Active-Site Oxygen Accessibility and Catalytic Loop Dynamics of Plant Aromatic Amino Acid Decarboxylases from Molecular Simulations

Aromatic amino acid decarboxylases (AAADs) are pyridoxal-5′-phosphate (PLP)-dependent enzymes that catalyze the decarboxylation of aromatic amino acid l-amino acids. In plants, apart from canonical AAADs that catalyze the straightforward decarboxylation reaction, other members of the AAAD family function as aromatic acetaldehyde synthases (AASs) and catalyze more complex decarboxylation-dependent oxidative deamination. The interconversion between a canonical AAAD and an AAS can be achieved by a single tyrosine-phenylalanine mutation in the large catalytic loop of the enzymes. In this work, we report implicit ligand sampling (ILS) calculations of the canonical l-tyrosine decarboxylase from Papaver somniferum (PsTyDC) that catalyzes l-tyrosine decarboxylation and its Y350F mutant that instead catalyzes the decarboxylation-dependent oxidative deamination of the same substrate. Through comparative analysis of the resulting three-dimensional (3D) O2 free energy profiles, we evaluate the impact of the key tyrosine/phenylalanine mutation on oxygen accessibility to both the wild type and Y350F mutant of PsTyDC. Additionally, using molecular dynamics (MD) simulations of the l-tryptophan decarboxylase from Catharanthus roseus (CrTDC), we further investigate the dynamics of a large catalytic loop known to be indispensable to all AAADs. Results of our ILS and MD calculations shed new light on how key structural elements and loop conformational dynamics underlie the enzymatic functions of different members of the plant AAAD family.


■ INTRODUCTION
−3 In plants, these enzymes are involved in the biosynthesis of secondary metabolites, 4 i.e., compounds that are not essential for growth or development but render plants competitive in their own environment.While mammals have only a single AAAD that catalyzes the decarboxylation of aromatic amino acids, the plant AAAD family has undergone extensive evolutionary diversification, giving rise to a series of paralogous enzymes with different substrate preferences and catalytic mechanisms. 3For instance, tryptophan decarboxylases (TDCs) and tyrosine decarboxylases (TyDCs), known as canonical AAADs, catalyze the straightforward decarboxylation of tryptophan (Trp) and tyrosine (Tyr), respectively.Two other members of the plant AAAD family, phenylacetaldehyde synthases (PAASs) and 4-hydroxyphenylacetaldehyde synthases (4HPAASs), known as aromatic aldehyde synthases (AASs), catalyze the more complex decarboxylation-dependent oxidative deamination 5 of phenylalanine (Phe) and Tyr, respectively.
PLP-dependent enzymes primarily participate in the biosynthesis of amino acids and amino acid-derived metabolites. 6rivaled in the versatility of reactions they catalyze, these enzymes are estimated to collectively account for ∼4% of all enzyme activities. 7Despite their diverse reaction types, PLPdependent enzymes employ a unified mechanism where the cofactor acts to stabilize the negative charge on the substrate C α atom developed during the reaction transition state. 6pecifically, the PLP cofactor is initially linked to the ε-amino group of a conserved active-site lysine, forming an internal aldimine (LLP, see Figure 1); upon binding of an aminecontaining substrate, LLP undergoes a Schiff base exchange reaction (transimination), during which a substrate-PLP linkage replaces the enzyme-PLP one, and the resulting external aldimine serves as a common intermediate for all PLP-dependent reactions.At this point, divergence in reaction types begins. 1,8For decarboxylases, the bond between the C α atom and the carboxylate group is cleaved, producing CO 2 and a carbanion, with the negative charge of the latter stabilized by PLP in the form of a quinonoid intermediate.−11 Mutagenesis and structural studies have established the critical role of a conserved tyrosine in AAADs. 11,12Upon mutation of this tyrosine to a phenylalanine, canonical plant AAADs are converted into AASs.Conversely, a Phe-to-Tyr mutation converts AASs to canonical AAADs.Notably, the decarboxylase activity of the resulting mutant is even faster than the aldehyde synthase activity of the wild-type (WT) AAS. 11A single Tyr-to-Phe mutation has also been reported to turn the mammalian AAAD L-Dopa decarboxylase (DDC) and the human histidine decarboxylase (HDC) into acetaldehyde synthases. 12,13Such a dramatic impact of a single amino acid has been attributed to the role of tyrosine as a proton donor to the reaction intermediate: this key tyrosine in decarboxylases acts to protonate the quinonoid intermediate; when it is replaced by a phenylalanine that cannot serve as a proton donor, molecular oxygen steps in, forcing the quinonoid intermediate to undergo oxidative deamination.
The importance of the key tyrosine and the additional role played by a conserved histidine is further illustrated in a recent work that combines structural and functional studies with molecular dynamics (MD) simulations of plant AAADs. 14In this work, X-ray crystal structures of four plant AAADs were resolved, including Catharanthus roseus TDC (CrTDC), Papaver somniferum TyDC (PsTyDC), Arabidopsis thaliana PAAS (AtPAAS), and Rhodiola rosea 4HPAAS (Rr4HPAAS).All four enzymes were found to be homodimers with two symmetric active sites located at the dimer interface (Figure 1).The key Tyr residue (Y348 in CrTDC or Y350 in PsTyDC) was located on a large loop that had been missing in X-ray crystal structures of mammalian DDCs with the exception of the recently resolved human aromatic amino acid decarboxylase. 15In CrTDC, this loop adopted an open conformation, exposing the active site.In PsTyDC, this loop adopted a closed conformation that sealed the active site, although a small portion (residues 354−359) of the loop was unresolved.Another important structural difference between the two canonical AAADs involved a small loop harboring a conserved histidine (H203 in CrTDC or H205 in PsTyDC, see Figure 1).In CrTDC, this small loop was rotated outward, while in PsTyDC, it adopted a conformation that enabled the aforementioned histidine to form a π−stacking interaction with LLP's pyridine ring.Our previous MD simulations spanning hundreds of nanoseconds 14 showed that initiated from the CrTDC-crystal structure in the open state, the large loop could undergo a swing motion that brought the key tyrosine within hydrogen bonding distance of the small-loop histidine.Based on the above structural and dynamic data as well as enzyme assays and transgenic yeast expressing WT, Y350F, or H205N mutant of PsTyDC, the small-loop histidine was identified as an important partner of the large-loop tyrosine, whereby the former amino acid facilitated the latter's proton transfer to the quinonoid intermediate. 14hile the above study has shed new light on the mechanistic basis of the functional divergence within the AAAD family, much remains to be learned about these enzymes.For instance, the Tyr-to-Phe mutation has been suggested to produce a larger cavity due to the absence of the hydroxyl group in phenylalanine, which may in turn facilitate the entrance of molecular oxygen into the active site. 14Given that the presence of O 2 at the active site is a prerequisite of AASs' oxidative deamination activity, it is conceivable that AAADs and AASs may have utilized the different sizes and hydrophobicities of the two amino acids to achieve differential oxygen accessibility within their active sites.Whether this is indeed the case, however, cannot be solely determined from static structural information.Another feature of the plant AAAD family that remains to be further investigated is the dynamics of their large loops that harbor the conserved tyrosine.Our previous work on CrTDC demonstrated considerable flexibility of this large loop, revealing its transition from an open to a semiclosed state in ∼500 ns MD simulations.However, the limited time span of these calculations and the lack of counterpart simulations initiated from the closed state hindered sampling of the loop conformational space and estimation of the characteristic correlation time of loop movement.Here, we report implicit ligand sampling (ILS) calculations on the closed-state PsTyDC in both the wild-type enzyme and its Y350F mutant.Contrary to the aforementioned hypothesis, the three-dimensional (3D) free energy profiles from ILS performed on both apo-and holo-forms of PsTyDC reveal similar pathways and comparable energy barriers of molecular oxygen entering the active sites of the WT, canonical AAAD, and its Y350F mutant that has been converted into an AAS.We further report microsecond MD simulations of CrTDC initiated from both the open and closed states of the large loop, demonstrating a considerably broader

Biochemistry
sampling of loop conformations than those recovered from our previous, hundreds of nanoseconds simulations.Based on these results, we discuss the impact of the key tyrosine/ phenylalanine on differentiating the reaction specificities of canonical AAADs and AASs, the correlation between loop secondary structure and dynamics, and sampling consideration for future modeling studies of the plant AAAD family.
■ METHODS Protein Structure Preparation.The crystal structures of CrTDC and PsTyDC were obtained from the Protein Data Bank (PDB codes: 6eew, 6eem 14 ).The large catalytic loop, in which the key tyrosine is located, adopts an open conformation in CrTDC (residues 342−361) and a closed conformation in PsTyDC (residues 344−363), where a small segment of the PsTyDC loop (residues 354−359) is missing.To construct the closed-state CrTDC, we turned to AlphaFold2 (AF2), 16 which was also used to model the missing segment in the PsTyDC loop.As shown in Figure 2, apart from the region around the missing loop segment, the AF2 prediction of PsTyDC is nearly indistinguishable from its crystal structure.The predicted missing segment, which forms a short α-helix, resembles the corresponding segment in the crystal structure of human HDC (data not shown).For CrTDC, the AF2-predicted closed conformation is highly similar to that of the PsTyDC crystal structure (Figure 2).To construct MD simulation systems, monomeric structures of CrTDC and PsTyDC obtained from AlphaFold Protein Structure Database 17 were aligned to their crystal structures in order to produce the corresponding homodimers.A CrTDC residue distant from the active site, Gly401, is replaced by an alanine in the crystal structure (PDB code: 6eew).As the crystal and AF2 structures were nearly indistinguishable except for the large loop, the CrTDC-crystal model was built with its large loop from the crystal structure and the rest of the protein from the AF2 structure so that Gly401 was consistently adopted in both of our CrTDC models.While most residues were found in their default protonation states, several were modeled in their neutral forms based on PROPKA-3.1 18,19 results.These residues are Asp268-A/B, Asp397-A/B, Lys208-A/B, and Glu169-A for CrTDC, and Asp270-A/B and Lys379-A/B for PsTyDC.The single amino acid mutants (PsTyDC-Y350F) were constructed using PyMOL-Mutagenesis Wizard. 20Unless otherwise stated, all simulation systems were constructed with the proteins in their apo-states, i.e., without the substrate or PLP cofactor.

MD Simulations and Analysis.
To prepare for O 2 accessibility calculation via ILS, five replicas of 100 ns plain MD simulations of apo-PsTyDC were first carried out for the WT protein or the Y350F mutant using NAMD2.14 21and the CHARMM36 force field. 22Apart from apo-PsTyDC, five replicas of 100 ns plain MD simulations were also performed for a holo-PsTyDC in complex with both the internal aldimine LLP and the substrate Tyr.−25 In enzymes with a strategically placed acidic residue (Asp289 in PsTyDC), a strong hydrogen bond involving the pyridine nitrogen favors the ketoenamine form of LLP by shifting the proton in its intramolecular hydrogen bond from the phenolic oxygen to the aldimine nitrogen. 23owever, unlike the enolimine form that has a good analogy with existing compounds in the latest CHARMM General Force Field (CGenFF, version 4.6) as reflected in its low penalty values of predicted parameters, the ketoenamine form lacks such a good analogy (data not shown) and the protein environment that stabilizes this tautomer cannot be readily included in its parametrization protocols.For this reason, we adopted the enolimine form of LLP in the ILS calculations, along with a neutral pyridine nitrogen and a neutral Asp289 (Figure S1A). 23The substrate Tyr is modeled in its zwitterionic form, mimicking a state prior to the formation of the Michaelis complex. 26The enolimine form of LLP was parametrized using CGenFF 27−29 with the conjugated lysine 321 atom slightly adjusted in its atomic partial charge to render the overall charge of the molecule an integer.Finally, five replicas of 100 ns plain MD simulations were also performed for a second holo-PsTyDC in complex with the external aldimine (Figure S1B) that has been crystallized in one of its monomers. 14Parameters for this external aldimine were again generated by CGenFF and its high-penalty charges and dihedrals were optimized using the Force Field Toolkit plugin of VMD. 30 Final parameters of LLP and external aldimine used in our simulations along with their corresponding structures are available at GitHub repository: https://github.com/TianjieLi-Jason/PsTyDC.git.The apo-and both holo-PsTyDC systems were solvated in a water box with 0.15 M NaCl to neutralize the systems, resulting in a total of ∼74,000 atoms.A 12 Å cutoff was applied for nonbonded interactions, while

Biochemistry
long-range electrostatics were computed using the particle mesh Ewald (PME) method 31,32 with a grid spacing of 1 Å.The temperature was maintained at 300 K by Langevin dynamics and the pressure was kept at 1 atm by Nose−Hoover Langevin piston. 33,34To maintain the closed-state structure of PsTyDC, positional restraints with a force constant of 5 kcal/ (mol•Å 2 ) were applied to the C α atoms of its large catalytic loop as well as the α-helices and β-sheets of the rest of the protein.In holo-PsTyDC, the positional restraints were also applied to the C α atoms of LLP and Tyr as well as the external aldimine.All simulation trajectories were written every 10 ps, which were then utilized in subsequent ILS calculations.
All CrTDC simulations were performed using GROMACS 2022.5 35 with the CHARMM36 force field. 22To distinguish simulations initiated from the loop-open conformation of the CrTDC-crystal structure and the loop-closed conformation of its AF2 prediction, hereon we refer to these two sets of simulations by their initial structures (CrTDC-crystal and CrTDC-AF2, respectively).Each initial dimer structure was solvated in a dodecahedron box with TIP3P water, and 0.05 M NaCl was added to neutralize the system (Figure S2), resulting in a total of ∼120,000 atoms.After a 2000-step energy minimization, the systems underwent a 1 ns NVT and a 1 ns NPT equilibration successively.Subsequent 8 μs production simulations were carried out in the NPT ensemble in two replicas for CrTDC-crystal and CrTDC-AF2, respectively, resulting in a total of 32 μs simulations for these two systems.The temperature was coupled to 300 K using velocity rescaling, 36 while the pressure was coupled to 1 bar using the c-rescaling method. 37Bonds involving hydrogen atoms were constrained using the LINCS algorithm. 38,39Nonbonded interactions were cut off at 12 Å, and the PME method 31,32 was applied for electrostatic interactions.To analyze loop conformations from MD trajectories, the longest principal axis of the CrTDC dimer was aligned with the z axis using VMD. 40The minimum distance d between atoms of Tyr348 and His203 in the same active site was computed by using GROMACS (gmx mindist).The principal component analysis (PCA) was performed on the C α atoms of the large loop by using gmx covar and gmx anaeig.Clustering analysis was conducted on combined trajectories of aligned CrTDC monomers using the root-mean-square deviation (RMSD)based GROMOS method with a cutoff of 6.2 Å.In all of the analysis regarding loop dynamics, the two monomers in a CrTDC dimer were treated equivalently.Secondary structural contents of the large catalytic loop were calculated using the DSSP program (version 4). 41,42The total percentage of residues classified as α-helix or 3-helix was taken as the loop's α-helical content.
Implicit Ligand Sampling (ILS).Instead of simulating actual oxygen migration events, the ILS approach makes use of the small size of O 2 to conduct a single-step free energy perturbation (FEP) in a postprocessing manner, yielding a complete 3D free energy map of the gas molecule within a given system. 43,44Specifically, the 3D potential of mean force (PMF) of a gas molecule is estimated as where ΔE m,k is the interaction energy of adding a gas molecule at position r with a given orientation (k) in simulation frame m (the explicit dependence of ΔE m,k on protein and solvent coordinates as well as ligand orientation is omitted for clarity), N stands for the total number of simulation frames, and C represents the number of ligand orientations explored at each frame. 44The resulting 3D PMF depends on the gas molecule's position r only, while its orientational degrees of freedom have been averaged out.With ΔE m,k = 0 in a vacuum, the thus obtained free energy value represents the cost of moving the gas molecule from a vacuum to the given location.For instance, the free energy value obtained for the bulk water region corresponds to the solvation free energy of the gas molecule.The difference between the free energy value obtained within a protein cavity and this solvation free energy can then be utilized to analyze the energetic cost, if any, of transferring the gas molecule from bulk water to the given position within the protein.Further details of the ILS method can be found in the work of Cohen et al. 43,44 The VolMap command in VMD was used to conduct the ILS calculation, where a total of N = 10000 snapshots were extracted from each 100 ns MD simulation of the WT or Y350F mutant of PsTyDC.The ILS calculation was performed at 300 K with a grid spacing of 1 Å.The Lennard-Jones parameters for molecular oxygen were σ = 1.7 Å and ε = −0.12kcal/mol. 43At each grid point, the free energy of placing an O 2 molecule was computed by setting the parameter orient to 7, which generated 21 probe orientations after exploiting the C2 rotary symmetry of an oxygen molecule.Further increasing the values of parameter orient to 8 or 9 did not significantly improve the ILS result (data not shown).For each of the five replicas of 100 ns simulations used in ILS calculations, analysis of the two PsTyDC monomers was performed separately, yielding n = 10 samples of O 2 free energy barriers in either apo-, holo-(LLP + substrate), or holo-(external aldimine) PsTyDC.Additionally, in each simulation replica, a region far from the protein was selected, yielding altogether n = 10 (five from WT and five from Y350F mutant) estimation of the free energy cost of placing O 2 in the bulk water.Statistical significance of the difference between two systems, e.g., apo-WT and apo-Y350F mutants, was assessed from all samples of each system using a two-sample t-test.All reported errors represent standard deviation.The propagation of error was achieved by taking the square root of the sum of the individual errors.In addition to ILS calculation, 3D water occupancy maps were computed using the same MD simulation trajectories at a resolution of 1 Å by the VolMap plugin of VMD.

■ RESULTS
Active-Site O 2 Accessibility in WT and Y350F Mutant of PsTyDC.Mutagenesis experiments have shown that substituting a tyrosine on its large catalytic loop by a phenylalanine is sufficient to transform a canonical TyDC into an AAS, reflecting the key role of the conserved Tyr residue. 11The removal of a hydroxyl group upon the Tyr-to-Phe mutation enlarges the active-site cavity, which has been hypothesized to facilitate molecular oxygen to occupy the position originally taken by this proton-donating group. 14ere, in order to quantify O 2 accessibility to the active site, we determined its 3D free energy profiles in both the WT enzyme and the Y350F mutant of PsTyDC via implicit ligand sampling (ILS).To comprehensively explore potential pathways utilized by O 2 to enter the active site, ILS calculations were performed on five replicas of 100 ns simulations for each of the apo-and two holo-forms of PsTyDC, with the latter in complex with Biochemistry either the internal aldimine LLP and substrate, or the external aldimine (Figures S1 and S3).As detailed in the Methods Section, ILS takes advantage of the relatively weak interaction between a gas molecule and its environment to conduct a single-step FEP calculation. 43A plain MD simulation of the solvated protein, without any O 2 , is conducted and ILS is performed in a postprocessing manner to extract the free energy cost of virtually placing a molecular oxygen at a given position of a 3D grid spanning the simulation system, i.e., treating O 2 as an implicit ligand.The computed free energy value at each point of the 3D grid reflects the probability of finding an oxygen at that position relative to the vacuum (set to have a free energy value of 0 kcal/mol).Isosurfaces connecting points with the same free energy can be used to highlight migration pathways that the gas molecule may take to enter the protein active site from the bulk solution.
Our ILS calculations revealed two major entrance pathways of oxygen into the active sites of both the WT protein and the Y350F mutant of PsTyDC, the representative snapshots of which are shown in Figure 3A,B.One of the two pathways (labeled a) is near the side chain of the key Tyr350 or its Phe350 substituent in the mutant protein, while the other (labeled b) is on the "back" of the first pathway and exits the protein surface near the loop residue Val351.By comparing the maximum free energy value along each pathway, we obtained the minimum free energy barrier that an oxygen molecule must  (A, B)) from the exterior of the protein to the active site in apo-PsTyDC, holo-PsTyDC with LLP and substrate, and holo-PsTyDC with external aldimine, respectively.The difference between WT and Y350F lacks statistical significance at the 0.05 level (denoted by ns) in all three simulation systems.

Biochemistry
overcome in order to take the "easier" path of the two and enter the active site.From five replicas of 100 ns plain MD simulations of each apo-or holo-PsTyDC dimers, we obtained 10 ILS maps that yielded the energy barriers as scattered dots in Figure 3C.Their mean and standard deviation are listed in Table S1, along with the O 2 solvation free energy computed for bulk water from the corresponding simulations.Based on the two-sample t-test (see the Methods Section), no statistically significant difference (at the 0.05 level) is found in any of the WT vs Y350F mutant comparison, i.e., they present similar barriers to O 2 permeation in apo-as well as both holo-forms investigated here (Figure 3C).Our computed O 2 solvation free energy (2.05 ± 0.02 kcal/mol) is in good agreement with previous ILS calculations (1.97 ± 0.02 kcal/mol) 43 performed at the same temperature (300 K) and comparable with the experimental value measured at 20 °C (1.78 kcal/mol). 45ubtracting this computed solvation free energy from the ILS results of a given PsTyDC dimer yields the energy barrier that an O 2 molecule must overcome in its migration from bulk solution into the protein's active site.The thus obtained barriers of the WT PsTyDC and Y350F mutant are 0.89 ± 1.02 and 0.65 ± 1.06 kcal/mol, respectively, the difference between which is again insignificant according to the two-sample t-test.Furthermore, although these barrier heights have statistically significant (P < 0.05) differences from zero, their values are comparable to the thermal energy k B T (∼0.60 kcal/mol at 300 K), suggesting that they can be readily overcome by a diffusing oxygen molecule.The above results indicate that molecular oxygen encounters similarly small barriers entering the active sites of the WT and Y350F PsTyDC.In other words, to this gas molecule, both enzyme active sites are nearly as readily accessible as the bulk solution.
In both the WT protein and Y350F mutant, neither pathway a nor b consistently dominates over the other to always represent the "easier" path for O 2 migration.This result and the aforementioned small energy barrier values suggest that the "breathing motion" of PsTyDC readily opens an entrance path, one way or the other, for the small gas molecule.Further analysis of the 3D O 2 free energy profiles identifies residues lining each of the two major pathways: pathway a is guarded by the key tyrosine (or its phenylalanine substitute), as well as His205 and Cys206.The narrowest segment of pathway b is

Biochemistry
lined by a number of aromatic residues, including Trp90, Phe122, and Tyr98 as well as the hydrophobic Val351 that neighbors Tyr350.Overall, the absence of a large energy barrier against the entrance of oxygen to the PsTyDC active site may be attributed to the relatively short length of pathway a and the largely hydrophobic nature of pathway b.Comparison with the 3D water occupancy maps obtained from the same simulations suggests that these two pathways are at least partially solvated, albeit to different extent (Figure S3).
Dynamics of the Large Catalytic Loop from CrTDC.In order to serve as the proton donor to the quinonoid intermediate, the key tyrosine in canonical AAADs must be positioned near the substrate for which the large catalytic loop needs to adopt its closed conformation.In the crystal structure of CrTDC, the large loops in both monomers are in an open state.Previously, we conducted 36 sets of 100 ns MD simulations (with one of them extended to ∼500 ns) on this structure to explore the initial transition of the loop from an open to a semiclosed conformation, which was found to be independent of the protonation states of the substrate and LLP. 14However, the relatively short simulation time and the sole initial structure from which the simulations were launched limited the sampling of loop dynamics.In this work, we performed multimicrosecond MD simulations of apo-CrTDC initiated either from its open state revealed by the crystal structure (CrTDC-crystal) or from a closed state predicted by AlphaFold2 (CrTDC-AF2) that resembles the crystal structure of the closed-state PsTyDC (Figure 2).As our goal is to extract characteristic structural and dynamic features in the large loop's exploration of its conformational space (as opposed to being locked in a given state), the apo-form of CrTDC in the absence of substrate and LLP was employed, since the loop was found to be highly mobile in the apo-form from our previous study. 14Specifically, two replicas of 8 μs MD simulations were initiated with either the open (CrTDC-crystal) or closed (CrTDC-AF2) state of the loop as its initial structure.With both monomers in the homodimeric enzyme treated equivalently, these simulations collectively provided 64 μs trajectories for the large catalytic loop that were subsequently analyzed.
To quantitatively characterize conformations of the large loop in CrTDC, we performed principal component analysis (PCA) on the combined simulation trajectories initiated from its crystal structure and AF2 prediction.The first two principal components (PC1 and PC2, Figure 4A,B) captured approximately 72% of the variance in loop movement, which then served as two basis vectors onto which the simulation trajectories were projected.The top six clusters from subsequent clustering analysis collectively accounted for 84% of the loop conformations, the projections of which onto PC1 and PC2 were also indicated in Figure 4A,B.To further assist the analysis of loop conformations, we computed two additional metrics, namely, the minimum distance d between atoms from the key residue Tyr348 and its proton transfer partner His203, as well as an opening angle θ that measured the extent to which the loop opened from the mouth of the active pocket (Figure 5A,B).Specifically, θ was defined as the angle between the vector from Leu342 to Tyr348 and the vector from Leu342 to the z-axis projection of residues 342− 361 C α atoms' center according to the CrTDC-AF2 structure (Figure S4).The Tyr348-His203 distance d and the loop opening angle θ therefore complement the PCA and clustering analysis in characterizing the diverse conformations sampled by the large loop during the aggregated tens of microsecond MD simulations.
Initiated from either the AF2 (black square) or the crystal structure (black triangle), the CrTDC large loop explored a wide range of conformations, displaying a broad distribution over the 2D coordinate space spanned by PC1 and PC2 (Figure 4A,B).Overall, the range of conformations sampled by simulations initiated from the CrTDC-AF2 structure appeared

Biochemistry
to be slightly broader than those initiated from the CrTDCcrystal structure, as evidenced by the former's wider coverage of the PC1-PC2 (Figure 4A) as well as the θ−d coordinate space and the individual distributions in θ and d (Figure S5).Their difference in these profiles is a clear indication that simulations initiated from the two states each sampled the loop conformational space to a different extent.While individually they were insufficient to sample the entire conformational space of the catalytic loop well, collectively these simulations could be used to explore characteristic structural and dynamic features of this space.As shown in Figure 4B,C, highly populated loop conformations emerged as dark regions with low free energy in the 2D PC1-PC2 space and can be identified by RMSD-based clustering analysis.Centroid structures obtained from the clustering analysis overlapped reasonably well with the low free energy regions on the PC1-PC2 map.Clusters 1 and 2, which represented ∼35% and ∼15% of the simulation trajectories, fell in the vicinity of the closed (CrTDC-crystal) and open (CrTDC-AF2) state of the large loop, respectively, although a clear distinction with the corresponding initial state could be found in both centroid structures.Specifically, compared to its closed state in the CrTDC-AF2 structure, the large loop in cluster 2 lost the precise positioning of Tyr348 (Figure 4C), likely due to the absence of substrate and LLP.Interestingly, this cluster centroid had an even smaller θ than the initial CrTDC-AF2 structure (θ = 37.7°vs 45.9°), suggesting that the loop capped the active site even more "deeply" in this conformation.Compared with the CrTDC-crystal structure, cluster 1 appeared to be slightly "more open" (θ = 113.6°vs100.3°), a difference that may have arisen from the considerable flexibility of the large loop in the apoenzyme.Such loop flexibility is consistent with our previous simulations 14 of CrTDC and has been examined in depth by a recent study on the human aromatic amino acid decarboxylase. 15Previous simulations on the enzyme triosephosphate isomerase have revealed that a highly flexible loop may adopt many different open states. 46Here, based on their locations on the 2D PC1-PC2 map as well as comparison with the CrTDC-AF2 and CrTDC-crystal structures (Figures 4C and S5B,C To evaluate the challenge posed by the vast conformational space accessible to the CrTDC catalytic loop, we determined the correlation time τ of the angle θ and distance d from each independent replica of 8 μs MD simulations.As shown in Table S2, the large fluctuation in τ and its hundred−nanosecond mean clearly point at the challenge of sampling the loop conformation via plain MD simulations.This issue can be exacerbated by the correlation between loop secondary structure and its dynamics.As shown in Figure 4C, a small portion of the large loop (residues 346 to 350) formed a short α-helix in both cluster 1 and the CrTDC-crystal structure.With the exception of cluster 4, the loop completely lost its helical content in the remaining conformations captured by clusters 2 to 6.Further analysis revealed no statistically significant difference in the average loop α-helical content between the four simulations initiated from CrTDC-AF2 and those initiated from the CrTDC-crystal structure (Figure 5).Nonetheless, for an individual simulation where the loop had a high α-helical content, sampling over the PC1-PC2 space was found to be markedly reduced (Figure S6), in line with the previously reported hindrance on loop movement posed by this short helix. 14

■ DISCUSSION
Given its electron-sink nature, PLP is known to slowly catalyze many reactions in the absence of an enzyme. 1,6Apart from enhancing this innate catalytic power of the cofactor, a key task of all PLP-dependent enzymes is to enforce substrate and reaction-type specificities through carefully selected and positioned active-site residues.On the control of reaction type, the Dunathan hypothesis 47 explains how the substrate orientation relative to the PLP pyridine ring determines which bond to C α is broken upon formation of the external aldimine.The reaction specificity control does not end here: as the quinonoid intermediate can react in a number of possible pathways, the enzyme must continue to govern subsequent steps to promote the desired reaction while minimizing the unwanted, side ones. 6,26Different strategies may be devised, depending on the competing reactions at play.For instance, alanine racemase has been shown to selectively destabilize the quinonoid intermediate to promote racemization over the transamination side reaction. 8For the mammalian DDC, O 2involved oxidation is known to be a side reaction. 48,49In plants, however, AASs have evolved to fully couple decarboxylation with oxidative deamination, making it their primary reaction instead. 5,9−11 A number of factors, including the ease of O 2 access, 5,14 have been hypothesized to play a role in the above switch of reaction-type specificity between the two categories of closely related enzymes.The experimentally demonstrated conversion of a canonical plant AAAD to an AAS upon a tyrosine to phenylalanine mutation (and vice versa) further focuses one's attention on the impact of this single amino acid switch.
Our ILS calculations of the closed-state PsTyDC reveal O 2 migration pathways in both the WT and Y350F mutant protein, allowing for detailed analysis of oxygen accessibility to their active sites.The difference in the free energy cost of the entrance of the O 2 into WT PsTyDC and its Y350F mutant is found to be statistically insignificant.Furthermore, the barrier heights in both proteins are comparable with the thermal energy, indicating that their active sites can be readily accessed by a diffusing oxygen molecule.These results suggest that oxygen accessibility is unlikely a significant contributor to different reaction-type specificities of the canonical plant AAADs and AASs.In other words, the dominance of decarboxylation over oxidation in WT PsTyDC or vice versa in the Y350F mutant does not simply arise from their distinction in O 2 accessibility.The conversion of a canonical AAAD to AAS upon the Tyr-to-Phe mutation can therefore be primarily attributed to the proton-donating ability of the former amino acid and the lack of such an ability of the latter: replacing the tyrosine by phenylalanine leaves the quinonoid intermediate without a proton donor, giving the oxygen molecule an opportunity to attack and direct the reaction toward oxidative deamination.Based on the ease of the gas Biochemistry molecule entering the PsTyDC active site revealed by our ILS calculations, a competition between protonation by Tyr350 and oxidation by O 2 should always be present once the quinonoid intermediate is produced.This suggests that the former reaction is likely faster than the latter and therefore kinetically favored by the WT PsTyDC.Such a difference would be consistent with kinetic measurement in other plant canonical AAADs and AASs: on the one hand, the Y348F CrTDC mutant is 17 times slower in its aldehyde synthase activity than the decarboxylase activity of the WT protein; on the other hand, the F338Y mutant of AtPAAS is found to be 3fold faster as a converted decarboxylase than the original, WT aldehyde synthase. 11Mechanistically, oxygen consumption by PLP-dependent enzymes has yet to be fully understood.The normal, triplet state of O 2 is more stable than its singlet state, and enzymes that take oxygen as a substrate typically employ transition state metal-containing cofactors or organic redox cofactors such as flavin.While the reactivity of the quinonoid intermediate with oxygen has been demonstrated in DDC, 50,51 exactly how AASs promote decarboxylation-dependent oxidative deamination as their primary reaction and chemical details following the decarboxylation step await further investigation.
Structural information on plant AAADs has opened the door to employ MD simulations to explore their conformational dynamics.With microsecond simulations initiated from both the open-state crystal structure and the closed-state AF2 model, we set out to uncover the dynamics of the CrTDC large catalytic loop not captured by previous hundreds of nanoseconds simulations initiated from the crystal structure alone.
Apart from the open and closed states, as revealed by the crystal structure and AF2 prediction, these simulations uncover a wide range of loop conformations.Loop movement, as measured by a characteristic angle θ and the distance d between the key partners Tyr348 and His203, has an average correlation time of up to ∼500 ns.While similar loop dynamics may be expected for PsTyDC given the indispensable role played by the conserved Tyr, sequence variation of the catalytic loop between these two enzymes may produce conformational and/or secondary structural differences, which remains to be investigated by future studies.Mechanistically, given that decarboxylases have a low to moderate turnover rate on the order of 1−10 s −1 , 5,12 representative of an "average" enzyme, 52 loop movement with the above correlation time may still be far removed from being rate limiting.However, such a correlation time already presents a significant challenge to MD modeling, suggesting that tens of microseconds or even longer simulations may be needed to fully sample an AAAD's loop conformational space.This challenge is neither new nor uncommon, as loop sampling in other enzymes has been shown to be computationally demanding. 46,53,54As a result, apart from multiple replicas of simulations sufficiently longer than the correlation time of loop movement, enhanced sampling approaches 53,55,56 should be explored in future studies of the plant AAAD family.

■ CONCLUSIONS
In this work, we examined key structural and dynamic elements underlying the differential enzymatic functions of plant AAADs through a combination of modeling approaches.Our implicit ligand sampling calculations on PsTyDC revealed similar oxygen accessibility in both the wild-type enzyme and its Y350F mutant.The calculated 3D O 2 free energy profiles identify two major migration pathways and negligible barriers for oxygen entrance into both the WT and mutant enzymes, suggesting that a competition between the straightforward decarboxylation and decarboxylation-dependent oxidative deamination is always present in their active sites, and the dominance of one reaction type over the other cannot be attributed to oxygen accessibility.Along with previous experiments, 11 these results strongly indicate that the ability to donate a proton to the quinonoid reaction intermediate by the conserved tyrosine in canonical AAADs and the lack of such an ability by the corresponding phenylalanine in AASs predominantly underlies the divergence in catalytic pathways of the two types of enzymes from the same AAAD family.Through microsecond MD simulations on CrTDC initiated from its open-state crystal structure and closed-state AF2 prediction, we extensively examined the intricate conformational and dynamic details of its large loop housing the key catalytic tyrosine residue.The correlation time of loop movement highlights the challenge to conventional MD and the need for enhanced sampling approaches to comprehensively explore the conformational space of this indispensable catalytic loop of plant AAADs.

Figure 1 .
Figure 1.Crystal structures of CrTDC and PsTyDC. 14(A) Superimposed crystal structures of CrTDC (blue) and PsTyDC (lime).(B) Magnified snapshot showing different conformations of the large loop and the small loop in the crystal structures of CrTDC (blue) and PsTyDC (lime).For clarity, only the substrate tyrosine of PsTyDC is shown, while the substrate tryptophan of CrTDC was omitted.

Figure 3 .
Figure 3. 3D PMF of O 2 calculated by ILS in the WT protein (A) and the Y350F mutant (B) of PsTyDC.The PMF profiles are represented by black wireframes with energy isosurfaces at 3.9 kcal/mol (WT) and 2.6 kcal/mol (Y350F), respectively.The two pathways (a, b) leading to the protein active site from bulk water are indicated by red arrows, with residues lining each pathway being highlighted.Common to all images are the protein surface (blue cavity), the catalytic large loop (cyan), and the key Tyr350/Phe350 residue (licorice).(C) Free energy values of O 2 in water and at barrier locations along their migration pathways (see(A, B)) from the exterior of the protein to the active site in apo-PsTyDC, holo-PsTyDC with LLP and substrate, and holo-PsTyDC with external aldimine, respectively.The difference between WT and Y350F lacks statistical significance at the 0.05 level (denoted by ns) in all three simulation systems.

Figure 4 .
Figure 4. MD simulations of the CrTDC-AF2 and CrTDC-crystal systems.Projection of the large loop movement onto the first two principal components PC1 and PC2.The yellow and blue scattered dots represent trajectories from CrTDC-AF2 and CrTDC-crystal simulations, while the black square and triangle indicate the initial closed (CrTDC-crystal) and open (CrTDC-AF2) states of the loop, respectively (pentagrams: centroid structures of the top six clusters from clustering analysis of the large loop).(B) Free energy surface over PC1 and PC2 determined from the combined CrTDC-AF2 and CrTDC-crystal trajectories.(C) Centroid structures of the large catalytic loop.The initial CrTDC-AF2 (yellow) and CrTDC-crystal (blue) structures are shown as transparent cartoon representations for reference.

Figure 5 .
Figure 5. Conformation of the CrTDC large loop during MD simulations initiated from CrTDC-AF2 and CrTDC-crystal structures.(A) θ, (B) d, and (C) the α-helical content of the catalytic large loop during the simulations.The horizontal dashed lines represent the corresponding values of the initial CrTDC-AF2 (blue) and CrTDC-crystal (orange) structures, respectively.Results from individual 8 μs simulations are merged and displayed in all panels separated by vertical dashed lines.
), clusters 4 and 6 appeared to represent alternative open states of the catalytic loop.In particular, the catalytic loop flipped further away from the active pocket located at the center of the homodimeric enzyme in cluster 6 (θ = 125.9°),indicative of a "wide open" conformation.As shown in Figure S5C, angle θ even reached ∼150°during simulations of both CrTDC-AF2 and CrTDC-crystal.Along with the broad range of conformations explored by the loop over the PC1-PC2 space, this result again reflects the remarkable flexibility of the CrTDC catalytic loop and its access to multiple open (including wide open) conformations.