HMF 1 (98%) and cyclohexanone (99%) were purchased from Aladdin (Shanghai, China). Furfural 2 (99%), BHMF (98%), and propylamine (98%) were obtained from Macklin Biochemical Co., Ltd. (Shanghai, China). 5-Methoxymethylfurfural 3 (97%) was bought from Adamas Reagent Ltd. (Shanghai, China). 5-Methylfurfural 4 (97%) and furfuryl alcohol (98%) were purchased from TCI (Japan). NAD(P)H was obtained from Yuanye Biotech (Shanghai, China). 5-Methylfurfuryl alcohol (98%) was purchased from Apollo Scientific Ltd. (United Kingdom). 5-Methoxymethylfurfuryl alcohol was synthesized and purified according to a recent method (Cheng et al., 2020). Other chemicals are of analytical grade and commercially available.

Escherichia coli BL21(DE3) and plasmid pET-28a were obtained from Novagen Inc. (Madison, United States). Protein marker, restriction endonucleases XhoI and NdeI, and T4 DNA ligase were purchased from Thermo Fisher Scientific GmbH (Schwerte, Germany). Q5 High-Fidelity DNA polymerase was obtained from New England Biolabs Inc. (United States). DpnI endonuclease was purchased from TaKaRa Biotechnology Co., Ltd. (Dalian, China). Isopropyl β-d-1-thiogalactopyranoside (IPTG), DNA marker, and ampicillin were purchased from Sangon Biotech (Shanghai, China). Glucose dehydrogenase (GDH) from Bacillus megaterium (20 U/mg) was prepared according to a previous method (Jia et al., 2019a). Cloning, heterologous expression, site-directed mutagenesis, and protein purification of AspRedAm and variants were available as Supplementary Material.

The ADH activities were determined spectrophotometrically by measuring the decrease in the NAD(P)H absorbance at 340 nm under pH 8/9 and 30°C. The reaction mixtures contained 43 μg/ml AspRedAm/37 μg/ml N93A, 5 mM HMF, and 0.2 mM NAD(P)H in Tris–HCl buffer (100 mM, pH 8 for N93A; pH 9 for AspRedAm). One unit (U) was defined as the amount of enzyme that catalyzes the oxidation of 1 μmol of NAD(P)H per minute under the above reaction conditions. The protein concentration was measured with Bradford method (Bradford, 1976), using bovine serum albumin as the standard.

Typically, 2 ml of Tris–HCl buffer (100 mM, pH 8) containing 25 mM of furans, 30 mM glucose, and 50 mg/ml of whole cells (wet weight) was incubated at 30°C and 150 rpm. Aliquots were withdrawn from the reaction mixtures at specified time intervals and diluted with the corresponding mobile phase prior to HPLC analysis. The conversion was defined as the ratio of the consumed substrate amount to the initial substrate amount (in mol). The selectivity was defined as the ratio of the amount of the desired product to the sum of all the products (in mol). All the experiments were conducted at least in duplicate, and the values were expressed as the means ± standard deviations.

Molecular docking was performed for both the WT and variant N93A in complex with substrate HMF and NADH. Modeller (version 9.13) was used to generate the 3D structure of variant N93A with NADH using the WT enzyme (PDB ID: 5G6R) as the template. Then AutoDock (version 4.2) was used for substrate docking, and 250 docking poses were obtained for each system. The simulated systems were named WT-NADH-HMF and N93A-NADH-HMF, respectively. Configuration clusters were analyzed for the reasonable binding mode with the highest docking energy.

The reaction mixtures were analyzed on a Zorbax Eclipse XDB-C18 column (4.6 mm × 250 mm, 5 μm, Agilent, United States) by using a reversed phase HPLC (Waters, United States) equipped with a 996 PDA detector. Unless otherwise stated, the mobile phase for HPLC analysis was a mixture of acetonitrile/0.4% (NH₄)₂SO₄ aqueous solution (pH 3.5, 1:9, v/v) at a flow rate of 0.6 ml/min. A mixture of acetonitrile/0.4% (NH₄)₂SO₄ aqueous solution (pH 3.5, 2:8, v/v) at a flow rate of 0.8 ml/min was used for monitoring the transformation of compound 3, while the mixture of acetonitrile/0.02% H₃PO₄ solution (4:6, v/v) acted as the mobile phase at a flow rate of 0.3 ml/min for monitoring the transformation of 4. HPLC chromatograms are available as Supplementary Material.

Based on the crystal structure of AspRedAm (Aleku et al., 2017) and the results of this protein docking with HMF and cofactor (Supplementary Figure 1), the residues within 4 Å around substrate HMF such as N93, I118, M119, A120, V121, and D169 were identified as the potential hotspots for mutation, since they may interact with substrate and cofactor. The catalytic performances of variants showing high selectivities toward BHMF were summarized in Figure 1, with the WT enzyme as control. In addition to the reductive amination product (N-propyl furfurylamine), the reduction product BHMF (the selectivity of around 23%) was also produced in reductive amination of HMF by AspRedAm, indicating the promiscuous ADH activity of this enzyme. Interestingly, these variants did not show the RedAm activities with HMF as substrate, but ADH activities (Figure 1), since no formation of N-propyl furfurylamine was observed. Notably, the RedAm activities of some variants (e.g., N93A) toward cyclohexanone remained (data not shown). Besides N93 and D169 (Aleku et al., 2017), other residues I118 and M119 were also recognized as the new important sites for reductive amination reactions. Among the mutants tested, variants N93A and N93S provided the best results in the reduction of HMF, with the substrate conversions of >99% and the BHMF selectivities of >99% (Figure 1). The ADH activities of the two variants were further compared in the absence of amine donor (Supplementary Table 2). Interestingly, mutation of N93 to alanine resulted in a 6.6-fold increase in the ADH activity, while no significantly improved activity was found with variant N93S. Like asparagine, serine has a more hydrophilic and larger side chain compared to alanine, which may result in unfavorable interactions with NADPH and/or substrate. Thus, variant N93S also showed a low activity. So, the enzyme properties of variant N93A were explored in the subsequent studies.

Both RedAms and ADHs are NAD(P)H-dependent enzymes, and cofactors have usually a great effect on their activities. On the other hand, NADPH is costlier than NADH. Therefore, the cofactor preference of the WT enzyme and variant was examined (Table 1). Unexpectedly, the WT enzyme preferred NADH as cofactor in the reduction reactions regardless of substrates (Table 1). However, it favored NADPH in the reductive amination reactions (Supplementary Table 3), which is consistent with the previous results (Aleku et al., 2017). It suggests that there may exist the mechanistic differences between the reduction and reductive amination reactions. According to Borlinghaus et al. (Borlinghaus and Nestl, 2018), all characterized members of the IRED family showed strict NADPH preference in the reductive amination. Variant N93A showed the NADH preference in the reduction of both HMF and cyclohexanone. Although the improvement in the ADH activities was achieved upon mutation, the activity of variant N93A toward HMF remained unsatisfactory (<0.1 U/mg). And it is much lower than those of some natural ADHs in the HMF reduction (Almeida et al., 2008; Laadan et al., 2008). Therefore, further improvement in its ADH activity by molecular modification is required for its future practical applications.

Effect of pH and temperature on the ADH activities and stability was evaluated (Figure 2). Figure 2A shows that the WT enzyme has the highest activity at pH 9 in the reduction of HMF, which is consistent with that in the reductive amination (Aleku et al., 2017). Its optimal pH slightly decreased to 8 upon mutation of N93 to alanine. Besides, the variant exhibited the relative activities of more than 70% within pH 7–9, indicating its good pH tolerance. As shown in Figure 2B, the optimal temperature of both the WT and variant was 30°C in the reduction, which is in agreement with that in the reductive amination (Aleku et al., 2017). Additionally, variant N93A displayed higher activities than the WT enzyme at 40–45°C (Figure 2B). The thermostability of the WT enzyme and mutant was assessed (Figure 2C). Intriguingly, variant N93A was found to be more thermally stable than the WT enzyme, regardless of whether cofactor is present or absent. In the absence of NADH, for example, variant N93A retained 65% of the residual activity after incubation of 3 h, while only 12% of the residual activity was observed with the WT enzyme. It may be one of the promising features of this variant for its application in synthetic chemistry.

Table 2 shows the effect of metal ions on the ADH activities of AspRedAm and variant N93A. Except for Al³⁺, almost all the metals tested exerted a great inhibitory effect on the ADH activities of the WT enzyme. The precipitation (possibly protein) occurred in the presence of Co²⁺ and Ni²⁺, resulting in no observed activities. A very low activity of AspRedAm was recorded in the presence of Mn²⁺. Likewise, the activities of variant N93A were significantly inhibited by Mn²⁺ and Co²⁺, although no precipitation happened.

The reaction kinetics were studied in the HMF reduction catalyzed by AspRedAm and variant N93A, and the kinetic constants were determined based on the Lineweaver–Burk plots (Table 3). It is worth noting that no substrate inhibition toward enzymes is observed within the substrate concentration range tested (data not shown). The K _m value of the WT enzyme was approximately 1.3 μM toward HMF, while the corresponding value of variant N93A approached 4.1 μM. The increased K _m value of the mutant indicates its lower substrate affinity toward HMF compared to that of the WT enzyme. Nonetheless, variant N93A provided a greatly improved k _cat value compared to the WT enzyme (96.7 vs 17.5 ms⁻¹). Therefore, the catalytic efficiency of the mutant (k _cat/K _m) was approximately 2-fold higher than that of the WT enzyme. To get mechanistic insights into the increased catalytic activity upon N93A mutation, molecular docking was performed for both the WT enzyme and variant N93A in complex with substrate HMF and NADH (Figure 3). As shown in Figure 3, the distance between C_NADH and O_LIGAND decreased from 5.3 ± 0.3 Å in WT-NADH-HMF to 3.9 ± 0.2 Å in N93A-NADH-HMF, and the shorter distance is more appropriate for hydride transfer. Besides, the increased docking energy was observed in N93A-NADH-HMF (−4.0 ± 0.1 kcal/mol) compared to that (−3.7 ± 0.2 kcal/mol) in WT-NADH-HMF, indicating stronger binding of substrate to variant N93A. The reason for the above-mentioned results may be that the smaller side chain of Ala93 allows substrate to bind more strongly in the binding pocket, due to its reduced steric hindrance.

It is well known that the substrate tolerance of enzymes is of great importance for their synthetic applications. Broad substrate specificity enables enzymes to be facilely applied to produce a variety of chemicals. Therefore, the substrate scopes of the WT and mutant were investigated (Table 4). AspRedAm was reported to have broad substrate specificity in reductive amination (Aleku et al., 2017). Also, the WT enzyme could accept all the tested chemicals as substrates in the reduction reactions, in spite of low promiscuous ADH activities (<35 mU/mg), and compound 11 was considered optimal. Increased activities were observed with variant N93A, and the highest activity of approximately 96 mU/mg was obtained when compound 7 acted as substrate. In general, variant N93A displayed higher activities toward phenyl-containing substrates (compounds 9–14) than furan aldehydes (1–8). The similar phenomenon was also found in the case of the WT enzyme. Recently, we also found that aldehyde dehydrogenases preferred benzylaldehydes to furan aldehydes as substrates (Zhang et al., 2020b). Therefore, it seems to be a common catalytic feature of enzymes, likely because of the higher richness of benzene-based chemicals in nature. In addition, the mutant as well as the WT enzyme could accept aliphatic ketones (15 and 16) as substrates. Like the WT enzyme, variant N93A had broad substrate specificity, indicating its application potential in synthetic chemistry.

To demonstrate the applicability of the engineered enzyme, biocatalytic reduction of biobased furans 1–4 to the target alcohols by whole cells harboring variant N93A was performed (Figure 4), with the cells harboring the plain vector as controls. It was found that HMF 1 could be reduced to BHMF in 18 h in the presence of whole cells harboring variant N93A, with the conversion of 98% and selectivity of 99%. A slightly lower conversion (95%) and a slightly lower selectivity (93%) were obtained in control. It suggests the presence of inherent ADHs in E. coli cells that enable the HMF reduction. Similarly, the substrate transformation occurred with other biobased furans as substrates in controls. However, the substrate conversions (86–>99%) were higher in the presence of whole cells harboring variant N93A compared to those (35–69%) in controls harboring the plain vector. For example, whole cells harboring variant N93A provide a conversion up to 96% within 6 h in the reduction of furfural 2, while the conversion of 58% was obtained in control. Higher conversions might be attributed to the introduction of variant N93A to E. coli cells. Whole cells harboring variant N93A exhibited significantly different catalytic efficiencies toward furans 1–4, although the activities of variant N93A toward these substrates were comparable (70–83 mU/mg, Table 4). For example, the reaction rate of furfural 2 was much higher than that of HMF 1 (times required for almost complete substrate conversion: 18 vs 6 h), which is in good agreement with our recent results (Cheng et al., 2020). In addition to the activities of variant N93A, the transport rates of substrates across hydrophobic cell membranes might influence the reaction rates in whole-cell catalysis as well (Almeida et al., 2007; Wierckx et al., 2011). More hydrophobic furfural 1 (log p value, 0.75) might be more readily transported into the cells for subsequent biotransformation than HMF (−0.10), which may make partial contribution to its higher reaction rate.