All chemicals were obtained from commercial suppliers Sigma-Aldrich, Acros Organics, Supelco, Thermo Scientific and Combi-Blocks with the highest purity available. Glucose dehydrogenase GDH-105 was generously donated by Codexis^®. GC-FID analysis were performed either on a Shimadzu GC-2010 gas chromatographs (Shimadzu, Japan) with an AOC-20i Auto injector equipped with a flame ionization detector (FID) and using nitrogen as the carrier gas (GC-FID 1), either on a Gas Chromatograph Trace 1,300 (Thermo Scientific) equipped with an autosampler injector AI/AS1310, a Flame Ionization Detector (FID) and a H₂ Generator Alliance (F-DGSi) (GC-FID 2). All samples were injected with GC quality solvents. The AmDHs used in this study have already been described (Caparco et al., 2020; Ducrot et al., 2021; Jongkind et al., 2022; Bennett et al., 2022; Ducrot et al., 2022). New batches were produced, according to the protocol already described, for IGCAmDH1, IGCAmDH1-W144A, RgnaAmDH, RgnaAmDH-W155A, MATOUAmDH2, ApauAmDH, ApauAmDH-W141A, PortiAmDH, PortiAmDH-W140A, MicroAmDH, MsmeAmDH-W141A, SgorAmDH, IGCAmDH5 and AcolAmDH (Supplementary Figure S1). For CfusAmDH, CfusAmDH-W145A, ChatAmDH, MATOUAmDH2-C148A and MsmeAmDH, previous batches obtained from previous studies and stored at −80°C were used (Jongkind et al., 2022; Ducrot et al., 2022). Spectrophotometric assays were recorded on a Safas UVMC2 (Safas, Monaco) thermostated at 30°C with a refrigerated/heating circulator Corio CD-200F (Jubalo^®, Seelbach, Germany) using microcells high-precision cell quartz with 10-mm light path (Hellma Analytics, Müllheim, Germany).

On GC-FID 1, the samples were injected on a CP-Sil 5 CB (25 m × .25 mm × 1.20 µm) with the following parameters: Injection at 340°C, split ratio 100, linear velocity 30 cm/s, column flow 4 mL min⁻¹, nitrogen as carrier gas. Details on the oven temperature program and retention times are given in Supplementary Table S1. For chiral GC-FID analyses, the samples were injected on CP-Chirasil Dex CB (Agilent J&W) (25 m × .32 mm × .25 µm, injection at 250°C, split ratio 25, linear velocity 30 cm/s, column flow 1.68 mL/min, helium as carrier gas) for compounds 9a, 9c, 16a, and 16c and on Hydrodex β-TBDM (50 m × .25 mm × .15 µm, injection at 250°C split ratio 25, linear velocity 38 cm/s, column flow 2.35 mL/min, helium as carrier gas) for 11a and 11c. Details on the oven temperature program and retention times are described in Supplementary Table S2. On GC-FID 2, samples were injected on TG-35MS AMINE column (30 m × .25 mm × 1 μm; Thermo Scientific), TG-624SILMS column (30 m × .25 mm × 1.4 µm; Thermo Scientific), ZB-XLB column (20 m × .18 mm × .18 µm; Phenomenex) or CP-Chirasil Dex-CB column (25 m × .32 mm × .25 μm; Agilent) with H₂ as carrier gas (30 mL min⁻¹). Details on the oven temperature programs, split ratio, detector and inlet temperature and retention times are detailed in Supplementary Table S3.

The reactions were performed in 100 mM NaPi buffer at pH ranging from 6.0 to 9.0 in a final volume of 1.0 mL. A solution of 10 mM substrate 15a, .2 mM NADP⁺, 12 mM glucose, 3 U mL⁻¹ glucose dehydrogenase (GDH-105), .1 mg mL⁻¹ MsmeAmDH, 1% v/v DMSO was stirred in a thermomixer at 400 rpm and 25°C for 4 h. Extraction was carried out with 1 mL of ethyl acetate (EtOAc), the organic layer was dried over MgSO₄, and analyzed on GC-FID 1. Cyclohexanol (15c) concentrations were deduced from a calibration curve equation using 5 mM dodecane as an internal standard and 15c reference standards. GC-FID chromatograms of reference standards, control reaction and one example of reaction with MsmeAmDH are provided in Supplementary Figure S2.

All the selected AmDHs were screened for conversion of a range of carbonyl-containing substrates to corresponding amines and alcohols as follows. To a reaction mixture (200 μL in 500 μL Eppendorf tubes) containing 10 mM carbonyl-containing substrate (for substrates 3, 4, 8, 11, 13, and 14, 200 mM stock solutions were prepared in DMSO rather than H₂O, with 5% v/v in the final reaction mixture), .2 mM NAD⁺, .2 mM NADP⁺, 3 U mL⁻¹ GDH-105, 12 mM D-glucose in the appropriate buffer (for reaction with ammonia/ammonium species: 2 M NH₄HCO₂/NH₄OH buffer at pH 8.5, for ammonia-free reactions: 100 mM sodium phosphate buffer at pH 7.5), was added .5 mg mL⁻¹ of purified enzymes. For each substrate, a blank mixture was prepared in the same manner but lacking the enzyme replaced by an appropriate volume of desalting buffer used for protein purification. The reaction mixtures and the calibration points were stirred at 25°C for 24 h at 400 rpm. To 80 µL of mix reaction were added 20 µL of 10 M NaOH and an extraction with 200 µL was performed with solvent (EtOAc or methyl tert-butyl ether (MTBE)) containing known amount of internal standard (1.5 mM dodecane or toluene) related in Supplementary Table S3. The organic layers were then analyzed on GC-FID 2 according to conditions detailed in Supplementary Table S3. Product b and c concentrations were deduced from calibration curves equation prepared with various ratio of ketone/alcohol/amine (or ketone/alcohol in the case of ammonia-free reactions) in the corresponding buffer and in presence of an appropriate volume of desalting buffer. Examples of GC-FID chromatograms are provided in Supplementary Figure S3. In addition, the organic layers were analyzed on GC-FID 1 according to conditions detailed in Supplementary Table S2 with chiral columns to determine the enantiomeric and diasteroisomeric excesses of chiral alcohols. Chiral GC-FID chromatograms are provided in Supplementary Figure S4.

Kinetic parameters were determined by spectrophotometric NADPH-monitoring in the forward ketoreduction direction at 340 nm. The reactions were performed in 100 mM potassium phosphate buffer pH 7.5 in a final volume of 100 μL at 30°C in a spectrophotometer cell with optical paths of 1 cm. Initial rates of the reaction were measured after addition of the purified enzyme (.1 mg mL⁻¹) with various concentrations of 15a (or NADPH) and saturated concentrations of NADPH at .2 mM (or 15a at 2 mM). The uncertainties are those generated by the fitting and the data are averages of three independent experiments. The conditions, fitting and determination of kinetic parameters were performed with Sigma Plot software and plots are provided in Supplementary Figure S5.

The reactions were conducted in duplicate at 30°C in a thermostated spectrophotometric cell (10 mm light path) in a final reaction volume of 100 μL. For the KRED activity, the reactions were performed in potassium phosphate buffer pH 7.5 with .2 mM of NADPH and .1–.3 mg mL⁻¹ of enzyme (MsmeAmDH, MATOUAmDH2, CfusAmDH, IGCAmDH1, MicroAmDH). Initial rates of the reaction were measured at 340 nm after addition of 2–10 mM of substrate (6a, 7a, and 15a) to determine the specific activity of the enzyme according to Beer—Lambert’s law and the molar absorptivity of β-NADPH (ε = 6,220 M⁻¹ cm⁻¹) after subtraction of the slope obtained under the same conditions except without enzyme. The calculated specific activities are detailed in Supplementary Table S4. For alcohol oxidation activity, the reaction were performed in 100 mM Tris-HCl pH 8.6 buffer with .2 mM of NADP⁺ and 2 mM of 7c or 15c. Initial rates of the reactions were measured at 340 nm after addition of 1 mg mL⁻¹ of MsmeAmDH to determine the specific activity as mentioned above after subtraction of the slope obtained under the same conditions but lacking substrate.

For study cases in ammoni-free buffer, the products 6a, 6c, 7a, 7c, and 15c were docked into the closed conformations of structures of MATOUAmDH2-NADP (PDB ID: 7ZBO), MsmeAmDH-NADP (PDB ID: 6IAQ) and CfusAmDH-NADP-15b (PDB ID: 6IAU) after removal of the amine 15b, and on models of RgnaAmDH-NAD, MicroAmDH-NAD(P) and IGCAmDH1-NAD (Mayol et al., 2019; Caparco et al., 2020), minimized beforehand to obtain the more favored positioning of the nicotinamide cofactor from its copied position in the CfusAmDH-NADP structure. The latter were previously minimized without restraints, using the YASARA2 force field within YASARA, after addition of the nicotinamide cofactor copied from the CfusAmDH-NADP structure (Krieger et al., 2009). For study cases in presence of ammonia/ammonium with RgnaAmDH, the model of CfusAmDH-15a-NH₄, already published, was used as a template to paste the ammonium molecule, after alignment of models using PyMOL Molecular Graphics System (Version 2.0 Schrödinger, LLC) (Mayol et al., 2019). NAD cofactor was built from the CfusAmDH-NADP structure by replacing the phosphate group (P2B, O1X, O2X and O3X atoms) with a hydrogen atom. The ligand PDB files were generated using the CORINA Molecular Online Tool (https://demos.mn-am.com/corina.html). With AutoDockTool, the docking simulations were performed on rigid structures, with no flexibility given to any catalytic pocket residues (Morris et al., 2009). The box was centered in the active site with the following size: 34, 36, 30 (36, 38, 30 in the case of dockings in presence of NH₃) (Autodock parameters). The number of Genetic Algorithm (GA) runs was fixed at 10 using the Lamarckian GA (4.2). The 10 ligand conformations obtained were then analyzed on PyMOL Molecular Graphics System (Version 2.0 Schrödinger, LLC).

We conducted this study with 12 characterized nat-AmDHs, i.e. ChatAmDH, IGCAmDH1, RgnaAmDH, MATOUAmDH2, CfusAmDH, ApauAmDH, PortiAmDH, MicroAmDH, MsmeAmDH, SgorAmDH, IGCAmDH5 and AcolAmDH, in addition to the recently described seven variants corresponding to CfusAmDH-W145A (Ducrot et al., 2022). We tested these 19 enzymes toward a large panel of carbonyl substrates at 10 mM concentration (Figure 1). Group A comprises alkyl linear aldehydes (isobutyraldehyde (1a), butanal (2a), hexanal (3a), heptanal (4a) and octanal (5a)), group B substituted aldehydes (furfural (6a), cyclopentanecarbaldehyde (7a) and benzaldehyde (8a)), group C linear ketones (pentan-2-one (9a), hexan-2-one (10a), heptan-2-one (11a), hexan-3-one (12a), acetophenone (13a) and 4-phenylbutan-2-one (14a)) and group D cyclic ketones (cyclohexanone (15a) and 2-methylcyclohexanone (16a)). The alcohol 1c–16c and/or amine 1b–16b products were semi-quantified in two reaction conditions (condition a: Buffer with ammonia/ammonium species; conditions b: Buffer without ammonia) by GC-FID monitoring, with a mix of NAD⁺ and NADP⁺ cofactors, internally recycled with glucose dehydrogenase GDH-105. The specific activities of some of the enzyme/substrate pairs were also determined (Supplementary Table S4).

For ammonia-free reactions, previous data reported by (Tseliou et al., 2020) demonstrated higher apparent rates of reduction at pH 7.0 compared to higher pH, and no preferred type of buffer in cofactor-recycling conversion studies. With the objective of achieving the best conversions into alcohols in condition b, we studied the effect of pH on the alcohol formation with the couple 15a/MsmeAmDH in sodium phosphate buffer. The reaction conditions (quantity of enzymes and duration) were deliberately adapted to deviate from quantitative conversions and thus to better visualize the differences according to the pH. As depicted in Figure 2, the alcohol amount 15c was quite similar in the pH range 6.0–7.5 and decreased progressively above pH 8.0. The screening for KRED activity was also done at pH 7.5 unlike the one for AmDH activity performed at pH 8.5 as already published (Mayol et al., 2019).

All the conversions deduced from calibration curves of standard alcohols and amines are presented in Figure 3. Despite alcohols peaks observed in blank reactions lacking enzymes, especially with aldehydes, the alcohol formations are enzymatic and present after subtraction of these blank controls (Supplementary Figure S3). Importantly, in 2 M ammonium buffer (conditions a), the AmDH activity strongly outperformed the KRED one, as, in most cases, no or only traces of alcohol could be detected (<1% conversion). The saturated ammonia/ammonium concentration promotes the amination reaction, presumably minimizing the “free” carbonyl substrate in the active site and so its direct reduction (Mayol et al., 2019). This KRED activity, specific to ammonia-free buffer, differs from the KRED side activity of wild-type or some engineered LysEDHs observed in ammonium buffer for some substrates. For example, 8a was converted into 17% of 8b and 23% of 8c in 2 M ammonium formate buffer with wild-type LysEDH (Tseliou et al., 2020). Here, none of the tested enzymes was revealed to display higher promiscuous KRED activity than AmDH one in ammonium buffer, except PortiAmDH, which displayed 3.4% conversion of 5a into 5c but no conversion into 5b. More notable KRED activities (conversions of 3.1%–9.3%) were reached with variants, even if they were still lower than the AmDH ones.

On the contrary, in ammonia-free buffer (condition b), alcohols were detected, even up to significant amounts. It appeared that, in general, the amounts of alcohol formed were very variable. This was not only linked to the identity of the enzyme itself, but also to the enzyme/substrate couple. Among all the tested substrates, each enzyme displayed notable KRED activity (>.5 mM of alcohol formed) with at least one substrate. Higher alcohol formation and a higher number of accepted substrates were obtained with some enzymes compared with others. RgnaAmDH, MATOUAmDH2, PortiAmDH, MicroAmDH and MsmeAmDH display high promiscous KRED activity in ammonium-free buffer, with up to 96% of alcohols formed. Generally, for wild-type enzymes, we observed much higher AmDH activity in ammonia-rich buffer than ketoreduction in ammonia-free buffer, but with variabilities depending on the type of substrate. For the short substrates 1a and 2a of group A, the conversions into alcohols in condition b were at least twice lower than the conversions into amines in condition a. Up to around conversions of 40% were nevertheless obtained, for example for 1a into 1c with MsmeAmDH, RgnaAmDH and CfusAmDH. For enzymes active toward longer aldehydes, the AmDH activity was 6–20 times higher than the KRED one; no more than 10% of alcohol 3c were obtained in condition b. For group B, unsaturated aldehydes 6a and 8a were not reduced, or only in very low amounts, despite many of the tested enzymes displayed high conversions to the corresponding amines 6b and 8b in ammonia-rich buffer. Specific activities towards 6a were all below .6 mU mg⁻¹, thus in accordance with conversion results (Supplementary Table S4). The KRED activity was observed only for cyclopentanecarbaldehyde 7a, with differing AmDH/KRED activity ratios dependent on the enzymes. RgnaAmDH and MATOUAmDH2 appeared as the enzymes with the higher KRED promiscuous activity for this substrate with 16%–32% and 23%–27% conversions of 7a into 7c respectively, and IGCAmDH1 and ChatAmDH the less active ones for the direct reduction while forming 63%–76% of cyclopentanecarbamine (7b). The specific activities determined with 7a in ammonia-free buffer correlates with such KRED activity data, with .4 and 14.6/58.4 mU mg⁻¹ measured for IGCAmDH1 and RgnaAmDH/MATOUAmDH2 respectively. In group C, for 9a, the KRED activity was significant in the cases of enzymes displaying high conversions in reductive amination, with AmDH/KRED activity ratios of around 4/1 for CfusAmDH and MicroAmDH to 2/1–1/1 for PortiAmDH and MsmeAmDH respectively. For example, 53% conversion into 9b vs. 54% into 9c, in the respective ammonia-rich and ammonia-free conditions, were obtained with the couple 9a/PortiAmDH. This trend was not shared for hexan-3-one (12a), for which no alcohol 12c was detected despite high amine 12b formation for PortiAmDH, MicroAmDH, and MsmeAmDH. For the other substrates of this group displaying activities only with variants, the amounts of alcohol were only significant with CfusAmDH-W145A, with 17%–36% of 11c and 14c, when 54% and 53%–70% of corresponding amines 11b and 14b were formed. For ApauAmDH-W141A and MicroAmDH-W141A, no or only low amounts of these alcohols were detected, while the AmDH activities induced conversions above 20%. For cyclic ketones of group D, all the highyl active AmDHs displayed high promiscuous KRED activity, giving particularly nearly total conversion in cyclohexanol 15c with CfusAmDH, MicroAmDH, MsmeAmDH and PortiAmDH in ammonia-free buffer. These results were in line with the high KRED specific activities of 39.4, 510.6 and 235.2 mU mg⁻¹ measured in ammonia-free buffer with 15a for the first three, respectively (Supplementary Table S4).

In the case of prochiral substrates (group C and D except 15a), the enantiomeric excesses (ee) were measured when >1 mM of alcohols (>10% conversion) were formed. As for the AmDH activity, the (S)-configuration was obtained with ee of (S)-alcohols 9c and 11c >99%. We can conclude a high (S)-selectivity for the promiscuous KRED activity of these AmDHs. In the case of the chiral substrate 16a, the cis-configuration was the major but not exclusive form (cis:trans ratio from 60:40 to 82:18) of the alcohol product 16c, with a moderate to high preference for the (S)-configuration at the quaternary carbon created (ee (S,R) = 73–83%; ee (S,S) = 80–97%)) (Figure 3; Supplementary Figure S6).

The Michaelis-Menten parameters were determined for MsmeAmDH for the reduction of cyclohexanone 15a to cyclohexanol 15c to provide a comparison to the values previously reported for its reductive amination into 15b (Table 1; Supplementary Figure S5) (Mayol et al., 2019). Indeed, MsmeAmDH was able to catalyze the conversion of this substrate for both reactions in high yields. In the absence of ammonia and at pH 7.5, the K _m for cyclohexanone 15a was slightly lower than in amination condition (.25 mM and .55 mM respectively). A similar observation was made for the K _m of the cofactor NADPH (.04 and .06, respectively). The k _cat being seven times lower, the resulting catalytic efficiency was approximately two times lower for 15a for the KRED activity over the AmDH one, and nearly identical for NADPH. MsmeAmDH displayed very slow catalysis of oxidation with a specific activity for 15c reaching only .4 mU mg⁻¹, preventing any accurate determination of kinetic parameters. No oxidation activity could even be detected for 7c, whose corresponding ketone 7a nevertheless emerged as a very good substrate in reduction condition.

In an attempt to link the in vitro data to the active-site characteristics of the enzymes, some structural analysis and docking experiments were performed on the closed conformations of structures of MATOUAmDH2-NADP, MsmeAmDH-NADP and CfusAmDH-NADP, and on minimized models of RgnaAmDH-NAD, MicroAmDH-NAD(P), IGCAmDH1-NAD and RgnaAmDH-NAD-NH₄ ⁺ for studies in presence of ammonia/ammonium species (see Material and methods). We confirmed the minimization step by comparing the conformations before and after minimization for the X-raystructure of CfusAmDH with NADP⁺. The greatest difference was seen for the ribose of NADP⁺ with a shift of 1.2 Å, but more importantly, no shift in orientation of all the residues involved in the pocket and the nicotinamide part were observed (Supplementary Figure S6). In addition, the minimized model of MATOUAmDH2 was compared to its X-raystructure (PDB ID: 7ZBO) recently published (Bennett et al., 2022). Some important shifts were observed in the more flexible part such as Y176, residue too far from the reacting part to be discussed in this paper. For the residues surrounding the substrate/product, no drastic shifts were generated, but the notable depicted ones (.7–1.5 Å) have to be considered for discussion results (Supplementary Figure S7). The following molecules were docked: 6a–c, 7a–c, and 15a–c; and the top ten conformations were considered (Table 2; Supplementary Figure S8). We noticed that the docked aldehydes or ketones a should be analyzed more carefully, as the obtained conformations were often not in agreement with the conformations of the corresponding products and mechanism. For example, the docked conformations of 15a displayed the carbonyl function “in the lower front” of the pocket, certainly due to interactions with the NH backbone of Q137 (MsmeAmDH numbering) and amide of the cofactor. In these conformations, the distances between the reactive carbon and the C4 of the nicotinamide are in the 4.6–5.1 Å range, therefore in disagreement with observed reduction (Supplementary Figure S9). We can notice that in the case of MATOUAmDH2, the docked 7a and 15a were flipped, with the carbonyl oriented to the “right” of the pocket, i.e., to His184 and Tyr171. The higher space available in this bigger pocket enables such orientation, in a conformation more similar to the one of the alcohol product, which may induce higher efficiency of MATOUAmDH2 for the reduction compared to other enzymes such as MsmeAmDH (Supplementary Figure S10) (Bennett et al., 2022). Precisely for cyclopentanecarbaldehyde 7a, KRED specific activities were ten-fold higher for MATOUAmDH2 compared to MsmeAmDH (58.4 vs. 5.2 mU mg⁻¹) (Supplementary Table S4).

First, looking at the distance between the carbon bearing either the alcohol or the amine and the C4 atom of the nicotinamide (hydride donor), we noticed that they are quite similar for a couple alcohol/amine and in an appropriate range for hydride delivery in many cases (3.3–4.2 Å) (Agarwal et al., 2000). The main interaction promoting lower energy conformations is indeed the hydrogen bond between labile hydrogen of alcohol or amine with the oxygen atom of the catalytic residue Glu104 (MsmeAmDH numbering), which “sticks” the carbon atom carrying this function at a very similar position, and therefore at a similar distance with the NADPH C4 atom hydride. Despite being in the same range, notable trends can be established from small differences for this key distance, all the more so if we coupled the criterion of calculated energies of the top conformations (Table 2; Supplementary Figure S8). Globally, the alcohol conformations were of higher energies than the amine ones, and the gain of energies between carbonyl substrates and alcohol vs. amine were lower (.05 to −.22 vs. −2.05 to −2.29 kJ mol⁻¹), which could be in agreement with the higher amount of amine formed compared to alcohol. In the case of 7c, lower average distances of 3.90 and 3.91 Å were measured for the favored “up” conformations for MsmeAmDH and RgnaAmDH respectively, enzymes with high promiscuous KRED activities, whereas slightly higher distances (3.8–4.6 Å, average value = 4.11 Å) were measured for IGCAmDH1 displaying very low conversion and specific activity toward 7a (Table 2; Supplementary Table S4). Having in mind that a distance of 4.3 Å is already too distant to postulate a direct hydride transfer, most of the conformations obtained with the latter were not in accordance with a possible reduction. Moreover, for the enzyme IGCAmDH1 with low promiscous KRED activity, energies were all higher than −4.97 kJ mol⁻¹, unlike the others which were below −5.1 kJ mol⁻¹, reaching −5.46 kJ mol⁻¹ for MsmeAmDH with good promiscuous KRED activity. For MATOUAmDH2, and to a lesser extent MicroAmDH, bearing bigger catalytic pockets, these relationships were not found. Such observation could also be made for cyclohexanol (15c) where distances of 3.6–3.8 Å, associated with lower energies (−5.42 to −5.83 kJ mol⁻¹), were calculated for nearly all the top ten conformations for MicroAmDH, MsmeAmDH, CfusAmDH displaying high promiscuous KRED activities, whereas longer distances (3.9–4.6 Å) with higher energy (−5.19 to −5.15 kJ mol⁻¹) were measured for IGCAmDH1, for which only traces of 15c and specific activitiy of .5 mU mg⁻¹ were obtained in in vitro experiments (Figure 4; Supplementary Figure S8. In the case of RgnaAmDH, five of the ten top conformations of 15c displayed distances of 3.4–3.5 Å whereas only 15.5%–17.2% alcohol were formed experimentally and a moderate specific activity of 2.6 mU mg⁻¹ was calculated with 15a. The remaining other five conformations were inappropriately flipped which may reflect an inferior suitability of this active site for 15c, as observed for CfusAmDH for 7c. Interestingly, the average energy for 7c (−4.71, −5.41, and −5.23 kJ mol⁻¹) and 15c (−5.00, −5.82, and −5.53 kJ mol⁻¹) were much lower than the ones for 6c (−3.67, −4.23, and −4.21 kJ mol⁻¹) with, respectively MATOUAmDH2, MsmeAmDH, and CfusAmDH. Combined with the fact that the energy difference between ketone/aldehyde substrates and alcohol products are negative for 7 and 15 (−.07 to −.21 kJ mol⁻¹), while being positive for 6 (.05 kJ mol⁻¹) unlike the furfural 6a/furfurylamine 6b one (−2.05 kJ mol⁻¹), this trend can be linked to the low specific activities with 6a compared to 7a and 15a and the absence of 6c in conversion experiments, compared to notable amounts of 7c and 15c (conversions of 10.6%–23.2% of 7a and 30.5—>99% of 15a into 7c and 15c, respectively).

To give some insights into the differences observed in conversions into alcohol in the presence or absence of ammonia, docking experiments were performed on RgnaAmDH with 6a and 7a, with or without NH₄ ⁺ docked inside the active site, using the model of CfusAmDH-15a-NH₄ ⁺, already published, as a template (Mayol et al., 2019). With docked NH₄ ⁺, the reactive carbon of 7a was too far from the C4 nicotinamide in all the conformations (−4.4–5.1 Å) to allow reduction, whereas these distances (3.7 Å for correct conformations) in absence of NH₃ were in accordance with experimentally observed KRED activity (conversion 32%) and specific activities (14.6 mU mg⁻¹). For 6a, the calculated distances (4.3–4.4 Å) were in each case incompatible with KRED activity both in ammonia-free or ammonia-rich conditions, as observed in the in vitro experiments (Figures 3, 5; Supplementary Figure S4).