The ALG-2-interacting Protein Alix Associates with CHMP4b, a Human Homologue of Yeast Snf7 That Is Involved in Multivesicular Body Sorting*

Alix (ALG-2-interacting protein X) is a 95-kDa protein that interacts with an EF-hand type Ca2+-binding protein, ALG-2 (apoptosis-linked gene 2), through its C-terminal proline-rich region. In this study, we searched for proteins that interact with human AlixΔC (a truncated form not containing the C-terminal region) by using a yeast two-hybrid screen, and we identified two similar human proteins, CHMP4a and CHMP4b (chromatin-modifying protein; charged multivesicular body protein), as novel binding partners of Alix. The interaction of Alix with CHMP4b was confirmed by a glutathione S-transferase pull-down assay and by co-immunoprecipitation experiments. Fluorescence microscopic analysis revealed that CHMP4b transiently expressed in HeLa cells mainly exhibited a punctate distribution in the perinuclear area and co-localized with co-expressed Alix. The distribution of CHMP4b partly overlapped the distributions of early and late endosomal marker proteins, EEA1 (early endosome antigen 1) and Lamp-1 (lysosomal membrane protein-1), respectively. Transient overexpression of CHMP4b induced the accumulation of ubiquitinated proteins as punctate patterns that were partly overlapped with the distribution of CHMP4b and inhibited the disappearance of endocytosed epidermal growth factor. In contrast, stably expressed CHMP4b in HEK293 cells was observed diffusely in the cytoplasm. Transient overexpression of AlixΔC in stably CHMP4b-expressing cells, however, induced formation of vesicle-like structures in which CHMP4b and AlixΔC were co-localized. SKD1E235Q, a dominant negative form of the AAA type ATPase SKD1 that plays critical roles in the endocytic pathway, was co-immunoprecipitated with CHMP4b. Furthermore, CHMP4b co-localized with SKD1E235Q as punctate patterns in the perinuclear area, and Alix was induced to exhibit dot-like distributions overlapped with SKD1E235Q in HeLa cells. These results suggest that CHMP4b and Alix participate in formation of multivesicular bodies by cooperating with SKD1.

Alix (ALG-2-interacting protein X; also named AIP1) was found as a protein that interacts with the Ca 2ϩ -binding protein ALG-2 (apoptosis-linked gene 2) (1) by the yeast two-hybrid method using mouse cDNA libraries by two independent groups (2,3). The interaction between Alix and ALG-2 was shown to be Ca 2ϩ -dependent. Human Alix is a 95-kDa protein that consists of 868 amino acids. It possesses no obvious structural motifs except for coiled-coil regions and a long prolinerich region on the C-terminal side. ALG-2 is one of the penta-EF-hand Ca 2ϩ -binding proteins (4,5) and interacts with the N-terminal domains of annexins VII and XI, which show some similarities with the proline-rich region of Alix (6,7). ALG-2 forms a homodimer as well as a heterodimer with another penta-EF-hand protein named peflin (8,9). In a subcellular fractionation experiment, most of Alix was recovered in the cytosolic fraction of fibroblast cells (3). Results of immunofluorescent microscopic analysis showed that overexpressed Alix was present in the cytoplasm but was concentrated in the perinuclear region and also localized at the cell periphery in lamellipodia and filopodia (10). Moreover, proteomic analyses have revealed that Alix is also present in extracellular microvesicles called exosomes, which are derived from dendritic cells (11), and phagosomes isolated from macrophage-like cells (12).
Homologues of Alix have been found in many organisms. The Xenopus homologue of Alix, Xp95, was identified as a protein of 95 kDa that was phosphorylated from the first to second meiotic divisions during progesterone-induced oocyte maturation (13). Overexpression of Hp95, a human homologue of Xp95 identical to human Alix, induced G 1 arrest, promoted detachment-induced apoptosis (anoikis), and reduced tumorigenicity in HeLa cells (14,15). PalA, an Aspergillus nidulans Alix homologue, is required for alkaline adaptation of the filamentous fungi whose pH regulation of gene expression is mediated by the zinc finger transcription factor PacC (16,17). Budding yeast has two Alix homologues, Rim20p (18) and Bro1 (19). Genetic analysis of Rim20p has revealed that Rim20p is essential for processing of the zinc finger protein Rim101p, a yeast homologue of PacC (18). Bro1 mutations result in a temperature-sensitive osmoremedial growth defect (19). Although their functions are not known, Alix homologues have also been found in Caenorhabditis elegans (named YNK1 (20)) and in Arabidopsis (GenBank TM accession number AC007591, protein ID AAD3964201).
In addition to ALG-2, some binding partners of Alix have been identified. They are the glioma-associated protein SETA (21), which is one of the isoforms of CIN85/Ruk (22,23), and * This work was supported by a Grant-in-Aid for Scientific Research B (to M. M.) and a Grant-in-Aid for Young Scientists B (to H. S.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM /EBI Data Bank with accession number(s) AB100261 and AB100262 (at the early stage of this study, we designated CHMP4a and CHMP4b as Shax2 and Shax1, respectively).
ʈ To whom correspondence should be addressed. Tel.: 81-52-789-4088; Fax: 81-52-789-5542; E-mail: mmaki@agr.nagoya-u.ac.jp. endophilins (10). These proteins, including ALG-2, associate with the proline-rich region in Alix by Src homology 3 domains in the case of SETA and endophilins or by a penta-EF-hand domain in the case of ALG-2. Conservation among the homologues, however, is lower within their C-terminal proline-rich regions, and yeast Rim20p lacks this region. In contrast, the N-terminal parts of Alix homologues show strong similarities, particularly in the regions containing a consensus site of potential tyrosine phosphorylation by Src family tyrosine kinases. We speculated that the N-terminal region of Alix plays an important role in functions common to all Alix homologues.
In this study, we screened for proteins binding to the Nterminal region of human Alix by the yeast two-hybrid method, and we isolated two clones encoding proteins that are similar to each other. One is identical to HSPC134/CHMP4 (chromatinmodifying protein; charged multivesicular body protein) (24,25), which is designated as CHMP4a in this paper, and the other is a novel protein designated as CHMP4b. These CHMP4 proteins are highly homologous to yeast Snf7/Vps32 (vacuolar protein sorting) (26,27). Yeast Snf7/Vps32 is a member of class E Vps proteins, which are required for sorting of membrane proteins into inner vesicles that originate by inward invagination in the lumen of the late endosome (multivesicular body (MVB) 1 ) (28). Subcellular localization of CHMP4b and the effect of CHMP4b overexpression were analyzed, and the obtained results suggest that Alix and CHMP4b participate in the endosomal sorting pathway.
Plasmid Constructions-Human Alix cDNA was cloned from Human Fetus Marathon-Ready TM cDNA (Clontech) by the PCR method and inserted into a pCR2.1-TOPO vector (Invitrogen). pCR2.1-TOPO-Alix Y319F, which has a point mutation at amino acid 319 (from tyrosine to phenylalanine), was created by PCR-based site-directed mutagenesis according to the provided instruction for QuickChange site-directed mutagenesis kit from Stratagene using two complementary primers (5Ј-AATGACTTCATTTTTCATGATCGAGTT-3Ј and 5Ј-AACTCGAT-CATGAAAAATGAAGTCATT-3Ј), and the mutation was confirmed by DNA sequencing. To obtain the cDNA fragment encoding amino acids 1-660 (Alix⌬C), the palindromic universal translational terminator, 5Ј-GCTTAATTAATTAAGC-3Ј, was inserted into the blunt-ended NdeI site of the Alix cDNA. A cDNA fragment corresponding to Alix⌬C was inserted into the yeast expression vector pGBKT7 (Clontech), the glutathione S-transferase (GST) fusion vector pGEX-4T-2 (Amersham Biosciences), and a mammalian expression vector pcDNA6/V5-His (Invitrogen) to construct pGBKT7-Alix⌬C, pGST-Alix⌬C, and pAlix⌬C-V5, respectively. Using the SmaI site located upstream of the stop codon in the Alix cDNA, a cDNA fragment corresponding to amino acid residues 1-847 was inserted into pcDNA6/V5-His to construct pAlix-V5. To generate truncated forms of GFP-Alix fusion proteins, various Alix cDNA fragments obtained by digestion with restriction enzymes were inserted into pEGFP-C2 or pEGFP-C3 (Clontech) vectors. To delete the putative 5Ј-untranslated region of CHMP4b by the PCR method, the forward primer containing a translational initiation codon (underlined), 5Ј-GCGAATTCCATGTCGGTGTTCGGGAAG-3Ј, and the reverse primer containing the stop codon (underlined), 5Ј-GCCTCGAGCCATT-ACATGGATCCAGCCCA-3Ј, were designed on the basis of the CHMP4b cDNA sequence. The PCR product was cloned into a pKF3 (TAKARA Shuzo, Kyoto, Japan). To create a pCMV-3ϫFLAG vector for eukaryotic expression, forward oligonucleotides, 5Ј-GATCGACTACAAAGACCAT-GACATCGATTATAAGGATGACGACGAT-3Ј and 5Ј-GGGATCCCTGC-A-3Ј, and reverse oligonucleotides, 5Ј-TCATCCTTATAATCGATGTCA-TGGTCTTTGTAGTC-3Ј and 5Ј-GGGATCCCATCG-3Ј, were inserted between BamHI and PstI sites of the pCMV-Tag2C vector (Stratagene). A fragment of the full-length CHMP4b cDNA was inserted into a pCMV-3ϫFLAG vector and a pEGFP-N1 vector (Clontech), and the resultant plasmids, designated pFLAG-CHMP4b and pCHMP4b-GFP, were used to express 3ϫFLAG-tagged protein and GFP fusion protein, respectively. The plasmid encoding GFP-LC3 protein was constructed as described previously (30). A Myc-SKD1 E235Q fragment from pEGFP-MSKD1 EQ (31) was inserted into the EcoRI site of a pCI-neo vector (Promega) to generate pMyc-SKD1 E235Q .
Yeast Two-hybrid Screening-The MATCHMAKER two-hybrid system, including yeast strains and vectors, was obtained from Clontech, and the screening was performed as described previously (6). Positive clones were sequenced using an automated fluorescent sequencer, ABI PRISM 310 (PE Applied Biosystems). A computer homology search using the advanced BLAST program and coiled-coil region prediction were performed on the World Wide Web. Phylogenetic analysis using Clustal X was performed as described previously (32).
Cell Culture and Transfection-HEK293 and HeLa cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum, penicillin (100 units/ml), and streptomycin (100 g/ml) at 37°C under humidified air containing 5% CO 2 . One day after HEK293 cells or HeLa cells had been seeded, the cells were transfected with the expression plasmid DNAs by the conventional calcium phosphate precipitation method or by using Fu-GENE6 (Roche Applied Science). After 24 h, cells were harvested and analyzed. To generate HEK293 stable transfectants that constitutively express 3ϫFLAG-tagged CHMP4b (FLAG-CHMP4b/HEK293), HEK293 cells transfected with the expression vector (pFLAG-CHMP4b) were selected by culturing in the presence of 0.5 mg/ml of G418 for 2 weeks.
Expression and Purification of GST-Alix⌬C-Escherichia coli BL21 cells were transformed with the plasmid pGST-Alix⌬C. The GST fusion protein was purified by binding to glutathione-Sepharose 4B (Amersham Biosciences) according to the manufacturer's instructions. GST-Alix⌬C was eluted from Sepharose beads with 10 mM reduced glutathione in 50 mM Tris-HCl, pH 8.5, dialyzed against 50 mM Tris-HCl, pH 7.5, and then stored at 4°C until use.
Pull-down Assay-One day after FLAG-CHMP4b/HEK293 (6 ϫ 10 6 ) cells had been seeded, the cells were harvested and washed with PBS (137 mM NaCl, 2.7 mM KCl, 8 mM Na 2 HPO 4 , 1.5 mM KH 2 PO 4 , pH 7.3) and then lysed in 450 l of lysis buffer (10 mM HEPES-NaOH, pH 7.4, 142.5 mM KCl, 0.2% Nonidet P-40, 0.1 mM pefabloc, 25 g/ml leupeptin, 1 M E-64, and 1 M pepstatin). Supernatants (each 200 l, obtained by centrifugation at 10,000 ϫ g) were incubated with 10 g of GST or GST-Alix⌬C protein that had been immobilized on glutathione-Sepharose for 3 h at 4°C. After complexes had been washed three times with the lysis buffer, they were subjected to Western blotting analysis using anti-FLAG monoclonal antibody (mAb). Signals were detected by either the chemiluminescent method using Super Signal West Pico Chemiluminescent Substrate (Pierce) or diaminobenzidine staining.
Immunoprecipitation-FLAG-CHMP4b/HEK293 cells transfected with the expression vector of either Alix⌬C-V5, GFP-Alix mutants, or Myc-SKD1 were lysed in the lysis buffer and then subjected to centrifugation at 10,000 ϫ g. The supernatants were incubated with anti-V5 mAb, anti-GFP pAb, or anti-FLAG mAb for 1 h and then incubated for 1 h after the addition of Protein G-Sepharose 4 Fast Flow (Amersham Biosciences). The Sepharose beads were then washed with the lysis buffer three times and subjected to Western blotting analysis using either anti-V5 mAb, anti-FLAG mAb, anti-GFP mAb, or anti-Myc mAb.
Epidermal Growth Factor (EGF) Uptake Assay-The transfected HeLa cells were incubated with Dulbecco's modified Eagle's medium for 1 h at 37°C and then with 0.5 g/ml tetramethylrhodamine-EGF (Molecular Probes, Inc., Eugene, OR) in 0.5 mg/ml bovine serum albumin/ Dulbecco's modified Eagle's medium for 1 h at 4°C. The cells were washed and incubated in 10% fetal bovine serum/Dulbecco's modified Eagle's medium for 30 min or 6 h at 37°C.

RESULTS
Yeast Two-hybrid Screening-To search for Alix-interacting proteins, the N-terminal region (amino acids 1-660) of human Alix (Alix⌬C) that was fused to the Gal4 DNA binding domain was used as bait, and a HeLa cell cDNA library was used as prey in a yeast two-hybrid screen system. Sixty-two positive clones were isolated from 6 ϫ 10 5 transformants. A search of DNA data bases revealed that the positive clones corresponded to 12 different proteins. Two of them were presumed to be human homologues of yeast Snf7/Vps32 (26,27) (Fig. 1), and they were also similar to each other (60.7% identity; 136 of 224 amino acid residues). One was identical to HSPC134 (24), which was referred to as CHMP4 by Howard et al. (25), and the other was very similar to a hypothetical protein registered in the sequence data bases (DDBJ/GenBank TM /EMBL accession number AL050349). In this paper, we designated HSPC134/ CHMP4 as CHMP4a and the other as CHMP4b, respectively. Both CHMP4a and CHMP4b were predicted to contain coiledcoil regions (Fig. 1A). Their N-terminal regions are rich in basic residues, whereas their C-terminal regions are acidic. The CHMP4 proteins are part of a large CHMP family (25). The data base search revealed an additional CHMP4 homologue in this subfamily in the human genome (Fig. 1B).
GST Pull-down Assays of Alix⌬C and CHMP4b-We performed a GST pull-down assay to determine whether the CHMP4b protein from mammalian cells has the capacity to bind to Alix. First, we constructed several plasmids for expressing various fusion proteins and generated HEK293 cell lines constitutively expressing FLAG-CHMP4b (FLAG-CHMP4b/ HEK293). After incubation of the lysate of FLAG-CHMP4b/ HEK293 with the GST fusion protein of Alix⌬C (GST-Alix⌬C) that had been immobilized on glutathione-Sepharose beads, the proteins bound to the beads were subjected to Western blotting using an anti-FLAG mAb as a probe. Whereas CHMP4b did not bind to the negative control GST beads, it bound to the GST-Alix⌬C beads ( Fig. 2A).
Co-immunoprecipitation of CHMP4b and Alix⌬C-When the lysate of FLAG-CHMP4b/HEK293 cells transfected with the expression vector of V5-tagged Alix⌬C (pAlix⌬C-V5) was immunoprecipitated with anti-FLAG mAb, the precipitates contained Alix⌬C-V5 as revealed by Western blotting using anti-V5 mAb (Fig. 2B). Moreover, in a reciprocal experiment, FLAG-CHMP4b was also co-precipitated with Alix⌬C-V5 using anti-V5 mAb for immunoprecipitation (data not shown).
Co-immunoprecipitation of CHMP4b with Alix Segments-To narrow the CHMP4b binding region in Alix⌬C, we performed co-immunoprecipitation analysis of FLAG-CHMP4b and GFP fusions of full-length or various truncated forms of Alix. The lysates of FLAG-CHMP4b/HEK293 cells transfected with the expression vectors of GFP-Alix mutants were immunoprecipitated with anti-GFP pAb, and the precipitates were subjected to Western blotting. As shown in Fig. 3, A and B, and summarized in Fig. 3C, we observed interaction of FLAG-CHMP4b with Alix mutants 1-868, 1-660, and 1-423 and weaker interaction with Alix mutant 115-660. No interaction was detected with Alix mutants 1-329, 221-660, and 329 -660. Bro1-rhophilin conserved domain located in the N-terminal region of Alix was essential but not sufficient for CHMP4b interaction. Since the Tyr 319 of Alix is highly conserved among Alix homologues in various organisms and serves as a potential phosphorylation site for Src-type tyrosine kinases, we replaced Tyr 319 of Alix⌬C with phenylalanine to examine whether this putative phosphorylation site is needed for the interaction. The interaction between CHMP4b and Y319F⌬C was still detectable with a slight decrease in signal.
Fluorescence Microscopic Analysis of Overexpressed CHMP4b-We investigated the subcellular localization of CHMP4b using HeLa cells transfected with pFLAG-CHMP4b. Double immunofluorescent staining was performed using an anti-FLAG antibody (Fig. 4, A, E, I, and M) and either an antibody of protein that is a marker of early endosome (EEA1) (Fig. 4B), late endosome and lysosome (Lamp-1) (Fig. 4F), or the trans-Golgi network and late endosome (CI-MPR) (Fig. 4J). We also used GFP-LC3 as an autophagosomal membrane marker instead of staining with an antibody (Fig. 4N). We found that FLAG-CHMP4b was distributed in a punctate manner mainly in the perinuclear area (Fig. 4, A, E, I, and M) and slightly diffused in the cytoplasm. However, only the diffuse pattern was observed in some cells (data not shown). As shown in merged images of the same panels ( Fig. 4, C, G, and K) and in images of higher magnification (Fig. 4, D, H, and L), the distribution of FLAG-CHMP4b partly overlapped in the perinuclear region with the distributions of EEA1 and Lamp-1, respectively, in the pFLAG-CHMP4b-transfected cell (Fig. 4, C, left cell, and G, bottom right cell), and the degree of overlapping was greater than that of CI-MPR (Fig. 4K, center cell). On the other hand, GFP-LC3 showed both punctate and diffuse patterns (Fig. 4N, inset), and its fluorescent signals were also detected in the nucleus under the present conditions, in which autophagy was not enhanced by subjecting the cells to nutrient starvation. The distribution of GFP-LC3 was not noticeably changed by overexpression of FLAG-CHMP4b (Fig. 4N) and did not overlap with that of FLAG-CHMP4b in cells expressing both GFP-LC3 and FLAG-CHMP4b (Fig. 4, O and P).
To determine whether the distribution of ubiquitinated proteins was affected by CHMP4b overexpression, we examined the localization of ubiquitinated proteins by staining HeLa cells transfected with pCHMP4b-GFP using anti-ubiquitin mAb (FK2) that recognizes mono-and polyubiquitinated proteins but not free monoubiquitin (33,34). In most of the transfected cells, the distribution of CHMP4b-GFP was essentially similar to that of FLAG-CHMP4b (Figs. 4 and 5A). As shown in Fig. 5B, fluorescent signals of ubiquitinated proteins became stronger in cells expressing CHMP4b-GFP, particularly in the perinuclear area, and showed punctate patterns. Furthermore, ubiquitinated proteins in this area were partially co-localized with CHMP4b-GFP (Fig. 5, C and D). When exogenously expressed FLAG-tagged ubiquitin was monitored, accumulation and partial co-localization of FLAG-tagged ubiquitin with GFPtagged CHMP4b were similarly observed (data not shown).
Accumulation of ubiquitinated proteins in the cells overexpressing CHMP4b suggests that overexpression of CHMP4b may disturb the endosomal sorting pathway. To provide functional data supporting this speculation, we monitored the fate of endocytosed EGF. HeLa cells transfected with pCHMP4b-GFP were incubated with Tetramethylrhodamine-EGF at 4°C for 1 h, washed, and then allowed to uptake EGF at 37°C for 30 min or 6 h. Internalized EGF was observed at 30 min, and noticeable change was not detected between the CHMP4b-GFP expressing cells and the untransfected cells (Fig. 5, E-G). At 6 h, fluorescent signals of tetramethylrhodamine-EGF were undetectable in most of the untransfected cells, whereas EGF remained in the perinuclear region of the cells expressing CHMP4b-GFP (Fig. 5, H-J). However, there was a variation in the amount of remaining EGF among the CHMP4b-GFP transfectants (data not shown).
Change in Alix Distribution by Overexpression of CHMP4b-To determine whether CHMP4b affects the subcellular localization of Alix, we performed immunostaining of HeLa cells transfected with pAlix-V5 (which encodes V5-tagged Alix) alone or both pAlix-V5 and pFLAG-CHMP4b (Fig. 6). Alix-V5 was diffusely distributed in the cytoplasm and par-tially concentrated in the perinuclear region of the Alix-V5 transfectants (Fig. 6B). In contrast, in the co-transfected cells, Alix-V5 was distributed in the perinuclear region (Fig. 6E), and most of the Alix was co-localized with CHMP4b (Fig. 6, D and  F). A similar effect was observed using the pAlix⌬C-V5 construct (data not shown).
Formation of Vesicle-like Structures-As shown in Fig. 7A, unlike obvious punctate patterns of localization of FLAG-CHMP4b in transiently overexpressed HeLa cells (Fig. 4) and in HEK293 cells (data not shown), FLAG-CHMP4b showed diffuse cytoplasmic distribution in stably expressing cells (FLAG-CHMP4b/HEK293), but vesicle-like structures were observed in ϳ5% of the cells (data not shown). On the other hand, Alix-V5 and Alix⌬C-V5 were localized in the cytoplasm, even in HEK293 cells transiently transfected with pAlix-V5 (Fig. 7B) or pAlix⌬C-V5 (Fig. 7C). In the FLAG-CHMP4b/HEK293 cells transfected with pAlix⌬C-V5, FLAG-CHMP4b and Alix⌬C-V5 showed punctate patterns of distribution (Fig. 7, G, H, J, and K) in 94% of the cells examined, and these two proteins were co-localized in vesicle-like structures (Fig. 7, I and L, indicated by the arrowheads). In contrast, the effect of overexpression of Alix-V5 in FLAG-CHMP4b/HEK293 cells on CHMP4b localization was not significant (Fig. 7, D, E, and F), and vesicle-like structures stained with FLAG-CHMP4b and Alix-V5 were observed in only 14% of the cells examined (data not shown).
Accumulation of CHMP4b and Alix at the E235Q Compartment-Howard et al. (25) reported that CHMP1 interacts with SKD1, which is an AAA-type ATPase homologous to yeast Vps4, and that CHMP1 accumulates in the so-called E235Q compartment in the cells expressing SKD1 E235Q , a dominant negative form of SKD1. To examine whether CHMP4b also interacts with SKD1 and accumulates in the E235Q compartment, we performed immunoprecipitation experiments. The lysate of FLAG-CHMP4b/HEK293 cells transfected with the expression vectors of either Myc-SKD1 or Myc-SKD1 E235Q was immunoprecipitated with anti-FLAG mAb, and the precipitates were subjected to Western blotting. As shown in Fig. 8A, whereas the interaction of Myc-SKD1 with FLAG-CHMP4b was not detectable under the condition used, the interaction of Myc-SKD1 E235Q with FLAG-CHMP4b was detected. We then performed immunostaining of HeLa cells transfected with expression vectors of Myc-SKD1 E235Q and CHMP4b-GFP. As shown in Fig. 8, B-D, the distribution of CHMP4b-GFP was very similar to that of Myc-SKD1 E235Q , and there were considerable overlaps between their distributions. To investigate whether Alix also accumulates in the E235Q compartment, we further performed immunostaining using HeLa cells expressing both Myc-SKD1 E235Q and GFP-Alix. The accumulation of a fraction of GFP-Alix was observed, and the increased GFP-Alix signals overlapped those of Myc-SKD1 E235Q (Fig. 8, E-G).

DISCUSSION
By a yeast two-hybrid screen using Alix⌬C (amino acids 1-660) as bait, we isolated two clones, which are similar to each other, as novel Alix-interacting proteins. By data base analysis, one of them was found to be identical to HSPC134, whose cDNA was isolated as one of the 300 previously undefined genes expressed in CD34 ϩ hematopoietic stem/progenitor cells (24), and HSPC134 was designated as CHMP4 by Howard et al. (25). In this paper, we designated HSPC134/CHMP4 as CHMP4a and the other similar protein as CHMP4b. These two proteins are homologues of yeast Snf7/Vps32 (Fig. 1), which is one of the class E Vps proteins and is essential for vacuolar protein sorting and transport (28). In addition to CHMP4a and CHMP4b, we found another CHMP4 homologue in the human genome data base (Fig. 1B). Their expression patterns in tissues and during development need to be analyzed in future studies.
In general, coiled-coil regions mediate the protein-protein interactions. Since both Alix and CHMP4b have coiled-coil regions, at first we speculated that these regions were essential for the interaction between the two proteins. However, as shown in Fig. 3, the coiled-coil regions in Alix were not required for their interaction, and the determined CHMP4b-interacting region possesses a potential phosphorylation site by Src-type tyrosine kinases. Indeed, substitution of Tyr 319 with Phe (Y319F mutant) has been shown to lose the ability to interact with both focal adhesion kinase and PYK-2 (35). The Y319F mutant, however, could substantially interact with CHMP4b (Fig. 3A). Xu et al. (18) reported that Rim20p, which is another yeast homologue of Alix, associates with Rim101p regardless of substitutions of Ile-Tyr to Ala-Ala in this highly conserved region. Although Tyr 319 in human Alix is widely conserved from lower to higher eukaryotic cells, the importance of Tyr 319 for protein complex formation appears to be dependent on interacting partners.
In yeast cells, class E Vps proteins are essential for carboxyl peptidase S sorting, and a group of class E Vps proteins form three complexes, named ESCRT-I, -II, and -III (endosomal sorting complex required for transport) (36 -38). ESCRTs work in the sorting of ubiquitinated proteins in the process of the invaginating endosomal membrane toward the lumen, and this compartment containing inner vesicles is called the MVB. ES-CRT-III is composed of four similar yeast CHMP family proteins, among which Snf7 forms a subcomplex with Vps20 and Vps2 forms a subcomplex with Vps24. It has recently been FIG. 2. Interaction between CHMP4b and Alix⌬C. A, the lysate of HEK293 cells constitutively expressing FLAG-CHMP4b was incubated with glutathione-Sepharose beads that carried GST or GST-Alix⌬C, and the beads then were pelleted by centrifugation. The pellets were analyzed by Western blotting with anti-FLAG mAb. B, HEK293 cells constitutively expressing FLAG-CHMP4b were transfected with pAlix⌬C-V5. After 24 h, transfectants were lysed and immunoprecipitated (IP) with anti-FLAG mAb (FLAG) or control mouse IgG (con). The immunoprecipitates were analyzed by Western blotting (WB) with anti-V5 mAb or anti-FLAG mAb. Mouse immunoglobulin heavy chains (IgG-H) and light chains (IgG-L) in the immunoprecipitates were also detected by subsequent Western blotting using peroxidase-conjugated anti-mouse IgG as a secondary antibody. The molecular masses of standard proteins are indicated on the left.
reported that Bro1 (yeast homologue of Alix) is identical to Vps31, which is also a class E Vps protein (28,39). Further, the endosomal association of Bro1 has been shown to be specifically dependent on Snf7 (40). Although direct interaction between Bro1/Vps31 and Snf7/Vps32 has not been demonstrated, the association of these proteins was suggested by results of systematic analysis of protein complexes by mass spectrometry (41). Moreover, the interaction between another yeast Alix homologue, Rim20p, and Snf7 was also suggested by results of genome-wide two-hybrid analysis (42). Since both CHMP4 and Alix are widely present from lower to higher eukaryotic cells, the function mediated by the interaction between the two proteins is presumed to be conserved and seems to be involved in the regulation of sorting in the endosomal pathway.
In HeLa cells (Figs. 4 and 5) and also in HEK293 cells (data not shown), most of the transiently overexpressed CHMP4b proteins are distributed in punctate patterns mainly in the perinuclear area. On the other hand, in HEK293 cells stably expressing FLAG-CHMP4b (FLAG-CHMP4b/HEK293), the diffuse distribution of CHMP4b in the cytoplasm was observed (Fig. 7A). By Western blot analysis using similar numbers of cells, the amount of the FLAG-CHMP4b protein in the transiently expressed HEK293 cells was estimated about 1.5-2 times as much as the amount of the protein expressed in the FLAG-CHMP4b/HEK293 cells (data not shown). Considering the transfection efficiency of 25-50%, the FLAG-CHMP4b protein in an individual transient transfectant was calculated to be present 3-8 times more per cell than in the stably expressing cells. Thus, the punctate subcellular localization of exogenously expressed CHMP4b may depend on the expression levels but not on the types of the cells. Indeed, diffuse cytoplasmic distributions were also observed in some HeLa cells by transient expression of the N-terminally tagged proteins (FLAG-CHMP4b and GFP-CHMP4b), particularly, using lesser amounts of expression vectors (data not shown). Since the C-terminally tagged CHMP4b-GFP protein is more prone to exhibit cytoplasmic punctate distributions under the similar transfection condition, tagging at the N terminus of CHMP4b might affect its stability, folding, and/or interaction with other proteins.
As in the yeast system, where the dissociation of Snf7 from the limiting membrane of MVB has been shown to depend on FIG. 3. Immunoprecipitation of CHMP4b with Alix truncated mutants. A and B, HEK293 cells constitutively expressing FLAG-CHMP4b were transfected with the expression vectors for various GFP-Alix truncated mutants. After 24 h, transfectants were lysed and immunoprecipitated with anti-GFP pAb. The immunoprecipitates were analyzed by Western blotting (WB) with anti-GFP mAb or anti-FLAG mAb. The numbers indicate the N-terminal and C-terminal Alix residues of the truncation mutants. C, a schematic representation of Alix truncated mutants is shown. The numbers indicate the N-terminal and C-terminal Alix residues of mutants. Y, the tyrosine residue in a highly conserved potential recognition site for Src type tyrosine kinases. F, the substituted phenylalanine residue by the point mutation. The relative strengths of the interactions are indicated as follows. ϩϩϩ, equivalent to full-length; ϩϩ, slightly weak; ϩ, weak; Ϫ, not detectable. BRD, Bro1-rhophilin conserved domain (NCBI conserved domain data base accession number 17278; pfam03097); CC, coiled-coil; PRR, proline-rich region.
the ATPase activity of Vps4 (27), the localization of CHMP4b may be also regulated by SKD1, a mammalian homologue of Vps4 (43,44). In the present study, this hypothesis was examined first by analysis of the interaction between CHMP4b and SKD1 and then by analysis of their intracellular co-localiza-tion. CHMP4b was co-immunoprecipitated with SKD1 E235Q , which is a SKD1 mutant defective in ATP hydrolysis, but not with wild type SKD1 (Fig. 8A). This result agrees with the previous report that blocking ATP hydrolysis of an AAA type ATPase by mutation (Glu to Gln mutation in the Walker B box) results in high affinity interactions of the AAA protein with its   protein targets (45). We observed the co-localization of CHMP4b with SKD1 E235Q in HeLa cells expressing CHMP4b-GFP and Myc-SKD1 E235Q (Fig. 8, B-D). Furthermore, the colocalization of CHMP4b-GFP with endogenous SKD1 was observed using anti-SKD1 antibody (data not shown). Since excessive CHMP4b in the transiently expressed cells may sequester endogenous SKD1, CHMP4b molecules may not be able to dissociate from the endosomal membranes and may keep associating with other ESCRT components. The accumulation of Alix in the E235Q compartment was observed as in the case of CHMP4b (Fig. 8, E-G). This result suggests that the dissociation of Alix from endosomes might also depend on the ATPase activity of SKD1.
The distributions of endosomal marker proteins such as EEA1 (early endosome) and Lamp-1 (late endosome and lysosome) were affected by overexpression of CHMP4b (Fig. 4).
CHMP4b was partially co-localized with EEA1 and Lamp-1 and to a lesser degree with CI-MPR (a marker protein for late endosome and trans-Golgi network), suggesting localization of CHMP4b from early to late endosomes. Ubiquitinated proteins accumulated in cells overexpressing CHMP4b, and CHMP4b was partly co-localized with ubiquitinated proteins like other endosomal marker proteins (Fig. 5, A-D). The electrophoretic mobilities of the FLAG-CHMP4b proteins analyzed by Western blotting using anti-FLAG antibody were not different between the lysates prepared from diffusely stained cells (FLAG-CHMP4b/HEK293) and those from transiently overexpressed HeLa cells that exhibited punctate distributions, and no higher molecular mass FLAG-CHMP4b proteins indicative of ubiquitination were observed in either cell lysates by Western blotting (data not shown). Thus, the observed co-localization of the ubiquitinated proteins and CHMP4b-GFP shown in Fig. 5, A-D, may not be due to ubiquitination of the CHMP4 protein itself. Ubiquitination of proteins is not only a signal of proteasomal degradation of cellular proteins but also a triggering signal for sorting of plasma membrane proteins such as growth factor receptors to lysosomal degradation through the endosomal sorting pathway, where Hrs and TSG101 play roles in directing ubiquitinated proteins via their ubiquitin-binding domains to the components of MVB sorting on endosomal membranes (28,46). Thus, the accumulation of the ubiquitinated proteins (Fig. 5, A-D) and the finding that overexpression of CHMP4b affected the distribution of endosomal markers (Fig.  4) suggest disturbance of the endosomal protein sorting pathway by overexpressed CHMP4b proteins. To provide experimental evidence in favor of this speculation, we monitored a fate of endocytosed EGF. Disappearance of the fluorescent signals of labeled EGF was inhibited in the HeLa cells transfected with CHMP4b-GFP (Fig. 5, H-J), raising a possibility that overexpression of CHMP4b disturbs the regulation of membrane trafficking from endosomes to lysosomes.
Overexpression of CHMP4b caused accumulation of Alix from the cytoplasm to the perinuclear area, where CHMP4b was co-localized (Fig. 6). We also found that overexpression of Alix⌬C-V5 induced formation of vesicle-like structures in cells stably expressing FLAG-CHMP4b (FLAG-CHMP4b/HEK293) (Fig. 7, G-L). These findings suggest that Alix cooperates with CHMP4b and participates in the endosomal pathway. Since no vesicular structures were observed in HEK293 cells transfected with pAlix⌬C-V5 (Fig. 7C), the level of endogenous CHMP4b seems to be too low to induce vesicle formation. In contrast, transient overexpression of CHMP4b both in HeLa cells (Figs. 4 and 5) and in HEK293 cells (data not shown) induced the accumulation of CHMP4b in the perinuclear area as punctate patterns without exogenous Alix, suggesting a sufficient expression level of endogenous Alix for CHMP4b interaction. Previously, Chatellard-Causse et al. (10) reported that overexpression of Alix-CT, the C-terminal half of Alix, led to cytoplasmic vacuolization and formation of tubulo-vesicular structures and that overexpression of Alix-CT affected the distribution of an endoplasmic reticulum marker protein, Grp78, but not the distributions of endosomal marker proteins (EEA1 and Lamp-1). Taken together with the results of our present study, Alix seems to have at least two different functions: a role in protein sorting of ubiquitinated endosomal cargoes and an unknown function involving the morphology of the endoplasmic reticulum. The former and the latter functions may be governed by the N-terminal and the C-terminal halves of Alix, respectively.
We have shown that CHMP4b interacts with Alix by coimmunoprecipitation experiments (Figs. 2B and 3) as well as by pull-down assays ( Fig. 2A). In these experiments, we observed that FLAG-CHMP4b proteins from FLAG-CHMP4b/ HEK293 cells recovered in the soluble fraction in lysis buffer containing 0.2% Nonidet P-40, but when an Alix⌬C-V5-expressing vector was transfected to FLAG-CHMP4b/HEK293 cells, approximately half of the FLAG-CHMP4b proteins were recovered in the insoluble fraction using the same buffer. In contrast, we observed little change in the recovery of FLAG-CHMP4b in the soluble fraction in the case of using transfectants of Alix-V5 (data not shown). This fact may be explained by the differences in the percentages of cells that formed vesicle-like structures induced by overexpression of Alix-V5 or Alix⌬C-V5 in FLAG-CHMP4b/HEK293 cells (14% versus 94%). CHMP4b-Alix complexes located in the vesicle-like structures are probably insoluble in lysis buffer containing 0.2% Nonidet P-40. Since the expression levels of Alix⌬C-V5 and Alix-V5 were not significantly different (data not shown), other molecules, which interact with the C-terminal region of Alix, may have an important role for dissociation from endosomal membranes. Our recently obtained data indicate that the binding sites for ALG-2, CIN85, and endophilin are all located at three different sites within the C-terminal proline-rich region. 2 Therefore, association or dissociation of Alix with these factors may trigger a conformational change of Alix and may affect the ability of Alix to interact with CHMP4b. Interestingly, CIN85 and endophilin are also factors involved in membrane receptor internalization triggered by Cbl, which has intrinsic ubiquitination activities as a RING-type E3 ligase (47,48).
Springael et al. (39) reported that Npi3, which is identical to Bro1, is required for vacuolar sorting of Gap1 permease under ubiquitin-dependent control in budding yeast. Compared with the genetically well studied yeast system, information on the mammalian MVB sorting pathway is still limited. Recent data, however, indicate that factors involving ubiquitin actions in the pathway are conserved between yeast and mammals (e.g. Vps27 versus Hrs, Vps23 versus TSG101, and Rsp5 versus Nedd4) (28,49). Interestingly, TSG101 was isolated as a candidate cDNA clone of the Alix-interacting partner using a fragment containing the proline-rich region of Alix as bait by the yeast two-hybrid screen. 3 Alix might link components of ubiquitin-dependent internalization machinery to components of the MVB sorting complexes by interacting with multiple factors.
Although the exogenously expressed FLAG-CHMP4b proteins exhibited cytoplasmic punctate patterns, they were also observed in the nucleus in some cells (Figs. 4 and 7). CHMP1, which is implicated in MVB formation (25), was first found as a partner of Polycomb group protein Pcl (Polycomblike) by a yeast two-hybrid screen, and a role in stable gene silencing within the nucleus was suggested (50). A post-translationally modified form of CHMP1 is associated with nuclear matrix, and overexpressed CHMP1 localizes to a punctate subnuclear pattern and arrests cells in S-phase (50). On the other hand, some mammalian class E Vps homologues were identified as components of nuclear factors. For instance, Eap20, Eap30, and Eap45, which are mammalian homologues of yeast ES-CRT-II components, Vps25, Vps22, and Vps36, respectively, are the subunits of RNA polymerase elongation factor, ELL (28,51). Thus, some mammalian class E Vps homologues might commonly play dual roles in the regulation of the gene expression in the nucleus and in the regulation of the MVB sorting in the cytoplasm.
Exosome is a specific form of the MVB inner vesicles secreted from various cells (52). Théry et al. (11) have revealed that Alix is present in exosomes. This fact agrees with our proposal that Alix is involved in the MVB sorting pathway. Alix may be recruited to the inner vesicles of MVB. Future studies, including electron microscopic analysis and tracing of endogenous proteins of Alix and CHMP4b with specific antibodies during cell surface receptor down-regulation, should clarify the specific stage of endosomal sorting regulated by these mammalian Vps protein homologues. was absent in Vps32/Snf7. We confirmed the absence of this motif in the CHMP4 proteins.