Proteomics Characterization of Abundant Golgi Membrane Proteins*

A mass spectrometric analysis of proteins partitioning into Triton X-114 from purified hepatic Golgi apparatus (84% purity by morphometry, 122-fold enrichment over the homogenate for the Golgi marker galactosyl transferase) led to the unambiguous identification of 81 proteins including a novel Golgi-associated protein of 34 kDa (GPP34). The membrane protein complement was resolved by SDS-polyacrylamide gel electrophoresis and subjected to a hierarchical approach using delayed extraction matrix-assisted laser desorption ionization mass spectrometry characterization by peptide mass fingerprinting, tandem mass spectrometry to generate sequence tags, and Edman sequencing of proteins. Major membrane proteins corresponded to known Golgi residents, a Golgi lectin, anterograde cargo, and an abundance of trafficking proteins including KDEL receptors, p24 family members, SNAREs, Rabs, a single ARF-guanine nucleotide exchange factor, and two SCAMPs. Analytical fractionation and gold immunolabeling of proteins in the purified Golgi fraction were used to assess the intra-Golgi and total cellular distribution of GPP34, two SNAREs, SCAMPs, and the trafficking proteins GBF1, BAP31, and α2P24 identified by the proteomics approach as well as the endoplasmic reticulum contaminant calnexin. Although GPP34 has never previously been identified as a protein, the localization of GPP34 to the Golgi complex, the conservation of GPP34 from yeast to humans, and the cytosolically exposed location of GPP34 predict a role for a novel coat protein in Golgi trafficking.

The molecular mechanisms by which newly synthesized secretory cargo is transported across the Golgi complex have remained elusive since the discovery of the organelle (1). A cell-free transport assay designed to reconstitute transport of secretory cargo through the organelle has been instrumental in uncovering a dynamic coat complex (via ARF/COPI coatomer) (2) and a membrane fusion machinery (via NSF/SNAPs/ SNAREs) (3). These constituents have been proposed to regulate cargo selection, membrane budding, vesicular formation, and membrane fusion to effect the anterograde delivery of newly synthesized secretory proteins between adjacent Golgiflattened cisternae (4). Although recent studies have challenged this view (5)(6)(7), the relevance of the above molecular machinery to the regulation of membrane trafficking in the early secretory pathway remains unquestioned. Parallel progress in understanding membrane traffic in the synaptic terminal has been largely a consequence of a systematic analysis of the major proteins identified therein largely as a consequence of the cloning of their cDNAs. This has led to considerable insight into the mechanisms of membrane targeting, fusion, budding, and vesiculation at the synaptic terminal with several of the same proteins that were identified via the Golgi cell-free transport assay (8 -10).
Recent innovations in the use of mass spectrometry to characterize proteins enable a rapid assignment of major proteins to cellular structures, bypassing the requirement for protein characterization via cDNA cloning. Hence, mass spectrometrybased protein assignments to centrosomes (11) and the nuclear pore (12) have led to new views on the function of these structures.
Edman degradation has been used previously to identify low molecular weight integral membrane proteins of a highly purified hepatic Golgi fraction. This led to the uncovering of four distinct members of the p24 family of integral membrane proteins located largely in the cis Golgi network and probably forming an intermolecular complex (13). This approach has now been extended to the use of mass spectrometry complemented by Edman degradation to identify all bands visualized by one-dimensional SDS-PAGE 1 of Golgi proteins partitioning into Triton X-114 and therefore expected to be membrane proteins (14). The approach unambiguously identified 81 proteins. A combination of electron microscope (EM) immunolabeling and analytical centrifugation was used to visualize the subcellular distribution of selected proteins identified by mass spec-trometry and Edman sequencing. Most abundant were Golgi resident enzymes and membrane trafficking proteins including Rabs and SNAREs. A novel peripheral membrane protein of 34 kDa was uncovered by this approach and localized to the isolated Golgi apparatus by cryoimmune labeling and to whole cells by confocal immunofluorescence. By systematically defining the protein composition or proteome of each organelle of the eukaryotic cell, a functional protein map of the cell may be realized.

EXPERIMENTAL PROCEDURES
Isolation of Golgi Fractions-The WNG fraction was isolated exactly as described by Dominguez et al. (7) and characterized for protein yield and galactosyl transferase marker enzyme enrichment also as described by Dominguez et al. (7). Galactosyl transferase relative specific activity was 122 Ϯ 32-fold (n ϭ 3) enrichment over the homogenate. The protein in the fraction corresponded to 0.043 Ϯ 0.018% of the original homogenate protein.
SDS-PAGE, Triton X-114 Partitioning, and N-Terminal Edman Sequencing of Golgi Membrane Proteins-Triton X-114 phase partitioning of Golgi integral membrane proteins was by the method of Bordier (14), as described by Dominguez et al. (13). For N-terminal Edman sequencing, phase-partitioned Golgi integral membrane proteins were separated by SDS-PAGE, electrotransferred onto polyvinylidene difluoride, and stained with Coomassie Brilliant Blue (15). Bands were excised from the polyvinylidene difluoride blot and washed with 20% methanol prior to automated microsequencing (16) on an ABI 476A protein sequencer (PE Biosystems), employing a blot cartridge and standard pulsed liquid blot protocol (17). Data were collected and analyzed both by ABI 610A software and by manually overlaying successive traces, and the resulting sequences were searched against a nonredundant protein data base using FASTA (18), blastp (19), and ProteinProspector MS-Edman programs. Signal peptides and the number of transmembrane domains (four different algorithms) were predicted according to algorithms available through the ExPASy Molecular Biology Server. The algorithms were SignalP, HMMTOP, SOSUI, TMHMM, and TMpred.
In-gel Enzymatic Digestion-Protein bands were excised from a Coomassie Blue-stained gel and destained prior to digestion according to the published procedure by Shevchenko et al. (20). The destaining was achieved by rinsing the excised gel pieces in 50% acetonitrile in water (2 ϫ 100 l). Each rinse involved briefly vortexing the sample prior to shaking at 37°C for 5 min before pipetting off the acetonitrile and adding 100 mM ammonium bicarbonate (50 l). The sample was vortexed again and shaken at 37°C for 15 min before the addition of acetonitrile (50 l). The completely destained gel was then subjected to dithiothreitol (1 mg/ml) to reduce cystinyl residues and then to iodoacetamide (10 mg/ml) to effect alkylation. These reagents were removed, and acetonitrile was added to re-shrink the gel pieces. The acetonitrile was removed, and gel pieces were dried in a vacuum centrifuge. The dried gel pieces were re-swollen in digestion buffer (50 mM ammonium bicarbonate (pH 8.5), 5 mM CaCl 2 ) containing 125 ng/10 l trypsin (Roche Molecular Biochemicals, sequencing grade). Digestion was allowed to proceed overnight at 37°C.

Peptide Analysis by Delayed Extraction Matrix-assisted Laser Desorption Ionization (MALDI) Mass Spectrometry (MS)-
The supernatant liquid from the digest was sampled directly as described by Jensen et al. (21). An aliquot (0.4 l) of the supernatant was added to an equal volume of 5% formic acid previously spotted onto a stainless steel target precoated with matrix. The matrix solution was prepared by mixing a saturated solution of cyano-4-hydroxycinnamic acid in acetone in a 4:1 ratio with a solution of nitrocellulose at 10 mg/ml in acetone/propan-2-ol (1:1, v/v). The target was allowed to air dry before being washed with 1% aqueous trifluoroacetic acid (2 l). Excess wash solution was blown off, and the target was dried using compressed air. MALDI mass spectra were obtained using a ToFSpec instrument (Micromass, Manchester, UK) fitted with a 337-nm nitrogen laser. Spectra were acquired using the instrument in reflectron mode and calibrated using a standard peptide mixture.
Peptide Analysis by Nanoelectrospray Ionization MS-In cases where the identity of the proteins present could not be established by MALDI analysis alone, the peptides were further analyzed by nanoelectrospray ionization MS/MS. Here the peptides were extracted from the gel pieces using 100 mM ammonium bicarbonate (2 ϫ 50 l) and acetonitrile (2 ϫ 50 l) followed by 5% formic acid in 50% methanol (2 ϫ 50 l). The samples were then centrifuged before removing the liquid into labeled tubes. The combined extracts for each sample were then dried using a vacuum centrifuge. Dried protein digests were redissolved in 5% formic acid containing 5% methanol (10 l) and desalted using a pulled-out glass capillary containing a small amount of POROS R2 resin (PerSeptive Biosystems). The desalting columns were prepared by slurry-packing a minute amount of resin into the capillary. The columns were equilibrated using 5% formic acid in 5% methanol (2 ϫ 5 l). After loading the sample, the columns were washed using 5% formic acid in 5% methanol (2 ϫ 5 l). The peptides were then eluted directly into the nanospray needle using 5% formic acid in 50% methanol (5 ϫ 0.25 l). The amount of packing material required for this application is far less than that contained in commercially available pre-packed columns, allowing the columns to be employed as single use disposable items and avoiding the risk of sample loss and contamination between samples. A Q-Tof TM hybrid mass spectrometer (Micromass) fitted with a Z-spray source was used to acquire the mass spectra. In MS/MS mode the quadrupole is used to select the precursor ion, which is then passed into a collision cell where fragmentation is induced by collision with argon gas molecules. The energy of collision is typically between 30 and 60 eV, depending on the mass and charge of the precursor ion. Ions formed by the cleavage of backbone bonds are designated a, b, c if the charge is retained on the N-terminal fragment and x, y, z if the charge resides on the C-terminal fragment (nomenclature according to Mann et al. (22)). Product ions higher in m/z value than a doubly or triply charged precursor ion are often part of a series of yЉ ions (23). In many cases it is possible to read part of the sequence from the pattern of yЉ ions. This, together with the masses bracketing the sequence, forms a Peptide Sequence Tag (24) that can be used to identify the peptide. Ions in the low mass region of the MS/MS spectrum typically include b series ions as well as other internal fragments. The latter are less specific but characteristic of the sequence (25).
Antibodies to GPP34 -Rabbit anti-peptide antibodies were raised against a peptide sequence (LKDREGYTSFWNDC; see Fig. 3B) derived from the human GPP34 sequence. The peptide was coupled via the C-terminal cysteine residue to keyhole limpet hemacyanin using mmaleimidobenzoyl-N-hydroxysuccinimide ester as the cross-linker. An initial subcutaneous injection of an emulsion of peptide in complete Freund's adjuvant was followed by four boosts of peptide (100 mg each) in incomplete adjuvant. Total IgG was purified by protein A-Sepharose affinity chromatography (26). Rabbit antiserum was also raised to a bacterial recombinant GST chimera of the human GPP34. Polymerase chain reaction products encoding GPP34 cDNA (amplified from IMAGE clone 664740, GB number AA232616, Research Genetics) were purified by agarose gel electrophoresis, ligated into pGEX2T (Amersham Pharmacia Biotech) and pTrcHisA (Invitrogen) using EcoRI/BamHI as restriction sites, and transformed into Escherichia coli DH 5␣. Positive clones were identified by polymerase chain reaction, and expression of the fusion protein was confirmed by SDS-PAGE and Coomassie R-250 staining. His-tagged GPP34 and GST-GPP34 were purified from isopropyl-1-thio-␤-D-galactopyranoside-induced bacterial cultures and affinity-purified according to the manufacturer's instructions on NiNTA resin (Qiagen) or glutathione-Sepharose beads (Amersham Pharmacia Biotech), respectively. Antibodies were raised as above, employing complete Freund's adjuvant. Anti-GST-GPP34 antibodies were further purified by affinity chromatography employing His-tagged GPP34 bound to Sepharose 4B (Amersham Pharmacia Biotech), as described by Harlow and Lane (26).
EM Immunolocalization and Morphometry-Freshly prepared Golgi fractions were incubated with primary antibodies and processed for EM immunolocalization as described by Lavoie et al. (27). In the case of GPP34, cryosections of Golgi fractions were prepared using the protocol description in Lavoie et al. (27) and incubated with the anti-peptide antibodies to GPP34.
Confocal Laser Scanning Microscopy of Anti-GPP34 -GPP34 antigenicity was visualized in rat FR3T3 fibroblasts and in primary cultures of rat hippocampal neurons. For neurons, the CA3 region of PO-P1 rat hippocampi was dissected out, and the neurons were cultured for 8 days as described in Baranes et al. (28). For immunocytochemistry, neurons were fixed for 10 min with 4% paraformaldehyde, 4% sucrose and incubated with affinity-purified rabbit antibody to recombinant GPP-34 (1:10) and mouse anti-MG160 (1:100; a kind gift of Drs. A. Beaudet, McGill University, and N. Gonatas, University of Pennsylvania). Neurons were then labeled with goat anti-rabbit Cy3-conjugated (1:300) and goat anti-mouse fluorescein isothiocyanate-conjugated (1:200) secondary antibodies (Jackson ImmunoResearch Laboratories Inc.). Optical sections were obtained using an LSM 410 confocal microscope (Zeiss), and images were taken under nonsaturating conditions.
Analytical Fractionation of Total Membranes-Livers from rats fasted overnight were minced and homogenized in 0.25 M sucrose, 4 mM imidazole buffer (pH 7.4) (homogenization buffer) with a motor-driven Potter-Elvejhem homogenizer with five strokes. The homogenate was filtered through two layers of cheese cloth, and the volume was adjusted (with homogenization buffer) to 20% (w/v) of the starting liver wet weight. The homogenate was centrifuged at 700 ϫ g max for 10 min. The resulting supernatant was centrifuged at 200,000 ϫ g max for 40 min. The pellet (combined Mitochondrial, Light mitochondrial, Particulate fraction (MLP)) was resuspended in 0.25 M sucrose imidazole buffer to 1 g/ml original liver weight, and 0.5 ml of this sample was layered on top of a continuous 0.5-2.3 M sucrose gradient buffered in imidazole (pH 7.4) (total volume 12.2 ml) and centrifuged at 110,000 ϫ g max for 17 h. Sixteen fractions were collected from the top, with the weight and density measured in an Abbe Mark II refractometer. Galactosyl transferase was assayed using [ 3 H]UDP-galactose and ovomucoid as substrate, as described previously (29). Protein was estimated by the Bradford method (30). Proteins of each fraction were precipitated with trichloroacetic acid, washed with 70% ethanol, and then resuspended in 5 mM Tris-HCl (pH 8) with the aid of a sonicator on ice in preparation for SDS-PAGE. Equal volumes of Laemmli sample preparation buffer (31) were added, and the sample was heated in a boiling water bath for 5 min. The resulting solution was clarified by centrifugation in a microcentrifuge. For Western blotting, proteins in each fraction were separated on a 10% gel by SDS-PAGE, and transferred to nitrocellulose electrophoretically for 75 min at 12 V, employing a Genie electrotransfer unit (IDEA Scientific Company, Minneapolis, MN). The blots were blocked with 5% skim milk in 10 mM Tris, 150 mM NaCl, 0.5% Tween buffer (pH 7.5) (32) and then incubated at 4°C overnight with primary antibodies: 1:1000 for protein A-purified anti-GPP34; 1:500 antibody dilution for SCAMP1 and SCAMP3 (kind gifts of Dr. D. Castle, University of Virginia); 1:1000 dilution for ␣ 2 p24, calnexin (CNX), and ribophorin II rabbit polyclonal antibodies; 1:500 for the monoclonal antibody to GS28; and 1:1000 for the chicken antibody to BAP31 (a kind gift of Dr. G. C. Shore, McGill University). After each antibody reaction, the blots were washed with 1% skim milk in 10 mM Tris, 150 mM NaCl, 0.5% Tween buffer (pH 7.5). In the case of GS28 monoclonal antibody or BAP31 chicken antibody, the washed blots were incubated with rabbit anti-mouse or rabbit anti-chicken antibodies for 1 h at room temperature. For visualization and quantitation, washed blots were incubated with a mixture of 125 I-labeled goat anti-rabbit (2 Ci per blot) and 1:5000 dilution of goat anti-rabbit alkaline phosphatase, developed with alkaline phosphatase reagent (32), and then exposed to x-ray film; the respective bands were excised from each lane of the blot, and the radioactivity was measured by a gamma counter. Distribution and frequency were calculated as described in Dominguez et al. (13) using the methodology according to Beaufay et al. (33).
For some experiments the peroxidase chemiluminescent system (34), as described by PerkinElmer Life Sciences, was used for visualization and quantitation. These experiments included GPP34 for Fig. 7A and mannosidase II, ␣ 2 p24, calnexin, and ribophorin II for Fig. 7B. For quantitation by chemiluminescence, the Bio-Rad GS-710 densitometer linked to the Multianalyst program was used. Protein A-peroxidase and goat anti-mouse peroxidase were employed for detection of mannosidase II, ␣ 2 p24, and CNX and ribophorin II, respectively, after reaction with primary antibodies (1:1000).

Identification of Integral Membrane Proteins by Gas Phase
Sequencing and Mass Spectrometry-The Golgi fraction was selected on the basis of its prior characterization, including a high enrichment in the marker enzyme, galactosyl transferase, and diminution in endosomal contamination (7). The WNG fraction was also characterized by morphometry. Using a random sampling methodology (35), electron microscopy of the filtered Golgi fractions revealed, as deduced from the point hit method (36), that 84% of the profiles were Golgi complexes. The remaining profiles were endoplasmic reticulum or plasma membrane (14%), mitochondria (1%), or peroxisomal cores (1%), as based on the analysis of 100 micrographs (ϫ 27,000 final magnification) and a point hit methodology employing 42 points per grid (per micrograph).
N-terminal Edman sequencing ( Fig. 1, Table I) of integral membrane proteins that partitioned into Triton X-114 identified 18 proteins unambiguously, of which 11 were clearly Golgi resident or trafficking membrane proteins. Readily identifiable were the ER to Golgi trafficking proteins p58 (ERGIC 53), four members of the p24 family, two ERD2-like proteins (i.e. Elps or KDEL receptors), mannosidase II, a GalNAc transferase, a sialyl transferase, and a nonconventional GST (37). Three cargo proteins were found (paraxonase, apoE, and apoC III). Two putative endosomal proteins (p76 and EMP70) and two novel proteins (25 DX and GS3786) were found.
In contrast to N-terminal Edman sequencing, mass spectrometry unambiguously identified 72 proteins (Fig. 1, Table I). As shown in Table I, these were categorized into Golgi resident proteins, trafficking proteins, contaminant (microsomal, mitochondrial, peroxisomal, endosome, or plasmalemma) proteins, and cargo proteins. No proteins were deleted from the reported data either for analysis by Edman sequencing or mass spectrometry.
Data analysis was effected as exemplified for band 16 (see Fig. 2, A-C, and Table II) of the gel used for MS studies (Fig. 1, right side). The major component of this band was the Rab6 protein, based on the tryptic peptide mass map. The Protein-Prospector MS-Fit search algorithm for the m/z values of the masses of the peptide fragments confirmed that the masses shown in Fig. 2A indeed corresponded to Rab6, as deduced from the 26 peptide masses listed in Table II. As seen in Fig. 2B, these peptides identified in the complete sequence covered 75% of the total sequence. However, only the human Rab6 complete sequence is in the data base, whereas the starting material is from rat liver. Only 26 of the 85 peptide masses identified for band 16 corresponded to human Rab6. Removing those 26 and searching the data base with lower stringency, 6 more peptides were assigned to Rab6, and peptide masses unique to Rab1a were identified. Some peptide masses for Rab1a coincided with masses of peptides for Rab6, confirming the similarity of predicted tryptic peptide fragments. MS/MS sequence tag data (Fig. 2C) for the doubly charged precursor at m/z 566.7, which relates to a peptide of mono-isotopic molecular mass 1131.4 Da ((M ϩ H) ϩ 1132.631, Fig. 2A) confirmed sequence variation between the rat Rab6 protein and the DNA-predicted human Rab6 (Fig. 2B) and identified post-translational modifications of the mature rat Rab6 protein. Fig. 2C shows the rat Rab6 N-terminal sequence, identifying that the initiation methionine was removed and that the protein is N-terminally acetylated. This is the first evidence for N-acetylation of Rab proteins, although the consensus motifs for removal of the N-terminal methionine and N-acetylation at the Ser residue for Rab6 has been predicted (38). Screening the mouse EST data base (data not shown) confirmed the substitution of Ala at position 2 of the mature Rab6 protein for mouse as compared with the human sequence (Fig. 2, B and C). For all other proteins identified by tryptic peptide mass mapping (Fig.  1, Table I), total coverage was between 15 and 60%.
GPP34 -One protein of previously unknown function was identified as a sequence conserved from yeast to humans (Fig.  3). One sequence tag (Fig. 3A) identified a tryptic fragment (shown in italics in Fig. 3B) that was found in the EST data base. Further searches led to the alignment of the sequences shown in Fig. 3B. These corresponded to two human gene products, two mouse gene products, and single gene products in Drosophila melanogaster, Caenorhabditis elegans, and Saccharomyces cerevisiae. Not shown are partial sequences matched to the data base of Schizosaccharomyces pombe, Kluveromyces lactis, Aspergillus nidulans, Danio rerio, and Xenopus laevis. Even though the sequence has not been cDNA-cloned in any known species, the corresponding gene (YDR372c) (Yeast Proteome Data Base) (39) has been deleted from S. cerevisiae as part of the Saccharomyces Deletion Project (40). The gene is not essential for viability, with no effect from gene deletion on growth (Saccharomyces Deletion Project strain reference numbers: 4208, 14208, 24208, 34208) (40).
The absence of a signal sequence and transmembrane domain predicts that GPP34 is a peripheral membrane protein.
Other peripheral proteins partitioning into the membrane phase were GM130, p115, actin, myosin, and ankyrin. GM130 and p115 are clearly , cytoplasmically oriented, and bound to GRASP65 or GRASP55. GRASP65 and GRASP55 are cytosolic proteins postulated to link adjacent Golgi cisternae (42, 43) that were not detected in our analysis. The presence of ankyrin in the fraction is expected (44), but its partitioning into Triton X-114 implies tight association with an integral membrane protein. In a similar way, the detection of myosin and actin may be relevant to Golgi function. An unconventional GST with three predicted transmembrane domains was identified (Table I). This microsomal class of GSTs is unrelated by sequence to the conventional GSTs, as deduced by blastp analysis.
Golgi resident proteins (Table I) including the Golgi marker MG160 (of as yet unknown Golgi function but postulated to be a fibroblast growth factor receptor (45)) and the processing enzyme mannosidase II, also frequently utilized as a Golgi marker, were prominent. Other N-and O-linked glycosyl processing enzymes were found as expected, as was the Golgi-or cis Golgi-located lectin VIP36 (46,47).
Trafficking proteins were identified. A 200-kDa protein with a sec7 domain was identified. This protein is the rat orthologue of a guanine nucleotide exchange factor (GEF) for ARFs termed GBF1 and related to products of the yeast genes GEA1 and GEA2 encoding Sec7 domains (48 -50). The Sec7 domain of this class of proteins effects GDP-GTP exchange and when associated with ARF-GDP can be targeted by the drug brefeldin A (48). However, the sequence-related protein p200ARF-GEF previously identified (51) and recently renamed BIG1 (52) was not identified. In addition, only three Golgi v-SNAREs (GS15, GS28, and Sec22b) were found. The Rab family GTPases were FIG. 1. Triton X-114 phase partitioning of the Golgi fraction. N-terminal Edman sequencing and summary of mass spectral characterization for identification of Golgi resident, trafficking, contaminant, cargo, and novel proteins partitioning into the Triton X-114 phase from the parent Golgi fraction. On the left are indicated the proteins identified by N-terminal Edman degradation and the assignment to bands from a Coomassie Blue-stained polyvinylidene difluoride membrane blotted from an SDS-PAGE gel (10% acrylamide in the resolving gel). On the right, integral membrane proteins were electrophoresed in a 5-15% polyacrylamide gradient gel, and Coomassie Blue-stained bands were characterized by mass spectrometry. The numbers refer to band numbers identified as prominent Coomassie Blue-stained polypeptides in separate gels (left, Edman degradation; right, mass spectrometry). The mobilities of molecular weight markers are indicated on the right of the stained polyvinylidene difluoride membrane used for Edman sequencing and on the left of the stained gel used for mass spectrometry.

TABLE I
Characterization of WNG fraction proteins partitioning into Triton X-114 Molecular masses were derived from one-dimensional SDS-PAGE (Fig. 1). Bands were assigned by either MS or Edman sequencing (Edman) and categorized into Golgi resident, trafficking, contaminant (including cargo), or novel. Indicated is the band number from Figure 1 for each assigned protein. Also indicated are the accession numbers for the sequences identified and the protein names (rat sequences if not indicated otherwise). Predictions of signal (signal peptide (SP), signal anchor (SA), targeting peptide (TP)) and transmembrane domains (TMD) are also shown, as well as the predicted number of transmembrane domains as described under "Experimental Procedures." For the latter, four different programs were used. A single number indicates consensus among the programs; more than one number indicates the different predictions. h, human; m, mouse; r, rabbit; b, bovine; d, dog; mic, microsomal; mit, mitochondrial; per, peroxisomal; end, endosome; PM, plasmalemma. The question mark for "similar to EMPp70" suggested a sequencing error in the data base, with a signal peptide expected but not predicted. also observed, of which Rab6 has been clearly Golgi-localized (53,54) whereas Rab5 is endosomal (55). Rabs 1a, 1b, and 2 (56 -60) are involved in ER to Golgi transport, and Rab7 is late-endosomal (61). Rab8b is involved in TGN trafficking events (62). Rab10 has been considered Golgi-located (63), and Rab13 may be in the plasma membrane (64). In addition, p76 and the protein designated "similar to EMP70" were found. An N-terminal fragment of the EMP70 protein was originally identified in endosome fractions isolated from yeast and called p24a (65). Two different gene products that have been identified as homologues to yeast EMP70 were found in our study, i.e. p76 (66) and the protein designated in the data base "similar to EMP70" (67). Sequences corresponding to p76 and "similar to EMP70" were observed by Edman degradation at 50, 147, and 165 kDa. Remarkably, one of the two KDEL receptors (Elp-1b, Table I) was also found at 147 and 165 kDa (besides its monomeric mobility at about 24 kDa). Conceivably, p76, "similar to EMP70," and the KDEL receptor may associate into SDSresistant complexes. Unexpectedly, SCAMPs 1 and 3 were observed, as was a membrane protein, BAP31, previously implicated as a regulator of apoptosis and suggested to cycle between the ER and Golgi apparatus (68, 69). Members of the p24 family, as well as p58, were found. Mammalian constituents (Sec23A and Sec13) of the COPII complex postulated to bind to p58 and the p24 family (13) were also identified. Remarkably, Sec22b, a v-SNARE regulating ER to Golgi secretory cargo transport, was found (70 -72). Also found were the NSF-associated protein ␣ SNAP and two TGN trafficking proteins, i.e. the cation-dependent mannose 6-phosphate receptor as well as TGN38. MS/MS data were obtained from 40 individual peptides from 10 different bands that we were unable to assign to a data base entry (data not shown). This unassigned data could represent as many as 40 novel proteins, but it is more likely that several peptides originate from the same proteins, and therefore the number of novel proteins is probably less than 40. Nevertheless, this represents an unexpectedly high number of novel proteins whose significance remains to be elucidated.
Localization of Selected Trafficking Proteins and GPP34 -To assess whether selected trafficking proteins were indeed in Golgi membranes or contaminants, an immunolocalization study was effected. We elected to study the intra-Golgi and cellular distribution of selected trafficking proteins and GPP34 by electron microscope immunolabeling of the Golgi fraction in situ and by analytical fractionation of total liver membranes, respectively. Neither approach has previously been used to address the distribution of these proteins.
Antibodies were raised to GPP34. Western blotting confirmed that the antigen was membrane-associated as well as cytosolic (data not shown). Immunolabeling of cryosections of the WNG fraction with the peptide-specific antibody to GPP34 revealed specific labeling at the periphery of the Golgi stack. Antigenicity was found on the cis and trans sides as well as at the lateral edges of stacked cisternae (Fig. 4, indicated by arrowheads). Controls without primary antibody but with secondary antibody conjugated to gold revealed no detectable gold labeling in the fraction (data not shown). Furthermore, confocal laser scanning confocal microscopy of rat hippocampal neurons (Fig. 5) revealed a juxtanuclear staining (Fig. 5A) as well as a more diffused cytosolic labeling. The juxtanuclear staining corresponded to that of the Golgi marker MG160 (Fig. 5B), with partial overlap observed (Fig. 5C). Labeling with anti-GPP34 alone gives identical staining to that seen in Fig. 5A (Table II). AA's, amino acids. C, nanoelectrospray ionization MS/MS of the doubly charged (m/z 566.7) 1131.4-Da tryptic peptide reveals this sequence tag fragmentation pattern, which identifies the N-terminal tryptic fragment of rat Rab6. This peptide, indicated by an asterisk in A ((M ϩ H) ϩ 1132.6), was not assigned by MALDI peptide mass mapping because of post-translational modifications (removal of the initiation Met, acetylation of the N-terminal Ser (S-acet)) and rodent-specific sequence variation of Ala for Thr at position 3 of the predicted human sequence. The experimentally determined sequence of the N-terminal tryptic peptide of mature rat Rab6 is acetyl-SAGGDFGNPLR. This alanine to threonine sequence difference in the N-terminal tryptic peptide (compare human Rab6, B) for rodent Rab6 was confirmed in nine mouse ESTs (NCBI nr.05.07.99).  16 (Fig. 1, right-hand side) identifying Rab6 The 26 matched peptides cover 75% (157 of 208 amino acids) of the human Rab6 protein. Shown are the 85 delayed extraction MALDI-MS (m/z) values submitted to ProteinProspector MS-Fit; the monoprotonated masses (MH ϩ ) of the 26 matched tryptic peptides derived from human Rab6; the mass differences (Delta); and the list of 59 unmatched masses. Start and end residue numbers, the amino acid residues before and after (in parentheses), and the corresponding tryptic peptide sequences are shown. Assigned modifications indicated: oxidized methionine, Met-ox; pyroglutamic acid, pyroGlu. (data not shown). Hence GPP34 is in part Golgi-localized. Direct immunolabeling of whole fractions by the method of Lavoie et al. (27) revealed that the Golgi SNAREs identified here, i.e. GS15 and GS28, were both Golgi-located but with different distributions (Fig. 6). GS15 was associated with edges of distended cisternae, whereas GS28 was primarily in Golgiassociated smooth membranes as well as in Golgi cisternae and vesicles. SCAMP1 was found on small Golgi-associated structures, whereas SCAMP3 was found in larger structures. Stacked flattened cisternae showed only low labeling. The ARF-GEF identified here (i.e. GBF1) was largely in smooth membranes clearly associated with stacked Golgi cisternae as well as in larger lipoprotein-filled structures. BAP31 was found on ER contaminants but also on associated vesicles and flattened cisternae of Golgi stacks. As controls, the p24 family member ␣ 2 p24, which is terminally N-glycosylated in this fraction, was cis located as predicted from previous studies (13), and calnexin was primarily found in ER contaminants (Fig. 6).
To identify the steady state proportion of these trafficking proteins, which codistributed with the Golgi markers galacto-syl transferase or mannosidase II, analytical fractionation was employed (Fig. 7, A and B). The rationale of analytical fractionation is that the sedimentation properties of marker proteins define compartmental boundaries, with median densities used as an index of comparison (73). Here GPP34 revealed a median density of 1.096, slightly less than that of the Golgi markers galactosyl transferase (median density 1.122) and mannosidase II (median density 1.120) (Fig. 7). Most of BAP31 and GS28 was found at higher median densities than that of the Golgi marker (galactosyl transferase or mannosidase II), although some GS28 was deduced to be Golgi-associated, as was the ␣ 2 member of the p24 family. SCAMP 1 revealed a distribution similar to the Golgi marker, as did SCAMP3. No GBF1 signal was detected, presumably because of its dissociation from the MLP fraction as a consequence of the preparation method and the imidazole buffer that was used (data not shown). The GS15 signal was beyond detection in the MLP fraction by Western blotting, presumably because the antibody is inefficient at recognizing denatured protein in Western blots. We have, however, previously detected GS15 by a chemiluminescence method of detection of Western blots, but only in the purified Golgi (WNG) fraction (7) and not in the total membrane fraction used for analytical fractionation.

DISCUSSION
The elucidation of the membrane protein complement of an organelle, i.e. the Golgi complex, was attempted by organelle isolation, phase partitioning to extract the integral membrane proteins, and protein characterization from one-dimensional SDS-PAGE by Edman degradation and mass spectrometry. Although two-dimensional gels are the preferred approach for proteomics, there remains an unresolved experimental difficulty in the solubility of membrane proteins during isoelectric focusing (74).
In the present study, the identified proteins were further characterized for their location by immunolabeling and analytical fractionation. Indeed, for the SCAMP proteins, the Golgi SNARE GS28, and the apoptosis-related protein BAP31, this study represents their first characterization by these approaches.
The methodology identified 81 proteins. These represented only the most abundant membrane proteins, and further refined analysis suggests a far greater complement in this Golgi fraction. 2 Of the 81 proteins characterized, 49 were considered as integral membrane proteins on the basis of having one or more predicted transmembrane domains (Table I). In addition to the transmembrane proteins are those proteins predicted to have covalent lipid modification motifs for insertion into the membrane by lipid tails (about 12). Of all proteins identified, 45 were considered to be in whole or in part Golgi-located. These included 17 resident membrane proteins and 28 trafficking proteins. Contaminants mainly from the ER and mitochondria were readily identifiable and represented 24 proteins. In addition, 40 sequence tags representing between 10 and 40 different novel proteins were elucidated. The single full-length novel protein GPP34 identified by gene building of mammalian EST sequences (Fig. 3) revealed deduced sequence information from human to yeast. Although the full-length cDNA cloning of this gene has not been reported, the gene has already been shown to be nonessential for viability of S. cerevisiae (40), which does not exclude a regulatory role for GPP34 in Golgi trafficking.
GPP34 revealed no targeting motifs, which might predict a Golgi localization. Comparison of its distribution between membranes and cytosol revealed it to be membrane-associated as well as cytosolic, as deduced from subcellular fractionation (data not shown). Cryosections were used to ensure maximum availability of potential antigenic epitopes to the applied antibody. By the criterion of localization of antigenicity on cryosections of the Golgi fraction, GPP34 was deduced to be Golgilocalized. Furthermore, by confocal fluorescence microscopy GPP34 was found in hippocampal neurons and fibroblasts to be concentrated in regions overlapping that of the Golgi marker MG160. The EST-derived primary sequence revealed no potential signal sequence or transmembrane domain. Hence, the protein is most likely synthesized on cytoplasmic ribosomes. Analytical subcellular fractionation revealed a distribution and median density to slightly lower densities than that of the two Golgi markers employed (galactosyl transferase and mannosi-dase II). Most recently, Lin et al. (5) have identified a recycling pathway for Golgi enzymes from Golgi cisternae to the p24containing cis Golgi compartment. The higher density tail in analytical gradients of galactosyl transferase and mannosidase II (1.12-1.19) could well correspond to a compartment harboring recycling Golgi proteins, i.e. the cis Golgi network. The peripheral membrane protein GPP34 appears to be absent from this latter compartment. GPP34 antigenicity at the top of the gradient may represent protein that dissociated from Golgi membranes during centrifugation. These explanations may account for the slightly lower median density of GPP34 than that of the Golgi markers. Two other novel proteins (i.e. 25DX and GS3786) were identified by Edman degradation. These integral membrane proteins have both been cDNA-cloned; their function remains largely unknown, and their sequences appear to be restricted to mammals. No information on their cellular localization is known.
The choice of Golgi fraction analyzed, i.e. the WNG fraction, was a consideration of its prior characterization (7). In our study, cargo was not depleted from the Golgi apparatus by agents such as cycloheximide, in order not to compromise the distribution of any membrane protein whose Golgi location may depend on cargo. Indeed, continued protein synthesis is FIG. 7. Distribution of SNAREs, SCAMPs, trafficking proteins, ER, and Golgi markers by analytical fractionation. Two separate experiments (A and B) are illustrated in which a combined total membrane preparation from rat liver homogenates (MLP fraction) was fractionated by isopycnic sucrose gradient centrifugation. In experiment A, the distribution of total protein is compared with that of the Golgi marker enzyme galactosyl transferase (GalT), the novel Golgi protein GPP34, SCAMP3, SCAMP1, the p24 family member ␣ 2 p24, the Golgi SNARE GS28, the ER marker CNX, and the rough ER marker ribophorin II. The median densities are indicated as vertical arrows. In B is shown a second experiment in which the distribution of mannosidase II is compared with that of ␣ 2 p24, BAP31, CNX, and ribophorin II. The distribution of ␣ 2 p24 revealed a variation in experiment A compared with B, whereas other control markers, i.e. ribophorin II, calnexin, and the Golgi markers (galactosyl transferase (A), mannosidase II (B)), reveal similar median densities. required for the localization of membrane proteins such as BAP31 in post-ER compartments (68). Independently of its effect on protein synthesis, cycloheximide also affects the secretory pathway directly, and this could also affect the protein distribution in the Golgi (75). The problem of increased cargo in the fraction was readily accommodated because few were expected to be recovered in the detergent phase following extraction with Triton X-114. A total of 10 candidate cargo proteins were identified by Edman degradation or mass spectrometry, all of which were either amphipathic or integral membrane proteins. However, whether endosomal proteins, i.e. p76 and "similar to Emp70," are cargo or contaminants requires further analysis.
It is noteworthy that by N-terminal Edman degradation, a different cohort of integral membrane proteins was identified from that found by mass spectrometry. N-terminal Edman degradation was superior for identifying multipass integral membrane proteins. Hence, the two KDEL receptors (i.e. Elp-1a and Elp-1b) and the two EMP70 homologues were identified by N-terminal sequencing but not by mass spectrometry. Multiple transmembrane-containing proteins, such as Elp-1a, Elp-1b, p76, and "similar to EMP70," may not be efficiently in-gel-digested, and/or the tryptic peptides may not be efficiently extracted after digestion for mass spectrometry. Indeed, none of the tryptic fragments of any of the transmembrane proteins identified by mass spectrometry (Fig. 1, right side) covered hydrophobic transmembrane domains.
Implications for Membrane Trafficking Models-The prominence of SCAMPs was unexpected. Although they have been implicated in endocytosis and exocytosis (76), no clear function is known for these tetra span membrane proteins. The localization of SCAMP1 to Golgi components predicts a role in membrane trafficking. Indeed, this is the only speculative function for this family of membrane proteins, i.e. in trans Golgi cycling compartments (76). Remarkably, both SCAMPs showed a different intra-Golgi distribution by electron microscopy (Fig.  6, C and D), whereas analytical fractionation (Fig. 7A) revealed a similar distribution for both and similar to that of the Golgi marker enzyme galactosyl transferase.
The Golgi complex is exquisitely sensitive to the drug brefeldin A, with the action of this drug on Golgi structure and the inhibition of secretion (77, 78) a consequence of the interaction of brefeldin A with an ARF-GEF (50,79). Two mammalian ARF-GEFs of 200 kDa with a Sec7 domain have been characterized to date (48,49,51). Only one of these, GBF1 (48), was identified here as partitioning into Triton X-114. Whether the other (BIG1) is also present requires further evaluation. Interestingly, the majority of GBF1 was localized to elements apposed to either side of the stacked Golgi cisternae (Fig. 6). Previous EM studies have shown a cis Golgi location in flattened cisternae by cryosectioning of intact liver (49). Efforts at more precisely localizing GBF1 within the Golgi complex are underway. Also found in domains apposed to one pole of the Golgi stack was ␣ 2 p24, a membrane protein previously identified in the cis Golgi network and involved in ER to Golgi recycling mechanisms and in COPI binding (5,13,27). The limited number of SNAREs, single ARF-GEF partitioning into Triton X-114, and recycling membrane proteins from the Golgi to ER (two Elps, p58, p24 family members) as prominent constituents are unexpected. Indeed, the coincident distribuition of the apoptosis-related protein BAP31 (69) with the cis Golgi marker ␣ 2 p24 implicates the former in an ER to Golgi recycling pathway. Characterization of the full-length sequences corresponding to the 40 sequence tags with no corresponding sequences yet found in current data bases may extend consider-ably the molecular machineries regulating the function of the Golgi complex in vivo.