INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Enveloped viruses acquire their lipid membranes by the budding of the viral nucleocapsid (NC) through cellular membranes. Although little is known about the molecular details of this process, it has become clear that the roles played by the viral membrane proteins in the formation of the viral envelope vary tremendously among different viruses. At one extreme, these proteins are not required at all. Viruses such as rhabdoviruses and retroviruses bud normally in the absence of their glycoproteins to form the characteristic bullet-shaped and rounded particles, respectively. At the other extreme, the viral membrane proteins are all that is required for envelope formation. Here, these proteins have the capacity by themselves to carry out the budding of particles devoid of an NC. While such "empty" particles are often smaller than authentic virionssubviral particles have been demonstrated for flaviviruses (1, 26, 37, 47) and hepadnaviruses (41, 49)their dimensions can perfectly match those of normal virions, as we and others have observed for coronaviruses (4, 56). Intermediate between these extremes are the many viruses for which the membrane proteins are essential but not sufficient to form the viral envelope. Here, internal components are also required: they act together with the membrane proteins to accomplish the budding. In this category, alphaviruses are the best-studied examples. (For a recent review of the topic, see reference 19.)

As for large biological complexes in general, molecular interactions between the structural components generate the free energy that drives virus assembly. In view of the widely differing roles of the viral membrane proteins in budding, the significance of the interactions between these proteins is also likely to vary greatly. Thus, while associations between the envelope glycoprotein trimers of retroviruses may be weak or even absent, protein-protein interactions are probably crucial for coronaviruses. Unfortunately, information about such interactions is largely lacking, particularly due to the technical difficulties of obtaining ultrastructural data for these viruses, which for the nonenveloped viruses has proved so valuable. An exception is the alphaviruses: cryoelectron microscopy and image reconstruction of Semliki Forest virus (59) and Sindbis virus (53) revealed among others the icosahedral surface symmetry (T=4) of both their nucleocapsids and their envelopes, as well as the trimeric nature of their spikes. In addition, and more recently, the reconstruction of the Ross River virus particle (9) visualized the tight association between the heterodimeric subunits of neighboring spikes.

Coronaviruses carry three or four proteins in their envelopes. The M protein is the most abundant component; it is a type III glycoprotein consisting of a short amino-terminal ectodomain, three successive transmembrane domains, and a long carboxy-terminal domain on the inside of the virion (or in the cytoplasm) (44). The small E protein is a minor but essential viral component (4, 5, 17, 48, 56). In cells, it accumulates in and induces the coalescence of the membranes of the intermediate compartment (IC), giving rise to typical structures (43). A fraction of the proteins appear extracellularly in membranous structures of unknown identity (35). The trimeric spike (S) protein forms the characteristic viral peplomers. These peplomers are involved in virus-cell attachment and in virus-cell and cell-cell fusion (8). A subset of coronaviruses contains a hemagglutinin-esterase (HE) protein, which occurs as a disulfide-linked homodimer (6).

For assembly of the coronavirus envelope, only the M protein and the E protein are needed (4, 5, 11, 21, 56). Expression in cells of the genes coding for these proteins leads to the formation and release of viruslike particles (VLPs) similar in size and shape to authentic virions. The S protein is dispensable for the formation of these particles. This has now been demonstrated for mouse hepatitis virus (MHV) (5, 11, 56), transmissible gastroenteritis virus (4), and feline infectious peritonitis virus (21). Particularly in MHV, the E protein is only present in trace amounts; though essential for their formation, the protein is barely detectable in VLPs of this virus (56). Thus, the protein component of the envelopes of these particles essentially consists of M molecules. We hypothesize that the coronavirus membrane basically consists of a dense matrix of laterally interacting M proteins, which in some way requires the E protein for budding and in which the S and HE glycoproteins are incorporated, if available, by specific interactions with M (13, 39, 40, 56).

The existence of M-M interactions has already been inferred from data obtained using sucrose gradient analysis. When expressed on its own, the M protein was found in large heterogeneous complexes in the Golgi apparatus (31). The S protein, which by itself is transported to the plasma membrane (40), appeared to associate with these M protein complexes when coexpressed, resulting in its retention in the Golgi complex. Further support for the existence of M-M protein interactions came from our recent observation that assembly-incompetent M protein mutants could be rescued into VLPs (11).

In view of the presumed importance of M proteins for the formation of the coronavirus envelope, the present study was undertaken to provide convincing evidence for the occurrence of interactions between them. In addition, we analyzed which domains of the M molecule are involved in these interactions and investigated where in the cell association of M proteins takes place. Finally, we studied the accuracy with which the M protein framework is composed by analyzing the sorting of foreign membrane proteins.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Cells, viruses, and antibodies. Recombinant vaccinia virus encoding the bacteriophage T7 RNA polymerase (vTF7-3) and OST7-1 cells were obtained from B. Moss. OST7-1 cells (16) were maintained as monolayer cultures in Dulbecco's modified Eagle's medium containing 10% fetal calf serum, 100 IU of penicillin/ml, and 100 µg of streptomycin/ml (all from Life Technologies, Ltd., Paisley, United Kingdom). The hybridoma line OKT8 producing the OKT8 monoclonal antibody against human CD8 (anti-CD8) was purchased from ECACC (Salisbury, United Kingdom). The rabbit polyclonal MHV strain A59 antiserum (K134; anti-MHV) (45), the rabbit polyclonal vesicular stomatitis virus (VSV) antiserum (K114; anti-VSV) (57), and the rabbit polyclonal peptide serum raised against the 18 carboxy-terminal amino acids of MHV M (anti-M_c) (30) have been described earlier. The monoclonal antibody J1.3 against the amino terminus of MHV M (anti-M_N) (52) was kindly provided by J. Fleming.

Expression vectors and site-directed mutagenesis. All of the expression vectors used contain the genes under control of bacteriophage T7 transcription regulatory elements. Expression construct pTM5ab contains the MHV strain A59 open reading frames 5a and 5b, the latter coding for the E protein, in pTUG31 (56, 58). The construction of M genes coding for the mutant proteins C, (a+b) and (b+c) (30), A2A3 and 18 (11), and M-KK and 3AT5-KK (12) has been described before (Fig. 1). Also, the constructs encoding the VSV G protein (58) and the equine arteritis virus (EAV) hybrid protein M+9A have been described before (12). The latter protein has an insertion of 9 amino acids, corresponding to the MHV M amino-terminal sequence (residues S2 to P10), behind the initiating methionine of EAV M. The construct coding for the MHV M protein LT, which has a deletion of 5 amino acids (L108 to T112), was fortuitously obtained during the construction of the gene coding for the mutant M protein Sap1 (13). To make the M gene encoding the mutant protein RK, which lacks amino acids R188 through K207, pLITMUS38 (New England Biolabs) containing the gene coding for the M protein Sap (13) was digested with BssHII and StyI, treated with mung bean nuclease (Pharmacia), and religated. The construct was treated with BamHI, and the resulting fragment was cloned into expression vector pTUG3. In hybrid protein VGM, the amino-terminal ectodomain of MHV M was replaced by that of VSV G. In order to make the construct encoding this protein, an SstI restriction site was engineered in the MHV M gene by PCR mutagenesis using primers 891 (5'-GTTCAGAGCTCTAAGGAATGGAACTTCTCG-3') and 746 (5'-CGTCTAGATTAGGTTCTCAACAATGCGG-3'), corresponding to the region coding for the carboxy-terminal part of the ectodomain (and introducing the SstI restriction site) and the 3' end of the MHV M gene, respectively. The PCR product obtained was cloned into the pNOTA/T7 shuttle vector (5 prime3 prime, Inc.) and subsequently excised from the plasmid with BamHI and cloned into pTUG3, resulting in construct pTUG3MSacI. The fragment encoding the VSV G ectodomain was excised from pSV045R-ts (18) (a kind gift from J. K. Rose) by using XhoI and SstI and cloned into pTUG3MSacI treated with the same enzymes, resulting in expression construct pTUG3VGM. Plasmids S83 and S84 were a kind gift from S. Munro (38). Plasmid S83 encodes a human CD8 protein, in which the cytoplasmic tail has been replaced by four foreign amino acids (KRLK), while plasmid S84 encodes human CD8 protein which contains the CD8 cytoplasmic tail starting with these 4 amino acids. The sequence coding for KRLK contains an AflII site which facilitates the exchange of cytoplasmic tails. The expression cassettes of plasmids S83 and S84 were excised by using HindIII and XbaI and cloned into pNOTA/T7 treated with the same enzymes, resulting in expression vectors pNOTACD8tr and pNOTACD8, respectively. The construct encoding hybrid protein CD8Mc contains the sequence encoding the extracellular and transmembrane domains of CD8 followed by the MHV M cytoplasmic domain sequence starting with the codon for residue S105. In order to make this construct, an AflII restriction site was engineered in the MHV M gene by PCR mutagenesis using primers 586 (5'-GTATTTTCTTAAGAGCATTAGGTG-3') and 495 (5'-TTAGATTCTCAACAATGCGG-3'), corresponding to the region coding for the amino-terminal part of the cytoplasmic domain (and introducing the AflII site) and the 3' end of the MHV M gene, respectively. The PCR product obtained was cloned into the pNOTA/T7 vector and subsequently excised using AflII and XbaI and cloned into pNOTACD8tr treated with the same enzymes, resulting in pNOTACD8Mc. In hybrid protein CD8N, the amino-terminal ectodomain of MHV M was replaced by that of CD8. The region encoding the MHV M transmembrane and cytoplasmic domains was excised from pTZ19RMN (30) by using PvuII and BamHI and cloned into pNOTACD8tr treated with EcoRV and BamHI, resulting in pNOTACD8N. All constructs were verified by sequencing.

View larger version (29K): [in this window] [in a new window]   FIG. 1.   Overview of mutant M proteins. A schematic linear representation of the M protein, with its three transmembrane domains (a, b, and c) indicated, is shown on top. Mutant proteins with deletions in the transmembrane region [(a+b) and (b+c)], the amphipathic domain (LT, C, and RK), and the extreme carboxy terminus (18) are depicted at the top. Gaps represent deletions; the numbers indicate the deleted amino acids. Mutant proteins with amino acid substitutions in the amino terminus and/or carboxy terminus (A2A3, KK, and 3AT5-KK) are also depicted. The six amino-terminal and carboxy-terminal residues are shown. Below, the membrane structures of MHV M, EAV M+9A, VSV G, and CD8, as well as their chimeric forms, are drawn. The black lines represent amino acid sequences derived from MHV M; the oval symbolizes the amphipathic domain. The gray lines and symbols designate sequences derived from EAV M, VSV G, or CD8. The intracellular localization of the mutant proteins (Local.), their abilities to coimmunoprecipitate indicator M proteins (coIP), and their abilities to become incorporated into VLPs when coexpressed with M protein A2A3 (Rescue) are indicated at the upper right and bottom. Golgi, ER, and PM indicate localization of the proteins in the Golgi complex, in the ER, and in the plasma membrane, respectively (references 11, 12, and 13 and data not shown). The semiquantitative scores ++, +, +/, and  indicate efficient, moderately efficient, inefficient, and no coIP of the indicator proteins M-18 (for 3AT5-KK) and M-A2A3 (for the others). The semiquantitative scores +, +/, and  indicate efficient, inefficient, and no rescue of the M proteins into VLPs as determined by immunoisolation of intact VLPs. ND, not determined. Pulse-chase analysis demonstrated that the stabilities of all mutant M proteins were similar to that of WT M, with the exception of the M protein RK, which was slightly less stable.

Metabolic labeling and immunoprecipitation. Subconfluent monolayers of OST7-1 cells in 10-cm² tissue culture dishes were inoculated with vTF7-3 (t = 0 h) and subsequently transfected 1 h later with plasmid DNA by using lipofectin (Life Technologies) as described previously (11). At t = 2 h, the cells were placed at 32°C. At t = 4.5 h, the cells were washed with phosphate-buffered saline and starved for 30 min in cysteine- and methionine-free modified Eagle's medium containing 10 mM HEPES, pH 7.2, and 5% dialyzed fetal calf serum. The medium was then replaced by 600 µl of similar medium containing 100 µCi of ³⁵S in vitro cell-labeling mixture (Amersham), and the cells were labeled for the indicated time periods. In some experiments, the radioactivity was chased by incubating the cells with culture medium containing 2 mM methionine and 2 mM cysteine for 2 h. Proteins were immunoprecipitated from cell lysates as described before (40). Culture media were prepared for immunoprecipitation (IP) in the presence or absence of detergents by addition of 1/4 volume of five-times-concentrated lysis buffer or by addition of 2.5 volumes of TEN buffer consisting of 40 mM Tris-HCl (pH 7.6), 50 mM NaCl, and 1 mM EDTA, respectively. The immune complexes were adsorbed to Pansorbin cells (Calbiochem) for 30 min at 4°C and were subsequently collected by low-speed centrifugation. The pellets were washed three times by resuspension and centrifugation using 50 mM Tris-HCl (pH 8.0)-62.5 mM EDTA-0.5% Nonidet P-40-0.5% Na-deoxycholate or TEN buffer. The final pellets were suspended in electrophoresis sample buffer. The immunoprecipitates were analyzed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis in 15% polyacrylamide gels.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Demonstration of M-M interaction. The monoclonal antibody J1.3 is directed against the amino terminus of the MHV M protein (52). Fine mapping of the epitope recognized by this antibodydesignated anti-M_Nenabled us to develop a coimmunoprecipitation (coIP) assay for the detection of interactions between the M molecules. We found out recently (11) that recognition of this epitope by the antibody is critically dependent on the presence of the serine residues at positions 2 and 3. A mutant of the M protein, named A2A3, in which these residues have been replaced by alanines, was not recognized by the monoclonal antibody. This mutant protein otherwise behaved identically to the wild-type (WT) M protein in every aspect studied, including its ability to assemble VLPs (11, 12). The coIP assay is thus based on the coexpression of the A2A3 protein with mutant M proteins carrying an intact anti-M_N epitope: association of the proteins is monitored by the coprecipitation of A2A3 M molecules by the monoclonal antibody. The assay is demonstrated in Fig. 2A. In this experiment, the A2A3 M gene was either expressed alone or in combination with the gene encoding the carboxy-terminal deletion mutant M18 (Fig. 1) or with the gene encoding a control protein, the chimeric EAV protein M+9A (12). This protein consists of the EAV M protein extended at its extreme amino terminus by inserting the 9-residue amino-terminal sequence of MHV M (residues S2 to P10). As a result of this extension, the EAV protein acquired the epitope recognized by the MHV-specific antibody anti-M_N. The EAV M protein is a triple-membrane-spanning protein with a topology similar to that of the MHV M protein but is slightly smaller (15). The genes were expressed in OST7-1 cells by using the vTF7-3 expression system. The cells were labeled for 2 h with ³⁵S-labeled amino acids starting at 5 h postinfection. Cell lysates were prepared and subjected to IP with either an anti-MHV serum or the monoclonal antibody anti-M_N. Mutants A2A3 and 18 were well expressed both in the single expression and in the coexpressions, as was demonstrated by IP using the anti-MHV serum (Fig. 2A, lanes 1, 3, 5, and 7). The A2A3 protein appeared as the well-known set of O-glycosylated forms described before (29, 54), with the unglycosylated form (M₀) and the Golgi-modified form containing galactose and sialic acid (M₃) being the most prominent species. The M protein mutant 18 also becomes O glycosylated normally (11). Its M₀ form runs slightly faster in the gel than A2A3, while its M₃ form comigrates with the unglycosylated form (Fig. 2A, cf. lanes 1 and 5). Importantly, the mutant protein A2A3 was clearly not recognized by anti-M_N, in contrast to the M protein 18, as seen after single expression (Fig. 2A, lanes 2 and 6). Analysis of the lysate from cells expressing protein A2A3 and the M mutant 18 revealed the formation of M-M complexes. The anti-MHV serum precipitated both proteins A2A3 and 18 (lane 3). The monoclonal antibody anti-M_N not only precipitated protein M18 but also the glycosylated forms of A2A3 (lane 4). Analysis of the lysate from cells expressing protein A2A3 and the EAV M protein revealed the specificity of the assay. While the anti-MHV serum only precipitated protein A2A3 (lane 7), the monoclonal antibody only precipitated the EAV M+9A protein (lane 8). No coIP was observed. As another control for the specificity of the interactions measured, lysates of cells singly expressing the M protein mutants A2A3 and 18 were pooled and subsequently processed for IP using monoclonal anti-M_N. No coIP was observed from the pooled lysates (not shown).

View larger version (30K): [in this window] [in a new window]   FIG. 2.   Demonstration of M-M interaction. Genes coding for the mutant M proteins A2A3 and 18, the chimeric protein EAV M+9A, and the E protein were expressed in OST7-1 cells in various combinations, as indicated above each lane (, absent), by using the vTF7-3 expression system. For the plasmid encoding protein A2A3, 5 µg was transfected, while for the plasmid encoding the E protein, 1 µg was used (A and B); 5 µg of the plasmid encoding the M protein 18 was used for panel A, and 1 µg was used for panel B, while 3 µg of the plasmid encoding the chimeric protein EAV M+9A was used for panel A and 1 or 5 µg was used for panel B. Cells were labeled for 2 (A) or 3 (B) h. Cell lysates were prepared and subjected to IP with either the anti-MHV serum (MHV) or the monoclonal antibody to the amino terminus of M (MN). When the E protein was coexpressed (B), culture media were also collected and processed for IP or for affinity isolation of VLPs. The affinity isolations were performed by using the monoclonal antibody anti-MN in the absence of detergents. The positions of the different proteins are indicated at the left, while the molecular mass markers are at the right. Only the relevant parts of the gels are shown.

Mapping of M protein domains involved in homotypic interactions. The assays were subsequently used to investigate the involvement of different domains of the M molecule in M-M interactions. To this end, a number of M protein mutants were evaluated (Fig. 1). Mutant proteins (a+b) and (b+c) have a deletion of the first and second transmembrane domains and of the second and third transmembrane domains, respectively, resulting in M proteins with only the third or only the first transmembrane domain left. With their amino termini in the lumen and their carboxy termini in the cytoplasm, these proteins have the same membrane topology as WT M (30). When expressed, they appear mainly in an unglycosylated form, which is indicative of their inefficient transport out of the endoplasmic reticulum (ER), as we verified by immunofluorescence (not shown). The mutant proteins LT, C, and RK each lack a different part of the amphipathic domain which encompasses the region of the M molecule between the transmembrane cluster and the approximately 20-residue hydrophilic carboxy-terminal tail. The disposition of the amphipathic domain has not yet been resolved. While the M protein mutant LT does not become glycosylated and localizes to the ER, the mutant proteins C and RK acquire O-linked sugars, which is indicative of their transport to the Golgi complex (references 11, 12 and unpublished results). Furthermore, we also tested a mutant M protein with an ER retrieval signal (M-KK). This protein carries a cytoplasmic KKXX ER retrieval and retention signal (2, 24) which localizes it to the ER. While this protein can become O glycosylated under artificial conditions (e.g., during treatment with brefeldin A [BFA]), no trace of glycosylation can be detected in standard pulse-chase experiments even after 3 h of chase (12). Apparently, the protein is either retained very efficiently in the ER or rapidly retrieved from pre-Golgi compartments, where no O glycosylation takes place (12). The mutant protein 18 was used as a positive control. Importantly, all these mutant proteins were found to be deficient in VLP assembly when coexpressed with the E protein gene (reference 11 and unpublished data).

View larger version (52K): [in this window] [in a new window]   FIG. 3.   Mapping of M protein domains involved in homotypic interaction. The genes encoding the M protein A2A3 and the E protein were coexpressed together with different amounts of mutant M genes as described in the legend to Fig. 2. The different combinations are indicated above the gel, as are the amounts (in micrograms) of the mutant M plasmids transfected (, absent). Cells were labeled for 3 h. (A) The cells and culture media were collected separately and subjected to IP with anti-MHV (MHV) and anti-MN (MN) antibodies in the presence of detergents. The positions of the mutant M proteins in the gel are indicated by black squares at the right sides of the lanes. The positions of the M0 and M3 forms of protein A2A3 are also indicated. (B) In an otherwise-identical experiment, the culture media were analyzed in the absence of detergents to allow isolation of intact VLPs.

Replacement of the MHV M ectodomain. The MHV M protein contains a short (25 amino acids) ectodomain that is located in the lumens of intracellular organelles or on the outside of the virion. Mutations in this domain render the protein deficient in VLP assembly (11). In the present study, we evaluated whether replacement of the MHV M ectodomain by heterologous ectodomains affects M-M interactions. In the experiment shown in Fig. 4, we tested the hybrid protein VGM, which contains the ectodomain of the VSV G protein (Fig. 1). WT VSV G protein was included as a control. When coexpressed with the E protein, neither the hybrid protein nor the VSV G protein was productive in VLP assembly (not shown). It has been established that transport of VSV G protein out of the ER requires the formation of G protein trimers (27). Immunofluorescence analysis indicated the VGM hybrid protein to be located in the ER and the Golgi compartment (not shown). The oligomeric state of this protein was not studied.

View larger version (39K): [in this window] [in a new window]   FIG. 4.   Replacement of the MHV M ectodomain by the VSV G ectodomain. Genes encoding the M protein A2A3 and the E protein were expressed together or in combination with genes encoding VSV G or the hybrid protein VGM, as described in the legend to Fig. 2. The amounts of the plasmids encoding VGM and VSV G that were transfected are indicated (in micrograms; , absent). Cells were labeled for 3 h. The cells and media were collected separately and subjected to IP by using the anti-MHV serum (MHV) or the anti-VSV serum (VSV). When IP on the medium was performed with the anti-VSV serum, no detergents were added to allow affinity isolation of VLPs. The positions of the different proteins are indicated at the left, while the molecular mass markers are at the right.

View larger version (70K): [in this window] [in a new window]   FIG. 5.   Replacement of the MHV M ecto- and transmembrane domains by the corresponding domains of CD8. Genes encoding the M protein A2A3 and the E protein were expressed together or in combination with genes encoding either CD8, hybrid protein CD8Mc, or hybrid protein CD8N, as described in the legend to Fig. 2. The amounts of the CD8 constructs transfected are indicated (in micrograms; , absent). Cells were labeled for 3 h. The cells and media were collected separately and subjected to IP by using the anti-MHV serum (MHV) or a monoclonal antibody directed against CD8 (CD8). When IP on the medium was performed with the anti-CD8 antibodies, detergents were omitted to allow isolation of intact VLPs. The positions of the unglycosylated (CD8N and CD8Mc) forms and of the fully glycosylated (CD8N, CD8Mc, and CD8) forms of the CD8 (hybrid) proteins are indicated by black squares at the right of the lanes. The position of M protein A2A3 is also indicated. The molecular mass markers are at the right.

M proteins interact in pre-Golgi compartments. The results obtained with the ER-retained M proteins suggest that M proteins interact already in the ER. To further study the kinetics and first site of these interactions, we expressed the genes encoding the mutant M proteins A2A3 and 18 individually and together and carried out a pulse-chase experiment in which we labeled the cells for 15 min followed by a 105-min chase. As shown in Fig. 6A, both proteins were mainly present in their unglycosylated forms (M₀) immediately after the labeling (lanes 1 and 9), while during the chase the slower-migrating (Golgi-modified) M protein species appeared (lanes 3 and 11). The analysis of the lysates from cells coexpressing the two proteins revealed that little association had occurred during the labeling period, as hardly any of the unglycosylated (pre-Golgi) A2A3 protein appeared in the precipitate prepared with the anti-M_N antibody (lane 6). During the chase, a significant fraction of the proteins had associated, as illustrated by the efficient coIP of the glycosylated (Golgi-modified) A2A3 forms by the monoclonal antibody (lane 8).

View larger version (49K): [in this window] [in a new window]   FIG. 6.   M proteins interact in pre-Golgi compartments. Genes encoding the M proteins A2A3, 18, and 3AT5-KK were expressed as described in the legend to Fig. 2. The different combinations are indicated at the top. When indicated, BFA (6 µg/ml) was present from t = 3 h. Cells were pulse-labeled for 15 min (P) followed by a 105-min chase (C) (panel A) or were labeled for 2 h (panel B). The positions of the different proteins are indicated. , absent.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

The formation of progeny virions in coronavirus-infected cells involves two main processes, assembly of the helical nucleocapsids and of the viral envelopes. These processes are spatially separated, occurring in the cytoplasm and in intracellular membranes, respectively, and they apparently take place independently of each other. Obviously, the M protein is the key player in virion assembly, as it not only directs envelope formation but in addition provides the matrix to which the NC can attach for budding. The molecular interactions between the M molecules are most likely essential to the functioning of the M protein. Here we provide direct evidence for such interactions, which appear to occur through multiple contact sites and which generate a framework in the membrane from which foreign proteins are selectively excluded.

For the study of M-M interactions, we established two assays, both based on a mutant M protein named A2A3. The protein behaved like WT M in all relevant respects but was immunologically distinguishable due to the lack of an epitope caused by two subtle mutations. This property allowed its use as a reporter in the coIP assay after coexpression with other mutant M proteins. In the VLP incorporation assay, it enabled the sensitive detection of mutant M proteins coassembled into A2A3 protein-based particles. The two assays clearly demonstrated the existence of M-M interactions. In addition, the coprecipitation of relatively large amounts of reporter protein by only trace amounts of bait protein indicated the occurrence of large M protein complexes. This result confirmed our earlier observation obtained by sucrose-gradient analysis of singly expressed proteins (31). In these studies, we found that WT M protein accumulated in Golgi membranes in large heterogeneous complexes consisting of up to 40 M molecules. Somewhat smaller complexes appeared when a cytoplasmic tail truncation mutant was analyzed.

The involvement of the different domains of the M molecule in M-M interaction was investigated by evaluating several M protein mutants. Strikingly, mutant M proteins with deletions in the transmembrane domains, in the amphipathic domain, or in the carboxy-terminal hydrophilic tail or hybrid M proteins with heterologous ectodomains were still able to interact with the reporter M molecules, resulting in their incorporation into envelope particles. Only when all three transmembrane domains had been replaced by a heterologous transmembrane domain were interactions with A2A3 M protein and subsequent incorporation into particles severely reduced. Apparently, the M molecules interact with each other through multiple contact sites along the polypeptides. These sites may not be limited to the transmembrane region. For instance, while the replacement of the MHV M ectodomain by heterologous ectodomains hardly affected M-M interaction, involvement of this ectodomain in homotypic interactions cannot be excluded. Rather, such interactions have actually been demonstrated for the M protein of the human coronavirus 229E, where a cysteine residue in the short ectodomain gives rise to the formation of disulfide-linked homodimers (3). Thus, interactions (though generally noncovalent) between ectodomains may be a common feature of coronavirus M proteins.

An important conclusion from our observations is that M-M interactions are essential for coronavirus envelope assembly but that they are not sufficient. While the different mutant M proteins studied were each able to associate with the A2A3 protein, none of them was able by itself to assemble into particles when coexpressed with the E protein (reference 11 and data not shown). Obviously, additional requirements have to be met. We hypothesize that the full complement of interactions between the M molecules is required for efficient particle formation. Conceivably, the interactions at the various contact sites along the M polypeptides provide the free energy needed to generate and stabilize membrane curvature. In this respect, the E protein is unlikely to contribute significantly due to its numerical underrepresentation. In addition, the M proteins may need to interact with viral (E) and host proteins.

Associations between the M proteins appeared to take place in early (i.e., pre-Golgi) compartments. This is not surprising, since coronaviruses are assembled at the membranes of the IC (25, 28, 54). Consistently, the M protein has also been shown to engage in its interactions with the other viral membrane proteins (S and HE) in early compartments, most likely in the ER (13, 39, 40). The resulting higher-order complexes are thought to be maintained primarily by the M-M interactions (40). When expressed alone, the M protein accumulates in the Golgi compartment (25, 29, 46). However, coexpressed ER-retained mutant M proteins were found to interact with O-glycosylatedi.e., Golgi-modifiedM molecules. This implies that M proteins recycle from the Golgi complex back to early compartments. Recycling of Golgi-resident membrane proteins is not without precedent. Both Golgi-resident glycosyltransferases (50) and proteins equipped with Golgi-targeting signals (10) have been shown to recycle through the ER. Furthermore, an inhibitor of sphingolipid synthesis shifted the steady-state distribution of infectious bronchitis virus M protein from the Golgi complex to the ER, suggesting that this M protein is at least in part localized by retrieval mechanisms (34). Interesting as this recycling process may be by itself, relocation of M proteins back to the ER and IC is probably functionally important in coronavirus-infected cells. Retrograde transport of escaped M molecules offers these proteins another opportunity to become assembled into progeny virions. In addition, recycling may provide a clearance mechanism to prevent saturation of the Golgi system with M molecules and subsequent impaired passage of progeny virions on their way out of the cell.

Our studies with the ER-retained M protein mutants not only show that the M proteins interact in early compartments; the observation that ER-retained M proteins can be rescued into VLPs also indicates that VLP budding can occur in these compartments, as is the case for coronavirions. It is not yet clear which factors determine the site of budding. Neither WT M proteins nor M-S complexes are retained by themselves in the budding compartment (25, 29, 40). A good candidate for controlling the site of budding is the E protein. When this protein is expressed independently, it appears to accumulate in membranes of the ER and IC (43). Thus, through its interaction with the M protein in infected cells, the E protein might be able to retain the M and M-S complexes in these early compartments where the viral particles are formed.

Sorting of membrane proteins plays an important role in the assembly of virus envelopes. In coronaviruses the membrane proteins S and HE are specifically incorporated into the budding particle via lateral interactions with M proteins (13, 39, 40, 56). Foreign membrane proteins seem to be efficiently excluded. The effective segregation of the VSV G and the EAV M proteins from the budding VLP indeed indicates that envelope assembly is a very selective process. This process is, however, not perfect, as was apparent from the low but detectable level of incorporation of CD8 molecules into virus particles. Inclusion of foreign membrane proteins has been shown before when MHV pseudotypes containing the murine leukemia virus envelope determinants were observed after propagation of MHV in cells persistently infected with the leukemia virus (60). Other enveloped viruses exhibit different sorting stringencies, in keeping with their mechanisms of assembly. Thus, retrovirusesdependent for budding only on the Gag proteinare not very selective against foreign membrane proteins, allowing incorporation of substantial amounts of proteins of host and viral origin (19, 55). In contrast, the extensive and specific interactions between the spikes themselves and with the NC during the budding of alphaviruses seem to leave little room for other proteins to steal into particles (19, 51).

An intriguing question that remains is how the selectivity in the incorporation of membrane proteins in coronaviruses is realized. Our working hypothesis is that the M proteins form a molecular matrix, a geometric framework in which vacancies occur at regular positions. Budding does not require that these vacancies be filled by proteins, but the positions can be taken by the S and/or HE proteins via specific interactions with M. Foreign membrane proteins generally will not associate with the M protein and will thus not be taken into the matrix. Nonspecific incorporation may occur, though rarely, by accidental fit, as we observed for the CD8 protein. Interestingly, while the S protein occurs in the form of trimers (14), the HE protein is present in disulfide-linked homodimeric form (for a review, see reference 6). This suggests that there may actually be two types of vacancies, one for each oligomeric structure. How the E protein finds its way into the viral envelope is still enigmatic.

Like coronaviruses, hepadnaviruses and flaviviruses exhibit an NC-independent budding mechanism, which in these cases leads to formation of subviral particles. Interactions between the envelope proteins have also been demonstrated for these viruses. The hepadnavirus small envelope S protein, the single requirement for subviral particle formation (41, 49), forms disulfide-linked oligomers (23). It was found that secretion-deficient mutant S proteins can either retain secretion-competent S protein in the cell (36, 42) or be rescued into secreted particles (7). For flaviviruses, heterodimer formation between the envelope proteins E and prM is required to allow assembly and secretion of subviral particles (1). Clearly, we are only beginning to tackle the many fundamental questions regarding the generation of all these viruses.