)
Bond angle rms (°)
Structure comparison studies of the monomers using the Dali server  classified CDH as a member of the pyruvate oxidase subfamily documented in low rmsd of 2.3 Å, 2.4 Å, 2.3 Å and 2.3 Å (for 93%, 91%, 89% and 85% of the Cα atoms) to Arabidopsis thaliana acetohydroxy acid synthase , Escherichia coli glyoxylate carboligase , Pseudomonas fluorescens benzaldehyde lyase  and E. coli pyruvate dehydrogenase , respectively. The sequence identity between CDH and the mentioned most related ThDP enzymes is 23%, 23%, 24% and 22%, respectively. Substantial differences between them exist solely at their 40 C-terminal residues (Fig. 2). In CDH, the segment between residues 550 and 564 is an essential component of the active side funnel (see below) whereas the following stretch (565–581) as well as a short 310-helix (582–586) link the C-terminal segment with the opposing monomer and the second dimer. The oligomeric contact area is substantially increased by the two C-terminal stretches (565–581) from different dimers that are arranged in a parallel manner over a length of 30 Å (Fig. S1). The interface area buried by the C-terminal arm (565–586) is approximately 1550 Å2 which is more than 25% of the total interdimer contact area.
The active site in front of the C2 atom of ThDP is located at the bottom of a 15 Å long, partly narrow funnel (Figs 2, 3) composed of a unique hydrophobic region at its entrance and a more hydrophilic base. The latter is mainly formed by loops following strands B21:B25, B71:B76 and B96:B102 of domain α, by parts of helix A482:A491 of domain γ and by the cofactors ThDP and FAD. The hydrophobic region is composed of the loop connecting strand A253:A256 and helix A263:A271 of helix A482:A490 and, in particular, of a semi-circularly arranged loop close to the C-terminal segment (Fig. 3). This funnel region is coated by apolar side chains of PheA259, LeuA487, LeuA551, ProA561, LeuA563 and LeuA564 that also define the size of the entrance. Interestingly, the funnel architecture in CDH is designed significantly differently than in other ThDP dependent enzymes where the funnel instead corresponds to a cleft characterized by a wide entrance which is mostly displaced by 5–10 Å and lies directly at the monomer–monomer interface (Fig. 3). In CDH, this access is essentially blocked by the semi-circularly shaped loop and the bulky side chain of TrpA285. These new features provide the molecular basis to enlarge the substrate specificity of the ThDP dependent enzyme family to alicyclic diketones. A comparison between CDH, benzaldehyde lyase and benzoylformate decarboxylase indicates no correlation between the geometry of the substrate binding funnel and their similarly large substrates.
Notably, only one substrate funnel is open and provides an oval entrance with a diameter of 7 Å which is sufficiently wide to be passed by CDO (Fig. S1). The other funnel is closed due to small rearrangements (below 2.5 Å) of PheA259, LeuA487, LeuA551 and LeuA563 positioned at the bottleneck of the substrate funnel. These define a cavity that is inaccessible for bulk solvent and contains sufficient space to accommodate the substrate CDO (Fig. 3). In the structure derived from the cubic crystal form the size of the entrance lies between those observed in the tetragonal form.
Despite the architectural differences of the substrate binding sites the shuttling between open and closed conformations and the crucial function of the C-terminal segment are also reported for other family members. For example, in yeast acetohydroxy acid synthase  the partly disordered C-terminal segment in the open state becomes ordered in the closed state to a two-stranded β-sheet followed by a loop-helix structure. Likewise, in oxalyl-CoA decarboxylase  a similar rigidification of the C-terminal segment is postulated. In contrast, in CDH both the open and the closed states are conformationally well defined, presumably due to the stabilization of the semi-circularly arranged loop by the C-terminal arm firmly anchored to other monomers of the tetramer (Fig. S1). This fixation concept opens perspectives for a specific reconstruction of the mentioned pronounced loop to realize new substrate specificities for ThDP dependent enzymes in a region between two rigid entities.
Cofactor binding sites
ThDP is strongly anchored to the protein matrix by a divalent metal ion determined as Mg2+ in CDH by inductively coupled plasma mass spectrometric measurements . Mg2+ ligates to two phosphate oxygens of the pyrophosphate tail, a water molecule, the main chain oxygen of SerA480 and the side chains of AspA451 and AsnA478, the latter being constituents of the conserved ThDP fingerprint motif (Fig. 4) . The thiazolium and methylaminopyrimidine rings are arranged in the characteristic V conformation held in place by a large hydrophobic aliphatic side chain as a leucine in CDH, transketolase  and benzoylformate decarboxylase  that projects between the two rings and contacts both. The V conformation forces the N4′ atom of pyrimidine in close proximity to the C2 atom of the thiazolium (3.2 Å in CDH) that favours an abstraction of the C2 proton by the N4′ atom and might destabilize, in parallel, the simultaneous presence of an amino form at N4′ and a protonated C2 atom . An invariant hydrogen bond between the GlyA424 carbonyl oxygen and the 4′-amino/imino group (2.8 Å in CDH) may assist by fine-tuning the position of the imino group (Fig. 4). The conserved residue GluB52 stabilizes the 4′-imino tautomeric form of the pyrimidine ring [7,8] by a strong hydrogen bond between the OE1 and N1′ atoms (their distance is 2.7 Å in CDH). Interestingly, in CDH (and in benzaldehyde lyase) the hydroxyl group of TyrA456 of the partner monomer is hydrogen bonded to the OE2 atom of GluB52, thereby adjusting its position and its pKa value (Fig. 4). The OE1 atom is further linked to the polypeptide by invariant interactions.
The ThDP conformation and protein–ThDP interactions can be described very accurately due to the excellent resolution of the CDH structure. Although refinement was performed without any restraints between the ThDP atoms regarding their bond lengths and bond angles, the two ThDPs in the asymmetric unit are nearly identical (Fig. S2). The asymmetry of the two active sites suggested by the selective MPD binding (see below) is not reflected in the ThDP geometry at a resolution of 1.26 Å. Likewise, no significant bond length differences are found between enzyme-bound ThDPs and between the enzyme-bound ThDP and free ThDP derived from high-resolution X-ray structures (Fig. S2).
The FAD cofactor is bound in an extended conformation at the C-terminal end of the six-stranded β-sheet of the domain β as reported for pyruvate oxidase  and acetohydroxy acid synthase . FAD binding in CDH shows the characteristic features of the pyruvate oxidase structure family  where the conserved motif GxGxxG for dinucleotide binding is absent and the adenosine moiety of FAD binds to the second βαβαβ unit in the opposite direction. Nevertheless, the FAD binding mode varies substantially among the family members (in contrast to the strictly conserved ThDP binding) suggesting that its presence but not its exact binding characteristics is required. First, the structure of the irregular segment following strand A274:A279 in subunit β is substantially rearranged which considerably influences the location and orientation of the isoalloxazine ring. Second, the isoalloxazine ring of FAD is planar in CDH and acetohydroxy acid synthase but bent by 15° across the N5–N10 axis in pyruvate oxidase. Third, solely in CDH FAD is largely shielded by the polypeptide scaffold. The access to the catalytic N5 atom is blocked by SerA240, PheA259, CysA260, TrpA285, PheA406 and LeuA551.
Substrate binding site
Kinetic studies under experimental conditions of crystal growth (40% (v/v) MPD, 15 mm Na+ acetate, 200 mm NaCl, 1 mm ThDP, 1 mm NAD+, 5 mm MgCl2 and 100 μm CDO in 5 mm HEPES buffer, pH 7.0) revealed no specific activity for CDH raising the question which of the three main components of the crystallization buffer – NaCl, MPD or Na+ acetate – is inhibiting catalysis.
The strongest inhibitory effect was observed for MPD (Fig. 5A), with 5% MPD leading to a decrease of the specific activity close to 60%; in the presence of 20% MPD, just 5% activity remained. This finding is reflected in the CDH structure where an MPD molecule is embedded into the potential substrate binding site of the open funnel (Fig. 3A). MPD is oriented in such a manner that its apolar part is directed towards the hydrophobic entrance. It contacts besides GlyA399 the apolar side chains of PheA259, TrpA285, LeuA551, LeuA487 and LeuA563. The first hydroxyl group of MPD is hydrogen bonded to HisB28 and the second interacts with the catalytic active C2 atom of ThDP, both at distances of 3.1 Å. Soaking experiments with substrate (3.7 mm CDO, 17 h) or co-crystallization experiments with up to 2 mm CDO did not lead to a displacement of the MPD molecule by the substrate. The strong binding of MPD to the active site is not too astonishing, since (a) MPD preferentially binds to hydrophobic sites , and (b) MPD resembles the monohydrated form of CDO not only in size but also with regard to the arrangement of the functional diol/dione substituents.
Interestingly, MPD is not present inside the closed substrate funnel that contains instead of the inhibitor three water molecules (Fig. 3B). This structural asymmetry in the two opposing active sites might be explained by a negative cooperativity with respect to the binding of MPD that would be in line with the concept of two communicating active sites in some ThDP enzymes . A coupling between the active sites is also supported by the catalytic inactivity of CDH in an MPD containing solution although one active site funnel is not occupied by MPD (albeit blocked for the substrate CDO in the crystalline state). Binding of substrate analogues exclusively into one active site were also found for yeast pyruvate decarboxylase  and the E1 component of Thermus thermophilus pyruvate dehydrogenase . A mechanism of communication between the two active sites is not obvious in CDH because the ∼ 20 Å long route between the aminopyrimidine rings does not appear to be optimized for an efficient proton transfer that is necessary for running two alternating catalytic cycles coupled by long-range acid/base catalysis [10,11]. In CDH this route is only lined up with a few protonable residues and not continuously water-filled as reported for the E1 component of pyruvate dehydrogenase  but rather blocked by Leu53, Leu82, Ala85, Ile425 and Tyr456 of both subunits.
Enzymatic activity studies with CDH indicated a significant but weaker inhibition by NaCl compared with MPD (Fig. 5). With 400 mm NaCl the specific activity was reduced to 10% which can be correlated with the structural finding of a putative chloride ion 3.3 Å apart from the C2 atom of ThDP in both substrate funnels (Figs 3 and 4 ). This assignment is consistent with the composition of the crystallization solution, the height and the spherical shape of the 2Fo − Fc electron density, the bond distances between 3.15 and 3.3 Å to neighbouring atoms, and the positively charged character of the protein environment. Polar interactions are formed between the chloride anion and the C2 and N4′ atoms of ThDP, the side chains of HisB76 and GlnB116 and a water molecule (Fig. 4). Notably, the chloride ion also interacts with the hydroxyl groups of MPD (distances 3.4 Å and 3.6 Å) in the open funnel that presumably enhances the affinity of both MPD and chloride in a synergistic fashion (Fig. 3A). The positions of the chloride ion and MPD might be normally occupied by the substrate CDO whose binding mode has, so far, not been structurally characterized. It is very likely, however, that HisB28, HisB31, HisB76, GlnB116 and AsnA484 are crucial for substrate binding and activation.
In contrast, Na+ acetate substantially influences the enzyme activity only at concentrations higher than 15 mm and is not responsible for the inactivity of the enzyme in the crystallization solution (data not shown). However, at higher concentrations, it might occupy the assumed binding site for the chloride anion. Additionally, we found one MPD molecule, one chloride anion and one acetate at the dimer–dimer interface near crystal packing contacts emphasizing their importance for crystal formation.
Towards the reaction mechanism of CDH
CDH catalyses two reactions: the hydrolytic cleavage of the CDO ring to 6-oxohexanoate and the NAD+ dependent oxidation of the semialdehyde 6-oxohexanoate to adipate (Fig. 1). Although the binding of CDO could not be experimentally established we assume that the orientation of CDO is largely determined since its hydrophobic portion probably points to the hydrophobic funnel entrance and its polar carbonyl groups to the hydrophilic base as found for MPD. The chloride ion might occupy the position of the C1 atom of CDO which is negatively charged in an intermediate state of the established catalytic cycle of ThDP dependent enzymes . In line with the mechanism reported for other family members [1,2] we postulated a nucleophilic attack of the C2 anion of activated ThDP onto the carbonyl group of CDO, a subsequent cleavage of the aliphatic ring and the release of the product 6-oxohexanoate accompanied by proton transfer steps (Fig. 6). HisB28, HisB76 and GlnB116 are attractive candidates to increase the electrophilic character of the attacked carbonyl carbon of CDO and to supply/uptake a proton towards/from the substrate–ThDP adduct during 6-oxohexanoate formation (Fig. 4).
The NAD+ dependent reaction does not appear to proceed as expected since several experimental observations strongly suggest that the FAD cofactor does not participate in the redox process. First, with known reductants including ascorbate, NADH, Na+ dithionite, Ti(III) citrate or NaBH4, FAD cannot, or can only partially, be reduced . Second, the N5 atom of the isoalloxazine ring is completely shielded by residues PheA259 and CysA260 and therefore inaccessible for both NAD+ and 6-oxohexanoate, the latter being about 5 Å away from FAD. Third, in the structurally related ThDP dependent flavoenzymes glyoxylate carboligase and acetohydroxy acid synthase [15,16] (but not in pyruvate oxidase) the flavin cofactor is considered as an evolutionary relict that is important for structural reasons but does not participate in a redox reaction. This is not too uncommon and has been reported for other flavoenzymes . However, solely in CDH, the paradox situation exists that a redox reaction is performed by the enzyme carrying FAD at the active site but without its participation. Although the current data clearly point to a direct hydride transfer between 6-oxohexanoate and NAD+ and not to stepwise electron transfer via FAD (as proved for pyruvate oxidase and acetohydroxy acid synthase ), there is no space for NAD+ in the active site funnel. Thus, either larger conformational changes to let NAD+ in or a movement of 6-oxohexanoate out of the funnel must occur. Undoubtedly, the elucidation of the mechanism of the NAD+ dependent reaction requires further studies, in particular the determination of the structure of the CDH–6-oxohexanoate–NAD+ ternary complex.
Material and methods
Purification, enzymatic assay and crystallization
CDH was purified from Azoarcus sp. strain 22Lin  and stored in 50 mm MES buffer, pH 6.5, containing 1 mm MgCl2 and 0.5 mm ThDP. Enzymatic activity was determined by monitoring the reduction of NAD+ at 365 nm. Assays contained 100 mm HEPES buffer, pH 8.0, 1 mm NAD+ and 0.1 mm CDO. After incubation for 5 min at 37 °C, reactions were started by addition of CDH . Crystals grew at 25 °C by the hanging-drop method. The drop consisted of equal volumes of protein storage and reservoir solution, the latter composed of 60% MPD, 20 mm Na+ acetate and 200 mm NaCl. The tetragonal crystals belong to the space group P41212 with axes lengths a = 123.6 Å and c = 144.6 Å. A second cubic crystal form was obtained at 20 °C with a reservoir solution containing 17–18% (w/v) poly(ethylene glycol) (PEG 8000), 100 mm Na+ acetate, 10 mm MnCl2 and 2% (w/v) isopropanol in 0.1 m HEPES buffer, pH 7.5.
Data collection and phase determination
Native and MAD/SAD data were collected at the ID29 and ID14-4 beamlines at the European Synchrotron Radiation Facility (ESRF) (Grenoble) and MIR data inhouse using a Rigaku 200 rotating anode generator (see Table 1). Data processing was performed with the hkl and xds packages [39,40]. Initially, two mercury positions were determined from the peak data set measured from CDH crystals soaked with Hg(II) acetate (shelxd ). Three platinum sites from CDH crystals soaked with K2PtCl4 were obtained from anomalous difference Fourier calculations. Due to the low occupancy of both heavy atom compounds two additional derivate data sets with ‘pip’ and ‘terpy’ soaked crystals were necessary for successful phasing (Table 1) using sharp . The figure of merit was 0.57 (0.58) for the centric (acentric) reflections over the resolution range 20–4 Å. Phases were improved by the solvent flattening procedure of solomon  implemented into sharp. On the basis of the resulting electron density map, the position of the heavy atoms and an approximate superposition with L. plantarum pyruvate oxidase one monomer could be masked and the non-crystallographic symmetry operator determined. After two-fold molecular averaging using dm  30% of the polypeptide chain could be automatically incorporated into the electron density map with maid . The missing residues were built within o . Data of the cubic crystal form were phased by the molecular replacement method .
Refinement and quality of the structure
The resulting model was initially refined with cns-1.1 , using established molecular dynamics and least-squares minimization protocols. Non-crystallographic symmetry restraints were not applied. Water molecules and alternative amino acid conformations were incorporated and anisotropic temperature factor values were calculated with shelxl-97  (Table 1). The final model consists of 2 × 587 amino acid residues, two FAD molecules, two ThDP molecules, two Mg ions and 1185 water molecules. Additionally, three chloride ions, three MPD molecules and one acetate molecule from the crystallization buffer were detected in the crystal structure of CDH. A second conformation was obtained for 68 amino acid residues. According to procheck  no dihedral angle of non-glycine and non-proline residues was located in disallowed regions. The structure of the cubic crystal form was refined using refmac5 .
This work was supported by the Max-Planck-Gesellschaft (JH, UE) and by the Deutsche Forschungsgesellschaft (PK, UE). We thank Christina Probian for the cultivation of Azoarcus, the staff of the beamlines ID29 and ID14-4 at ESRF (Grenoble) for help during data collection and Hartmut Michel for continuous support.