We have determined the crystal structure of recombinant vvUDG by SIRAS phasing in trigonal space group P3₂21. The structure was refined to an R value of 24.1% (R_free of 29.9%) at 2.4 Å resolution (see Tables 1 and 2). The recombinant protein used for crystallization contained an N-terminal His-tag. The final model consists of two subunits A and B with a total number of 440 protein residues (subunit A: residues -1 to 171, 174 to 186 and 189 to 218; B: -8 to 172, 174 to 186 and 189 to 218), 1 chloride and 1 sulfate ion, 146 water, 4 glycerol (GOL) and 3 imidazole (IMD) molecules. Parts of the N-terminal His-tag are visible in both subunits, A (2 residues: -1 to 0) and B (9 residues: -8 to 0). Several residues in two loop regions could not be fitted into the electron density in each subunit (A: 172,173, 187,188; B: 173, 187,188) presumably due to disorder. In addition, some residues have truncated side chain density in both subunits (A: -1, 0, 170, 171 and 195; B: -8, -6, 0, 185 and 195). 3 residues in subunit B (S7, N17 and Q203) show alternate conformations. The Ramachandran plot shows 96.3% of all protein residues in the allowed regions, with 12 residues (3.2%) in generously allowed and 2 residues (0.5%) in disallowed regions. Electron density for several of the residues in generously allowed regions (A9, Y11, F79 and N206 in both subunits) and the two residues in the disallowed region (R167 and A168 in subunit B) is good. Of these residues A9 and Y11 are part of a β-hairpin turn connecting strands 1 and 2. N206 is in the turn following helix 9 (residues 189–205) in the C-terminal part of the protein. Refinement statistics are shown in Table 2.

We also crystallized vvUDG (without His-tag) in orthorhombic space group P2₁2₁2₁. This structure was determined by molecular replacement using the refined model of the trigonal crystal form, and was refined to an R value of 25.3% (R_free of 30.2%) at 2.3 Å resolution (Tables 1 and 2). The final model consists of 8 subunits A through H with a total number of 1708 residues (subunits A: 214 residues; B: 212; C: 215; D: 215; E: 216; F: 214; G: 213; H: 209), 1 chloride ion, 573 water, 2 Hepes (EPE) and 18 glycerol (GOL) molecules. Several residues in the same two loop regions could not be fitted into the electron density in each subunit (A: 185–188; B: 168,169, 186–189; C: 187–189; D: 193–195; E: 185,186; F: 164, 165, 186,187; G: 188–192; H: 171–173, 185–190) and are missing from the final model. The Ramachandran plot shows 96.4% of all protein residues in allowed regions, while 2.6% of the residues are found in generously allowed and 1.2% (18 residues) in disallowed regions. Here, the same residues (A9, Y11, F79 and N206) with good electron density as mentioned for the trigonal crystal form are also found in generously allowed and disallowed regions. Refinement statistics are shown in Table 2. The average rms deviation for all Cα atoms between subunits of the two different crystal forms is ~0.5 Å.

The overall structure of the vvUDG protein in both crystal forms is similar. vvUDG adopts an α/β fold described as DNA glycosylase fold in the SCOP database 6 that is composed of a parallel β-sheet of 4 strands (order 2134) with 3 layers of α/β/α. In Fig. 1 the secondary structure assignments and topology diagrams for vvUDG are compared with those for human UDG. In vvUDG the central β-sheet is surrounded on either side by a total of 5 larger helices. Two helices are observed on one side (residues: 45–49, 50–55; 188–205), and three helices on the other side (residues: 19–32, 33–39; 86–101; 133–149). Additional features are seen at each terminus. At the N-terminus the vvUDG structure exhibits a β-sheet made up of two anti-parallel β-strands (residues: 1–7; 11–17). At the C-terminus the polypeptide chain folds back to form another small anti-parallel β-sheet (residues: 107–109; 215–217) and displays the pairing of two small helices (residues: 110–113; 211–214). The active site groove is visible at the C-terminal edge of the central parallel β-sheet. Active site residues D68, Y70, F79, N120 and H181 are lined up at the edge of the groove. The tertiary structure of a vvUDG subunit is shown in Fig. 2A.

The asymmetric unit in trigonal space group P3₂21 contains a dimer (type I) with subunits A and B related by non-crystallographic symmetry two-fold symmetry (rms deviation of 0.06 Å for all Cα atoms between individual subunits). This NCS related dimer is shown in Fig. 2B. The total buried solvent accessible surface area (SASA) is approximately 4%, corresponding to an interface area of 806 Å ² between individual subunits. Analysis of protein-protein interactions was performed using the ProFace server 78. In each subunit there are 15 interface residues. For subunit B the contact area is confined to the N-terminal residues 1–4, 6, 9–12 and 14, and residues 38, 45–47 and 54. For subunit A most of the interface residues are found in two regions (residues: 54–56, 58–60; 111–114) with a few additional residues at the termini (residues: 1–2, 16; 211, 214).

The packing of vvUDG in the unit cell of the trigonal form gives rise to a second type of dimer (type II) formed by subunits that are related by crystallographic 2-fold symmetry. The buried surface area in these dimers amounts to 6.5–7% of the total SASA, corresponding to an interface area of 1310 Å ² between individual subunits. This type II dimer interface has 16–18 residues (167–169; 175–178, 180; 190–191, 194–195, 198, 201–202, 204–206) from each subunit. The contact residues are in the large conserved C-terminal helix 9, the loop (residues 165–167) connecting strands 8 and 9 and in strand 9. Since strand 9 is part of a conserved parallel β-sheet in the central core, interactions involving this strand extend the four-stranded β-sheet to an anti-parallel eight-stranded β-sheet in the dimer. This type II dimer is also observed in the orthorhombic space group P2₁2₁2₁ (Fig. 2C), and is likely the physiological dimer observed in solution. In the orthorhombic crystal form, subunits A through H in the asymmetric unit are arranged as four dimers related by non-crystallographic 2-fold symmetry (rms deviation for all Cα atoms between individual subunits is 0.39–0.45 Å). Based on the subunit-subunit interactions these dimers are of type II as seen in the trigonal crystal form.

Considering that vvUDG is a dimer in solution (see Methods) the dimeric assembly in both crystal forms is unlikely to be an artifact of crystal packing. The protein-protein interactions in the dimers may be important in fulfilling vvUDG's role as a component of the DNA polymerase processivity factor. It is tempting to speculate that the interactions in the type I dimer only seen in the trigonal crystal form might mimic the interaction between UDG and A20 in the heterodimeric complex (see later) while the other set of interactions hold the homodimeric assembly of vvUDG.

vvUDG was crystallized in the absence of any substrate. Glycerol molecules from either the crystallization solution or the cryoprotecting reagent occupied the active site in both crystal forms. The quality of electron density for the active site residues and the glycerol molecules is excellent in each case (see Omit Map; Fig. 3A). Glycerol is an inhibitor of UDG and kinetic studies showed that 200 mM glycerol inhibited the reaction rate of E. coli UDG by ~50% 9. A glycerol molecule (from the cryoprotectant) was located in the uracil binding pocket in the crystal structure of E. coli UDG 9. In this structure the three hydroxyl groups of glycerol mimicked atoms O2, O4, and N3 of uracil (URA) in their interaction with the enzyme. Glycerol forms hydrogen bonds directly or through water molecules with active site residues (Q63, D64, Y66, F77, N123 and H187). In the vvUDG trigonal crystal form, one glycerol molecule is located in the active site of subunit B and occupies the same position as seen in E. coli UDG (see Fig. 3B). This glycerol makes interactions with three of the five active site residues (D68, F79 and N120) and their immediate neighbors (see Fig. 3A). In subunit A an imidazole molecule occupies the active site making interactions with D68 and displaying hydrophobic contacts with F79. The chloride ion (Cl^-) in subunit A shows distances of 3.3–3.5 Å to backbone nitrogen atoms of active site residues Y70 and F79 and the ND2 atom of N120. In subunits A, C, E and G of the orthorhombic crystal form one glycerol molecule is located in each active site and exhibits similar interactions with active site residues D68, Y70, F79 and N120 as described for subunit B of the trigonal crystal form. Details of the contacts involving these ligands in both crystal forms are listed in Table 3.

Figure 3 shows a comparison of the active sites of vvUDG (with bound glycerol) and E. coli UDG complexes (with bound glycerol and uracil). We have modeled a uracil molecule in the active site of vvUDG in an orientation as found in other crystal structures (Fig. 3C).

In both crystal forms additional glycerol molecules are located away from the active site. Contacts formed by non-active site glycerol molecules and other ligands are provided in Additional file 1.

The vvUDG sequence exhibits considerable differences in the characteristic motifs utilized by other UDGs for recognizing and flipping the uracil moiety in the substrate DNA during the catalytic activity (see Table 4 for a comparison of motifs in vvUDG with E. coli and human UDG). The vvUDG has five of the six conserved active site residues (D68, Y70, F79, N120 and H181), but lacks the conserved Leu residue (see Table 4). In other UDGs the Leu residue is part of the 'Leu intercalation loop', which has the characteristic motif (-HPSPLSXXR-). The 'intercalation loop' (also called catalytic loop) is substantially altered in poxvirus UDGs (see Table 4). Only 3 residues (H181, P182 and R187) match with residues in human and E. coli UDG. Residue R185 in vvUDG corresponds to L191 in E. coli UDG (see Fig. 4A).

Uracil DNA glycosylase catalyzes the hydrolytic cleavage of the N-glycosidic bond of premutagenic uracil residues in DNA by base flipping. Results from a study by Drohat et al. 10 in E. coli support a mechanism for catalysis that emphasizes catalytic residue Asp64 as the general base activating a water molecule for nucleophilic attack at C1' of the deoxyribose, and catalytic residue His187 as a neutral electrophile, stabilizing a developing negative charge on uracil atom O2 in the transition state. Stivers et al. 11 demonstrated that base flipping contributes little to the free energy of DNA binding but provides a substantial contribution to specificity through an induced-fit mechanism. For the binding of DNA substrates UDG uses a number of residues that are not part of the active site 12. According to Tainer et al. 131415 the DNA repair mechanism of UDG involves pinching of the phosphodiester backbone of damaged DNA using hydroxyl side chains of four conserved serine residues (S88, S166, S189 and S192 in E. coli UDG; S169, S247, S270 and S273 in human UDG). This results in flipping of the deoxyuridine from the DNA helix into the enzyme active site. These authors propose that strain induced by serine pinching is used to lower the activation barrier for glycosidic bond cleavage. Results based on S88A, S189A, and S192G "pinching" mutations described by Werner et al. 16 indicated a role for these serine-phosphodiester interactions in uracil flipping and preorganization of the sugar ring into a reactive conformation. The 'Pro-rich' and 'Gly-Ser' loops that contain Ser residues in other UDGs are missing in vvUDG. In addition, the two Ser residues are also missing from the 'Leu intercalation loop' (see Table 4). In the pinch-push-pull uracil detection mechanism, the conserved Leu residue of the 'Leu intercalation loop' penetrates into the DNA minor groove to push the uracil base into the active-site pocket. Based on the structure of the L272A complex of human UDG with DNA, the L272 side chain push is not essential for nucleotide flipping, although it plays a key role in efficient activity 13. Results of another study 17 suggest that the Leu residue within the -HPSPLS-motif is crucial for the uracil excision activity of UDG.

The side chain of this conserved Leu residue in UDG is also inserted into the hydrophobic cavity of the specific uracil-DNA glycosylase inhibitor (Ugi) from Bacillus subtilis 1819. Putnam et al. 18 pointed out that a significant fraction of the buried surface area (~10%) in the Ugi-complex results from the complementarity between the conserved Leu residue and the Ugi hydrophobic cavity. Poxvirus UDGs contain instead an Arg residue at this position in the 'Leu intercalation loop' (see Fig. 4A). It was shown that vvUDG activity is not inhibited by Ugi 3. Superimposition of vvUDG with E. coli UDG in the E. coli UDG-Ugi inhibitor complex (1UUG) provides a possible explanation for the lack of inhibition (Fig. 4B). Size and electrostatic property of the Arg residue are incompatible with insertion into the hydrophobic cavity of Ugi.

Since the conserved Leu residue and the 'Leu intercalation loop' are critical components for the conventional UDG catalytic mechanism, poxvirus UDGs may utilize a different yet unknown reaction mechanism for carrying out the DNA repair activity.

The conserved motif for uracil specificity (-LLLN-) in UDGs is also altered in the vvUDG protein (-IPWN-). Only N120 as part of the active site residues is conserved. Nonetheless, poxvirus UDG is still highly specific for uracil and does not act on other modified bases 1. In addition, the 'catalytic water-activating loop' is different in vvUDG. This loop (-GIDPYP-) shows two changes compared to the conserved motif (-GQDPYH-).

Two temperature sensitive mutants, which were mapped to the D4 ORF by Ellison et al. 3, have been described. Of these, Dts30 (G179R substitution), is of particular interest since this mutation confers defective DNA replication and demonstrates a reduced ability of the mutant D4 protein to interact with A20. Residue G179 is the C-terminal residue of strand 9 in the central parallel β-sheet. As shown in Fig. 5 the substitution of G179 residue by an Arg residue (modeled) would force a large basic residue into the hydrophobic pocket made up of residues Y156, I177, F195 and I198. Y156 is part of the preceding strand 8, I177 is part of strand 9, while F195 and I198 belong to helix 9 following strand 9. Accommodating the side chain of Arg179 will require considerable structural rearrangement. It is likely, although speculative, that the mutation leads to a rearrangement of secondary structure elements (β-strands and α-helices) at and close to the C-terminus to accommodate this residue, which in turn might interfere with binding to A20 and the formation of the A20:UDG processivity factor. Another possibility is a destabilizing effect of this mutation by modulating the dimer interface resulting in a potential interference with the dimer formation.

The other temperature-sensitive mutation, Dts27, creates a L110F substitution. Although this residue points into a hydrophobic pocket formed by residues F79, I89, I92 and Y108, the bulky aromatic side chain of the mutated residue is expected to cause steric hindrance and disrupt the local environment as shown in the model in Fig. 6.

vvUDG shows ~20% sequence identity with E. coli and human UDGs. Sequence homology with herpes simplex virus1 (HSV1) UDG is also in the same range (21% identity). The structural homology between these proteins is low (rms deviation of 2.0 Å for 149 Cα atoms and a match rate of 66% to vvUDG). On the other hand, HSV1 UDG is very similar to human UDG and E. coli UDG in terms of sequence (39% identical to human; 49% identical to E. coli), fold (rms deviation of 1.1 Å for 210 Cα atoms and a match rate of 94% to human; rms deviation of 1.2 Å for 199 Cα atoms and a match rate of 88% to E. coli) and characteristic motifs 2021.

The core of the vvUDG structure, however, is similar to other UDGs. The conserved four-stranded central parallel β-sheet, a small second β-sheet made from two anti-parallel β-strands, and six helices in the vvUDG structure match with the observed topology in E. coli and human UDG (see Table 5). Superimposition of the vvUDG structure with E. coli UDG (PDBId: 2EUG) using program TOPP 22 revealed an rms deviation of 2.1 Å for 136 Cα atoms in the matching secondary structure elements (match rate 64.5%; sequence identity 21.3%). The superimposition with human UDG (PDBId: 1AKZ) gave an rms deviation of 2.0 Å for 138 Cα atoms in the matching secondary structure elements (match rate 61.9%; sequence identity 20.3%). The vvUDG structure shows some new features that are unique among known UDG structures as discussed earlier and shown in Fig. 1. Compared to vvUDG the C-termini in E. coli 9 and human 23 UDG do not show any well defined secondary structure. Other structural differences of vvUDG to E. coli and human UDG include fewer α-helices surrounding the central β-sheet, a distinct bend of the N-terminal helix (residues 19–39), some tighter turns in loop regions and the movement of C-terminal residues in strands 8 and 9. These two strands are shifted towards the center of the parallel β-sheet with respect to the other two structures.

A structure-based alignment of UDG sequences that included also the vvUDG sequence 13 demonstrates the pitfalls of this approach when the sequence identity drops to about 20%. The new features especially at the termini are not recognized, and the corresponding residues in the viral sequence are instead lined up with the previously observed conserved secondary structure elements in UDGs of other species. It is intuitive that the described novel features in the vaccinia virus UDG structure play a role in the unique function of poxvirus UDG in replication.

A plausible model for the function of poxvirus UDG as part of the processivity factor is shown in Fig. 7. The model shows two dimers arranged around a central channel in a homotetrameric arrangement that is observed in both crystal structures (Figs. 7A and 7B). A similar molecular assembly has been noticed in previously studied sliding clamps 242526. The diameter of the central channel in vvUDG is approximately 27 Å (distance measured between corresponding residues on either side). This compares well with diameters in the heterotrimeric sliding clamp of PCNA (Fig. 7C) in Sulfolobus solfataricus (PDBId: 2IX2) and the homodimeric sliding clamp of the polymerase β-subunit in E. coli (PDBId: 1MMI) that are ~29 Å and 30–35 Å, respectively. However, the central channel must have sufficient flexibility in order to accommodate various binding components. In this analogy A20, which acts as a scaffold with binding regions for UDG, E9 and other factors such as D5 and H5 5, would seem to function as a clamp loader. Protein regions at the type I dimer interface only observed in the trigonal crystal form may be involved in the binding of A20. A part of this dimer interface includes N-terminal residues 1–16 and C-terminal UDG residues 208–218. These 27 residues are almost exclusively hydrophobic. Interestingly, the N-terminal 50 residues of A20 that constitute the minimal interacting binding site for UDG are also predominantly hydrophobic. A hydrophobic interaction in the proposed binding surface is consistent with the observed stability of the heterodimeric A20:UDG complex at high ionic strength (up to 750 mM NaCl) 2. Although previous experiments have suggested a 1:1 stoichiometry of binding between A20 and UDG 2, the true composition of a functional polymerase unit remains to be determined.

The D4 gene sequence (AAA48100; Western Reserve 109) encoding for uracil-DNA glycosylase (218 a.a.; Mr ~25 kDa) was subcloned into pET15b vector (Novagen), and transformed into E. coli BL21(DE3)pLysS Rosetta cells (Invitrogen). DNA sequencing showed a single substitution (D17N) when compared to the vaccinia virus sequence in the database. The resulting recombinant protein contains a 20-residue insert at the N-terminus comprising of a hexahistidine tag and a thrombin cleavage site. Bacterial cells were grown in Luria-Bertani (LB) broth in the presence of ampicillin to an OD₅₉₅ of 0.7 at 37°C. After induction by isopropyl-β-D-thiogalactopyranoside (IPTG) the protein was expressed for 16 hrs at 18°C. The cell culture was centrifuged at 6400 g for 15 min at 4°C, and cell pellets were stored at -80°C until further use.

Frozen cells suspended in a lysis buffer (50 mM Tris, pH 8.0; 100 mM NaCl; 1 mM benzamidine; 0.1 mM phenylmethylsulfonyl fluoride; 5 mM β-mercaptoethanol) were lysed by multiple cycles of freezing and thawing. The cell free extract was prepared by centrifugation at 39200 g for 30 min at 4°C. The recombinant protein was purified from the bacterial extract using affinity chromatography on a Ni-NTA column (Amersham Biosciences). After the protein was eluted from the column with 200 mM imidazole, the hexahistidine tag was removed by treatment with thrombin. The digestion mixture was concentrated and subjected to gel filtration on a Superdex 200 column equilibrated with elution buffer (50 mM Tris, pH 9.0; 100 mM NaCl; 3 mM dithiothreitol). The major portion of the protein eluted as a dimer (calculated from the elution volume of protein standards and from dynamic light scattering experiments). Fractions representing this major peak were concentrated to 8 mg/ml by ultrafiltration.

We also employed a rapid single step affinity purification protocol using a TALON™ (BD Biosciences) column for purification of His-tagged protein from the bacterial extract. Briefly, the cell free extract (pH 7.3) was applied to the TALON™ column containing immobilized cobalt ions. The column was extensively washed with buffer containing 300 mM NaCl and 5 mM imidazole. The tightly bound protein was eluted in a gradient at about 150 mM imidazole in Tris buffer (50 mM, pH 7.3; 100 mM NaCl; 10 mM β-mercaptoethanol). According to the SDS polyacrylamide gel the protein purity was better than 95%. The N-terminal His-tag was not cleaved, and protein in the peak fractions was concentrated to approximately 7 mg/ml without buffer exchange.

The thrombin cleaved protein crystallized in two different conditions (condition 1: 5% PEG6000, 7.5% MPD, 0.1 M Hepes, pH 7.25 at 4°C; condition 2: 5% PEG3000, 0.1 M NaCl, 0.1 M Hepes, pH 7.5 at 4°C). Crystals grew to about 0.2–0.3 mm in 1–3 days. These crystals belong to orthorhombic space group P2₁2₁2₁ with unit cell dimensions of a = 117.77 Å, b = 134.06 Å, c = 139.10 Å (see Table 1). In this crystal form, there are 8 subunits of UDG in the asymmetric unit (V_M ~2.7 Å ³/Da corresponding to 55% solvent). Results from dynamic light scattering (DLS) verified that the protein exists predominantly as a dimer (estimated MW ~57 kDa, Stokes radius ~3.4 nm).

The protein purified in the one-step procedure without buffer exchange (pH 7.3) was crystallized in 100 mM Hepes buffer, pH 7.25, 12% glycerol and 1.5 M ammonium sulfate as precipitant. The size and quality of crystals were significantly improved using a microseeding protocol (2 μl drops with a 1:1 ratio of protein to seed solution) using the hanging-drop vapor diffusion experiment in NeXtal plates (QIAGEN). The protein crystallized in trigonal space group P3₂21 with unit cell parameters of a = b = 85.20 Å, c = 139.72 Å, γ = 120° (see Table 1). The asymmetric unit contains two subunits related by non-crystallographic symmetry (V_M ~2.7 Å ³/Da corresponding to 54% solvent).

Heavy-atom derivatives were prepared by transferring crystals from seeding experiments into a stabilizing solution (100 mM Hepes buffer at pH 7.25, 12% glycerol and 1.7 M ammonium sulfate) containing in addition varying amounts (1–5 mM) of different heavy atom salts. In this fashion, we obtained successfully a uranyl derivative from an overnight soak in 3 mM uranyl nitrate, which contains U⁺⁴ ions in form of the bivalent radical UO₂²⁺ group. The dataset of this heavy-atom derivative was isomorphous to the dataset of the native protein with slightly different unit cell parameters (see Table 1). Data were collected at 100 K on cryoprotected (same as crystallization solution but with 25% glycerol) crystals in house (R-Axis IV image plate detector) and at BioCryst (R-Axis IV⁺⁺).

Glycerol was used as cryoprotectant for both crystal forms. For the trigonal form crystals were transferred directly from the crystallization condition (containing 12% glycerol) into the same solution with 25% glycerol, while for the orthorhombic form cryoprotection required a step-transfer protocol (5% to 25% glycerol). All diffraction data were indexed and processed using HKL2000 27 and DTREK 28 program packages. Diffraction data are summarized in Table 1.

Efforts to solve the structure by molecular replacement using known UDG structures of E. coli, human and herpes simplex virus failed. The structure of vvUDG in trigonal space group P3₂21 was determined using SHELX 29 by the method of single isomorphous replacement with anomalous scattering (SIRAS) with phase information from a single heavy atom derivative (see Table 1). Isomorphous (to the native dataset) and anomalous differences of the uranyl dataset were good to 2.8 Å resolution. With the help of the graphical interface "hkl2map" determination of the heavy atom substructure (U sites) and initial phasing was successful at 2.8 Å using SHELXD 30. A correlation coefficient of 46% between E_obs values (from ΔF) and E_calc values (from heavy atoms) indicated an excellent quality of this solution. The proper enantiomorph and the right space group (P3₂21) were clearly established. After density modification and phase extension to 2.5 Å resolution in SHELXE 31 the SIRAS phases for space group P3₂21 had an overall figure of merit of 0.60 and a connectivity index of 0.91.

The structure of the orthorhombic crystal form was solved by molecular replacement with the program MOLREP 32 using the refined model of the trigonal crystal form. The structure solution shows eight subunits in the asymmetric unit arranged as 4 homodimers.

The obtained reflection file with SIRAS phases was converted to CCP4 mtz format. The resultant map allowed the placement of 59% of the amino acid sequence into the electron density by automated model building using PHENIX 33, and clearly established the presence of the two expected subunits in the asymmetric unit. The remaining residues were fitted into electron density maps calculated with combined SIRAS and model phases. Manual model building was performed with QUANTA (Accelrys, Inc.) and COOT 34. Model building and stages of subsequent refinement included the use of CNS 35 simulated annealing omit maps. This procedure was especially necessary to follow the chain correctly in the regions of the dimer interfaces because of the close proximity of NCS and symmetry related molecules, and also because these regions contain the residues that are disordered in the final model. In addition, the use of omit maps and difference electron density maps was standard practice throughout model building and refinement. Ligands and water molecules were added using programs CNS 35, REFMAC 36 and COOT 34 and also manually into difference electron density maps (F_o-F_c maps, 3σ level). All water molecules that showed low occupancies or high B-factors after refinement and did not satisfy distance constraints for hydrogen-bonding to protein residues were subsequently removed. For the ligands the real space R fit and the quality of electron density was the deciding factor.

Refinement for the trigonal crystal form (subunits A, B) was carried out at 2.4 Å resolution using CNS 35 and REFMAC 36. Various NCS models (from tight NCS to no NCS) were used during refinement stages. The final NCS restraints produced the lowest R_free and the smallest difference between R and R_free values. In addition to restrained refinement by maximum likelihood with tight NCS restraints for main chain atoms (rms deviation of distances is 0.18 Å and of B-factors is 0.29 Å ²) and medium NCS restraints for side chain atoms (rms deviation of distances is 0.50 Å and of B-factors is 0.44 Å ²), we also used the TLS refinement option in REFMAC 36. Each protein subunit (subunits A and B) was divided into three TLS groups (residues 1–97, 98–162 and 163–218) based on analysis by the TLS motion determination (TLSMD) server 37. Release of NCS restraints and independent refinement of subunits led to an increase in R values and worsened the geometry.

Refinement for the orthorhombic crystal form was carried out at 2.3 Å resolution using CNS 35 and REFMAC 36. Restrained refinement by maximum likelihood with medium NCS restraints was combined with TLS refinement in REFMAC 36.

Table 2 shows the refinement statistics for both structures. Figures were prepared using PyMOL (DeLano Scientific LLC) and GRASP 38.

Coordinates and structure factors for the trigonal and orthorhombic crystal forms of vvUDG have been deposited in PDB (PDBIds: 2OWQ, 2OWR).