Evidence of a sequestered imine intermediate during reduction of nitrile to amine by the nitrile reductase QueF from Escherichia coli

In the biosynthesis of the tRNA-inserted nucleoside queuosine, the nitrile reductase QueF catalyzes conversion of 7-cyano-7-deazaguanine (preQ0) to 7-aminomethyl-7-deazaguanine (preQ1), a biologically unique four-electron reduction of a nitrile to an amine. The QueF mechanism involves a covalent thioimide adduct between the enzyme and preQ0 that undergoes reduction to preQ1 in two NADPH-dependent steps, presumably via an imine intermediate. Protecting a labile imine from interception by water is fundamental to QueF catalysis for proper enzyme function. In the QueF from Escherichia coli, the conserved Glu89 and Phe228 residues together with a mobile structural element composing the catalytic Cys190 form a substrate-binding pocket that secludes the bound preQ0 completely from solvent. We show here that residue substitutions (E89A, E89L, and F228A) targeted at opening up the binding pocket weakened preQ0 binding at the preadduct stage by up to +10 kJ/mol and profoundly affected catalysis. Unlike wildtype enzyme, the QueF variants, including L191A and I192A, were no longer selective for preQ1 formation. The E89A, E89L, and F228A variants performed primarily (≥90%) a two-electron reduction of preQ0, releasing hydrolyzed imine (7-formyl-7-deazaguanine) as the product. The preQ0 reduction by L191A and I192A gave preQ1 and 7-formyl-7-deazaguanine at a 4:1 and 1:1 ratio, respectively. The proportion of 7-formyl-7-deazaguanine in total product increased with increasing substrate concentration, suggesting a role for preQ0 in a competitor-induced release of the imine intermediate. Collectively, these results provide direct evidence for the intermediacy of an imine in the QueF-catalyzed reaction. They reveal determinants of QueF structure required for imine sequestration and hence for a complete nitrile-to-amine conversion by this class of enzymes.

different chemical steps of the enzymatic reaction are coordinated exactly with the intermediate physical steps of coenzyme binding and release, and these in turn must be precisely timed with the disengagement of free imine from the covalent thiohemiaminal. The way of how QueF orchestrates these multiple steps (4 -6) immediately suggests strategies possibly utilized for intermediate sequestration.
QueF enzymes are oligomeric proteins built from subunits that possess a small, so-called tunneling-fold (T-fold) domain (18). The core T-fold domain consists of a ␤␤␣␣␤␤ arrangement of secondary structure (4,5,9). The known QueF enzymes are divided into unimodular (e.g. QueF from Bacillus subtilis) (3,5,19,20) and bimodular groups (e.g. QueF from Escherichia coli (ecQueF)) (4,7,(21)(22)(23)(24), depending on whether they comprise subunits containing only a single T-fold domain or two T-fold domains in tandem repeat, respectively. Both types of QueF have their active sites located at the structural interface of two T-fold domains. The active site residues, and their structural arrangement into a functional catalytic center, are highly conserved in both QueF types (4,5,21). In addition to the catalytic cysteine (Cys 190 in ecQueF), there is an invariant pair of acidic residues of which the aspartic acid (Asp 197 in ecQueF) plays a role in proton transfer, as shown in Scheme 1, and the glutamic acid (Glu 89 in ecQueF) is involved in binding recognition of preQ 0 .
Crystallographic studies of QueF from B. subtilis (5,20) and the bimodular QueF from Vibrio cholerae (vcQueF; (4)) show that preQ 0 binding involves an induced-fit conformational change in protein structure, with the consequence that the substrate binding site, which is quite open in the apoenzyme, is closed up completely in the holoenzyme. The bound preQ 0 becomes sequestered completely from the solvent as a result (Fig. 1).
In vcQueF, for which crystal structures of apoenzyme (PDB code 3RJ4), noncovalent complex of a C194A variant with preQ 0 (PDB codes 3RZP and 3UXV), and covalent complexes of H233A (PDB code 4GHM) and R262L (PDB codes 3S19 and 3UXJ) variants with preQ 0 were determined, the induced-fit binding of preQ 0 comprises three main elements in particular. A characteristic device, generally referred to as the "QueF motif," is embedded structurally in a ␣-helix flanking the active site (5,9). At its N-terminal end, this helix has a glutamic acid (Glu 89 in ecQueF), which forms a hydrogen bond with preQ 0 (2.8 Å) in the enzyme-substrate complex, as shown in Fig. 1. The glutamic acid is positioned opposite to an adaptable element of the QueF structure (Fig. 1, B and C) contributing the catalytic cysteine to the active site. This "Cys element," which is completely conserved within bimodular QueF enzymes, additionally comprises Leu 191 and Ile 192 (ecQueF numbering). The Cys element rearranges to cover the active site in the preQ 0enzyme complex structure (PDB code 4GHM; Fig. 1C). Finally, a conserved phenylalanine (Phe 228 in ecQueF), which adopts a position to open up the preQ 0 binding pocket in the apoenzyme, closes down over the bound substrate, by forming pistacking interactions with it (3.7 Å), in the holoenzyme (Fig.  1C). Two key residues of preQ 0 binding, Glu 89 and Phe 228 , and the catalytic Cys 190 thus get "locked" onto the bound substrate at the end of the structural rearrangement. These features of vcQueF structure important for enzymatic function appear to be completely conserved in ecQueF. Overall, the sequence identity between the two enzymes is 65%. SCHEME 1. The proposed mechanism of ecQueF reducing preQ 0 to preQ 1 is shown. Residues involved in catalysis are indicated. Also shown are residues involved in closing up the binding site of preQ 0 and targeted by mutational analysis in this study. The amino acid residues are arranged to indicate their relative positions in the enzyme structure. Amino acid numbering of ecQueF is used. The dashed line indicates a hydrogen bond.

Imine intermediate during nitrile reduction by QueF
Kinetic studies of ecQueF indicate that NADPH binds to QueF already at the noncovalent complex with preQ 0 (7). The presence of NADPH enhances ϳ4-fold the thioimide formation rate compared with reaction of free enzyme and preQ 0 . The substrate-binding pocket has a narrow aperture, made up from Glu 89 and the Cys element, Ile 192 in particular (Fig. 1), that appears to be used for threading the nicotinamide moiety of NADPH into the active site (21). Structurally, therefore, substrate cannot be released before the coenzyme has dissociated. Accordingly, an important feature of QueF efficiency in intermediate sequestration, suggested from kinetic analysis, is the extremely tight binding of NADPH by the enzyme at the stage of the thiohemiaminal/imine (7).
The evidence just summarized suggests that the residues Glu 89 and Phe 228 and also the Cys element are significant for intermediate sequestration in ecQueF. Anchoring the proposed imine through interactions from these QueF structural elements might be crucially important. In this study, therefore, we performed mutagenesis of ecQueF, replacing the "anchoring" Glu 89 and Phe 228 by residues (Ala or Leu) unable to fulfill the analogous function. We additionally replaced Leu 191 and Ile 192 by Ala with the aim of perturbing the active site-closing movement of the Cys element. Overall, we considered that because of a substrate-binding site not properly closable any more, the ecQueF variants might no longer behave as highfidelity nitrile reductases. The putative imine intermediate might become intercepted by water able to enter an opened-up active site in the variant enzymes. Detailed characterization of these ecQueF variants reveals two-electron reduction of nitrile to imine for the first time in a nitrile reductase, thus providing direct evidence of the intermediacy of an elusive imine in the native catalytic reaction, and provides a structural and mechanistic explanation for complete nitrile-to-amine conversion by QueF enzymes, emphasizing the critical importance of intermediate sequestration.

Isothermal titration calorimetry study of preQ 0 binding to ecQueF variants
Purified ecQueF variants were shown to bind preQ 0 , and the corresponding thermodynamic characteristics were determined, in ITC experiments. Binding of preQ 0 by ecQueF is a two-step process in which an initially formed noncovalent enzyme-preQ 0 complex is converted to a covalent thioimide adduct (Scheme 2 and Ref. 7). Note that with ITC, it was not A, a superimposition of apoenzyme and holoenzyme structures of the vcQueF homodimer is shown. B and C, a close-up view of the binding site for preQ 0 in apoenzyme (B) and holoenzyme (C) is shown. The orientation of NADPH binding is from the crystal structure of the ternary complex of enzyme, preQ 0 , and NADPH (vcQueF R262L variant; PDB code 3UXJ). Despite missing density for the nicotinamide ring in the protein-bound NADPH (dark gray), the likely orientation of the nicotinamide moiety can be inferred from the binding positions of preQ 0 and the structurally resolved portion of NADPH. Binding of NADPH in ecQueF was also examined with structure modeling (21). The ecQueF residues corresponding to the vcQueF residues shown in A-C are Glu 89 , Ser 90 , Cys 190 , Leu 191 , Ile 192 , Asp 197 , Phe 228 , His 229 , and Glu 230 . The preQ 0 , NADPH, and amino acids in the binding pocket are indicated by element-based colors.

Imine intermediate during nitrile reduction by QueF
possible to distinguish between the two steps of preQ 0 binding and the data report on the binding process as a whole. The results are shown in Fig. 2, and the parameters calculated from the data are summarized in Table 1.
Strong heat release on titrating enzyme solution with preQ 0 solution (Fig. 2), similarly as it occurred with the wildtype enzyme (7), was good evidence that ecQueF variants had retained the ability to bind preQ 0 . Substitution of Glu 89 caused substantial weakening of the preQ 0 binding as compared with wildtype enzyme, reflected in a smaller negative value of ⌬G binding (⌬⌬G binding ϭ ϩ14 kJ/mol) and increased overall apparent dissociation constant K d (292-fold). The degree of disruptive effect was similar in both enzyme variants. The F228A variant also showed weakened preQ 0 binding, reflected in a smaller negative value of ⌬G binding (⌬⌬G binding ϭ ϩ10.8 kJ/mol) and 90-fold increased K d . In terms of ⌬G binding and K d , preQ 0 binding was unaffected by the replacement of Leu 191 or Ile 192 with Ala. Interestingly, the ⌬G binding of the E89A, E89L, and F228A variants was comparable to the ⌬G binding of the pre-  Table 1. The c values (c ϭ [dimeric protein]/K d ) obtained from the experiments were 5 for the E89A, 6 for the E89L, 235 for the L191A, 435 for the I192A, and 11 for the F228A variant. SCHEME 2. The two-step kinetic mechanism of preQ 0 binding is shown. E is free ecQueF. E⅐preQ 0 indicates the noncovalent complex. E-preQ 0 indicates the covalent thioimide adduct.

Imine intermediate during nitrile reduction by QueF
viously reported C190A variant, which can bind preQ 0 just noncovalently (7). However, contrasting their similarity overall, the ⌬G binding values of these ecQueF variants involved distinct relative contributions from enthalpy (⌬H binding ) and entropy (ϪT⌬S binding ). The ⌬H binding became more negative (favorable of binding), and the ϪT⌬S binding became more positive (nonfavorable of binding) in going from C190A via the variants. A covalent thioimide formation more pronounced in E89L compared with the E89A and F228A variants and lacking completely in C190A variant could explain these trends in ⌬H binding and ϪT⌬S binding .
In comparison to the wildtype enzyme, the ⌬⌬G binding of the E89A variant was primarily due to the effect on ⌬H binding . In the E89L and F228A variants, however, the ⌬⌬G binding involved effects, of a similar degree, on ⌬H binding and ϪT⌬S binding . The ⌬G binding of L191A variant involved a ⌬H binding lower and a ϪT⌬S binding higher than the corresponding preQ 0 binding parameters of the wildtype enzyme. However, the ⌬H binding and ϪT⌬S binding terms compensated each other in an overall unchanged ⌬G binding . A ϪT⌬S binding term substantially higher in the ecQueF variants (E89L and L191A) is therefore worth noting. In the I192A variant, the ⌬H binding and ϪT⌬S binding terms of preQ 0 binding were comparable to what they were in the wildtype enzyme.

Kinetic analysis of covalent thioimide adduct formation in ecQueF variants
Absorbance at 380 nm (⑀ ϭ 10.02 Ϯ 0.14 mM Ϫ1 cm Ϫ1 ) indicates the covalent thioimide intermediate of wildtype ecQueF (7). Titration of the enzyme variants with preQ 0 also gave rise to a "thioimide" absorbance band, as shown in Fig. 3. Maximum absorption was however shifted to a slightly higher wavelength (Յ5 nm) in the Glu 89 variants. In wildtype ecQueF based on half-of-the-sites reactivity of the enzyme dimer, the molar equivalent of preQ 0 was sufficient to convert all enzyme present into the covalent adduct (7). In ecQueF variants, especially the ones having Glu 89 replaced, excess preQ 0 (ϳ10-fold) was required for full conversion of the enzyme (Fig. 3A).
In wildtype ecQueF as shown recently (7), the thioimide formation involves a noncovalent enzyme-preQ 0 complex in rapid equilibrium, which reacts only relatively slowly to the covalent adduct (Scheme 2). Absorbance time courses from preQ 0 binding experiments with the E89A, L191A, I192A, and F228A variant were best fit with single exponentials. The overall fit was not improved when double exponentials were used (data not shown). The corresponding rate constants and amplitudes (data not shown) were hyperbolically dependent on the preQ 0 concentration, as expected if the ecQueF variants reacted in two distinct steps, as shown in Scheme 2, analogous to the wildtype enzyme (7). A global fit of the reaction time courses was therefore performed using the kinetic mechanism in Scheme 2. Experimental data are superimposed in Fig. 3 with the fitting results. Parameters of preQ 0 noncovalent binding (K d kinetic ϭ k 2 /k 1 ) and thioimide formation (k 3 , k 4 ) by the ecQueF variants are summarized in Table 2 along with the corresponding parameters of the wildtype enzyme.
Compared with the wildtype enzyme, the K d kinetic was increased in all ecQueF variants (F228A, 59-fold; E89A, 41-fold; and L191A, 7.1-fold), except for the I192A variant that exhibited the same K d kinetic as the wildtype enzyme. The k 3 was decreased in all enzyme variants, most strongly so in the F228A variant (10-fold). The ratio k 3 /k 4 describes the tightening of preQ 0 binding in consequence of covalent thioimide formation. The effect was more pronounced in the L191A variant (k 3 /k 4 ϭ 440) than in the E89A (k 3 /k 4 ϭ 53), I192A (k 3 /k 4 ϭ 26), and F228A variant (k 3 /k 4 ϭ 40). Including the wildtype enzyme in this comparison is difficult because only a rough lower limit of 68 could be given for k 3 /k 4 . Because in kinetic experiments the k 4 was not different from 0 (7), the relationship K d overall ϭ K d kinetic /(1ϩk 3 /k 4 ) was used to calculate the upper bound of k 4 from the apparent K d overall of the wildtype enzyme determined in ITC experiments. The values of K d overall determined kinetically and obtained from ITC data agreed for the L191A and  The DMSO concentration did not exceed 2%. The black dashed lines are the averaged data from triplicate measurements. Global fitting was performed using the two-step binding mechanism in Scheme 2. The fits are shown with orange lines. The obtained kinetic constants are summarized in Table 2.

Imine intermediate during nitrile reduction by QueF
F228A variant. However, they differed by as much as 5-fold for the E89A and I192A variant. We explain this difference by the fact that in the ITC measurement, both the covalent and the noncovalent complex contribute to the recorded heat signal, whereas absorbance is completely specific for detecting the thioimide adduct. Therefore, only absorbance measurements allow for a clear-cut determination of K d overall .

Reduction of preQ 0 by the ecQueF variants
Direct analysis of the reaction mixtures by 1 H NMR ( Fig. 4) and HPLC (Fig. 5) was used to examine preQ 0 reduction by the ecQueF variants. Initial evidence was that although F228A and both Glu 89 variants showed conversion of preQ 0 at a slow rate, they hardly produced preQ 1 . We considered that any imine intermediate exposed to water after the first reduction would be detected as 7-formyl-7-deazaguanine. Using authentic reference material from chemical synthesis (7), we identified 7-formyl-7-deazaguanine as the main product of preQ 0 reduction by E89A, E89L, and F228A variants, as shown in Figs. 4 and 5. Proton signals characteristic of preQ 1 were below detection in these reactions. Conversely, 7-formyl-7-deazaguanine was completely absent from, and preQ 1 the only detectable product in, reactions of the wildtype enzyme. Reaction of the L191A variant gave a mixture of products comprising mainly preQ 1 (80 -90%), but also 7-formyl-7-deazaguanine that accounted for the rest of preQ 0 converted. The preQ 0 reduction by the I192A variant also proceeded to form both preQ 1 and 7-formyl-7-deazaguanine, roughly at a ratio of 1:1.
LC and LC-MS analysis confirmed the ecQueF variants to exhibit the particular change in product formation from preQ 0 (Fig. 5). When the preQ 0 reduction with the variants were done in Tris buffer, LC traces revealed a putative product peak of unknown identity, appearing always together with the expected product peak for 7-formyl-7-deazaguanine (Fig. 5B). The unknown peak exhibited absorbance at 262 and 340 nm, just like 7-formyl-7-deazaguanine but contrary to preQ 0 and preQ 1 , which both show absorbance only at 262 nm. When the same enzymatic reactions were done in phosphate buffer, the second peak was missing, and only the peak for 7-formyl-7-deazaguanine was found (Fig. 5, C and D). This together with the characteristic mass of the unknown peak (1H ϩ , 282) suggested an imine adduct between 7-formyl-7-deazaguanine and Tris, as shown in Fig. 5, that may have formed in solution already during the enzymatic reaction or during sample preparation.
Ability of the LC and LC-MS analytic method to quantify even small amounts of the secondary reaction product (e.g. 5% preQ 1 in preQ 0 conversion by the E89A variant) made it possible to determine whether the NADPH or the preQ 0 concentra-tion had an influence on the product distribution in the enzymatic conversions. Whereas high concentrations of NADPH might prevent the loss of imine intermediate from the enzyme and so pull the enzymatic reaction toward preQ 1 formation, the opposite effect, namely preference for formation of aldehyde, could be expected at high preQ 0 concentration (Scheme 3). In each enzyme including wildtype ecQueF, there was no effect from variation in [NADPH] between 1 and 10 mM. Note that relatively high NADPH concentrations were used in the experiment to maximize the possible "pull" toward preQ 1 . The case of the L191A and I192A variants, in which on average one imine was lost in every five (L191A) and two (I192A) preQ 0 conversion events, clearly shows that exchange between NADP ϩ and NADPH for complete 2-fold reduction of preQ 0 was still somewhat functional after the replacements of Leu 191 and Ile 192 . This result immediately suggests the possibility that imine was intercepted by water in, or escaped into solution from, the L191A and I192A variants before the new NADPH had the chance to bind, implying in turn an early disengagement of the imine from the covalent thiohemiaminal. The alternative possibility is that imine was intercepted by water in, or was released from, the ternary complexes of L191A and I192A variants with NADPH.
Contrary to NADPH, the preQ 0 substrate, on variation of its concentration, affected strongly the product distribution from nitrile reduction by the different ecQueF variants. In each enzyme as shown in Fig. 6, the percentage of imine in the total reaction product released increased upon increase in the molar ratio [preQ 0 ]/[enzyme dimer] used in the experiment. The effect was most pronounced in the E89A and F228A variants, which switched from forming primarily preQ 1 (80%) when preQ 0 was used at the same concentration as the enzyme dimer to being almost completely selective for imine formation (Ն95%) at higher [preQ 0 ]/[enzyme dimer] ratios. The E89L variant gave predominantly imine at all conditions used. In the I192A variant, the percentage of imine in total product increased steadily dependent on the [preQ 0 ]/[enzyme] ratio and reached a limiting value of ϳ50% when preQ 0 was added in excess. The L191A variant formed imine as the minor product (Յ20%). In all enzyme variants except for F228A, the percentage of imine in total product typically exhibited a roughly hyperbolic dependence on the [preQ 0 ]/[enzyme] ratio used (Fig. 6, A-D), from which the maximum imine formation at saturating preQ 0 was determined. In the F228A variant, maximum imine formation was established from the data, but the underlying dependence of product selectivity on [preQ 0 ]/[enzyme] ratio appeared complex (Fig. 6E).

Imine intermediate during nitrile reduction by QueF Kinetic analysis and KIE studies
To analyze the enzyme kinetics in more detail, and determine primary isotope effects (KIE) from the use of (4R)-[ 2 H]NADPH as reductant, we monitored the preQ 0 conversion by wildtype and variant enzymes by in situ proton NMR and also by LC. Kinetic data obtained from time-course analysis are summarized in Table 3. Material balances were fully consistent with the chemical reaction, preQ 0 ϩ NADPH ϩ H 3 O ϩ 3 7-formyl-7-deazaguanine ϩ NADP ϩ ϩ NH 3 , for the variants (E89A, E89L, F228A) carrying out only a single NADPH-dependent reduction of preQ 0 . At long reaction times (Ն24 h), however, decomposition of NADPH became significant as a spontaneous side reaction. The reaction of wildtype ecQueF was confirmed as preQ 0 ϩ 2 NADPH ϩ 2H ϩ 3 preQ 1 ϩ 2 NADP ϩ . Utilization of NADPH for preQ 0 reduction by the L191A and I192A variants was fully consistent with the preQ 1 and 7-formyl-7deazaguanine obtained. Product selectivity of the enzyme variants was confirmed and shown to be constant over the whole time course with prevailing steady-state conditions ([preQ 0 ] Ͼ 15 [enzyme dimer]). The specific rate of preQ 0 conversion was slowed in the enzyme variants compared with wildtype ecQueF, most strongly in E89L (700-fold), then E89A (160-fold), F228A (140-fold), L191A (24-fold), and I192A (2.7-fold). Using (4R)-[ 2 H]NADPH instead of NADPH, we showed that the deuterium was transferred from coenzyme to form 7-formyl-7deazaguanine (all enzyme variants) and preQ 1 (L191A and I192A variants), indicating the same pro-R stereoselectivity for NADPH conversion in the variants as in wildtype enzyme (7).
The KIE determined from a direct comparison of reaction rates with NADPH and (4R)-[ 2 H]NADPH was 2.9 (Ϯ0.7) for the E89A variant, 3.0 (Ϯ0.6) for the E89L variant, and 2.8 (Ϯ0.4) for the F228A variant. The "reference KIE" for the reaction of the wildtype enzyme was 1.8 (Ϯ0.5). The shown KIEs are means (Ϯ S.D.) from five independent determinations and represent the averaged isotope effects on the rates of substrate consumption (preQ 0 , NADPH) and product formation (preQ 1 or 7-formyl-7-deazaguanine, NADP ϩ ). The L191A and I192A variants differed from the wildtype enzyme and the Glu 89 and Phe 228 variants in that their reaction rates based on substrate consumption and product consumption were all affected by a different KIE, as shown in Table 3.
In the reaction of the L191A variant, the KIE on preQ 0 consumption was 3.7 (Ϯ0.2), higher than the KIE of 2.3 (Ϯ0.2) on the release of 7-formyl-7-deazaguanine but lower than the KIE of 5.1 (Ϯ0.2) on the main reaction path of the enzyme to preQ 1 . The KIE on conversion of NADPH to NADP ϩ was 4.4 (Ϯ0.2). Using (4R)-[ 2 H]NADPH as coenzyme, the molar ratio between preQ 1 and 7-formyl-7-deazaguanine was just 1.8, significantly smaller than the ratio of 4 received from preQ 0 reduction by NADPH with the L191A variant.
There is only one isotope-sensitive step involved in 7-formyl-7-deazaguanine formation: that of reduction of preQ 0 to imine. The KIEs for the E89A, E89L, and F228A variants therefore reflect the slowing down of the enzymatic reactions at just this

Imine intermediate during nitrile reduction by QueF
step. The individual KIEs associated with reductions of the preQ 0 substrate and the imine intermediate by wildtype ecQueF were determined previously as 3.3 and 1.8, respectively (7). Global fitting of reaction time courses of preQ 0 conversion were used in our earlier study (7) to obtain these KIEs. The KIEs on conversion of preQ 0 to 7-formyl-7-deazaguanine by the SCHEME 3. The proposed kinetic mechanism of preQ 0 conversion by L191A and I192A variants of ecQueF via partitioning of the imine intermediate between reduction to preQ 1 and hydrolysis to 7-formyl-7-deazaguanine is shown. Covalent (-) and noncovalent (⅐) binding of substrates/products and coenzymes to the enzyme are indicated. Rate constant numbering starts from Scheme 2 with the assumption that enzyme was saturated with NADPH. Nitrile reduction to imine (k 5 ) and imine reduction to preQ 1 (k 6 ) are isotope-sensitive steps. Coenzyme exchange in imine intermediate is assumed to be in rapid equilibrium. The dotted line is a hyperbolic fit to the relative product proportion of 7-formyl-7-deazaguanine, and its maximum relative content is shown.

Table 3 Catalytic reaction rates and their associated KIEs for ecQueF wildtype and variants thereof
The catalytic reaction rates are given as mol product released (or substrate converted)/(mol enzyme ϫ min). They were determined individually for each compound participating in the reaction. The reaction rates of substrate (preQ 0 , NADPH) consumption are negative, and those of product (preQ 1 , 7-formyl-7-deazaguanine, NADP ϩ ) formation are positive. D V is the primary KIE, because of deuteration of NADPH, on the catalytic rate. It is the average of the five individual KIEs determined from preQ 0 , preQ 1 , 7-formyl-7-deazaguanine, NADH, and NADP ϩ . Each individual KIE is the average of five independent reactions. Note that in reactions of the E89A, E89L, and F228A variants, which perform a single-step reduction of preQ 0 to produce 7-formyl-7-deazaguanine, the experimental D V equals D k 5 according to Scheme 3. In the wildtype enzyme, D V involves contribution from the KIEs on k 5 and k 6 . a In parentheses is shown the primary KIE on the consumption rate or the formation rate of the compound indicated. b The KIE on imine reduction (k 6 in Scheme 3) is shown in square brackets. As explained in text, this KIE is the same as the KIE on the experimental product ratio R, preQ 1 / 7-formyl-7-deazaguanine, formed in the reaction. c NA, not applicable. Contrary to the wildtype enzyme and the E89A, E89L, and F228A variants in which D V was not dependent on whether substrate consumption or product formation was analyzed and so an average value could be given, the L191A and I192A variants necessitated KIE evaluation separately for each compound involved in the reaction. The concept of an average KIE was therefore not applicable.

Imine intermediate during nitrile reduction by QueF
Glu 89 variants and the F228A variant were of a similar magnitude as the first-step KIE, i.e. preQ 0 reduction to imine, for the reaction of wildtype ecQueF.
In the L191A and the I192A variant, partitioning of the imine intermediate between reaction to preQ 1 and reaction to 7-formyl-7-deazaguanine constitutes the point of departure for analyzing the different KIEs in these two enzymes. The molar ratio (R) of preQ 1 and 7-formyl-7-deazaguanine reflects the rate constant ratio of imine reduction/imine hydrolysis (k 6 /k 7 ), as shown in the kinetic mechanism depicted in Scheme 3. Because only the imine reduction is sensitive to isotopic substitution in coenzyme, the isotope effect on the product ratio (4/1.8 ϭ 2.22 for L191A; 2.6/0.48 ϭ 5.4 for I192A) is also the KIE on k 6 . A KIE on preQ 0 conversion smaller than on preQ 1 formation arises because partitioning of the imine intermediate is also subject to an isotope effect. The fraction of imine intermediate reacting to 7-formyl-7-deazaguanine is defined from Scheme 3 as With F H and F D known, the KIE on preQ 0 consumption obtains, according to Scheme 3, from the KIE on preQ 1 formation as KIE_preQ 0 ϭ KIE_preQ 1 (1 Ϫ F D )/(1 Ϫ F H ). The calculated value of KIE_preQ 0 is therefore 4.08 for the L191A variant. The same value is obtained for the I192A variant. The KIE on formation of 7-formyl-7-deazaguanine obtains from the relationship, KIE_preQ 0 F H /F D , as 2.27 and 1.70 for the L191A and the I192A variant, respectively. Finally, the KIE on NADPH consumption, which is the same as the KIE on NADP ϩ formation, is calculated with the relationship, KIE_preQ 0 (2 Ϫ F H )/ (2 Ϫ F D ), as 4.48 and 5.28 for the L191A and the I192A variant, respectively. These calculated KIE data agree well the experimentally obtained values.
Evidence summarized in Table 3 shows that in preQ 1 formation the complete reaction of the L191A or the I192A variant was affected by a substantially larger KIE than the partial reaction of imine reduction (k 6 ) and identifies nitrile reduction to imine as the rate-determining step of the overall preQ 0 conversion by these two ecQueF variants. Linear time courses of NADPH consumption by the enzyme, with no sign of a presteady state burst, were consistent with this notion (data not shown).

Discussion
Conversion of preQ 0 to 7-formyl-7-deazaguanine represents the first reaction of this kind catalyzed by an enzyme. Chemically, it consists in a single hydride transfer reduction of a nitrile group from NADPH followed by the addition of water to the imine thus obtained. Catalysis from the enzyme is required only for nitrile reduction. Imine hydrolysis is a spontaneous reaction, well established from other enzymatic transformations, like those of amino acid dehydrogenases for instance (25)(26)(27). Mechanistically, the "interrupted" (two-electron) reduction of preQ 0 is significant, for it provides first-time direct evidence on the occurrence of an imine intermediate in the reaction pathway of QueF nitrile reductases. In addition to reinforcing the current mechanistic proposal for QueF (Scheme 1) (3, 5-7), these results advance comprehension of the mode of action of the enzyme, demonstrating how critically important imine intermediate sequestration is for complete 2-fold reduction of preQ 0 to preQ 1 . The finding that site-directed substitutions interferring most significantly with intermediate sequestration (e.g. Glu 89 and Phe 228 ) were also strongly disruptive on the hydride transfer revealed that the major physical steps of the enzymatic process proceeded tightly interconnected with the catalytic act(s). It furthermore indicated that ecQueF made parsimonious use of its active-site structural devices to accomplish the different tasks in the overall multistep catalytic reaction. The thermodynamic signatures of preQ 0 binding by the different ecQueF variants appear consistent with the enzymesubstrate interactions revealed in the crystallographic evidence on vcQueF (Fig. 1).
The hydride transfers from NADPH are the slowest steps in the overall conversion of preQ 0 to preQ 1 by ecQueF (7). Evidence that catalytic rates were lowered by up to ϳ700-fold in the ecQueF variants, whereas their associated KIEs caused by deuteration of the coenzyme were not decreased as compared with the wildtype KIEs, strongly supports the suggestion that catalysis was also rate-limiting in the enzyme variants. The crucial importance of substrate positioning at the stage of imine reduction was most vividly evident from the fact that substitutions of Leu 191 and Ile 192 , and even more so those of Glu 89 and Phe 228 , caused loss of imine from the enzyme during the reaction. The fraction of preQ 0 , which on conversion by the ecQueF variants did not make it through to the fully reduced amine preQ 1 , increased dependent on the substrate amount present. A simple model, having preQ 0 compete with imine intermediate for binding to the enzyme in rapid equilibrium, fails to explain the experimental product distributions, for it predicts that at "saturating" [preQ 0 ], no preQ 1 will be formed. This is most evidently refuted from the behavior of the L191A and the I192A variants.
Formally, a general model of molecular complex dissociation, induced by a competing ligand as shown in literature (28), would be consistent with the effect of [preQ 0 ] on the product distribution in the ecQueF variants (Scheme 4). The model involves ligand dissociation effectively from a preassociated complex, which enables rapid rebinding of the ligand to give the actual biological form of the complex. The model predicts that competitor molecules will accelerate the breakdown of SCHEME 4. A hypothetical kinetic mechanism of ecQueF-imine complex dissociation induced by preQ 0 is shown. E is the free enzyme. The enzymebound preQ 0 and imine are indicated by subscript. Eϳimine indicates a preassociated complex of ecQueF with imine intermediate. The model is adapted from literature (28).

Imine intermediate during nitrile reduction by QueF
enzyme-ligand complexes by occluding the rapid rebinding of enzyme and ligand from the level of their preassociated complex. It is therefore worth noting that the model does not imply the formation of a ternary complex between enzyme, ligand and competitor. The ligand dissociation rate (k off ), enhanced in the presence of competitor (C), is thus given by the relationship, derived from Scheme 4, k off (C) ϭ k d ϫ (k esc ϩ k on C) Ϭ (k a ϩ k esc ϩ k on C), where k a and k d are first-order association and dissociation rate constants for (rapid) formation of the fully associated complex from the preassociated complex, k esc is the first-order rate constant of ligand release into solution, and k on is the rate constant for competitor binding (28). Note that the k esc step involves hydrolysis of imine. Scheme 4 does not distinguish mechanistically between whether the hydrolysis occurs in solution or, more likely, with imine still bound to the enzyme.
The results in Fig. 6 suggest that the ecQueF variants differed in the relative contribution of k esc and k on to k off . Whereas in the E89L variant the escape of imine appeared to be largely controlled by k esc (because k off was hardly dependent on the preQ 0 concentration), the product formation by the L191A and I192A variants reflected a prominent effect from k on . The E89A variant adopted an intermediate position between the two extremes, with both k esc and k on affecting the product distribution dependent on the competitor preQ 0 . A molecular account of the proposed kinetic scenario for ecQueF is not easy to conceive, and the model is essentially independent of such details. However, a preassociated enzyme-imine complex in a somewhat open conformation might make possible the kinetically implied displacement of the imine by preQ 0 .
In conclusion, identification of residues in the catalytic apparatus of ecQueF essential (Glu 89 and Phe 228 ) or auxiliary (Leu 191 and Ile 192 ) for complete 2-fold reduction of nitrile to amine has led to uncover the imine intermediate, and the necessity of its efficient sequestration in the active site, as central features of the enzymatic mechanism. Structure-based interpretation of the effect of site-directed substitutions on the reaction selectivity of ecQueF is that because of a substratebinding site (Fig. 1, B and C) not fully closable any longer, the enzyme variants lose the imine intermediate to hydrolysis, completely in E89A, E89L, and F228A and partly in L191A and I192A. Although not proven by the evidence shown, a likely scenario is that the imine is captured by water now able to enter the partly opened-up active site in the variant enzymes. The hydrolyzed imine, 7-formyl-7-deazaguanine, is not recognized by ecQueF as a substrate of NADPH-dependent reduction, as demonstrated previously (7). It would therefore be released into solvent.

Chemicals and materials used
NADPH (purity, Ͼ98%) and NADP ϩ (purity, Ͼ97%) were from Carl Roth (Karlsruhe, Germany). 2-Propanol-d 8 (99.5 atom % D) was from Sigma. All other materials were of the highest purity available from Carl Roth and Sigma. The preQ 0 and 7-formyl-7-deazaguanine were synthesized as described previously (7,19). A pET-28a(ϩ) expression vector encoding Thermoanaerobium brockii alcohol dehydrogenase as an N-terminally His-tagged protein was ordered from GenScript (Piscataway, NJ). Standard expression in E. coli BL21-DE3 and His-tag purification were used to prepare the enzyme.

Site-directed mutagenesis
Mutagenesis leading to site-directed substitution of Leu 191 by Ala (L191A) was performed according to a standard twostage PCR protocol (34, 35). A pEHISTEV expression vector comprising the ecQueF gene was used as the template. The oligonucleotide primers used are shown with the mismatched bases underlined: L191A forward, 5Ј-GCTGAAATCAAACT-GCGCGATCACCCATCAACC-3Ј; and L191A reverse, 5Ј-GGTTGATGGGTGATCGCGCAGTTTGATTTCAGC-3Ј. Mutagenesis to substitute Glu 89 by Ala or Leu was reported in an earlier study (21). Replacement of Ile 192 or Phe 228 by Ala was ordered from Genscript. All mutations were verified by gene sequencing.

Enzyme preparation
The ecQueF wildtype and site-directed variants thereof (E89A, E89L, L191A, I192A, and F228A) were obtained as N-terminally His-tagged proteins using expression in E. coli BL21-DE3 as described previously (7,21). All enzymes were purified by a reported two-step procedure, consisting of immobilized metal ion affinity chromatography and gel filtration (7). Enzyme purity was confirmed by SDS-PAGE. The background of the E. coli expression host did not contain QueF activity in amounts detectable with the assays used. Contamination of ecQueF variants with endogenous wildtype activity could thus be ruled out firmly. The protein concentration was measured with a Pierce BCA protein assay kit (Thermo Fisher Scientific). Enzyme stock solutions (0.8 -1.2 mM) were stored at Ϫ20°C and used up within 3 weeks because of limited stability.

ITC and spectrophotometric analysis of preQ 0 binding in ecQueF variants
A VP-ITC micro calorimeter from Microcal (Malvern Instruments Ltd., Malvern, UK) was used at 25°C. The enzyme was gel-filtered twice to phosphate buffer (100 mM Na 2 HPO 4 -NaH 2 PO 4 , pH 7.5, 50 mM KCl) using illustra NAP 5 columns (GE Healthcare). A DMSO concentration of maximally 2.0% (v/v) was used in both the enzyme and the preQ 0 solution. The experiments were conducted, and data evaluation done as described previously (7). The enzyme molar concentration was based on the protein concentration assuming a functional The covalent thioimide adduct between preQ 0 and ecQueF in wildtype or variant form was detected from its characteristic absorbance with maximum absorption at ϳ380 nm. Absorbance titrations were carried out with a Beckman DU 800 spectrophotometer (Beckman Coulter, Pasadena, CA) as described previously (7). Kinetic study of the thioimide formation was done with a SX.18 MV Stopped flow spectrophotometer (Applied Photophysics; Leatherhead, UK). The experimental progress curves of absorbance (380 nm) increase recorded at different protein, and preQ 0 concentrations were analyzed by a global fitting procedure, as previously described (7), that employed the program COPASI (version 4.11_build 65) and used the mechanism in Scheme 2. A least-squares fitting routine implemented in COPASI was used to determine the kinetic rate constants and their standard deviations. In the fitting, k 1 and k 2 were not determined individually, but only their ratio K d kinetic (which equals k 2 /k 1 ) was obtained. We showed that the fitting always converged to a unique and well defined solution that was independent of the parameter start values used. All fitted parameters (K d kinetic , k 3 , and k 4 ) were well determined as regards their associated standard deviation. An overall binding constant (K d overall ) was calculated from the results, using the relationship K d overall ϭ K d kinetic /(1 ϩ k 3 /k 4 ). Note that K d overall thus includes the covalent thioimide formation.
Binding constants were transformed into binding energies using the relationship ⌬G ϭ ϪRT lnK d . A 1 M standard state for all reactants was assumed. R is the gas constant (8.314 J/mol⅐K), and T is the temperature (298.15 K).

Enzymatic reduction of preQ 0
Experiments were performed to characterize the preQ 0 conversion by the different ecQueF enzymes. In addition to determining the specific rate of substrate consumption, determination of the product(s) formed by preQ 0 reduction was of main interest. Reactions were carried out by incubating buffered solutions of enzyme, preQ 0 , and NADPH on a Thermomixer Comfort (Eppendorf, Hamburg, Germany) at 25°C with agitation at 700 rpm for ϳ24 h. The DMSO concentration used was below 2.5%. The concentrations of preQ 0 and NADPH were varied as indicated. The enzyme concentration was 50 M (E89A, E89L, I192A, and F228A variants), 10 M (L191A variant), and 1.8 M (wildtype enzyme). The reactions were stopped by precipitating the enzyme with methanol (10%, by volume) at 70°C for 10 min (1000 rpm in a Thermomixer Comfort) (7). Samples were applied to 1 H NMR, HPLC, and LC-MS from comprehensive analysis.

HPLC analytics
The samples were analyzed using an Agilent 1200 HPLC system (Santa Clara, CA) equipped with a 5-m SeQuant ZIC-HILIC column (200 Å, 250 ϫ 2.1 mm; Merck), the corresponding guard column (20 ϫ 2.1 mm; Merck), and a UV detector ( ϭ 262 and 340 nm). A 20 mM ammonium acetate buffer (pH 6.67) was used. A decreasing gradient in acetonitrile was applied for compound separation over a 20-min run time. Specifically, the acetonitrile concentration was decreased only slightly from 80 to 78% over 5 min, and a steeper ramp from 78 to 60% was then used from 5 to 20 min. The column was washed with 60 and 80% acetonitrile for 5 min after each analysis. The flow rate was 0.5 ml/min. The column temperature was 30°C.
Optionally, a mass detector (Agilent 6200 Quadrupole) was coupled to the HPLC system, which in this case was equipped with a 2.7-m Poroshell 120 EC-C18 column (120 Å, 100 ϫ 3 mm; Agilent) and a UV detector ( ϭ 210, 262, and 340 nm). The masses of the products from the enzymatic reactions were scanned in a mass range of 100 -850 using a positive mode. The masses of preQ 0 (1H ϩ , 176), preQ 1 (1H ϩ , 180), and 7-formyl-7-deazaguanine (1H ϩ , 179) were also analyzed in SIM mode. A linear gradient of acetonitrile was used in water containing 0.01% formic acid. The gradient was from 0% to 25% acetonitrile over 5 min. The column was washed with 90% acetonitrile for 1 min and water for 0.5 min. The post time was 1 min. The flow rate was 0.7 ml/min. The column temperature was 30°C.

H NMR measurements
NMR spectra were recorded at 499.98 MHz and 30°C using a Varian INOVA 500-MHz spectrometer (Agilent Technologies) with VNMRJ 2.2D software. The spectra of coenzymes, preQ 0 and preQ 1 in D 2 O/H 2 O solution were reported previously (7,36,37). The reduction of preQ 0 by the different ecQueF enzymes was monitored by in situ 1 H NMR measurements that involved presaturation of the residual water signal. Tris buffer (30 mM, pH 7.5) in 120 mM KCl was used that also contained an internal standard solution of 30 l (trimethylamine hydrochloride, 1.82 ϫ 10 Ϫ5 mol/liter; trimethylsilylpropanoic acid, 0.05% in 99.5% D 2 O) to obtain data for correct integration. The D 2 O content was 14% (v/v), and the total reaction volume was 0.6 ml. The reactions were started through addition of the enzyme. Spectra of the reaction with wildtype ecQueF or the L191A variant were collected at every 5-20 min for 5-20 h in the magnet. Because the reactions with the Glu 89 variants were relatively slow, the spectra were collected for 48 h in the magnet, and the solutions were temporarily kept at room temperature. The data were analyzed in the software program ACD/NMR processor academic edition (version 12.01_build 39104). The total amount of 7-formyl-7-deazaguanine formed in the reaction was determined from the free aldehyde product present and the corresponding imine adduct between 7-formyl-7deazaguanine and Tris (3.82 ppm in 1 H NMR spectra, not shown).

KIE studies
NADPH and (4R)-[ 2 H]NADPH were prepared by reducing NADP ϩ in the oxidation of 2-propanol and 2-propanol-d 8 with T. brockii alcohol dehydrogenase. A reported protocol was used for their synthesis (7,38). Primary KIEs on preQ 0 reduction were obtained by comparing initial rates measured with NADPH and (4R)-[ 2 H]NADPH, both synthesized as described above. Enzymatic reactions were monitored by HPLC and for certain enzymes (wildtype, E89A, E89L, and L191A) additionally by in situ 1 H NMR. The initial rates were obtained from linear fits to HPLC or spectral data. Note that because all sub-

Imine intermediate during nitrile reduction by QueF
strates and products of the reaction were measured in the analysis, a total of four or five reaction rates was obtained: preQ 0 and NADPH consumption and preQ 1 , 7-formyl-7-deazaguanine, and NADP ϩ formation. The enzyme concentrations used in the reactions were as follows: wildtype, 1.5 M; E89A and E89L, 25 M; L191A, 8 M; I192A, 10 M; and F228A, 20 M. The preQ 0 concentration was saturating, that is, 500 M using the wildtype enzyme, the L191A, and the I192A variant and 800 M using the E89A, the E89L and the F228A variant. The NADPH concentration was 1.5 mM. The temperature was 30°C. The DMSO concentration was always at or below 2%. Phosphate buffer (100 mM Na 2 HPO 4 -NaH 2 PO 4 , pH 7.5) containing 50 mM KCl was used.
Author contributions-J. J. and B. N. designed the research and wrote the paper. J. J. conducted experiments and analyzed data together with B. N. All authors agreed on the final version of the paper.