Structure and Function of Threonine Synthase from Yeast*

Threonine synthase catalyzes the final step of threonine biosynthesis, the pyridoxal 5′-phosphate (PLP)-dependent conversion ofO-phosphohomoserine into threonine and inorganic phosphate. Threonine is an essential nutrient for mammals, and its biosynthetic machinery is restricted to bacteria, plants, and fungi; therefore, threonine synthase represents an interesting pharmaceutical target. The crystal structure of threonine synthase from Saccharomyces cerevisiae has been solved at 2.7 Å resolution using multiwavelength anomalous diffraction. The structure reveals a monomer as active unit, which is subdivided into three distinct domains: a small N-terminal domain, a PLP-binding domain that covalently anchors the cofactor and a so-called large domain, which contains the main of the protein body. All three domains show the typical open α/β architecture. The cofactor is bound at the interface of all three domains, buried deeply within a wide canyon that penetrates the whole molecule. Based on structural alignments with related enzymes, an enzyme-substrate complex was modeled into the active site of yeast threonine synthase, which revealed essentials for substrate binding and catalysis. Furthermore, the comparison with related enzymes of the β-family of PLP-dependent enzymes indicated structural determinants of the oligomeric state and thus rationalized for the first time how a PLP enzyme acts in monomeric form.

Together with tryptophan synthase (TRPS), threonine deaminase (TDA), O-acetylserine sulfhydrylase (OASS), cystathione ␤-synthase (CBS), and 1-aminocyclopropane-1-carboxylate deaminase (ACCD), TS constitutes the core of the fold-type II family of PLP enzymes (4,5) (also referred to as ␤-family (6)). Detailed amino acid sequence alignments revealed that TS can be grouped into a plant and a fungal subfamily (7). The former one (class I subfamily) comprises TS from higher plants, cyanobacteria, archaebacteria, and the eubacterial groups of Mycobacteria, Aquificaceae, and Bacillus species. The second subfamily (class II subfamily) contains the enzymes from fungi and from the eubacterial groups of Proteobacteria and corneyform bacteria. Only 5 from ϳ500 residues are invariant between both subfamilies, including the PLP-binding lysine as part of a phenylalanine-lysine-aspartate consensus sequence.
During the last decades, TS was purified and characterized from several bacteria and fungi (8 -12) and from Arabidopsis thaliana (7,13). In plants, the substrate of TS, OPHS, is the branching point for threonine and methionine biosynthesis. Flux coordination between both synthetic pathways is accomplished by allosteric activation of plant threonine synthase. The allosteric effector is S-adenosyl methionine (SAM) (14,15), a product of methionine decomposition. When the intracellular levels of methionine and SAM are high, TS partitions OPHS away from methionine toward threonine synthesis. TS from higher plants occurs in solution as homodimers of about 110 kDa, and detailed studies with the A. thaliana (aTS) enzyme indicated that the extended N terminus is responsible for allosteric activation (7,16). In bacteria and fungi, homoserine is the last common precursor of threonine and methionine, and the activity of TS seems not to be regulated by any effector. Instead, homoserine kinase is feedback-inhibited by threonine (17,18). TS from the fungal class are unique among PLP-dependent enzymes, because they fulfill their physiological function in monomeric state (8,10). Recently, the crystal structure of the aTS apoprotein structure has been solved (19) indicating its domain organization and its quaternary assembly. However, neither the PLP cofactor nor the allosteric activator SAM could be identified in the highly disordered active site of aTS, restricting mechanistic conclusions that could be derived from the structural data.
Because threonine biosynthesis is restricted to bacteria, fungi, and plants, TS is an interesting target enzyme for the development of novel antibiotics and herbicides. Consequently, a number of substrate analogues have been tested as putative mechanism-based inhibitors of TS (7,20,21). As a first step toward structure-based drug design, we report here the crystal structure of TS from Saccharomyces cerevisiae (yTS) at 2.7 Å resolution. This is the first structure of a PLP enzyme that is active as monomer. Furthermore, comparison with related enzymes and modeling of the physiological enzyme-substrate complex give new insight into the reaction mechanism of threonine synthases.

EXPERIMENTAL PROCEDURES
Protein Overexpression and Purification-Briefly, the PCR-amplified S. cerevisiae thrC gene was subcloned into the Novagen pET22b(ϩ) vector NdeI/XhoI restriction sites and transformed into Escherichia coli BL21(DE3) cells via electroporation. Transformed bacteria were grown in Luria broth supplemented with 100 mg liter Ϫ1 ampicillin (LB/Amp), induced with 0.5 mM isopropyl-␤-D-thiogalactopyranoside, and harvested after 3 h of induction. The protein was purified in a three-step procedure. Frozen cells were ruptured by sonification, and the supernatant was loaded onto a DEAE-Sepharose FF column buffered with 50 mM potassium phosphate, pH 7.5, 2 mM EDTA, and 10 M PLP (buffer A). The flow-through containing yTS was collected and ammonium sulfate (AS) was added up to a concentration of 1.3 M. The protein was charged on a phenyl-Sepharose-HP column equilibrated with 1.3 M AS in buffer A and eluted with a linear gradient from 1.3 to 0.25 M AS in buffer A. Fractions containing yTS were subjected to size-exclusion chromatography (Sephacryl S-200 HR, 10 mM potassium phosphate, pH 7.5, 10 M PLP).
To perform a selenomethionine (SeMet) multiple-wavelength anomalous dispersion experiment, the yTS enzyme was expressed in the (metϪ) E. coli strain B834(DE3)(hsd metB), growing in new minimal medium (NMM (22)). To overcome the problem of methionine incorporation and of slow growth due to the presence of toxic SeMet, bacteria grown overnight in LB/Amp were harvested by centrifugation, washed and resuspended in NMM, and used to inoculate 10 liters of NMM. Induction of overexpression and purification of SeMet-yTS was performed as described for the native protein. The level of SeMet incorporation in purified yTS was almost 100%, as determined by electrospray mass spectroscopy.
Crystallization and Data Collection-Prior to crystallization, the protein was concentrated with Centripreps to 10 mg ml Ϫ1 . Crystals were grown at 4°C by the sitting-drop, vapor diffusion method with 0.1 M Hepes/NaOH, pH 7.6, 20% polyethylene glycol 4000, and 10% dioxane as crystallization liquor in a 2:1 ratio (3 l of protein solution and 1.5 l of reservoir). Drops were pipetted in CrysChem plates containing a volume of 500 l in the reservoir. In the case of SeMet-yTS, successful conditions were similar to those of the native protein with the difference that a mixture of 0.1 M BaCl 2 and 20% benzamidine was used as additive instead of dioxane. During the crystallization trials of yTS, two main problems were encountered: (i) obtainment of single non-twinned crystals, and (ii) slow crystal growth. Unfortunately, both problems could not be overcome by other crystallization techniques (e.g. seeding). Both native and SeMet-yTS crystallized in the monoclinic space group P2 1 with almost identical unit cell parameters, a ϭ 97.6 Å, b ϭ 52.5 Å, c ϭ 109.5 Å, ␤ ϭ 99.6°and a ϭ 95.3 Å, b ϭ 51.6 Å, c ϭ 106.9 Å, ␤ ϭ 99.6°, respectively. Assuming two monomers per asymmetric unit, the Matthews parameter (23) for the yTS crystals was calculated as V M ϭ 2.4 Å 3 Da Ϫ1 corresponding to a solvent content of 45%. The cryoprotectant consisted of the crystallization solution supplemented with 16% (v/v) of 2-methyl-2,4-pentanediol. Crystals were soaked for a few seconds in the cryobuffer and then frozen in a nitrogen stream at 100 K (Oxford Cryosystems Cryostream). With the selenium-substituted crystals, a three-wavelength data experiment was performed from a single crystal to 2.7 Å resolution at the BW6 synchrotron beamline at Deutsches Elektronen Synchrotron (Hamburg, Germany) on an MAR345 image plate. Three data sets were collected: one at the SeMet fЉ-maximum ( 1 ϭ 0.9874 Å), the second data set at the fЈ-minimum ( 2 ϭ 0.98 Å), and remote data ( 3 ϭ 0.92 Å) to maximize the anomalous fЉ and the disperse fЈ contributions of the K-shell absorption edge of selenium.
Structure Determination and Refinement-Data were processed and scaled using the programs DENZO and SCALEPACK (24). Anomalous difference Patterson maps were calculated from the f Љ data set and allowed identification of twelve selenium sites using the program RSPS (25). The selenium sites were refined with SHARP (26) and used to calculate protein phases and electron densities. After solvent flattening with SOLOMON (27) the electron density showed several secondary structure elements and allowed building the molecular mask with MAMA (28). Using NCS operators extracted from the heavy atom sites with FINDNCS (29), a 2-fold averaged map, in which almost the complete backbone could be inserted, was calculated with AVE (28,30). The model building was performed using the program O (31). After building the first monomer with a poly-alanine chain and including some of the side chains, the model was extended to the second monomer with the NCS operator. Most of the remaining side chains were included in the model after a rigid body and positional minimization with CNS (32) using the protein parameter set of Engh and Huber (33). Bulk solvent, overall anisotropic B-factor corrections, and NCS restraints were introduced depending on the behavior of the free R-factor index. Simulated annealing, individual B-factor refinement, and three rounds of automatic water building and deleting steps were followed by a visual inspection of the resulting 2F o Ϫ F c and F o Ϫ F c electron density maps. After several cycles, the refinement converged at a R-factor of 21.2% with an R free of 25.2%. All refinement procedures were performed using the maximum-likelihood target (34).
Molecular Modeling-Molecular modeling was done using the modules Viewer, Builder, Docking, Delphi, and Discover3 of the program package Insight II (Version 97.0, MSI, Los Angeles, CA). One monomer of the refined yTS x-ray structure was chosen as the starting point for model building and energy minimization of the modeled complex. OPHS and subsequently its external aldimine with PLP were generated and positioned into the active site based on the binding of aminoethoxy vinylglycine in cystathionine ␤-lyase (35). The Consistent Valence Forcefield of Insight II program was used for energy minimization with Discover3, which was run over 1000 minimization steps until convergence (0.1 kcal/mol tolerance). The protein monomer and the ligand molecule were minimized simultaneously.

RESULTS AND DISCUSSION
Quality of the Structure-Bright yellow crystals of yTS were obtained in the monoclinic space group P2 1 with two molecules in the asymmetric unit, corresponding to a solvent content of 45%. The structure of yTS was solved by the multiwavelength anomalous dispersion method using incorporated selenomethionine (SeMet) as an anomalous scatterer. The original density obtained with SHARP (26) was improved by density modification protocols implemented in SOLOMON (27) and 2-fold realspace averaging with AVE (28,30). The PLP cofactor was inserted during the last round of model refinement when its orientation could be unequivocally derived from the 2F o Ϫ F c omit density (Fig. 1a). The final model comprises 2 yTS monomers, 2 PLP molecules, and 334 water molecules and displays good stereochemistry (Table I). The two yTS molecules are related by a non-crystallographic 2-fold axis, and a superposition performed with the program O (31) resulted in a root mean square (r.m.s.) deviation of 0.2 Å for 509 C ␣ main-chain atoms.
In the final electron density, most of the 514 residues from both monomers were clearly defined. Only the electron density of residues 146 -151, 228 -233, and 513-514 was of insufficient quality to construct them. Therefore, these residues are omitted from the model. Similarly, the side chains of residues 219 -227 were highly disordered. Interestingly, the segment 219 -233 precedes the residues of the active site, which appear to be involved in substrate binding. Thus the flexibility of this segment seems to be of functional importance. The remaining disordered regions are highly solvent-exposed and not fixed by any intramolecular interaction. Furthermore, they are not involved in any crystallographic contacts. During the refinement the R-factor dropped from 43.8 to 21.2%, while the free R-factor decreased continuously from 45.0 to 25.2%. In the Ramachandran plot, calculated with PROCHECK (36), 84.3% of the residues lie in the most favored, 13.9% in the additional allowed, and 1.1% in the generously allowed regions. One residue, Val-96, is found to have a forbidden combination of dihedral angles of the peptide backbone. However, the electron density of this residue is unambiguous. Val-96 is part of a tight ␤-turn wedged between several bulky residues that define its orientation. Because this ␤-turn is located far off the active site (ϳ20 Å) and does not interact with any active site residue, the Ramachandran outlier Val-96 seems to be of structural importance.
Both crystallographic and non-crystallographic contact surfaces are small (Ͻ0.5% of the 40,135 Å 2 total surface area). The most intimate contacts were observed between the two molecules in the asymmetric unit. At the corresponding interface, FIG. 1. Structural features of the yTS monomer. a, stereoview of the electron density for the PLP cofactor (with Lys-124 in orange) and the immediate vicinity (colored by atom type). The 2F o Ϫ F c simulated annealing omit map (2.7-Å resolution, contour level 1.0 ) was calculated by omitting the PLP and Lys-124 from phase calculation. b, domain organization. A topology diagram of the secondary structure is shown on the left side, with the A, B and C domains colored in purple, orange, and gray, respectively. The PLP bound to helix ␣5 is shown in green. On the right side, the solvent-accessible surface of the monomer is colored according to its domain architecture. The PLP cofactor is bound in the deep cleft located at the BC domain interface. c, stereo ribbon presentation of the yTS fold. The nomenclature of the secondary structure elements is given, ␣-helices are colored in blue, ␤-sheets in red, and loops in yellow. The PLP is colored by atom-type. a was produced with SETOR (48). b was produced with DINO (www.dino3d.org); c was produced with MOLSCRIPT (49) and RASTER3D (50).
residues Glu-72, Asp-77, Ala-78, and Lys-81 and some welldefined water molecules undergo a few polar interactions establishing a small delimited interaction patch. Therefore, the crystal structure of yTS confirms biochemical results indicating that the monomer is indeed the biocatalytically active form of fungal TS.
Overall Structure of yTS-The dimensions of the globular yTS monomer are ϳ64 ϫ 62 ϫ 58 Å 3 . The overall fold of yTS is presented schematically together with a surface presentation and a ribbon plot in Fig. 1 (b and c), the participating residues in Table II. The structure can be described in terms of three functionally and spatially different domains A, B and C: a small N-terminal domain (A), the PLP-binding domain, which comprises also the C terminus of the polypeptide chain (B), and the so-called large domain (C). Regarding the amino acid sequence of yTS, domains B and C are not sequential. While domain A contains residues 1-94, B is composed of residues 120 -262 and 477-512, and the C domain comprises residues 95-119 and 263-476. Common to all domains is an open ␣/␤ sandwich structure as a central building element. The N-terminal domain consists of a two-stranded antiparallel ␤-sheet comprising ␤-strands ϩ␤a and Ϫ␤b. These strands are connected by a short ␣-helix, ␣1, and are followed by helices ␣2, ␣3, and ␣4, which pack on the solvent-accessible side of the sheet. The PLP-binding B domain contains a five-stranded parallel ␤-sheet (ϩ␤h, ϩ␤e, ϩ␤f, ϩ␤g, and ϩ␤o) that is shielded from solvent by helices ␣5, ␣6, and ␣7 on one side and helices ␣8 and ␣20 on the other side. The PLP cofactor is bound at the Nterminal end of helix ␣5, which is the longest helix in the entire structure, constructed by 22 amino acids corresponding to a length of 38 Å. The cofactor is covalently attached to the ⑀-amino group of Lys-124 yielding the internal aldimine frequently observed in PLP-dependent enzymes (37). Domain C displays a complex architecture, which can be described as a combined ␤-sheet with an in-between helix. The mainly paral-lel six-stranded ␤-sheet, Ϫ␤d, ϩ␤c, ϩ␤i, ϩ␤n, ϩ␤j, and ϩ␤m, is separated by helix ␣11 from the short, antiparallel ␤-sheet, ϩ␤k and Ϫ␤l. Both sheets are flanked by helices ␣10, ␣12, ␣13, ␣14, ␣15, and ␣16.
At the individual domain interfaces, both main-chain and side-chain atoms are involved in a multitude of polar and hydrophobic interactions. The A-C domain interface resembles a heterodomain ␣␤␣ sandwich. Helices ␣12 and ␣13 of domain C constitute one ␣-layer flanking the ␤-sheet of domain A on one side. Numerous hydrogen bonds and one strong salt bridge between Arg-12 and Glu-340 were observed at this interface. The interactions between domains A and B are established by a mixed four-helix bundle comprising helices ␣3, ␣4, ␣5, and ␣9, which binds helix ␣1 at its center. A large cluster of hydrophobic residues forms the core of this interaction patch that is enclosed by a total of 12 hydrogen bonds and two salt bridges.  The third interdomain contact between B and C is observed at the C-terminal ends of the individual ␤-sheets. Similar to the other domain interfaces, several hydrogen bonds are formed between the two large domains. Furthermore, the non-sequential construction couples both domains closely to each other. In summary, an intensive net of interactions defines the orientation of domains A, B and C relative to each other, thereby restricting domain flexibility and reorientation.
Comparison with Threonine Synthase from A. thaliana-Although the active unit of the ␤-family PLP enzymes is the monomer, i.e. the active site exhibits no contributions from the neighboring subunit, most of the ␤-family enzymes exist as dimers or tetramers. Presumably, oligomerization is a prerequisite for allosteric regulation. Consistently, regulatory domains have been identified for the structurally characterized TDA, OASS, TRPS, and CBS and the plant threonine synthase aTS. Although the activity of fungal TS is not regulated by any known effector molecule, the dimeric plant TS is subject to positive allosteric regulation by SAM (14,15). To illustrate these activation differences and to determine the main structural differences of threonine synthase of classes I and II, a detailed structural comparison was performed between yTS and the recently determined crystal structure of aTS. The sequence identity shared by the enzymes from A. thaliana and S. cerevisiae is 23%. Several of the conserved residues of the two central ␤-sheets were chosen to determine the initial superposition matrix with the program O, which subsequently was improved by extending the alignment to the overall structure resulting in an overall r.m.s. deviation of 2.05 Å for 270 C ␣ s (Fig. 2a).
Most of the central part of the yTS structure, comprising main regions from all three domains, is conserved in aTS. Structural differences arise generally by the incorporation of extra secondary structure elements. Particularly, ␣13, ␤l, and ␣14 of yTS have no counterparts in aTS. Furthermore, the N-terminal domain adopts a completely different fold in both subfamilies. In yTS, these extra structural elements interact with each other and bury the regions that set up the dimer interface of aTS, thus explaining the differences in the oligomeric states within the subfamilies inside the TS family. Furthermore, the aTS-specific swap domain, which is of central importance for dimerization, has no equivalent in yTS. It has been proposed that the N-terminal extension, only present in TS from higher plants (i.e. A. thaliana), and the ability to dimerize are necessities for transmitting the allosteric effect of SAM. Due to the absence of the allosteric effector from the aTS crystal structure, a hypothetical SAM binding pocket had been suggested on the basis of structural comparisons with other SAM-binding proteins. Consistent with the biochemical data, the SAM-binding motif is located in the N-terminal domain, which is a unique feature of the plant threonine synthases. However, the mode of action of the allosteric activator SAM is still not understood.
We also performed an alignment of the respective active site structures. Unfortunately, the aTS structure represents the apoform of the enzyme, in which the cofactor is absent. The superposition with the active site of yTS clearly indicates the limitation of the aTS structure to deduce mechanistic features of threonine synthesis. Large parts of the active site are disordered, the orientation of the PLP cannot be unequivocally inferred, and the orientation of catalytic and substrate-binding residues is elusive. It was proposed that the loss of the cofactor underlines the high degree of disorder in the aTS active site region, in which even the mechanistically important loop comprising residues 347-363 (319 -335 in yTS) is absent. In related enzymes, this loop forms one border of the substrate binding pocket. Thus Biou and coworkers (19) drew their mechanistic conclusion solely from a sequence alignment with TDA, TRPS, and OASS.
As these authors already mentioned, several active site residues that play important catalytic function obtain different conformations compared with the related enzymes TDA, TRPS, and OASS. These residues include, for example, Asp-194 (Asp-163 in yTS), which had been proposed to interact with PLP-O3, and Thr-432 (Thr-449), which is bound to the pyridine nitrogen N1. Reorientation of these residues by binding the PLP cofactor can be envisaged by the yTS structure. Further structural adjustments of the active site to fine tune the reactivity of the cofactor includes residues Phe-123 (Phe-162 in aTS), Arg-246 (Arg-267), and His-422 (His-404) as indicated in Fig. 2b. Arg-246 is of outstanding importance, because it is the residue that binds the substrate ␣-carboxylic group, as revealed by modeling an enzyme-substrate complex of yTS (see below). The other two residues, Phe-123 and His-422, fix the orientation of the cofactor in the active site by impeding any movements in the vertical direction (respective to the pyridine ring plane). Interestingly, His-422 of yTS is not involved in binding N1 of the pyridine and, presumably, neither is its aTS counterpart. Instead, Thr-449 plays this role in yTS. Another notable difference concerns the phosphate binding pocket. In the apo-form of aTS, the phosphate-binding loop comprising residues 277-281 protrudes more into the active site, occupying the place of the cofactor phosphate group. The crystal structure of yTS indicates how this region re-orientates upon cofactor binding. In summary, the well-defined active site architecture of yTS provides for the first time the molecular frame with which to postulate the detailed reaction mechanism of threonine formation.
Structural Comparison with Related Structures-The DALI algorithm was employed to compare the crystal structure of yTS with structures deposited in the Protein Data Bank (PDB) (38). The six members of the fold-type II family of PLP enzymes (also know as ␤-family (6)) yielded the highest similarity scores (Table III): aTS (PDB code 1E5X), TDA (1TDJ), OASS (1OAS), TRPS (2TYS), CBS (1JBQ), and ACCD (1F2D). The DALI superposition of yTS with these enzymes resulted in r.m.s. deviations of 2.7-3.6 Å for 288 -354 aligned C ␣ s. As for the alignment with aTS, most of the central part of yTS is conserved in the other ␤-family enzymes. Structurally differences arise generally by the incorporation of secondary structure elements into yTS (Fig. 2c). At its C terminus, the additional helices ␣18, ␣19, and ␣20 and strand ␤o occupy the region that forms the monomer-monomer interaction patch of TDA. Furthermore, residues 12-20 and strand ␤a at the N terminus of yTS and residues 346 -384, comprising helices ␣13 and ␣14 and strand ␤l, cover the surface that is essential for TDA dimer-dimer interactions. Thus, potential oligomerization patches involved in dimer (TDA, OASS, TRPS, aTS, and ACCD) and/or tetramer formation (TDA, CBS) are occluded in yTS. Remarkably, these extra elements are highly conserved in the subfamily of fungal TS. A notable difference in the conservation pattern of these segments is that the extra terminal segments are strictly conserved, whereas the ␣13-␤l-␣14 region is absent in a few members of the fungal TS family. Thus the unique N-and Cterminal extensions of fungal TS appear to be the main determinants of the monomeric state, which seems to be a unique property in the large family of PLP enzymes.
Active Site Architecture and Mechanistic Implications-The monomer has a "Pac Man"-like shape due to the presence of a canyon crossing the whole molecule (Fig. 1b). This substratebinding groove is 43 Å wide, 19 Å deep, and allows access to the PLP cofactor, which is fixed at the bottom of this canyon at the C-terminal end of the large ␤-sheet of domain B. The yTS substrate binding funnel is lined primarily by basic residues creating a positive surface potential. Participating residues originate from all three domains and are well conserved within the fungal TS family, suggesting a general substrate-trapping mechanism. Binding of the cofactor is mainly achieved by the PLP-binding lysine and the highly specific pocket that anchors the PLP phosphate group. These two elements originate from the B and C domains, respectively. The organization of the active site is shown schematically in Fig. 3a, its detailed spatial structure in Fig. 3b, and the alignment with the TDA active site in Fig. 4a. yTS exhibits the characteristic absorption maximum of a protonated Schiff base at 422 nm that does not significantly change in intensity in the pH range of 6 -9 (data not shown). Thus the pK a value of the internal Schiff base formed between Lys-124 and PLP appears to be greater than 9. A hydrogen bond between the protonated nitrogen of this aldimine and PLP-O3Ј stabilizes the negative charge of the latter and keeps the aldimine linkage in the same plane as the pyridine ring system. The environment of the PLP-O3Ј mainly determines the pK a of the internal aldimine. In yTS, Asp-163 is oriented FIG. 2. Structural comparison. a, stereoview of the C ␣ trace superposition of yTS (color code as in Fig. 1b) and aTS (green). The N and C termini are indicated, as well as the extra elements for yTS ␣13-␤l-␣14 and ␣17-␣18. b, structural alignment of the active site of yTS (colored-coded by atom) and aTS (green). c, ribbon representation of the yTS summarizing the extra secondary elements of yTS (residues 1-20: ␣13, ␣14; ␤l: ␣18, ␣19, ␣20; ␤o: highlighted in red) relative to aTS and TDA. a and b were produced with DINO; c was produced with MOLSCRIPT (50) and RASTER3D.
properly to form a hydrogen bond with the negatively charged O3Ј, suggesting that the carboxylate group of this aspartate is protonated. Obviously, the interaction between Asp-163 and PLP-O3Ј does not lower the pK a of the internal aldimine. Thus, Asp-163 is proposed to bind the amino group of the incoming substrate to orient it properly for transaldimination. A welldefined water molecule (B-factor of 38 Å 2 ) in close proximity to the carboxylate group of Asp-163 may point to the binding position of the substrate's amino group in the Michaelis complex.
The PLP pyridinium ring is sandwiched between Ala-330, Met-331, and His-422 at the solvent-exposed side and Phe-123 at the side directed toward the protein interior. In concert these residues impede any movement of the pyridine ring in a vertical direction (Fig. 3a). Due to their orientation, the ring systems of His-422 and Phe-123 are unable to undergo ring-stacking interactions with the pyridine ring. The PLP-N1 nitrogen is in hydrogen bonding distance to Thr-449 located at the Cterminal end of strand ␤n. The position of the threonine side chain is fixed by a hydrogen bond between its hydroxyl group and the main-chain amide of Ala-450. Furthermore, its methyl group is sandwiched between the side chains of Phe-123 and Pro-275. The PLP-C2Ј methyl group shows no significant interaction with the protein. It is buried in a hydrophobic pocket constructed by residues Ala-450, Phe-455, and His-422.
The phosphate group of PLP represents the main anchor of the cofactor and is attached to the protein by a total of eight hydrogen bonds. Individual interactions are undergone between OP1 and the main-chain amide nitrogens of Gly-280 and Asp-281, which are located at the N terminus of helix ␣10. Asn-278 forms a strong hydrogen bond to OP2, which receives a second hydrogen bond from a well-defined active site water. OP3 is hydrogen bonded to the main-chain amide nitrogens of Gly-277 and Phe-279 and another water molecule that is fixed by the main-chain amide group of Ile-282. Finally, OP4 is hydrogen-bonded by a water molecule that is held in place by Pro-275. Binding of the PLP phosphate is further improved by polar interactions with the positive end of the macrodipole of helix ␣10, a feature frequently seen in PLP-binding proteins (for a review see Ref. 39). In the family of fungal TS, all of the described active site residues are well conserved; residues Ser-161, Asp-163, Thr-164, Pro-275, Thr-449, and Ala-450 exhibit even an invariant pattern.
Modeling of the Substrate Binding Mode-Due to problems with crystal growth and crystal stability, it was not possible to obtain experimental data for yTS enzyme-substrate or enzyme- FIG. 3. Active site of yTS. a, schematic representation of the functionally important interactions within the active site. On the left side, hydrogen bonds between protein, water (dark balls) and cofactor (light gray) are indicated, whereas the fixation of the PLP pyridine ring is shown on the right side. b, detailed active site architecture. The internal aldimine is seen in light gray, water molecules are shown as dark balls, and the macrodipole of helix ␣10 is indicated. b was produced with DINO. inhibitor complexes. Therefore, we decided to model the putative binding mode of the substrate OPHS (Figs. 4b and 5a). As a starting point, OPHS was pre-orientated into the active site as the external aldimine, with its ␣-carboxylate group pointing toward Thr-164 and its side chain stretching into the active site pocket without undergoing any specific interactions. The position of the cofactor was not changed. The situation before and after energy minimization is de-picted in Fig. 4b. The external aldimine-enzyme complex was minimized to a total energy of 3400 kcal/mol after 1000 steps and tilted the pyridinium ring of PLP by ϳ10°. Thereafter, the ␣-carboxylate group of OPHS moved toward Thr-164, obviously one of the main substrate binding residues, and formed an additional hydrogen bond to Ser-161. Both carboxylate binding residues are conserved in about 28 out of 30 threonine synthase sequences. During energy minimization, the phosphate group of OPHS moved toward the side chain of Arg-246 and formed two hydrogen bonds with its guanidinium group, which are strengthened by charge-charge interactions. Additional hydrogen bonds were formed with Asn-240, Ser-241, and Asn-278. All four phosphate-binding residues are strictly conserved in the fungal subfamily of TS and also show a high degree of conservation in the plant subfamily, where they are conserved in 13 out of 16 sequences. Therefore we propose that Asn-240, Ser-241, Arg-246, and Asn-278 constitute a TS-specific phosphate-binding pocket.
In our calculated model, formation of the ES complex led to structural adjustments in the active site, particularly of residues interacting with the phosphate group of the substrate. The greatest structural differences between the free enzyme and the modeled enzyme-substrate complex were observed for the phosphate-binding loop comprising residues 238 -242. The average r.m.s. deviation between the corresponding C ␣ atoms of both structures is 1.5 Å, which is rather higher than the overall r.m.s. deviation of 0.4 Å. Interestingly, the phosphatebinding loop and its preceding segment 215-237, including helix ␣8, exhibit disorder as implied by high thermal factors in the native structure indicating the flexibility of this active-site region. This should ease the acceptance of the substrate via an induced-fit mechanism. Thus the inherent flexibility of the segment 238 -242 seems to be a requirement for facile loop reorientation to set up the phosphate-binding site.
Detailed mechanistic proposals have been put forward for the yTS-related enzymes TRPS and OASS (40,41). Remarkably, an open-closed transition has been reported for both enzymes. In OASS, a domain movement comprising strands ␤4 and ␤5 as well as helices ␣3 and ␣4 is triggered upon substrate binding and moves a so-called "asparagine loop" (residues 67-70) deeper into the active site, thereby creating several new hydrogen bonds with the carboxylate group of the substrate. Besides improved substrate anchoring, the domain movement also shields the active site from solvent. However, the mobile elements of OASS have no counterparts in yTS. Furthermore, a large conformational change upon binding of a substrate analogue has been reported for the K41A mutant (PLP-binding lysine exchanged by alanine) of OASS (42), where the subdomain comprising residues 87-131 moves as a rigid body. Similarly, in TRPS the so-called COMM domain (residues 102-189) undergoes a rotation promoted either by bound ligands or by binding of monovalent cations (43,44). These subdomains align to ␣6, ␤f, ␣7, ␤g, ␣8, and ␤h of yTS. Because this part of the B domain is fixed by the extra structural elements ␣19, ␤o, and ␣20 to the C domain, the domain movements observed in OASS and TRPS seem irrelevant for yTS, thus confirming our modeling experiment.
Reaction Mechanism of yTS-TS is the only PLP-dependent enzyme that catalyzes a ␤,␥-replacement reaction on an amino acid substrate. The complex reaction mechanism of TS, which formally involves abstraction of the C ␣ and C ␤ protons of the substrate OPHS, non-hydrolytic removal of the phosphate group, protonation of C ␥ , and addition of water at C ␤ , has challenged scientists for many decades. A minimum number of two residues acting as acid-base catalysts has been proposed to be involved in this reaction. However, the present crystal structure reveals, that Lys-124 is the only residue in the immediate environment of the bound substrate that can act as an acidbase catalyst. Therefore, the complex chemistry of threonine formation seems to be catalyzed by the ⑀-amino group of Lys-124 as the single catalytic residue. Our modeled structure of the external aldimine indicates that this amino group can be brought by slight rotations of the torsion angles chi1, chi2, chi3 and chi4 into hydrogen-bonding distance to all relevant atoms. The likely reaction mechanism is shown in Fig. 5b.
After productive binding of OPHS, which could be dominated by interaction of the phosphate group of the substrate with the phosphate binding pocket (Asn-240, Ser-241, Arg-246, Asn-278) and of the substrate's carboxyl group with Ser-161 and Thr-164, the amino group has to be deprotonated to initiate the transaldimination reaction. For this process three different scenarios can be envisaged: (i) Deprotonation may be achieved by the deprotonated form of the internal aldimine, as in aspartate aminotransferase (45). (ii) Deprotonation by a protein residue acting as an active site base represents a similar mechanism and has been proposed for several trans-sulfuration enzymes (46). (iii) Depending on the pK a of the amino group in the environment of the active site, a significant portion of the substrate may already exist in deprotonated form (47). Because the former two proposals can be excluded for yTS, the latter mechanism seems to be fitting. To test this hypothesis we analyzed the pH optimum of yTS and, in consistency with the supposed activation mechanism, found the pH optimum to be quite basic, around pH 8.2. Thus, a sufficient amount of "activated" substrate should be present in solution to account for the observed reaction rates.
After transaldimination, the newly formed external aldimine (II) is in proper orientation for C ␣ proton abstraction. Accordingly, Lys-124 catalyzes the abstraction of the C ␣ proton and its transfer to the PLP C4Ј position (III). In the next step of the reaction, Lys-124 stereospecifically removes the ␤-pro-S hydrogen, thereby triggering the non-hydrolytic elimination of the ␥-substituent as inorganic phosphate. The ␥-methylene group of the resulting vinylglycine ketamine intermediate (IV) is reprotonated, thus yielding the PLP-derivative of E-aminocrotonate (V). The addition of water at C ␤ and, finally, reverse transaldimination yields L-threonine with retention of configuration.
In accordance with the fact that the structurally characterized ␤-family enzymes yTS, TDA, TRPS, OASS, CBS, and ACCD utilize the same cofactor but catalyze different chemical reactions, their active sites show obvious similarities and substantial differences. The similarities include: (i) the binding mode of the phosphate group of the cofactor, which is fixed by a glycine-rich region at the N-terminal end of helix ␣10; (ii) the identical orientation of the always protonated external aldimine such that its re side is facing the solvent; (iii) the interaction of the positively charged pyridine nitrogen with the hydroxyl group of either threonine (yTS), serine (TDA, OASS, TRPS, CBS), or glutamate (ACCD); and (iv) the absence of any ring-stacking interactions with the PLP pyridine ring. The latter two features should reduce the electron withdrawing capacity of the cofactor. The most striking difference in the active site architecture of the respective enzymes is related to substrate binding. The comparison of yTS with TDA provides a nice example of how both enzymes exploit the steric and electrostatic design of their active sites to achieve proper selection of their substrates, which differ in size and charge. The active site entrance of TDA is much narrower due to some interdomain-spanning interactions, as for example the salt bridge between Lys-117 and Glu-240. Furthermore, the active site cleft of TDA is lined by a mixture of acidic and basic residues, whereas the pocket of yTS is primarily lined with basic residues (Lys-222, Lys-228, Lys-233, Arg-246, Arg-319, Lys-371) and amide nitrogens that generate a strikingly positive surface potential, which could guide the phosphate group of the OPHS substrate down to the active site. Most notably, the pocket that binds the phosphate group of OPHS is occupied in TDA by the loop comprising residues 153-158 (Fig. 4c). We therefore propose that the substrate and reaction specificities of the ␤-family of PLP-dependent enzymes are determined by structural differences of their substrate binding pockets, whereas the arrangement of the catalytic important functionalities is strikingly similar.