Biochemical and structural insight into the chemical resistance and cofactor specificity of the formate dehydrogenase from Starkeya novella

Formate dehydrogenases (Fdhs) mediate the oxidation of formate to carbon dioxide and concomitant reduction of nicotinamide adenine dinucleotide (NAD+). The low cost of the substrate formate and importance of the product NADH as a cellular source of reducing power make this reaction attractive for biotechnological applications. However, the majority of Fdhs are sensitive to inactivation by thiol‐modifying reagents. In this study, we report a chemically resistant Fdh (FdhSNO) from the soil bacterium Starkeya novella strictly specific for NAD+. We present its recombinant overproduction, purification and biochemical characterization. The mechanistic basis of chemical resistance was found to be a valine in position 255 (rather than a cysteine as in other Fdhs) preventing the inactivation by thiol‐modifying compounds. To further improve the usefulness of FdhSNO as for generating reducing power, we rationally engineered the protein to reduce the coenzyme nicotinamide adenine dinucleotide phosphate (NADP+) with better catalytic efficiency than NAD+. The single mutation D221Q enabled the reduction of NADP+ with a catalytic efficiency kCAT/KM of 0.4 s−1·mm−1 at 200 mm formate, while a quadruple mutant (A198G/D221Q/H379K/S380V) resulted in a fivefold increase in catalytic efficiency for NADP+ compared with the single mutant. We determined the cofactor‐bound structure of the quadruple mutant to gain mechanistic evidence behind the improved specificity for NADP+. Our efforts to unravel the key residues for the chemical resistance and cofactor specificity of FdhSNO may lead to wider use of this enzymatic group in a more sustainable (bio)manufacture of value‐added chemicals, as for instance the biosynthesis of chiral compounds.

Formate dehydrogenases (Fdhs) mediate the oxidation of formate to carbon dioxide and concomitant reduction of nicotinamide adenine dinucleotide (NAD + ). The low cost of the substrate formate and importance of the product NADH as a cellular source of reducing power make this reaction attractive for biotechnological applications. However, the majority of Fdhs are sensitive to inactivation by thiol-modifying reagents. In this study, we report a chemically resistant Fdh (Fdh SNO ) from the soil bacterium Starkeya novella strictly specific for NAD + . We present its recombinant overproduction, purification and biochemical characterization. The mechanistic basis of chemical resistance was found to be a valine in position 255 (rather than a cysteine as in other Fdhs) preventing the inactivation by thiol-modifying compounds. To further improve the usefulness of Fdh SNO as for generating reducing power, we rationally engineered the protein to reduce the coenzyme nicotinamide adenine dinucleotide phosphate (NADP + ) with better catalytic efficiency than NAD + . The single mutation D221Q enabled the reduction of NADP + with a catalytic efficiency k CAT /K M of 0.4 s À1 ÁmM À1 at 200 mM formate, while a quadruple mutant (A198G/D221Q/H379K/S380V) resulted in a fivefold increase in catalytic efficiency for NADP + compared with the single mutant. We determined the cofactor-bound structure of the quadruple mutant to gain mechanistic evidence behind the improved specificity for NADP + . Our efforts to unravel the key residues for the chemical resistance and cofactor specificity of Fdh SNO may lead to wider use of this enzymatic group in a more sustainable (bio)manufacture of value-added chemicals, as for instance the biosynthesis of chiral compounds.

Introduction
Formate utilization is increasingly being explored as a sustainable alternative in the production of electricity and platform chemicals [1][2][3][4]. Low cost, high solubility, permeability to biomembranes and low reduction potential are among the main characteristics that make this C 1 compound one of the most attractive feedstocks for the development of 'formatotrophic' cell factories [5,6]. Micro-organisms that use formate as source of both carbon and reducing power must be equipped not only with a metabolism machinery for its assimilation (i.e. serine pathway or reductive acetyl-CoA pathway), but also with specific enzymes to exploit its reducing power, producing CO 2 as a waste product. The released carbon dioxide could then be electrochemically reduced back to formate, allowing to feed the biomanufacturing microbes in a continuous and circular manner [7].
NAD + -dependent formate dehydrogenases (Fdhs) (EC 1.17.1.9) [8] couple the oxidation of formate with the reduction of NAD + , producing CO 2 and NADH, which can be used for reductive metabolism. Fdhs from bacteria and methylotrophic yeasts have found applications for the regeneration of NADH in vivo [9][10][11], as well as in vitro [12][13][14]. Studies on the Fdh from Pseudomonas sp. 101 (Fdh PSE ) have laid the foundation for biochemical and mechanistic understanding of the enzyme family [15] and have been followed up by protein engineering efforts [12,[16][17][18][19][20] to make Fdhs more effective biocatalysts for the largescale production of optically active compounds [21][22][23]. However, Fdhs are well known for their sensitivity to chemical modification [15], which limits their potential use as regenerators of reduced redox cofactors in the field of biocatalysis. For instance, the a-haloketone substrates used in the reductive biosynthesis of ahalohydrins, which are important building blocks for chiral compounds [24][25][26], inhibit Fdhs. In 2003, Hasegawa and collaborators reported for the discovery of a native Fdh resistant to a-haloketones and thiolmodifying reagents, purified from Thiobacillus species, strain KNK65MA [27]. The chemical resistance of Fdh from T. sp KNK65MA (Fdh TBA ) was hypothesized to be caused by the presence of only five cysteines in the amino acid sequence instead of seven, as found in most of the bacterial Fdhs characterized so far (Table 1) [28]. The residues replacing the two 'missing cysteines' are a valine in position 255 and an alanine in position 288. The presence of a valine in position 255 indeed plays an important role in conferring high resistance to thiol reagents, as demonstrated by replacing Cys255 in Fdh PSE [29]. The involvement of the residue in position 288 is less clear. Notably, the eukaryotic Fdhs from Candida boidinii (Fdh CBO ) and Candida methylica are also sensitive to thiol-modifying compounds as p-chloromercuribenzoate (PCMB) [30] and 5,5 0 -dithiobis-(2-nitrobenzoate) DTNB [31], but they lack a cysteine in the active site at the position corresponding to Cys255 in bacterial Fdhs. These yeast Fdhs have only two cysteines, which correspond to positions 52 and 288 in the bacterial enzymes.
In order to use Fdhs as effective NAD(P)H regenerators in biomanufacturing, it is necessary to unambiguously elucidate which cysteines are sensitive to chemical modifications, leading to protein inactivation. Here, we present the biochemical characterization of a novel NAD + -dependent formate dehydrogenase (Fdh SNO ) from the soil bacterium Starkeya novella recombinantly overproduced in Escherichia coli cells. Like its close homologue Fdh TBA , Fdh SNO exhibits a native resistance to thiol-modifying reagents. The protein has reduced sensitivity to inactivation to the ahaloketone ethyl 4-chloroacetoacetate (ECAA) (halftime of more than 4 h). We restored the missing Table 1. Overview of the cysteine positions in the most studied Fdh-proteins. In the case of the yeast sequences from C. boidinii and C. methylica, the original position of the amino acids are reported in brackets. Positions 49 and 86 are also shown since they represent the main difference between the enzymes from Thiobacillus sp. KNK65MA and Starkeya novella. Total  cysteines  Ref  5  49  52  86  145  182  248  255 288 354 cysteines in Fdh SNO to establish the amino acids responsible for its chemical resistance. Crystallographic investigation of Fdh SNO provides an insight into the functional characteristics of the chemical resistance, in comparison with the structures of known homologues.

Amino acid position
To make the enzyme also suitable for NADP +dependent reactions, we investigated the effects of a single (D221Q) and quadruple mutant (A198G/ D221Q/H379K/3380V), with the latter resulting in a better catalytic efficiency towards NADP + than NAD + as an electron acceptor. The cofactor-bound structure of the quadruple mutant Fdh SNO offers a mechanistic explanation for the change in cofactor specificity.

Gene overexpression and Fdh SNO purification
To obtain an enzyme with improved chemical stability, we chose the gene  (Table S1). The gene Snov_3272 was initially cloned into in an expression vector suitable for the recombinant overexpression in E. coli, but the high GC content [34] of the native gene from S. novella turned out to be incompatible with the codon usage of E. coli (Table S2). Thus, we switched to a codon-optimized sequence of the gene (synFdh) for the expression in E. coli (a detailed analysis of the codon-optimization is available in Table S3 and Fig. S1). Fdh SNO was then overproduced and isolated from the soluble fraction of E. coli MC 1061 [35] cells (as described in the Materials and Methods section), as evidenced by a band around 46 kDa on the Sodium-Dodecyl-Sulfate polyacrylamide gel (SDS/PAGE in Fig. 1A) in agreement with the calculated monomeric molar mass based on sequence. The protein was purified by a combination of immobilized metal affinity (IMAC) and size exclusion (SEC) chromatographies (Fig. 1B), with a reproducible yield around 90 mg of protein per litre of culture. The elution volume of the protein from the SEC corresponds to a mass of roughly 90 kDa, consistent with a dimeric state typical of bacterial Fdhs [15].

Biochemical characterization
We characterized the purified Fdh SNO in terms of substrate specificity, inhibitor sensitivity, pH and temperature dependence, and stability. Fdh SNO has one of the highest affinity for formate among known bacterial Fdhs [28], with a K M value of 2.2 AE 0.4 mM (at an NAD + concentration of 2.0 mM), and a k CAT value of 0.9 s À1 . Fdh SNO has a strong preference for NAD + over NADP + . At a fixed concentration of 20.0 mM formate, the K M for NAD + was 115 AE 11 lM, and k CAT 1.1 s À1 . We could not obtain kinetic values for NADP + as an electron acceptor, although a marginal NADP + reduction was observed using a large excess of the compound (Fig. S2).
Since the inhibitory capacity of small linear anions on NAD + -dependent Fdhs has been documented in the past [15], we verified their inhibitory potential and quantified the inhibitory constants (IC 50 ) of azide and thiocyanate (Fig. S3). We determined IC 50 values of 100 nM for azide and 48 lM for thiocyanate. The kinetic parameters of Fdh SNO are summarized in Table 2. Next, we evaluated the effect of the pH on both enzymatic activity and protein stability. Fdh SNO showed the highest activity around pH 7.5, but retained more than 70% of the maximal activity at a wide pH range (between 4.5 and 10.2 ( Fig. 2A)). We tested the pH stability by incubating the protein for 20 h in different buffers (Fig. 2B), and found that potassium phosphate buffer at pH 7.5 was the best to preserve the protein activity. At pH values between 4.5 and 6.0, the enzymatic activity decreased over the 20 h incubation, but still between 50% and 75% of activity was retained. The incubation in carbonate (pH 9.2-10.2) and acetate at pH 4.0 led to the complete inactivation of Fdh SNO. Table 2. Kinetic parameters of Fdh SNO . In the case of formate and NAD + , the kinetic parameters come from biological quadruplicates (n = 4, the errors correspond to the standard deviation). For the inhibitory experiments, IC 50 values are the average of biological triplicates (n = 3, the errors indicate s.e.m.). All the assays were carried out in 50 mM KPi at pH 7.5. N.D. stands for not determined in experimental conditions. The melting temperature (T m ) of Fdh SNO in the absence of any nicotinamide cofactor was estimated by thermal shift assay at 52.5°C (Fig. 2C), which agrees well with the T m measured for Fdh TBA [27]. As expected, the addition of NADP + barely affected the protein T m (T m NADP+ = 53.5°C), while the presence of NAD + led to an increase in T m of nine degrees (T m NAD+ = 61.5°C), confirming binding and thus the strict specificity of Fdh SNO towards NAD + .
Storage at 4°C preserved activity over long timescales, with 90% activity retained after 35 days of storage in the fridge (Fig. 2D, black line). At 30°C, Fdh SNO remained close to be fully active over the first week of storage (blue line). Freeze-thaw cycles (Fig. 2E) had a larger inactivating effect on Fdh SNO , with a linear loss of activity as a function of the number of freeze-thaw cycles, consistent with cold denaturation of the purified protein progressively taking place, as reported for numerous other enzymes [36].

Restoration of sensitivity to thiol-modifying reagents by mutagenesis
We tested the effect of thiol reagents on activity (Fig. 3A,B) by incubating the protein for 2 min prior to the start of the reaction by the addition of formate. We chose a set of chemicals with different charges, sizes and binding mechanisms: n-ethylmaleimide (NEM), PCMB, 2-(Trimethylammonium)ethyl methanethiosulfonate (MTSET) and 5,5 0 -dithiobis-(2nitrobenzoate) (DTNB). We also tested the effect of the reducing agents dithiothreitol (DTT) and 2mercaptoethanol (BME), which we used to ensure that Fdh SNO did not undergo detrimental oxidation events during the purification and storage procedures. Both DTT and BME slightly improved the protein activity, respectively, to 107% and 114%. As predicted based on the sequence, Fdh SNO was resistant to inactivation by the thiol-modifying reagents, retaining most of its activity in the presence of NEM (87%), MTSET (81%) and DTNB (79%). PCMB was the only exception, as it caused a decrease of the activity to 27%. Likely, interactions take place with side chains of other amino acids than cysteine (imidazole and carboxylic groups), which are prone to form complexes with organomercurial compounds [37,38], affecting the protein functionality.
In order to get more detailed insight into the amino acids responsible for the chemical resistance, we compared the cysteines present in Fdh SNO with those in the homologous bacterial and eukaryotic Fdhs reported as sensitive to NEM, DTNB and PCMB (Table 1). Besides the valine in position 255 and the alanine in position 288 that distinguish the chemically resistant Fdh SNO and Fdh TBA , we also tested position 52 which is occupied by serine in Fdh SNO , whereas thiol-sensitive Fdhs in yeast species have a cysteine (C23 in the native sequence). In these eukaryotic Fdhs, only two cysteines are part of the monomeric sequence, and intriguingly the other cysteine is not located at the position equivalent to 255 in the bacterial enzymes, but instead is found at position 288 (natively C262).
We purified to homogeneity single mutants of Fdh SNO introducing cysteines at positions 52, 255 and 288 in order to 'restore' susceptibility to thiol-reagents (Fig. S4). For each mutant, we evaluated the effect of reducing agents (DTT and BME), azide, thiocyanate and thiol-modifying reagents (Fig. 3B). The activity of the three mutants did not show significant changes from that of the wild-type (WT) in the presence of DTT and BME. Also the inhibition mediated by azide and thiocyanate remained unaltered in the mutants. In contrast, the cysteine restoration in position 255 caused susceptibility to thiol reagents, leading to complete protein inactivation upon incubation with 2.0 mM MTSET and 0.2 mM PCMB, and almost complete inhibition with 2.0 mM NEM or 0.2 mM DTNB (residual activity of 3%). Instead, S52C and A288C retained the resistance to all tested thiol-modifying reagents, showing no difference with the wild-type Fdh SNO . Since only V255C displayed the loss in chemical resistance, we proceeded to quantify its chemical stability by longer incubations with the thiol compounds (Table 3, time courses in Fig. 3C), and compared it with the WT Fdh SNO . As expected, the half-life values (t ½ ) of V255C for all the tested thiol-reagents were lower than 1 min (circles in Fig. 3C). In contrast, the chemical stability of the WT protein (squares in Fig. 3C) proved to be higher than V255C, albeit to a variable extent that depended on the specific compound.
Due to the potential use of Fdhs in the synthesis of valuable organic building-blocks, we also assessed the activity of Fdh SNO and its cysteine mutants in the presence of ECAA and ethyl (S)-(À)-4-chloro-3hydroxybutyrate (ECHB), as shown in Fig. 3D,E. Both the a-haloketone and a-haloalcohol are precursors to statins, a broad class of HMG-CoA (3hydroxy-3methylglutaryl-CoA) reductase inhibitors for lowering blood cholesterol [39,40]. As with the thiol reagents, the activity of WT Fdh SNO was not significantly affected by either 20 mM ECAA or ECHB. This chemical resistance was similarly conserved also in the S52C and A288C mutants, but the restoration of the cysteine in position 255 led to a drop in activity in the presence of ECAA, with only 20% residual activity compared with the wild-type upon 2 min incubation. The V255C sensitivity to ECAA was also evident from the estimation of the half-life value (~1 min). By contrast, the chemical stability of WT Fdh SNO against the a-haloketone was 300 times higher than V255C, with a t ½ of almost 5 h ( Table 3). The wild-type enzyme was even more resistant to inactivation by ECHB, for which we could not calculate a t ½ value even after 24-h incubation (Fig. S5). Replacing Val255 with a cysteine made instead the enzyme susceptible to the a-haloalcohol (t ½~4 5 min).

Structural insight into the chemical resistant Fdh SNO
To gain further insights into the functional features of Fdh SNO , we solved the crystal structure of the apoenzyme to a resolution of 2.1 A (PDB entry 7QZ1, more details in the Materials and Methods section). Crystallographic data and refinement statistics are summarized in Table S4.
Like its structurally characterized homologues (summarized in Table S5), Fdh SNO forms a homodimeric assembly of two identical monomers (Fig. 4A). Each monomer consists of two domains that adopt the Rossmann fold, a structural motif typical of nucleotide binding proteins [41]. Upon structural superimposition of the structures of Fdh SNO and Fdh PSE in complex with NAD + (PDB entry 2NAD), it was possible to identify the binding site of NAD + . The main amino acids involved in the binding of the cofactor are Ala198, Asp221, Val255 and Ser380, as highlighted in Fig. 4B.
By comparison with the structure of homologues sensitive to thiol reagents, FDH SNO shows no major reorganization in the secondary or tertiary structure, as shown in the right panel of Fig. 4A. In all of the positions under examination (residues 52, 255 and 288), the orientation of the side chains remained unchanged, regardless of the presence of the thiol of the cysteine or of the specific substituent group (hydroxyl in Ser52, isopropyl in Val255, methyl in Ala288). Moreover, the mutations do not affect the accessible surface area (ASA) of each of these residues (Table S6). Val255 was more exposed to the solvent (highest ASA) than Ala288 and Ser52, which have little-to-no ASA for solvent. Val255 was the only residue to be located in the coenzyme binding pocket. Its proximity with the NAD + binding site explains why the residue in this position plays a critical role in allowing or preventing the cofactor binding when a thiol-modifying reagent is present.

Cofactor specificity can be changed by mutations
Because numerous enzymes with industrial biomanufacturing potential (e.g. cytochrome P450 and Table 3. Half-time values (t ½ ) of the Fdh SNO activity for the thiolmodifying reagents employed in this study. The data come from biological triplicates (n = 3, the errors are reported as standard deviation). N.Q. means "not quantifiable", as the protein inactivation was complete in less than a minute. N.D. stands for "not determined" after 24 h incubation.

Compound
Concentration (mM) cyclohexanone monooxygenases) require NADPH rather than NADH as a cofactor [43,44], we decided to explore the possibility to change the co-substrate specificity of Fdh SNO . An NADP + -variant of this chemically resistant Fdh may be coupled with NADPH-dependent biocatalytic processes to provide regeneration of the expensive NADPH by oxidation of formate. We selected the point mutation D221Q and the quadruple mutation A198G/D221Q/H379K/S380V (GQKV-Fdh SNO from now on). The neutralization of the negative charge of Asp221 has long been known to change the coenzyme specificity [45,46]. The other three mutations were chosen by combining the main interactions observed with the cofactor from our superimposed crystallographic structure (Fig. 4B) with the observations from other studies on NADP +dependent Fdhs [20,47,48]. We determined the kinetic parameters of the purified D221Q and GQKV mutants (SEC profiles in Fig. S6), for NAD + , NADP + and formate in the presence of both the cofactors (Table 4, see Fig. S7 for the graphs).
D221Q Fdh SNO acquired the ability to reduce NADP + in addition to NAD + , with an 11-fold lower K M for the phosphorylated cofactor (K M of 0.3 mM for NADP + , 3.3 mM for NAD + ). At the same time, the affinity of the mutant for formate in the presence of NAD + was more than doubled (K M of 5.5 mM) compared with the WT enzyme. Therefore, we conclude that the substitution of aspartate 221 with glutamine leads to broadened specificity with both NAD + and NADP + tolerated as electron acceptors, at the expense of the affinity for NAD + .
The quadruple mutant GQKV showed a twofold increase in k CAT compared to D221Q when NADP + and formate are cosubstrates, and even a better affinity for NADP + resulting in a K M value of 0.1 mM that practically matches the one for NAD + in WT Fdh SNO . The three additional mutations did not alter the K M for formate in the presence of NADP + (15.6 mM for GQKV, 16.7 mM for D221Q), but NAD + became a poorer substrate for the GQKV variant, as not only the K M for NAD + was doubled compared with D221Q, but also the affinity for formate (K M of 22.1 mM) increased fourfold with NAD + as a cosubstrate. Finally, the improvement of the quadruple mutant over the single one for the change of specificity appears evident in terms of catalytic efficiency: the k CAT /K M for NADP + was five times higher than D221Q.
A structural insight into the NADP + dependency of GQKV-Fdh SNO To further understand the effect of the chosen mutations on cofactor specificity, we determined the crystallographic structure of GQKV-Fdh SNO in complex with NADP + and azide at a resolution of 2.5 A (Fig. 5A-E and Table S4, PDB entry 8OQ2). To assess any structural differences, the quadruple mutant was compared against the cofactor-bound structure of the closest homologue specific for NAD + , namely FDH PSE (PDB entry 2NAD).
The replacement of Ala198 with a glycine (Fig. 5B) enlarged the pocket accommodating the nicotinamide adenosine ring, resulting in the loss of a direct interaction between the cofactor and the amino acid side chain (4.14 A distance between the oxygen atom of the ribose ring and the a-carbon atom of Gly198 in the mutant Fdh, vs 2. 74 A between the main oxygen of the ribose ring and the side chain carbon of Ala198 in FDH PSE ). Table 4. Kinetic parameters of the NADP + -dependent Fdh SNO constructs. The values were obtained from independent triplicates (n = 3), the errors represent the s.e.m. N.D. not determined. Asp221 plays a key role in the NAD + dependency of Fdhs by establishing hydrogen bonds with the hydroxyl groups in positions 2 0 -and 3 0 -of the ribose moiety, and electrostatically repelling the negative phosphate substituent of the NADP + cofactor. Neutralizing aspartate by mutation to glutamine (Fig. 5C) allowed NADP + recognition as a preferred cosubstrate. Besides, the insertion of Gln221 led to the distinct reorientation of the side chain of the adjacent arginine in position 222, enabling not only the formation of a salt-bridge interaction with the 2 0 -phosphate group of NADP + , but also a further hydrogen bond with the N1 atom of the adenosyl ring.
The introduction of Lys379 did not seem to establish a direct interaction with the cofactor (Fig. 5D); however, a closer proximity with one of the terminal amines of Arg222 (2.56 A) was observed in contrast with His379 in the NAD + -bound structure (2.93 A): this might suggest a stabilizing function in the favour of the specific reduction of NADP + .
Following the point mutation S380V, a reorientation of the side chain was observed in comparison with the Fdh PSE structure. Within the homologue structure, the hydroxyl group of Ser380 points directly towards the unoccupied binding site and is separated by 2.72 A from the closest phosphate group of the nicotinamide pyrophosphate. In contrast, the mutation to Val380 leads to a near 180°flip in the side chain position while the orientation of the amino acid backbone remains largely unchanged. The backbone hydroxyl group is now the closest moiety to the same phosphate, whereas the Val380 side chain is 7.87 A from the phosphate, enabling an opening of over 5 A at this location of the binding site. The mutation of S380V does not represent a significant steric alteration to promote or inhibit the binding of substrate; however, it is likely that the extremely polar NADP + substrate would have a repulsive influence on the hydrophobic valine side chain, yielding this reorientation of side chain positioning.

Discussion
Starkeya novella is known for its extensive metabolic flexibility, since it is a mesophilic organism able to grow not only on single carbon compounds (such as methanol and formate) and sulfur derivatives (e.g. thiosulfate) [49], but also in soils contaminated with toxic organophosphate insecticides [50] and organic fuel additives [51]. Bacteria that survive a variety of environmental stresses from extreme natural conditions to artificial pollutants, could harbour valuable biocatalysts for the manufacture of platform chemicals [52,53]. Thus, the rich diversity in soil prokaryotic communities [54,55] represents a promising source to explore for a new arsenal of enzymes for (bio)chemical syntheses [56,57].
In this study, we report the isolation and characterization of a novel chemically resistant Fdh encoded by the soil bacterium S. novella. Fdh SNO displayed similar kinetic parameters to the previously characterized bacterial Fdhs [28], with a K M for formate in the low millimolar range and a strict dependence on NAD + , for which the affinity lies in the micromolar range. In particular, the K M value for formate of 2.2 mM is one of the most favourable affinity constants among Fdhs, around 3-7 times lower than the characterized homologues from Pseudomonas sp.101, Mycobacterium vaccae N10 and C. boidinii. On the contrary, the k CAT around 1/s is lower than to the values in Fdh PSE and Fdh CBO , respectively, with a k CAT of 7/s and 4/s. The biochemical characteristics of Fdh SNO match well with those of the other characterized Fdhs [28]: high activity in a wide pH range (4.5-10.2), an optimum temperature around 50°C and good storage capacity at 4°C. Fdh SNO gradually loses activity when subjected to multiple rounds of freezing and thawing, a procedure often used to deliver enzymes into the lumen of liposomes.
Azide was found to be a powerful Fdh SNO inhibitor, with an estimated IC 50 of 10 À7 M, four orders of magnitude lower than the K M for formate (10 À3 M). This means that traces of azide (commonly used in the range of 0.02-0.1%, corresponding to 5 9 10 À3 M À 2.5 9 10 À2 M, as an antimicrobial agent [58]) in reagents and equipment that will come into contact with Fdh need to be carefully removed to prevent protein inhibition. Crystal structures of bacterial Fdhs (including the quadruple mutant of Fdh SNO reported in this work) co-crystallized with NAD(P) + and azide (Fig. S8) reveal that the azide molecule forms a dense network of hydrogen bonds with Ile122, Asn146, Arg284 and His332 in the active site. These interactions lead to the formation of a protein conformation analogous to the one adopted in the transition state during formate conversion [59]. Although thiocyanate shares with azide the characteristics of a three-atom linear anion, it is 500 times less effective as an inhibitor. In the absence of a structure of the ternary complex with the cofactor and thiocyanate, we speculate that thiocyanate is less suitable to form hydrogen bonds with the aforementioned residues, probably because hydrogen bond formation with the molecule is more sensitive to the relative orientation of the bond donor, and due to the lower electronegativity of the sulfur atom compared to the nitrogen atoms of azide. Hence, the restricted ability of the sulfur atom to coordinate hydrogen bonds with donors leads to a weaker inhibition than the triatomic nitrogen compound [60].
Since we observed that Fdh SNO is chemically resistant to various thiol-modifying compounds, we designed 'cysteine-restoring' mutants in those positions suspected to lead to sensitivity [27]. We tested the mutations S52C and A288C corresponding to the only cysteines present in homologous Fdhs in Candida species, which were reported as sensitive to compounds with high affinity for thiol groups [30,31]. In addition, Slusarczyk and coworkers highlighted how replacing these cysteines in Ser52 (C52S) and Ala288 (C288A) resulted in a four-to fivefold increase in resistance to Cu 2+ ions [17]. We also introduced a cysteine at position 255, of which the oxidation or the chemical modification is the main cause of chemical inactivation of the enzyme [15]. Our work pinpoints that the presence of a valine at position 255 is responsible for the chemical resistance of Fdh SNO , with no measurable roles of Ser52 and Ala288. This experimental conclusion is reinforced by the crystal structure. The residue at position 255 is located in the cofactor-binding pocket in the proximity of the adenine part without directly participating in the catalytic action [42]. Upon mutation of Val255 into a cysteine, the thiol group of Cys255 is accessible to thiol-modifying reagents. Steric hindrance resulting from the cysteine modification prevents the binding of the cofactor, culminating in protein inhibition. Therefore, the absence of the thiol group in Fdh SNO represents the key feature underlying the chemical resistance of this enzyme. Valine and cysteine display similar ASAs to the solvent in position 255 (Table S6); consequently, the different chemistry between the thiol and the isopropyl groups seems to be the only relevant determinant. The amino acids at positions 52 and 288 are both too distant from the active site and not accessible enough to the solvent to play an effective role in the chemical resistance mechanism.
The increased resistance of Fdh SNO against the alpha-halo compounds ECAA and ECHB, and the thiol-modifying reagents NEM, DTNB, MTSET and PCMB makes the enzyme a more versatile long-lasting supplier of reducing power for industrial biomanufacturing. It may be possible to boost the stability of Fdh SNO further. Strategies for increasing the operational stability of Fdhs have been reviewed elsewhere [8,28].
With improved chemical resistance, Fdh SNO also is an ideal template to change the specificity for NAD + into NADP + in order to make a flexible toolset of enzymes for regeneration of the reduced cofactor of choice. The single mutation of aspartate at position 221 into glutamine was found to allow the reduction of NADP + , while maintaining the ability to convert NAD + into NADH, albeit with a 30 times lower affinity than the WT enzyme. In order to obtain a NADP + -dependent version of Fdh SNO with better catalytic efficiency, we selected the mutations of the amino acids in positions 379 and 380, similarly to the recent work described by Calzadiaz-Ramirez et al. [20]; the role of Gly198 in altering the K M values for the cofactors was already known, as it minimizes the steric hindrance to accommodate the phosphate group in position 2 0 of the ribose ring belonging to NADP + [18,20]. Together with A198G and D221Q, the replacements H379K and S380V have been identified as 'epistatic mutations' that act synergistically to improve both the affinity and the turnover number for NADP + . In particular, His379 is known to stabilize the stacking of the adenosine moiety of NAD + , while Ser380 binds the pyrophosphate group in the proximity of the adenine ribose [42]. The occurrence of a lysine instead of histidine was previously observed in the Fdh-homologue from Granulicella mallensis MP5ACTX8, which displayed double cofactor specificity with a preference for NADP + [48]. Its structure in complex with the cofactor revealed how the presence of Lys379 disrupts any possible interaction with the adenosine moiety, as the side chain of the amino acid is directed towards the solvent rather than the substrate: our structure confirmed this observation and further suggested a possible stronger interaction with Arg222 (Fig. 5D). On the other hand, the substitution S380V introduces a side chain that cannot interact anymore with the nicotinamide pyrosphosphate, but that allows the steric accommodation of the phosphate group indicative of NADP + . The result of these synergistic mutations can explain the higher catalytic efficiency for NADP + of our GQKV-Fdh SNO compared with the single mutant D221Q Fdh SNO , as well as the worsening of the K M value for both the substrates when NAD + is the cosubstrate.
In conclusion, unlike multisubunit Fdh complexes containing iron-sulfur clusters and various metal ions [61,62], NAD + -dependent Fdhs are versatile enzymes that do not require a prosthetic group to produce NADH or NADPH from formic acid, releasing CO 2 as a final product. With a minimal mutagenic approach of a chemically resistant Fdh specific for NAD + (Fdh SNO ), we highlighted the residues critical to modulate the specificity for NAD + and NADP + . Wild-type or mutant enzymes can be chosen to regenerate NAD(P)H, depending on the requirement of the specific cofactor or of the intermediates formed in the biosynthesis process of interest, taking advantage of the oxidation of the low-cost chemical formate.
Particularly, Fdh SNO has recently been successfully used as a pivotal component of a minimal enzymatic pathway triggered by formate ensuring the redox regeneration in the field of bottom-up synthetic biology [14,63]. Furthermore, the manufacture of valuable enantiomers for different industries [25] could represent an application for this biocatalyst, both in in vivo and in vitro systems. The compatibility of chemically resistant Fdhs in productive processes has already been shown with the precursors of statins [12], encouraging the use of such enzymes for pathways whose synthetic steps have the same chemistry, such as adrenergic receptor agonists or selective serotonin reuptake inhibitors [64][65][66]. Likewise, Fdh SNO could prove to be a suitable tool for enzymatic or microbial bioremediation of sites contaminated by chemical and metal pollutants [67][68][69], avoiding the exhaustion of the reduced cofactor pool that supports the biodegradation process. Alternatively, the biochemically and structurally explored mutations on Fdh SNO could be applied to high turnover number Fdh-homologues to combine catalytic efficiency, chemical resistance and desired cofactor specificity.

Bioinformatic analysis
The screening of the homologues of Fdh TBA (UniProt accession number: Q76EB7) was carried out with the HMMER web service (https://www.ebi.ac.uk/Tools/hmmer/). Once Fdh SNO was identified as the target protein, we obtained the encoding sequence by inquiring the Genome Database Collection BioCyc (https://biocyc.org/).
The numbers referring to the codon bias of the microorganisms Starkeya novella DSM 506 and E. coli K-12 (Table S2) are publicly available from the online resource (https://www.kazusa.or.jp/codon/) of the Kazusa DNA Research Institute [70].
Accessible surface area of amino acids (Table S6)

Expression plasmid construction
The native gene of the Fdhs Snov_3272 (https://bacteria. ensembl.org/Starkeya_novella_dsm_506_gca_000092925/Gene/ Summary?g=Snov_3272;r=Chromosome:3436073-3437278; t=ADH90546) was PCR-amplified using the primers snofdh001 and snofdh002 (all the primers used in this study are listed in Table S7), while the template was constituted by the genomic-DNA of Starkeya novella purchased from the DSMZ collection (DSM 506). The codon-optimized version of the same gene (synfdh) for the expression in E. coli was designed and then acquired on the 'Gene Art gene synthesis' portal (Thermo Fisher Scientific, Inc., Waltham, MA, USA), which provided the synthetic gene into the pMA-RQ cloning vector. This plasmid was used as template for the PCR-amplification by the primers snofdh003 and snofdh004. Both the PCR-products Snov_3272 and synfdh were consequently cloned first in the pINIT vector and then in the expression plasmid pBXC3H of the FX cloning kit [71], as previously described [14]. For the purpose of selection of the correct plasmid sequences, E. coli MC1061 cells (ATCCÒ 53338) [35] were used. The resulting constructs were verified by DNA-sequencing.

Site-directed mutagenesis
The introduction of the single mutations S52C, V255C, A288C and D221Q was made according to the protocol of the QuikChange Site-Directed Mutagenesis Kit (Agilent Technologies, Inc., Santa Clara, CA, USA). The primers employed for the construction of each mutant are reported in Table S7. As pBXC3H_synfdh was the selected vector for the gene overexpression, we adopted it as DNA-template for the mutagenic PCR. Following the amplification in 50 lL as the final volume, each reaction mixture was treated with the restriction enzyme DpnI (10 UÁlL À1 ) to digest the parental non-mutated DNA for 1 h at 37°C. 5 lL of the digestion mixture was used to transform MC1061 chemocompetent cells by heat shock. The successful outcome of the sitedirected mutagenesis was assessed by sequencing the isolated plasmids. The construction of the quadruple mutant GQKV was carried out similarly to the single mutants, but starting from pBXC3H_synfdhD221Q as template for a double round of site-directed mutagenesis. We first inserted the single mutation A198G using the primers snofdh011 and snofdh012, obtaining the vector pBXC3H_synfd-hA198GD221Q. After verifying the correct insertion of the second mutation, we repeated the PCR-mutagenesis procedure using the double mutant vector (pBXC3H_synfd-hA198GD221Q) as a template, amplifying the DNA with the primers snofdh015 and snofdh016. Also in this case, the resulting vector pBXC3H_synfdhA198GD221QH379KS380V was confirmed by sequencing.

Protein overproduction and purification
Fdh from Starkeya novella was overproduced in MC1061 E. coli cells as heterologous host, similarly to what was previously described [14]. Briefly, the pBXC3H_synfdh strain was grown at 37°C, 200 rpm, in LB medium + 100 lgÁmL À1 ampicillin, up to the optical density ffi 0.6, to be then induced with final 0.01% v/v L-arabinose. Following the induction, the culture was moved at 20°C and left overnight shaking at 200 rpm. Around 20 h after the arabinose addition, the cells were spun down (15 min, 6000 g, 4°C), and washed with buffer A (50 mM KPi pH 7.5, 150 mM NaCl), centrifuging with the same parameters. The resulting pellet was resuspended in an approximate volume of 30 mL buffer A, supplementing it with 100 lgÁmL À1 DNase, 1 mM MgSO 4 , 1 mM phenylmethylsulfonyl fluoride, to be consequently sonicated for 10 min (3 s on, 6 s off cycle, 70% amplitude) using a VCX130 Vibra-Cell sonicator (Sonics & Materials, Inc., Newton, CT), keeping the sample chilled over the cell disruption. After adding 1 mM -K-EDTA pH 7.0, the insoluble fraction was removed by further centrifugation at 48 254 g, 4°C for 30 min, while the supernatant was flash-frozen with liquid nitrogen to be stored at À80°C.
At the moment of the protein purification, the cytosolic fraction was quickly thawed and incubated in a Econo-Pac column (Bio-Rad Laboratories, Inc., Hercules, CA, USA) for 30 min at 4°C with 1 mL of Ni 2+ -Sepharose resin (Ni SepharoseÒ 6 Fast Flow Cytiva) and 10 mM imidazole pH 7.5, on continuous shaking. The resin material was prewashed with 20 column volumes (CV) MQ and 20 CV buffer A. After the incubation, the soluble fraction was left flowing through by gravity, leaving the His-tagged enzyme trapped in the nickel resin. Then, the resin material was washed with 20 CV buffer A + 50 mM imidazole pH 7.5, and eventually the bound protein was eluted in several fractions of buffer A + 500 mM imidazole pH 7.5 (the first fraction corresponded to 70% of the packed CV, the sequential ones to 50% CV). The fractions containing the protein were pooled, centrifuged for 5 min at 21 230 g, 4°C with EDTA 1.0 mM, pH 7.0, and loaded on a Superdex 200 Increase 10/300 GL column (GE Healthcare, Chicago, Illinois, USA), attached to a NGC Quest TM 10 Chromatography System (Bio-Rad Laboratories, Inc.) with an incorporated UV detector. The protein was eluted with buffer A and its concentration was assessed afterwards spectrophotometrically (NanoDrop Technologies, Inc., Wilmington, DE, USA) using the extinction coefficient (e 280 nm = 62 340 M À1 Ácm À1 ) calculated with the online tool Protparam (https://web.expasy.org/protparam/). Last, Fdh SNO was concentrated on a 10 kDa centrifugal filter device AmiconÒ Ultra 0.5 mL (Merck Millipore Ltd., Darmstadt, Germany) to 2-4 mgÁmL À1 , supplemented with 10% glycerol (v/v) and stored at À80°C, after flashfreezing aliquots of 60-80 lL. As further controls of the different protein isolation steps, we harvested several fractions over the purification process for the SDS/PAGE (Fig. 1A) and eventually loaded them on SDS blots made of a 15% separating gel and a 5% stacking gel.
The same protocol for the protein overproduction and the purification was used for the S52C, V255C, A288C, D221Q and GQKV mutants.

Activity and kinetic assays
The activity of Fdh SNO and its mutants was monitored following the reduction of NAD(P) + at 340 nm (e NAD(P)H , 340 nm = 6.22 mM À1 Ácm À1 ) in a SPARK 10 M plate reader (Tecan) at 30°C. The standard assay conditions include in the reaction mixture 50 mM KPi pH 7.5, 0.2 lM Fdh SNO , 0.5 mM NAD + and 20 mM formate. All the biochemical assays shown in Fig. 2A-E were performed on biological triplicates (n = 3, the error bars represent the s.e.m.) coming from different purification batches, unless differently stated.
The effect of different pH and buffers was evaluated in terms of activity and stability using 100 mM acetate (pH 4.0-5.5), KPi (pH 6.0-8.0), Tris (pH 7.0-9.0) and carbonate (pH 9.2-10.4). 100% activity was represented in both cases by the initial velocity obtained by the condition of 100 mM KPi at pH 7.5. For the pH activity experiments ( Fig.2A), the standard assay conditions were modified employing 0.3 mM NAD + . In the pH stability set-up (Fig. 2B), the incubation of the protein with the buffer of choice was carried out for 20 h at 30°C. After that, the mixture was transferred in the 96-well plates, final 0.5 mM NAD + was added and the reaction was started upon the addition of 20 mM sodium formate.
The T m of Fdh SNO (Fig. 2C) was determined in the presence and absence of NAD + and NADP + substrates using the SYPRO Orange method [72]. The optimal protein:dye ratio was found after initial screening tests by varying the ratios of protein (1-20 lM) and dye  in the presence of 50 mM KPi, pH 7.5. The experiments in the presence of the cofactors were carried employing the final nicotinamide concentration of 10 mM. In 96-well PCR plates, the apo-, NAD + -and NADP + -Fdh SNO conditions were prepared in technical quadruplicates starting from 50 lM protein in buffer supplemented with 59 SYPRO Orange dye. Each sample was directly diluted in the wells with MQ water to obtain final 5 lM protein, and then resuspended. The T m experiments were performed using a BioRad iCycler iQ instrument by measuring at the wavelength of 575 nm. A temperature range of 25-95°C was chosen with a dwell time of 1 min between each datapoint collected.
To determine the influence of the temperature storage (Fig. 2D), 3.0 lM Fdh SNO aliquots were conserved in the dark at 4 and 30°C. For each time point, part of these aliquots was used at the final concentration of 0.3 lM in standard assay conditions.
The effect of several freeze-thaw cycles (Fig. 2E) was assessed in standard assay conditions, following the flashfreezing in liquid nitrogen and consequent rapid thawing in a 10°C water bath of 2.0 lM Fdh SNO aliquots.
The kinetic parameters of the wild-type Fdh SNO were spectrophotometrically determined in 50 mM KPi pH 7.5, at 30°C, as previously described [14], using 0.2 lM enzyme. In short, keeping the NAD + concentration constant at 2.0 mM, we calculated the micromoles of NADH formed per minute per milligram of protein by adding different amounts of formic acid (0.3-18.0 mM). By varying the NAD + concentration from 10 to 1000 lM, we estimated the kinetics for the cofactor by adding 20 mM formic acid. Measurements were made in quadruplicate (n = 4, the error indicates the standard deviation) and the corresponding initial velocities were fitted through the Michaelis-Menten equation. In the case of the NADP + -specific variants of Fdh SNO D221Q and GQKV (A198G/D221Q/H379K/ S380V), we fixed NADP + at 3.0 mM and added the range of 0.3-200.0 mM sodium formate to start the reactions. To calculate the kinetic parameters for NADP + , variable amounts of cofactor (10-3000 lM) were employed keeping the formate concentration constant at 200 mM. In the presence of NAD + , we changed the range of formate to 1-350 mM, while the amount of NAD + was fixed at 20 mM. Vice versa, we estimated the kinetic parameters for NAD + (0.25-40.0 mM) adding the constant concentration of 350 mM sodium formate for both the mutants. The kinetics of Fdh SNO D221Q and GQKV are the result of measurements on biological triplicates (n = 3, the errors correspond to the s.e.m.), plotted according to the Michaelis-Menten model (Fig. S7).
In the IC 50 experiments (Fig. S3), we incubated the inhibitors with the mixture reaction for a few min before adding the formate to start the assay under standard conditions. Azide was tested in the 10 À1 to 10 À8 mM range, while thiocyanate was used in the 10 À5 to 10 2 mM range. The relative activity for each condition was determined by comparison with the initial velocity displayed by Fdh SNO in the absence of the inhibitor (arbitrarily set at 100%).

Thiol-resistance and stability experiments
The experiments shown on Fig. 3B,E on Fdh SNO WT and its cysteine mutants S52C, V255C, A288C were carried out at 30°C, as described in the above-mentioned standard assay conditions, including 50 mM KPi pH 7.5, 0.2 lM Fdh SNO , 0.5 mM NAD + and 20 mM formate. All the thiol compounds used were freshly prepared, since they were solubilized in 50 mM KPi pH 7.5 buffer and immediately added to the reaction mixture. After 2 min of incubation with the nicotinamide cofactor and the enzyme, formate was added to trigger the reaction, which was spectrophotometrically monitored at 340 nm. BME, DTT, NEM and MTSET were used at the final concentration of 2.0 mM, while DTNB and PCMB were tested at the final concentration of 0.2 mM. DTNB was freshly prepared and stored in an eppendorf tube covered with aluminium foil to prevent photo-damage events [73]. PCMB was solubilized according to the procedure described for pmercuribenzoates by KK Han et al. [74]. In the case of ECAA and ECHB, we diluted them directly in 50 mM KPi pH 7.5 buffer at the concentration of 200 mM, and finally tested at 20 mM in the same manner described for the other thiol compounds.
The chemical stability experiments (Table 3, Fig. 3C and were then added to trigger the enzymatic reaction in 200 lL of final volume. Setting the initial velocity in the absence of any compound as 100% relative activity, we plotted the obtained initial velocities using the one-phase decay equation built-in in the GRAPHPAD PRISM software to calculate the protein half-life (t ½ ) for each tested compound.

Crystallization conditions and data collection
In order to prepare the protein for crystallographic purposes, the overproduction and purification procedures were carried out as described above, with the only exception being the elution buffer during SEC, which was composed of 20 mM HEPES pH 7.0, 50 mM NaCl. Upon initial screening with commercial kits, Fdh SNO was crystallized using the hangingdrop vapour diffusion technique. An equal volume (1 lL) of apo-protein solution (12 mgÁmL À1 ) was mixed with the crystallization solution [0.1 M Bis-Tris pH 5.5, 5-20% polyethylene glycol (PEG) 3350, 0.1 M MgCl 2 ] and allowed to equilibrate at room temperature. Crystals appeared within a week and stopped growing after 3 weeks. X-ray diffraction data were collected at beamline P13 at the European Molecular Biology Laboratory (EMBL) outstation at Deutsches Elektronen-Synchrotron, Hamburg, Germany.
Following size exclusion chromatography, the quadruple mutant of GQKV-Fdh SNO was concentrated to 18 mgÁmL À1 using a 10 000 MWCO concentrator. The protein was mixed with a custom screen crystallization buffer composed of 0.05 M citric acid pH 4.5, 5-20% PEG 3350 and 0.05-0.1 M MgCl 2. This buffer mixture was predetermined from thermal shift assay results and the protein mixed with the buffer in a ratio of 2 : 1, respectively. Long, colourless rectangular prism crystals were obtained via the sitting drop vapour diffusion technique which were of suitable quality for X-ray diffraction studies. The crystals were soaked for 20 s in a 0.01 M solution of NADP + and NaN 3 and were then further soaked in 30% (v/v) PEG 400 for cryo-protection. The crystals were subsequently flash-frozen in liquid nitrogen in preparation for diffraction experiments. Data were collected at beam line ID30A-1 (ESRF, Grenoble).
All data were processed with XDS [75]. WT Fdh SNO crystallized in P1 space group with four molecules in the asymmetric unit, and diffracted to the resolution of 2.1 A. The structure of apo-Fdh PSE from the Protein Data Bank (PDB entry: 2NAC) was used as a search model for the molecular replacement. A preliminary Fdh SNO model was obtained running the BALBES pipeline [76]. The electron density map was manually inspected and all residues from position 2 to 373 could be unambiguously modelled. GQKV-Fdh SNO diffracted to 2.5 A resolution (P6 1 space group), allowing the unambiguous placement of the amino acids 2-382 within the experimental density map. The structure of holo-Fdh PSE from the Protein Data Bank (PDB entry: 2NAD) was used as a search model for Molecular Replacement with Phaser [77]. The model was built using Coot [78], and the refinement and validation steps were carried out with Phenix [79]. Data collection and refinement statistics for all proteins are reported in Table S4. All models were deposited into the PDB repository with the following IDs: 7QZ1 for Fdh SNO and 8OQ2 for GQKV-Fdh SNO .

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1. Amino acid occurrence in three positions of close homologues of Fdh TBA . Table S2. Genomic GC content comparison. Table S3. Codon-optimization for the expression in E. coli of the gene Snov_3272 from S. novella. Table S4. Data collection and refinement statistics of apo-and holo-Fdh from Starkeya novella, PDB entry: 7QZ1 and 8OQ2, respectively. Table S5. List and main features of the available structures of NAD(P) + -dependent Fdhs. Table S6. Accessible Surface Area and Accessibility calculations of positions displaying cysteines from the crystal structures of Fdh SNO , Fdh PSE and Fdh CBO . Table S7. List of the primers used in this study. Fig. S1. Codon-optimization for the recombinant overproduction of Fdh SNO . Fig. S2. NAD + -specificity of Fdh SNO . Fig. S3. IC 50 plots for azide (N3) and thiocyanate (SCN). Fig. S4. Size-exclusion chromatography (SEC) on a Superdex 200 Increase 10/300 GL of the cysteine mutants of Fdh SNO following Immobilized Metal Affinity Chromatography. Fig. S5. Chemical stability of WT Fdh SNO and V255C against ethyl 4-chloroacetoacetate (ECAA) and ethyl (S)-(À)-4-chloro-3-hydroxybutyrate (ECHB). Fig. S6. Size-exclusion chromatography (SEC) on a Superdex 200 Increase 10/300 GL column of the mutants of Fdh SNO following Immobilized Metal Affinity Chromatography. Fig. S7. Kinetic data for NADP + -dependent Fdh SNO constructs D221Q and GQKV. Fig. S8. Conservation of the hydrogen-bond network formed between azide and active site residues in different Fdhs.