The Crystal Structures of Apo and Complexed Saccharomyces cerevisiae GNA1 Shed Light on the Catalytic Mechanism of an Amino-sugar N -Acetyltransferase*

The yeast enzymes involved in UDP-GlcNAc biosynthesis are potential targets for antifungal agents. GNA1, a novel member of the Gcn5-related N -acetyltransferase (GNAT) superfamily, participates in UDP-GlcNAc biosynthesis by catalyzing the formation of GlcNAc6P from AcCoA and GlcN6P. We have solved three crystal structures corresponding to the apo Saccharomyces cerevisiae GNA1, the GNA1-AcCoA, and the GNA1-CoA-GlcNAc6P complexes and have refined them to 2.4, 1.3, and 1.8 Å resolution, respectively. These structures not only reveal a stable, b -intertwined, dimeric assembly with the GlcNAc6P binding site located at the dimer interface but also shed light on the catalytic machinery of GNA1 at an atomic level. Hence, they broaden our understanding of structural features required for GNAT activity, provide structural details for related aminoglycoside N -acetyltransferases, and highlight the adapta-bility of the GNAT superfamily members to acquire various specificities. m M The activity of these enzymes was verified. Crystallization— Crystals of the native and protein

Although the GNAT AcCoA binding site is well documented, the binding site of the acceptor substrate has been characterized in only two cases: the histone binding site in the tGCN5-CoA-H3 peptide complex structure (7) and the serotonin binding site in the AANAT-bisubstrate analog complex structure (10). These two structures also provide the first glances into the catalytic machinery of GNAT. Yet, a better understanding of the diverse modes of acceptor substrate recognition and the catalytic mechanism of the GNATs as well as further insights into the evolution of this superfamily still await additional structural studies of GNAT-substrate complexes.
Glucosamine-6-phosphate N-acetyltransferase 1 (GNA1) is a novel amino-sugar N-acetyltransferase member of the GNAT superfamily. GNA1 holds a key position in the pathway toward de novo synthesis of the essential metabolite UDP-GlcNAc and is present in various eukaryotic organisms (Fig. 1A). The GNA1 murine homologue, EMeg32, which possesses an extra 31-residue NH 2 -terminal region compared with the yeast homologues, has recently been characterized (13). EMeg32 is essential for embryonic development, and its inactivation in mouse embryonic fibroblasts generates resistance to apoptotic stimuli and defects in cell proliferation (14). GNA1 has also been characterized in yeast (15) and shown to be essential to the survival of Saccharomyces cerevisiae in which it controls multiple cell cycle steps (16). In addition, Candida albicans GNA1 null mutants exhibit reduced virulence when injected into mice (17), making GNA1 a potential target for the development of new antifungal agents.
We present here three high resolution crystal structures of GNA1. The apo GNA1 and the binary GNA1-AcCoA complex structures were solved independently using MAD techniques and refined to 2.4 and 1.3 Å resolution, respectively (Table I). The structure of the GNA1-CoA-GlcNAc6P ternary complex was solved by molecular replacement and refined to 1.8 Å resolution (Table I). These three structures reveal GNA1 catalytic features and provide the first complete picture of an amino-sugar GNAT active site as a first step toward the development of specific inhibitors.

EXPERIMENTAL PROCEDURES
Expression and Purification of the Native and Selenomethionyl Proteins--Full-length GNA1 was amplified from genomic S. cerevisiae DNA, cloned into the pQE30 expression vector, and transformed into M15pREP4 (Qiagen) Escherichia coli cells. Protein expression was induced with 0.1 mM isopropyl-1-thio-␤-D-galactopyranoside for 20 h at 37°C. Selenomethionyl GNA1 was produced as previously published (18). The recombinant His-tagged native and selenomethionyl enzymes were purified via nickel affinity and anion exchange chromatographies, dialyzed against 10 mM Tris-HCl, pH 8, 150 mM NaCl, 1 mM dithiothreitol, and concentrated to 10 mg/ml. The activity of these enzymes was verified.
Crystallization-Crystals of the native and selenomethionyl protein were grown at 20°C using the hanging-drop vapor diffusion method by mixing equal volumes of the protein solution and the reservoir, which contained 17-22% polyethylene glycol 600 and 0.2/0.4 M imidazole/ malate, pH 5.1. Crystals obtained from the apo protein belong to the orthorhombic space group P2 1 2 1 2 1 and contain 6 molecules/asymmetric unit. Those obtained from GNA1 preincubated overnight with 1 mM AcCoA or CoA belong to the monoclinic space group C2 and contain 4 molecules/asymmetric unit. For data collection, the crystals were transferred to a reservoir solution containing 32% polyethylene glycol 600 and flash-frozen at 100 K in gaseous nitrogen. Data Collection-A three-wavelength MAD experiment was performed at the ESRF beamline ID14-4 (European Synchroton Radiation Facility, Grenoble, France) on both the AcCoA-complexed and apo selenomethionyl protein crystals at 2.8 and 3.2 Å resolution, respectively. High resolution data sets were collected at 2.4 and 1.3 Å resolution on an apo GNA1 crystal and a GNA1-AcCoA complex crystal, respectively, on beamline ID14-2 (ESRF, Grenoble). A 1.8 Å resolution data set was collected at EMBL-X31 (DESY, Hamburg) from a GNA1-CoA complex crystal soaked in the reservoir solution supplemented with 1 mM GlcNAc6P for 6 days.
Structure Solution, Model Building, and Refinement-All data were processed and reduced using DENZO (19) and the CCP4 program suite (20). The apo and AcCoA-complexed GNA1 structures were solved independently using SOLVE (21). The experimental MAD electron density maps were improved by solvent flattening, non-crystallographic symmetry averaging, and phase extension with the program DM (20). The apo GNA1 model was built manually using TURBO-FRODO (22). The GNA1-AcCoA complex model was built automatically using ARP/ wARP (23). These two models were refined against their respective high resolution data set using the program CNS (24), including bulk solvent and anisotropic B-factor corrections. NCS restraints were used only for the apo GNA1 model. High temperature factors and weak electron density maps are associated with residues Gln-52 to Lys-57 in the two models. The 3 intertwined dimers of the apo GNA1 model have an average root mean square deviation of 0.7 Å for all C␣ atoms. In the GNA1-AcCoA complex model, the root mean square deviation value between the 2 dimers is 0.4 Å for all C␣ atoms. The GNA1-CoA-GlcNAc6P complex structure was obtained from a rigid body refinement using the GNA1-AcCoA complex as a starting model. Fourier difference maps clearly revealed the location of the bound CoA and GlcNAc6P in two of the four molecules. The structure of the GNA1-CoA-GlcN6P complex was also solved at 2.5 Å; superimposition of the two ternary complexes (GNA1-CoA-GlcNAc6P and GNA1-CoA-GlcN6P) revealed that GlcN6P (the substrate) and GlcNAc6P (the reaction product) were positioned similarly. Because the GNA1-CoA-GlcNAc6P structure was obtained at a higher resolution than that of GNA1-CoA-GlcN6P, we only considered in the analysis the GNA1-CoA-GlcNAc6P complex structure. The stereochemistry of the refined models was analyzed by PROCHECK (25); no residue was found in the disallowed regions of the Ramachandran plot. The coordinates of apo, AcCoA-, and CoA-GlcNAc6P-complexed GNA1 have been deposited in the Protein Data Bank (accession codes 1I21, 1I12, and 1I1D).

RESULTS AND DISCUSSION
Overall Structure-The three-dimensional structures of GNA1 in its apo state and complexed forms with AcCoA or with CoA and GlcNAc6P have been solved and refined at 2.4, 1.3, and 1.8 Å resolution, respectively. Overall, the electron density is well defined for these structures (Fig. 1B) except for a surface loop comprising residues Gln-52 to Lys-57 (cf. "Experimental Procedures"). As predicted from sequence analysis, GNA1 shares structural similarities with other GNAT superfamily members (1,11). The GNA1 fold consists of a central core, composed of a mixed 5-stranded ␤-sheet flanked by 4 ␣-helices, and a COOH-terminal strand ␤6, which is projected away from the central core ( Fig. 2A).
The GNA1 structure is dimeric in the crystal as well as in solution, as attested from gel filtration data (not shown). The crystalline dimer is made of two intertwined GNA1 monomers in which strand ␤6 of one subunit exchanges with the identical strand from the other subunit ( Fig. 2A). A ␤-strand exchange between subunits in a dimer is an unusual feature among GNATs and has been observed only in the HAT Hpa2 structure (9). In all other structurally characterized GNAT, except Hat1 that lacks a ␤6 strand (5), the hinge loop preceding strand ␤6 folds back onto its own subunit. This difference is reminiscent of three-dimensional domain-swapped proteins in which the loop that precedes the exchanged domain can switch from a closed to an opened conformation thereby leading to either a monomeric or a dimeric form (29). In the case of GNA1 or Hpa2, the ␣4 -␤6 loop is too small to undergo such a conformational switch, and the dimeric assembly is further stabilized by a hydrophobic interface, two features that make three-dimensional domain swapping unlikely. Nonetheless, the monomeric GNATs and the intertwined dimers of GNA1 and Hpa2 are most probably related by divergent evolution from a common ancestor, and the evolutionary mechanisms that have led to dimer formation may have included three-dimensional domain swapping.
The Cofactor Binding Site-In each subunit of the GNA1 dimer, AcCoA is positioned in a large hydrophobic cleft located at the site where the two parallel strands, ␤4 and ␤5, diverge because of a ␤-bulge in strand ␤4 that positions the side chains of Glu-98 and Asp-99 on the same face of the ␤-sheet. The presence of this ␤-bulge is remarkably well conserved among GNATs, which suggests a critical role for this structural element in the formation of the AcCoA binding site.
AcCoA adopts a conformation similar to that described in other AcCoA-complexed GNAT structures (1). The acetyl group of AcCoA, which marks the active site, is located between strand ␤5 and the ␤-bulge and is largely stabilized by contacts with the protein; the two carbon atoms contract hydrophobic interactions with residues Ile-100, Leu-133, and Tyr-143, and the carbonyl oxygen inserts into an oxyanion hole formed by the backbone amides of residues Asp-99 and Ile-100. Such an oxyanion hole has been observed in the structurally related N-myristoyltransferase (30) but is a unique feature within the structurally characterized GNATs.
Superimposition of the apo and AcCoA-complexed GNA1 structures shows that AcCoA binding induces subtle structural rearrangements that are confined to the edges of the cleft and result in a slightly narrower cleft. Residues 102-109 in the ␣3-␤5 loop and 134 -143 in the ␤5-␣4 loop plus the N-cap of ␣4 move by ϳ1.3 and 1.1 Å, respectively, toward the center of the cleft ( Fig. 2A). Whether these conformational changes, induced upon cofactor binding, are a prerequisite for acceptor substrate binding as shown for other GNATs (31)(32)(33) needs to be ascertained by kinetic studies. A detailed comparison with other GNATs reveals that these rearrangements differ from those reported for (i) tGCN5, in which the cofactor-binding cleft opens slightly upon AcCoA binding to accommodate the histone tail (7); and (ii) AANAT, in which a major rearrangement of the ␣1-loop-␣2 region occurs upon AcCoA binding to complete the serotonin binding site (10). Therefore, although the binding of AcCoA is similar among GNATs, it induces different conformational changes that can contribute to the specific binding of the acceptor substrate.
The Acceptor Substrate Binding Site-The GNA1 aminosugar binding site exhibits an atypical architecture, as it is built at the dimer interface and involves residues from the exchanged ␤-strand, two features found only in a few intertwined oligomeric structures such as that of bovine seminal ribonuclease (34). GlcNAc6P binds at the base of the AcCoA cleft within a small pocket that is lined mostly with electronegative residues except for a patch of positively charged residues that specifically accommodate the 6-phosphate group (Fig. 2B). Remarkably, the GlcNAc6P acetyl group is positioned similarly to the cofactor acetyl group in the GNA1-AcCoA complex structure (Fig. 2B). The sugar-6-phosphate establishes numerous hydrogen bonds, mainly via side chain atoms, together with a few hydrophobic contacts such as that found between Leu-27 and the ␤-face of the sugar ring (Fig. 2C). Superimposition of the structures of the cofactor-complexed SmAAT (2) or EfAAT (3) (the two structurally characterized aminoglycoside GNATs) on that of the GNA1-CoA-GlcNAc6P complex shows that SmAAT Phe-51, EfAAT Trp-25, and GNA1 Leu-27 are identically positioned, an observation that supports a common functional role for these residues, thereby identifying an aminoglycoside recognition feature of GNATs.
Catalytic Mechanism-Several reports on GlcN6P N-acetyltransferases suggest that catalysis requires sulfhydryl groupcontaining residues, such as a cysteine that could act as a nucleophile in a two-step mechanism involving the formation of a covalent acetyl-cysteine enzyme intermediate (35,36). In contrast, structural and kinetic data available for GNATs support a mechanism proceeding through a direct nucleophilic reaction of acceptor substrate on AcCoA (1,32,33).
In the GNA1-AcCoA complex structure, the cysteine residue closest to the AcCoA acetyl group lies 6.5 Å apart, too far to play a role in acetyl transfer. Furthermore, no other appropriate nucleophile residue is found in the proximity of the acetyl group, which makes the formation of an acetyl-enzyme intermediate very unlikely and supports the hypothesis of a singlestep mechanism as suggested for GNATs. Consistent with this hypothesis is the position of the amino group of the product GlcNAc6P (similar to that of the substrate GlcN6P; cf. "Experimental Procedures"), which is ideal to allow a direct nucleophilic attack at the AcCoA carbonyl (Fig. 2B). In addition, the nucleophilic character of the amine is enhanced by the hydrogen bond it establishes with the backbone carbonyl of Asp-134 (Fig. 2C). The AcCoA carbonyl is polarized via hydrogen bonds to the backbone amides of Asp-99 and Ile-100, located in the oxyanion hole, a feature that facilitates the nucleophilic attack FIG. 1. Sequence conservation and quality of the GNA1 structure. A, sequence alignment of GNA1 homologues. Conserved and similar residues are highlighted with black and gray backgrounds, respectively. GNA1 secondary structure elements forming the structurally conserved GNAT core are shown in black. The sequence alignment for SmAAT and Hpa2 is based on a structural comparison with GNA1. Subunit 1 residues involved in the GlcNAc6P and AcCoA binding sites are identified by filled circles and filled triangles, respectively, and are shown in black for the residues making GNAT conserved interactions. The unfilled circles indicate subunit 2 residues that complete the GlcNAc6P binding site of subunit 1. The four GNAT sequence motifs are boxed. B, stereoview of the 1.3 Å resolution experimental, solventflattened, averaged electron density map contoured at 1.25 around an AcCoA molecule. and stabilizes the negative charge building up on the oxygen atom of the tetrahedral reaction intermediate (Fig. 3). Finally, the Tyr-143 hydroxyl group, which lies within hydrogen bond distance of the AcCoA sulfur atom (Fig. 2B), could serve to stabilize the thiolate anion of the departing CoA molecule. Tyr-143 also establishes hydrophobic contacts with the acetyl group, probably playing a role in correctly positioning the acetyl group for the reaction. A critical role of Tyr-143 in catalysis is supported by mutagenesis data (15). A close inspection of the active site of GNA1 also pinpoints two significant structural and functional differences with others members of the GNAT superfamily.
The first striking difference resides within the GNA1 ␤-bulge which, when compared with the ␤-bulges in other GNAT structures, shows a markedly different hydrogen bonding pattern. The GNAT ␤-bulge is an irregularity of the antiparallel ␤ structure in which two residues on one strand are facing a single residue on the other strand (37). Such a ␤-bulge is formed when two consecutive residues in a ␤-strand direct their backbone carbonyl or amide toward the same side of the strand, thereby breaking the typical alternated pattern of a ␤-structure. In GNA1, the backbone amides of Asp-99 and Ile-100 are projected toward the active site and form the oxyanion hole, whereas the carbonyls of Glu-98 and Asp-99 point toward ␤3. In all other members of the GNAT superfamily (except EfAAT in which a proline perturbs the bulge conforma- FIG. 2. Structure of GNA1 and Ac-CoA, GlcNAc6P binding sites. A, ribbon representations of the GNA1 fold (left) and the intertwined GNA1 dimer (right). In subunit 1, the GNA1 secondary structure elements forming the structurally conserved GNAT core are shown in green, the exchanged ␤-strand in yellow, and the remaining structural elements in cyan. Subunit 2 is shown in magenta with its exchanged strand ␤6 in red. The molecular surface of AcCoA-(B) and CoA-GlcNAc6P-(C) complexed GNA1, oriented as in Fig. 1B (left view) and color-coded (B) as in Fig. 1B, with the regions undergoing small structural rearrangements upon AcCoA binding displayed under a transparent surface (the cyan and yellow bonds refer to the apo and AcCoA-complexed GNA1 models, respectively). Ac-CoA is shown with carbon (white), nitrogen (blue), sulfur (green), oxygen (red), and phosphorous (purple) atoms. C, the color code is according to the electrostatic potential with positive and negative charges shown in blue and red, respectively. The essential catalytic Tyr-143 is displayed through a transparent surface. CoA and GlcNAc6P are shown with yellow carbon atoms. D, stereoview of the GlcNAc6P binding site with residues from subunit 1 and 2 shown in cyan and magenta, respectively. The dotted lines indicate hydrogen bonds. Residues within the GNAT conserved ␤-bulge are displayed in green. tion (3)), the situation is reversed; the oxyanion hole is absent because the two consecutive backbone amides are now directed toward strand ␤3 (establishing an hydrogen bond with the backbone carbonyl of the facing residue), whereas two consecutive main chain carbonyls are found pointing into the active site (Fig. 4). These two carbonyls have been suggested to play a role in acceptor substrate binding in tGCN5 (7) and in the stabilization of catalytic water molecules in AANAT (10); this indicates that, in addition to a common role in structuring the cofactor binding site, the ␤-bulge could also fulfill other nonconserved catalytic functions.
The second important difference concerns the deprotonation of the acceptor substrate amino group prior to the reaction. In the case of tGCN5 and AANAT, a chain of well ordered water molecules, or "proton wire," connecting the acceptor substrate amino group to the proposed catalytic bases tGCN5 Glu-122 or AANAT His-120 was suggested to be involved in this proton removal (7,10). In the GNA1-CoA-GlcNAc6P complex, a similar proton wire is observed, leading to Glu-98 in which the side chain occupies a similar position as that of tGCN5 Glu-122 and AANAT His-120. However, the E98A mutation does not abolish the GNA1 activity (15), suggesting that Glu-98 might not function as the general base. Nevertheless, deprotonation prior to the reaction might not be necessary in the case of GNA1, because the pK a of GlcN6P (ϳ7.75) is lower than that of other GNAT acceptor substrates such as lysine (8.95) or serotonin (ϳ10). This hypothesis is also supported by the fact that the optimum pH of purified mammalian GlcN6P N-acetyltransferases lies in the alkaline range (35,36) and by the lower K m value of S. cerevisiae GNA1 for GlcN6P at pH 8 than at pH 7.5 (13), suggesting that GNA1 may preferentially bind the basic/ deprotonated form of GlcN6P.
Substrate Specificity among GNATs-Although the GNAT enzymes share structural similarities, they have distinct acceptor specificities, consistent with their implication in various biological processes. A comparative analysis of the complexes of GNA1-CoA-GlcNAc6P, tGCN5-CoA-H3 peptide (7), and AANAT-bisubstrate analog (7, 10) highlights the structural determinants responsible for the substrate specificities among GNATs. Importantly, this knowledge is essential for the design of specific inhibitors for medical applications.
The structural comparison of these three complexes reveals that both the NH 2 -and COOH-terminal regions diverge between the different GNAT structures and are important for substrate specificity. The NH 2 -terminal structural differences concern the ␣1-loop-␣2 region and are relatively minor, whereas more dramatic changes occur in the COOH-terminal end. In AANAT, the ␣4 -␤6 loop orients toward the active site as it folds back on its own subunit. This loop, along with the ␣1-␣2 loop, almost covers the active site, thus facilitating the binding of a hydrophobic substrate (Fig. 5A). In tGCN5, the 20-residue segment inserted between ␣4 and ␤6 contributes to one side of the substrate binding canal, providing specific binding residues for the histone tail (Fig. 5B).
In GNA1, the shorter ␣4 -␤6 loop does not participate directly in acceptor substrate binding, but it forces strand ␤6 to extend and exchange with the identical strand of the other subunit in the dimer. Instead of using this loop, GNA1 exploits its intertwined oligomeric state to achieve specific binding, because an important part of the GNA1 active site consists of residues from the other subunit in the dimer (Fig. 5C). In contrast, this region is partly replaced in monomeric AANAT by a long loop inserted between strands ␤3 and ␤4, which contributes largely to one wall of the serotonin binding site (Fig. 5A). For the monomeric tGCN5, the short ␤3-␤4 turn contributes to the canal-shaped active site designed to accommodate a long peptidic chain (Fig. 5B).
Role in the Cell Cycle-GNA1 was shown to control multiple cell cycle steps in S. cerevisiae (16). It is still unclear whether this role is related to the N-acetyltransferase activity of GNA1 in UDP-GlcNAc biosynthesis (which implies a physiological link between UDP-GlcNAc and cell cycle progression) or if it is the consequence of an additional function of GNA1.
The hypothesis of an additional HAT activity for GNA1 was addressed, but no HAT activity could be detected in vitro (15). Interestingly, a comparison of GNA1 with the related GNAT structures reveals that the closest structural homologue of GNA1 is the HAT Hpa2, which also adopts an intertwined dimeric structure (9). Superimposition of the two structures reveals differences in the relative arrangement of the two subunits, resulting in different acceptor substrate binding sites. A narrow open-ended channel, in which the histone tail could insert, is found in the Hpa2 structure instead of the rounded pocket of GNA1, which seems unlikely to accommodate an extended and bulky histone tail.
Could GNA1 fulfill an additional function via the noncovalent association with a particular cell compartment or with a protein partner? An association with the cytoplasmic face of organelle membranes has been described for EMeg32 (the GNA1 murine homologue) which also co-purifies with the cdc48 homologue protein (p97/VCP) (13). Double-hybrid systematic experiments performed in S. cerevisiae revealed interactions between GNA1 and a priori unrelated or unknown proteins (38). Further biochemical experiments are needed to determine the biological relevance of these protein/membrane interactions.
UDP-GlcNAc is a key precursor of chitin (a component of the yeast and fungal cell wall) as well as of the glycosylphosphatidylinositol anchor of membrane-bound proteins and is essential to N-linked glycosylation and O-GlcNAc modification of proteins. Glycosyltransferases involved in N-glycosylation, such as the yeast GPT/alg7, which uses UDP-GlcNAc as a substrate, have been suggested to play a role in the cell cycle (39). In addition, a recent report shows that EMeg32-dependent UDP-GlcNAc levels influence cell cycle progression and apoptosis signaling (14). Hence, the role of GNA1 in cell cycle progression appears to be linked to its key GlcN6P N-acetyltransferase activity in de novo UDP-GlcNAc biosynthesis. The structural data presented here have allowed us to propose a catalytic mechanism for GNA1, as well as providing a structural template for GNA1 homologues and related aminoglycosides GNATs. Finally, these results further exemplify the remarkable diversity of the GNAT superfamily and represent a critical step toward the development of specific inhibitors.