‘Dopamine-first’ mechanism enables the rational engineering of the norcoclaurine synthase aldehyde activity profile

Norcoclaurine synthase (NCS) (EC 4.2.1.78) catalyzes the Pictet–Spengler condensation of dopamine and an aldehyde, forming a substituted (S)-tetrahydroisoquinoline, a pharmaceutically important moiety. This unique activity has led to NCS being used for both in vitro biocatalysis and in vivo recombinant metabolism. Future engineering of NCS activity to enable the synthesis of diverse tetrahydroisoquinolines is dependent on an understanding of the NCS mechanism and kinetics. We assess two proposed mechanisms for NCS activity: (a) one based on the holo X-ray crystal structure and (b) the ‘dopamine-first’ mechanism based on computational docking. Thalictrum flavum NCS variant activities support the dopamine-first mechanism. Suppression of the non-enzymatic background reaction reveals novel kinetic parameters for NCS, showing it to act with low catalytic efficiency. This kinetic behaviour can account for the ineffectiveness of recombinant NCS in in vivo systems, and also suggests NCS may have an in planta role as a metabolic gatekeeper. The amino acid substitution L76A, situated in the proposed aldehyde binding site, results in the alteration of the enzyme's aldehyde activity profile. This both verifies the dopamine-first mechanism and demonstrates the potential for the rational engineering of NCS activity.

NCSs from Thalictrum flavum (TfNCS) and Coptis japonica (CjNCS2) were shown to have a broad aldehyde substrate scope, enabling the enzymatic formation of diverse (S)-tetrahydroisoquinolines (THIQs) [5][6][7][8]. Both enzymes accept most phenylacetaldehydes but do not turn over a-substituted aldehydes. Particularly notable is the turnover of the linear aliphatic aldehydes by CjNCS2 and the large 1-napthylacetaldehyde by TfNCS because these substrates have properties that are very different from those of the natural substrate 4-HPAA. By contrast to the aldehyde, little variation in dopamine structure is tolerated: the 3hydroxy moiety is crucial for successful turnover [6,9].
Progress in the field of synthetic biology has led to the development of a number of microbial BIA pathways which include recombinant NCS. However, it is not clear whether NCS has activity in any of these in vivo settings as a result of the presence of a high non-enzymatic background Pictet-Spengler reaction [10,11]. In some of these synthetic pathways, chirality is established in the final products courtesy of the selectivity of enzymes downstream of NCS [12]. The apparent ineffectiveness of recombinant NCS in vivo contrasts with its natural in planta behaviour, where its removal results in significant reduction of alkaloid production [13]. The non-enzymatic background reaction observed in the recombinant systems is likely to be caused by phosphates: inorganic phosphate is capable of catalyzing the Pictet-Spengler reaction between dopamine and aldehydes, forming racemic THIQs [14]. Consequently, the use of phosphate buffer has adversely affected a number of in vitro investigations into the NCS mechanism and kinetics [1,9,[15][16][17].
Solution of the TfNCS crystal structure enabled the identification of the enzyme active site residues ( Fig. 2A) [16]. An enzyme mechanism for the natural substrates was developed based on the holo crystal structure containing dopamine and a nonproductive aldehyde (Fig. 3A) [18]. This mechanism involves the binding of 4-HPAA to the enzyme prior to dopamine (the 'HPAA-first' mechanism). The mechanism features the aldehyde buried in the active site, and so does not appear to account for the aldehyde promiscuity of the enzyme. Subsequently, an alternative mechanism was developed based on computational docking, in which dopamine binds to the enzyme prior to 4-HPAA (the 'dopamine-first' mechanism) (Figs 3B and 4) [6].
A clearer understanding of the mechanism and kinetics of NCS is required not only to understand the early steps of BIA biosynthesis, but also to further develop NCS as a tool for both synthetic biology and biocatalysis. This knowledge will enable the future rational engineering of the enzyme, leading towards the optimization of the recombinant enzyme for in vivo activity, and also the broadening of the enzyme's substrate scope.
In the present study, we reassess the mechanism and kinetics of NCS; this enables us to rationally engineer the activity of the enzyme, increasing its activity towards an unnatural substrate. First, we compare the two proposed mechanisms: the HPAA-first Overall mechanism of tetrahydroisoquinoline Pictet-Spengler reaction. NCS reaction mechanism as proposed previously [9]. A similar mechanism is postulated for the phosphate catalysed reaction, giving a racemic product [14]. The reaction is initiated by formation of an aminol, followed by elimination of water to form an iminium cation. The subsequent rate-limiting step, deprotonation of the dopamine 3-hydroxy, triggers cyclization onto the iminium. The final, irreversible, step is deprotonation of the quinone 8a-H to form norcoclaurine.
mechanism, based on the holo X-ray crystal structure, and the dopamine-first mechanism, based on computation docking. Amino acid substitutions are used to probe the role of active site amino acids; their activities support the dopamine-first mechanism. Suppression of the non-enzymatic background chemical reaction reveals new kinetic parameters for NCS, showing it to be a catalytically inefficient enzyme with remarkably high apparent K m values. Finally, the substitution L76A in the aldehyde binding site proposed by the dopamine-first mechanism results in the modification of the enzyme's aldehyde activity profile.

Results and Discussion
The holo crystal structure provides ambiguous data The HPAA-first mechanism was based on the observed ligand binding modes in the holo NCS crystal structure. This structure was obtained by soaking NCS crystals with solutions of dopamine and para-hydroxybenzaldehyde (PHB), an electron deficient non-productive 4-HPAA analogue. Subunit A of the holo crystal structure is shown to bind PHB in two different conformations, whereas subunit B binds both dopamine and PHB simultaneously ( Fig. 2A) [16]. The ligands modelled into the active site of the holo crystal structure do not fit the electron density convincingly (Fig. 2B). The density observed in subunit A does not support the presence of two conformations of PHB, as modelled (Fig. 2B, subunit A). In subunit B, a methylene from dopamine does not fit in the density, and oxygen atoms from both ligands are in regions with no density (Fig. 2B, subunit B). Generally, the electron density appears to have insufficient definition to describe the identity of ligands present. This observed ambiguity may be a result of low occupancy, or perhaps a heterogeneous sample.
The difference in the binding behaviour of the two subunits is likely to be a consequence of the different rotameric conformations of Phe112 ( Fig. 2A). Evaluation of the crystal structures revealed that the Phe112 conformation present in subunit B is unusual; it is only present in 1% of phenylalanines in the Protein Data Bank [19]. This analysis is supported by molecular dynamics simulations suggesting that, in solution, the Phe112 conformation in subunit B reverts to that observed in subunit A (Fig. 5).
The differences in the active sites of the subunits may be a knock-on effect from the tight crystal packing. The asymmetric dimer is held together by a nonnative b-strand formed from the first nine N-terminal residues of subunit A in the adjacent dimer (Fig. 6). This N-terminal sequence is not necessary for enzyme activity [20,21] and is possibly part of a cleaved signal peptide (predicted by SOSUISIGNAL) [22].
Overall, the binding modes derived from the holo structure appear to be speculative on the basis of the observed electron density. Proposed HPAA-first mechanism cannot account for NCS activity The HPAA-first mechanism was proposed based on the arrangement of the substrates in subunit B (Fig. 3A) [16,18]. A major drawback of the HPAAfirst mechanism is that it cannot account for NCS reacting with a wide variety of aldehydes. First, the proposed interaction between Asp141 and the phenol of 4-HPAA would not be possible with many aldehydes. Second, it is unlikely that the active site can incorporate large aldehydes (such as 1-napthylacetalde-hyde or citronellal) and dopamine simultaneously in the proposed stacked formation.
In this mechanism, the intermediate imine undergoes a rearrangement which includes an iminium cis-trans isomerization and the rotation of the entire aldehyde R-group. This step is vital for the subsequent ratedetermining cyclization and deprotonation, which establish the chirality of the product. However, there is no evidence for this reorganization: the holo crystal structure could only provide evidence for the initial arrangement of the substrates.   Table 3 respectively. The chemical basis of the mechanism also has limitations. Water is not sufficiently basic (pK b = À1.7) to deprotonate the 3-hydroxy of dopamine in the ratedetermining step; a basic active site residue is required [23]. Finally, the hydrogen bond between the quinone and Tyr108 would not provide electrostatic stabilization to the intermediate, and cannot account for the successful reaction when the 4-hydroxy moiety is not present.

Computational docking reveals dopamine-first mechanism
To explore alternative enzyme mechanisms, reaction intermediates were docked into the active site of TfNCS (2VQ5, subunit A) ( Fig. 7 and Table 1) [6]. This revealed binding modes that form the basis of the dopamine-first enzyme mechanism (Figs 3B and  4). Subunit A appears to be a better candidate for docking studies than subunit B on the basis of molecular dynamic simulations, which suggest the subunit A Phe112 conformation present is prevalent in solution (Fig. 5). The holo crystal structure 2VQ5 (with ligands removed) was used for computational docking because it has a superior resolution to the apo structure 2VNE.
A key factor in the biocatalytic utility of NCS is its ability to turn over a wide variety of structurally diverse aldehydes. Docking calculations provide a structural explanation for this observation: the most favourable binding modes for reaction intermediates show that only the aldehyde R-group is partially exposed to the solvent ( Fig. 4 and Table 1). The poor tolerance of a-substituted aldehydes by NCS may be rationalized on steric grounds because the a-carbon is buried in the active site. Large aldehydes (but with no a-substitutions) may be accepted because the R-group bulk can be in the solvent, away from the enzyme.

Behavior of variants support the dopamine-first mechanism
To probe the validity of the dopamine-first mechanism, the activities of particular NCS variants were investigated. The residues Tyr108, Glu110, Lys122 and Asp141 were selected for substitution based on both previous investigations and the in silico docking results [16]. Initially, two aldehydes, the natural substrate 4-HPAA and the aliphatic hexanal, were selected for investigation, aiming to probe whether any amino acid substitution affected the enzyme activity in a substrate-specific manner.  Table 1. Images generated using Chimera. All variants were judged to be folded under experimental conditions on the basis of their melting points ( Table 2). All enzymes were more active with 4-HPAA than hexanal (Fig. 8). Enzymes, with sufficient activity, were observed to perform the reactions in Hepes buffer in a stereoselective manner [(S)-isomer, > 95% enantiomeric excess] (Fig. 9). The type of buffer used in the non-enzymatic reactions impacted greatly on the conversion. In phosphate buffer, conversions with 4-HPAA and hexanal were 69% and 54%, respectively (racemic product). By contrast, Hepes buffer reached only 5% and 3%, respectively (Fig. 8B).
The variant K122L demonstrated no activity with 4-HPAA or hexanal, supporting the identity of Lys122 as the residue catalyzing the rate-determining cyclization step. Lys122 is able to catalyze this step because of its low pK a (prediction~7.2), which is the result of burying a charge in a hydrophobic environment ( Table 3). The stability increase of 9.1°C conferred by the K122L substitution is a demonstration of the same effect ( Table 2). The pK a2 phosphate, which is able to catalyse the racemic reaction, is also 7.2; this suggests    [16,31]. Both isomers of the aminol, iminium, quinone and iminium-citronellal intermediates were investigated: the wavy bonds show where the stereochemistry varied. For observed docking conformations, see Figure 4 and Table 1. that both Lys122 and phosphate catalyze the same step in the reaction mechanism [24].
Of the two carboxylic acid residues investigated, NCS variant activities show that only Glu110 is necessary for the reaction to proceed. That both a removal of charge (E110Q) and a change in charge position (E110D) eliminate activity supports the suggestion that Glu110 acts as a base, abstracting the quinone 8a-H (Fig. 3B, step e). On the other hand, the behaviours of Asp-141 variants suggest that the residue is not responsible for base catalysis, as, whilst D141N has 10% of WT activity, D141E retains 50% activity. This suggests that Asp141 may provide general electrostatic stabilization, rather than base catalysis.
Tyr108 appears to have a dual role: it not only contributes to the electrostatic properties of the active site, but also defines the shape of the cavity entrance (Fig. 4). This is reflected in a decrease in both rate and apparent K m values in the Y108F variant ( Table 4). The removal of the hydroxy moiety upon amino acid substitution opens up the space around the active site entrance, perhaps aiding aldehyde substrate binding. This change is more pronounced for 4-HPAA than hexanal, as the former has greater steric bulk.

Kinetics reveals NCS operates with low catalytic efficiency
The use of phosphate buffer in previous kinetic experiments with NCS led to the generation of inaccurate kinetic parameters as a result of the non-enzymatic background reaction [1,6,9,[15][16][17]. In this study, we performed all enzyme assays in Hepes buffer, in which there is far less non-enzymatic background (Fig. 8B). However, because of the increased background reaction rate at high substrate concentrations, kinetic parameters were determined using nonsaturating concentrations of the nonvaried substrates.
The apparent kinetic parameters determined in the present study differ considerably from those determined Table 1. Binding modes calculated from in silico docking. Predicted binding affinity and protein-ligand interaction distances of docked reaction intermediates in the NCS active site (2VQ5, subunit A). Conformations predicted by AUTODOCK VINA [31]. For full structures of intermediates and atom numbering information, see Fig. 7. Figure 4 shows the structures of selected binding modes. All binding modes described were the highest ranked by the software.

Ligand
Binding affinity     (Fig. 9). Enzyme melting points were measured showing all enzymes were folded in the assay conditions ( Table 2). Ligand structures are shown in Fig. 7. Docked structures are shown in Figure 4 and described in Table 1. in previous investigations ( Fig. 10 and Table 4). For wild-type (varying dopamine with 4-HPAA), we observed an apparent k cat and K m of 24 s À1 and 22 mM respectively, which are substantially larger than the previously determined values of 4.5 s À1 and 0.4 mM [16]. Also notable is an absence of cooperativity from the kinetics, with respect to both dopamine and aldehydes (Fig. 10). This is in accordance with the observation that D29TfNCS is monomeric at concentrations below 10 lM [21]. The kinetic parameters determined here describe an enzyme that is catalytically inefficient. NCS exhibits an apparent k cat /K m of 1.0 s À1 ÁmM À1 , which is 100-fold below the median of 125 s À1 ÁmM À1 calculated across all enzymes (Table 4) [25]. The K m,app of dopamine with 4-HPAA is 22 mM, showing that high substrate concentrations are required for the enzyme to demonstrate significant turnover. With this in mind, it is expected that NCS operates in vivo with substrate concentrations far below a saturating level. For comparison, absolute intracellular metabolite concentrations calculated in Escherichia coli showed only three of 103 primary metabolites had concentrations above 10 mM, and over 70% had concentrations below 1 mM [26]. This observed catalytic inefficiency may explain the poor activity of recombinant NCS in vivo [10,11].
Plant secondary metabolism enzymes often display low catalytic efficiency [27]. This may be a result of low Table 4. Apparent kinetic parameters for TfNCS. Corresponding velocity/substrate concentration curves can be found in Fig. 10. The assay methodology is as described in the Materials and methods. WT, wild-type.   Table 4. The substitution L76A causes change in aldehyde activity profile The ability to rationally engineer NCS activity will enable the facile biocatalytic syntheses of diverse (S)-THIQs. The dopamine-first mechanism predicts that Leu76 is proximal to the aldehyde R-group (Fig. 4) and so, based on that mechanism, it was hypothesized that amino acid substitutions of Leu76 would show modified activities with different aldehydes. A change in the aldehyde activity profile of a Leu76 variant would validate the dopamine-first mechanism because the aldehyde binding site for the HPAA-first mechanism is in a different location. The substitution L76A improves the activity of NCS towards both (S)-and (R)-citronellal but reduces activity towards 4-HPAA and hexanal (Table 4). This change is most notable with (S)-citronellal where the apparent k cat is doubled both with respect to dopamine and the aldehyde. Interestingly, apparent K m values are largely unaffected by the substitution. The data may also suggest that citronellal inhibits the enzyme at high concentrations ( Fig. 10 and Table 4), which would be in accordance with the NMR observation that dopamine does not bind to the enzyme subsequent to aldehyde binding [21].
Docking calculations of the imine-citronellal reaction intermediates suggest a molecular origin to this activity increase (Fig. 11). The methyl groups of the (S)-and (R)-citronellal intermediates are relatively close to Leu76: at 3.65 and 3.75 A respectively. Removing the steric bulk close to this methyl moiety may enable greater conformational freedom in the intermediate, entropically reducing the reaction energy barrier. The enhancing effect of the amino acid substitution may be more pronounced for (S)-rather than (R)-citronellal simply as consequence of proximity of the methyl group to Leu76.
The substitution L76A has modified the aldehyde activity profile of NCS. This supports the dopaminefirst reaction mechanism and also demonstrates the potential for engineering NCS towards other substrates and activities.

Conclusions
In the present study, we have described and assessed two different mechanisms of NCS activity: the HPAAfirst mechanism and the dopamine-first mechanism. Activities of TfNCS variants appear to support the dopamine-first mechanism. By avoiding the use of phosphate buffer in enzyme assays, we observed novel kinetic parameters that show NCS operates with low catalytic efficiency. This can account for its apparent ineffectiveness in recombinant in vivo systems, and may indicate a possible in planta role as a BIA gatekeeper enzyme. The amino acid substitution L76A, in the proposed dopamine-first aldehyde binding site, resulted in a change in the aldehyde activity profile, strongly supporting the mechanism in question.
Further experiments are required to fully characterize NCS behaviour both in vitro and in vivo. For example, double-reciprocal plots with an appropriate inhibitor will provide a kinetic assessment of the compulsory-order binary mechanism proposed in the present study. It may also be possible to monitor the interaction between NCS and substrates via direct methods such as NMR [21]. Future progress with in planta proteomics and metabolomics may reveal more details about the involvement of NCS in the control of flux into the BIA pathway [28]. The work reported from the present study has wide implications: not only have we gained insight into the early steps of in planta BIA biosynthesis, but also we have demonstrated that it is possible to modify the aldehyde activity profile of NCS via amino acid substitution. Our determination of the NCS mechanism will give rise to the future generation of NCS variants capable of catalyzing the stereoselective formation of diverse THIQs.

Molecular Dynamics
Molecular dynamics simulations were conducted in GRO-MACS using the CHARMM27 forcefield [29,30]. Each subunit of 2VQ5 was prepared by removing waters, ligands and Nand C-termini. The protein subunits were placed in a cube, then water (tip3p) and counter ions were added. The systems were energy minimized (F max < 250 kJÁmol À1 Ánm À1 ). This was followed by NVT equilibration (100 ps) and NPT equilibration (100 ps). Three simulations for each subunit ran for 100 ns, at 300 K. The Phe112 conformation was determined by calculating the angle between the Ca and Cf of Phe112 and the Ca of His106. A value for this angle was calculated for every tenth frame.

Docking
Potential reaction intermediates (Fig. 7) were docked into the active site of subunit A from the TfNCS crystal structure 2VQ5 (Table 1) [16]. Docking was performed using AUTODOCK VINA (exhaustiveness = 8) [31]. Unless noted otherwise, only the top ranked binding modes were used. pK a predictions PROPKA, version 3.1 [32] was used to predict pK a values of residues in the enzyme active site. The TfNCS enzyme structure model was obtained from crystal structure 2VQ5 subunit A [16]. Protein-ligand structures derived from docking calculations (see above) were also analyzed in this manner.