Structural Insights into the Quaternary Catalytic Mechanism of Hexameric Human Quinolinate Phosphoribosyltransferase, a Key Enzyme in de novo NAD Biosynthesis

Quinolinate phosphoribosyltransferase (QPRT) catalyses the production of nicotinic acid mononucleotide, a precursor of de novo biosynthesis of the ubiquitous coenzyme nicotinamide adenine dinucleotide. QPRT is also essential for maintaining the homeostasis of quinolinic acid in the brain, a possible neurotoxin causing various neurodegenerative diseases. Although QPRT has been extensively analysed, the molecular basis of the reaction catalysed by human QPRT remains unclear. Here, we present the crystal structures of hexameric human QPRT in the apo form and its complexes with reactant or product. We found that the interaction between dimeric subunits was dramatically altered during the reaction process by conformational changes of two flexible loops in the active site at the dimer-dimer interface. In addition, the N-terminal short helix α1 was identified as a critical hexamer stabilizer. The structural features, size distribution, heat aggregation and ITC studies of the full-length enzyme and the enzyme lacking helix α1 strongly suggest that human QPRT acts as a hexamer for cooperative reactant binding via three dimeric subunits and maintaining stability. Based on our comparison of human QPRT structures in the apo and complex forms, we propose a drug design strategy targeting malignant glioma.


Results
Overall structure of human QPRT. To investigate the molecular basis of mammalian QPRT in the hexamer state upon substrate binding, we determined the structures of HsQPRT in the apo form and in complex with its reactant QA as well as the reaction product NAMN at 2.8, 3.1 and 2.6 Å, respectively. The data collection and refinement statistics are summarised in Table 1. The HsQPRT monomer consists of 12 α -helices and 12 β -strands arranged into two structural domains: an N-terminal open-faced α /β -sandwich domain (residues 1-112 and 279-288; six-stranded antiparallel β -sheets and five α -helices) and a C-terminal open TIM α /β -barrel domain (residues 113-278; six β -strands and seven α -helices) (Fig. 1a). The domain architecture and secondary structure elements were nearly the same in both the apo form and the substrate complexes (Fig. 1b). The dimer of HsQPRT is formed by a head-to-tail interaction of the N-terminal domain of one monomer and the C-terminal domain of the other monomer (Fig. 1c). The active sites of the twisted bow tie-shaped HsQPRT dimer are located at the interface between the α /β -barrel of one subunit and the β -sandwich of the other subunit. The active site consists of the C-terminal α /β -barrel with one N-terminal β -sheet for dimerisation, which is consistent with the dimeric interfaces found in other type II PRTs 27,34 . Each monomer extensively connects to three other monomers, indicating that HsQPRT forms intra-and inter-contacting arrangements in the hexameric assembly.
Hexameric HsQPRT is formed by the trimerisation of dimers in a three-fold axis, which is similar to other eukaryotic enzymes such as SsQPRT and ScQPRT ( Fig. 1d and Supplementary Fig. S2). Size exclusion chromatographic analysis revealed that HsQPRT forms a hexamer with an estimated molecular weight of approximately 190 kDa. Based on biochemical approaches, this hexamerisation is consistent with QPRTs from other mammals such as rat and porcine [28][29][30][31] .

Structural comparison of human and other eukaryotic QPRTs. Various structures of eukaryotic
QPRTs have been reported, such as those from human (apo and complexes with tartrate or phthalate, an analogue of QA with a benzene ring rather than a pyridine ring) 26,29 , porcine (NAMN complex) 28 and yeast (apo, complexes with QA, PRPP or phthalate) 27 . The apo HsQPRT structure was compared with previously reported HsQPRT and ScQPRT apo structures (PDB codes 4KWV and 3C2E) 27,29 . However, the apo structure of HsQPRT described by Malik et al. is relatively similar to that of ScQPRT. Moreover, our apo HsQPRT structure showed a different conformation from other structures, particularly in loop L (H160-L167) and loop O (A191-V198) (Fig. 1e,f). Two loops were located in not only the active site but also the dimer-dimer interface. The orientations of these loops in our apo structure resembled an open structure, whereas that in the other apo structure appeared as a more closed dimer-dimer interface, referred to as HsQPRT-open and HsQPRT-closed, respectively. Conformational changes in loops L and O suggest that apo HsQPRT exists as a dynamically equilibrated structure around six active sites located in three dimer-dimer interfaces, potentially affecting substrate binding and enzyme turnover rate. In addition to analysing complex structures, we also compared the structures of QPRT in complex with QA (reactant of HsQPRT), tartrate (PDB code 2JBM) and phthalate (PDB code 4KWW) to investigate the differences in the binding mode between the substrate and its mimic. Both loops L and O in the two structures of HsQPRT in complex with QA analogues were more similar to HsQPRT-closed than to HsQPRT-open. Tartrate has a hydroxyl ethyl group rather than the pyridine ring of QA, which induces conformational changes in H160 and R161 (Fig. 1g). Although R161 showed an ionic interaction with a carboxylate in both QA and tartrate, R161 of the tartrate complex protruded into the pyridine moiety of QA and created an additional interaction with E246 and S268. Alternatively, R161 of HsQPRT-QA rotated by approximately 50° to avoid the steric hindrance against the pyridine moiety and resulted in a conformational change of the imidazole side chain of H160 in addition to its interaction with the carboxylate moiety of QA. Moreover, the ε -amino group of K171 was rotated by approximately 45° because of the difference in the position of the carboxylate moiety of QA and tartrate molecules of 2.4 Å. Phthalate showed a highly similar mode of binding to that of QA, making HsQPRT the stable intermediate state that could not combine with PRPP to produce NAMN.
In addition to in humans, the NAMN-bound QPRT structure from porcine (SsQPRT-NAMN, PDB code 4I9A) showed structural and sequence conservation, particularly near the active site. Although the NAMN binding site is similar between human and porcine, residues near the dimer-dimer interface showed different conformations ( Supplementary Fig. S3). In MolA, the residue corresponding to R190 of SsQPRT-NAMN is glutamine in human and was rotated by approximately 180°. In addition, the human protein contains D193 rather than N193 from porcine, enabling higher ionic contacts with H133′ in MolF and indicating that HsQPRT forms a more stable hexameric configuration compared to the porcine enzyme.

Dataset
HsQPRT-open HsQPRT-QA HsQPRT-NAMN Changes in the dimer-dimer interaction upon substrate binding. Although the overall folds of hex- americ HsQPRT-open and substrate complexes were highly similar, specific regions of the structures differed, particularly the dimer-dimer interfaces (Fig. 2a and that R161 is an allosteric residue that not only functions as a key player in the active site in the presence of a ligand but also as a mediator between the intermolecular interactions to release the product and form a hexamer in the HsQPRT-open structure. This is consistent with the kinetic analysis of the effects of various mutations around the active sites, which showed that the R161A mutation could abolish substrate binding and enzyme activity with a 20% capacity of wild-type 26 . Consequently, substrate binding of one protomer in one dimer caused conformational changes in the adjacent protomer in the other dimer via loops L and O. In addition, changes in the interaction between the two dimers upon substrate binding also supports that HsQPRT exists as a hexamer that induces conformational change in the adjacent two dimers while the substrate binds to one dimer to enable reactant binding. This was further explored and is described in the next section. The mode of binding of the product NAMN molecule to human QPRT also showed a distinctive structural difference near the dimer-dimer interface compared to that of the reactant QA (Fig. 3c). Two hydroxyl groups of D193 in MolA of HsQPRT-QA showed an ionic interaction with H133′ of MolF. In contrast, D193 in MolA of HsQPRT-NAMN moved by approximately 4.9 Å to form ionic contacts with R189 in MolA, which enabled only one hydroxyl group to interact with H133′ of MolF. These changes in the dimer interactions caused the product complex to have a more relaxed hexamer. The contact areas between the two dimers in HsQPRT-QA and HsQPRT-NAMN were 2127 and 2010 Å 2 , respectively. These results suggest that QA-bound HsQPRT maintains a stable hexamer to mediate the subsequent reaction, such as PRPP binding and phosphoribosyl transfer into QA to generate NAMN. After the reaction, a weaker dimer-dimer interaction between HsQPRT and NAMN compared to QA appeared to be favourable for release of the reaction product. Changes in the interaction between two dimeric subunits were found not only in reactant binding but also in the conversion from the reactant to the product.

Significance of the N-terminal alpha helix in hexamer formation in human QPRT. The
N-terminal region of eukaryotic QPRTs from yeast to human mainly consists of highly conserved hydrophobic residues such as leucine and proline (Fig. 4a). Although some of the prokaryotic QPRTs such as HpQPRT and TtQPRT appear to exist as hexamers, most exhibit a dimeric configuration. This finding supports the hypothesis that eukaryotic QPRTs maintain the stability of hexamer assembly by generating a hydrophobic interaction with the adjacent dimer within the hexamer structure through their N-terminus.
Interestingly, structures of mammalian QPRTs have unique features compared with that of the yeast enzyme, such as an α -helix of approximately two turns at the N-terminus (helix α 1), which is located in the dimer-dimer interface such as loops L and O (Fig. 4b). Consistent with this hypothesis, helix α 1 of MolA (A3-P11) in the HsQPRT structure interacts with helix α 3′ (Y32′ -S37′ ) in the adjacent dimer (Fig. 4c). The conformation of helix α 1 is very similar among apo and complex structures with substrates, the root mean square deviations of which are less than 1.0 Å, as shown in Fig. 3b and Supplementary Figure S5. Irrespective of substrate binding, highly conserved N-terminal hydrophobic residues L6-P11 in MolA form extensive hydrophobic contacts with L30′ -A39′ in another subunit of the adjacent dimer, MolE. Because the interface between the dimers is mostly hydrophobic, interactions mediated by hydrophobic residues in helix α 1 are highly conserved around mammalian QPRTs and contribute to the formation of the dimer-dimer interface. Hydrophobic interactions in helix α 1 in HsQPRT occupy approximately 30% of the total contact area between the two dimers. Thus, two intermolecular interactions, such as MolA-MolE (helix α 1-helix α 3′ ) and MolA-MolF (loop L-loop O′ ), possibly play an essential role in hexamerisation of the HsQPRT structure with sustained dimeric interfaces. To examine the role of hydrophobic residues in the N-terminus of HsQPRT involved in the oligomeric state, we generated serial N-terminal deletion mutants (NΔ 4, NΔ 8, NΔ 9, NΔ 10 and NΔ 12) (Fig. 4d). NΔ 4 eluted as a hexamer that was similar to wild-type, while NΔ 8, NΔ 9, NΔ 10 and NΔ 12 were eluted as dimers, indicating that HsQPRT hexamerisation may be dependent on three hydrophobic residues, L8, L9 and P11. These residues are highly conserved throughout mammalian enzymes, presuming that truncated HsQPRT at the N-terminus resembles the dimer found in yeast enzyme. This finding is consistent with the result of the same experiment using full-length ScQPRT and its N-terminal truncated mutant (NΔ 10 Sc ) ( Supplementary Fig. S6). The full-length enzyme exists as a hexamer, but NΔ 10 Sc was found to be in the dimeric state, which is consistent with the size distribution from hexameric to dimeric HsQPRT. This indicates that the N-terminal hydrophobic residues of eukaryotic QPRTs are important for hexamer formation.
To determine how helix α 1 is involved in hexamer stabilisation in HsQPRT, we performed a heat aggregation test using the spectrophotometric method. Interestingly, the half-aggregation temperature of full-length HsQPRT (83.6 °C) was approximately halved for the mutant missing helix α 1, which included only nine residues (NΔ 9, 48.3 °C) (Fig. 4e). Combined with structural information, these results indicate that elimination of the hydrophobic interaction by helix α 1 sufficiently makes HsQPRT sufficiently unstable to maintain the hexameric configuration. However, how conversion of hexameric HsQPRT to a dimeric protein by eliminating helix α 1 affects the enzyme function remains unclear, although the NΔ 9 mutant has residues forming the intact active site. Additionally, Liu et al. reported that the reactant QA binds to HsQPRT prior to PRPP binding 26 . Thus, we evaluated the binding affinity of both the full-length protein (hexamer) and the NΔ 9 mutant (dimer) to the first-order reactant QA using isothermal titration calorimetry (ITC) (Fig. 5). The K d value between QA and the NΔ 9 mutant (401 ± 56 μ M) was decreased by approximately 80-fold than that of the full-length protein (4.8 ± 0.4 μ M) and was similar to the wild-type ScQPRT (450 ± 30 μ M) 27 . The QA binding into full-length HsQPRT is endothermic and entropically driven, suggesting that the dynamicity between the open and closed conformations of the hexameric HsQPRT is important for binding. Such dynamicity is due to the flexible loops formed the dimer-dimer interface conserved only in mammalian QPRT (Supplementary Fig. S4). The dimeric NΔ 9 mutant may not have structural flexibility because it lacks the dimer-dimer interaction, which may affect formation of the closed conformation and make the reaction exothermic. Although ScQPRT forms a hexamer, it showed only a closed conformation even though the apo form, which may hinder QA accessibility. This explains why the NΔ 9 mutant showed similar binding affinity to ScQPRT. In contrast, full-length HsQPRT stabilised the hexamer, which could form an open conformation (Supplementary Fig. S7). Thus, hexameric HsQPRT enhances QA accessibility, improving binding affinity. The results of the ITC experiment indicated that HsQPRT exists as hexamer to enhance the efficiency of reactant binding by stabilizing the three dynamic dimer-dimer interfaces of hexameric HsQPRT to accommodate the reactant. Moreover, we measured the binding affinity and calculated the Hill coefficient to analyse the cooperativity of binding to the hexameric HsQPRT for QA, phthalic acid, PRPP, phthalic acid with saturated PRPP and PRPP with saturated phthalic acid. We used phthalic acid rather than QA for bi-substrate binding, as the reaction interfered with the affinity measurement in the presence of the two substrates. We calculated the Hill coefficient by analysing the Hill plot. The K d value of QA was 4.8 ± 0.4 μ M with a Hill coefficient of 1.28 ± 0.09, indicating that QA binds to hexameric apo HsQPRT with positive cooperativity (Fig. 5a and Supplementary Fig. S8). The binding property of phthalic acid to HsQPRT was very similar to that of QA with a K d value and Hill coefficient of 5.2 ± 0.2 μ M and 1.25 ± 0.02, respectively ( Supplementary Fig. S9). Thus, phthalic acid is a good analogue of QA for measuring bi-substrate binding and the cooperativity of substrate binding. We could not measure the binding properties of PRPP likely because of the low binding affinity of PRPP ( Supplementary Fig. S10a). The K d value of PRPP for the yeast enzyme was reported to be approximately 10 mM, which exceeds the limit of ITC measurement 27 . In order to investigate the binding affinity and cooperativity for the binding of the second substrate, we titrated PRPP under saturated conditions of phthalic acid and vice versa ( Supplementary Fig. S10,S11). Under phthalic acid saturated conditions, PRPP binding showed positive cooperativity, with a K d value and Hill coefficient of 124 ± 5 μ M and 1.14 ± 0.01, respectively. These results indicate that PRPP binding affinity increases in the presence of the first-order substrate QA. Cooperativity of the second substrate (PRPP) binding was lower than that of the first substrate (QA). In the reverse experiment, under PRPP saturated conditions, phthalic acid binding affinity dramatically decreased by approximately 6-fold (K d = 33.3 ± 3.7 μ M) and the positive cooperativity decreased to 1.11 ± 0.01. This result indicates that PRPP inhibits first-order substrate (QA) binding. Based on structural and biochemical evidence, mammalian QPRTs including the human enzyme may exist as a hexamer by the trimerisation of dimers assisted by helix α 1, consisting of only nine residues in its N-terminus. Moreover, the hexameric configuration is essential for not only stabilizing the hexameric enzyme through intermolecular hydrophobic interactions between the two dimers independently of substrate binding, but also contributes to cooperative reactant binding.

Discussion
In this study, we examined the molecular basis of the unexpected rearrangement of hexamer assembly caused by conformational changes in three dimer-dimer interfaces during the reaction process, as well as hexamer stabilisation of HsQPRT by the N-terminal short helix α 1. Thus, we propose a reaction model for HsQPRT (Fig. 6). Although the apo structure of HsQPRT showed dynamic movement between the open and closed conformations in its active site and its dimer-dimer interface, it can exist as a hexamer through extensive hydrophobic contacts with helix α 1. Some pentosyltransferases (EC 2.4.2) adopt a hexameric configuration through the trimerisation of dimers, such as nucleoside phosphorylases (PDB codes 4E1V, 4D8Y and 4R2X) [35][36][37] ; however, it is unusual for the short α helix of HsQPRT to contribute to hexamer stability while its flexible loops show dynamic conformational changes. First, loops L and O of HsQPRT were closed upon QA binding. Next, PRPP bound to the carboxylate moiety of QA to produce NAMN. Because PRPP is less stable than QA because of its highly reactive pyrophosphate group and therefore cannot remain in the active site for a long period of time, it is rational that QA can occupy the active site before PRPP enters. This is also consistent with the results of a previously reported kinetic analysis performed by Liu et al. 26 . As the reaction progresses, HsQPRT changes the conformation of loop O in the dimer-dimer interface, making the hexamer slightly less stable. Once an NAMN molecule is generated after the phosphoribosyl transfer reaction, HsQPRT fully unfastens its active site and regains the apo form, which can adopt open and closed states. The dynamicity of these dual conformations may provide the driving force to release NAMN.
NAD + plays a key role in cell survival and growth, particularly by maintaining redox homeostasis and cellular energy generation. Given the necessity of excessive NAD + levels in tumour cells to maintain their rapid growth, inhibiting NAD + biosynthesis is an attractive target for the development of anticancer agents. In this context, extensive efforts have been exerted to design compounds that induce apoptosis of tumour cells with NAD + depletion by inhibiting enzymes associated with NAD + synthesis. However, the design of anticancer agents that inhibit NAD + biosynthesis has only focused on the salvage pathway, in which nicotinamide is a precursor. For instance, FK866 (APO866), a low-molecular-weight compound, has been reported to exhibit potent efficacy in a series of cancers such as breast, gastric and liver cancer, inhibiting nicotinamide phosphoribosyltransferase, a rate-limiting enzyme in the salvage pathway of NAD + synthesis, and is currently being examined phase II clinical trials. Importantly, a recent study suggested that QPRT is highly involved in de novo NAD + synthesis in glioma in which approximately 30% of tumours are derived from the brain and central nervous system and 80% from all malignant brain tumours 20 . Interestingly, QPRT is exclusively expressed in the World Health Organization high-grade (WHO grade III-IV) glioma and produces NAD + using QA taken from normal microglial cells. This enables glioma to survive under conditions of oxidative stress and NAD + depletion caused by therapeutic approaches, including irradiation and alkylating agent treatment 21 . In this study, we determined that HsQPRT structures are present in apo and substrate-bound forms. In particular, the novel open conformation first observed in our apo structure provides information that can be used in the development of anticancer agents for treating high-grade glioma. Thus, agents targeting QPRT in glioma should be developed using information from the structures. Despite the substitution of QA with phthalate, which has an inhibitory effect in response to HsQPRT, the low selectivity related to its simple structure makes NAMN more useful as a starting point for developing a novel anti-glioma agent.
To design an NAMN-based QPRT inhibitor with higher specificity compared to substrates with extra moieties, NAMN was placed in the HsQPRT-open structure, demonstrating novel conformational changes in the NAMN binding site compared with the HsQPRT-NAMN structure (Fig. 7). No hindrance was observed when NAMN was placed in the HsQPRT-open structure. The orientations of the residues interacting with the ribose and phosphate groups were nearly the same in both HsQPRT-open and HsQPRT-NAMN structures such that the structures did not have to be further optimised. However, there was a difference in the interaction with the nicotinate moiety between the two QPRT structures of both structures. In HsQPRT-NAMN, nicotinate interacted with H160, R161, M169 and K171 to form a cavity that blocked another interaction with nicotinate, specifically through the close ionic contact of the R161 in loop L and K171 with 3-caboxylate in nicotinate. In contrast, the ligand binding cavity remained open in the HsQPRT-open structure because of the conformational changes in both R161 and K171. This cavity consisted of two positively charged patches wrapped around 3-carboxylate and one negatively charged patch. Modification of the carboxylate group in the NAMN molecule with polar moieties connected in the aliphatic chain generated an additional hydrogen bond with the cavities in the HsQPRT-open structure. These findings suggest that the scaffolds are promising for developing a novel anti-glioma agent that selectively targets the first structure of HsQPRT-open with unique conformational properties of loops in the substrate binding site around the dimer-dimer interface.
In summary, we reported the structures of HsQPRT associated with de novo NAD biosynthesis in the apo form with a novel open conformation and in complex with QA or NAMN as a reactant or product. We determined the physiological relevance of HsQPRT-QA compared with tartrate as a QA mimic. Apo and complex structures bound to substrate revealed that loops L and O, located in the dimer-dimer interface of hexameric HsQPRT, altered their conformations upon substrate binding and played a key role in reactant binding by changing the interaction between the two dimers. Moreover, to investigate the function of N-terminal hydrophobic residues in eukaryotic QPRTs, we performed structural comparisons and several biochemical studies, including size exclusion chromatography, a heat aggregation test and ITC measurements. We propose that conserved hydrophobic residues in helix α 1 stabilise the hexameric configuration of eukaryotic QPRTs by reinforcing the dimer-dimer interaction. Based on structural and biochemical studies, we determined that HsQPRT exists and functions as a hexamer to bind the reactant cooperatively and to maintain its stability. Based on the extra cavity of HsQPRT-open compared with HsQPRT-NAMN, we also suggest a strategy for drug development targeting HsQPRT to treat malignant glioma with high specificity. The results of this study provide biological information related to the functional insights of HsQPRT and anticancer agent design.

Methods
Protein expression and purification. DNA encoding full-length HsQPRT (residues 1-297) was amplified from human cDNA (GenBank code 30583300) by polymerase chain reaction using a PCR system 2720 thermocycler (Applied Biosystems, Foster City, CA, USA) and subcloned into the NdeI and XhoI sites of the pET21a (Novagen, Madison, WI, USA) expression vector. To scale up the over-expressed cells, HsQPRT protein was expressed in Escherichia coli BL21 (CodonPlus) strain at 37 °C with 1 mM isopropyl-D-thiogalactoside (Pharmacia, Uppsala, Sweden) for 6 h after induction. After disrupting the cells by sonication, the cell lysate was harvested by low-speed centrifugation at 4 °C and loaded onto a gravity-flow column (Bio-Rad, Hercules, CA, USA), which was packed with nickel-nitriloacetic acid agarose resin (Peptron, Daejeon, Korea) and pre-equilibrated with lysis buffer (50 mM NaH 2 PO 4 , pH 7.5 and 300 mM NaCl). After washing with lysis buffer, the HsQPRT protein was eluted using lysis buffer additionally containing 250 mM imidazole. The protein was concentrated using a Centriprep YM-3 (Millipore, Billerica, MA, USA) and then subjected to Hiload 16/60 Superdex 200 prep grade (GE Healthcare, Little Chalfont, UK) size exclusion chromatography equilibrated with 20 mM HEPES-NaOH, pH 7.5 and 100 mM KCl. The fractions containing the recombinant protein were confirmed using SDS-PAGE and subsequently pooled and concentrated to 15 mg/mL for initial crystallisation. The N-terminal truncated HsQPRTs (NΔ 4, NΔ 8, NΔ 9, NΔ 10 and NΔ 12) were prepared using the same procedures as used for the full-length protein. All proteins were obtained in > 98% purity as confirmed by SDS-PAGE. Concentrations for HsQPRT were determined using the following theoretical extinction coefficients (26470 M −1 cm −1 ) with PROTPARAM 38 and were calculated with an additional leucine, glutamate and hexahistidine tag in the C-terminus.
Crystallisation and data collection. The crystallisation of apo HsQPRT has been previously described 39 .
Briefly, plate-like crystals of apo HsQPRT were grown at 21 °C in 2 μ L hanging drops consisting of equal volumes of protein solution (15 mg/mL) and mother liquor consisting of 100 mM MES-NaOH, pH 5.0, 7-15% (w/v) PEG MME 2K and 10 mM potassium thiocyanate. The crystals of HsQPRT-QA were grown under mother liquor consisting of 100 mM TRIS-HCl, pH 8.0, 6-10% (w/v) PEG 8K, 100 mM magnesium acetate and 6 mM PRPP (1:10 molar ratio). Next, the crystal was soaked with 6 mM QA for 15 h. Although we performed co-crystallisation of QA only, this crystal diffracted poorly. In addition, we also collected datasets of the co-crystal of HsQPRT-PRPP and obtained the electron density, but we could not find electron density corresponding to PRPP. PRPP may contribute in maintaining the crystal packing, although PRPP was not detected in the electron density map. During the process of soaking in QA, it is likely that PRPP was exhausted by hydrolysis and the reaction observed was QA binding at the active site. HsQPRT-NAMN was co-crystallised under a reservoir solution under the same conditions for crystallisation of the apo enzyme except 100 mM sodium acetate, pH 5.0, rather than MES-NaOH was used with protein-NAMN (1:10 molar ratio) pre-incubated at 4 °C for 15 h. The crystals grew to a maximum size of 0.3 × 0.3 × 0.1 mm over one week. The crystals were cryoprotected in the reservoir solution supplemented with 25% (v/v) glycerol and flash-frozen under N 2 gas at 95 K. Native datasets of apo crystals were collected using an ADSC Quantum 210 CCD detector at beamline NW12A in PF-AR of the Photon Factory, Japan at a wavelength of 1.0000 Å. Datasets of HsQPRT-QA and HsQPRT-NAMN were collected using an ADSC Quantum 315 CCD detector at beamline 5C in the Pohang Accelerator Laboratory, Korea at a wavelength of 0.9795 Å.
Structure determination. The apo crystal belonged to the space group P2 1 with unit cell dimensions of a = 76.2, b = 137.1, c = 92.7 Å and β = 103.8°. All datasets were processed and scaled using HKL2000 40 . The asymmetric unit contained one hexamer (MolA-MolF) with a Matthews coefficient of 2.46 Å Da -1 and solvent content of 49.8%, as calculated using the Matthews coefficient 41 in the CCP4i 6.5.0. suite 42 . The apo HsQPRT structure was determined at 2.8 Å resolution using the molecular replacement method with the Phaser 43 program using the structure of the HsQPRT-tartrate complex (PDB code 2JBM) as a search model. Next, refinement cycles were performed using the REFMAC5 44 program in the CCP4i 6.5.0. suite 42 . No electron density was observed in some regions of each molecule (residues 159-165 and 192-197 in MolD and 159-166 in MolE). Phases of HsQPRT-QA or HsQPRT-NAMN were obtained by molecular replacement with Phaser 43 using the HsQPRT-tartrate complex as a template. Coordinates of nine QA and twelve NAMN molecules were obtained from RCSB Ligand Expo (http://ligand-expo.rcsb.org) and placed into the density corresponds to them. Multiple rounds of refinement were performed using REFMAC5 44 to final 2.8 (apo), 3.1 (HsQPRT-QA) and 2.6 (HsQPRT-NAMN) Å resolutions, respectively. The models were rebuilt manually using COOT 45 Fig. S12,S13). The Ramachandran plots, which were calculated using the program PROCHECK 46 , showed nearly no residues with torsion angles in forbidden areas. Detailed refinement statistics are provided in Table 1. All molecular graphics were generated using PyMOL (The PyMOL Molecular Graphics System, version 1.3, Schrödinger, LLC, New York, NY, USA).