Cloning and Characterization of the Murine Glucosamine-6-phosphate Acetyltransferase EMeg32 DIFFERENTIAL EXPRESSION AND INTRACELLULAR MEMBRANE ASSOCIATION*

N -Linked glycosylation is a post-translational modifi-cation occurring in many eukaryotic secreted and sur-face-bound proteins and has impact on diverse physio-logical and pathological processes. Similarly important is the generation of glycosylphosphatidylinositol linkers, which anchor membrane proteins to the cell. Both protein modifications depend on the central nucleotide sugar UDP- N -acetylglucosamine (UDP-GlcNAc). The enzymatic reactions leading to generation of nucleotide sugars are established, yet most of the respective genes still await cloning. We describe the characterization of such a gene, EMeg32 , which we identified based on its differential expression in murine hematopoietic precursor cells. We further demonstrate regulated expression during embryogenesis. EMeg32 codes for a 184-amino acid protein exhibiting glucosamine-6-phosphate acetyltransferase activity. It thereby holds a key position in the pathway toward de novo UDP-GlcNAc synthesis. Surprisingly, the protein associates with the cytoplasmic side of various intracellular membranes,

Asparagine (N)-linked oligosaccharides are components of most secreted and surface-bound proteins of mammalian cells. Carbohydrate structures on cell surface glycoproteins are instrumental for cell-cell or cell-matrix interactions, immune reactions and tumor development, whereas sugar modifications on secreted proteins are important for their transport, their biological activity and their clearance from the circulation (1,2).
The diversity of sugar units within complex or hybrid Nlinked oligosaccharides is generated by the controlled action of specific glycosyltransferases. The process starts in the endo-plasmic reticulum (ER) 1 with the transfer of an oligomannose precursor from the lipid carrier dolichol phosphate to asparagines within the nascent protein chain (3). Maturation of the oligomannose precursor occurs during passage of the attached protein through cis-, medial, and trans-Golgi and is mediated by several trimming enzymes (e.g. mannosidases) and glycosyltransferases (for a review, see Ref. 2).
At the basis of each N-linked carbohydrate is N-acetylglucosamine (GlcNAc), whose initial transfer to dolichol phosphate is achieved in the form of its nucleotide sugar donor UDP-GlcNAc. GlcNAc not only constitutes the first aminosugar residue linked to asparagine in all N-glycosylated proteins, but also serves as an important module in generating complex or hybrid N-linked oligosaccharide structures as well as O-linked oligosaccharides (3). In addition, a number of proteoglycans contain GlcNAc linked to the respective core protein or other aminosugar residues. It also serves in generating glycosylphosphatidylinositol (GPI) linkers, which are responsible for anchoring a variety of cell surface molecules to the plasma membrane (4 -6). The initial reaction of GPI assembly (the transfer of GlcNAc from UDP-GlcNAc to phosphatidylinositol) is uniformly distributed to the cytoplasmic leaflet of the ER (7).
Generation of UDP-GlcNAc is possible by both de novo synthesis and salvage pathways and the respective enzymes have been identified (for a review, see Ref. 8). In eukaryotic cells the sequence of intermediates is fructose 6-phosphate 3 glucosamine 6-phosphate 3 N-acetylglucosamine 6-phosphate 3 Nacetylglucosamine 1-phosphate 3 UDP-GlcNAc. In addition, two alternative routes are described for generating UDP-Glc-NAc. In one case, GlcNAc from degradation of nutritional constituents or from lysosomal degradation of oligosaccharidemoieties is phosphorylated by N-acetylglucosamine kinase (9) to generate N-acetylglucosamine 6-phosphate. This then feeds into the pathway described above. The second route uses the sequence galactosamine 3 N-acetylgalactosamine 1-phosphate 3 UDP-N-acetylgalactosamine 3 UDP-GlcNAc (for a review, see Ref. 8).
It is not clear at present to which extent the redundancy in generating UDP-GlcNAc is employed by cells to respond to various extracellular stimuli, nor how the expression of the respective enzyme genes is regulated, because most of the corresponding cDNAs have not been cloned. We have recently developed a PCR-based approach to comprehensively analyze gene expression in rare normal murine hematopoietic precursor cells and have used this system to identify novel genes specific for certain precursor types (10 -12). We report here the cloning and characterization of a murine cDNA (EMeg32), which displayed a strong differential expression pattern in adult hematopoietic precursor cells. EMeg32 encodes a protein of 184 amino acids exhibiting glucosamine-6-phosphate acetyltransferase activity, which is known to be a key step toward UDP-GlcNAc synthesis. Despite the fact that nucleotide sugars are synthesized in the cytoplasm, we find EMeg32 associated with Golgi and other intracellular vesicle membranes. We also demonstrate that EMeg32 protein copurifies with the evolutionary conserved cdc48 homolog valosin-containing protein (VCP), a 97-kDa ATPase involved in the reassembly of mitotic Golgi fragments. Whether this is important in light of the accumulation of EMeg32 protein at the G 2 /M boundary of the cell cycle is discussed.

EXPERIMENTAL PROCEDURES
Identification of Difference Tags and Screening of cDNA Libraries-Generation of difference sequences by poly(A) PCR from single hematopoietic cells and subtractive hybridization to enrich for cDNA species present in the "tracer" sample (i.e. bipotent erythroid/megakaryocytic precursor, E/Meg) but not in the "driver" sample (i.e. bipotent neutrophil/macrophage precursor, G/M), is described (10,12). 5 ϫ 10 5 plaqueforming units from a ZAPII cDNA library generated from S17 stroma cells were screened with the EMeg32 difference tag. DNA was prepared from single positives and digested with NotI to release 2.5-kb inserts, which were subcloned into pBlueScriptIISKϩ (Stratagene) thereby generating pSKII-EMeg32. Both strands of the respective inserts were sequenced to completion. The original tag corresponds to EMeg32 cDNA positions nt 2067-2344. Alignment of the two sequences revealed four point mutations in the tag, probably generated by errors in the singlecell PCR reaction.
RACE-PCR-Synthesis of double-stranded cDNA for Marathon TM RACE (CLONTECH) starting from 5 g of total RNA from spleen or FDC-P1 cells, ligation of cDNA adaptors, and 5Ј RACE using "touchdown PCR" were as described (12) with 10 pmol of Marathon TM -adaptor primer AP1 (5Ј-CCA TCC TAA TAC GAC TCA CTA TAG GGC-3Ј) and  20 pmol of EMeg32-specific 3Јprimer 133m (5Ј-CTG TAT TCT GAC TCC  AGT CCA CTT CTT TGA, 451m (5Ј-CTC TGC ACT CGT CAC TAA CGA  CGA-3Ј), or 746m (5Ј-TAC TAC ACA CTG GCA CAC CAC A-3Ј) (see Fig.  1). 3Ј RACE was done with AP1 primer and 5Ј primers 403p (5Ј-GCT AAG AGA GGG AGA GTA GAA GAT GT-3Ј) or 2068p (5Ј-CAA GTA CAA TGG CAC AAG AG-3Ј) accordingly. Several independent clones from each primer combination were sequenced. Point mutations in the RACE clones as compared with the ZAPII cDNA insert were mutually exclusive. 5Ј RACE products did not extend the 5Ј end of the original cDNA insert; 4 out of 15 clones analyzed started with its first nucleotide. 3Ј RACE products extended the original cDNA insert by 70 bp, revealing a cluster of poly(A) signals at the extreme 3Ј end. Poly(A) tails at cDNA positions 1506, 1851, and 1855 were obtained from some 3Ј RACE subclones indicative of degenerate poly(A) signals (ATTAAA) at positions 1472 and 1832 (also see Fig. 1).
RNA Preparation and Northern Analysis-Cell lines were cultured as described previously (11). Total RNA was prepared from murine tissues and cell lines and subjected to Northern blot analysis (12 g/ sample) as described (13). The EMeg32-specific tag probe was an EcoRI fragment of the original difference clone and therefore flanked by 26(A) at its 3Ј end and 34(T) at the 5Ј end. Nested cDNA probes A, B, and C in Fig. 2 were generated from pSKII-EMeg32 by restriction enzyme digest using NotI-SacI (probe A), SacI-SspI (probe B), or SspI (probe C). In addition, a NcoI-SmaI fragment corresponding to the extreme 3Ј end of EMeg32 cDNA downstream of the original tag was used.
Plasmids and Constructs-A NotI site was introduced downstream from the EMeg32 STOP codon and used to fuse a trimerized hemagglu-tinin (HA) tag followed by 10 histidines in frame to the EMeg32 ORF. The fusion construct was subcloned into pBlueScriptIISKϩ generating pSK-EH. For MSCV-EMeg32/HAHIS, an EcoRI fragment of pSK-EH was inserted into the respective site in MSCV-neo (14). The amino acid fusion sequence starting at EMeg32 aa 181 is RFLKSGRIF YPYDVP-DYA G YPYDVPDYA GS YPYDVPDYA AQRG(H) 10 , with the HA tags underlined. MSCV-EMeg32 was generated by inserting an EcoRI-BamHI fragment of pSKII-EMeg32 ORF into the respective sites of MSCV-neo. To obtain pSKII-EMeg32 ORF , an EcoRI site upstream of the first ATG and a BamHI site downstream of the EMeg32 ORF were introduced by PCR and the respective EcoRI-BamHI fragment subcloned into pBlueScript II SKϩ. pGEX-EMeg32 was generated by subcloning an EcoRI-NotI fragment of pSKII-EMeg32 ORF in frame into the respective sites of pGEX-4T2 (Amersham Pharmacia Biotech). pGEX2T-ScGNA1 is described (15). To generate pFASTBAC-His/ EMeg32, an EcoRI-XbaI fragment of pSKII-EMeg32 ORF was subcloned in frame into pFASTBAC1-HTb (Life Technologies, Inc.). The amino acid sequence preceding the EMeg32 ORF is MSY Y(H) 6 D YDI PTT ENL YFQ GAM GSG IPG KMK, with the first EMeg32-specific amino acids underlined and the His tag in bold. To generate pKS-UTR, a SacI-SspI fragment (nt 695-1953 in EMeg32 cDNA) of pSKII-EMeg32 was subcloned into the SacI/SmaI sites of pBluescript KSϩ (Stratagene).
In Situ Hybridization-For the generation of antisense probes, pSKII-EMeg32 ORF was linearized with EcoRI and pKS-UTR with SacI followed by in vitro transcription using T3 RNA polymerase in the presence of digitoxigenin-coupled UTP. In situ hybridizations on sections of 129/Sv or CD1 embryos were done as described (16).
Cell Fractionation and Salt Extraction-Fractionation of EpH4 cells and sucrose density gradients were done as described (17). For salt extraction, postnuclear supernatant was prepared as described for the density gradients. Equal volumes of postnuclear supernatant were gently mixed with homogenization buffer (HB) or 1:1 with 0.6 or 2 M NaCl or KCl in HB, to give final concentrations of 0.3 and 1 M, respectively. After 30 min of incubation on ice, membranes were separated from cytosolic and extracted proteins by ultracentrifugation (30 min, 100,000 ϫ g). Membrane pellets were resuspended in the same volume of HB. Soluble and membrane proteins were precipitated with methanol/HCCl 3 , boiled in Laemmli buffer, and analyzed by SDS-PAGE and immunoblotting.
Protein Purification and Ion Trap MS/MS-GST fusion proteins were generated by transformation of pGEX-2T (Amersham Pharmacia Biotech), pGEX-ScGNA1, or pGEX-EMeg32 into Escherichia coli DH5␣. 50-ml cultures were induced with 1 mM isopropyl-1-thio-␤-D-galactopyranoside (3 h; 30°C), centrifuged, and pellets were washed in ice-cold PBS prior to sonification in 8 ml of AT buffer (see below) containing protease inhibitors. Supernatants of cleared lysates were incubated (2 h; 4°C) with glutathione-coupled Sepharose beads. Following three washes in 5 ml of AT buffer, bound proteins were eluted twice using 600 l of E buffer (20 mM glutathione, 100 mM Tris-Cl, pH 8.0, 120 mM NaCl) and dialyzed overnight against 1 liter of AT buffer.
For analysis of the 200-kDa band co-purifying with His-EMeg32 (but not with His-AIF (His-tagged apoptosis inhibitory factor)) from baculoextracts, its stable, monomeric 100-kDa form was reduced with dithiothreitol, carboxyamidomethylated with iodoacetamide, and separated on SDS-PAGE. The 100-kDa band was cut out, in-gel digested with Lys-C endoprotease, and run on a capillary liquid chromatography connected in line to an ion trap-tandem mass spectrometer (MS/MS). Acquired MS/MS spectra were searched against the OWL data base using SEQUEST ( Centrifugal Elutriation-Following one wash, 10 8 FDC-P1 cells were resuspended in 10 ml of EL buffer (PBS without Ca 2ϩ and Mg 2ϩ ; 1 mM dithiothreitol) and loaded on a Beckman JE-5.0 elutriator with a flow rate of 10 ml/min at 2000 rpm (4°C). Flow rate was increased by 2 ml/min for every second 50-ml fraction collected. The fractions used for analysis covered a range from 16 ml/min to 28 ml/min. For the last 2 fractions, 200-ml aliquots were collected. Cells were pelleted and either resuspended in EL buffer plus 1% Triton X-100 for protein analysis, or in 50% ethanol for standard propidium iodide staining and cell cycle analysis on a FACScalibur (Becton Dickinson) using CellQuest software.

Molecular
Cloning of EMeg32-We generated a set of cDNAs from single murine hematopoietic cells with defined developmental potential (10) and applied this system to identify genes specific for bipotent precursors of the erythroid and megakaryocytic lineages (E/Meg) using subtractive hybridization. A cDNA sample from a bipotent neutrophil/macrophage (N/M) precursor (driver) was subtracted from a bipotent E/Meg precursor (tracer), tracer-enriched cDNA material was subcloned, and 56 random difference clones were screened on our hematopoietic hierarchy cDNA blots. Of these, clone 32 was of particular interest, because it exhibited a striking differential expression pattern (also see Fig. 7 in Ref. 10).
This 3Ј sequence tag hybridized to most tripotent erythroid/ megakaryocytic/mast cell (E/Meg/Mast) precursors, strongly to the bipotent E/Meg precursors, and to none of the bipotent N/M precursors analyzed. (Fig. 1A). In descendents of bipotent E/Meg precursors, tag expression is only found in the megakaryocytic, but not in the erythroid branch (Fig. 1A). Expression was also detectable in monopotent macrophage and neutrophil precursors. Besides being expressed in mature megakaryocytes, tag expression is present in mature macrophages, while cDNA from other mature populations did not hybridize (Fig. 1A). Sequencing followed by initial data base searches revealed that this difference tag, EMeg32, was novel.
Subsequently, a cDNA library generated from S17 stroma cells was screened with the EMeg32 subtraction tag. We identified three clones containing a 2.5-kb cDNA insert which contains an open reading frame (ORF) of 184 amino acids (aa) at its 5Ј end (nt 64 -615, Fig. 1B). RACE was performed using RNA from FDC-P1 cells and spleen and various combinations of gene-specific primers. As a result, (i) the original cDNA sequence could be confirmed and extended on the 3Ј end by several nucleotides (Fig. 1B, see "Experimental Procedures"). RT-PCR using primers corresponding to the extreme ends of EMeg32 cDNA confirmed that the additional sequences are valid. (ii) There was no STOP codon in frame upstream of the first ATG chosen. However, the region around this ATG matches the Kozak consensus sequence well.  (Fig. 1B).
Characterization of EMeg32 mRNA Expression-Northern analysis on RNA from several murine tissues and cell lines with probes covering different areas of the cDNA revealed that three mRNA species of 0.9, 1.6, and 2.6 kb in size contain EMeg32 ORF-specific sequences (Fig. 1C). However, the original tag (nt 2067-2344) or a probe derived from sequences downstream (nt 2420 -2536) of the 3Ј tag hybridized to two RNA species of 2.6 and 4.4 kb in size (data not shown). Importantly, the 4.4-kb mRNA species does not hybridize to cDNA probes specific for the EMeg32 ORF or 3Ј-untranslated region up to nucleotide position 1953 (Fig. 1C). While the length of the cloned EMeg32 cDNA (2536 bp) is in good accord with the 2.6-kb mRNA species, the nature of the 4.4-kb mRNA species still remains elusive. Noteworthy, a number of 5Ј ends of murine expressed sequence tags overlap with the extreme 3Ј end of EMeg32 cDNA and we are therefore currently pursuing the possibility that the 4.4-kb transcript represents a mRNA species transcribed in the opposite direction relative to EMeg32.
Together with the RACE and RT-PCR experiments performed, the Northern analysis confirms the integrity of the EMeg32 cDNA and argues for alternative polyadenylation rather than alternative splicing as the likely mechanism gen- erating the three major mRNA species (Fig. 1, B and C, and data not shown). Despite the differential expression of EMeg32 mRNA in the hematopoietic hierarchy, it is expressed in all tissues analyzed by Northern blotting. The 1.6-kb species is generally more abundant than the other two transcripts. Its expression in adult mice is highest in liver, kidney, and salivary glands; intermediate in intestine, thymus, and lung; but low in heart, spleen, brain, bone marrow, and testis ( Fig. 1C and data not shown).
In situ hybridization to sections of murine embryos at different stages of development revealed EMeg32 mRNA to be widely expressed at day 12.5 of embryonic development (E12.5; Fig.  2A). Beginning with E14.5, expression is more confined (data not shown). With onset of osteogenesis at E16.5, hybridization was strongest in vertebrae, calvaria, and craniofacial and long bones (Fig. 2B) and also in skin (Fig. 2D), whereas other cell types showed low expression. These observations were confirmed at the protein level by an immunofluorescence analysis using frozen sections of E16.5 embryos and EMeg32-specific antiserum (data not shown). In limbs of newborn mice, EMeg32 mRNA was clearly expressed in osteoblasts, chondrocytes, and hematopoietic cells located in the bone cavities, while expres-sion in skeletal muscle was undetectable (Fig. 2C).
In conclusion, EMeg32 is expressed widely at early stages of embryonic development but is confined to bones, skin, and the hematopoietic system at later developmental stages. Expression resumes in most tissues around birth and is maintained in adult tissues. However, the ubiquitous expression pattern obtained by Northern analysis may obstruct the fact that tissues comprise different cell types, often with varying expression levels for a given gene. A comprehensive in situ analysis on adult tissue sections will clarify whether EMeg32 is differentially expressed in different cell types as demonstrated for the adult hematopoietic system.
EMeg32 ORF and Protein Expression-Data base searches revealed proteins from Caenorhabditis elegans, Saccharomyces cerevisiae, Schizosaccharomyces pombe, and Candida albicans with significant homologies to EMeg32 (Fig. 3A). In addition, several murine and human expressed sequence tags matched parts of the EMeg32 cDNA (data not shown). While the C. elegans and S. pombe entries emerged from the respective genome projects without a demonstrated function, the S. cerevisiae and C. albicans proteins were recently identified to be glucosamine-6-phosphate acetyltransferase 1 (GNA1; Ref. 15). A closer inspection of the EMeg32 ORF revealed matches to known protein motifs (Fig. 3B); at its N-terminal half (aa 38 -115), EMeg32 exhibits homology to aa 280 -369 of annexin VII of Dictyostelium discoideum (26% identical, 47% similar), a Ca 2ϩ -dependent phospholipid-binding protein (23). The homology covers helices IID-IIID in the C-terminal tetrad repeat of annexin VII (Fig. 3C), but contains a gap at the position of a conserved Ca 2ϩ binding motif (24). Furthermore, EMeg32 harbors a putative leucine zipper between aa 141 and 155, which is well conserved in the five homologs (Fig. 3A). An acetyltransferase (AT) signature motif consisting of two domains (25) was identified between aa 119 and 138 (domain I, Fig. 3D) and between aa 161 and 170 (domain II, Fig. 3D).
Furthermore, based on multiple alignment and structural analysis, a consensus motif, (R/Q)-X-X-G-X-G, could be established for the whole GCN5-related N-acetyltransferase family of proteins (26 -28). The corresponding motif in EMeg32 is R-X-X-Q-X-G (aa 129 -134, Fig. 3D). Moreover, a similarly important conserved glycine in domain II is replaced by aspartate in EMeg32 (aa 169), and it was therefore of importance to demonstrate whether EMeg32 is able to bind acetyl-coenzyme A or derivatives. Since these changes distinguish EMeg32 also from GNA1, we asked whether EMeg32 is murine glucosamine-6-phosphate AT (see below).
Next, we affinity-purified anti-P1 and -P2 antisera and demonstrated that they detect proteins of 21 kDa in mouse, rat, monkey, and human cell lines (Fig. 4B). Since antiserum 2087 was about 20 times more sensitive in Western blot analysis than serum 497, we used it for detection of low EMeg32 protein levels, although it also recognizes a second protein of 18 kDa unrelated to EMeg32 (see below).
Functional Characterization of EMeg32 Protein-To address the question whether EMeg32 is indeed an AT, we conducted several tests. First, we demonstrated that purified baculovirusexpressed His-EMeg32 protein binds to CoA coupled to Sepharose beads, while it did not interact with beads conjugated to palmitoyl-CoA or glutathione. Endogenous EMeg32 protein from cellular extracts behaved identically (data not shown). The 18-kDa protein, which cross-reacts with antiserum 2087, is probably not an AT because it did not interact with either type of beads.
Next we analyzed whether EMeg32, due to its strong similarity to S. cerevisiae GNA1 (ScGNA1), could acetylate GlcN6P. For this assay, GST-EMeg32 protein was produced in bacteria in parallel to GST-ScGNA1 and GST only (Fig. 5A). In addition, baculovirus-expressed purified His-EMeg32 and a control protein, His-AIF, purified under identical conditions were employed (Fig. 5A). With 0.1 g of purified protein, 200 M acetyl-CoA and varying amounts of GlcN6P per reaction GST-ScGNA1 and His-EMeg32 exerted GlcN6P-AT activity, while the control proteins did not (Fig. 5, B and C). While several independently generated samples of purified His-EMeg32 always exerted GlcN6P-AT activity, GST-EMeg32 did not (Fig.  5B). However, GST-EMeg32 also bound to palmitoyl-CoA beads (data not shown), indicating that the bacterially expressed GST fusion protein may acquire a different conformation, which impairs its acetyl-CoA binding pocket and activity. Alternatively, posttranslational modifications are needed to generate functional EMeg32 protein, which were not exerted in procaryotic cells. We therefore used baculovirus-expressed His-EMeg32 for further studies on substrate specificity and kinetics. We also tested baculovirus-expressed EMeg32-HAHIS (corresponding to the fusion used in retrovirus expression; see below). It behaved identically to His-EMeg32 in the assays employed (data not shown). Fig. 5D shows that His-EMeg32 preferably uses GlcN6P as a substrate, since GlcN1P or GalN1P did not lead to release of CoA from acetyl-CoA in the assay used. The enzymatic activity of EMeg32 is robust and survives several freeze/thaw cycles or storage for up to 6 months at Ϫ70°C. We determined the average Michaelis-Menten K m value for GlcN6P as 40 ϫ 10 Ϫ6 M and the K m value for acetyl-CoA as 120 ϫ 10 Ϫ6 M. Both values are in the range described for this class of enzymes (15,28).
In summary, EMeg32 clearly exerts GlcN6P-but not GlcN1P-AT activity, which is in contrast to GST-ScGNA1 (15). In addition, its N-terminal half resembles a domain of a phospholipid-binding protein (annexin VII) not found in ScGNA1. For these reasons we refer to this murine (m) enzyme as EMeg32/mGlcN6P-AT.
Subcellular Localization of EMeg32-Since the synthesis of nucleotide sugars is believed to take place in the cytoplasm, we studied the subcellular localization of EMeg32/mGlcN6P-AT in NIH3T3, Rat-1, or COS-1 cells using indirect immunofluorescence and confocal microscopy. Remarkably similar perinuclear staining was obtained with all three cell lines, likely reflecting EMeg32 localization to Golgi apparatus (Fig. 6, A-C; data not shown). The EMeg32-specific signal was abolished by preincubation of the antiserum with baculovirus-expressed, purified His-EMeg32 (data not shown). As expected, preimmune serum did not generate a specific signal (data not shown). To confirm localization to Golgi, double immunofluorescence was performed using affinity-purified serum 479 and a monoclonal antibody against ␣-mannosidase II, a medial Golgi marker (29). As can be seen in Fig. 6C, signals for EMeg32 and ␣-mannosidase II clearly overlapped. However, EMeg32 protein was also expressed in vesicular structures different from Golgi.
To determine the nature of the additional compartment, we used a murine mammary epithelial cell line, EpH4, for which the localization of organelles in 10 -40% sucrose gradients has been established (17,30). The distribution of EMeg32 in the gradient was determined by immunoblotting using antisera against known organelle markers (Fig. 6D). In accord to the colocalization with ␣-mannosidase II in vivo, EMeg32 distribution overlaps with the peaks of nucleobindin (Nuc-1), a recently described Golgi-associated Ca 2ϩ -binding protein (31), and p58/ ERGIC53, a transmembrane protein in the ER to Golgi intermediate compartment (32). Transferrin receptor was used to show the distribution of the recycling compartment, early endosomes, and the plasma membrane at similar densities as Golgi-derived membranes. However, the second peak of EMeg32 in the upper region of the gradient (Fig. 6D, lanes   FIG. 4. Characterization  3-10) matches the distribution of Rab7 and other late endocytic markers (17) and suggests the association of EMeg32 to membranes of the late endosome/lysosome. Double immunofluorescence performed with NIH3T3 cells using affinity-purified 479 and a monoclonal antibody against murine LAMP-1 (lysosome-associated membrane protein 1; Ref. 33) revealed that only a small fraction (ϳ1%) of LAMP-1positive vesicles in the periphery of the cell were also EMeg32positive (data not shown). This indicates that the overlap with the Rab7 distribution in the sucrose gradient probably reflects mainly endosomal structures.
To test how EMeg32 is associated with the respective organelles we first performed sequential ultracentrifugation of EpH4 homogenates at 17,000 ϫ g and 100,000 ϫ g to separate Golgi cisternae from lighter membranes and cytosol, respectively. EMeg32 was found in both pellets and in the cytosolic supernatant (Fig. 6E). Furthermore, postnuclear supernatants were subjected to increasing salt concentrations, thereby stripping proteins from the cytoplasmic face of membranes into solution. To control for leakiness of organelles during this process, pellets and supernatants were analyzed for calreticulin, a soluble luminal protein. Further controls used were p58/ER-GIC53 and Rab5 (a protein anchored to the cytoplasmic leaflet of membranes via a lipid-modification), which should be retained in the membrane pellets, as well as ␤COP, which is expected to elute into the 0.3 M salt supernatants. Fig. 6E clearly shows that with increasing salt concentrations more EMeg32 protein is found in the supernatant, indicating that it is associated with the cytoplasmic face of organelle membranes. It is possible that the hydrophobic annexin VII-like domain aids in that respect. Localization of control proteins was as expected (Fig. 6E).
Cell Cycle Regulation of EMeg32-ScGNA1 was described previously as PAT1 (putative acetyltransferase 1; Ref. 34) which was not deposited in any of the relevant data bases. PAT1 mutant S. cerevisiae exhibited defects in cytokinesis, chromosome loss, and generation of aberrant spindles (34). Notably, the temperature-sensitive PAT1/ScGNA1 mutation pat1-3 can be complemented by mutated SAC1, an integral membrane protein of ER and Golgi, which was first identified as a suppresor for actin mutations in yeast (34,35). Certain yeast SAC1 mutants show growth retardation (36). It remains to be determined how the enzymatic activity of ScGNA1/PAT1 protein links to the multiple defects observed in cell cycle, and whether ScGNA1/PAT1 mutant cells exhibit a defect in glycosylation or GPI linker formation.
Toward understanding a possible role of EMeg32 in cell cycle regulation, we first tested whether its expression is controlled during cell cycle using the IL-3-dependent cell line FDC-P1 and centrifugal elutriation. DNA content of cells in the elutriated fractions was determined by fluorescence-activated cell sorting analysis and normalized protein extracts were analyzed by immunoblotting using EMeg32 antiserum 2087 or an antibody against GSK3 (glycogen synthase kinase 3) as a control. Surprisingly, EMeg32 protein levels were about 4-fold elevated in late G 2 /early M phase cells as compared with cells in the G 1 phase of cell cycle (Fig. 7, A-C). This elevation was observed in three independent experiments.
EMeg32 Copurifies with the cdc48 Homolog p97/VCP-In the course of purifying His-EMeg32 from baculovirus-infected SF9 cells (Fig. 8A), we repeatedly identified a 200-kDa protein, which bound to His-EMeg32 on nickel-chelate columns but not to His-AIF or other His-tagged proteins (Fig. 8B, data not shown). Upon storage, boiling, or one freeze/thaw cycle, this 200-kDa band was completely converted to a single 100-kDa form (Fig. 8C). The latter was analyzed by liquid chromatography and ion trap-tandem mass spectrometry. A total of 6 peptide spectra were identified (for sequences, see "Experimental Procedures"), which matched mouse VCP, a 97-kDa ATPase associated with diverse cellular activities (AAA), which readily forms soluble, ringlike hexameric complexes. Thus, the 200-kDa band originally observed most likely represents a dimer consisting of two p97 monomers. p97 represents the homolog of S. cerevisiae cdc48 (37) and has been implicated to function in homotypic membrane fusion events including fusion of ER membranes (38) and reassembly of Golgi cisternae from vesicles and tubules (39,40) at mitosis. p97/VCP is a protein highly conserved throughout evolution (e.g. one amino acid exchange between mouse and human p97), and we have preliminary evidence that epitope-tagged EMeg32 also forms a complex with mammalian p97/VCP (data not shown).
Given the membrane association of EMeg32, its accumulation prior to mitosis, and the finding that it copurifies with p97/VCP, it is tempting to assume that EMeg32 is involved in the regulation of membrane fusion events at mitosis.
To address this possibility, we tested whether retroviral overexpression of EMeg32 would affect the growth properties of e.g. FDC-P1 cells. G418-resistant clones were generated following infection of FDC-P1 cells with MSCV-EMeg32, MSCV-EMeg32/HAHIS, or MSCV-neo retroviruses, respectively (Fig.  9A). All clones analyzed (5 for MSCV-EMeg32, 5 for MSCV-EMeg32/HAHIS, 3 for MSCV-neo) exhibited similar growth properties independent of the level of exogenously expressed EMeg32(/HAHIS). Further, we did not observe a change in the rate of cell death occurring after IL-3 withdrawal (data not shown). However, maximum steady state RNA and protein levels achieved for ectopically expressed EMeg32 or EMeg32/ HAHIS were only equal to endogenous EMeg32 levels in FDC-P1 cells ( Fig. 9B; data not shown). Similar results were obtained following infection of NIH3T3 cells, which also express endogenous EMeg32 RNA and protein ( Fig. 9C; data not shown). An analysis of EMeg32 function in cell cycle therefore awaits the generation of cells with undetectable expression of this enzyme, preferably by genetic inactivation (see "Discussion"). . C, overlay of images in A and B. D, subcellular fractionation of EpH4 cells on a continuous sucrose gradient (10 -40%) followed by immunoblot analysis is shown. As references Golgi-associated nucleobindin (Nuc-1), ERGIC53 (p58, specific for ER/Golgi intermediate compartment), a late endosomal marker (Rab7), or transferrin receptor (TfnR) were used. Affinity-purified anti-P2 serum was employed; only p21 EMeg32 is shown (EMeg32). E, homogenized EpH4 cells were subjected to sequential ultracentrifugation. The 17,000 ϫ g (17K) and 100,000 ϫ g (100K) pellets as well as supernatant (C) were used. Postnuclear supernatant from EpH4 cells was extracted in 0.3 or 1.0 M NaCl or KCl. Addition of plain homogenization buffer (HB) is shown as control. Membranes were pelleted at 100,000 ϫ g (P) and analyzed by immunoblot together with extracted and cytosolic proteins (S). Antibodies against ␤COP, calreticulin, Nuc-1, ERGIC53 (p58), Rab5, and EMeg32 (as in D) were used as indicated.

DISCUSSION
We have identified and characterized a novel murine cDNA, which codes for a 184-aa protein exhibiting GlcN6P-AT activity. Using acetyl-CoA and GlcN6P as substrates, the enzyme generates N-acetylglucosamine 6-phosphate, which, following isomerization to N-acetylglucosamine 1-phosphate, is converted to UDP-GlcNAc. We demonstrate for the first time for this class of enzyme association with Golgi and other intracellular membranes, regulated expression during cell cycle, and copurification with p97/VCP, a protein implicated in mitotic membrane fusion events.
EMeg32/mGlcN6P-AT was cloned as a differentially expressed entity using our single cell PCR-based subtractive hybridization approach. Absence of EMeg32 protein expression in mature murine erythrocytes and its presence in mature macrophages established from long term bone marrow cultures (data not shown) corroborates, at least on the maturing cell side, the specificity of EMeg32 protein expression originally observed on the RNA level (Fig. 1A). The retrovirus system described above can now be used to assess whether constitutive expression of EMeg32 in bone marrow cells would influence the differentiation potential or growth properties of hematopoietic precursors. Such experiments are in progress.
The fact that EMeg32 is differentially regulated in the hematopoietic system and shows differential expression during embryogenesis argues for the dispensability of the enzyme's function at least at some cell and developmental stages. The successful generation of EMeg32 knock-out embryonic stem cells and fibroblasts 2 further strengthens this notion. This aspect is unanticipated, because inactivation of the yeast glucosamine-6-phosphate acetyltransferase GNA1/PAT1 is lethal in S. cerevisiae (15,34), which might indicate that the pathway for synthesis of N-acetylglucosamine 6-phosphate is buffered by redundancy in mammalian cells or that the two proteins exert additional, non-overlapping functions.
Biochemical Features of EMeg32/mGlcN6P-AT-Glucosamine-6-phosphate AT (EC 2.3.1.4) has been purified previously from several sources (41)(42)(43)(44). In rat liver, three different forms 2 G. Boehmelt and T. W. Mak, manuscript in preparation.  (1-7) were lysed, normalized to total protein content, and analyzed using ␣-P2 serum or a monoclonal antibody against GSK3. Only p21 EMeg32 (EMeg32) and the GSK3-␣ (GSK3) form are shown. B, the intensities of the signals obtained from the immunoblotting analysis in A were quantified using a densitometer. Relative protein levels of EMeg32 normalized to GSK3 in each fraction (1-7) are plotted. C, 50% of elutriated fractions were fixed and analyzed by fluorescence-activated cell sorting following RNase treatment and propidium iodide incorporation. The percentage of cells in the G 1 , S, and G 2 /M phase of the cell cycle was determined for each fraction (1-7) and plotted. of GlcN6P-AT were described, which eluted from a Sephadex G-75 column with the apparent molecular mass of 23,000 Ϯ 2000 (41). This is in perfect agreement with the molecular mass of EMeg32 (21,000 Da) and its rat homolog on SDS-PAGE gels (Fig. 4B).
It had been repeatedly suggested that sulfhydryl groups are present at or near the active site of the enzyme (41,45). EMeg32 harbors 6 cysteine residues, one of which (aa 128) lies in its AT signature domain I (aa 119 -138), but is not conserved in the homologs aligned (Fig. 3A). The only consistently conserved cysteine is in the spacer region (aa 157). Mutational analysis is now possible to identify which cysteine is crucial for activity.
Using recombinant purified histidine-tagged EMeg32 expressed in baculovirus, we determined the Michaelis-Menten K m value for GlcN6P to be 40 ϫ 10 Ϫ6 M, which is comparable to the one described for rat GlcN6P-AT form III (70 ϫ 10 Ϫ6 M; Ref. 41). The K m value of ScGNA1 for GlcN6P was found to be 124 ϫ 10 Ϫ6 M at pH 7.5 (15), whereas we measured it identical to the average EMeg32 value (Fig. 5B; data not shown) at pH 8.0. Since the pH optimum for the purified GlcN6P-ATs lies in the alkaline range (41,43,44), it is conceivable that this slight difference can be ascribed to the different reaction conditions used.
Concerning substrate specificity, we could not detect activity of EMeg32/mGlcN6P-AT toward GalN1P or GlcN1P, the latter of which can be used by ScGNA1 at high (1 mM) substrate concentrations. Similarly, Oikawa and Akamatsu (41) reported no activity of purified rat liver GlcN6P-AT for GlcN1P acetylation. In E. coli, UDP-GlcNAc is synthesized by a bifunctional enzyme, GlmU, having both GlcN1P-AT and UDP-GlcNAc pyrophosphorylase activity (46). In yeast and in higher eukaryotes, these functions are separated and amended by introduction of GlcN6P-AT into the pathway. However, N-acetyl-GlcN1P remains the only substrate for the respective UDP-GlcNAc-pyrophosphorylase. Thus, ScGNA1, being the evolutionary "older" version, may still maintain part of the prokaryotic substrate specificity that is lost in higher eukaryotes.
Localization of EMeg32 and Cell Cycle Regulation-The attachment of EMeg32 to the cytoplasmic membrane leaflet of Golgi and other intracellular membranes (Fig. 6) is important for several reasons: First, the final steps of nucleotide sugar synthesis are believed to take place in the cytoplasm, and proteins have been identified that transport the respective end products through the membrane into the lumen of ER and Golgi (47), where they are used in further glycosylation steps. Release of N-acetyl-GlcN6P by EMeg32 will take place into the cytoplasm, and no further factors are needed to make this compound available for the last two steps toward UDP-GlcNAc synthesis.
Second, the membrane association of EMeg32 and its accumulation prior to mitosis might be relevant in terms of the Golgi/ER transmembrane localization of SAC1, which has been identified in a suppressor screen to complement the cell cycle defects in ScGNA1/PAT1 mutants. In yeast, SAC1 functions as a polyphosphoinositide phosphatase, which is able to convert the phosphatidylinositol (PtdIns) phosphates PtdIns (4)P, Pt-dIns (3)P, and PtdIns (3, 5)P 2 into phosphatidylinositol (48). Actin assembly, intracellular vesicle trafficking, and budding processes as well as the secretory pathway from Golgi are dependent on some of the phosphatidylinositides, which serve as a substrate for the SAC1 lipid phosphatase (49,50). Further, the end product of the enzymatic reaction performed by SAC1 is phosphatidylinositol, which could be used to generate, together with UDP-GlcNAc, novel GPI linkers (7). Both GPI linker intermediates and extracellularly GPI-linked receptors may be important for signaling events occurring just prior to mitosis (51-53) and could provide a reason for the increase of EMeg32 protein levels at that cell cycle phase. It will be of considerable interest to assess how EMeg32 feeds into the SAC1 regulatory pathways, whether it directly interacts with the SAC1 homologue, and whether intracellular pools of phosphatidylinositols or GPI moieties are altered in response.
Third, we demonstrated that EMeg32 copurifies with p97/ VCP. This AAA-type hexameric ATPase is highly conserved throughout evolution and is implicated to function not only in the endocytic cycle based on its binding to clathrin (54), but also in the reassembly of Golgi cisternae from vesicles generated either by treatment with specific drugs like ilimaquinone (40) or by treatment with mitotic cytosol (39). p97/VCP resembles in its function N-ethylmaleimide-sensitive factor (NSF), another member of the AAA-type ATPase family. Vesicle fusion is thought to be mediated by the binding of v-SNAREs (soluble N-ethylmaleimide-sensitive factor attachment protein receptor) on the vesicle to t-SNAREs on the target membrane. The formation of the SNARE pairs is believed to bring the two membranes together and catalyze the fusion reaction, which is at the heart of vesicle-mediated cellular transport processes. The SNARE pairs, now residing in the same membrane, are possibly broken by the ATP-hydrolyzing action of NSF. p97/ VCP/cdc48 might function similarly, albeit its targets are pairs of t-SNAREs mediating homotypic membrane fusions as necessary, for example, for organelle biogenesis (55). ATPase-spe- FIG. 9. Retroviral expression of EMeg32. A, retroviral constructs used. LTR, long terminal repeats; gray box, EMeg32 ORF; black box, HAHIS tag. The phosphoglycerate kinase promotor (PGK), the neomycin resistance cassette (neo), splice donors (SD), and acceptors (SA), respectively, as well as the packaging signal (⌿ϩ) are indicated. Black triangle, polylinker (PL); large arrows depict transcriptional start sites. B, Western blot analysis of FDC-P1 clones stably expressing MSCV-neo (neo-6C) or MSCV-EMeg32/HAHIS (clones EHH-11D, EHH-15C, EHH-19C) retrovirus. Blots were incubated with anti-HA antibody 12CA5 (anti-HA) or EMeg32 antiserum 2087 (anti-P2). p26 EMeg32/HAHIS and p21 EMeg32 are indicated by arrows and arrowhead, respectively. C, Northern blot analysis on RNA from NIH3T3 clones expressing MSCVneo (neo-2), MSCV-EMeg32 (E-3.3), or MSCV-EMeg32/HAHIS (EHH-3.1, EHH-4.4), respectively. The blot was hybridized with probe A (see Fig. 1). The positions of endogenous (arrowheads) and retrovirus-driven (arrows) EMeg32-specific transcripts are indicated. M, molecular mass standards in kb. cific cofactors mediate the attachment of NSF or p97/VCP to the respective SNAREs (56 -58). One way to explain regulated Golgi fragmentation and reassembly at mitosis might be via the dominance of the p97/VCP-dependent organelle biogenesis pathway over the NSF-dependent membrane traffic pathway (for reviews, see Refs. 59 -61).
EMeg32, a protein associated with Golgi and endosomal membranes and able to form a complex with p97/VCP, could exert a regulatory or a scaffolding function for this important ATPase. Further studies will clarify whether EMeg32 always interacts with cytoplasmic p97/VCP (also see Fig. 8B), whether it affects the enzymatic function of p97/VCP (or vice versa), and what the cell biological response of such regulation might be. The possible involvement of EMeg32 in mitotic membrane fusion events, either via the genetic link to the SAC1 phosphatidylinositol phosphatase or via the interaction with p97/VCP, renders this protein an intriguing entity whose detailed functional analysis will profit from the availability of mammalian cells that have the gene or its product inactivated. The successful generation of embryonic stem cells and fibroblasts that have the EMeg32 gene inactivated by homologous recombination 2 now provides an important tool to address these questions.