Crystal Structure and Biophysical Analysis of Furfural-Detoxifying Aldehyde Reductase from Clostridium beijerinckii

It has previously been suggested that Cbei_3974 may be useful to alleviate the toxicity of furfural during the fermentation of acid-treated lignocellulose lysates (7). NADPH-dependent activity toward furfural was previously reported for this enzyme, but no characterization of the reaction product was shown (7). To confirm that the enzyme indeed generates the less-toxic alcohol from the aldehyde, the reaction product from an enzyme-substrate-NADPH incubation was analyzed by gas chromatography-mass spectrometry (GC-MS) in parallel with controls containing no NADPH or no enzyme. After 5 h of incubation, a new compound could be detected on the GC trace (Fig. 1). The retention time and fragmentation pattern were identical to those of a commercial standard of furfuryl alcohol. This compound was not detected in either of the controls, showing that its formation was enzyme catalyzed and NADPH dependent.

Cbei_3974 has previously been shown to also exhibit activity with hydroxymethyl furfural, benzaldehyde, and butyraldehyde (7). To more fully explore the substrate scope of the enzyme, a selection of aldehydes, ketones, and alcohols was chosen. These putative substrates were incubated at 2 mM with the enzyme and an excess of NADPH, NADH, or NADP⁺. The change in cofactor concentration was measured continuously by UV spectroscopy to give the reaction rate (Table 1). Surprisingly, l-glyceraldehyde-3-phosphate gave only negligible activity despite the enzyme sharing 57.7% identity with Escherichia coli YghZ, which converts l-glyceraldehyde-3-phosphate to l-glycerol-3-phosphate as part of a novel triose phosphate isomerase (TIM) bypass that allows the formation of dihydroxyacetone phosphate under gluconeogenic conditions, when TIM is genetically inactivated (9). Purified YghZ was shown to reduce l-glyceraldehyde-3-phosphate to l-glycerol-3-phosphate, which can be converted to dihydroxyacetone by l-glycerol-3-phosphate dehydrogenase, thus complementing TIM deficiency. Rather unexpectedly, YghZ is stereospecific for the l-enantiomer of the substrate, whereas the TIM substrate is d-glyceraldehyde-3-phosphate. It was therefore proposed that a spontaneous reaction may interconvert the two enantiomers (9). On this basis, YghZ and enzymes with similar sequences, including Cbei_3974, are annotated in the KEGG database (https://www.kegg.jp/) as l-glyceraldehyde-3-phosphate reductases (10, 11). Our results demonstrate that this annotation is incorrect for Cbei_3974.

Only minimal activity was found for the previously identified substrates, furfural and butyraldehyde (7), while the enzyme had no measurable activity with benzaldehyde at the concentration tested (2 mM). In contrast, 4-pyridinecarboxaldehyde, which differs from benzaldehyde only by the presence of a nitrogen atom in the aromatic ring, was the most kinetically efficient among all of the substrates examined. Turnover was 17 times faster than for l-glyceraldehyde-3-phosphate and 11 times faster than for furfural. Similarly, propionaldehyde did not show activity at 2 mM, while the more polar methylglyoxal, a dialdehyde of the same chain length, gave strong activity 6.8 times faster than furfural. No activity could be detected with ketones or alcohols at the concentrations tested. The enzyme was specific for NADPH, with no activity detectable with 0.4 mM NADH. To determine whether the differences between substrates were caused by differences in k_cat or K_m, steady-state kinetics were measured for the 5 fastest substrates, not including 4-nitrobenzaldehyde, which was not soluble enough to achieve saturation. All the aldehyde substrates resulted in Michaelis constants in the millimolar range, but NADPH had higher affinity as indicated by a lower K_m of 32 μM (Table 2). The best substrate was 4-pyridine carboxaldehyde, with a k_cat of 10 s⁻¹ and a K_m of 3.87 mM. This was closely followed by methylglyoxal, with a similar k_cat of 8.52 s⁻¹ but a higher K_m of 12.9 mM. Furfural has previously been shown to have an extraordinarily high k_cat of 1.4 × 10⁵ s⁻¹ at 40°C (7). In our experiments the k_cat was measured at 19°C and was 2.72 s⁻¹. This was lower than expected, even considering the lower temperature, but is more realistic. The K_m of 34.9 mM measured here is in agreement with the literature value (7).

The hydride transfer in aldehyde reductases occurs from either the pro-R or pro-S hydrogen on C-4 of the nicotinamide ring of NADPH. Typically, short-chain reductases (SDRs) transfer the pro-S hydrogen, while aldo-keto reductases (AKRs) transfer the pro-R hydrogen (12). To determine the stereospecificity of Cbei_3974, the enzyme was incubated with NADPH or (4R)-[4-²H]NADPD with an excess of the substrate 4-pyridinecarboxaldehyde. The reaction products were analyzed to determine whether the deuterium had been incorporated into the alcohol product or remained on the nicotinamide cofactor. The alcohol reaction product was extracted with chloroform and analyzed by GC-MS. The incubation with (4R)-[4-²H]NADPD gave a product which was 1 atomic mass unit larger than that from the incubation with NADPH, consistent with the incorporation of deuterium. In a duplicate reaction, the nucleotide product, NADP⁺, was purified by ion-exchange chromatography, freeze-dried, and dissolved in D₂O. Nuclear magnetic resonance (NMR) analysis of NADP⁺ from the reaction compared with commercial NADP⁺ showed identical spectra, confirming that the deuterium at the pro-R position had been transferred from (4R)-[4-²H]NADPD (Fig. 2). Cbei_3974 therefore transfers the pro-R hydrogen from NADPH, in accordance with other members of the AKR superfamily.

To ascertain if hydride transfer is the rate-limiting step, the substrate kinetic isotope effect (KIE) for NADPH versus (4R)-[4-²H]NADPD was determined for a range of substrates. A KIE on k_cat of 2.13 to 2.58 was observed across the 5 substrates tested (Fig. 2), suggesting that the catalytic step is at least partially rate limiting for all substrates tested.

The effect of protein dynamics on catalysis was investigated by heavy-enzyme production, where the nonexchangeable carbon and nitrogen atoms were replaced with their heavy counterparts (¹³C, ¹⁵N) to slow protein motions without affecting the electrostatics. A reactivity difference between the “heavy” (labeled) and “light” (natural-abundance) enzymes indicates that protein motions affect the catalysis (13). As the substrate KIE measurements indicated that hydride transfer is partially rate limiting, steady-state measurements were used to determine any effect that slower protein motions in the heavy enzyme may have on the catalytic step.

Heavy enzyme (¹⁵N, ¹³C) was produced in M9 medium with labeled feedstocks and purified to homogeneity (see Fig. S3 in the supplemental material). The incorporation of the heavy isotopes was confirmed by mass spectrometry on the purified enzyme, which revealed a 5.5% mass increase (see Fig. S4 in the supplemental material). To determine if the protein was correctly folded, the circular dichroism (CD) spectrum and melting temperature were recorded and compared for both the “heavy” and “light” enzymes (see Fig. S5 in the supplemental material). The two enzymes gave identical spectra and had nearly identical melting temperatures of 62.4°C ± 0.1 and 63.8°C ± 0.2, respectively, indicating that isotopic labeling does not significantly alter protein folding.

Steady-state kinetics were used to determine the k_cat values for the “heavy” and “light” enzymes with a range of substrates at 19°C (3). All substrates gave an enzyme KIE of near unity, implying that there were no mass-dependent effects and that dynamic coupling was minimal (Fig. 3). Although some authors have proposed that enzymes use “promoting motions” to drive catalysis (14–17), this result is consistent with a growing body of literature that shows that dynamic effects become significant only outside physiological conditions and only when poorly tolerated substrates that necessitate rearrangement of the active site are utilized (18–20). The enzyme does not, therefore, use dynamic motions as a part of its catalytic mechanism.

A recent study on the thermophilic Geobacillus stearothermophilus alcohol dehydrogenase (BsADH) showed that significant heavy-enzyme KIEs manifest only below its physiological temperature (40°C) and only with poor substrates (19). The temperature dependency of heavy-enzyme KIEs has been suggested to be an indicator of whether an enzyme is optimized for utilization of a particular substrate (19). The temperature dependence of the KIE for Cbei_3974-catalyzed reduction of 3-pyridine carboxaldehyde was constant over the temperature range from 11 to 44°C (Fig. 4), suggesting that the active-site architecture of the enzyme is optimized for this substrate.

The protein was cocrystallized with NADPH and the structure solved by molecular replacement using PDB entry 5T79, which is the crystal structure for STM2406, an AKR from Salmonella enterica serovar Typhimurium of unknown physiological function but with a substrate profile similar to that of Cbei_3974 (21). The two proteins have 60.91% sequence identity and a root mean square deviation of 0.89. The structure was refined at 1.75 Å to an R_factor of 16.5% (R_free of 19.3%). The structure consists of alternating α-helices and β-strands forming an 8-stranded TIM barrel with some extra helices (Fig. 5).

This motif is conserved across the AKR superfamily (22). Both this structure and STM2406 have an unusual N terminus consisting of a long loop and a β-hairpin. Most AKR structures, including the structure of Coptotermes gestroi AKR1 (another AKR known to reduce furfural), have a shorter N-terminal tail consisting of only the β-hairpin or, in the case of the AKR7 family, have no N-terminal tail (21, 23, 24). The function of this extra sequence is unclear. Conversely, the C terminus is truncated and is lacking a loop that is present in many AKRs, leaving the active site exposed to solvent (24). AKRs which omit this loop have low catalytic efficiency consistent with the measured millimolar Michaelis constants (21, 25, 26). Deletion of the C-terminal loop from human aldose reductase AKR1B1 (26), Bacillus subtilis YhdN and YvgN, and Pseudomonas aeruginosa PA1127 (21) resulted in a dramatic loss of catalytic efficiency.

NADPH sits in a mostly open cleft with a hydrophobic center and polar residues at the ends where the adenine base and nicotinamide ring bind. The adenine base is held in place by hydrogen bonds to Glu-307 and Asn-308. The nucleotide 2′-phosphate that distinguishes NADPH from NADH is hydrogen bonded to Gln-304 and Ser-300. The diphosphate makes hydrogen bond contact with the backbone oxygen of Leu-225.

There is an area of missing electron density between residues Ile-238 and Leu-256. In Coptotermes gestroi AKR1 and human aldose reductase, this region forms a mobile loop that would strap the cofactor in place across the diphosphate bridge (23, 27). The lack of density in Cbei_3974 indicates that the region is disordered and suggests that the loop is not trapping the cofactor.

The canonical mechanism of AKRs involves hydride transfer from NADPH to the carbonyl acceptor (12). This is followed by protonation from an active-site tyrosine as part of a proton relay from histidine and bulk water (12). Neighboring aspartate and lysine residues lower the pK_a of tyrosine to enable it to function as an acid (12). In the close homologue STM2406, the catalytic tetrad consists of Tyr-66, Asp61, Lys-97, and His-138 (21). All these residues are conserved in Cbei_3974 (identical numbering). It was not possible to obtain crystal structures of protein-product complexes due to the low affinity of the ligands. Therefore, docking was used to predict the possible binding of substrates. AutoDock Vina (28) was used to dock furfural and the best substrate, 4-pyridine carboxaldehyde, into the active site. The best pose was selected on the basis of proximity to NADPH and the proposed catalytic residues. These poses are illustrated in Fig. 6. Both substrates are orientated toward the pro-R hydrogen of NADPH, consistent with the experimentally determined stereochemistry. The carbonyl oxygen of 4-pyridine carboxaldehyde is within hydrogen bonding distance of the exocyclic amide of NADPH and makes hydrophobic contacts with residues Asn-65, Trp-33, and Tyr-100. These residues are conserved in STM2406 and have been shown to be important for binding in that enzyme (21). Furfural docked into the active site in a similar location but with a different orientation, possibly due to its smaller size. The active site has many polar residues, which may explain why the more hydrophobic aldehydes such as benzaldehyde are less favored. Asn-65 and Tyr-100 contribute to the polar surface of the active site, and therefore alteration of these residues to more hydrophobic ones may help improve activity for hydrophobic substrates. In STM2406, which has a very similar active site, the variant Asn-65-Met (both enzymes have the same residue numbering) gave a 341% increase in activity toward 3-pyridinecarboxaldehyde compared with that of the wild type and a 2-fold decrease in K_m (21). Alterations of Tyr-100 to aspartate, leucine, isoleucine, or valine mostly resulted in insoluble proteins, while Tyr-100-Ala showed decreased activity, but this may have been due to a loss of steric bulk by replacing a phenyl group with a hydrogen atom (21).

A pET-14b vector harboring a codon-optimized gene encoding Cbei_3974 was purchased from GenScript (the sequence is given in Fig. S6 in the supplemental material). This also encodes a 6× His tag and thrombin cleavage site upstream of Cbei_3974.

NADPH was obtained from Fisher or Apollo Scientific. (4R)-[4-²H]NADPD was prepared from NADP⁺ (Melford) and isopropanol-d8 (Acros) according to the published protocol (33). ¹⁵NH₄Cl₂ and [¹³C]glucose were obtained from Goss Scientific, Cheshire, UK. Furfural was obtained from VWR, methylglyoxal from Apollo Scientific, and 3-pyridine carboxaldehyde from Acros. All other chemicals were obtained from Sigma-Aldrich.

Cbei_3974 was overproduced in BL21(DE3) cells and purified by Ni affinity chromatography as previously described (7). The protein was dialyzed against 10 mM HEPES-NaOH (pH 7.5)–300 mM NaCl and concentrated to 10 mg/ml. Crystallization trials were performed using the sitting-drop vapor diffusion method by mixing 0.5 μl protein stock solution and 0.5 μl of a seed stock with 0.5 μl reservoir solution. The seed stock was obtained from microcrystals grown in 100 mM morpholinepropanesulfonic acid (MOPS) (pH 7.3)–13% polyethylene glycol (PEG) 8000–750 mM NH₄Cl with 2 mM NADPH and 2 mM 4-pyridine methanol. Diffracting crystals were obtained from 90 mM MOPS (pH 7.6)–271 mM NH₄Cl–2.7% PEG 8000, with 10 mM NADPH and 2 mM 4-pyridine methanol added to the protein prior to crystallization. The crystals were transferred to cryoprotectant (90 mM MOPS [pH 7.6], 271 mM NH₄Cl, 2.7% PEG 8000, 25% ethylene glycol) and flash-frozen with liquid nitrogen.

The X-ray diffraction data were collected at 100 K at Diamond Light Source (Oxfordshire, UK) on beamline I04-1 and integrated with XDS (34) in the xia2 package (35). The data were scaled, reduced, and analyzed with AIMLESS and TRUNCATE in CCP4i (36). The structure was solved by molecular replacement with PHASER (37) using coordinates from PDB no. 5T79 as a searching model (21). The structure model was adjusted with COOT (38) and refined with REFMAC5 (39). Graphical representations were prepared in Chimera (40), PyMOL (The PyMOL Molecular Graphics System, version 1.8.X [Schrödinger, LLC]), YASARA View (41) and LigPlot+ (42). X-ray crystallography data and refinement statistics are given in Table S1 in the supplemental material.

Ligand structures were downloaded from PubChem (https://pubchem.ncbi.nlm.nih.gov/) (43) as SDF files and converted into mol2 format using Chimera (40). Ligand and protein were converted into PDBQT format using AudoDockTools1.5.6 and docked using AutoDock Vina (28). Graphical representations were prepared in PyMOL (The PyMOL Molecular Graphics System, Version 1.8.X [Schrödinger, LLC]), Chimera (40), and LigPlot+ (42).

Arctic Express DE3 cells harboring pET-14b_Cbei3974 were grown in 20 ml M9 minimal medium containing 100 μg/ml carbenicillin overnight at 37°C with shaking. This was diluted 1:50 into 0.5 liters of M9 medium containing either natural-abundance isotopes or [¹³C]glucose and ¹⁵NH₄Cl₂ for heavy-enzyme production. Cultures were grown to an optical density at 600 nm (OD₆₀₀) of 1.0 at 37°C and 220 rpm. Cultures were cooled to 12°C, and 0.4 mM IPTG (isopropyl-β-d-thiogalactopyranoside) was added to induce gene expression. Cells were harvested at 4,000 rpm after 16 h of growth at 12°C and 200 rpm. The protein was purified as previously described (7).

Cbei_3974 (0.585 μM), 0.4 mM NADPH, NADH, or NADP⁺, and 2 mM putative substrate were mixed in 20 mM potassium phosphate buffer, pH 7.0. The subsequent change in NADPH concentration was monitored at 340 nm (ε₃₄₀ = 6,220 M⁻¹cm⁻¹) for 1 min, using a Shimadzu UV-2401PC spectrophotometer in 5-mm quartz cuvettes, to give the reaction rate.

Kinetic parameters were determined with 117 nM Cbei_3974 with one of the two substrates; one was held at a saturating level, while the other was varied across a concentration range of 0 to 75 mM (aldehydes) or 0 to 0.2 mM (NADPH). Rates were measured as described above and the data fitted to the Michaelis-Menten equation using GraphPad Prism version 7.00 for Windows (GraphPad Software, La Jolla, CA, USA). Each data point is the average from 3 repeats. For heavy-enzyme kinetics, a minimum of two data sets of triplicates were collected.

Liquid chromatography-mass spectrometry (LC-MS) was performed on a Waters Synapt G2-Si quadrupole time-of-flight mass spectrometer coupled to a Waters Acquity H-Class ultraperformance liquid chromatography (UPLC) system. The column was an Acquity UPLC protein BEH C4 (300 Å, 1.7 μm by 2.1 mm by 100 mm) operated in reverse phase and held at 60°C. The gradient employed was 95% A to 35% A over 50 min, where A is H₂O with 0.1% HCO₂H and B is acetonitrile (ACN) with 0.1% HCO₂H. Data were collected in positive ionization mode and analyzed using Waters MassLynx software version 4.1. Deconvolution of protein charged states was obtained using the maximum entropy 1 processing software.

Mixtures containing 5 μM Cbei_3974, 4.6 mM NADPH or (4R)-[4-²H]NADPD, and 28 mM aldehyde in a final volume of 40 μl 20 mM potassium phosphate (pH 7.0) were incubated at 40°C for 1 h. Aliquots (4 μl) were removed at time zero and at 5 or 7 h and quenched with 1 ml of chloroform, which also served to extract the organic molecules. A 5-μl aliquot of the organic layer was injected onto a PerkinElmer Clarus 680 gas chromatograph. The initial temperature was 40°C, which was held for 1 min, and elution was with a gradient rising to 150°C at 15°C/min, holding at 150°C for 1 min. After a 3-min solvent delay, mass spectra were collected over the range of 45 to 200 E+. Control reactions with either enzyme or NADPH omitted were also performed, and a standard of furfuryl alcohol was run on the GC-MS.

Cbei_3974 (1 μM), 2 mM (4R)-[4-²H]NADPD, and 25 mM 4-pyridine carboxaldehyde were incubated at 37°C for 3 h. NADP⁺ was purified from the reaction on a SAX-10 column using published methodology (33), freeze-dried, and redissolved in D₂O. Commercial standards of NADPH and NADP⁺ were also dissolved in D₂O. NMR spectra of standards and reaction product were collected on a Bruker 400 instrument.

Circular dichroism (CD) measurements were performed on an Applied Photophysics Chirascan spectrometer using 7 μM protein in 20 mM potassium phosphate buffer (pH 7.0)−20% glycerol. Spectra were recorded over a temperature range of 5 to 85°C from 200 nm to 400 nm. The melting temperature was calculated by fitting the data in SigmaPlot (Systat Software, San Jose, CA).

The X-ray structure solved in this study was deposited into the Protein Data Bank (http://www.rcsb.org/pdb/) with accession code 6HG6.