The Role of a 21-kDa Viral Membrane Protein in the Assembly of Vaccinia Virus from the Intermediate Compartment*

We have recently provided morphological evidence that a key event in the assembly of vaccinia virus is the formation of a novel cisternal domain of the intermedi- ate compartment (IC) between the endoplasmic reticu-lum and the Golgi complex (Sodeik, B., Doms, R. W., Ericsson, M., Hiller, G., Machamer, C. E., van’t Hof, W., van Meer, G., Moss, B., and Griffiths, G. (1993) J. Cell Biol. 121, 521–541). This tightly apposed cisternal domain incompletely surrounds the spherical immature virus that matures into the first of the two distinct infectious forms of vaccinia, the intracellular mature vi- rus (IMV). In this study we describe the characterization of an abundant membrane protein of the IMV, the gene product of A17L, a 21-kDa protein that has recently been shown to be essential for the formation of the viral mem- branes (Rodriguez, D., Esteban, M., and Rodriguez, J. R. (1995) J. Virol. 69, 4640–4648). Upon translation in vitro , p21 associated with rough microsomal membranes in a co-translational manner. Using NH 2 - and COOH-termi- nal specific antibodies, we show that both in

Vaccinia virus, the best characterized member of the orthopoxviridae is a double-stranded DNA virus, whose genomic size approaches 200 kilobases, that encodes for over 200 proteins. Approximately half of these proteins are part of the structure of the infectious particles, while the non-structural proteins carry out a variety of complex viral functions (Moss, 1990). This virus family is unique in having two different infectious forms, the intracellular mature virus (IMV) 1 and the more complex extracellular enveloped virus (EEV) which is enclosed by one additional membrane relative to the IMV (Griffiths and . The first morphological signs of the infection process is the appearance, at approximately 3 h after infection, of large structures enriched in DNA in the infected host cell, easily visualized by DNA dyes (Cairns, 1960). These structures are the sites of DNA replication and transcription (Moss, 1990). The first ultrastructural evidence for viral membrane assembly is the presence of rigid, curved membrane structures (the crescents) that, although still open to the cytoplasm (Sodeik et al., 1993) can give rise to perfectly spherical profiles (the so-called immature virus (IV)) as seen by thin section EM (Dales, 1963). These structures then enclose the viral DNA (Morgan, 1976) to form the IMV whose DNA is completely protected from the outside environment.
A long standing dogma in the vaccinia field proposed that the viral membranes, were the result of a poorly defined de novo mechanism of membrane synthesis (see Dales and Pogo (1981), Mohandas and Dales (1995), and Wilton et al. (1995)). We have recently challenged this view and provided evidence that the crescents and the IVs are composed of not one, but two, tightly apposed membranes reminiscent of cellular junctions. Moreover, structural and immunocytochemical data showed that the viral-induced cisternal membranes are continuous with membranes of the intermediate compartment (IC) between the rough ER and the Golgi complex. In conjunction with lipid analysis showing that the lipids of the IMV membranes resembled the ER more than other organelles (Sodeik et al., 1993), these data led to a unique model of the assembly of vaccinia postulating that the process depended on the formation of a newly synthesized, vaccinia-dependent, cisternal domain arising from the IC. One consequence of this mechanism is that the outer surface of the IMV is topologically equivalent to the cytoplasmic surface of cellular membranes, rather than the luminal domain, as in most other enveloped viruses that bud at intracellular membranes.
A prediction of this assembly mechanism based on known cell biological principles (Griffiths and Rottier, 1992) was that the process would require the presence of vaccinia-encoded membrane protein(s) that behaved as markers of the IC. Thus, the model required that at least one vaccinia-encoded mem-brane protein would insert into the rough ER and, following folding and assembly, be transported to and retained in the IC. Since the IVs curve in one direction, a further requirement of this model is the need for the cisterna to be polarized such that two distinct classes of vaccinia integral membrane proteins would be sorted to the outer and inner membrane profile of the IC-derived cisterna, that is evident in thin section EM. Supporting this view is our recent data showing that two peripheral membrane proteins, p14 (gene A27L) and p65 (D13L; the target of the drug rifampicin, which reversibly blocks IV assembly) (Moss et al., 1969;Baldick and Moss, 1987;Tartaglia et al., 1986), are differentially targeted to the outer and inner membrane, respectively (Sodeik et al., , 1995. Presumably, these peripheral membrane proteins are binding to two distinct integral membrane proteins in the two cisternal bilayers. The IC (also referred to as the cis-Golgi network and ERGIC) is the focus of considerable controversy, since the realization that a number of antigens localized specifically to membrane structures distal to the rough ER and proximal to the first Golgi compartment (Hauri and Schweizer, 1992;Saraste and Svensson, 1991;Hobman et al., 1992;Griffiths and Rottier, 1992). We define the latter as the site where the trimming of mannose residues on N-linked oligosaccharides down to five mannose residues occurs, as well as the compartment where the acquisition of sensitivity to digestion with endoglycosidase-D by this class of oligosaccharides takes place . A key finding in this field has been to show that the entry from the ER into the first Golgi compartment is blocked at 15°C (Saraste and Kuismanen, 1984) or by energy blocks (Balch et al., 1986). Furthermore, genetic, biochemical, and electron microscopical data argue strongly for a minimum of one vesicular transport step between the ER and the Golgi in both mammalian cells and in yeast (see Beckers et al. (1990) and Rexach and Schekman (1991)). We have recently shown that, at 15°C or in the presence of GTP␥S, ␤ cop-containing (or cop I) buds accumulate in the IC Griffiths et al., 1995) and have provided data arguing that the membranes of the IC are physically and functionally continuous with the rough ER. Moreover, these cop I buds contained the G protein of vesicular stomatitis virus when the entry of the latter was blocked at 15°C . This model of structural and functional continuity between the rough ER and a complex IC network, extending into the periphery of the Golgi stack, is also supported by a large number of classical ultrastructural studies (see Claude (1970), Ellisman (1985a, 1985b), Sesso et al. (1994), and Clermont et al. (1994), as well as a recent study using defined markers (Stinchcombe et al., 1995)). In spite of these extensive ultrastructural data, alternative views persist (Pelham, 1994;Aridor et al., 1995).
The membrane protein p21 (A17L) of vaccinia seemed a good candidate to test our model on the production of VV membranes from the IC. This protein was recently shown to be essential for virus assembly, and in its absence no IVs are formed (Rodriguez et al., 1995). It is a non-glycosylated integral membrane protein made as a 23-kDa precursor that is cleaved at a later step of IMV assembly (Rodriguez et al., 1993;Whitehead and Hruby, 1994). Moreover, this protein has been shown to form specific complexes with the peripheral, outer membrane protein p14 (Rodriguez et al., 1993). In this study we show that p21 of vaccinia has the hallmarks of a resident protein of the IC that is co-translationally inserted (SRP-dependent) into the rough ER.

MATERIALS AND METHODS
Cells, Viruses, and Antibodies-HeLa cells (CCL2) were obtained from ATTC and grown in Dulbecco's modified Eagle's medium (DMEM) containing penicillin and streptomycin and 5% heat-inactivated fetal calf serum (DMEM/FCS). For most experiments a deletion mutant of VV strain Western Reserve, vRB12 (a kind gift of Dr. B. Moss) was used. This mutant virus lacks the 37-kDa EEV protein and therefore cannot make IEV and EEV (Blasco and Moss, 1991). The virus was propagated in HeLa cells and purified as described (Roos et al., 1996). Since vRB12 does not make plaques on cell monolayers, the stocks were titrated by immunofluorescence. Peptide antibodies were raised against the following peptides: SYLRYYNMLDDFSAG (NH 2 -terminal, residues 2-16) and KPYTAGNKVDVDIPTFNSLNTDDY . These peptides were coupled to keyhole limpet hemocyanin. Rabbits were immunized and bled as described and the serum collected and stored at Ϫ20°C. 2 Antibodies recognizing a 39-kDa core protein (A4L) were from Dr. M. Esteban (Maa and Esteban, 1987). The NH 2 -and COOH-terminal specific antibodies recognizing the mouse hepatitis virus M protein have been described previously (Krijnse Locker et al., 1992).
Metabolic Labeling, SDS-PAGE, and Isoelectric Focusing-VRB12 infected cells, grown in 6-cm dishes, were starved in methionine/cysteine-free DMEM at 6.5 h post-infection. The cells were pulse-labeled for 5 min at 7 h post-infection with 200 Ci/dish EXPRE 35 S 35 S (DuPont NEN), washed once with PBS, and chased in DMEM for 30, 60, and 120 min. The cells were lysed as before in 500 l lysis buffer containing protease inhibitors (Krijnse Locker et al., 1992). The lysate was immunoprecipitated using 5 l of the NH 2 -terminal antibody as described by Krijnse Locker et al. (1992). The immunoprecipitates were analyzed in a 15% SDS-polyacrylamide gel. For the two-dimensional gel blot of p21, purified IMV was incubated for 30 min at 37°C in first dimension lysis buffer (9.8 M urea, 2% ampholines (Pharmacia, Uppsala, Sweden) pH 7-9, 4% Nonidet P-40, 100 mM DTT) and subsequently isoelectric focused using a Bio-Rad mini-two-dimensional gel system and a pH gradient from 3 to 7 and run in the second dimension on 15% SDSpolyacrylamide gel.
Western Blotting, Nonidet P-40-DTT, and Protease Treatment of the IMV and Post-nuclear Supernatant-Two 10-cm dishes of HeLa cells were infected with vRB12 for 1 h at 37°C, after which the inoculum was replaced by DMEM/FCS containing 100 g/ml rifampicin. At 9 h postinfection, the cells were washed twice with ice-cold PBS, scraped from the dish in PBS, and gently pelleted by centrifugation at 2000 rpm for 1 min. The cell pellet was resuspended in 350 l of 10 mM Tris-Cl, pH 9, and the cells broken by 15-20 strokes of a 2-ml Dounce homogenizer. The nuclei were removed by centrifugation for 2 min at 3000 rpm. The post-nuclear supernatant was treated with 100 and 200 g/ml trypsin (Sigma) and protease K (Merck, Darmstadt, Germany), for 30 min at 37°C or at 4°C, respectively. Trypsin treatment was stopped with aprotinin (4 g/ml; Sigma) and trypsin inhibitor (200 g/ml; Sigma), and to the protease K-treated samples PMSF (4 g/ml; Sigma) was added. The digested and untreated post-nuclear supernatant were run on SDS-PAGE and p21 was detected by Western blotting using the NH 2 -terminal antibody at 1:400 and the COOH-terminal antibody at 1:1000 as described . The protease treatment of purified IMV was carried out as described before (Roos et al., 1996). Purified vRB12 was briefly sonicated and incubated for 30 at 37°C with 20 mM DTT in 10 mM Tris-Cl, pH 9. Then an equal volume of 2% Nonidet P-40, 40 mM DTT was added and the incubation continued for another 30 min. After a brief second sonication, the sample was layered onto 50 l of 36% (w/v) sucrose in 10 mM Tris-Cl, pH 9, and centrifuged for 30 min at 24 p.s.i. in a Beckman Airfuge. The upper phase was concentrated by acetone precipitation and, like the pellet, dissolved in LSB. The samples were run on 15% SDS-PAGE and blotted onto nitrocellulose, and p21 was detected as described above.
Cloning of p21-A recombinant plasmid, pUCA17L, containing the gene A17L of VV strain WR, was obtained from Dr. M. Esteban. The gene had been polymerase chain reaction-amplified from the VV WR genome using the following primers: 5Ј-CCGGATCCCCCATGAGTTA-TTTAAGATATT-3Ј (upstream) and 5Ј-TATCTGCTATTCCTGGTG-3Ј (downstream). The BamHI restriction site in the upstream primer and a PvuII site downstream of the gene were digested and used to clone the gene into BamHI/SmaI-digested pUC19. Subsequently, the gene was cut from this vector using SalI and SacI and positioned downstream of the T7 RNA polymerase promotor of plasmid pBSSK (Stratagene) to generate T7 expression vector pBSA17L. Sequence analysis of the A17L gene in pBSA17L showed that its sequence was completely identical to the previously published sequence of VV Copenhagen strain (Goebel et al., 1990). To create a second expression vector, which allowed expression driven by a cellular promotor, the A17L gene was cut from pBSA17L using XbaI and KpnI and cloned downstream of the ␣-globin promotor of vector pCMUII, a derivative of pC81G (Pääbo et al., 1986) a kind gift of Dr. T. Nilsson. The generation of pTz vector bearing the MHV-M protein has been described before (Krijnse Locker et al., 1992).
Immunofluorescence and Electron Microscopy of Infected and Transfected Cells, Pre-embedding Labeling, and Immunofluorescence of Streptolysin O-permeabilized Cells-For the indirect immunofluorescence, HeLa cells grown on coverslips were infected with vRB12 and fixed at 7.5 h, post-infected by a 5-min incubation at Ϫ20°C with precooled methanol. The cells were rinsed with PBS and blocked with 5% FCS in PBS for 10 min at room temperature. To detect p21 by indirect immunofluorescence, we used the COOH-terminal antibody at 1:150 and a goat anti-rabbit conjugated to fluorescein isothiocyanate from Dianova (Dianova, Hamburg, Germany). The labeled cells were subsequently incubated with 5 g/ml Hoechst B2883 (Sigma) for 30 min at room temperature to label the DNA.
Cryosections of HeLa cells infected for 7.5 h were prepared as described before . Transfection of p21 was performed as follows. HeLa cells either grown on coverslips (for IF) or in 6-cm dishes (for EM) were infected with VVT7-3 (a kind gift of Dr. B. Moss) for 45 min at 37°C. The inoculum was removed and replaced by serumfree DMEM containing 5 mM hydroxyurea (Sigma). The cells were left for 15 min at 37°C before transfection. The infected cells were transfected using DOTAP (Boehringer GmbH, Mannheim, Germany) and 2 g of CsCl purified pBsA17L/3.5-cm dish, in serum-free DMEM containing hydroxyurea. Infected and transfected cells were fixed for either IF or EM at 6 h post-transfection.
The pre-embedding labeling was performed as follows. Six-cm dishes of HeLa cells, infected for 6 h, were incubated for approximately 5 min at room temperature in water. When the cells were sufficiently swollen, the water was quickly replaced by 4% paraformaldehyde in 250 mM Hepes-KOH, pH 7.4. This sudden change of osmolarity caused the cells to break open while being fixed. The cells were fixed in this solution for 10 min at room temperature, rinsed extensively with PBS and 20 mM glycine, blocked with PBS/glycine containing 5% FCS for 2 h at 4°C, and incubated overnight at 4°C with the antibodies (diluted in PBS/ glycine-FCS; NH 2 -terminal 1:100, COOH-terminal 1:300). After thorough washing with PBS/glycine, the cells were incubated for 3 h at 4°C with protein A-gold diluted in PBS/glycine-FCS, washed extensively before fixation with 1% glutaraldehyde and embedding into Epon. The quantitation of the labeling was done as follows. For each antibody, 100 IVs or crescents were counted that were labeled by at least one gold particle. Gold particles were classified as being either on the inner or outer membrane when the labeling was not further than 20 nm from the IV membrane (Griffiths, 1993).
For the immunofluorescence in permeabilized cells, infected and rifampicin-blocked cells were permeabilized with streptolysin O and immunolabeled as described previously , except that 2 units/ml SLO was used instead of 3 units/ml and that GTP␥S was omitted. The cytoplasmically exposed domains were visualized using the NH 2 -(1:100) and COOH-terminal (1:150) antibodies.
In Vitro Translation-In vitro translation of pBSA17L was done using the T7-coupled transcription and translation system provided by Promega (Promega, Madison, WI). Dog pancreatic rough microsomes were a kind gift of Katja Schroeder and Dr. Bernard Dobberstein. To remove endogenous mRNAs, the microsomes were treated with micrococcal nuclease. Rough microsomes were diluted in RM buffer (50 mM Hepes-KOH, pH 7.5, 50 mM KOAC, 2 mM MgAc, 250 mM sucrose, and 1 mM DTT) to an A 280 of 20. To this CaCl 2 to 1 mM, 150 units/ml micrococcal nuclease (Sigma) and 2 mM PMSF were added and incubated for 10 min at 25°C. The reaction was stopped by addition of 2 mM EGTA. The microsomes were layered on top of 500 l of 0.5 M sucrose in RM buffer and spun for 30 min at 35,000 rpm in a TLS55 Beckman tabletop centrifuge rotor. The pellet was resuspended in 450 l of RM buffer and the A 280 measured against the untreated microsomes. As a control we used the mouse hepatitis M protein, which has been shown to associate with efficiencies of up to 95% with rough microsomal membranes (Rottier et al., 1984(Rottier et al., , 1985. The M and p21 protein were translated in vitro according to the instructions of the manufacturer using a 25-l system. When microsomes were added they were used at 100 A 280 units/translation. To test for the membrane insertion, translation mixtures were loaded on top of a 75-l cushion of 0.5 M sucrose in RM-buffer. The samples were spun for 10 min at 26 p. s. i. in a Beckman Airfuge. The supernatant and the pellet were diluted in 500 l of lysis buffer (see above) containing 0.2% SDS, and the proteins they contained immunoprecipitated with the COOH-terminal specific antibodies to p21. When testing for the post-translational insertion of p21, 200 g/ml cycloheximide and microsomes were added at the end of the translation and the incubation continued for 45 min at 30°C, after which the samples were spun as above. For protease treatment of the microsomes, the samples were mixed after translation with an equal volume of protease K diluted in RM buffer at the indicated concentrations and incubated for 30 min on ice. After addition of PMSF, the samples were immunoprecipitated with the NH 2 -and COOH-terminal specific antibodies.

Generation of Antibodies and Identification of p21 on Two-
dimensional Gels-In order to characterize the vaccinia gene A17L, a 21-kDa membrane protein of the IMV, we produced peptide antibodies recognizing its 15 NH 2 -and 24 COOHterminal amino acids, respectively. Purified IMV was separated on two-dimensional gels, blotted onto nitrocellulose, and probed with the NH 2 -terminal peptide serum. Several spots were detected, all running around 21-25 kDa and ranging in pI between 4.5 and 4.1 (Fig. 1). This is consistent with the predicted molecular mass of the protein of around 21 kDa and an isoelectric point of 4.2. The COOH-terminal antibody, however, seemed to recognize only the spots running at 25 kDa (not shown), suggesting that the 21-kDa form may be a processed form lacking the extreme COOH terminus (see below).
Upon in Vitro Translation, p21 Inserts into Microsomes-As pointed out in the introduction a critical test of our assembly model is the need to find at least one membrane protein which behaved as a typical marker of the IC and that is targeted to the IC after insertion into the rough ER. Although this may seem a trivial point, we noticed during the mapping of the membrane proteins of the IMV that the virus lacks any typical class I or class II (von Heijne and Gavel, 1988) membrane proteins (see "Discussion"). Also, several of the IMV proteins, which behave like membrane proteins in infected cells, failed to associate with rough microsomal membranes upon in vitro translation. Examples of such proteins are p39 (A4L) 2 and p32 (D8L; data not shown) (Niles and Seto, 1988), which only associate with membranes in the context of a viral infection.
The sequence of p21 contains a very hydrophobic central domain long enough to span a bilayer four times. A recently described secondary structure prediction program was applied to the p21 sequence, focusing especially on predicting putative transmembrane domains (PHDhtm; Rost and Sander (1994)). This analysis predicted the four stretches of the sequence between the amino acids 61-77, 81-98, 118 -138, and 141-161 to be ␣-helical, a property of transmembrane domains, suggesting that the protein may span the membrane four times. Moreover, these predicted ␣-helical regions coincide with the hydrophobic stretches as determined by a Kyte and Doolittle plot (see Rodriguez et al. (1993)). The p21 protein is not predicted to have a hydrophobic region at its extreme NH 2 terminus, nor does a program designed to predict consensus sequences for signal FIG. 1. Detection of p21 by two-dimensional gel Western blot. Purified IMV was dissolved in first dimension lysis buffer and the proteins separated by isoelectric focusing in the first dimension and SDS-PAGE in the second dimension. As a marker on the far right side of the gel, the same sample was run in the second dimension only (1D). After electrophoresis the gel was blotted unto nitrocellulose and probed with the NH 2 -specific antibody. The isoelectric focusing is shown from basic (left; B) to the acid side (right; A).
peptidase (according to von Heijne rules) predict that p21 has an NH 2 -terminal cleavable signal. The p21 sequence thus resembles that of a class of multi-spanning membrane proteins that generally lack cleaved NH 2 -terminal sequences and that make use of internal hydrophobic domains for insertion (Friedlander and Blobel, 1985;Krijnse Locker et al., 1992) (see Rapoport (1986) for a review).
To assess whether p21 was able to insert into micosomal membranes, the protein was translated in vitro either in the presence or absence of dog rough microsomes. p21 clearly inserted into microsomes; about 80% pelleted with the membranes, while the protein remained in the supernatant when microsomes were not added to the translation mixture (Fig. 2). If the membranes were added after blocking translation with cycloheximide p21 did not pellet, indicating that the insertion into microsomes only occurred co-translationally (Fig. 2). Following co-translational insertion and treatment of the microsomes with increasing amounts of either protease K or trypsin (not shown), reactivity for both the NH 2 -and COOH-terminal antibody was lost; this occurred at a relatively low protease concentration of 10 g/ml, showing that both the termini were exposed on the outside of the membranes (Fig. 3). As a control we used the mouse hepatitis virus M protein, a triple-spanning membrane protein whose topology has been well established (Rottier et al., 1985). Upon translocation in vitro as well as in vivo, its NH 2 terminus is luminal and its COOH terminus cytosolic (Krijnse Locker et al., 1992). In agreement with these results, when MHV-M was translated in vitro in the presence of microsomes, its NH 2 terminus was clearly protected, even at the highest concentration of protease K, while the COOH terminus, similar to the results for p21, was lost at the lowest concentration (Fig. 3).
p21 Adopts the Same Topology in Vivo as in Vitro-We then wanted to confirm the topology of p21 in vivo. For this we infected cells in the presence of rifampicin, a drug that blocks viral assembly prior to the IV stage (Moss et al., 1969) but does not interfere with the synthesis of late proteins nor with the membrane insertion of IMV membrane proteins. 3 Cell homogenates were prepared and treated with trypsin or protease K, and the putative loss of the NH 2 or COOH terminus was assayed by Western blotting. Consistent with the in vitro experiment, both termini were lost after digestion (Fig. 4A). To establish these data more firmly, we next performed an indirect immunofluorescence assay in semi-intact cells (Dupree et al., 1993;Krijnse Locker et al., 1995). Infected cells were permeabilized with streptolysin O, using a two-step procedure of permeabilization described previously to permeabilize the plasma membrane only (see Dupree et al. (1993)). To assess cytoplasmically exposed epitopes, the permeabilized cells were incubated with both antibodies. Both the NH 2 -and COOHterminal antibodies labeled discrete structures in the perinuclear region of the cell, which we recognized as the viral factory (see below; Fig. 4B), thereby confirming the cytoplasmic exposure of both termini also in vivo.
p21 Occurs in Several Forms with Different Molecular Weights-The two-dimensional gel immunoblot (Fig. 1) showed that p21 migrates with different molecular weights as well as isoelectric points. To document this more clearly, p21 migration 3 O. N. Jensen, T. Houthaeve, A. Shevchenko, S. Cudmore, M. Mann, G. Griffiths, and J. Krijnse-Locker, submitted for publication.

FIG. 2. p21 inserts into microsomes in a co-translational man-
ner. p21 was translated in vitro in the presence (ϩ) or absence (Ϫ) of microsomal membranes (RM). Soluble p21 (S) was separated from membrane integrated p21 (P) by centrifugation, and each sample subjected to immunoprecipitation with the COOH-terminal antibody. The protein only pellets when microsomes are present during translation. When microsomes are added after translation is blocked with cycloheximide (RM Ϫ/ϩ), p21 does not associate with the membranes. FIG. 3. Topology of p21 in microsomal membranes. p21 (pBSA17L) or the MHV-M protein (pTzMHV-M) were translated in vitro in the presence of microsomes. After translation the membranes were incubated for 30 on ice with 10 -200 g/ml protease K (Prot. K). Protease treated or untreated (0) samples were immunoprecipitated with NH 2 -(N) or COOH (C)-terminal specific peptide sera. Reactivity to p21 with both antibodies is lost at the lowest (10 g/ml) protease concentration. The NH 2 terminus of the MHV-M protein, however, is protected, even at the highest protease concentration of 200 g/ml.

FIG. 4. Topology of p21 in infected cells. a, protease treatment of post-nuclear supernatant of infected and rifampicin-blocked cells.
Postnuclear supernatant of cells infected for 9 h was treated with 100 and 200 g/ml trypsin (tryp.) or protease K (Prot. K). Samples were prepared for Western blotting and probed with the NH 2 -(anti-N) and COOH-(anti-C) terminal antibodies. b, immunofluorescence. Infected and Rifblocked cells were permeabilized with SLO and incubated with the NH 2 -(A) and COOH-(B) terminal antibodies, followed by a goat antirabbit coupled to fluorescein isothiocyanate. Both antibodies label discrete structures in the perinuclear region, which may represent the viral factories. in 1D gels was compared when expressed in different ways. The pBSA17L plasmid enabled us to do in vitro translation in a reticulocyte lysate as well as transient expression in vivo using the recombinant VV/T7 system (Fuerst et al., 1986; see below). p21 expressed in HeLa cells or translated in vitro in the presence or absence of rough microsomes showed the same pattern: a doublet that ran at about 30 kDa (Fig. 5). In lysates of infected and rifampicin-blocked cells, p21 migrated predominantly as the lower of these two forms, with some also running much faster, at about 20 kDa. In 35 S-labeled and purified IMV, p21 occurred predominantly in this latter 20-kDa form. These different forms of the protein were named p21, p21Ј, and p21Љ (see Fig. 5) and their precise molecular weight was calculated by determining their migration in three separate experiments on 15% SDS-PAGE. p21 migrated at 29 kDa, p21Ј at 25.5 kDa, and p21Љ at 22 kDa. The details concerning the processing of p21 will be described elsewhere. 4 Topology of p21 in the Virus-Having established that both the NH 2 and COOH termini of p21 were exposed in the cytosol, we next wanted to know whether the protein localized to the inner or outer membrane of the IMV. For this purified IMV was treated with both trypsin and protease K, and the putative loss of the NH 2 or COOH terminus was assayed by Western blotting. Since we wanted to follow the putative loss of both the NH 2 and COOH termini, we chose to use a virus stock in which p21 migrated (as assayed by Western blotting) in equal amounts at the position of p21Ј and p21Љ. Fig. 6 shows a representative experiment showing that upon digestion for 30 min, the reactivity to the two antibodies was lost. When the same blot was probed with antibodies to the gene product of A4L (or to the core protein 4a; data not shown), a 39-kDa protein of the viral core (Maa and Esteban, 1987;Roos et al., 1996) 2 no loss of this protein could be observed, showing that the proteases apparently did not digest internally located core proteins (Fig. 6). However, since preliminary EM data had shown that the COOH-terminal antibody preferentially labeled the inside of the immature viruses, we sought to confirm these protease data in a different way.
For this the plasma membrane of infected cells was permeabilized, and after a brief fixation they were incubated with both antibodies and protein A-gold before embedding into Epon, a procedure used previously to localize the protein p65 (D13L) to the concave side of the crescents and to the inside of the IVs . Infected cells permeabilized and labeled in this way showed many IVs that obviously, as expected from our earlier study (Sodeik et al., 1993), had lost their electron dense contents, while the IMVs appeared not to be affected. Intracellular organelles, such as the rough ER and the Golgi complex, still appeared to have their luminal contents intact, indicating that only the plasma membrane was permeabilized (data not shown). In these preparations the electrontransparent IVs, labeled with both antibodies preferentially on their inside (Fig. 7, a-c). In contrast, IMVs did not appear to label on their outside (data not shown). Quantitation showed that 77% of the labeling for the NH 2 -terminal antibody was associated with the inner membrane, while for the COOHterminal antibody this percentage was 90% (Table I). These morphological data strongly suggest that p21 resides predominantly, or even exclusively, in the inner membrane cisterna of the IV and IMV. The apparent discrepancy between the morphological and the biochemical data is addressed under "Discussion." Localization of p21 in Transfected Cells-The co-translational insertion into rough microsomes suggested that, during infection, the p21 most likely becomes inserted into the rough ER before being targeted to the membranes of the viral factory. We therefore sought to confirm this assumption both by localizing p21 in infected cells and by expressing the protein independently. To express p21 in HeLa cells, we initially tried to use a transient expression system, using a vector with a cellular promotor. To detect enough expressed protein by a technique such as immunofluorescence, such expression systems generally require transfection times of at least 24 h. Upon expression of p21 using an ␣-globin promotor (Pääbo et al., 1986), we observed that the (apparently) transfected cells rounded up and detached from the coverslip before expression could be detected. Apparently, the protein when expressed independently was very toxic to cells, since another, cytosolic, VV protein expressed in the same way did not show this cytotoxicity. 2 p21 was therefore expressed using the vaccinia T7 system (Fuerst et al., 1986). Since p21 is a late protein, hydroxyurea was used to block viral DNA-and p21 synthesis, thus allowing the synthesis of only the transfected protein expressed by pBSA17L. Indeed, VVT7-infected and hydroxyurea-treated cells that were fixed at 6 h post infection did not show any detectable levels of p21 by immunofluorescence (data not shown). Upon infection and subsequent transfection of pBSA17L, however, p21-transfected cells showed an intense reticular labeling, typical of the ER (Fig. 8c). By electron microscopy of such transfected cells, strong labeling was observed of the nuclear envelope and the ER (Fig. 9). In some cells labeling was also seen of the Golgi stack and at the plasma membrane (not shown). The morphological preservation of FIG. 5. Comparison of the migration of p21 in SDS-PAGE expressed by different systems. p21 was immunoprecipitated using the NH 2 -terminal antibody from 35 S-labeled IMV (V), from post-nuclear supernatant of infected Rif-blocked cells (P), from 35 S-labeled transfected cells (T), or from in vitro translations carried out in the absence (IVϪ) or presence (IVϩ) of microsomal membranes. The molecular weight markers (M) of 14, 30, and 45 kDa, respectively, are shown on the right. p21 occurs in three different forms: p21 f representing the slowest migrating form that is very prominent in transfected cells or upon in vitro translation. p21Ј, the middle form, is the major band in Rif-blocked infected cells, while the virus contains mainly p21Љ, the fastest migrating form.
FIG. 6. Protease treatment of purified IMV. Purified IMV was treated with 100 and 200 g/ml trypsin (Tryp.) and protease K (Prot. K) and the treated samples analyzed by Western blotting. Reactivity to both the NH 2 -terminal (␣-N) and the COOH-terminal (␣-C) antibody is lost at all protease concentrations. The core-associated protein A4L, however, is not affected by the protease treatment (␣-A4L). these transfected cells (most likely influenced by cytotoxicity of the protein when expressed; see above) made it difficult to identify the IC and well preserved Golgi stacks. Importantly, in these preparations we saw no evidence of any areas of tight membrane-membrane interactions (see "Discussion").
Localization of p21 in Infected Cells-In infected cells the immunofluorescence pattern for p21 was significantly different from that seen following transfection; the COOH-terminal antibody showed a punctate labeling that always seemed to colocalize with the viral DNA, as visualized by Dapi-staining (Fig. 8, a and b). In some cells faint labeling could also be detected in extended reticular structures generally considered to represent the ER, but this labeling was generally not very convincing.
By immuno-electron microscopy using thawed cryosections of fixed cells, p21 strongly labeled the membranes of the cresents and the IVs (predominantly on their inside; Fig. 10, a and  c). The IMVs were only weakly (but specifically) labeled, most likely because the internally located epitopes of p21 are not readily accessible in the densely packed particle (Fig. 10b; see also Ericsson et al. (1995)). Of the cellular organelles, a relatively low level of labeling could be detected on the nuclear envelope and the ER (Fig. 11a). In contrast, we could detect strong labeling for p21 on structures reminiscent of the intermediate compartment. To unequivocally establish the identity of these membrane structures, infected cells were double-labeled with p21 and antibodies to the small GTPase Rab1, which has been shown to localize to the IC ;   Saraste et al., 1995). Co-localization of both proteins was very easily detectable in tubulo-vesicular structures close the to nucleus as well as in more peripheral sites (Fig. 11, a, d, and e).
Often the labeling for both markers appeared close to the Golgi complex, and in most, but not all, random profiles through Golgi stacks, this organelle itself was labeled on one or two cisternae (Fig. 11, b and c), mostly on one (cis-) side, which we believe to be part of the intermediate compartment . These data are fully in agreement with our earlier model of the ER-Golgi boundary, although we again emphasize that, at present, this boundary cannot be unequivocally determined due to the absence of appropriate bona fide markers of the downstream Golgi compartment such as antibodies to the Golgi mannosidase I (see Krijnse Locker et al. (1994) for discussion). In general, more distal compartments of the biosynthetic pathway lacked any significant labeling. Apparently, upon insertion into the rough ER p21 gets efficiently transported to and retained in the membranes of the IC. DISCUSSION We have previously shown that the first observable event in the assembly of vaccinia virus membranes involves the formation of a new cisternal domain of the IC between the ER and the Golgi complex (Sodeik et al., 1993). That analysis was based on EM data that were supported by lipid analysis of the IMV. In the present study we provide, in agreement with a recent genetic approach (Rodriguez et al., 1995), more support for our assembly model at the molecular level. Specifically, we show that the protein, p21, fulfills all the requirements that could be predicted from our earlier data. The protein inserts into the rough ER in a co-translational, presumably SRP-dependent manner and accumulates in the IC in infected cells.
We have recently made a major effort to identify all the membrane proteins of the IMV. 3 A striking observation made during the characterization of these membrane proteins has been that, until now, we have not identified a single membrane protein of the class I or class II (see von Heijne and Gavel (1988) for a definition). Such membrane proteins typically contain a NH 2 -terminal hydrophobic sequence that is used for SRP-dependent translational arrest and subsequent insertion into the rough ER (for a review see Sabatini et al. (1982) and Wickner and Lodish (1985)). Consistent with this observation is the widely held assumption that the IMV, in contrast to the EEV, does not contain any typical N-or O-linked glycoproteins (Garon and Moss, 1971). Some of the membrane proteins we have characterized so far are predicted to be multispanners, like p21 described in this paper. Others seem to belong to a FIG. 9. p21 mainly localizes to the ER in transfected cells. Transfected cells were permeabilized, incubated with anti-p21 antibodies and protein A-gold before embedding into Epon. Several ER elements are shown, all of which label for p21 (small arrows). Bars, 100 nm.

FIG. 10. p21 localizes to the viral membranes on cryosections of infected and fixed cells.
A shows a viral factory (F), from which crescents (CR) bud. The membranes of the crescents label for p21, on their concave side. On the upper left side membrane structures, which seem attached to the factory, are also labeled (arrowheads). B shows IMVs that poorly label for p21. In C several IVs are shown, all of which label predominantly on their inner membrane. In continuity with the IVs are membranes, possibly of the IC, that are also labeled. Bars, 100 nm. novel class of membrane proteins, containing COOH-terminal anchor sequences (Kutay et al., 1993) 2 that probably insert in a post-translational manner. Finally, there is a class of mem-brane proteins that lack any apparent hydrophobic sequence long enough to span a lipid bilayer and whose mechanism of membrane association in infected cells is not clear at present. The significance of this lack in the IMV of "typical" membrane proteins and its possible implications for the generation of the virally-modified IC membranes remain to be determined.
Using both in vitro and in vivo assays, we show in this study that p21 inserts co-translationally into membranes, leaving both the NH 2 and COOH termini on the cytosolic side of the membrane. This would be consistent with the protein spanning the membrane either two or four times, a model consistent with the results of a recently described secondary prediction program that predicts that the protein has four ␣-helical hydrophobic domains long enough to span a membrane (17-21 amino acids long). Moreover, the second loop between the second and third hydrophobic domain, contains an arginine that, following the rules predicting the topology of membrane proteins (von Heijne and Gavel, 1988;Hartmann et al., 1989), can be expected to be cytosolic. Therefore, we believe that p21 spans the membrane four times.
If our prediction of the topology of p21 is correct, then it would bear structural similarity to other membrane proteins spanning the membrane four times. Among this collection of membrane proteins are the connexins (see Beyer et al. (1990) for a review), occludin (Furuse et al., 1993), and myelin proteolipid protein (Popot et al., 1991;Pfeiffer et al., 1993), none of which share any homology to p21 at the sequence level. Not only are these membrane proteins predicted to span the membrane four times and, like p21, leave both termini cytosolic. Strikingly, these proteins are involved in tightly apposing two membranes (tight junctions, gap junctions, myelin sheath formation). It is therefore tempting to speculate that p21 might be important in the tight apposition of IMV membranes. The topology of p21 in the membrane leaves two loops on the luminal side that are available for a putative intra-luminal interaction of p21 to form two tightly apposed membranes of the IC. Upon close inspection of these loops, however, they appear too short (4 and 3 residues respectively) and the residues they contain make it difficult to imagine such an interaction. This would be consistent with our failure to detect any hint of tightly apposed membranes of the ER/IC upon independent expression of p21. We propose therefore that p21 may be necessary, but not sufficient for the process of bringing the two cisternal membranes of the IC into tight apposition during the assembly.
Genetic data have shown that p21 plays an essential role in the assembly of the VV membranes (Rodriguez et al., 1995). An important role for p21 in the generation of the viral membranes could also be predicted from its relative abundance, as well as the fact that it is a very hydrophobic protein, having about half of its amino acids associated with the viral membrane. One possible function of p21 could be the formation of large oligomers ("patches"), as has been shown recently for the M protein of mouse hepatitis virus, another virus that uses IC membranes for its assembly (Opstelten et al., 1995). Such putative p21 oligomers may provide a structural scaffold for part of the crescent structure.
In an attempt to map the protein to either the inner or outer membrane, we used two assays. The pre-embedding labeling unequivically demonstrated that both the NH 2 -and COOHterminal antibodies predominantly labeled the inside the crescents and IVs. This strongly suggests that p21 is located in the inner membrane. However, at first glance these data are not easily reconciled with two other observations. First is our protease experiments using purified intact IMV showing that both the NH 2 and COOH termini are lost after a 30 min protease treatment, a treatment that did not affect a core-associated FIG. 11. Co-localization of p21 with the IC marker Rab1. In this double-labeling cryosection experiment p21 (5 nm gold, small arrowheads) localizes along with Rab1 (10 nm gold, large arrowheads) to membranes of the IC. In A tubular vesicular structures close to the nuclear envelope (NE) label for both Rab1 and p21, while the nuclear envelope itself labels weakly for both markers. Occasionally, these IC structures, positive for Rab1, seem to emanate from the nuclear envelope. In B and C labeling for both markers is seen in IC structures close to the Golgi complex (G). Sometimes the stack itself is labeled on one or two cisternae. In B this labeling can be expected to be on the cis-Golgi side, since on the opposite side an IMV is seen in the process of being wrapped by TGN membranes to form the precursor of the extracellular enveloped virus (Schmelz et al., 1994). D and E show further typical tubular-vesicular structures of the IC, which label for both proteins. In E a section is shown through a IC tubule that is attached to a vesicular element of the IC. These larger vesicular structures probably correspond to the dotty pattern that is seen with many IC markers by immunofluorescence. Bars, 100 nm. protein. We have recently shown, however, that upon prolonged protease treatment of the IMV, the proteases may enter the particle (Roos et al., 1996). In the latter study, treatment of purified IMV with trypsin and protease K digested more than 90% of the core proteins p39 and 4a in a 4-h period. We have interpreted these results to mean, in conjunction with cryo-EM observations of immunogold labeled preparations, that, rather than being a particle that has completely sealed by a fusion of the cisterna with itself, the IMV outer surface may be interrupted by a proteinaceous structure that is sensitive to digestion by exogenous proteases (see Roos et al. (1996)). Therefore, we strongly suspect that over a 30-min digestion period, proteases may have sufficient access to the inner membrane of the virus for them to cleave both the NH 2 and COOH termini of p21, while proteins like p39 and 4a, which make up the core, may resist protease treatment for a longer time.
p21 has also been shown to interact in immunoprecipitation studies with a 14-kDa protein (Rodriguez et al., 1993) that is on the outer (cytoplasmic) surface of the IMV (Sodeik et al., 1995). Since p14 does not have a hydrophobic stretch that could anchor it into a bilayer, Rodriguez et al. (1993) suggested that p21 may bind directly the 14-kDa protein to the outer membrane, thus implying that p21 must be an outer membrane protein.
Based on our present results, we suggest instead that the interaction between p14 and p21 is indirect or that it reflects binding of p14 to a small pool of p21 that labels the outer membrane of the IV (see table I).
Collectively, these data strengthen the model first put forward based on EM studies, supported by a lipid analysis. They show that the multispanning, abundant, and essential vaccinia-encoded protein p21 has all the hallmarks of a key component in initiating the assembly of the IV from the membranes of the IC. Our current studies are focused on determining how this protein interacts with key vaccinia membrane proteins in order to initiate the tight-junction-like membrane apposition that appears to be crucial for the whole assembly process of the virus.