The Subnanometer Resolution Structure of the Glutamate Synthase 1.2-MDa Hexamer by Cryoelectron Microscopy and Its Oligomerization Behavior in Solution

The three-dimensional structure of the hexameric (αβ)6 1.2-MDa complex formed by glutamate synthase has been determined at subnanometric resolution by combining cryoelectron microscopy, small angle x-ray scattering, and molecular modeling, providing for the first time a molecular model of this complex iron-sulfur flavoprotein. In the hexameric species, interprotomeric α-α and α-β contacts are mediated by the C-terminal domain of the α subunit, which is based on a β helical fold so far unique to glutamate synthases. The αβ protomer extracted from the hexameric model is fully consistent with it being the minimal catalytically active form of the enzyme. The structure clarifies the electron transfer pathway from the FAD cofactor on the β subunit, to the FMN on the α subunit, through the low potential [4Fe-4S]1+/2+ centers on the β subunit and the [3Fe-4S]0/1+ cluster on the α subunit. The (αβ)6 hexamer exhibits a concentration-dependent equilibrium with αβ monomers and (αβ)2 dimers, in solution, the hexamer being destabilized by high ionic strength and, to a lower extent, by the reaction product NADP+. Hexamerization seems to decrease the catalytic efficiency of the αβ protomer only 3-fold by increasing the Km values measured for l-Gln and 2-OG. However, it cannot be ruled out that the (αβ)6 hexamer acts as a scaffold for the assembly of multienzymatic complexes of nitrogen metabolism or that it provides a means to regulate the activity of the enzyme through an as yet unknown ligand.

enzyme in primary nitrogen assimilation and in photorespiration (5,6). On the other hand, enhancing the GltS activity through protein and metabolic engineering is a viable option for modulating the NAD(P) ϩ /NAD(P)H ratio (or the 2-OG level) in bioconversions (e.g. see Ref. 7), the nitrogen assimilation pathway in nitrogen-fixing bacteria used as biofertilizers, or the response to osmotic or acid stress (8). NADPH-GltS from A. brasilense, a root-growth promoting nitrogen-fixing bacterium, is the best studied GltS. The three-dimensional structure of ␣GltS in complex with L-methionine sulfone (MetS, an L-Gln analog) and the 2-OG substrate has been solved by x-ray crystallography (9), providing insights into the glutaminase and the synthase reactions taking place in this subunit (Fig. 1A). A combination of structural and functional experiments has shown that at least five steps at three different locations are necessary to carry out the overall reaction (1-3) (Fig. 1A), namely NADPH oxidation (step 1) and transfer of reducing equivalents from FAD (on ␤GltS) to FMN (on ␣GltS) through at least two of the three [Fe-S] clusters (step 2) and binding and hydrolysis of L-Gln at the glutaminase site in the Type II (PurF) glutamine amidotransferase domain of ␣GltS (step 3) followed by transfer of the released ammonia molecule to the synthase site through the intramolecular tunnel (step 4). At the synthase site, the 2-iminoglutarate (2-IG) intermediate formed upon the addition of ammonia to 2-OG is reduced to L-Glu by reduced FMN (step 5). A striking feature of GltS is its ability to coordinate the activities of the NADPH-oxidizing site on ␤GltS, and the glutaminase and synthase sites on ␣GltS, the latter sites being connected by the intramolecular ammonia tunnel. In GltS, L-Gln hydrolysis occurs only when the enzyme is in its reduced state and bound to 2-OG, thus avoiding the waste of L-Gln (10).
Unfortunately, crystallization experiments of the NADPH-GltS ␣␤ holoenzyme have been so far unsuccessful. Thus, the structure of ␤GltS, its association with its related ␣ subunit partner within the ␣␤ protomer, and, possibly, its association with other subunits within the complex remain to be established. Furthermore, the location and the role of the [4Fe-4S] 1ϩ/2ϩ clusters of ␤GltS only rely on indirect observations. Interestingly, the sequence of ␤GltS and, in particular, the spacing of its N-terminal conserved cysteine residues are similar to those of several other proteins or protein domains (1), among which only the Pyrococcus furiosus sulfide dehydrogenase ␣ subunit (11) and the bovine dihydropyrimidine dehydrogenase (DPD) (12)(13)(14) have been characterized to different extents. Site-directed mutagenesis experiments allowed us to confirm the location of the DPD-like [4Fe-4S] clusters of GltS within ␤GltS. It was also shown that they are not only involved in the electron transfer between FAD (on ␤GltS) and FMN (on ␣GltS) but also in the structuring of the N-terminal subdomain of ␤GltS forming the interface between the ␣ and ␤ GltS subunits (15). Furthermore, the loss of coupling of the glutaminase and synthase catalytic subsites within the isolated ␣GltS suggests that a conformational change occurs in this subunit upon association with ␤GltS (16).
For all of these reasons, obtaining information on the threedimensional structure of the GltS ␣␤ protomer is a key step toward the understanding of the sophisticated mechanism of action and regulation of this essential enzyme.
In the present work, we used three-dimensional cryoelectron microscopy (cryo-EM), small angle x-ray scattering (SAXS), and modeling to investigate the stoichiometry and the structure of the active oligomeric NADPH-GltS species in solution. The combined use of cryo-EM and SAXS allowed us to obtain an electron density map of the GltS (␣␤) 6 hexamer at a subnanometer resolution, which has been reached only recently, and only in a few cases, for cryo-EM-based structure determinations (17)(18)(19)(20)(21)(22). Moreover, we propose a homology model of ␤GltS, and, more importantly, that of the ␣␤ protomer. The latter sheds light on the intramolecular electron transfer process from FAD to FMN along the enzyme iron-sulfur clusters. Finally, modeling of the (␣␤) 6 oligomer of NADPH-GltS into the 9.5 Å resolution cryo-EM-derived electron density shows how the C-terminal ␤ helical domain of ␣GltS acts as a structural spacer, which establishes interprotomeric ␣-␣ and ␣-␤ contacts, playing a key role in the oligomerization process.

EXPERIMENTAL PROCEDURES
Enzymes-GltS or the species formed by the wild-type ␣ subunit and a C-terminally His 6 -tagged variant of the ␤ subunit (GltS-His) were overproduced in Escherichia coli BL21 (DE3) cells and purified, quantified, and assayed as described previously (15,23). ␣GltS was prepared and characterized as described in Ref. 16. Steady-state kinetic analyses were carried out as in Ref. 15 and references therein. Construction of GltS deletion mutants and other techniques are described in the supplemental material.
Prior to each experiment, the protein solutions were transferred in 25 mM Hepes/KOH buffer, pH 7.5, containing 1 mM EDTA and 1 mM dithiothreitol by gel filtration through Sephadex G25 (medium) columns (PD10, prepacked disposable columns; GE Healthcare) equilibrated and eluted with the same buffer. If needed, the enzyme solution was concentrated to 10 -30 mg/ml and stored frozen at Ϫ80°C after flash freezing in liquid nitrogen. The same buffer was used for all dilutions.
Small Angle X-ray Diffraction Data Collection-Synchrotron x-ray scattering data from solutions of GltS in the presence and absence of ligands and/or substrates were collected at the X33 beamline (DESY, Hamburg) (24) at protein concentrations (c) ranging from 0.75 to 10.0 mg/ml. At a sample-detector distance of 2.7 m, the range of momentum transfer 0.1 Ͻ s Ͻ 5 nm Ϫ1 was covered (s ϭ 4 sin()/, where 2 is the scattering angle and ϭ 0.15 nm is the x-ray wavelength). The data were processed using standard procedures by the program package PRIMUS (25). The forward scattering I(0) and the radii of gyration (R g ) were evaluated by the program AUTORG (26) using the Guinier approximation (27), assuming that at very small angles (s Ͻ 1.3/R g ), the intensity is represented as I(s) ϭ I(0) exp(Ϫ(sR g ) 2 /3). The effective molecular mass of the solute (MM) was estimated by comparison of the forward scattering I(0) with that from reference solutions of bovine serum albumin (MM ϭ 66 kDa). For the NaCl-containing solutions, the MM estimates were appropriately corrected to account for contrast reduction (factors 1.5 and 2.4 for 1 and 2 M NaCl, respectively).
The scattering intensities for the monomeric, dimeric, and hexameric GltS computed by CRYSOL (28) from the atomic coordinates taken from the cryo-EM model were employed to analyze the oligomeric composition of all of the samples. The program OLIGOMER (25) was used to find the volume fractions of components minimizing the discrepancy 2 (normalized sum of the reduced S.D. values) between the linear superposition of the weighted intensities of the components and the experimental data from the mixture.
Specimen Preparation and Cryo-EM Data Collection-Grids with holey carbon films were prepared and vitrified by flash freezing in liquid ethane (29). For the initial three-dimensional reconstruction, a GltS sample (6.2 mg/ml) was observed in a Philips CM12 electron microscope using an acceleration voltage of 120 kV and a magnification of ϫ45,000. Each field was recorded with subsequent tilt angles of 45 and 0°and with defocus values of Ϫ1.6 and Ϫ1.2 m, respectively. Additional untilted specimen images of the wild-type GltS complex and of the isolated ␣GltS were recorded on the same instrument, with sample concentrations of 9.25 mg/ml and 7 mg/ml, respectively. For high resolution studies, the sample was observed in a JEOL JEM 2100F, using an acceleration voltage of 200 kV, a spherical aberration of 0.5 mm, and a magnification of ϫ50,000. Images were recorded under low dose conditions (10 electrons/ Å 2 ) on Kodak SO 163 film.
Three-dimensional Reconstruction-For the initial three-dimensional reconstruction, 437 pairs of particles were extracted from 12 pairs of tilted-untilted images, interactively, using the program WEB (30). A preliminary three-dimensional model was reconstructed with the software package SPIDER (30) using the random conical tilt series method (31). This model was refined using the projection matching method (32) with 1344 particle images extracted from five additional untilted specimen images. A few final refinement iterations were performed with imposed D3 symmetry until the resolution stabilized at 26 Å, as estimated with the Fourier shell correlation criterion (33). For the high resolution reconstruction, micrographs with anisotropic or no diffraction rings were removed using the enhanced power spectra sorting method (34). 60,000 particle images were selected automatically from 84 remaining micrographs using Roseman's particle selection algorithm (35). However, only 13000 particle images were kept after visual inspection and splitting in defocus groups. The parameters of the contrast transfer function of the electron microscope were computed for each particle with the program CTFTILT (36). An iterative refinement of the low resolution three-dimensional model was performed with SPIDER, using the projection matching method coupled with the method of correction of the contrast transfer function by Wiener filtering of volumes from focal series (37). D3 symmetry was imposed, and different filters were applied on the reference volume at different stages of the refinement until the resolution improved and stabilized at 9.5 Å (33). The Fourier amplitudes of the final, nonfiltered volume were enhanced using the GltS SAXS data as described in Ref. 38.
Atomic Structure Fitting and Modeling-Sequences of ␤GltS (SwissProt accession number Q05756) and DPD (Protein Data Bank code 1h7w (14)) were first roughly aligned using ClustalW (39,40). Then the secondary structure prediction methods SOPMA, PHD, and PREDATOR, integrated in the metaserver NPS@ (41), and hydrophobic cluster analysis (42,43) were applied to ␤GltS. The secondary structures predicted for ␤GltS were compared with the corresponding ones of DPD calculated by DSSP (44), allowing us to optimize the alignment. This alignment was used with MODELLER 7v7 (45) to generate homology models by satisfaction of spacial restraints derived from the alignment. The ␣ 2 GltS dimer (Protein Data Bank code 1ea0 (9)) was submitted to the algorithm "loop" of MODELLER to model the unresolved regions 305-307, 1172-1179, and 1194 -1202 of chains A and B. Only the seven C-terminal residues (residues 1473-1479) of ␣GltS, absent in the Protein Data Bank file, remain missing in the final pseudoatomic model of GltS.
To perform the fitting of atomic data into the cryo-EM map, the following programs were compared: (i) rigid body fitting in Fourier space with Colores and Colacor algorithms of SITUS (46,47); (ii) real space fitting with the FoldHunterP algorithm of EMAN (48,49); (iii) real space fitting using the operation "Fit Model in Map" of Chimera (50,51). A first ␣ 2 dimer was fitted, and the 3-fold symmetry was used to calculate the atomic coordinates of the other two. Due to the unambiguous dispositions of the subunits, these different programs yielded equivalent results. Then a ␤GltS model was fitted independently, and D3 symmetry was used to calculate the positions of the other ones. An automatic check with Chimera software showed no significant steric hindrance between the fitted subunits. This fitting was compared with another procedure, consisting of fitting independently the ␣and ␤GltS monomers. However, this approach did not improve the model. Molecular graphics images were produced using the UCSF Chimera package from the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (supported by National Institutes of Health Grant P41 RR-01081). The experimental density map (accession code EMD-1440) and the fitted atomic coordinates of GltS (Protein Data Bank code 2vdc) have been deposited in the Macromolecular Structure Database.

Three-dimensional Reconstruction Volume of GltS by Cryo-EM at 9.5 Å Resolution
To obtain a preliminary three-dimensional reconstruction volume, preparations of GltS were observed by cryo-EM with the specimen grid tilted at 45 and 0°, respectively (Fig. 1B). Digitized images were subjected to the random conical tilt series technique, a three-dimensional reconstruction method inspired by tomographic approaches (31,52). The volume obtained at this stage was solved at a resolution of 26 Å, which was sufficient to clarify two important points concerning the structure of the oligomeric complex. First, although no symmetry was enforced during the three-dimensional reconstruction process, the particles produced well defined rectangular side views (Fig. 1C) and triangular top views (Fig. 1D), specific for a 322 point group symmetry. Second, using the rigid body fitting techniques of SITUS (46,47) and EMAN (48,49) software, it was easy to fit the atomic structure of the crystallographic ␣ 2 dimers (Protein Data Bank code 1ea0 (9)) forming three large "pillars" within the complex (Fig. 1, C and D). Hence, the stoichiometry of the whole complex was unambiguously deter-mined with six ␣GltS, forming the three pillars of the complex, and six smaller masses, connecting these pillars, most likely corresponding to ␤GltS. Interestingly, the crystallographic asymmetric unit of ␣GltS contains ␣ 2 dimers, but the analysis of the geometry of adjacent dimers in the crystals revealed that they interact to form tetramers, with a structure essentially corresponding to that of two adjacent pillars of the cryo-EM-derived hexameric model of GltS (supplemental Fig. 1).
To improve the resolution of the three-dimensional reconstruction, as needed to propose a complete structural model of the (␣␤) 6 oligomeric GltS complex, additional cryo-EM images were collected with the specimen grid untilted, using a JEOL 2100F electron microscope under low dose conditions. Micrographs of the highest quality were selected as described in Ref. 34, and an automatic identification of particles was performed on the selected micrographs using the procedure described in Ref. 35. After additional sorting of the particle images by visual inspection, the kept images were split into defocus groups and used for the high resolution three-dimensional reconstruction. Using the preliminary volume ( Fig. 1, C and D) as the starting model, the new image set was subjected to an iterative process of three-dimensional projection alignment and contrast correction, until the resolution of the three-dimensional reconstruction volume stabilized at 9.5 Å. This better resolved structure has lost its smooth appearance but clearly shows finer details on the isosurface representations of the rectangular side view (Fig. 1E) and the triangular top view (Fig. 1F). This new structure was the template required for the complete modeling of the GltS complex.

Homology Modeling of the GltS ␤ Subunit
In order to generate a model of the (␣␤) 6 GltS hexamer from the cryo-EM-derived structure at 9.5 Å resolution, a structural model of ␤GltS was required. It has been previously shown (1,12,14) that ␤GltS and the N-terminal region of DPD are similar to each other. Thus, the DPD atomic coordinates (Protein Data Bank code 1h7w, chain A, residues 31-520) were used as a structural template to produce a homology model of ␤GltS ( Fig. 2A). Overall, the DPD and ␤GltS sequences are very similar to each other (26% identity and 46% similarity on the  Fig. 2E). B, cryoelectron micrograph of the GltS complex. The triangular top view and the rectangular side views (see below) are highlighted with circles and rectangles, respectively. C and D, cryo-EM volume at 26 Å with one (C) and three copies (D) of the structure of the crystallographic ␣ 2 dimer (Protein Data Bank code 1ea0) fitted into the three pillars of the GltS oligomer, indicating the presence of six ␣ subunits in the core region and of six ␤ subunits in the remaining peripheral areas. E and F, cryo-EM volume of the (␣␤) 6 complex at 9.5 Å resolution, where elements of secondary structure start to show. C and E, rectangular side views; D and F, triangular top views. The bar in F corresponds to 10 nm and refers to the scale of the images in C-F. whole sequence calculated with BLOSUM62 substitution matrix) taking into account in ␤GltS the deletion of a small 8-residue ␣ helix (residues 177-184; supplemental Fig. 2) and an insertion of 4 residues (residues 346 -349) between strands ␤9 and ␤10 of DPD (supplemental Fig. 2). Only the region spanning residues 1-25 of ␤GltS (corresponding to residues 31-55 of DPD, supplemental Fig. 2, stars) could not be reliably mod-eled due to the low similarity and to the fact that the corresponding region of DPD is an ␣ helix projected from subunit A toward subunit B, with which it interacts. Thus, the first 25-residue peptide of ␤GltS was omitted.
As previously suggested (12,14,15), the [4Fe-4S] clusters of ␤GltS are well modeled, assuming that one of these centers is formed by Cys 47 , Cys 50 , Cys 55 , and Cys 108 of ␤GltS, corresponding to Cys 79 , Cys 82 , Cys 87 , and Cys 140 , forming the first N-terminal cluster of DPD (nFeS I according to DPD nomenclature (14)), and the other (nFeS II in DPD) is formed by Cys 59 , Cys 98 , Cys 104 , and Glu 124 . These Cys residues correspond to Cys 91 , Cys 130 , and Cys 136 of DPD. Glu 124 of ␤GltS substitutes Gln 156 of DPD, in part justifying the different redox properties of the GltS [4Fe-4S] clusters with respect to the corresponding ones in DPD. Indeed, one of the GltS [4Fe-4S] clusters can be reduced by NADPH, and both clusters are reduced photochemically (53). On the contrary, none of the corresponding DPD clusters could be reduced even under mild denaturing conditions (13).

Rigid Body Modeling of the GltS (␣␤) 6 Hexamer
With the crystallographic structure of ␣GltS and a model of ␤GltS available, it was possible to fit them into the 9.5 Å cryo-EM model (Fig.  1, E and F), using a rigid body approach. Three copies of the crystallographic ␣ 2 dimer were positioned in the electron density of the three pillars of the oligomer to form the core of the complex. One copy of the ␤GltS model was positioned in each one of the triangular shapes extending at the periphery of the ␣GltS hexameric core. The best fit resulted in the model shown in Fig. 2, B and C (see also movie1.mov in the supplemental materials), which was fully supported by the SAXS data, as detailed below.
This observation agrees well with the results of site-directed mutagenesis of Cys residues of ␤GltS, which indicated a role of the ␤GltS [4Fe-4S] clusters in structuring the N-terminal region of this subunit allowing its association with ␣GltS to form the ␣␤ protomer (15).

SAXS and Stoichiometry of the NADPH-GltS Complex in Solution
Overall Parameters-The x-ray scattering patterns from GltS solutions are given in Fig. 4. The effective R g and MM computed from the SAXS data are presented in Table 1. Control measurements with GltSHis solutions (15) yielded, within the errors, the same scattering curves as the wild-type GltS at the given conditions. Thus, it was confirmed that the C-terminal His 6 tag engineered in ␤GltS yields an enzyme form indistinguishable from the native GltS. The MM values obtained for solutions of unliganded GltS at low ionic strength (  [17][18][19][20]. On the contrary, preincubation with NADP ϩ either alone or in combination with MetS and 2-OG (Table 1, lines 21-24 and 13-16) resulted in a decrease of both R g and MM, suggesting a shift in the oligomeric equilibrium toward smaller particles. For all of the above solutions, the R g and MM values at the highest protein concentrations (ϳ8 -10 mg/ml) were somewhat smaller than those obtained at the medium ones (ϳ2-4 mg/ml), indicating a concentration effect (namely repulsion between particles) at the higher protein concentrations.
In order to identify the smaller particles found in the GltS solutions, the effect of the ionic strength of the solvent on the oligomerization state of the enzyme was studied by SAXS. Preliminary analytical dynamic light scattering and gel filtration experiments showed that increasing ionic strength above 0.7 M brought about dissociation of GltS or GltSHis into a smaller species (i.e. the ␣␤ protomer; see below). The process was slow, reaching completion after 16 -20 h of incubation at 4 -15°C with greater than 85% retention of activity (supplemental Fig. 3, A-C). Removal of NaCl by either dialysis or centrifugal gel filtration established that the dissociation of GltS ␣␤ oligomer into ␣␤ protomers was reversible and fast (supplemental Table  1 and supplemental Fig. 3D), except for a small fraction of protein that had irreversibly dissociated into the free ␣ and ␤ subunits, as revealed by SDS-PAGE of individual fractions obtained by gel filtration chromatography (not shown).
The SAXS data from GltS solutions that had been preincubated with 1 or 2 M NaCl reveal a dramatic decrease of R g and MM, yielding a low mass species (Table 1, lines 5-12), presumably, the ␣␤ protomer (see below). These parameters show a clear correlation with protein concentration, with 2 M NaCl exhibiting an effect stronger than that of 1 M NaCl on protein dissociation (compare lines 5-8 with lines 9 -12 of Table 1).
Analysis of the Oligomeric Composition-The computed scattering from the cryo-EM model of GltS (Fig. 2, B and C) agrees well with the measured SAXS profile of salt-and ligandfree GltS at high protein concentrations. However, it displays some systematic deviations (Fig. 4, upper fit), which become very significant with the addition of salt and NADP ϩ (Fig. 4). To determine the nature of the deviations, quantitative analysis of ; plots 17-20, 7.9-1 mg/ml GltS preincubated for 16 h with MetS and 2-OG only; plots 21-24, 7.8-0.9 mg/ml GltS preincubated for 16 h with NADP ϩ (1 mM). Experimental data are denoted by dots, and fits from OLIGOMER are presented as solid lines. In the top curve, the computed scattering from the cryo-EM model (Fig. 2) is denoted by triangles. MARCH 28, 2008 • VOLUME 283 • NUMBER 13 the oligomeric state of GltS as a function of concentration and ligands was performed. It was possible to fit the entire set of data (Fig. 4) by the scattering from mixtures of GltS ␣␤ protomers, (␣␤) 2 dimers (i.e. the pillars forming the GltS oligomer) and (␣␤) 6 hexamers extracted from the cryo-EM model (Fig. 2,  B and C). The volume fractions of these species were computed by OLIGOMER (25), and are presented in Table 1. Even without salt or ligands and at the highest protein concentration, GltS is not fully hexameric. Rather, the solution contains 7-10% of GltS ␣␤ protomers. The same observations hold in the presence of MetS and 2-OG. The somewhat higher values obtained at the highest protein concentrations might be explained by the repulsive interactions between the particles already discussed above. The presence of salt significantly shifts the protomer-hexamer equilibrium toward the GltS protomer, possibly weakening the interprotomeric ␣-␣ and ␣-␤ interactions as well as the ␣-␣ interactions within each (␣␤) 2 dimer. OLIGOMER results confirm that GltS in 1 M NaCl completely dissociates into ␣␤ protomers at concentrations below 2 mg/ml ( Table 1, lines 11 and 12). Interestingly, only upon the addition of NADP ϩ (independently from the presence of the other two ligands), a small fraction of (␣␤) 2 dimeric particles appeared ( Table 1, lines 13-16 and 21-24). In the corresponding threecomponent mixtures, the volume fraction of ␣␤ protomers remained almost unchanged with respect to free GltS (Table 1, lines 1-4), but the amount of the dimeric particles systematically grew with the dilution of the GltS solution. This finding suggests that NADP ϩ acts on the interface between (␣␤) 2 dimers of adjacent pillars by disrupting to some extent the (␣␤) 6 hexamers (Figs. 2 and 3). This is at variance with the effect of ionic strength, where only intraprotomeric ␣-␤ interactions are maintained.

The Hexameric NADPH-GltS Structure
The possibility of dissociation of GltS into the isolated ␣ and ␤ subunits was also considered in OLIGOMER (not shown). In some cases, slightly better fits were obtained. However, the improvement was not systematic, and the volume fractions of the individual subunits were in all cases below 10%, in agreement with the Ͻ15% loss of activity observed with samples of GltS upon extended incubation with NaCl (supplemental Fig. 3 and Table 1).
Earlier SAXS results suggested a tetrameric (␣␤) 4 assembly of GltS in solution (54), assuming monodispersity of the GltS samples. Given the new evidence for the co-existence of different GltS oligomers, the previous data were reanalyzed, leading us to the conclusion that the samples studied at that time were actually mixtures of about 20% ␣␤ protomers and 80% (␣␤) 6 hexamers (supplemental Fig. 4) as opposed to more than 90% hexamers from the current measurements.
The oligomeric state of ␣GltS in solution was also reanalyzed (54) by taking into account the possibility of sample heterogeneity. At variance with the GltS holoenzyme, the salt-induced dissociation of ␣GltS was completed in ϳ10 -20 min (supplemental Table 2 and supplemental Fig. 5). This result indicated a lower stability of the high mass species of ␣GltS than that formed by the GltS holoenzyme. The SAXS patterns from solutions of ␣GltS were also measured, although not as titration series (supplemental Fig. 5). Primary analysis of the SAXS data (supplemental Table 3 and supplemental Fig. 5) suggested the presence of ␣ 6 hexameric particles in solution, but their percentage was systematically lower than that found with the GltS ␣␤ holoenzyme. The oligomeric composition of ␣GltS solutions was characterized by OLIGOMER (25) using mixtures of ␣ monomers, (crystallographic) ␣ 2 dimers, and ␣ 6 hexamers (extracted from the cryo-EM-derived GltS model, Fig. 2, B and  C). The maximum content of ␣ 6 hexamers did not exceed 84% (obtained for a 5 mg/ml solution). This finding indicates a higher stability of the GltS (␣␤) 6 hexamer compared with ␣GltS, presumably due to the additional interprotomeric ␣-␤

Overall parameters and volume fractions of the components in GltS solutions obtained by SAXS
Each line of the table corresponds to a SAXS profile in Fig. 4. R g and MM are the radius of gyration and molecular mass calculated from the scattering data, respectively. The observed polydispersity of the samples is expressed by relative percentages of ␣␤ protomers present as ␣␤ monomers, (␣␤) 2 dimers, and (␣␤) 6

hexamers in solution.
Discrepancy between the computed curve and the experimental data is denoted as . MetS, 2-OG, and NADP ϩ concentration was 1 mM each.

Additions
Plot in Fig. 4  interactions. Furthermore, the results are in full agreement with the size distribution of particles observed by cryo-EM using ␣GltS preparations (supplemental Fig. 6). The addition of salt led to dissociation of ␣ 6 hexamers similar to that observed with GltS, but at 2 mg/ml and in the presence of 1 M NaCl, ␣ 6 hexamers were almost fully dissociated into ␣ monomers (supplemental Table 3). The heterogeneity of the previous GltS preparation led to underestimating the oligomer stoichiometry (54). Both gel filtration and SAXS data pointed to a tetrameric assembly (the former due to the nonlinearity of calibration at high MM, the latter due to partial dissociation). Interestingly, even under incorrect stoichiometry and symmetry assumptions, the ab initio shape of GltS by (54) displays an overall hollow appearance similar to that of the present cryo-EM model (supplemental Fig.  7). Furthermore, assuming hexameric species and imposing D3 symmetry, ab initio shape reconstruction and rigid body refinement based on the present SAXS data yielded shapes of ␣ 6 and (␣␤) 6 hexamers similar to those derived from the cryo-EM model.

Effect of the Oligomerization State on the Catalytic Activity of GltS
In order to establish if the oligomerization state of GltS affects its catalytic activity, we first evaluated if we could generate variants of the ␣ or ␤ subunits capable of forming the ␣␤ protomer but unable to establish interprotomeric contacts yielding the ␣␤ protomeric species at all protein concentrations and at low ionic strength. Inspection of the GltS (␣␤) 6 model showed that interprotomeric ␣-␤ and ␣-␣ contacts (Fig. 3) involve several residues without, however, well defined couples of strongly interacting residues (e.g. facing positively/negatively charged residues or hydrophobic patches). Therefore, in order to attempt to produce a monomeric form of GltS for further studies, we generated GltS species formed by the ␤GltS variant carrying a C-terminal His 6 tag (15) and truncated variants of ␣GltS. Deletion of the C-terminal 40 residues of ␣GltS removed the 1438 -1453 ␣ helices resting on the ␤ helical core, which are engaged in the interprotomeric ␣-␤ contacts (Fig. 3, A and B). However, this deletion was too drastic, leading to no production of the truncated ␣GltS protein. On the contrary, removal of the C-terminal 1473-1479 heptapeptide, which is disordered in the ␣GltS crystal structure (9), led to a GltS form (GltSHis⌬7), which was essentially indistinguishable from the full-length GltSHis with respect to protein production levels, behavior, and yields during purification (supplemental Table 4), activity, cofactor content, aggregation state, and its dependence on ionic strength (supplemental Fig. 3).
In light of these results, we turned our attention to the fact that incubation of GltS in the presence of 1 M NaCl leads to full dissociation into ␣␤ protomers. Since dissociation is a slow process (supplemental Fig. 3), but reassociation upon removal of NaCl is fast (supplemental Table 1, supplemental Fig. 3), we had to carry out the determination of the apparent steady-state kinetic parameters (maximum velocity, V max , and K m values for the three substrates) in assay mixtures containing 1 M NaCl using GltS solutions (1 mg/ml) that had been preincubated for 16 -20 h in the absence or presence of 1 M NaCl. In the absence of NaCl, the (␣␤) 6 hexamer prevails, whereas preincubation with NaCl yields the monomeric ␣␤ species. The inclusion of 1 M NaCl in the steady-state kinetic measurements done with the hexameric enzyme form led to a 3-6-fold decrease of the V max value ( Table 2) with respect to that measured in the absence of NaCl (15,23,55). The K m values for NADPH and 2-OG increased by 2 orders of magnitude, and that for L-Gln increased 5-10 fold. The ␣␤ species exhibited the same V max values as the hexameric one but 3-fold lower K m values for 2-OG and L-Gln, which are fully reflected in the catalytic efficiency (V max /K m ) with these two substrates ( Table 2). That dissociation brought about an increase in the catalytic efficiency of the enzyme was confirmed by the correlation of the decrease of the GltS radius (as monitored by dynamic light scattering) and the increase of initial velocity of reactions containing subsaturating concentrations of the substrates and 1 M NaCl when GltS was incubated with 1 M NaCl (supplemental Fig. 8). The modest effect of the oligomerization state on the kinetic properties of the enzyme makes it unlikely that the monomer/hexamer equilibrium significantly affects the biological function of GltS.

DISCUSSION
By combining cryo-EM, SAXS, and molecular modeling, it was possible to propose a structural model of the (␣␤) 6 hexam-

Steady-state kinetic parameters of the glutamate synthase reaction catalyzed by the hexameric and monomeric forms of GltS
Assays were carried out at 25°C in 50 mM Hepes/KOH buffer, pH 7.5, 1 M NaCl by monitoring NADPH oxidation at 340 nm (⑀ ϭ 6.23 mM Ϫ1 cm Ϫ1 ) for NADPH concentrations up to 200 M or 374 nm (⑀ ϭ 2.05 mM Ϫ1 cm Ϫ1 ) for NADPH concentrations above 50 M. The substrate concentration or concentration range is shown. Each assay was done at least in duplicate. Aliquots (5 l) of solutions of GltS (0.5-1 mg/ml) preincubated in 25 mM Hepes/KOH buffer, pH 7.5, 1 mM EDTA, 1 mM DTT in the absence or presence of 1 M NaCl were used to initiate the reaction (1 ml). The apparent maximum velocity (V max ) and K m values for the varied substrate (K varied substrate ) were determined with the Grafit 5.0 software (Erythacus Software Ltd.). The V max /K m values calculated for the varied substrate are indicated. The Hexameric NADPH-GltS Structure MARCH 28, 2008 • VOLUME 283 • NUMBER 13 eric complex of the NADPH-dependent bacterial GltS, the essential enzyme of ammonia assimilation (Figs. 2 and 3). The combined approach allowed us to reconstruct the GltS particles and to solve any ambiguity that might have arisen in the reconstruction process. The high resolution reached (9.5 Å), which led to the observation of secondary structure elements, allowed us to fit the electron density with the atomic models of the GltS ␣ and ␤ subunits. Thus, we obtained information on the interprotomeric and intraprotomeric interface regions and, more importantly, on the structure of the ␣␤ protomer, which is here confirmed to be the minimal catalytically active unit. Our work exploited the complementarity of the two structural methods for the analysis of oligomeric assemblies, which is particularly important for dynamic situations where dissociation cannot be excluded. In cryo-EM analyses, images of individual particles can be selected and processed, allowing one to pick the species of interest. In SAXS, the scattering from the entire ensemble is recorded, and monodispersity has to be assumed for the ab initio or rigid body analysis. The particle symmetry cannot be directly deduced from the scattering data, and additional information is required. In Ref. 54, both gel filtration and SAXS data pointed to a tetrameric assembly, and a P222 symmetry was adopted for the reconstruction. The differences between the "old" and "new" SAXS models in supplemental Fig. 7 are not only due to a stronger dissociation of the old preparation but, mainly, to the different symmetry assumptions. The models generated without the symmetry from the previous and new data were rather similar to each other, albeit having lower resolution. Under the assumption of D3 symmetry, the independently generated ab initio SAXS model agrees well with the cryo-EM data. This example stresses the necessity of verification of sample monodispersity for the structural studies. Indeed, for a monodisperse system, SAXS is able to assess the oligomeric composition (thus, to provide a good estimate of the possible symmetry) simply by evaluating the MM of the solute. For mixtures, the experimental MM value from SAXS is an average over the ensemble, and, if dissociation is present, it underestimates the mass of the highest oligomer. At the same time, our work underlines the strength of SAXS in providing direct quantitative characterization of mixtures of different species in solutions as a function of concentration and of the presence of various factors.
In the GltS (␣␤) 6 hexamer, extended contacts are made within the ␣ 2 dimers forming the body of each one of the pillars (Figs. 1, E and F, and 2), which correspond to the crystallographic asymmetric unit. The interprotomeric contacts between ␣-␣ and ␣-␤ subunits are weaker. The model is consistent with several observations. In the crystals of ␣GltS, ␣ 2 dimers form the asymmetric unit. However, adjacent dimers interact with each other forming tetramers resembling the cryo-EM determined oligomer, missing one of the pillars (supplemental Fig. 1). The presence of interprotomeric ␣-␤ contacts explains why the ␣GltS oligomer, which also forms hexamers at high protein concentration and low ionic strength (supplemental Table 3; supplemental Figs. 4 and 5), dissociates more readily than the (␣␤) 6 hexamer. The same interprotomeric ␣-␤ contacts may rationalize the effect of NADP ϩ on the oligomerization state of the enzyme. NADP ϩ , although at a saturating concentration, induced partial dissociation of the (␣␤) 6 hexamer (with no activity loss; not shown) and led to the appearance of a species that could be identified with the (␣␤) 2 dimer forming the hexamer pillar. Indeed, NADP ϩ binding to ␤GltS might induce a conformational change sufficient to weaken the interaction with the ␣ subunit of the adjacent protomer. In support of NADP ϩ -induced conformational changes is the fact that it was found to modify the kinetics of proteolytic cleavage of both ␣ and ␤ subunits with trypsin and chymotrypsin in the native GltS (56). Finally, the minimal stable unit of GltS is the ␣␤ protomer, which is obtained at high ionic strength.
Thus, it appears that the C-terminal domain of ␣GltS, whose core is formed by a so far unique right-handed ␤ helical fold, serves as a structural spacer in three ways: (i) its ␤ helical core region participates in building the ␣ subunit keeping the amidotransferase domain and the synthase domain in place and allowing them to form the intramolecular ammonia tunnel (Fig. 3A) (9); (ii) the ␣ helices protruding from it and resting on its side mediate interprotomeric ␣-␤ contacts while (iii) its N-terminal part is implicated in the interprotomeric ␣-␣ contacts.
To our knowledge, no protein of known structure, other than GltS, has been found to contain this ␤ helical structure, so that it remains to be established if this fold plays a similar role in other proteins. Data bank searches revealed that the only proteins showing regions of significant sequence similarity with the C-terminal region of ␣GltS are the C subunits of tungstenor molybdenum-dependent formyl methanofuran dehydrogenases (see Interpro classification: IPR002489). However, to our knowledge, no data are available on the specific role of this region in these enzymes.
Because of the relatively low resolution of the current density map in terms of atomic interpretation and the absence of obvious residues that might be specifically responsible for the interprotomeric ␣-␣ and ␣-␤ interactions, efforts to induce monomerization of GltS by site-directed mutagenesis were done by constructing deletion mutants. The deletion of the C-terminal 7 residues of ␣GltS (residues 1473-1479), which are unresolved in the crystal structure of ␣GltS, led to a fully active enzyme, similar to the wild-type species with respect to stability and aggregation state. On the contrary, the deletion of the C-terminal 40 residues of ␣GltS led to no protein production. This result indicates that the deletion prevents protein folding, in agreement with the fact that the deleted protein fragment forms the 1438 -1454 ␣ hel-ices that cover a highly hydrophobic surface of the ␤ helix and completes the ␤ helical core contacting the seventh and final turn (Fig. 3A).
The structure of the ␣␤ protomer extracted from the (␣␤) 6 hexamer is consistent with the hypothesis that the ␣␤ protomer is the minimal catalytically active unit of GltS. In both the (␣␤) 6 hexamer and in the ␣␤ protomer extracted from it, the three catalytic subsites are accessible to substrates, and no interprotomeric electron transfer pathways can be observed, confirming previous spectroscopic observations (57). The spacial arrangement of the five cofactors is linear (Fig. 2E), implying that electron transfer from FAD to FMN occurs in two subsequent oneelectron transfer events, with the two electrons following the same path. The E m values of the oxidized and the hydroquinone forms of the FAD and FMN cofactors and of the oxidized and reduced [3Fe-4S] cluster have been determined (58). Estimates of the E m values of the oxidized/semiquinone and semiquinone/hydroquinone forms of the flavins have also been calculated (58) (supplemental Table 5).
With those values and assuming that the two [4Fe-4S] clusters are equipotential with an E m value as low as that of the NADP ϩ /NADPH couple (Ϫ340 mV) (53,58), the scheme shown in Fig. 5 can be built. This pathway differs from the previously proposed bifurcated electron transfer pathway (58) for a greater number of thermodynamically unfavored steps. The latter are, however, possible within an overall thermodynamically favored reaction with precedents found in, for example, the well characterized fumarate reductase (59).
Interestingly, Phe 54 of ␤GltS, a residue conserved as a Phe or a Tyr in essentially all ␤GltS, seems to be suitably positioned to mediate the electron transfer from the ␤GltS [4Fe-4S] cluster closest to the interface to the [3Fe-4S] center in ␣GltS (Fig. 2E).
The question of the physiological meaning of the protomer/ hexamer equilibrium in GltS remains open. We have measured the apparent steady-state kinetic parameters V max and K m for the three GltS substrates in the presence of 1 M NaCl using FIGURE 5. Electron transfer pathway in NADPH-GltS. The geometry of the five redox centers of GltS ␣␤ protomer extracted from the cryo-EM-determined hexameric complex (Fig. 2) and the estimates of their midpoint potential (E m ) values (supplemental Table 5) (53,58) allow us to propose that electron transfer from FAD to FMN takes place with two one-electron transfer processes that follow the same pathway. However, the energetics of transfer of the first (empty circle, continuous arrows, steps 2-4) and the second electron (full circle, dashed arrows, steps 5-7) differ due to the different E m values of oxidized/semiquinone and semiquinone/hydroquinone forms of both flavin cofactors. In these schemes, the two low potential [4Fe-4S] clusters of GltS (53) are assumed to be equipotential with E m values in the range of that of the NADP ϩ /NADPH couple (Ϫ340 mV). The two-electron transfer process from NADPH to FAD (step 1), and from FMN hydroquinone to the 2-iminoglutarate intermediate resulting from the addition of ammonia and 2-OG (step 8) are shown with the broad arrows containing the electron pair. In the top panel, the electron transfer process is depicted using the E m values determined for unliganded GltS, whereas in the lower panel, those determined for GltS in complex with 2-OG are used. It is not known if 2-OG binding precedes reduction of GltS cofactors, although it binds to oxidized GltS, and which is the effect of the 2-iminoglutarate intermediate on the redox properties of the cofactors. enzyme solutions that had been preincubated in the absence or presence of NaCl (1 M, 20 h). NaCl in reaction mixtures containing the hexameric species brought about 10 -100-fold increase of the K m values for L-Gln, 2-OG, and NADPH (Table  2). This was accompanied by a 3-6-fold decrease of the V max value. The same experiments carried out with enzyme that had been preincubated with 1 M NaCl to ensure dissociation into ␣␤ protomers showed V max values similar to those measured with the hexameric form. However, the K m values for 2-OG and L-Gln were ϳ3-fold lower than those measured for the hexameric form. The resulting modest (3-fold) increase of the catalytic efficiency with 2-OG and L-Gln of the ␣␤ protomer, as compared with the hexameric form, leads to the conclusion that the oligomerization behavior of GltS should only have a minor effect on its functionality in vivo. Rather, it may be suggested that the oligomerization state of GltS may be relevant for its interaction with (so far unknown) regulatory ligands or other proteins.