A Parsimonious Mechanism of Sugar Dehydration by Human GDP-Mannose-4,6-dehydratase

Biosynthesis of 6-deoxy sugars, including l-fucose, involves a mechanistically complex, enzymatic 4,6-dehydration of hexose nucleotide precursors as the first committed step. Here, we determined pre- and postcatalytic complex structures of the human GDP-mannose 4,6-dehydratase at atomic resolution. These structures together with results of molecular dynamics simulation and biochemical characterization of wildtype and mutant enzymes reveal elusive mechanistic details of water elimination from GDP-mannose C5″ and C6″, coupled to NADP-mediated hydride transfer from C4″ to C6″. We show that concerted acid–base catalysis from only two active-site groups, Tyr179 and Glu157, promotes a syn 1,4-elimination from an enol (not an enolate) intermediate. We also show that the overall multistep catalytic reaction involves the fewest position changes of enzyme and substrate groups and that it proceeds under conserved exploitation of the basic (minimal) catalytic machinery of short-chain dehydrogenase/reductases.


NahK expression and purification
The expression plasmid for a His 6 -tagged version of NahK 2 from Bifidobacterium longum JCM1217 was kindly provided by Dr. Motomitsu Kitaoka (National Food Research Institute, Ibaraki, Japan). The plasmid was transformed into E. coli BL21(DE3)pLysS. Recombinant NahK was produced in 1-L baffled shaken flasks (250 mL medium, inital OD 600 0.1) at 37°C and 110 rpm using Lennox media containing 0.05 mg/mL of kanamycin. At an OD 600 of 0.8, the temperature was decreased to 18°C and gene expression was induced with 0.4 mM IPTG for 20 h. Cells were harvested by centrifugation at 2,000 g at 4°C for 30 min using a Sorvall RC-5B refrigerated super speed centrifuge (Dupont Instruments). Cell disruption and immobilized Ni affinity chromatography (IMAC) were performed as described above for hGMD. Following affinity purification, an additional anion exchange (AEX) purification was performed to eliminate potentially contaminating phosphatases. Therefore, fractions from IMAC containing the NahK were pooled, and concentrated. Using Amicon Ultra-15 Centrifugal Filter Units (10 kDa cut-off; Millipore) buffer was exchanged to 50 mM HEPES, pH 7.5, and total volume concentrated to 10 mL. The sample was applied to a 5 mL HiTrap Q HP AEX column (GE Healthcare) mounted onto an ÄKTA prime plus (GE Healthcare) system and equilibrated in 50 mM HEPES buffer, pH 7.5. Bound protein was eluted using a linear gradient Centrifugal Filter Units (10 kDa cut-off; Millipore) were used. 70 µL portions (30 mg/mL) were stored at -70°C until further use.
The NahK assay was performed at 37°C in 50 mM HEPES buffer (pH 7.0) containing 2 mM MgCl 2 . The substrates used were ATP (10 mM) and Man (10 mM). One unit of NahK activity is the amount of enzyme producing 1 μmol of -mannose 1-phosphate/min under the conditions used. The enzyme preparation had a specific activity of 0.02 U/mg.

ManC expression and purification
A His 6 -tagged variant of E. coli ManC was used. Expression and purification were performed as described earlier. 3 The ManC assay was performed at 37°C in 50 mM HEPES buffer (pH 7.0) containing 2 mM MgCl 2 .
The substrates used were GTP (10 mM) and -mannose 1-phosphate (10 mM). One unit of ManC activity is the amount of enzyme producing 1 μmol of GDP-Man/min. The enzyme preparation used had a specific activity of 6 U/mg.

Crystallization and data collection
Purified wildtype hGMD and variants thereof (E157Q, S156D) were crystallized using the sitting drop vapour diffusion method. Co-crystallization was performed using 13 mg/mL enzyme, 1 mM NADP  Table S1.

Structure determination and refinement
The structure of hGMD was determined by molecular replacement with PHASER 8 using PDB ID 1T2A as a search model. Model building was performed with COOT 9 and refinement was carried out with phenix.refine 10 Table S1.

Molecular dynamics simulation
Molecular dynamics (MD) simulations were performed using YASARA dynamics. 14 The crystal structure of the E157Q variant (PDB ID: 6GPK) was used as the starting structure because the full protein, including the inhibitor loop, could be modelled in this case. To simulate the wildtype enzyme, Gln 157 was substituted in silico by Glu 157 and the resulting energy-minimized active-site architecture was evaluated using the crystallized wildtype complexes. No significant differences could be detected between the experimental and thus modelled active sites. Ligands present in the active site were modified; GDP-Man to GDP-mannos-4'',5''-ene and NADP + to NADPH, granting minimal perturbation of the crystallized complex. In each simulation, the hGMD dimer was used as this represents the native quaternary structure of the enzyme. Force field parameters of the ligands were assigned by the Yasara´s AutoSmiles approach 15,16 implemented in YASARA (version 18.2.7). Energy minimization was done prior to each MD simulation in explicit water using the AMGER15FB 17 force field and the TIP3P water model 18 . The system was immersed in a octahedral simulation cell, which extended 10 Å on each side. The AMBER15FB force field was assigned with 8 Å van der Waals interactions cut-off, and long range electrostatic interactions were calculated using the Particle Mesh Ewald (PME) algorithm. 19 All simulated systems were neutralized to pH 7.4 by counter ions using 0.9% NaCl. Solvent density was 0.997 g/mL (at 298 K). YASARA predicted pK a values for ionizable side chains (E, D, H and K) 20 . However, the protonation states of Glu 157 (protonated) and Tyr 179 (deprotonated) were chosen according to the simulated reaction step. The constant simulation temperature of 298 K was set by assigning random velocity vectors to all atoms using a Maxwell-Boltzmann distribution and a Berendsen thermostat, so that the resulting kinetic energy matched the specified temperature. The time average macroscopic temperature was kept at the requested value by rescaling the atom velocities. 21 Convergence of productive MD simulations to equilibrium were evaluated by the RMSD of total energy and C atoms (see Figs. S12 and S13). After 1 nsec of the simulation, the complex could be considered stable. The RMSD of Catoms was 1.3 (± 0.1) Å. The C atoms and key interatomic distances were evaluated every 100 psec S10 (Fig. S13) and average distances were calculated from 1 to 15 ns of simulation. Average distances include both simulated chains (Table S2).

Capillary zone electrophoresis
Capillary zone electrophoresis was performed at 50°C on an HP

Enzyme-bound NADPH at reaction steady state
Prior to each measurement, the enzyme-bound NADPH was oxidized to NADP + using GDP-4''keto-6''deoxy-mannose as the substrate, adopting a procedure previously developed for dTDP-glucose dehydratase. 22  immediately after the reaction start and after 3600 sec. Samples were quenched in methanol and analyzed by capillary electrophoresis (Fig. S18).

Preparative synthesis of GDP-Man
Synthesis and product isolation was performed as described by Pfeiffer et al. 3 Purity of the lyophilized GDP-Man was verified using HPLC, CE and 1 H NMR (Fig. S20).

NMR analysis of purified sugar nucleotides
Lyophilised sugar nucleotides were dissolved to 3 -10 mM in 50 mM potassium phosphate buffer (pD 7.9) in a total volume of 600 µL.  Fig. S20 -S24). de Compostela, Spain) was used for data processing and analysis.

In situ 1 H NMR analysis of hydrogen/deuterium exchange at the C''5 of GDP-4''-keto-6''deoxy-mannose
Enzymatic conversion of GDP-4''-keto-6''-deoxy-mannose (produced in H 2 O) to GDP-4''-keto-5''-deutero-6''-deoxy-mannose were analyzed by 1 H-NMR (Fig. S27 -S29                 . Note: all enzymes were pre-incubated with GDP-4''-keto-6''-deoxy-mannose to oxidize the NADPH present in the enzyme as isolated. This oxidation was possible for wild-type enzyme and the S156A and S156D variants (panels a, b and c). Due to lacking activity in the enzymes, this oxidation was not possible for the Y179F and the E157Q variants. These latter variants therefore retained enzymebound NADPH before being used in the assay with GDP-Man (panels d and e). For details of the procedures used, please see on page S11 the section Enzyme-bound NADPH at reaction steady state. a) Wildtype and b) S156A retain their (very low) amount of initial NADPH present in the enzymes while full conversion of GDP-Man occurs within 3600 s. c) S156D shows retention of the (very low) initial NADPH without conversion of GDP-Man within 3600 s. d) Y179F and e) E157Q were not oxidized and exhibit an initial NADPH occupancy of 70% an 49%, respectively. Both enzymes show a constant NADPH content, without conversion of GDP-Man during 3600 s of incubation. CE traces show peaks (blue trace measured at 253 nm) corresponding to (1) GDP-Man and (2) GDP-4''-keto-6''-deoxy-mannose; as injection volumes are not tightly controlled, areas under the peak of GDP containing compounds are used to calculate conversions. The red trace corresponds to the current during the measurement.      . Proposed mechanism of -elimination via an initial enolization step in hGMD and its possible application to the reaction of dTDP-glucose-4,6-dehydratase (RmlB). Both hGMD and RmlB utilize a conserved Glu as general catalytic base in the enolization step. In hGMD, the Glu is the general acid in the subsequent 1,4-elimination, describable according to this study as a vinylogous E2 reaction (see Gerlt and Gassman 23 ). Once the Glu has abstracted the proton from C5'', it changes its conformation slightly to be able act as general acid for the elimination of water. Note: our MD simulations have captured this conformational flexibility of the Glu for hGMD. RmlB differs from hGMD by the presence of an acidic residue (Asp) in the active site as shown. The Asp has been considered as the catalytic acid for the elimination of water (see ref 24 in this Supporting Information and references therein). However, the possibility that the Glu moves into the immediate vicinity of the Asp when it adopts the conformation required for general acid catalysis to water elimination was not considered. Assuming this to occur, the presence of Asp in RmlB is expected to increase the pK a of the Glu as compared to its value in hGMD. From previous analyses of similar residue arrangements in the active sites other enzymes (e.g., hexosaminidases; see main text) the effect on pK a enhancement is expected to be substantial (several log units). The mechanistically relevant consequence of the pK a increase in Glu according to the scenario proposed is that enolization of the ketone formed in the initial step of enzyme catalysis, specifically the proton transfer from C5'' to Glu, becomes thermodynamically more favorable. This is indicated in the figure as a change in the internal equilibrium for enolization of the enzyme-bound intermediates in RmlB as compared to hGMD. Note that the true equilibrium of the enolization is not known and we intend to show here only the relative effect between the two enzymes. In the case that the overall reaction rate constant (k cat ) is limited by oxidation-enolization, as in hGMD, modulating the pK a of the Glu appears to be a chemically sound strategy, perhaps used in natural evolution of dehydratases with just that selection principle, to enhance the enzymatic rate. Since an elevated pK a of Glu would make proton transfer to the water leaving group less favorable in the subsequent elimination step, as indicated in the figure, ability of the Glu to engage in both general base and general acid catalysis must be a balanced one. In RmlB, however, the Asp was proposed to function as the catalytic acid for elimination. 24 We note that stabilization of the protonated Glu in a conformation bonded to Asp, as shown in the figure, could have the consequence that enolization and elimination become strongly coupled (concerted) steps of the overall enzymatic reaction. This would allow -elimination via enolization to be reconciled with the experimental observations of concerted reaction of wildtype dTDP-glucose-4,6dehydratase. 24