Intramolecular Backfolding of the Carboxyl-terminal End of MxA Protein Is a Prerequisite for Its Oligomerization*

Mx proteins are large GTPases, which play a pivotal role in the interferon type I-mediated response against viral infections. The human MxA inhibits the replication of several RNA viruses and is organized in oligomeric structures. Using two different experimental approaches, the mammalian two-hybrid system and an interaction dependent nuclear translocation approach, three domains in the carboxyl-terminal moiety were identified that are involved in the oligomerization of MxA. The first consists of a carboxyl-terminal amphipathic helix (LZ1), which binds to a more proximal part of the same molecule. This intramolecular backfolding is a prerequisite for the formation of an intermolecular complex. This intermolecular interaction is mediated by two domains, a poorly defined region generated by the intramolecular interaction and a domain located between amino acids 363 and 415. Co-expression of wild-type MxA with various mutant fragments thereof revealed that the presence of the carboxyl-terminal region comprising the amphipathic helices LZ1 and LZ2 is necessary and sufficient to exert a dominant negative effect. This finding suggests that the functional interference of the carboxyl-terminal region is due to competition for binding of an as yet unidentified cellular or viral target molecules.

Mx proteins are interferon-induced GTPases, which are highly conserved among species and are present in many if not all vertebrates. Mx proteins belong to a subfamily of large GTPases that includes dynamin and VPS1 (1,2). Common features of these proteins are their high endogenous GTPase activity and their tendency to form oligomeric structures (3)(4)(5)(6)(7). Although highly conserved in their amino-terminal moiety comprising the GTP-binding domain, the various members of the family of dynamin-related GTPases appear to serve different cellular functions. Dynamin and VPS1 are involved in coated vesicle-mediated endocytosis and intracellular protein trafficking, respectively, while Mx proteins are effectors of the interferon type I-mediated antiviral function. In particular, human MxA, a cytoplasmic protein, shows intrinsic antiviral activity against various negative strand and at least one positive strand RNA virus (8 -16). The molecular mechanism of action and the viral target molecules remain largely unknown (for a review, see Refs. 17 and 18).
Mx proteins from interferon type I-treated cells or recombi-nant Mx from Escherichia coli have been shown to undergo extensive oligomerization (19 -23). In vitro, purified Mx1 is able to assume two distinct assembly states, which are interconvertible depending on the presence or absence of GTP (20). Similarly, dynamin self-assembles into ringlike structures, which are converted into helical stacks in the presence of GTP. These helical structures appear to enable dynamin to wrap around the neck of membrane vesicles and help pinch them off from the plasma membrane (4,5,24). For Mx proteins, however, the functional relevance of the oligomer formation and GTP-dependent assembly states remains elusive. Furthermore, several regions of Mx proteins have been proposed to be responsible for the oligomerization. These include a highly conserved region termed self-assembly motif in the aminoterminal moiety (20) as well as two amphipathic helices at the carboxyl-terminal end (19). Recently, Melen and Julkunen (22) reported that, in the case of the human MxB, the 71 carboxylterminal amino acids comprising the two conserved leucine repeats are responsible for its oligomerization (22). Using the mammalian two-hybrid transcription system and an interaction-dependent nuclear translocation approach (25), we show here that for MxA the proximal leucine repeat is necessary but not sufficient for oligomerization. Prerequisite for the oligomerization is the formation of an intramolecular interaction. In addition, we demonstrate that only MxA mutants containing both carboxyl-terminal leucine repeats are able to exert a dominant negative phenotype, strongly suggesting that these are required to interact with viral targets.

EXPERIMENTAL PROCEDURES
Construction of pQE30FMxA Plasmids-First, the sequence coding for the FLAG epitope N-AspTyrLysAspAspAspAspLys-C (26 -28) was inserted adjacent to the six-histidine tag of plasmid pQE30 (Qiagen GmbH, Hilden, Germany). The MxA cDNA fragments were then fused in frame at the 3Ј end of the FLAG sequence. Where necessary, the MxA cDNA fragments where joined to the FLAG sequence with a short linker. pQE30FMxA(L612K) and pQE30FMxA(L643K) were generated by site-directed mutagenesis of pQE30FMxA using the QuickChange™ mutagenesis kit (Stratagene, La Jolla, CA) according the manufacturer's protocol. Deletion of 4 amino acids (⌬81-84) in the first GTPbinding motif was achieved by digestion of plasmid pSP65TMxA (29,30) with SacI and subsequent re-ligation.
Construction of Expression Vectors Encoding GAL4-and VP16-MxA Fusion Proteins-The various MxA cDNA fragments were inserted in frame at the 3Ј end of the sequences coding for the GAL4 DNA binding domain (G) in pSG424 (31) and the transactivation domain of VP16 (V) in pAASV19NVP16 (32).
Construction of pQE16FMxA Mutants-For unknown reasons it was not possible to express MxA protein fragments containing the His tag as well as the FLAG-epitope in mammalian cell lines. To remove the His tag, the FMxA cDNA fragments of all pQE30FMxA constructs were excised and inserted into plasmid pQE16 (Qiagen GmbH). The carboxyl-terminal deletion mutant pQE16FMxA-(2-415) was generated by introducing a stop codon at position 416 of the amino acid sequence of MxA. For stable expression of the MxA-(577-662) carboxyl-terminal fragment, the mouse dihydrofolate reductase coding sequence present in pQE16 was excised and inserted between the sequences coding for the FLAG epitope and the MxA-(577-662) peptide.
Construction of pHMG-FMxA Mutants-To generate suitable constructs for the expression of the above described MxA variants in mouse 3T3 cells, the different MxA mutant fragments were released from the parental pQE16 vectors, blunted, and ligated into the unique EcoRV site of the eukaryotic expression vector pCL642 (33).
Transient Transfection of Swiss 3T3 Cells-The day before transfection, 1-3 ϫ 10 5 cells were seeded with 2 ml of complete growth medium into 35-mm culture dishes and incubated overnight at 37°C. Transfections were carried out with 1 g of plasmid DNA and 10 l of Unifectin-10 (kindly provided by A. Surovoy, University of Tü bingen, Tü bingen, Germany) according to the suggestion of the supplier. For the mammalian two-hybrid transcription system, 0.35 g each of a GMxA and a VMxA fusion protein expression vector were mixed with 0.35 g of the CAT 1 reporter plasmid pG5BCAT (34). The amount of DNA in solutions prepared for the transfection experiments was checked by running agarose gels. The cells were incubated at 37°C for 48 h prior to analysis.
CAT ELISA-Cytoplasmic extracts were prepared and the amount of CAT was determined in 50 g of total protein per sample with a CAT ELISA (Roche Molecular Biochemicals, Mannheim, Germany), which was carried out according to the protocol of the manufacturer.
For indirect double immunofluorescence analysis, infected cells were labeled with a mixture of monoclonal mouse anti-FLAG M2 antibody and of a polyclonal rabbit antiserum (dilution 1:500) directed against influenza A/Turkey virus. The staining was performed with a mixture of rhodamine isothiocyanate-conjugated goat anti-mouse (1:100) and fluorescein isothiocyanate-conjugated goat anti-rabbit (1:50) antibodies. Detection of immunostained cells was carried out with a confocal laser scanning microscope (Carl Zeiss, Oberkochen, Germany) equipped for data digitalization.
Virus and Infection Procedurs-The virus stock (6.8 ϫ 10 8 plaqueforming units/ml) of fowl plague influenza virus FPV-B, strain Bratislava (37), was prepared from supernatants of virus-infected Swiss 3T3 cells (8). The Swiss mouse 3T3 cell line was grown in Dulbecco's modified Eagle's medium containing 10% fetal calf serum. For immunofluorescence analyses of transfected 3T3 cells, 80%-confluent cell monolayers were infected for 1 h at room temperature with 5 infectious particles of FPV-B per cell, in medium containing 2% fetal calf serum and 20 mM HEPES, pH 7.3. The virus inoculum was removed by two washings with phosphate-buffered saline, and the cultures were incubated for 5 h at 37°C in Dulbecco's modified Eagle's medium containing 10% fetal calf serum.

RESULTS
The complete coding sequence of human MxA and various fragments thereof were cloned in frame downstream of the coding sequences of the GAL4 DNA-binding region (G, amino acids 1-147) and VP16 transactivating domain (V, amino acids 412-490) of the expression vectors pSG424 (31) and pAASVN19VP16 (32), respectively. Expression of all fusion proteins was verified by transient transfection into mouse Swiss 3T3 cells, followed by immunofluorescence analysis (data not shown).
The plasmids pGMxA and/or pVMxA were co-transfected with the CAT reporter plasmid pG5BCAT (34) into Swiss 3T3 cells. The cells were harvested 48 h after infection, and lysates were prepared. The amount of CAT enzyme present in the lysates was determined by ELISA. MxA-MxA interaction was easily detectable, yielding a high amount of CAT enzyme (on the average 110 pg of CAT/50 g of protein) comparable to the positive control consisting of the direct fusion of the V and G domains encoded by the plasmid pSGVP (126 pg of CAT/50 g of protein) (38). The MxA-MxA interaction was shown to be specific, since co-expression of the VP16 transactivation domain with GMxA fusion protein (plasmids pAASV19NVP16 and pGMxA) or the GAL4 DNA binding region with the VMxA fusion protein (plasmids pSG424 and pVMxA) did not lead to the activation of the CAT reporter gene (data not shown).
Oligomerization of MxA Is Mediated by the Carboxyl-terminal Half of MxA-In order to identify regions in MxA required for the interaction various mutants with progressive deletions from the amino-terminal or carboxyl-terminal end of MxA were constructed and tested for interaction. The interaction capacity of each MxA fragment was assessed by expressing it as GAL4 and VP16 fusion protein (exceptions are indicated in the text). The data shown in Figs. 1 to 3 represent the mean values of at least six independent transfection experiments and all CAT ELISAs were carried out in duplicate. We have observed little variation of the amount of CAT produced between individual transfection experiments using the same constructs (normally less than 10%). As shown in Fig. 1, the entire carboxyl-terminal moiety of MxA (MxA-(363-662)) is required for the MxA-MxA interaction. Deletion of 88 amino acids or more from the carboxyl-terminal end in both partners (MxA-(2-574)) led to the complete abrogation of the interaction while deletion of the entire amino-terminal half (MxA-(363-662)) had no effect. However, further deletions into the carboxyl-terminal moiety (MxA-(465-662)) destroyed the capability for interaction. These findings indicate that the MxA oligomerization domain(s) reside within the carboxyl-terminal half of the protein.
To test whether disproportionate expression of MxA frag- The carboxyl-terminal half of MxA mediates homotypic interaction. Swiss 3T3 cells were cotransfected with pG5CAT and pairwise combinations of plasmids expressing mutant and wild-type MxA (white boxes) fused to GAL4 DBD and to VP16 AD (data not shown). Cell lysates were tested in a CAT ELISA, and the amount of CAT enzyme measured (CAT SIGNAL) is given as a percentage of the wild-type MxA-MxA interaction, which is arbitrarily set to 100%. The data represent mean values of at least six independent transfection experiments carried out in duplicate. The amino acid sequence of the mutants is depicted in scale relative to MxA wild-type. ments which have the ability to self-assemble into complexes (MxA and MxA-(363-662)) would lead to a preferential interaction with itself thereby efficiently competing the interaction with other partners, we performed a series of transfection experiments where the amount of DNA of one interaction partner was kept constant and the amount of DNA of the other partner was increased up to 8-fold. Increasing amounts of DNA of one partner resulted in a decrease of the CAT signal (at maximum 70% reduction at an 8-fold excess of DNA of one partner). This decrease was independent of whether the interaction partner was able to self-assemble or not (data not shown). The most likely explanation of this finding is that sequestering of transcription factors (squelching) required for the expression of the MxA fragments rather than competition of MxA fragments for interaction is the cause of the observed decrease of the CAT signal.
The LZ1 Helix Interacts with a More Proximal Region Located between Amino Acids 363 and 574 -In order to further define the localization of the interaction domain(s) of MxA, amino-terminal deletion mutants were pairwise co-expressed with the corresponding carboxyl-terminal deletion mutants and assessed for their capability to transactivate the CAT reporter gene (Fig. 2). As expected, MxA-(2-362) was not able to interact with MxA-(363-662). The same result was obtained with MxA-(2-465) and MxA-(465-662). Remarkably, MxA-(2-574) and MxA-(574 -662) clearly were able to interact with each other (Fig. 2), indicating the existence of two distinct interaction domains within the carboxyl-terminal half of MxA, one residing within the 88 carboxyl-terminal amino acids, the other between amino acid positions 363 and 574.
In this context it is interesting to note that the two leucine zipper motifs LZ1 and LZ2, originally described by Melen and co-workers (19), are located within the 88 carboxyl-terminal amino acids, making them prime candidates for the interaction domain. To test this possibility we introduced a mutation into the constructs pGMxA-(574 -662) and pVMxA-(574 -662) by site-directed mutagenesis leading to the replacement of the leucine residue at position 612 to a lysine residue. This mutation destroys the amphipathic character of the helix LZ1 without affecting the overall ␣-helical structure. Indeed, MxA-(574 -662, L612K) was not able to bind to MxA-(2-574) (Fig. 2), lending strong support to the hypothesis that the LZ1 helix constitutes the carboxyl-terminal interaction domain.
In an attempt to better define the localization of the second interaction domain, MxA-(2-465) was co-expressed with MxA-(574 -662). The fact that the MxA-(2-465) was not capable to form a complex with MxA-(574 -662) suggests that the region between amino acid positions 465 and 574 may be important for the interaction (Fig. 2). Since two amphipathic helices located between amino acids 423 and 470 and between 495 and 522 are present within this region, a mutation was introduced into each of them by site-directed mutagenesis to destroy their amphipathic character. The resulting constructs pGMxA-(2-574, V458K) and pGMxA-(2-574, I509K) encode GAL4-MxA fusion proteins where the valine residue at position 458 or the isoleucine residue at position 509 was replaced by a lysine residue. However, co-expression of these proteins with VMxA-(574 -662) led only to slight reductions of the CAT enzyme transactivation, reaching 49% and 90%, respectively, of the value for MxA-MxA interaction (Fig. 2).
Intramolecular Interaction of MxA Is a Prerequisite for the Intermolecular Complex Formation-So far, the results obtained can be explained by an intermolecular interaction of two distinct domains leading to the dimerization or oligomerization of the protein as suggested by Schumacher and Stä heli (23). However, the data do not exclude the possibility that an intramolecular interaction between the LZ1 motif and the second interaction domain can occur. To examine whether MxA molecules engage in an intramolecular interaction, MxA variants comprising both interaction domains were co-expressed with various MxA fragments containing only one interaction domain.
As expected, MxA-(363-577) lacking the last 88 amino acids showed a strong interaction with the fragment MxA-(574 -662), which comprised the 88 carboxyl-terminal amino acids. However, MxA-(362-662) interacted neither with MxA-(574 -662) nor with MxA-(465-662) (Fig. 3A). Only co-expression of GMxA-(363-662) with VMxA-(363-662) resulted in the transactivation of the CAT reporter gene. Very similar results were obtained when MxA-(363-662) was replaced by the full-length MxA (Fig. 3A). These results clearly demonstrate that an MxA fragment containing only the LZ1 motif is not able to form a complex with an MxA fragment containing both interaction domains. However, when the leucine to lysine substitution at position 612 in the LZ1 motif was introduced into the fulllength protein (MxA(L612K)), interaction with MxA-(574 -662) was restored (Fig. 3A), demonstrating that the intermolecular interaction of LZ1 with the second interacting domain was efficiently blocked by a competing intramolecular interaction.
In a second set of experiments, we assessed the capability of full-length MxA to interact with MxA fragments lacking increasing portions of the carboxyl-terminal region (Fig. 3B). (MxA-(2-415)) amino acids from the carboxyl-terminal end still led to the transactivation of the CAT enzyme (65%, 97% and 85%, respectively) when compared with the full-length MxA-MxA interaction (100%). Only when an additional 53 amino acids from the carboxyl terminus were deleted (MxA-(2-362)) was the interaction destroyed (Fig. 3B). As expected, the fragment MxA-(363-577) was able to form a complex with fulllength MxA corroborating the previous results. Furthermore, the pivotal role of the LZ1 region in the intermolecular interaction was underscored by the results obtained by co-expres- protein-protein interaction. Only when this internal complex is formed can an intermolecular interaction with a region located between amino acids 362 and 415 of MxA occur. A prediction from this model is that the mutant MxA(L612K) would still be able to interact with MxA through its intact intermolecular interaction domain but would not be able to form a complex with itself because the intramolecular interaction cannot be formed by either partner molecule. As shown in Fig. 3C, this is indeed the case. Upon co-expression of MxA with MxA(L612K), high levels of CAT enzyme were detected (70%) whereas GMxA(L612K) did not interact with VMxA(L612K). Since MxA(L612K) is not able to interact with itself it is not able to form homodimers or oligomers and should therefore only exist as monomers. In an attempt to disrupt the intermolecular interaction domain located between amino acids 362 and 415, we introduced a leucine to lysine mutation at position 389 of MxA. This mutation interrupts the amphipathic character of an ␣-helix located between amino acids 371 and 398. However, this mutation had no effect on the capacity of this mutant to form a complex with itself or with wild-type MxA (Fig. 3C).
The Region between Amino Acids 362 and 415 Is Required for the Intermolecular Interaction-In order to further investigate the intermolecular MxA interaction, the nuclear translocation assay originally described by Ponten and co-workers (25) was used. This approach entails co-expression of TMxA, an MxA variant containing the nuclear localization signal of the SV40 large T antigen at its amino terminus (30), with FMxA mutants in mammalian cells. Since TMxA accumulates exclusively in the cell nucleus (30), cytoplasmic MxA mutants able to interact with TMxA will also be dragged to the nucleus. A series of plasmids were constructed that directed the expression of MxA protein moieties carrying the FLAG epitope (26) at their amino terminus (Table II). In order to stabilize the expression of MxA-(577-662), the coding sequence of the mouse dihydrofolate reductase (D) was cloned in frame between the FLAG tag and the MxA coding sequence. The resulting protein was designated FDMxA-(577-662). As eukaryotic expression vector, the plasmid pCL642 was chosen, which contains the upstream regulatory region of the mouse 3-hydroxy-3-methylglutaryl coenzyme A reductase gene (33).
To assess whether a given MxA protein fragment was able to interact with TMxA, the corresponding pHMG-FMxA plasmids were transfected with or without pHMG-TMxA into Swiss mouse 3T3 cells. After 48 h the subcellular localization of the FMxA protein fragment was determined by immunofluorescence analysis with a monoclonal antibody directed against the FLAG epitope. As expected, FMxA was able to form a complex with TMxA leading to its translocation to the nucleus. The immunofluorescence analysis for FMxA showed the typical punctate nuclear staining pattern for TMxA when expressed alone as a control (Fig. 4). Similarly, FMxA-(⌬81-84), a GTPbinding mutant lacking four amino acids in the first GTPbinding consensus element (39), was translocated to the nucleus. When FMxA-(2-576) and FMxA-(2-415) lacking the 86 and 247 carboxyl-terminal amino acids, respectively, were coexpressed with TMxA, they also accumulated predominantly in the nucleus, which is indicative for a tight association with TMxA. In contrast, FMxA-(2-362) lacking 300 carboxyl-terminal amino acids remained in the cytoplasm and was therefore not capable to form a complex with TMxA. These results are in complete agreement with the data generated by the mammalian two-hybrid system, corroborating the conclusion that the region between amino acid positions 362 and 415 is critical for the intermolecular interaction of MxA. This was further supported by the results obtained with FMxA-(362-662), FMxA-(362-626), and FMxA-(362-576), which showed that these pro-teins accumulated in the nucleus when co-expressed with TMxA (Fig. 4). In contrast, FDMxA-(577-662) was not able to interact with TMxA, which was expected, since the LZ1 domain can only interact with the second intramolecular interaction domain in trans when the LZ1 domain of the binding partner has either been deleted or mutated (Table II).
We also tested two MxA mutants carrying single amino acid substitutions, namely FMxA(L612K) and FMxA(L643K), for their ability to interact with wild type MxA. The exchange of the leucine residue at amino acid position 643 with a lysine residue in MxA(L643K) alters the distribution of charges in the LZ2 helix, which has been previously shown to be important for virus specificity (30). Moreover, the same amino acid substitution at the corresponding position in mouse Mx1 (Mx1(L612K)) completely abrogated its antiviral activity against influenza A virus (40). When FMxA(L612K) and FMxA(L643K) were coexpressed with TMxA, they were efficiently translocated to the nucleus, indicating that the mutations did not affect their capability to interact with wild-type MxA. This finding is in agreement with all previous observations of this study and was expected since the two mutants contain an intact intermolecular interaction domain (amino acids 362-415).
The 88 Carboxyl-terminal Amino Acids of MxA Are Sufficient to Confer a Dominant Negative Effect-We next tested whether expression of distinct MxA protein fragments would lead to an interference with the function of wild type MxA. For this purpose, cells of a stably transfected Swiss mouse 3T3-MxA clone were transiently transfected with MxA mutant proteins and subsequently assayed for their sensitivity to influenza A virus by double-immunofluorescence analysis. The presence of viral antigens was examined with a rabbit polyclonal anti-influenza A virus serum and the FMxA mutants were detected using a mouse monoclonal anti-FLAG antibody. To assess whether a given MxA mutant exerted a dominant negative effect, the percentage of mutant-expressing cells that were infected with influenza virus was determined (Table I).
The parental 3T3-MxA cell line which expresses physiological levels of MxA in virtually 100% of the cells is resistant to infection with influenza A virus (8). As expected, transfection of pHMG-FMxA directing the expression of wild-type FMxA had no influence on the antiviral activity of the endogenous MxA (9% infected cells). Similarly, the mutants FMxA-(2-576), FMxA-(2-363), FMxA-(363-626), and FMxA-(362-576) were not able to interfere with the function of wild-type MxA (Table  I). By contrast, the mutant proteins FMxA-(⌬81-84), FMxA-(362-662), and FDMxA-(577-662) exhibited a dominant-negative phenotype and reversed at least partially the antiviral block of MxA (79%, 63%, and 37% infected cells, respectively). In the case of FDMxA-(577-662), the lower percentage of infected cells is most likely due to its low expression level in the transfected 3T3-MxA cells. The common denominator of these proteins is that they comprise the carboxyl-terminal region, which includes the amphipathic helices LZ1 and LZ2. Intriguingly, the two mutants FMxA-(362-662, L612K), and FMxA-(362-662, L643K) did not exert a dominant negative effect upon MxA (12% and 17% infected cells, respectively) strongly suggesting that both the LZ1 and LZ2 helices have to be intact in order for the carboxyl-terminal region to exert a dominant negative effect. DISCUSSION One of the common features of Mx proteins is their tendency to form large aggregates in vitro and in vivo (19 -21). Although many efforts have been undertaken to define the regions that direct the oligomer formation, this issue is still highly controversial. Furthermore, the functional role of Mx oligomers in the antiviral activity remains elusive. One group reported the ami-no-terminal self-assembly domain of Mx proteins to be important for their oligomeric structure (20), while others favor the involvement of the carboxyl-terminal moiety in such homotypic interactions (19,21,22,25). In this study, we show that the carboxyl-terminal moiety of MxA (amino acids 363-662) con-tains all the structural components necessary to perform homotypic MxA-MxA interactions. Moreover, taking advantage of the mammalian two-hybrid system, we were able to identify three domains that appear to play an important role in oligomerization: an amphipathic helix designated LZ1 located at the carboxyl terminus (amino acids 595-622); a second, less well defined domain located between amino acids 363 and 574; and a third region located between amino acids 363 and 415 required for the intermolecular interaction.
First, a region comprising the 88 carboxyl-terminal amino acids (amino acids 574 -662) was found to be required for the interaction. This region contains two leucine zipper motifs, LZ1 and LZ2, located between amino acids 595 and 622 and between 640 and 660, respectively, which had been previously reported to be responsible for the formation of oligomeric structures (19). However, we observed that, although this carboxylterminal region (amino acids 574 -662) was necessary for a homotypic interaction, it was not sufficient. The carboxyl-terminal region was not able to form a complex with itself but engaged in an interaction with a more proximal region located between amino acids 362 and 574. These findings are in line with results obtained by Schwemmle and co-workers (21), who demonstrated that the 98 carboxyl-terminal amino acids, resulting from a mild proteinase K treatment of MxA, formed a Wild-type and all mutant forms of MxA proteins contain a FLAG tag at the amino terminus, except for TMxA, whose amino terminus carries the nuclear localization signal. The FLAG-tagged protein encoding plasmids were transfected into Swiss 3T3 cells either alone or together with pHMG-TMxA, and their subcellular distribution was monitored by indirect immunofluorescence analysis using the mouse monoclonal antibody M2, specific for the FLAG epitope. The nuclear localization of TMxA alone (control) was verified using the monoclonal mouse anti-MxA antibody 5-56. We also attempted to better define the more proximal region that binds to LZ1. Data obtained with the mammalian twohybrid system revealed that the amino-terminal mutant MxA-(2-465) was not able to interact with the carboxyl-terminal mutants MxA-(465-662) or MxA-(574 -662). This suggests but does not prove that the second interaction domain is located between amino acids 465 and 574. It is also possible that this proximal binding region is located around position 465 and might be therefore destroyed in the mutants MxA-(2-465) or MxA-(465-662).
So far, the results obtained could be explained by two different models: (a) an intermolecular head to tail interaction between MxA molecules or (b) an intramolecular interaction through a back-folding of the carboxyl-terminal region onto the body of the protein. Indeed, the carboxyl-terminal region was not able to compete for binding to the second interaction domain in trans when the carboxyl-teminal region was also provided in cis. Even co-transfection of an 8-fold excess of DNA encoding MxA-(574 -662) over DNA encoding full-length MxA did not lead to a detectable interaction of these two molecules (data not shown). Only when the intramolecular interaction was destroyed by mutation of LZ1 in the context of the fulllength MxA protein or the carboxyl-terminal half (MxA-(362-662)) was interaction in trans observed. These data clearly indicate that the intramolecular interaction of the proximal interaction domain with LZ1 was favored over the same interaction occurring in trans, making a direct involvement of the COOH-terminal region in the intermolecular interaction highly unlikely.
Therefore, if the two identified interaction domains were engaged in an intramolecular interaction, a third interaction domain has to be postulated, which could be responsible for the oligomerization of Mx proteins. Indeed, such an intermolecular interaction domain, located between amino acids 362 and 415 of MxA, was identified. All MxA mutants containing this region did form a complex with full-length MxA. A prerequisite for this interaction is, however, that at least one of the binding partners is able to form the intramolecular interaction.
These data strongly suggest that only when the carboxylterminal region of MxA containing the LZ1 helix (region 1) folds back on its proximal interaction domain (amino acids 363-574, region 2) the MxA molecule becomes competent to engage in a homotypic interaction with a second MxA molecule. This association is mediated by the intermolecular interaction domain (amino acids 363-415, region 3) of the second MxA molecule. Whether the proximal intramolecular interaction domain of MxA is directly involved in this intermolecular interaction remains to be determined. Since the intermolecular interaction domain (region 3) of the first MxA molecule is not involved in this dimer formation, interaction with a third MxA molecule and hence oligomer formation is possible (Fig. 5). This model also conforms with data obtained from parallel experiments with mouse Mx1 using the same mammalian two-hybrid system. 2 The results of the dissection of murine Mx1 protein revealed a similar organization of the structural elements that were of relevance for the Mx1 oligomerization, suggesting that in both proteins oligomerization may arise by very similar mechanisms.
Schumacher and Stä heli carried out a very similar study employing the yeast two-hybrid system to identify domains mediating the oligomerization of MxA (23). They came to the conclusion that the carboxyl-terminal region can fold back on a internal more proximal region of the same molecule. Taking advantage of our MxA(L612K) construct, they also observed that residue L612 is critical for this interaction. However, they suggest that the carboxyl-terminal region is directly involved in oligomerization by interacting with the internal domain of 2 H. P. Hefti, unpublished data.  other MxA molecules. This conclusion is not supported by our data, which predict that an intramolecular backfolding is a prerequisite for the formation of oligomers which are mediated by a domain between amino acids 362-415. The apparent discrepancy stems from the fact that Schumacher and Stä heli fail to see an interaction between the full-length MxA with the fragment MxA-(372-599) with the yeast two-hybrid system, while we clearly observe interaction of MxA with all fragments comprising the region between amino acids 362 and 415. It is therefore highly likely that the amino acids located between position 362 and 372 are critical for the oligomerization.
In the case of human MxB, oligomerization appears to be mediated by a different mechanism. Melen and co-workers (22) have recently shown that hetero-oligomerization between the cytoplasmic 76-kDa MxB protein and the nuclear 78-kDa variant of MxB containing the amino-terminal nuclear localization signal was directly mediated by the corresponding 71 carboxylterminal amino acids, which include the leucine repeats LZ1 and LZ2.
All available evidence indicates that MxA(L612K) exists in the form of monomers in the cytoplasm since it is not able to form homo-oligomeric structures. This is also evident from immunofluorescence analysis in transfected 3T3 cells, where a diffuse instead of a punctate staining pattern is observed. This mutant MxA protein provides us with a very important tool to evaluate the functional role of Mx proteins in their monomeric form. Preliminary evidence indicates that MxA(L612K) exhibits antiviral activity against influenza A virus despite the absence of GTP-hydrolyzing activity. 3 It has been previously demonstrated by in vitro studies that the GTP binding activity of Mx1 or MxA protein is sufficient to exert antiviral activity (12).
The putative role of the oligomerization of Mx proteins still remains elusive. It is certainly intriguing that the Mx homolog dynamin, which is involved in the endocytosis of synaptic vesicles, also forms intramolecular interactions that are involved in self-assembly of this protein. Moreover, intramolecular complex formation appears to be important for regulating the dynamin GTPase activity (41). So far, however, there is no experimental evidence for Mx proteins to be also functionally related to the dynamin family.
We also analyzed whether MxA mutant proteins lacking GTP binding and antiviral activity can exert a dominant negative effect on wild-type MxA. Dominant interfering effects can be explained either by a direct interaction of an inactive mutant with its functional counterpart or by an interaction of the inactive mutant with substrates or target molecules of the wild-type form. The results presented in this study clearly show that the carboxyl-terminal region (amino acids 576 -662) comprising the LZ1 and LZ2 motifs is both necessary and sufficient to interfere with the function of wild-type MxA protein. MxA mutants that were still capable of binding to MxA but that lacked the carboxyl-terminal moiety were neutral. MxA protein fragments containing only one of the two carboxylterminal amphipathic helices (LZ1 or LZ2) in an intact form also were not able to exert a dominant-negative effect. These findings answered two important questions. First, the formation of hetero-oligomers between wild-type MxA and inactive mutants per se was not sufficient to generate a dominantnegative effect. This finding is consistent with the results obtained by Ponten and co-workers showing that, although the MxA fragment MxA-(359 -572) is able to interact with wild type MxA, it cannot exert a dominant negative effect (25).
Moreover, it would fit to our model predicting that oligomeric structures constitute inactive reservoirs of Mx proteins. Second, the dominant negative effect is apparently mediated by an interaction of the carboxyl-terminal moiety with as yet unidentified viral or cellular target proteins. This study clearly demonstrates that the presence of both leucine repeats LZ1 and LZ2 is necessary for the functional interference to occur.
In order to better understand the molecular mechanism of the antiviral function of Mx proteins, it will be instrumental to further characterize the role of these two amphipathic helices and to identify their targets. Finally, a structural analysis is necessary to reliably assign the roles of the various segments of the MxA protein in the intra-and intermolecular interactions.