Crystal Structures Capture Three States in the Catalytic Cycle of a Pyridoxal Phosphate (PLP) Synthase*♦

Background: Pyridoxal phosphate (PLP) synthase generates PLP from three common metabolites in a reaction with three covalent intermediates. Results: Crystal structures capture the synthase active site at three steps of catalysis. Conclusion: Protein dynamics are correlated with formation of covalent intermediates. Significance: New information is provided about the complex catalysis in the synthase active site and the roles of conserved amino acids. PLP synthase (PLPS) is a remarkable single-enzyme biosynthetic pathway that produces pyridoxal 5′-phosphate (PLP) from glutamine, ribose 5-phosphate, and glyceraldehyde 3-phosphate. The intact enzyme includes 12 synthase and 12 glutaminase subunits. PLP synthesis occurs in the synthase active site by a complicated mechanism involving at least two covalent intermediates at a catalytic lysine. The first intermediate forms with ribose 5-phosphate. The glutaminase subunit is a glutamine amidotransferase that hydrolyzes glutamine and channels ammonia to the synthase active site. Ammonia attack on the first covalent intermediate forms the second intermediate. Glyceraldehyde 3-phosphate reacts with the second intermediate to form PLP. To investigate the mechanism of the synthase subunit, crystal structures were obtained for three intermediate states of the Geobacillus stearothermophilus intact PLPS or its synthase subunit. The structures capture the synthase active site at three distinct steps in its complicated catalytic cycle, provide insights into the elusive mechanism, and illustrate the coordinated motions within the synthase subunit that separate the catalytic states. In the intact PLPS with a Michaelis-like intermediate in the glutaminase active site, the first covalent intermediate of the synthase is fully sequestered within the enzyme by the ordering of a generally disordered 20-residue C-terminal tail. Following addition of ammonia, the synthase active site opens and admits the Lys-149 side chain, which participates in formation of the second intermediate and PLP. Roles are identified for conserved Asp-24 in the formation of the first intermediate and for conserved Arg-147 in the conversion of the first to the second intermediate.

PLP synthase (PLPS) is a remarkable single-enzyme biosynthetic pathway that produces pyridoxal 5-phosphate (PLP) from glutamine, ribose 5-phosphate, and glyceraldehyde 3phosphate. The intact enzyme includes 12 synthase and 12 glutaminase subunits. PLP synthesis occurs in the synthase active site by a complicated mechanism involving at least two covalent intermediates at a catalytic lysine. The first intermediate forms with ribose 5-phosphate. The glutaminase subunit is a glutamine amidotransferase that hydrolyzes glutamine and channels ammonia to the synthase active site. Pyridoxal 5Ј-phosphate (PLP), 2 a biologically active form of vitamin B 6 , includes a pyridine ring with hydroxyl, methyl, formyl, and phosphate substituents. The carbaldehyde at the 4Ј-position makes this B 6 vitamer the most versatile co-factor in nature through formation of Schiff-base intermediates with both enzymes and substrates. PLP-dependent enzymes are predominantly involved in amino acid metabolism, lipid metabolism, and gluconeogenesis where they catalyze transamination, decarboxylation, racemization, elimination, or replacement of an electrophilic group and phosphorolysis reactions (1,2). In addition to being an efficient co-factor, PLP has antioxidant properties, as the core aromatic pyridine ring of PLP is an efficient singlet oxygen scavenger (3)(4)(5).
PdxS and PdxT have distinct and spatially separated catalytic functions (7,8). PdxT, a glutamine amidotransferase (GAT) of the Triad (class I) family (7,8,14), uses a Cys-His-Glu catalytic triad to produce ammonia by hydrolysis of glutamine (Fig. 1A). PdxS is a remarkable synthase that produces PLP using readily available metabolites from the pentose phosphate pathway (R5P), from PdxT (NH 3 ), and from glycolysis (G3P) (7,8). The activity of PdxT is coupled to its PdxS synthase in a different manner than for other GATs (15). Typically, GATs have abrogated glutamine hydrolysis in the absence of their synthase substrates. PdxT is unique in that its glutaminase activity depends only on formation of the intact PLPS (PdxS 12 PdxT 12 ) and is independent of the PdxS substrates (8). As is typical of glutamine-dependent synthases, PLPS or the PdxS subunit can form PLP using exogenous ammonia, although glutamine is a more efficient nitrogen source (8,16).
The complex chemical reactions that occur in the PdxS active site have been the subject of intense investigation, and at least two covalent intermediates have been identified (Fig. 1B) (7,17). A PdxS catalytic lysine initially forms a Schiff base imine adduct with the R5P C1 atom in a manner that is independent of PdxT, glutamine, or G3P (7). The imine undergoes spontaneous isomerization to a stable amino ketone intermediate, here named I 1 (18,19). Ammonia, either exogenous or from glutamine hydrolysis, reacts with I 1 to release phosphate and water, forming a chromophore intermediate, here named I 2 , with an absorption maximum at 320 nm (16,17). Nitrogen addition from ammonia was shown to occur at the C2 atom, and a structure was proposed for I 2 , in which the covalent bond to the catalytic lysine is shifted from C1 to C5 (18,20). A third covalent intermediate (here named I 3 ), a Schiff base with the product PLP, has also been observed for B. subtilis PLPS (18). The I 2 chromophore has not been visualized directly; the binding site for G3P has not been identified, and the mechanism of the proposed C-N bond shift from C1 in I 1 to C5 in I 2 is unknown. Additionally, a second phosphate-binding site outside the synthase active site has been identified for PLP and is proposed to play a role in catalysis (11,19).
The overall architecture of PLPS (10,14,21) and the structures of PdxS (11,13,21) and PdxT (12,22) are well understood. The 24-subunit PLPS comprises 12 PdxS subunits and 12 PdxT subunits (14). PdxS forms a D6-symmetric cylindrical dodecamer of two opposing hexameric rings (13). The inner surface of the dodecameric cylinder is lined with the active sites of the 12 PdxS monomers, which possess the (␤/␣) 8 barrel fold (13). The 12 PdxT subunits dock at the outside of the PdxS cylinder with each PdxT active site facing the exterior end of a PdxS (␤/␣) 8 barrel. The hydrophobic interior of the PdxS (␤/␣) 8 barrel is thus inferred to be an ammonia channel, in an identical manner to imidazole glycerol phosphate synthase, another GAT (23). All crystal structures that include the PdxS synthase subunit lack 14 -28 amino acids at the C terminus (10,11,13,14,21). Despite their disorder, the last 21 amino acids are essential to PLP synthesis and become progressively ordered upon binding of PdxT and substrates (24).
To investigate the intriguing aspects of the PdxS mechanism, we characterized PLPS from G. stearothermophilus and obtained crystal structures for substrate complexes of PdxS alone (PdxS⅐I 1 and PdxS⅐I 1 /I 2 complexes) and of the intact enzyme (PLPS⅐I 1 ⅐Glu). The structures provide snapshots of PdxS at distinct steps in its complicated catalytic cycle and provide insights into the elusive mechanism and the structural elements that drive the conversion of I 1 to I 2 . PdxS alternates between open and closed conformations dependent on PdxT and substrates. Interaction of PLPS with glutamine and R5P induced FIGURE 1. PLP biosynthesis by the R5P-dependent pathway (PdxS/PdxT) in G. stearothermophilus. A, glutaminase reaction of PdxT. B, synthase reactions of PdxS. I 1 forms upon R5P addition to Lys-81, I 2 upon ammonia addition to I 1 , and PLP upon G3P addition to I 2 . A proposed structure (16) for the I 2 chromophore is shown. partial ordering of the C-terminal tail, which forms a lid over the R5P site in the closed state of the synthase subunit. Synthase closure is stabilized by several charged residues and appears to be essential for the formation of the I 2 chromophore. The structure of PdxS⅐I 1 /I 2 provides snapshots of PdxS from the I 1 to the I 2 state.

EXPERIMENTAL PROCEDURES
Cloning of G. stearothermophilus pdxS and pdxT-The plasmids pET0881 (13) for PdxS and pET0669 for PdxT were used for all mutageneses and subclonings. pET0669 was generated by subcloning the PdxT coding region into the pETTEV281 vector (13). The PdxT coding sequence was kindly provided by Dr. Boris Belitsky. PdxT H169N (here designated H169N T ) and some substitutions in PdxS (R53K, R53E, R53A, R53Q, E151S, E151A, E151Q, R288K, R288E, R288Q, R288A, D24N, D24A, D102A, R147K, R147A, and R147Q) were generated by sitedirected mutagenesis (QuikChange by Stratagene). PdxS truncations and substitutions made at 291 and 292 were generated by subcloning, and the remaining PdxS substitutions were generated by site-directed mutagenesis followed by subcloning and insertion into an appropriate expression vector. The pTrc-9 vector was constructed from pMCSG9 by cleaving pMCSG9 with restriction endonucleases SphI and NdeI. This removes the T7 promoter, lac operator, and ribosome-binding site sequences. These sequences were replaced by insertion of a synthetic DNA carrying the trp/lac fusion promoter trc (25), the translational enhancer sequence for the E. coli atp genes (26), and the ribosome-binding site (AGGAGG) spaced 8 bp from the ATG start sequence (27). All mutagenesis was verified by sequencing.
Expression and Purification of PdxS and PdxT-PdxS and PdxT were purified separately. Cells of Escherichia coli strain BL21(DE3) (or strain BL21 for pPdxSMBP_K149R and pPdxSMBP_K149Q) were transformed with the appropriate expression plasmid, grown at 37°C in either Luria-Bertani (LB) or Terrific Broth (TB) medium until A 600 reached 0.8 for LB and 1 for TB. The temperature was reduced to 20°C for 1 h, and expression was induced with either 200 M (TB) or 400 M (LB) isopropyl 1-thio-␤-D-galactopyranoside, and cultures were incubated 12-16 h. All purification steps were carried out at 4°C unless otherwise noted. Cells were harvested by centrifugation, resuspended in 40 ml of Buffer A (20 mM Tris-HCl, pH 7.9, 500 mM NaCl), lysed by sonication, and centrifuged at 18,000 ϫ g. The supernatant was filtered and loaded onto a Ni ϩ2 affinity column (HisTrap TM HP, GE Healthcare). Bound protein was eluted with a linear gradient of 0 -500 mM imidazole in Buffer A. The N-terminal His 6 tag of PdxS was removed by tobacco etch virus protease by 4 h digestion at room temperature followed by dialysis (Buffer A). Tobacco etch virus protease and uncleaved PdxS were removed by Ni 2ϩ affinity chromatography. Proteins intended for crystallization were further purified by gel filtration. PdxS was purified on an S300 column (HiPrep TM 16/60 Sephacryl TM S-300 HR, GE Healthcare) using 20 mM Tris-HCl, pH 8.5, 500 mM NaCl, 10 mM glutamine, 2 mM DTT. PdxT was purified on an S100 column (HiPrep TM 16/60 Sephacryl TM S-100 HR, GE Healthcare) using 20 mM Tris-HCl, pH 7.9, 500 mM NaCl, 2 mM DTT. Intact PLPS was reconsti-tuted by mixing PdxS and PdxT H169N in a 2:3 molar ratio in the presence of 10 mM glutamine, purified by gel filtration (S300 column with 20 mM Tris-HCl, pH 7.9, 500 mM NaCl, 10 mM glutamine, 2 mM DTT), dialyzed (20 mM Tris-HCl, pH 7.9, 10 mM glutamine), and stored at Ϫ80°C.
PLPS Assays-Published assays were used with slight modifications (8,17). Glutaminase activity was assayed in a coupled reaction with glutamate dehydrogenase. Reactions were carried out at 37°C in a total reaction volume of 250 l containing 50 M PdxS, 50 M PdxT, 100 mM Tris-HCl, pH 8.5, 20 mM glutamine, 50 mM KCl, 0.375 mM 3-acetylpyridine adenine dinucleotide, and 1.8 M glutamate dehydrogenase (Sigma). Glutamate formation was monitored by glutamate dehydrogenase reduction of 3-acetylpyridine adenine dinucleotide to the 3-acetylpyridine adenine dinucleotide reduced form detected at 363 nm using a SpectraMax M5 Multi-Mode Microplate Reader (Molecular Devices). Formation of PdxS⅐I 2 was assayed at 37°C in a total reaction volume of 50 l containing 50 M PdxS, 50 M PdxT, 50 mM Tris-HCl, pH 7.9, 20 mM glutamine, and 10 mM R5P. I 2 was detected by absorbance at 315 nm in a SpectraMax M5. PLP synthesis was assayed at 37°C in a total reaction volume of 50 l containing 50 M PdxS, 50 M PdxT, 50 mM Tris-HCl, pH 7.9, 20 mM glutamine, 10 mM R5P. After a 7-min incubation to form I 2 , PLP synthesis was initiated by addition of G3P (20 mM). PLP was detected by absorbance at 415 nm in a SpectraMax M5. I 2 formation and PLP synthesis for PdxS mutants at Lys-149 were assayed in a sub-micro quartz cuvette (Starna Cells, Inc.) with a total reaction volume of 100 l using an Ultrospec 3100 pro spectrophotometer (Amersham Biosciences). For determination of kinetic constants, 1 M PLPS was used, and the reactions were carried out as described above.
Intact Protein Mass Spectrometry-For detection of PdxS covalent intermediates, reaction mixtures with a total volume 100 l were incubated for 15 min at 37°C, quenched with 10% formic acid, and analyzed by LC-MS. Samples were eluted at a flow rate of 0.5 ml/min, with a linear gradient of 5-95% acetonitrile, over 15 min employing an Aeris widepore C4 column (3.6 m, 50 ϫ 2.10 mm) and directly analyzed by ESI mass spectrometry (Agilent 6520 Accurate Mass Q-TOF). Data were analyzed using the maximum entropy deconvolution algorithm within the Agilent Mass Hunter Qualitative Analysis software.
Assays of intermediate formation at Lys-81 in PdxS D24N, D24A, R147K, R147Q, and R147A were carried out with 150 M PdxS and 200 M PdxT, 20 mM Tris-HCl, pH 7.9, 10 mM R5P, 20 mM glutamine. Assays of adduct formation in wild type PLPS were carried out with 31.25 M PdxS, 60 M PdxT, 20 mM Tris-HCl, pH 7.9, 10 mM R5P, 0 -20 mM glutamine. Assays of adduct formation in wild type PdxS were carried out with 31.25 M PdxS, 10 mM R5P, 0 -50 mM (NH 4 ) 2 SO 4 , 20 mM Tris-HCl, pH 7.0; excess R5P was removed by gel filtration (PD-10 Desalting column, GE Healthcare); and the eluate was diluted to 1 mg/ml total protein prior to addition of (NH 4 ) 2 SO 4 and formic acid treatment. Assays of a covalent adduct of product PLP with PdxS were done under single turnover conditions according to a published protocol (18) in which the I 2 chromophore was formed by addition of R5P and ammonia to PdxS; excess substrates were removed during buffer exchange to phosphate, pH 7.4; G3P was added, and the protein was subject to MS analysis as described above.
X-ray Data and Structure Determination-Data were collected at the Advanced Photon Source on the GM/CA beam line 23ID-D and were processed and scaled with the HKL2000 suite (Table 1) (29). Initial phases for each structure were determined by molecular replacement. The structure of PLPS/ H169N T /R5P/Gln was solved using the program BALBES (30) and the structures of PdxS (13) (1ZNN) and PdxT (14) (2NV2). PdxS from the PLPS/H169N T /R5P/Gln structure was used to solve the PdxS/R5P (PdxS⅐I 1 ) structure using the program Phaser (31). The PdxS/R5P (PdxS⅐I 1 ) structure was then used to solve the PdxS/R5P/NH 3 (PdxS⅐I 1 /I 2 ) structure also using Phaser. Iterative model building was done in COOT (32), and refinement was carried out with REFMAC (33)(34)(35) in the CCP4 suite (36) and Buster (37). TLS groups for all three structures were determined using the webserver TLS MD (38,39). Link records for the covalent adducts PdxS Lys-81-I 1 , Lys-81-I 2 , Lys-149-R5P, and PdxT Cys-78 -Glu were created using J Ligand (40), and the restraints and coordinates were created using PRODRG (41) and the grade server (42). The final models were validated with MolProbity (43). The refinement statistics and the stereochemical quality of the final models are summarized in Table 1. Residues 47-56 are disordered in both PdxS structures. The C-terminal 23-25 residues are disordered in both PdxS structures; only residues 291-294 at the C terminus are missing in the PLPS structure. The N-terminal ϳ18 residues are disordered in the PdxS structures, but they become ordered by association with PdxT in the PLPS structure.

RESULTS
Kinetic Characterization of G. stearothermophilus PLPS-Several assays were used to evaluate PLPS activity, including glutamine hydrolysis by intact PLPS, chromophore formation by PLPS or PdxS, and PLP formation by PLPS and PdxS. Kinetic constants were determined for the glutaminase (K m ϭ 0.60 Ϯ 0.07 mM and k cat ϭ 0.060 Ϯ 0.001 min Ϫ1 ) and for PLP synthesis (R5P K m ϭ 10 Ϯ 2 M, G3P K m ϭ 1.06 Ϯ 0.4 mM, and k cat ϭ 0.010 Ϯ 0.005 min Ϫ1 ), and they yielded values roughly similar to the range reported for B. subtilis PLPS (8,19). PLPS displayed Michaelis-Menten behavior with no detectable cooperativity in any of its catalytic activities. The glutaminase activity of most GATs depends on or is accelerated by the synthase substrates (15) in a manner that tightly couples the glutaminase and synthase activities, efficiently generating ammonia only when the synthase substrates are present. In contrast, the PLPS glutaminase was neither dependent on nor accelerated by R5P or G3P (data not shown), but it was dependent on formation of intact PLPS from the glutaminase (PdxT) and the synthase (PdxS) subunit (8,14,16). The initial velocity was six times faster for glutamine hydrolysis than for PLP formation. This indicates weak coupling of the active sites under saturating conditions, which may not pertain in vivo. Although the synthase activity could be reconstituted using ammonia as a nitrogen source, the efficiency was 2-fold lower than with glutamine ( Fig. 2).
We attempted to capture distinct biochemical states of the synthase by crystallization of intact PLPS or the synthase subunit, PdxS, alone. Crystal structures were obtained at 2.7 Å for I 1 in PdxS (PdxS⅐I 1 ), at 2.3 Å for a mixture of I 1 and I 2 in PdxS (PdxS⅐I 1 /I 2 ), and at 2.7 Å for intact PLPS with I 1 in PdxS and the glutamate thioester intermediate in PdxT (PLPS⅐I 1 ⅐Glu) ( Table  1). These structures and additional probes by site-directed mutagenesis are presented below.
First Covalent Intermediate (I 1 )-We visualized the synthase active site with its first covalent intermediate (I 1 ) by co-crystallization of PdxS with R5P (Table 1 and Fig. 3). Covalent modification of PdxS in solution was confirmed by whole-protein mass spectrometry (MS) (Fig. 4A). The overall dodecameric structure is virtually identical to the structure of the PdxS free enzyme (13), although the crystal forms differ. Electron density corresponding to an R5P adduct at Lys-81 was present in the active sites of all six subunits in the crystallographic asymmetric unit (Fig. 3). The density was most consistent with formation of a Lys-81 adduct at the R5P C1 atom, based on the appearance of unbiased density and the distance between the lysine nitrogen atom and the I 1 phosphate. This differs from the C2 Schiff base adduct reported for T. maritima PLPS (10) but is consistent with NMR data (18). Refinement tests of both the 1-imino, 2-hydroxyl isomer and the 1-amino, 2-keto isomer resulted in a better fit to density for the amino-keto isomer, as expected (18). Lys-81 is at the top of strand ␤3 in the PdxS ␤ 8 /␣ 8 barrel. I 1 extends ϳ15 Å across the barrel with the phosphate near the N terminus of helix ␣8Ј. Despite formation of I 1 , the active site is open, perhaps due to the disorder of residues 49 -56, which were also disordered in the free enzyme structure (13). Other contacts of PdxS with I 1 include hydrogen bonds to phosphate oxygens from the amide nitrogens of Gly-153, Gly-214, and Gly-235. The conserved side chain of Asp-24 at the top of strand ␤1 is near the I 1 O2 atom. Asp-24 has been implicated in PLP synthesis (14), but no data address its function. The side chain is in position to shuffle protons during formation of an initial R5P Schiff base at Lys-81 and subsequent isomerization to the I 1 amino ketone. Consistent with this prediction, substitutions at Asp-24 were impaired for I 1 formation. I 1 did not accumulate in either D24A or D24N (Fig. 5, A and B), and I 2 formation was not detectable in the D24A variant and was negligible in the D24N variant (Table 2).
During crystallization of PdxS⅐I 1 , an R5P adduct also formed at Lys-149. We observed the doubly modified state in solution upon long term incubation of PdxS with an excess of R5P (Fig.  4B). Density for the adduct at Lys-149 was best fit with a connection to the R5P C2 atom (Fig. 3). In refinement tests, the 2-amino, 3-keto isomer at C2 (1-hydroxyl, 2-amino, 3-keto) was a better fit to density than were either the imine at C2 or the alternative amino-ketone at C2 (1-al, 2-amino, 3-hydroxyl).
When an R5P covalent intermediate was discovered (7), Lys-149 was identified as the site of the R5P adduct, but the site was reassigned to Lys-81 in light of a structure of PLPS co-crystallized with R5P (10). The Lys-149 adduct is unlikely to be on the reaction pathway given its long formation time and given that the Arg variant at this position (K149R) was competent for PLP synthesis, although at a reduced rate (Table 2). Lys-149 is at the top of strand ␤6 in the PdxS ␤ 8 /␣ 8 barrel with the side chain and R5P adduct directed outside the barrel away from the active site. Interestingly, in this position the adduct phosphate occupies an identical site as the phosphate of product PLP in a crystal structure of the yeast enzyme ( Fig. 3) (11), and the site is occupied by free phosphate or sulfate ions in several crystal structures. The phosphate has ionic contacts with Arg-137 and Arg-138, which are part of an extensive salt bridge network with Glu-103, His-115, Lys-118, and Glu-141. Catalysis was modestly reduced by mutagenesis of either Arg-137 or Arg-138, but double substitutions with Gln or Ala reduced both I 2 and PLP formation 10-fold. In contrast, substitutions at Lys-149 were deleterious to I 2 formation, and no PLP was formed without a positive charge at this position. Substitutions at other amino acids in the salt bridge network (Glu-103 and His-115) affected PLP synthesis only modestly (Table 2). A covalent adduct of product PLP with the synthase subunit of B. subtilis PLPS has been reported by two groups (18,19). We used singleturnover conditions and intact-protein MS to detect a covalent PLP adduct, but no such adduct accumulates with the G. stearothermophilus synthase subunit (data not shown).
Second Covalent Intermediate (I 2 )-A chromophoric intermediate (I 2 ) forms in the PdxS active site when ammonia

Structure of PLP Synthase
encounters I 1 (16,17). This can be accomplished by addition of glutamine to PLPS⅐I 1 or by addition of ammonia to PdxS⅐I 1 or PLPS⅐I 1 . PdxS⅐I 2 has a mass consistent with addition of ammonia to the R5P adduct and loss of phosphate and water (16,17). We used intact-protein MS to identify conditions for maximal accumulation of I 2 in either PdxS or PLPS, but all conditions yielded mixtures of I 2 , I 1 , and the free enzyme. In no condition was I 2 the predominant species (Figs. 4C and 6), consistent with previous results (16).
We were unable to obtain a structure of PdxS⅐I 2 in the context of intact PLPS, but replacement of lithium citrate with ammonium citrate in the conditions for crystallization of PdxS⅐I 1 produced crystals in a different form (Table 1 and Fig. 7) in which the 12 PdxS subunits in the crystallographic asymmetric unit included three states of the synthase active site: I 1 , I 2 , and no intermediate (Fig. 7). Thus, the 2.3-Å structure is a snapshot of PdxS in transition between I 1 and I 2 . I 1 is the predominant state in seven of the 12 subunits; two subunits contain  (11)). Key amino acids are shown as sticks in atomic coloring (blue, N; red, O; orange, P) with cyan, C, for PdxS residues hydrogen-bonded (yellow dashes) to ligands (white C for I 1 and R5P; green C for PLP). Residues in the extensive salt bridge network are also shown (orange C). The boundaries of disordered regions at residues 47-56 and the C terminus (271-294) are shown as cyan spheres.
density for I 2 with additional density for a phosphate ion disconnected from I 2 ; two subunits have density for a phosphate ion; and the 12th subunit has density consistent with five carbon atoms not covalently attached to Lys-81 as well as disconnected density for a phosphate ion. Density interpreted as I 2 is consistent with a five-carbon planar species, and the intermediate was modeled according to the proposed structure (16) shown in Fig. 1. In all subunits, the density for the intermediates was poorer than for the surrounding protein, probably due to a mixture of states in each subunit.   FEBRUARY 27, 2015 • VOLUME 290 • NUMBER 9

JOURNAL OF BIOLOGICAL CHEMISTRY 5233
The most significant change to the protein structure is the position of the Lys-149 side chain, which is directed into the active site where it may hydrogen bond with the O2 and/or O4 atoms of I 1 or the O3 atom of I 2 . Lys-149 does not interact with the phosphate nor with other polar side chains in the active site. Its position is supported by electron density in all 12 subunits, and rotations of the backbone add a hydrogen bond to the ␤5-␤6 connection (Thr-148 carbonyl and Ala-212 amide). This is a major change from the outward-pointing position of Lys-149 in all other structures of the PLPS synthase, whether from G. stearothermophilus or other sources. Crystallization conditions were similar (same buffer, same pH, and similar crystallization agent) for the two PdxS structures reported here and for our previous free enzyme structure (13). The free enzyme was crystallized with ammonia and no R5P, PdxS⅐I 1 with R5P and no ammonia, and PdxS⅐I 2 /I 1 with both R5P and ammonia, leading us to conclude that the inward conformation of Lys-149 is associated with formation of the chromophoric I 2 .
Apart from contacts with Lys-149, I 1 interacts with the enzyme identically in PdxS⅐I 1 /I 2 and in PdxS⅐I 1 , primarily through hydrogen bonds with the phosphate. In both structures, conserved residues Asp-102 (␤4) and Arg-147 (␤6) form a salt bridge beneath Lys-81-I 1 but do not contact the intermediate. Substitution of Asp-102 with alanine had a negligible effect on I 2 or PLP formation (Table 2). Conversely, mutagenesis of Arg-147 reduced I 2 formation and PLP synthesis more than 10-fold. Intact-protein MS revealed that Arg-147 substitutions prevented conversion of I 1 to I 2 (Fig. 5, C and D).
PLPS with PdxS⅐I 1 and PdxT-Glutamyl Thioester Intermediates-We obtained a crystal structure of intact PLPS (PdxS 12 PdxT 12 ) with two covalent intermediates, the glutamate thioester intermediate at Cys-78 in the glutaminase subunit (PdxT) and I 1 in the synthase subunit (PdxS). The crystal structure of the double covalent intermediate trapped the fully activated synthase active site and is informative about conformational changes that occur during the catalytic cycle (Figs. 8  and 9).
The glutamate thioester intermediate at the nucleophilic Cys-78 was stabilized by creating the inactive PdxT variant with Asn substituted for His-169 in the catalytic triad (PdxT/ H169N T ) (14). The thioester appears at full occupancy in all six PdxT subunits in the crystallographic asymmetric unit, as does I 1 at PdxS Lys-81 (Fig. 8B). Thus, the synthase active site is poised to receive ammonia. Formation of this synthase state apparently required both the glutaminase subunit (PdxT) and also catalysis in its active site, i.e. formation of the glutamate thioester intermediate. PdxS structural differences relative to the G. stearothermophilus PdxS-only structures (Figs. 3, 7, and 9A) (13), a B. subtilis intact PLPS structure (14), and a T. maritima intact PLPS (10) are limited to the synthase active site; there are no differences in subunit contacts or relative orientations among bacterial PLPS structures. Similarly, the glutaminase active site and subunit exhibit no changes relative to the B. subtilis PLPS structure with the glutamate thioester intermedi-  ate (14). Interaction of PdxT with PdxS sequesters the glutaminase active site from bulk solvent, in a manner that would slow the loss of labile ammonia generated in the glutaminase reaction. The ammonia generation site faces the entrance to the PdxS ␤ 8 /␣ 8 barrel at the opposite end from the synthase active site. The hydrophobic barrel interior is well suited to diffusion of uncharged ammonia to the synthase active site.
In contrast to the glutaminase active site, the synthase active site of the PLPS double covalent intermediate reflects a remarkable set of conformational changes relative to the other states of G. stearothermophilus PdxS. These changes sequester the site from bulk solvent and provide more specific contacts with I 1 (Fig. 8). Three elements of the synthase structure are changed. First and most dramatically, the C-terminal tail is partially ordered from residues 272 to 290, leaving only residues 291-294 without electron density. Second, an internal peptide (residues 49 -56) becomes ordered and forms a short helix (␣2a, residues 48 -54) over the active site. Finally, helix ␣8Ј (236 -240) shifts 3 Å toward the I 1 phosphate. These structural elements meet at the synthase active site and interact with one another.
The shift of helix ␣8Ј brings backbone amides at its N terminus (Gly-235 and Ser-236) within hydrogen bonding distance of the I 1 phosphate. The phosphate is thus enclosed by ␣8Ј and the loops containing Gly-153 and Gly-214, which also form hydrogen bonds with phosphate oxygens. An additional hydrogen bond is formed with the Ser-236 side chain. The shifted position of helix ␣8Ј is stabilized by the formation of helix ␣2a, which delivers the invariant Arg-53 side chain to the active site where it stabilizes the phosphate enclosure through hydrogen bonds with the Glu-151 side chain and the Pro-152 backbone carbonyl (Fig. 8C). The conserved "KGEPG" loop (residues Lys-149 -Gly-153) is central to this closure of the active site through interactions with phosphate and with Arg-53. The KEGPG loop is the only structural element that interacts directly with I 1 , helix ␣2a, and the C-terminal tail. Substitutions were made to disrupt the interaction between helix ␣2a (Arg-53) and the KGEPG loop (Glu-151) ( Table 2). The initial velocity of I 2 formation decreased significantly in comparison with the initial velocity of PLP synthesis. However, glutaminase activity was not affected (data not shown).
The C-terminal tail serves as a lid that covers helices ␣2a and ␣8Ј and the KGEPG loop, but it does not directly contact the covalent intermediate. The tail is not fully ordered, but combination of densities from the six independent subunits in the crystallographic asymmetric unit permitted all but the last four amino acids to be built into electron density (Fig. 8C). Tail residues Pro-285, Glu-286, and His-287 contact Arg-53 and the KGEPG loop.
The C-terminal tail interacts with the PdxS core in a zippered network of hydrogen bonds alternating between side chain and backbone atoms. For example, the His-287 side chain (tail) hydrogen bonds with the Gly-150 carbonyl (core, KGEPG). The Gln-290 side chain (tail) hydrogen bonds to the Glu-112 backbone carbonyl (core), and the His-115 side chain (core) hydrogen bonds to the Gln-290 backbone amide (tail). Additionally, the C-terminal tail hydrogen bonds to core residues residing on the adjacent subunit. Hydrogen bonds are formed between the carbonyl of Met-275 (tail) and the side chain of Arg-60 (helix ␣2a in the adjacent subunit), the carbonyl of Arg-276 (tail) and the amide nitrogen of Val-58 (neighboring helix ␣2a), and the carbonyl of Gly-155 (KGEPG loop) and the side chain of Arg-60 (neighboring helix ␣2a). Consistent with the lack of cooperativity, the orientation of each subunit relative to its neighbors is unchanged by active site closure. In all six monomers, density for the tail ends at Gln-290 (Fig. 8C). The Gln-290 backbone carbonyl is hydrogen-bonded to Arg-137 (core) at precisely the secondary phosphate site (phosphate of R5P in the adduct at Lys-149 shown in Fig. 3, sulfate ion in the PdxS free enzyme (13), phosphate ion in the Plasmodium PLPS structure (21), and PLP phosphate in the yeast synthase structure (11)). Thus, the ordered C-terminal tail is incompatible with phosphate-containing ligands in this site. Interestingly, the Lys-149 side chain is outside the active site and pointed toward the secondary phosphate site, unlike its position in the PdxS⅐I 1 /I 2 structure (Fig. 9B).
The full length of the C-terminal tail is essential to I 2 formation. We made a series of C-terminal truncations (PdxS⌬271-294, PdxS⌬279 -294, PdxS⌬287-294, PdxS⌬291-294, and PdxS⌬294) to evaluate the importance of the tail. All truncated variants had decreased I 2 formation and PLP synthesis, with a greater effect on I 2 formation in our initial velocity-based assay ( Table 2). The reduction in activity was correlated with the size of the truncation, but removal of even one amino acid (⌬294) reduced I 2 formation 5-fold and PLP synthesis 2-fold. The four C-terminal residues not observed in the PLPS⅐I 1 ⅐Glu structure were Glu-291-Arg-292-Gly-293-Trp-294, of which only Trp-294 and Arg-292 are conserved. Substitutions at Arg-292 had a greater effect on both chromophore and PLP formation compared with substitutions at the nonconserved Glu-291. Retaining the positive charge at position 292 (R292K) was least dele-terious. As expected, all truncations had wild type glutaminase levels (data not shown).

DISCUSSION
Our studies present new insights into the complex mechanism of the synthase subunit of PLPS and new information about the roles of three conserved synthase amino acids. Together with the structure of the G. stearothermophilus PLPS synthase subunit (13), the structures reported here provide snapshots of the synthase active site in four relevant states ( Fig.  1) as follows: the free enzyme (13); I 1 before formation of the glutaminase Michaelis complex (Fig. 3); I 1 after formation of the glutaminase Michaelis complex (Fig. 8); and I 1 /I 2 (Fig. 7). As the PLPS is from a single biological source, the structures are directly comparable. They are also highly complementary to the structures of various intermediate states of synthase homologs from T. maritima (10), Plasmodium berghei (21), and Pyrococcus horikoshii (44). These synthases have very similar I 1 states in which the intermediate extends across the active site from its covalent attachment at Lys-81 to the phosphate-binding site. I 1 is fixed in position at its ends, and the structures with strong density for the intermediate reveal a sole hydrogen bond between O2 and Asp-24, but none for O3 or O4. Apart from Asp-24, the active site between Lys-81 and the phosphate site is hydrophobic and spacious, with ample room for the substrate to change conformation in the complex reaction chemistry, particularly the proposed flip of the intermediate from a C1 to a C5 bond with Lys-81 (Fig. 1). We discovered that Asp-24, which was known to be critical for PLP synthesis (14), participates in the formation of I 1 . Substitutions at this position blocked accumulation of the intermediate (Fig. 5, A and B). The Asp-24 carboxylate is well positioned to assist in proton shuffling and dehydration during the initial formation of a Lys-81-R5P Schiff base and the subsequent isomerization to I 1 (Fig. 3). We also identified a role in formation of I 2 for conserved Arg-147, which lies below Lys-81 inside the (␤/␣) 8 barrel. Substitutions at Arg-147 had no impact on I 1 accumulation, but they blocked formation of I 2 upon addition of ammonia to I 1 (Fig. 5, C and D).
We visualized a mixture of states in the synthase active site following addition of ammonia to I 1 due to an inability to trap I 2 (Fig. 6). Crystals of this form of the synthase had 12 subunits in the asymmetric unit with differently populated active sites. In all subunits, either the I 1 phosphate or the free ion occupies the phosphate site. The Lys-149 side chain is directed into the active site of each subunit, in a position well supported by electron density (Fig. 7). Here, a comparison of the G. stearothermophilus synthase structures is important because under very similar crystallization conditions Lys-149 pointed into the active site in the presence of both R5P and ammonia, but it was directed outward in the presence of either substrate alone ( Fig.  3) (13). This is notable because Lys-149 points outside the active site in all other synthase structures (10,13,14,21,44). Thus, addition of ammonia to I 1 is the trigger for transfer of Lys-149 to the active site, where it is known to be critical for I 2 formation (Table 2) (17). We infer that after ammonia addition the synthase active site re-opens permitting entry of Lys-149.
The structure of intact PLPS with on-pathway intermediates in both the glutaminase and synthase active sites (PLPS⅐I 1 ⅐Glu) captures the enzyme as it is poised for ammonia attack on the synthase I 1 intermediate. This state has a closed synthase active site where the C-terminal tail and helices ␣2a and ␣8Ј together bury the I 1 intermediate. The closed synthase active site is an ideal environment to sequester labile ammonia en route to I 2 .
Ordering of helix ␣2a is a key part of active site closure, as it is disordered in the other G. stearothermophilus synthase structures. The ordering of helix ␣2a (residues 49 -56) and the C-terminal tail (270 -294) may be coupled, as part of the tail (275-276 backbone) contacts residues 58 and 60 in an adjacent subunit. This is consistent with the observation that the B. subtilis synthase C terminus is protected from proteolysis by addition of the glutaminase subunit and substrates (24). Although the final four amino acids (291-294) remained disordered in PLPS⅐I 1 ⅐Glu, these residues, particularly conserved Arg-292, are critical for the formation of I 2 and PLP ( Table 2). Strong electron density for the C-terminal tail ends precisely at the secondary phosphate site outside the (␤/␣) 8 barrel (Fig. 3,  inset). A role has been proposed for phosphate binding to this site during the catalytic cycle (19). Alternatively, the site may be a transient anchor for the synthase terminal carboxylate, perhaps to assist in shifting Lys-149 into the active site. Ordering of the C-terminal tail and closure of the synthase active site appear to have required both a Michaelis-like complex in the glutaminase and I 1 in the synthase, based on structures with the glutaminase intermediate in B. subtilis PLPS (14) and with I 1 in the P. horikoshii PLPS (44), both of which had an open synthase active site and a disordered C-terminal tail. FIGURE 9. Conformational changes in the synthase active site. A, closure around I 1 in PLPS⅐I 1 ⅐Glu. Superposition of PLPS⅐I 1 ⅐Glu (red) and PdxS⅐I 1 /I 2 (teal) illustrates how synthase closure forms helix ␣2a and shifts helix ␣8Ј toward I 1 relative to its position in the open structures, here represented by PdxS⅐I 1 /I 2 . The disorder of amino acids 48 -56 in PdxS⅐I 1 /I 2 is indicated by yellow dashes. B, comparison of Lys-149 positions. Lys-149, located on the invariant KGEPG loop, points into the active site in PdxS⅐I 1 /I 2 (teal) and outside the active site in all other structures, here illustrated by superposition with PLPS⅐I 1 ⅐Glu (red). The KGEPG loop also shifts with movement of Lys-149 into the active site. A phosphate ion is shown in the secondary site.
Several key questions remain concerning the synthase mechanism. Formation of the I 2 chromophore remains mysterious. The chemical steps to its formation have been proposed (17,20) but not demonstrated. A proposed structure of I 2 is indirectly supported by the structure of an acid breakdown product (18) and by the electron density in PdxS⅐I 1 /I 2 (Fig. 7), but neither is definitive. The mode of interaction of substrate G3P with the synthase is also unknown. We did not capture a G3P complex with either the free enzyme or the I 1 intermediate, and it is likely that G3P interacts only with the I 2 intermediate state. To trap a G3P complex, we searched for a residue that is required for PLP synthesis but not for formation of I 1 or I 2 . No conserved amino acid in the vicinity of the active site had this property. Given the instability and hence reactivity of I 2 , G3P binding near the intermediate will likely result in PLP formation with no catalytic assistance from the enzyme. We considered two possibilities for a G3P-binding site. The first, on the outside of the (␤/␣) 8 barrel, makes use of the secondary phosphate-binding site. G3P in this site would require an assist, perhaps through covalent attachment to Lys-149, to move into the active site upon I 2 formation. The observed shift of Lys-149 into the active site in PdxS⅐I 1 /I 2 makes this an attractive hypothesis. However, substitution of Arg for Lys-149 did not prevent PLP synthesis, arguing against this possibility. A more probable scenario is for the G3P phosphate to replace the free phosphate that forms upon ammonia attack in the I 1 to create I 2 , thereby binding in an optimal position for reaction with I 2 . This would also explain our (and others) inability to trap a meaningful PLPS⅐G3P complex.
This study also leaves open the classic question for all glutamine amidotransferases about how the synthase and glutaminase active sites communicate to couple their activities. For PLPS, the specific question is how engagement with a Michaelis complex of the glutaminase subunit leads to closure of the synthase active site in the I 1 state. The now four crystal structures of G. stearothermophilus PdxS provide no clues. Crystal structures of other PLPS enzymes (10,14,21) or synthase dodecamers (11,44) are equally opaque on this point. PLPS is the only characterized GAT in which the glutaminase activity does not depend on, nor is it accelerated by, the synthase substrates but depends only on the formation of the intact PLPS (8). We note that the PLPS substrates (glutamine, R5P, and G3P), unlike those of other GATs, are readily available metabolites and that glutamine may be the limiting substrate in vivo.
Up to 1.5% of prokaryotic genes encode for PLP-dependent enzymes; therefore, the demand for PLP as a cofactor is immense (45). The remarkable single-enzyme PLP biosynthetic pathway is a marvel of enzyme catalysis. Our studies provide additional glimpses into the mechanism, but there remains much to learn.