Structure and Catalytic Mechanism of Eukaryotic Selenocysteine Synthase*

In eukaryotes and Archaea, selenocysteine synthase (SecS) converts O-phospho-l-seryl-tRNA[Ser]Sec into selenocysteyl-tRNA[Ser]Sec using selenophosphate as the selenium donor compound. The molecular mechanisms underlying SecS activity are presently unknown. We have delineated a 450-residue core of mouse SecS, which retained full selenocysteyl-tRNA[Ser]Sec synthesis activity, and determined its crystal structure at 1.65Å resolution. SecS exhibits three domains that place it in the fold type I family of pyridoxal phosphate (PLP)-dependent enzymes. Two SecS monomers interact intimately and together build up two identical active sites around PLP in a Schiff-base linkage with lysine 284. Two SecS dimers further associate to form a homotetramer. The N terminus, which mediates tetramer formation, and a large insertion that remodels the active site set SecS aside from other members of the family. The active site insertion contributes to PLP binding and positions a glutamate next to the PLP, where it could repel substrates with a free α-carboxyl group, suggesting why SecS does not act on free O-phospho-l-serine. Upon soaking crystals in phosphate buffer, a previously disordered loop within the active site insertion contracted to form a phosphate binding site. Residues that are strictly conserved in SecS orthologs but variant in related enzymes coordinate the phosphate and upon mutation corrupt SecS activity. Modeling suggested that the phosphate loop accommodates the γ-phosphate moiety of O-phospho-l-seryl-tRNA[Ser]Sec and, after phosphate elimination, binds selenophosphate to initiate attack on the proposed aminoacrylyl-tRNA[Ser]Sec intermediate. Based on these results and on the activity profiles of mechanism-based inhibitors, we offer a detailed reaction mechanism for the enzyme.

Organisms that co-translationally incorporate selenocysteine (Sec) 2 into proteins in response to UGA codons in mRNA are frequently encountered in all three domains of life, prokaryotes, Archaea, and eukaryotes. The human genome contains 25 known genes encoding selenoproteins (1), several of which have essential functions. Selenoprotein biosynthesis requires two specialized metabolic pathways, the first to synthesize Sec and the second to incorporate Sec into proteins. The benefits of selenoproteins, e.g. their unique redox and catalytic properties, apparently outweigh selenium toxicity and the burden of maintaining intricate Sec synthetic and decoding machineries.
Notwithstanding some common themes, both Sec synthesis and decoding (for reviews, see Refs. [2][3][4] differ in bacteria and eukaryotes, and archaeal selenoprotein biosynthesis largely follows the eukaryotic schemes (5). In all organisms a special tRNA [Ser]Sec bearing an anticodon complementary to the UGA codon is central to both processes (6,7). All tRNA [Ser]Sec species exhibit a number of non-canonical features whereby they are exclusively utilized for the Sec synthesis and Sec insertion pathways (8).
For Sec insertion at UGA codons, selenocysteyl-tRNA [Ser]Sec is recognized by a special elongation factor, SelB in bacteria (9) and EFSec in eukaryotes (10,11), that replaces EF-Tu and EF-1␣ in escorting the aminoacylated Sec-tRNA [Ser]Sec to the ribosome. Bacterial SelB binds a stem-loop structure located within the coding region of selenoprotein mRNAs and directly downstream of a UGA codon (12). Such stem-loop structures thereby act as Sec insertion sequence elements that recode UGA, which normally signals translational termination. In contrast to bacteria, eukaryotic Sec insertion sequence elements are located in the 3Ј-untranslated regions of selenoprotein mRNAs (13), alleviating the burden of having to maintain functional secondary structures within the coding region and facilitating insertion of more than one Sec into a single protein chain. Such recognition from a distance requires a special adaptor protein, Sec insertion sequence-binding protein 2 (14). Additional factors appear to be involved in UGA recoding in higher organisms. For example, SECp43 was recently found to associate with a Sec-tRNA [Ser]Sec ⅐EFsec complex in vitro and to enhance the interaction between EFsec and Sec insertion sequence-binding protein 2 in vivo (15). SECp43 also influences the specific post-transcriptional modification of tRNA [Ser]Sec (16). Furthermore, a role for ribosomal protein L30 in Sec decoding on the ribosome has recently surfaced (17).
The molecular identity of eukaryotic/archaeal SecS has only recently been elucidated. An archaeal open reading frame, annotated as SecS, did not act on Ser-tRNA [Ser]Sec or PSer-tRNA [Ser]Sec (25). RNA-mediated interference technology provided the first direct evidence for an essential role of soluble liver antigen/liver and pancreas antigen (SLA/LP) in selenoprotein biosynthesis (16). SLA/LP was originally identified as the target of autoantibodies from patients with a severe form of autoimmune chronic active hepatitis (26,27). Indeed, SLA/LP interacted with SECp43 and tRNA [Ser]Sec as detected by coimmunoprecipitation (16,26,28). Structural homology modeling predicted that SLA/LP is a PLP-dependent enzyme of the aspartate aminotransferase family (29). Unequivocal evidence that SLA/LP embodied the elusive eukaryotic/archaeal SecS was finally provided independently by two groups (28,30) who directly demonstrated the conversion of PSer-tRNA [Ser]Sec to Sec-tRNA [Ser]Sec by the enzyme.
The identification of eukaryotic/archaeal SecS paved the way for understanding the molecular mechanisms underlying its catalytic activity. Xu et al. (28) showed that mouse SecS is able to dephosphorylate PSer-tRNA [Ser]Sec , indicating that aminoacrylyl-tRNA [Ser]Sec is a likely intermediate in the reaction. It was also unequivocally established that SecS employs SeP produced by selenophosphate synthetase 2 as the activated selenium donor (24,28). Apart from these aspects, the enzyme is presently enigmatic. In particular, it is not known (i) how SecS recognizes its two substrates (PSer-tRNA [Ser]Sec and SeP) and whether binding occurs concomitantly or sequentially, (ii) how SecS differs from related enzymes that recognize low molecular weight amino acid substrates, (iii) whether and how the enzyme discriminates against free PSer, and (iv) which residues participate in PSer to Sec conversion on tRNA [Ser]Sec .
Here we report on a combined structural and biochemical analysis of a mammalian SecS which illuminates the above questions. We determined high resolution crystal structures of SecS from mouse (mmuSecS) in which we find that the active site is constructed around a Lys-284-bound PLP cofactor at the interface of two protomers of a close dimer. A SecS-specific N terminus leads to association of two dimers into a homotetramer, which has possible relevance for PSer-tRNA [Ser]Sec positioning. The active site is complemented by a SecS-specific loop, which is disordered in the absence of ligands but contracts in the presence of a substrate-mimicking phosphate moiety. A glutamate neighboring the PLP cofactor is ideally positioned to deter substrates bearing a free ␣-carboxyl group. In line with the structural results, we show that changes of phosphate-coordinating residues, which are solely conserved in SecS orthologs, lead to reduced activity of the enzyme and that SecS does not act on free PSer. Using in addition the response of SecS toward mechanism-based inhibitors, we propose a detailed catalytic mechanism.

EXPERIMENTAL PROCEDURES
Protein Production and Site-directed Mutagenesis-A DNA fragment encoding the full-length secS gene from mouse (Gen-Bank TM accession number NM_172490) was amplified and cloned into pETM-13 vector to allow the expression of a C-terminal His 6 -tagged protein. The insert was verified by DNA sequencing. The resulting plasmid was termed pETM-13-secS.
Rosetta2(DE3) cells were transformed with the pETM-13-secS expression construct. Overproduction of the target protein was carried out at 289 K using auto-inducing medium (31). Cells were harvested when the maximum culture density was reached and resuspended in buffer A (50 mM HEPES-NaOH, pH 7.5, 500 mM NaCl, 20 mM imidazole, 2 mM ␤-mercaptoethanol). 1 mg of lysozyme and 3 l of DNase (1 mg/ml) were added per gram of wet cells, and the mixture was incubated for 30 min on ice with stirring. After completing cell rupture by sonication and removing cell debris by centrifugation, soluble fusion protein was captured on a nickel-nitrilotriacetic acid-Sepharose column (Qiagen), washed with buffer B (50 mM HEPES-NaOH, pH 7.5, 1 M NaCl, 20 mM imidazole, 2 mM ␤-mercaptoethanol), and eluted with a linear gradient of imidazole (20 to 250 mM) in buffer A. Fractions containing mmuSecS were pooled, concentrated, and further purified by gel filtration on a Superdex-200 HiLoad 26/60 column (GE Healthcare) equilibrated with buffer C (10 mM HEPES-NaOH, pH 7.5, 500 mM NaCl, 2 mM dithiothreitol). Purified target protein was concentrated to 14 mg/ml using Vivaspin 15 concentrators (30,000 MWCO; Sartorius Vivascience), separated into aliquots, flash-frozen in liquid nitrogen, and stored at 193 K until use. Using the above protocol we obtained ϳ1.5 mg of mmuSecS per liter of bacterial culture at a purity of greater than 95% as estimated by SDS-PAGE analysis (see Fig. 1A).
Mutants of mmuSecS Arg-313 and mmuSecS Gln-105 were generated in pETM-13-secS via the QuikChange protocol (Stratagene). The mutations were verified by DNA sequencing. All mutants of mmuSecS were purified in the same way as the wild type protein.
Limited Proteolysis and Analytical Gel Filtration Analysis-50 l of mmuSecS (14 mg/ml) were incubated with 3 g of elastase at 277 K for 80 min. The major band on an SDS gel originating from elastase treatment (mmuSecS elast ; see Fig. 1A) was in-gel-digested with trypsin, and fragments were analyzed by matrix-assisted laser desorption ionization-mass spectroscopy as described (32).
In Vitro Transcription and Filter Binding Assay-The template for in vitro transcription was prepared as described (21). Uniformly 32 P-labeled tRNA [ 10,000 cpm (100 fmol) of 32 P-labeled unacylated tRNA [Ser]Sec were incubated 30 min on ice with 2.5, 5, 10, or 20 M mmuSecS or mmuSecS elast in 20 l of buffer D (HEPES-NaOH, pH 7.5, 200 mM NaCl, 5 mM MgCl 2 , 1 mM dithiothreitol, 0.5 mg/ml total E. coli tRNA). 10-l aliquots of the reaction mixtures were loaded on a Protran BA 83 nitrocellulose membrane (Whatman) and washed with 50 ml of buffer C. For the detection of RNA-protein complexes, the membrane was exposed to a PhosphorImager screen overnight, which was then scanned using a Typhoon 8600 (GE Healthcare).
SecS Activity Assay-Synthetic tRNA [Ser]Sec was used in all reactions for assaying SecS activity. Synthetic tRNA [Ser]Sec was aminoacylated with serine by seryl-tRNA Ser synthetase, the seryl moiety was phosphorylated (21), and the PSer-tRNA [Ser]Sec was isolated (28). The extent of serylation of tRNA [Ser]Sec in the presence of seryl-tRNA Ser synthetase, serine, and other reaction components was 80 -90%, and subsequent phosphorylation in the presence of PSer-tRNA [Ser]Sec kinase, ATP, and other reaction components neared 100% (Refs. 21 and 28 and references therein). 3 The preparations of seryl-tRNA Ser synthetase and PSer-tRNA [ The activities of mmuSecS elast and of the mmuSecS point mutants relative to that of the full-length wild type mmuSecS were determined. Thioredoxin (Trx) was used as a negative control. Sec synthetic reactions were carried out as described (28). The selenium donor, SeP, was generated from chemically synthesized [(CH 3 ) 3 SiO] 3 PSe (28), which was a generous gift of Dr. Richard Glass, and 0.2 mM SeP was used in all Sec synthesis assays.
Dephosphorylation Activity of mmuSecS-Dephosphorylation of PSer-tRNA [Ser]Sec was carried out as described (28) with the following modifications. In the inhibition assays, mmuSecS was incubated with different concentrations of propargylglycine (PG) or trifluoroalanine (F 3 -Ala) for 3 min at room temperature in a 10-l volume of buffer (28), 100 ng of 32 P-labeled and 1.0 g of cold PSer-tRNA [Ser]Sec were added to bring the total volume to 20 l, and the reaction was incubated at 310 K for 15 min. To assess the activity of mmuSecS on free PSer, a stock solution of [ 32 P]PSer was prepared by deacylating 0.5 g of [ 32 P]PSer-tRNA [Ser]Sec in 25 l at pH 9.0 and 315 K for 1 h. 5 l of [ 32 P]PSer were added to each of the mmuSecS reactions and then incubated for 15 min at 310 K. 10 units of alkaline phosphatase were used to dephosphorylate [ 32 P]PSer, and the resulting phosphate was used as a positive control. Reactions were prepared for chromatography and chromatographed as given (28).
Crystallographic Procedures-For crystallization, 1.5 l of mmuSecS elast (14 mg/ml in buffer C) were mixed with an equal volume of reservoir solution (11% [v/v] ethylene glycol without other buffer components) in a 24-well Cryschem plate (Hampton Research) immediately after elastase treatment. Crystals were grown by sitting-drop vapor diffusion at 293 K. They appeared overnight and continued to grow for the next 2 days. For data collection, crystals of mmuSecS elast were shock-frozen in a 100 K nitrogen stream (Oxford Cryosystems) after transfer into a cryo-protecting buffer (100 mM HEPES-NaOH, pH 7.5, 250 mM NaCl, 1 mM dithiothreitol, 35% (v/v) ethylene glycol). All data were recorded on a Bruker-Nonius FR591 rotating anode generator producing CuK␣ X-radiation ( ϭ 1.54179 Å) at 45 kV and 100 mA equipped with Osmic mirrors and a MAR345 image plate (MARResearch). Data were processed with the HKL package (33) ( Table 1).
The structure of mmuSecS elast was solved via a single isomorphous replacement with anomalous scattering strategy. Crystals were soaked for 30 s in cryo-protecting buffer supplemented with 0.5 M sodium iodide and immediately shockfrozen in a 100 K nitrogen stream. Friedel pairs were kept separate during data reduction ( Table 1). The iodide-soaked crystals proved isomorphous to the native crystals. 42 iodide positions were found by using SHELXD (34) and used for initial phase calculations (Table 1). Solvent flattening with SHELXE clearly indicated the correct hand of the heavy atom substructure (Table 1) and yielded a high quality experimental electron density map (supplemental Fig. S1).
424 of the 438 residues located in the final structure were positioned in the first round of automatic model building with ARP/wARP (35). The structure was completed by manual model building and automatic refinement with Refmac5 (36). Water molecules were automatically placed with ARP/wARP. One solute species was identified as a chloride ion based on the presence of 250 mM NaCl in the crystallization buffer and residual positive electron density after placement of a water molecule. Another solvent molecule was interpreted as an ethylene glycol molecule originating from the crystallization or cryoprotecting buffer. TLS refinement (37) was conducted to model differential global anisotropic displacements of the three domains of mmuSecS elast . During all stages of refinement, a randomly selected set of 5% of the reflections was used for cross-validation ( Table 1). The iodide-soaked crystal structure was refined by the same strategy including the 42 iodide ions located by SHELXD. Additional, lower occupancy iodide positions were found in an anomalous difference Fourier map, obtained with phases calculated from the final refined native structure and the anomalous differences measured for the iodide data set.
For monitoring of phosphate binding, crystals were soaked for 1 min in cryo-protecting buffer supplemented with 0.5 M sodium phosphate, pH 7.5. Similarly, crystals could be derivatized with sulfate (not shown). Data were collected as described, and the structure of a phosphate-soaked crystal was solved by molecular replacement with MOLREP (38) using the structure coordinates of the native protein as a search model while omitting the solvent structure, the PLP cofactor, and alternative side chain conformations. Model building and refinement were conducted as described for the native protein (Table 1). Coordinates and structure factors have been submitted to the Protein Data Bank.

Limited Proteolysis Delineates a SecS Core Fully Active in PSer-tRNA [Ser]Sec to Sec-tRNA [Ser]Sec
Conversion-We expressed, purified, and crystallized full-length SecS from mouse (mmuSecS), but the crystals diffracted only to ϳ6 Å resolution. To explore the possibility that flexible regions hindered generation of well ordered crystals, mmuSecS (ϳ55 kDa) was digested with various proteases. Elastase gave rise to a stable fragment of about 49 kDa (Fig. 1A). Tryptic mass spectrometric fingerprinting showed that the elastase-resistant fragment encompassed residues 19 -468 (not shown). Thus, the protease removed most of the C-terminal portion that carries the main SLA/LP antigenic epitope (26,27), also lacking in archaeal SecS (Fig. 2). We refer to the elastase-resistant core of the enzyme as mmuSecS elast .
Gel filtration analysis revealed that mmuSecS is tetrameric in solution (Fig. 1B). Elastase treatment did not change the migration behavior of the protein detectably, showing that the proteolytic treatment left the quaternary structure of the enzyme intact (Fig. 1B). Because mmuSecS binds significantly to unacylated tRNA [Ser]Sec (28), we examined the tRNA [Ser]Sec binding activity of mmuSecS elast by nitrocellulose filter binding. The elastase-treated protein bound tRNA [Ser]Sec with an affinity and specificity comparable with those of the full-length enzyme (Fig. 1C). We next tested the Sec synthesis activity of mmu-SecS elast using a paper chromatographic assay (28). After removal of elastase by gel filtration chromatography, mmuSecS elast was incubated with buffer containing SeP and PSer-tRNA [ (21,28). As positive and negative controls, mmuSecS and Trx, respectively, were substituted for mmu-SecS elast . The efficiency at which mmuSecS elast converted PSer-tRNA [Ser]Sec to Sec-tRNA [Ser]Sec was indistinguishable from that of the full-length enzyme (Fig. 1D). The above results demonstrate that mmuSecS elast structurally and functionally closely resembles the full-length enzyme and constitutes a suitable platform on which to explore the structure-activity relationships of SecS-dependent Sec biosynthesis.
SecS Is a Member of the Fold Type I Family of PLP-dependent Enzymes with Distinct Structural Elements-mmuSecS elast crystallized readily after the addition of 11% (v/v) ethylene glycol at room temperature. The crystals diffracted to high resolution on a rotating anode x-ray generator and could be derivatized by quicksoaking in 0.5 M sodium iodide for structure solution by single isomorphous replacement with anomalous scattering (Table 1; Fig. S1). Refinement converged at R/R free factors of 16.8/19.8% with the final model exhibiting good overall stereochemistry ( Table 1). The only amino acids lacking well defined electron density were residues 19 -22 at the N terminus, residue 468 at the C terminus, and residues 98 -104, constituting a flexible loop. Structural homology searches (39) suggested that mmu-SecS elast exhibits significant similarity to the fold type I family of PLP-dependent enzymes (also referred to as the aspartate aminotransferase family) (40). Where appropriate, we will compare the structure of mmuSecS elast to those of Archaeoglobus fulgidus PSer-cysteine synthase (afuPSerCysS) (41), members of the NifS family of Cys/Sec lyases (42)(43)(44)(45)(46), and the cystine C-S lyase C-DES from Synechocystis (synC-DES) (47). A quantitative comparison with these proteins is given in Table 2. These enzymes act or can act on related substrates and may share some catalytic properties with SecS. In particular, PSerCysS from methanogenic Archaea affords a precedence for the tRNA-based amino acid synthesis via a PSer-tRNA intermediate. In these organisms the sole pathway for cysteine biosynthesis is via PSer-tRNA Cys , obtained by direct aminoacylation of tRNA Cys with PSer by PSer-tRNA Cys synthetase, and subsequent PSer-tRNA Cys to Cys-tRNA Cys conversion by PSerCysS (48). mmuSecS elast can be divided into three domains (Fig. 3, A and B). Domain 1 (blue scaffold in Fig. 3, A and B) is a composite of residues 23-130 and 313-330 and is purely ␣-helical (encompassing helices ␣1-␣4 and ␣12). The N terminus (residues 23-44; element I in Fig. 3, A and B) differs from that of other fold type I enzymes (Fig. 3, C and D). In mmuSecS elast , helix ␣1 is positioned at the protein surface, running approximately perpendicular to the scaffolding helices of the domain. Domain 1 also exhibits a long insertion between helices ␣2 and ␣4 (residues 62-108; element II in Fig. 3, A and B), which encompasses two loops separated by helix ␣3. The corresponding element in other enzymes of the fold type I family is significantly shorter (Fig. 3, C and D). Seven residues within the second loop of the insertion (residues 98 -104) are disordered due to intrinsic flexibility (bordered by spheres in Fig. 3A). Both the non-canonical N terminus and the unique insertion are conserved among SecS orthologs (Fig. 2) and, therefore, are expected to confer unique functions on the enzyme. Domain 2 of SecS is the largest module of the protein (residues 131-312; cyan scaffold in Fig. 3, A and B). It comprises a ␣/␤/␣ sandwich fold encompassing a sevenstranded ␤-sheet (␤1-␤9-␤8-␤7-␤6-␤2-␤3) characteristic of the fold type I family. The ␤-sheet is parallel except for strand ␤9 (Fig. 3, A and B). A short helix (␣10) between strands ␤8 and ␤9 carries a PLP cofactor in Schiff-base linkage to Lys-284 (Figs. 3A and 4). Helices ␣7, ␣8, and ␣9 line the ␤-sheet at the convex outside, and helices ␣5, ␣6, and ␣11 lie at the concave inside. A short ␤-hairpin (␤4 and ␤5) is inserted between strand ␤3 and helix ␣7. Domain 2 is connected to the second part of domain 1 by a short loop (residues 311-314; element III in Fig. 3, A and B).
Strands ␤11 and ␤12 thereby form one rim of an active site funnel leading from the surface to the PLP (Fig. 3A). In the NifS-like enzymes, the analog of the long domain 3 loop (element IV) is often disordered (43) and bears a conserved Cys that can be charged in the active site with elemental sulfur. The resulting persulfide is thought to donate a S 0 building block for iron-sulfur cluster biosynthesis. Element IV of mmuSecS elast does not contain a Cys.
Cross-strutting via the N Nerminus Leads to Homotetramers That Exhibit Surface Properties Suitable for Binding PSer-tRNA [Ser]Sec -mmuSecS elast crystals contained one protein molecule per asymmetric unit. Consistent with the gel filtration analysis, the orthorhombic crystal symmetry gave rise to tetramers in which the protomers are related by three orthogonal 2-fold axes (see Fig. 5A, left). Within a tetramer, two pairs of monomers (Mol I/II and Mol III/IV; Fig. 5A, left) interact intimately, burying 7343 Å 2 of combined surface area upon association. Two of these close dimers further associate into tetramers via less extensive interactions between Mol I and Mol III, viz. Mol II/Mol IV (1891 Å 2 combined surface area buried in each contact), and between Mol I and Mol IV, viz. Mol II/III (276 Å 2 combined surface area buried in each contact; Fig. 5A, left). The tetramers are held together by the formation of a short antiparallel coiled-coil between the ␣2 helices of Mol I and III (Mol II and IV), which is cross-strutted by helices ␣1 (Fig. 5B). In contrast, afuPSerCysS, NifS relatives, or synC-DES lack the surface-exposed N terminus (elements I in Fig. 3, C and D). Consistently, all these latter proteins exist as dimers. Fig. 5C shows the electrostatic potential mapped to the surface of a mmuSecS elast tetramer. Large patches of positive charge (blue) are visible, consistent with the overall basic pI of 8.3 calculated for the protein. In particular, the funnel leading to the active site is strongly positively charged. Therefore, the surface properties of mmuSecS elast appear to be designed to contact the sugar-phosphate backbone of tRNA [Ser]Sec at multiple positions. Consistent with this view, we observed avid binding of anions to mmuSecS elast . After soaking with NaI, we located 62 iodide ions per protomer bound to the surface of mmuSecS elast (Fig. 5D). One of these positions was always occupied by a chloride ion in structures not treated with iodide (not shown).
The PLP Cofactor Is Tightly Anchored by Non-canonical Contacts to Both Protomers of a Close Dimer-A close dimer exhibits two identical active sites at the protomer interfaces pinpointed by a PLP cofactor (Fig. 4A). Because the PLP was refined at full occupancy, leaving no residual difference density, all four subunits of a tetramer bear a cofactor. This situation is different from afuPSerCysS, where only one of two potential active sites in a dimer was equipped with PLP (41). We refer to the PLP attached to Lys-284 of a reference molecule and the surrounding active site as "cis"; the PLP attached to the opposite protomer and its surrounding active site are referred to as "trans." In mmuSecS elast , both monomers of a close dimer contribute side chains for PLP binding in an active site (Fig. 4A). Apart from the covalent linkage to Lys-284, PLP is additionally bound via multiple hydrogen bonds and electrostatic and van der Waals interactions in cis. These interactions FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9 JOURNAL OF BIOLOGICAL CHEMISTRY 5853 exclusively involve residues from domain 2. The PLP phosphate group is positioned over the N terminus of helix ␣5, interacting favorably with the helix macro-dipole and engaging in hydrogen bonds to the backbone amides of Thr-144 and Gly-145. The pyridine nitrogen maintains hydrogen bonds to the side chains of Cys-175 and Asn-252. Asn-252 is at variance with the vast majority of fold type I enzymes, in which an Asp at the equivalent position is the only strictly conserved residue apart from the PLP-bound Lys (40). An Asn is expected to support the electron sink character of the pyridine ring less than an Asp, possibly demanding a good leaving group such as a phosphate on the ␤-carbon. Interest-   FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9 JOURNAL OF BIOLOGICAL CHEMISTRY 5855 ingly, in afuPSerCysS, in which phosphate is also the leaving group, the pyridine nitrogen is again bound to an Asn (Fig. 2) (41). The PLP pyridine ring of mmuSecS elast is sandwiched between the side chains of Gln-172 and Ala-254 on the re and si faces, respectively. Archaeal SecS enzymes feature a His in place of Gln-172 (Fig. 2). A similar His in a NifS-like protein from Thermotoga maritima has been discussed as a tunable acid-base catalyst in the reaction mechanism (43). Thus, the enzymatic mechanisms of archaeal and eukaryotic SecS may differ in detail.

Structure and Mechanism of Eukaryotic Selenocysteine Synthase
Interactions with PLP in trans (Fig. 4, A and B) involve residues from the SecS-specific insertion in domain 1 (element II) and from the short element III connecting domain 2 and the second part of domain 1 (Fig. 3, A and B). Both elements primarily interact with the PLP phosphate. Arg-75 (originating from the first loop of element II) is deposited on the phosphate side of the PLP pyridine ring where its side chain can engage in two charged interactions with anionic phosphate oxygens. Arg-75 is appropriately positioned by a double salt bridge interaction with Asp-283 from the cis protomer (Fig. 4, A and B). The preceding residue of element II, Glu-74, comes to lie on the opposite side of the pyridine ring and is connected via water molecules to the C3 hydroxyl group of PLP and the nitrogen of the Schiff base. It is kept in place by van der Waals contacts to Tyr-255 of the cis protomer (Fig. 4, A and B). In addition, the backbone nitrogen of Arg-313 (from element III) hydrogen bonds to an anionic phosphate oxygen of the PLP.
In other fold type I PLP-dependent enzymes, element III is often significantly longer than in SecS (Fig. 3, C and D) and provides additional residues for binding the PLP phosphate. In contrast, the region corresponding to element II in afuPSerCysS is much shorter than in SecS (21 versus 47 residues) and is completely disordered in the structure (41), failing to provide stable PLP anchoring (Fig. 3C). In synC-DES, the equivalents of helices ␣2 and ␣4 are longer and place a very short element II, which is suspended between them, remote from the trans PLP and close to domain 3  of the other protomer (Fig. 3D). Thus, in other fold type I enzymes, element II serves to reinforce dimerization but does not contribute directly to the active sites. Instead, in afuPSerCysS, NifS relatives and synC-DES a portion of the N terminus is positioned between domains 1 and 3 and harbors residues, which in some cases contact the cis PLP (Fig. 3, C and D). In mmuSecS elast , the first loop of the long element II replaces this N-terminal part in trans (Fig. 3A).
In contrast to the situation within a close dimer, there is no crosscommunication at the active sites between molecules belonging to the two different close dimers of a tetramer. This situation suggests that the close dimers are sufficient to provide the chemical microenvironment required for catalysis.

Binding of Phosphate Triggers Disorder-Order Transition in an
Active Site Loop-Substrates of SecS contain a number of phosphates or phosphate-related groups, such as the phosphodiester backbone of tRNA [Ser]Sec , the ␥-phosphate of the PSer moiety, and SeP. We reasoned that phosphate could mimic binding of either of these groups at the active site of SecS and determined the crystal structure of mmuSec-S elast after soaking crystals for 30 s in 0.5 M phosphate buffer. Strikingly, we observed that a phosphate (P1 in Fig. 6) is cradled in the second loop of the domain 1 insertion (residues 98 -104 of element II; green in Fig.  6), which was previously disordered. Upon phosphate binding, this loop contracts and covers part of the trans active site (Fig. 6, A and B). Arg-313 originating from the loop that connects domain 2 to the second part of domain 1 (element III in Fig. 3, A and B) forms the base of the P1 phosphate binding site. To bind the P1 phosphate, Arg-313 and Gln-105 (neighboring the previously disordered element) are profoundly repositioned (Fig. 6, C and D). In addition, the P1 phosphate interacts directly as well as via a water bridge with the side chain of Arg-97 and with the side chain and backbone of  FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9

JOURNAL OF BIOLOGICAL CHEMISTRY 5857
Ser-98, both of which were disordered in the absence of phosphate (Fig. 6, C and D).
A poly-dentate anion appears to be required to engage in the observed interactions and elicit the structural transition. In agreement with this notion, a similar disorder-order transition was observed upon soaking with sulfate ions (not shown), whereas mono-dentate anions did not evoke any such change. For example, 250 mM NaCl were present in all crystallization and soaking experiments, and we did not observe any conformational changes in the structure soaked with an additional 0.5 M sodium iodide (Table 1). We, therefore, refer to the loop between Gly-96 and Pro-106 of element II as the "phosphate loop" (P-loop).
Upon binding of the P1 phosphate, the P-loop closes off part of the active site of mmuSecS elast (Fig. 6, A and B). It is likely that active site closure accompanies catalysis and may serve, e.g. to locally exclude bulk water. Therefore, the 3Ј-end of tRNA [Ser]Sec most likely gains access to the PLP cofactor from the side opposite the P-loop (arrow in Fig. 6B).
We observed a second phosphate binding site remote from the active site (P2 in Fig. 6B). This phosphate location could indicate a site of contact to the tRNA [Ser]Sec phosphodiester backbone. However, in contrast to the P-loop and Arg-313 (see below), residues contacting the P2 phosphate (Arg-199, His-368) are not conserved in SecS orthologs (Fig. 2). It is possible that the binding of the P2 phosphate is not functional and restricted to mouse and a few other SecSs.

Phosphate Binding Is Mediated by Conserved and SecS-specific Residues That Are Essential for Selenocysteine Synthesis-
The P-loop is a highly conserved element of eukaryotic and archaeal SecS, which is lacking from related enzymes with a different function (Fig. 2). In particular, Arg-97, Ser-98, and Gln-105 from the P-loop as well as Arg-313 from element III, which directly contact the P1 phosphate, are strictly conserved among SecS enzymes but not beyond. These observations are consistent with the idea that the P-loop carries out an essential function that is specific for SecSs.
To directly probe the importance of residues contacting the P1 phosphate, we generated mutant mmuSecS proteins in which Gln-105 was changed to Glu (mmuSecS Q105E ) or Arg-313 was changed to Ser or Glu (mmuSecS R313S ; mmuSecS R313E ). All mutants were expressed as soluble proteins in E. coli, migrated as tetramers in gel filtration, and exhibited a PLP complement comparable with that of the wild type protein, as indicated by their absorption maxima at 334 nm (ketimine form) and 418 nm (aldimine form; not shown). We tested the mutants for their ability to convert PSer-tRNA [Ser]Sec to Sec-tRNA [Ser]Sec . Strikingly, each of the point mutants severely corrupted the activity of mmuSecS (Fig. 7A). Although the mmuSecS Q105E and mmuSecS R313S mutants still exhibited activities of about 50 and 30% of the wild type protein, respectively, the R313E exchange rendered mmuSecS virtually inactive (Fig. 7A). The more severe effect with the negatively charged Glu compared with the neutral Ser in place of Arg-313 directly supports the functional relevance of phosphate binding at this position. These results directly link the ability of mmuSecS to bind a phosphate or a related group via induced fit of the P-loop to its catalytic competence. Based on these observations, we suggest that the P1 phosphate mimics binding of a substrate or of a functional portion of a substrate.
The P-loop Could Serve as a Binding Site for the PSer ␥-Phosphate and SeP-We scrutinized the possibility that the ␥-phosphate of PSer-tRNA [Ser]Sec could be bound by the P-loop. To this end we modeled the structure of an external aldimine comprising PLP in a Schiff-base linkage to a PSer esterified at the ␣-carboxylate. For modeling, we superimposed the structure of PSer-aminotransferase in complex with the substrate mimic ␣-methyl-L-glutamate (49) onto the mmuSecS elast -phosphate structure. We replaced the ␣-methyl-L-glutamate moiety with a ␣-carboxy ester of PSer, retained all side chain conformations as observed in the mmuSecS elast structure in complex with phosphate, and allowed the PLP moiety to adopt a slightly more inclined orientation (Fig. 6E). Even without adjustments of the protein matrix, the ␥-phosphate of the PSer ester could be accommodated approximately at the P1 phosphate position. These results suggest that the P-loop could serve to bind the ␥-phosphate of PSer-tRNA [Ser]Sec .
The similarity of phosphate and SeP suggests that the P1 phosphate could also mimic binding of the co-substrate SeP to the P-loop. Binding of the PSer moiety and SeP to the P-loop could occur sequentially and is not mutually exclusive (see "Discussion"). In contrast, we did not manage to fit a phosphate from the backbone of tRNA [Ser]Sec without clashes in the position of the P1 phosphate at the P-loop, consistent with our above suggestion that the tRNA [Ser]Sec 3Ј-end approaches the active site distal to the P-loop (Fig. 6B).
SecS Discriminates against Free O-Phospho-L-serine-Free PSer is produced, for example, as an intermediate in the biosynthesis of serine by transamination from 3-phosphohydroxypyruvate. Therefore, SecS should be safeguarded against using free PSer as a substrate. We tested this notion by attempting to dephosphorylate free PSer with mmuSecS. Indeed, the enzyme proved to be completely unreactive with respect to the free amino acid (Fig. 7B).
We inspected the active site of mmuSecS elast for possible filtering devices. Typically, PLP-dependent enzymes of the fold type I deploy a positively charged Arg in the neighborhood of the PLP to bind the negatively charged ␣-carboxylate of an amino acid substrate. In the absence of a substrate, a sulfate or phosphate group often binds at an equivalent position as the ␣-carboxylate (43). The ␣-carboxylate binding Arg originates in cis from the ␤-sheet in domain 3. An equivalent Arg is also strictly conserved in SecS orthologs (Arg-404; Fig. 4B). However, in mmuSecS elast , its side chain is turned away from the PLP. Arg-404 interacts instead by a cationinteraction with Phe227 (another residue strictly conserved only in the SecS orthologs; Fig. 2). It is additionally fixed by hydrogen bonds to the side chain of Asn-435 and the backbone carbonyl of Met-423 and by water-mediated hydrogen bonds to the backbone carbonyl of Ala-228 and to side chains of His-425 and Tyr-433 (Fig. 4B). The side chain of Arg-404 is thereby stably tugged away, since it does not change orientation in the presence of even 0.5 M phosphate (as seen in our phosphate-soaked structure), suggesting that Arg-404 is not involved in binding of a ␣-carboxylate.
As pointed out above, mmu-SecS elast harbors Glu-74 in trans next to the PLP moiety (Fig. 4, A and  B). Interestingly, Glu-74 occupies the same spatial position as the equivalent of Arg-404 in other fold type I enzymes (Fig. 4, B and C). This observation and our model of an external aldimine of mmuSecS with a PSer ester (Fig. 6E) demonstrate that Glu-74 would strongly disfavor productive placement of a substrate with a free (negatively charged) ␣carboxylate. In PSer-tRNA [Ser]Sec , however, the ␣-carboxylate of PSer is esterified to the 2Ј-or 3Ј-hydroxyl group of the 3Ј-terminal adenosine and does not carry a negative charge. Therefore, we expect that Glu-74 acts as a substrate filter by repelling compounds with a negatively charged ␣-carboxylate.
The Inhibition Profile of SecS Resembles That of ␤-Lyases-To gain additional insight into the reaction mechanism of SecS, we tested whether the dephosphorylation activity of mmuSecS is inhibited by the mechanism-based inhibitors PG and F 3 -Ala. PG preferentially inhibits PLP-dependent enzymes such as cystathionine ␥-lyase, which mediate lyase reactions at the ␥-carbon (50). In contrast, F 3 -Ala preferentially inhibits enzymes such as cystathionine ␤-lyase, which mediate replacement of a substituent at the ␤-car-bon (51). Strikingly, mmuSecS is unaffected by up to 25 mM PG (more than 12,500-fold excess over mmuSecS active sites; Fig. 7C). In contrast, partial enzyme inhibition was detected in the presence of 5 mM F 3 -Ala and above (Fig. 7C). These inhibition profiles are in agreement with SecS catalyzing a ␤-replacement reaction with a high specificity. Furthermore, the data suggest that the lytic half-reaction of SecS follows the cystathionine ␤-lyase scheme, strongly supporting aminoacrylyl-tRNA [Ser]Sec as an intermediate (28).
Although our data do not allow us to derive exact inhibition constants, we note that compared with E. coli cystathionine ␤-lyase (51), the inhibition of mmuSecS by F 3 -Ala is weak. For example, in a comparable setup, the halftime for inactivation of E. coli cystathionine ␤-lyase by 1 mM F 3 -Ala was less than 2 min (52), whereas mmuSecS is not measurably affected under these conditions (Fig. 7C). This observation is consistent with the suggested substrate discrimination by Glu-74, since F 3 -Ala exhibits a free ␣-carboxylate and is, therefore, expected to be discouraged from forming an external aldimine.

DISCUSSION
SecS Orthologs Constitute a Unique Subclass of Fold Type I PLP-dependent Enzymes-We have presented structural and functional analyses of a mammalian SecS demonstrating how SecS orthologs are set aside from other PLP-dependent enzymes. Although mmuSecS elast is composed of three domains whose structural scaffolds exhibit high similarity to the fold type I family of PLP-dependent enzymes (Fig. 3), distinguishing structural and functional characteristics are conferred by remodeled elements both within and outside of these conserved scaffolds (labeled I-IV in Fig. 3). These novel elements clearly define SecS orthologs as a special subclass of the family.
Of paramount importance for the function of SecS are two motifs that are unique to and highly conserved in SecS orthologs. First, a special N terminus (element I) serves as a tetramerization device. It reinforces the interaction between two close dimers by cross-strutting (Fig. 5, A and B). Neither afuPSerCysS nor NifS relatives nor synC-DES exhibit a comparable element, and all of these enzymes form dimers. As further detailed below, we suggest that tetramerization could be crucial for proper PSer-tRNA [Ser]Sec positioning. Second, a long insertion between helices ␣2 and ␣4 of domain 1 (element II) is involved in catalysis by SecS. Via a first loop, element II provides residues Glu-74 and Arg-75, which anchor the PLP cofactor in trans. It thereby positions the negatively charged Glu-74 ideally to act as a substrate filter. Via a second loop it dispatches the P-loop to the trans active site, which mediates binding of SecS-specific substrates.
Evidently, the evolution of the unique N terminus and of the long domain 1 insertion went hand in hand. Placing the N terminus at the outside of the protein, where it can engage in tetramer formation, liberated a binding site between domains 1 and 3 as seen, e.g. in afuPSerCysS, NifS relatives, and synC-DES proteins (Fig. 3). The first loop of the domain 1 insertion evolved to take advantage of the liberated binding site between domains 1 and 3 in trans. It thereby became ideally positioned to contribute residues to the trans active site (Fig. 3A).
tRNA Selection Strategy of SecS-Our results highlight a number of features that enable SecS to specifically recognize its substrates. The highly positively charged surface of mmu-SecS elast is apparently designed to interact at multiple sites with the phosphodiester backbone of tRNA [Ser]Sec . This expected mode of interaction is supported by our observations of numerous anion (iodide) binding sites (Fig. 5D), binding of a chloride ion in untreated crystals (not shown), and binding of the P2 phosphate ion distal to the active sites in phosphate-soaked crystals (Fig. 6B). tRNA [Ser]Sec exhibits a number of unique structural characteristics compared with canonical tRNAs, such as an elongated (9 ϩ 4 base pairs) helical stack between the acceptor stem and the T⌿C stem and an unusually long variable arm (8,53). Multiple latching points on the surface of SecS would allow the enzyme to recognize these global structural features and discriminate against other tRNAs, which are in vast excess in the cell. Thus, the tRNA selection strategy of SecS may resemble that of bacterial SelA, which very inefficiently converts Ser-tRNA [Ser]Sec mutants with a shortened, canonical acceptor stem (54).
The mmuSecS elast active site environments are built up entirely by residues originating from the two protomers of a close dimer. Why then does mmuSecS form tetramers? One possibility is that the tetramer provides an effective binding platform for the large tRNA [Ser]Sec molecule. Although we have no direct evidence for the mode of PSer-tRNA [Ser]Sec binding to mmuSecS, portions of tRNA [Ser]Sec could extend beyond the borders of the molecule to whose active site its 3Ј-end is bound. To illustrate the relative sizes of the molecules and how the SecS tetramer could serve as a binding platform for tRNA [Ser]Sec , we generated a hypothetical docking model (Fig.  5E). A model of tRNA [Ser]Sec (as derived in Hubert et al. (53)) was positioned on the mmuSecS elast tetramer with the 3Ј-end approaching the active site of one monomer distal to the P-loop, leaving the P-loop available for accommodation of the PSer moiety. The body of tRNA [Ser]Sec was then adjusted by rotation about the 3Ј-terminal nucleotide to avoid clashes with the protein. Significantly, the unique mode of tetramerization provides the mmuSecS tetramer with a unique, elongated shape (shown by the characteristic distances between the active sites;   5A). Other arrangements, such as the stubbier form of E. coli cystathionine ␥-synthase (Fig. 5A) (55), generate other relative dispositions of active sites. Thus, SecS may have evolved as a distinctly shaped tetramer to support efficient PSer-tRNA [Ser]Sec binding by one dimer to position its 3Ј-end appropriately in an active site of a neighboring dimer. A similar principle may underlie the decameric organization of bacterial SelA. Further experiments are required to test these ideas.
Mechanisms for Substrate Binding and Differentiation in the Active Site-Our structural analyses have shown that the long, conserved domain 1 insertion (element II) of mmuSecS does not merely serve to support PLP anchoring in trans. Part of this insertion, which we refer to as the P-loop, can undergo disorder-order transitions coupled to binding of poly-dentate ions such as phosphate or sulfate. We interpret this observation as direct evidence for the mode of substrate binding by mmuSecS. Our molecular modeling suggests that the P1 phosphate could resemble the binding of the ␥-phosphate of the PSer moiety of PSer-tRNA [Ser]Sec (Fig. 6E). Evidently, SeP could also bind to the P-loop in a similar fashion as the P1 phosphate. In contrast, we could not position the tRNA [Ser]Sec phosphodiester backbone in the same way at the P-loop. In any case, the P-loop in combination with Arg-313 (from element III) obviously binds and positions phosphate portions of substrates by an induced fit mechanism. This notion is perfectly borne out by our mutational analyses, which showed that the phosphate-coordinating residues are crucial for mmuSecS activity (Fig. 7A).
In addition, we suggest that SecS has exploited the domain 1 insertion to install a filtering mechanism that allows it to exclude free PSer and other free amino acids from its active site. Glu-74, positioned strategically next to the trans PLP (Fig. 4, A  and B), adopts a similar spatial position as the side chain of an Arg originating from the domain 3 ␤-sheet in related enzymes, which typically serves as an ␣-carboxylate recognition device (Fig. 4C). In SecS, an equivalent Arg is present, but it is turned away via interactions with other conserved residues (Fig. 4B), e.g. to engage in alternative interactions with the tRNA [Ser]Sec portion. We suspect that Glu-74 repels negatively charged carboxyl groups of free amino acids. Neutral ester moieties as in PSer-tRNA [Ser]Sec are presumably allowed to productively approach the PLP internal aldimine. Our finding that SecS does not convert free PSer supports the role of Glu-74 as a substrate discriminator. Similar to the P-loop, the elements constituting the putative substrate filter are highly conserved among SecS orthologs but not beyond. Thus, SecS has acquired specialized functional modules for substrate binding and differentiation.
␥-Phosphate binding at the P-loop provides a facile explanation for the observation that SecS binds PSer-tRNA [Ser]Sec preferentially over non-aminoacylated tRNA [Ser]Sec (28). However, presently we can only speculate why Ser-tRNA [Ser]Sec is bound with least efficiency (28). One clue is provided by the observations that the P-loop tends to bind substrate mimics such as phosphate or sulfate and that PSer-tRNA [Ser]Sec , but not Ser-tRNA [Ser]Sec , can compete efficiently with these molecules. The unloaded 3Ј-end of tRNA [Ser]Sec may fit next to the contracted P-loop without having to compete for binding at that place. Another possibility is that ␥-phosphate binding at the P-loop leads to proper accommodation of the PSer moiety and of the tRNA [Ser]Sec 3Ј-end in the active site, whereas Ser at the 3Ј-end may engage in alternative interactions, which could be mutually exclusive with proper fitting of tRNA [Ser]Sec . Clarification of Structure-based Reaction Mechanism-NifS-like enzymes mobilize sulfur for iron-sulfur cluster biosynthesis via a protein-bound persulfide using a conserved cysteine that lies in a long loop of domain 3 (element IV in Fig. 3, A and B) (43). synC-DES employs a related strategy by generating an external cysteine persulfide via cystine C-S cleavage, which remains non-covalently fixed at the active site (47). Furthermore, it has been discussed that PSerCysS could also employ a persulfide mechanism (41). Because NifS can support selenide delivery (56), SecS could function in an analogous fashion by using a perselenide intermediate. However, although SecS exhibits a domain 3 loop analogous to the persulfide loop of NifS enzymes, no cysteine that could serve as attachment site for selenium is present in that loop. The only cysteine that is conserved in eukaryotes in the active site cavity is Cys-226 (Fig. 2). However, its sulfur atom is still more than 7 Å away from any atom of the PLP and remote from the modeled substrate. In addition, this residue is not conserved in archaeal SecS (Fig. 2). Furthermore, in light of the observation that SeP delivered by selenophosphate synthetase 2 is the active selenium donor of SecS (24,28), internal or external perselenide production (via SeP) appears to be an off-pathway reaction. Under these circumstances it is unlikely that SecS functions via an intermediate perselenide moiety.
Instead, a mechanism that is consistent with all our findings evokes direct selenide delivery by SeP (Fig. 8). As suggested by the inhibition results, we based the first part of the scheme on the E. coli cystathionine ␤-lyase mechanism (51). In our work-  FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9

Structure and Mechanism of Eukaryotic Selenocysteine Synthase
ing model the process is initiated by PSer-tRNA [Ser]Sec binding and positioning of the ␥-phosphate at the P-loop (Fig. 8, I). Lys-284 is expected to be deprotonated after liberation upon external aldimine formation, constituting a strong base suitable for abstracting the ␣-hydrogen from the substrate (Fig. 8, II). Phosphate release would generate aminoacrylyl-tRNA [Ser]Sec as an intermediate (Fig. 8, III and IV). Next, the liberated phosphate is exchanged for SeP at the P-loop (Fig. 8, IV), which would then be ideally situated to donate Se 2Ϫ to the ␤-carbon of the aminoacrylate moiety (Fig. 8, IV and V). We suggest that attack of the aminoacrylyl-tRNA [Ser]Sec intermediate by SeP involves concomitant attack by a water molecule on SeP, again giving rise to a phosphate leaving group (Fig. 8, IV and V). A general base is proposed to activate this water molecule via a circular protonic shift involving Lys-284 (Fig. 8, IV). Subsequent reverse transaldimination (Fig. 8, V and VI) liberates the product, Sec-tRNA [Ser]Sec , and regenerates the internal aldimine.
Details of the above model are still in the dark. For example, we presently envision sequential binding and conversion of the substrates coupled to repeated contraction and relaxation of the P-loop since our phosphate-soaking experiments only revealed a single phosphate binding site. However, we cannot rigorously exclude the possibility for a separate binding pocket for SeP at the present time. Furthermore, the identity of the water-activating general base is unknown at the time. This function could involve residues from the P-loop/Arg-313 (for example, note the water bound to Arg-97 and phosphate in Fig.  6D). In archaeal SecS, the proposed water activation could be provided by the histidine residue stacking on the PLP. It is also conceivable that Lys-284 directly activates the water molecule without an intervening general base. Evoking a similar scenario, it has been shown that the PLP-binding lysine in E. coli cystathionine ␥-synthase can reach to distal regions of the substrate (55). Work is in progress to further corroborate and refine this mechanism. To this end, our high resolution crystal structures of mmuSecS elast provide excellent leads for further rational sitedirected mutagenesis.