Interaction of the Metalloprotease Disintegrins MDC9 and MDC15 with Two SH3 Domain-containing Proteins, Endophilin I and SH3PX1*

Metalloprotease disintegrins (a disintegrin and metalloprotease (ADAM) and metalloprotease, disintegrin,cysteine-rich proteins (MDC)) are a family of membrane-anchored glycoproteins that function in diverse biological processes, including fertilization, neurogenesis, myogenesis, and ectodomain processing of cytokines and other proteins. The cytoplasmic domains of ADAMs often include putative signaling motifs, such as proline-rich SH3 ligand domains, suggesting that interactions with cytoplasmic proteins may affect metalloprotease disintegrin function. Here we report that two SH3 domain-containing proteins, endophilin I (SH3GL2, SH3p4) and a novel SH3 domain- andphox homology (PX) domain-containing protein, termed SH3PX1, can interact with the cytoplasmic domains of the metalloprotease disintegrins MDC9 and MDC15. These interactions were initially identified in a yeast two-hybrid screen and then confirmed using bacterial fusion proteins and co-immunoprecipitations from eukaryotic cells expressing both binding partners. SH3PX1 and endophilin I both preferentially bind the precursor but not the processed form of MDC9 and MDC15 in COS-7 cells. Since rat endophilin I is thought to play a role in synaptic vesicle endocytosis and SH3PX1 has sequence similarity to sorting nexins in yeast, we propose that endophilin I and SH3PX1 may have a role in regulating the function of MDC9 and MDC15 by influencing their intracellular processing, transport, or final subcellular localization.

Metalloprotease disintegrins (also referred to as ADAMs, 1 a disintegrin and metalloprotease, or MDC proteins, metalloprotease, disintegrin, cysteine-rich proteins) are a family of glycoproteins related to snake venom metalloproteases and integrin ligands (1)(2)(3). They are composed of several domains including a pro-domain, a metalloprotease domain, a disintegrin domain, a cysteine-rich region, and in most cases a membrane-spanning region and cytoplasmic tail. Metalloprotease disintegrins have been shown to play a role in fertilization (4 -8), muscle cell binding and fusion (9), shedding of tumor necrosis factor ␣, and other membrane-anchored proteins from the plasma membrane (10 -14), regulating neurogenesis and the function of Notch (15)(16)(17)(18) and Delta (19), and processing of heparin-binding epidermal growth factor-like growth factor (HB-EGF) (20). Currently 28 ADAMs have been identified, 15 of which contain a catalytic zinc-binding consensus sequence (HEXXH) (3,21). Four of these ADAMs have been shown to be catalytically active metalloproteases (MADM/Kuz/ADAM 10 (22); tumor necrosis factor ␣-converting enzyme (TACE)/ADAM 17 (10,11,23); meltrin-␣/ADAM 12 (24); and MDC9/ADAM 9/meltrin-␥ (20,25)). Besides their roles as metalloproteases, ADAMs are also thought to mediate cell-cell interactions. This was first suggested by the similarity of both the ␣ and ␤ subunit of the heterodimeric sperm protein fertilin (ADAM1/ADAM2) to snake venom disintegrins (6), which are short soluble peptides that bind to integrins (␣ IIb ␤ 3 and ␣ v ␤ 3 ) and are potent inhibitors of platelet aggregation (26 -30). Indeed, the disintegrin domain of fertilin ␤ (ADAM2) apparently mediates sperm-egg binding by interacting with an integrin on the oocyte surface, and this interaction may be a prerequisite for membrane fusion (4,7,(31)(32)(33)(34)(35). Furthermore, the disintegrin domain of the widely expressed human MDC15 (ADAM15) contains the integrin binding consensus "RGD" at its putative integrin-binding site (36,37). Evidence of a role for MDC15 in cell-cell interactions comes from studies where the extracellular portion of human MDC15 supports cell adhesion via the ␣ v ␤ 3 integrin when it is expressed in Escherichia coli (38) or via the integrins ␣ v ␤ 3 and ␣ 5 ␤ 1 when it is expressed in COS-7 cells as a fusion protein with the Fc portion of human IgG (39).
Several ADAMs contain cytoplasmic signaling motifs, including proline-rich regions that resemble Src homology 3 (SH3) ligand domains (36,37,40,41). The proline-rich region of the cytoplasmic tail of MDC9 has been shown to bind to the SH3 domain of Src in vitro in a blot overlay assay but not to the SH3 domain of the related oncogene abl (40). The presence of SH3 ligand domains in the cytoplasmic domains of MDC9, TACE, MDC15, meltrin-␣ (ADAM12), meltrin-␤ (ADAM19), and other family members suggests potential interactions with cytoplasmic proteins containing SH3 domains. These interactions could possibly be linked to the functions of metalloprotease disintegrins by inside-out regulation of activity, by outside-in signaling, or by regulating other aspects of their function, such as their subcellular localization or maturation.
Several lines of evidence suggest that the cytoplasmic domain of metalloprotease disintegrins may have a role in regulating their function. Overexpression of the cytoplasmic tail of MDC9 in Vero cells blocks the phorbol 12-myristate 13-acetateinduced shedding of HB-EGF (20). Also protein kinase C␦, which is involved in the shedding of HB-EGF, has been shown to bind to and phosphorylate the cytoplasmic tail of MDC9 (20). Furthermore, activation of protein kinase C with phorbol 12myristate 13-acetate results in phosphorylation of the cytoplasmic domain of MDC9 in Chinese hamster ovary cells (25). Interestingly, overexpression of full-length MDC9 strongly increases the constitutive shedding of HB-EGF, whereas overexpression of MDC9 lacking both predicted SH3 ligand domains results in a much smaller increase in constitutive shedding of HB-EGF, suggesting that interactions with SH3 domain-containing proteins may regulate MDC9 function (20). Additional evidence for a cytoplasmic domain regulating the function of an ADAM is provided by studies where the effect of a dominant negative form of Kuz (MADM, ADAM10), lacking the metalloprotease domain, on Notch signaling is strongly attenuated by removing a short conserved cytoplasmic region (16).
To elucidate the function of ADAM cytoplasmic domains, we have sought to identify proteins that interact with the widely expressed MDC9 (40) and MDC15 (36,37,42), both of which contain cytoplasmic proline-rich SH3 ligand domains. By using a yeast two-hybrid screen, we identified two SH3 domaincontaining proteins that are able to interact with both MDC9 and MDC15 but not with ADAM10, meltrin-␣, or TACE. One of the interacting proteins, human endophilin I (also known as SH3GL2), is apparently an orthologue of mouse and rat SH3p4, which have been implicated in synaptic vesicle endocytosis (43,44). The second protein is novel and contains one N-terminal SH3 domain, followed by a phox homology (PX) domain (PX domains contain a proline-rich sequence that may interact with proline-rich sequence binding modules such as SH3 or WW domains (45)), and a C-terminal coiled-coil domain. Because the novel protein sequence includes an SH3 domain and a PX domain, it was given the name SH3PX1. Endophilin I and SH3PX1 are the first reported SH3 domain-containing binding partners for metalloprotease disintegrins in cells. Since endophilin I and SH3PX1 both contain domains that are known to regulate the intracellular sorting or subcellular localization of other proteins, these proteins may regulate the function of MDC9 and MDC15 by affecting their intracellular transport or subcellular localization.

EXPERIMENTAL PROCEDURES
Reagents-Restriction endonucleases, T4 DNA ligase, and Taq DNA polymerase were obtained from Roche Molecular Biochemicals. [ 32 P]dCTP was purchased from NEN Life Science Products. Reagents were obtained from Sigma unless stated otherwise.
Plasmids-Constructs encoding the cytoplasmic domain of either human MDC9 or human MDC15 fused in frame with the GAL4 DNAbinding domain were prepared as follows. The cytoplasmic domains were amplified from cDNA by PCR using primers designed to give a 5Ј EcoRI site and a 3Ј SalI site downstream of a stop codon. The resulting PCR products were digested with EcoRI and SalI and ligated into the corresponding sites of the pGBT9 vector (CLONTECH). The same approach was used to generate the appropriate cytoplasmic tail deletion mutants for MDC9 or MDC15. Constructs encoding GAL4 fusions with the cytoplasmic domain of mouse meltrin-␣, human TACE, and mouse meltrin-␤ were prepared in the same manner as above using genespecific primers. The construct encoding the cytoplasmic domain of human ADAM10 was prepared using primers designed to generate a 5Ј EcoRI site and a 3Ј PstI site. The construct encoding the glutathione S-transferase (GST) fusion protein with the cytoplasmic domain of human MDC15 was described previously (36). The construct encoding a GST-human MDC9 cytoplasmic tail fusion protein was created using primers generating a 5Ј EcoRI site and a 3Ј SalI site downstream of a stop codon; this was cloned into pGEX4T-3 (Amersham Pharmacia Biotech). HIS/T7-tagged fusion protein constructs were prepared by ligating the EcoRI/XhoI inserts from pGADGH yeast-two-hybrid positives into the EcoRI and XhoI sites of the pET28b(ϩ) vector (Novagen). Fusion protein constructs with an N-terminal FLAG-tag were prepared by ligating the EcoRI/XhoI inserts from pGADGH yeast two-hybrid positives into the EcoRI and SalI sites of pFLAG-CMV-2 (Eastman Kodak Co.), a vector that yields a fusion protein with a FLAG-tag at its N terminus. The pcDNA 3 constructs for eukaryotic expression of fulllength mouse MDC9 and mouse MDC15 have been described previously (40,42). All constructs were sequenced prior to use to rule out any mutations resulting from PCR amplification.
Yeast Two-hybrid Screen and ␤-Galactosidase Assay-Yeast twohybrid screening was performed using the Matchmaker two-hybrid system (CLONTECH). A HeLa cell Matchmaker cDNA library in pGADGH (CLONTECH) and bait constructs in pGBT9 were used to simultaneously transform the yeast strain HF7c (MATa, ura3-52, his 3-200, lys 2-801, ade 2-101, trp 1-901, leu 2-3, 112, gal4-542, gal80-538, LYS2::GAL1-HIS3, URA3::(GAL4 -17 mers) 3 -CYC1-lacZ). A HeLa cell library was used as these cells contain MDC9 and MDC15 protein (data not shown). Transformed yeast expressing prey proteins that interact with the bait construct were selected by the ability to grow on SD agar plates lacking L-tryptophan, L-leucine, and L-histidine. Transformed yeast that were positive by nutritional selection were further assayed by ␤-galactosidase filter assays (following growth in the presence of L-histidine) according to the CLONTECH Matchmaker protocol. The specificity of bait-prey interactions was determined by co-transforming HF7c yeast with prey plasmid and control vector pGBT9 (encoding the GAL4 DNA binding domain) or pLAM 5Ј (encoding a GAL4 DNA binding domain-lamin C fusion protein). Positive prey plasmids were isolated from yeast after removing the bait plasmid by "dropout" following culture in SD-leucine medium (as described in the CLONTECH Matchmaker protocol). Isolated plasmids were used to transform E. coli XL-1 Blue and were identified by cDNA sequencing (The BioResource Center, Cornell University, Ithaca, NY). A total of four cDNA clones were identified using MDC15 as a bait, whereas three clones were identified using MDC9 as a bait. Two of the cDNAs were found in both screens (endophilin I and SH3PX1, see below). A novel protein related to hs-MAD2 (46), which was termed MAD2␤, was found using MDC9 as a bait but also interacted with MDC15 (47). By using MDC15 as a bait, two clones that interacted only with MDC15, but not with MDC9, were identified. One of these clones was sequenced completely (cDNA sequence deposited in GenBank TM under the accession number AF130979) and appears to be the human orthologue of FAP52 (48). The sequence of the second cDNA clone was identical to bases 2980 -3773 of a cDNA with the GenBank TM accession number AF063308.
Isolation of cDNA Clones from a Zap cDNA Library-An MDA-MB-468 Zap cDNA library (36) was screened using a radiolabeled probe corresponding to a 1.9-kilobase pair EcoRI/XhoI cDNA fragment of one of the yeast two-hybrid isolates, pGADGH/K-9 (SH3PX1). To identify cDNA inserts with the longest 5Ј extension, PCR was performed with antisense primers corresponding to the 5Ј end of the yeast-two-hybrid clone pGADGH/K-9 and the T3 primer, which anneals 5Ј of cDNA inserts in the Zap vector. Clones with the longest inserts were plaquepurified, and in vivo excision of the cDNA insert was performed using the ExAssist/SOLR system according to the Zap cDNA synthesis kit protocol (Stratagene). Full-length clones were sequenced on both strands. The full-length K-9/SH3PX1 sequence was submitted to Gen-Bank TM and given the accession number AF131214.
Northern Blot-The mRNA transcription pattern of SH3PX1 was determined by probing a human multiple tissue Northern blot (CLON-TECH) with a random-primed 32 P-labeled probe corresponding to the 1.9-kilobase pair EcoRI/XhoI fragment from pGADGH/K-9. Blots were prehybridized and hybridized using ExpressHyb (CLONTECH) following the ExpressHyb protocol. The human ␤-actin probe template DNA was from CLONTECH.
Bacterial Fusion Protein Production-Plasmids encoding GST fusion proteins or HIS/T7 fusion proteins were used to transform E. coli BL21 (DE3). Fusion protein expression was induced using isopropyl-␤-D-thiogalactopyranoside. Soluble protein was released by rupturing the cells with three cycles of freeze-thawing in phosphate-buffered saline, pH 7.4, containing 1% v/v Triton X-100 and protease inhibitors (49). Debris was removed by centrifugation (13,000 ϫ g, 4°C, 20 min), and GST fusion proteins were captured from the supernatant on glutathione-Sepharose 4B beads (Amersham Pharmacia Biotech).
Antibody Production-Rabbit polyclonal antibodies against the cytoplasmic domains of mouse MDC9 and mouse and human MDC15 have been described previously (36,40,42). Anti-SH3PX1 rabbit polyclonal antibodies were raised against a GST fusion protein with amino acids 6 -595 of SH3PX1, prepared by ligating the EcoRI/XhoI inserts from pGADGH yeast two-hybrid positive K-9 into the EcoRI and XhoI sites of pGEX/4T-3 (Amersham Pharmacia Biotech). The GST fusion protein was purified from E. coli using a previously described protocol (40). Female New Zealand White rabbits were immunized with the purified fusion proteins in phosphate-buffered saline according to established protocols (50).
Western Blot Analysis-Proteins were separated by SDS-polyacrylamide gel electrophoresis and transferred to nitrocellulose (Schleicher & Schuell) as described (51). After blocking in 5% reconstituted dry milk (Carnation) and incubation with primary antibodies, bound primary antibodies were detected using horseradish peroxidase-conjugated secondary antibodies (Promega) and the enhanced chemiluminescence detection kit (Amersham Pharmacia Biotech).
Transient Transfection of COS-7 Cells and Co-immunoprecipitation-COS-7 cells at approximately 50% confluence were transfected using LipofectAMINE (Life Technologies, Inc.) and cultured for 48 h. Cells were washed in phosphate-buffered saline, lysed in cell lysis buffer (Tris-buffered saline containing 0.5% v/v Nonidet P-40, 1 mM Ca 2ϩ , and 1 mM Mg 2ϩ ), and the lysate was cleared by centrifugation (14,000 ϫ g, 15 min, 4°C). Co-immunoprecipitations were performed using lysates prepared from cells co-transfected with FLAG-tag fusion protein constructs and full-length mouse MDC9 or mouse MDC15 constructs in pcDNA 3. FLAG-tagged proteins were immunoprecipitated using anti-FLAG M2 monoclonal antibodies and protein G-Sepharose beads (Sigma). Beads were washed in cell lysis buffer, and bound proteins were eluted by boiling in SDS-polyacrylamide gel electrophoresis sample loading buffer containing 10 mM dithiothreitol. In these experiments, mouse MDC9 was used because it is expressed at higher levels in COS-7 cells than its human homologue (data not shown). Furthermore, mouse MDC15 was used because it is processed to its mature form by COS-7 cells more efficiently than the human form (data not shown). Separate experiments confirmed that mouse GST-MDC9 and mouse GST-MDC15 were both able to bind to bacterially expressed endophilin I and SH3PX1 (data not shown).

Interaction of MDC9 and MDC15 GST-Cytoplasmic Tails with Endophilin I and SH3PX1
Fusion Proteins in Vitro-His/T7-tagged endophilin I and SH3PX1 fusion proteins were expressed in E. coli. Bacterial cell lysates were prepared as described above. GST alone as a control or GST fusion proteins with the cytoplasmic domain of MDC9 or MDC15 were captured on glutathione-Sepharose beads, added to the bacterial cell lysate with either the His/T7-tagged endophilin I or SH3PX1 fusion proteins, and incubated at 4°C for 2 h. The beads were washed with lysis buffer. Bound proteins were eluted by heating to 95°C in SDS-polyacrylamide gel electrophoresis sample loading buffer containing 10 mM dithiothreitol for 5 min, separated on 10% w/v SDSpolyacrylamide gels, blotted onto nitrocellulose, and detected by Western blot with an anti-T7 monoclonal antibody (Novagen).

Identification of Cytoplasmic Proteins That
Interact with MDC9 and MDC15 by Yeast Two-hybrid-To identify proteins that interact with the cytoplasmic domains of human MDC9 and human MDC15, 6 ϫ 10 6 clones of a HeLa cDNA library were screened by yeast two-hybrid using the MDC9 or MDC15 cytoplasmic domains as bait constructs. Growth under nutritional selection and production of ␤-galactosidase were used as markers of an interaction between bait and prey proteins, and clones that were positive by both of these criteria were isolated. In independent screens with MDC9 and MDC15 cytoplasmic tails, two interacting proteins, initially designated as K-60 and K-9, were isolated. Neither K-60 nor K-9 interacted with a control bait protein, the GAL4 DNA binding domain fused to human lamin C. Two additional yeast two-hybrid clones were identified in the screen with MDC15 which did not interact with MDC9, but these were not characterized in this study (data not shown, see "Experimental Procedures").
In order to determine the specificity of the observed interactions, we tested whether K-60 or K-9 could interact with the cytoplasmic domains of other metalloprotease disintegrins using the yeast two-hybrid system. Nutritional selection (Fig. 1A) indicated that K-60 and K-9 both interact with the cytoplasmic tails of MDC9, MDC15, meltrin-␣ (ADAM12) and meltrin-␤ (ADAM19). However, in the more stringent ␤-galactosidase production assay, K-9 only interacted with the cytoplasmic tails of MDC9 and MDC15, whereas K-60 was able to interact with MDC9, MDC15, and meltrin-␤ (Fig. 1B). Neither K-60 nor K-9 interacted with the cytoplasmic tails of TACE, ADAM10, or meltrin-␣.
cDNA Cloning, Sequencing, and Analysis of the Expression Patterns of K-60 and K-9 -Sequence analysis showed that K-60 is identical to amino acids 23-352 of Src homology 3-containing Grb2-like protein transcript 2 (SH3GL2). Because SH3GL2 is apparently a human orthologue of mouse endophilin I (SH3p4) we will refer to it as endophilin I hereafter (52,53). Endophilin I is a 40-kDa protein that is composed of 352 amino acid residues and includes an SH3 domain resembling that of the adaptor protein Grb2 and a coiled-coil domain (43). Neither the cDNA sequence of the yeast two-hybrid clone K-9 nor its deduced protein sequence was identical to any other full-length sequence in the current GenBank TM data base, although several ESTs and one partial clone corresponding to this sequence were noted. To identify the full-length cDNA corresponding to the K-9 insert, several positive clones were isolated from an MDA-MB-468 Zap cDNA library and sequenced. The longest cDNA clone isolated had an insert of 2310 base pairs that encoded a protein of 595 amino acid residues with a predicted molecular mass of 67 kDa ( Fig. 2A). The clone isolated by the yeast two-hybrid screen corresponds to amino acids 6 -595 of this protein. Because the deduced protein sequence includes an SH3 domain (Fig. 2B) and a PX domain (45) (Fig. 2C), this novel protein was given the name SH3PX1. The C-terminal 26 residues of SH3PX1 are predicted to form a coiled-coil by the protein analysis program "COILS" (54). Fig. 2 presents alignments of the SH3 domain (Fig. 2B) and the PX domain (Fig. 2C) of SH3PX1 with the most closely related protein domains currently found in the GenBank TM data base. The two most closely related sequences are encoded by adjacent regions of the Caenorhabditis elegans genome, suggesting that they are parts of the same gene.
Northern blot analysis showed that SH3PX1 is widely expressed with transcript sizes of 4.4 and 3.1 kilobases (Fig. 2D). It remains to be determined whether these two transcripts represent splice variants of SH3PX1 or highly related genes. Highest transcription levels were observed in heart and placenta, whereas relatively low levels of expression were found in thymus and peripheral blood leukocytes.
Polyclonal antibodies raised against a GST-SH3PX1 fusion protein recognized proteins of approximately 57 and 78 kDa in HeLa, COS-7, and 293 (a transformed primary human embryonal kidney cell line) cell lysates (Fig. 3). Transfection of COS-7 cells with full-length SH3PX1 in pcDNA 3 resulted in an increase in the amount of 57-and 78-kDa proteins (Fig. 3), whereas preimmune serum did not recognize either of these proteins.
Distinct Proline-rich Sequences in the Cytoplasmic Domains of MDC9 and MDC15 Are Required for the Interaction with Endophilin I Versus SH3PX1-The cytoplasmic domains of MDC9 and MDC15 both contain two proline-rich regions that are predicted to serve as ligands for SH3 domains (Fig. 4A) (36,37,40,42). The sequences within the cytoplasmic tails of MDC9 and MDC15 that are sufficient for interactions with endophilin I and SH3PX1 were defined using specific cytoplasmic tail deletion mutants in a yeast two-hybrid assay. Interactions were scored by nutritional selection and ␤-galactosidase production as outlined above.
Binding of endophilin I to MDC9 and MDC15 was mapped to constructs containing the N-terminal proline-rich region (ami-no acid residues 785-800 of MDC9 and amino acid residues 758 -786 of MDC15; Fig. 4, A and B). In contrast, SH3PX1 required the C-terminal proline-rich region for interaction with the MDC9 or MDC15 cytoplasmic tails (amino acid residues 800 -819 of MDC9, and amino acid residues 787-814 of MDC15; Fig. 4, A and B).
To corroborate further the associations observed by yeast two-hybrid, an in vitro assay for interactions of the binding partners expressed in E. coli was performed. GST or GST fusion proteins with the MDC9 or MDC15 cytoplasmic tail were immobilized on glutathione-Sepharose 4B beads and used to capture His/T7-tagged endophilin I or SH3PX1 fusion proteins. Fig. 5 shows that GST fusion proteins with the cytoplasmic domain of MDC9 or MDC15 were able to bind bacterially ex- pressed His/T7-tagged endophilin I or SH3PX1 fusion proteins, whereas the GST control protein was not. Similar results were obtained when GST-MDC9 or GST-MDC15 fusion proteins were used to capture FLAG-tagged SH3PX1 or endophilin I from lysates of transiently transfected COS-7 cells (data not shown).
Interaction of SH3PX1 and Endophilin I with MDC9 and MDC15 in Eukaryotic Cells-To examine whether SH3PX1 and endophilin I can interact with MDC9 and MDC15 in eukaryotic cells, FLAG fusion proteins with endophilin I or SH3PX1, or a FLAG control, were co-expressed with mouse MDC9 or mouse MDC15 in COS-7 cells. FLAG-tagged proteins were immunoprecipitated from cell lysates using an anti-FLAG monoclonal antibody, and co-immunoprecipitated MDC9 or MDC15 was detected using antibodies against their cytoplasmic domains. Both MDC9 and MDC15 co-immunoprecipitated with FLAG-SH3PX1 and with FLAG-endophilin I from eukaryotic cells but not with FLAG controls (Fig. 6, A and B). Interestingly, both SH3PX1 and endophilin I associated predominantly with the pro-forms of MDC9 and MDC15, although the majority of detectable MDC9 and MDC15 in COS-7 cells is present in its mature form. The pro-forms of MDC9 and MDC15 still possess the pro-domains, which is thought to keep the protease inactive via a cysteine-switch mechanism (25,42,55) at least until the pro-domain is removed by a furin-type pro-protein convertase in the trans-Golgi network.

DISCUSSION
The cytoplasmic tails of several metalloprotease disintegrins, including MDC9 and MDC15, contain proline-rich regions that are proposed to function as SH3 ligand domains (36,37,40,41). In this study, we identified two SH3 domaincontaining proteins, endophilin I and the novel SH3PX1, as binding partners of MDC9 and MDC15. In both cases, the interactions were initially identified by yeast two-hybrid and then shown to be direct and specific using bacterially expressed proteins and by co-immunoprecipitation from COS-7 cells overexpressing both binding partners. Since rat endophilin I has been implicated in endocytosis and SH3PX1 contains a combination of domains that are implicated in intracellular protein trafficking or movement (see below), these proteins may also have a role in regulating the subcellular localization or function of MDC9 or MDC15.
Endophilin I (which is also referred to as SH3GL2 or SH3p4) contains a Grb2-like SH3 domain and a coiled-coil domain. Endophilin I is expressed at highest levels in brain, but it is also expressed at lower levels in other tissues (43). In contrast, MDC9 and MDC15 are both widely expressed (36,40). However, endophilin I belongs to a gene family with two other highly related members, one of which (endophilin II, SH3GL1, or SH3p8) is expressed ubiquitously, whereas the other (endophilin III, SH3GL3, or SH3p13) is expressed mainly in the brain, thymus, and testis (43). Since the SH3 domains of endophilins I, II, and III are highly conserved and can all bind to the SH3-ligand domain of synaptojanin, it will be interesting to determine whether endophilins II and III are also capable of interacting with MDC9 and/or MDC15. Thus, although endophilin I has a relatively specific expression pattern, the widely expressed MDC9 and MDC15 may be able to interact with different forms of endophilin in different tissues. Rat endophilin I binds synaptojanin, amphiphysin (I and II), and dynamin at synaptic termini (43,44,53,56) and may regulate synaptic membrane retrieval by its ability to compete with amphiphysin and dynamin for binding to synaptojanin (56). The finding that human endophilin I can interact with MDC9 and MDC15 raises the possibility that members of the endophilin protein family may have several distinct binding partners and functions. Consistent with this idea, endophilin III/ SH3GL3 was recently reported to bind to the Huntington's disease protein Huntingtin and to promote the formation of polyglutamine-containing aggregates (57).
Since endophilin I has only one SH3 domain, it should be able to interact with one SH3 ligand domain at a time. However, endophilins I, II, and III also contain N-terminal coiledcoil domains. Coiled-coils are amphipathic helixes that mediate protein-protein interactions and are also found in the SNAREs and other proteins with roles in intracellular membrane fusion and/or intracellular protein transport (58 -60). These coiled- . Numbers in parentheses refer to the amino acid residues at the N and C termini of each construct. Plasmids encoding the GAL4 DNA binding domain fused to portions of cytoplasmic tails and either pGADGH-K-9/SH3PX1 or pGADGH K-60/endophilin were co-transformed into yeast strain HF7c. ␤-Galactosidase assays were performed as described under "Experimental Procedures." Transformed yeasts were examined for their ability to grow under nutritional selection and to produce ␤-galactosidase; ϩ indicates those constructs able to interact with prey constructs by both criteria, and Ϫ indicates those constructs unable to interact with prey constructs by both criteria. ND, not determined. coil regions could either allow dimers between endophilin proteins to form or could link endophilins to other proteins containing coiled-coil domains. Thus, if endophilin can homo-or heterodimerize via its coiled-coil domain, one could envision a function in linking MDC9 or MDC15 to proteins with a role in endocytosis or to other proteins that could regulate the subcellular localization or function of MDC9 and MDC15. SH3PX1, the second protein found to interact with both MDC9 and MDC15, contains an N-terminal SH3 domain and a PX domain (45) in addition to a short C-terminal coiled-coil domain. The N-terminal part of SH3PX1, including the SH3 domain, is identical to a partial human cDNA isolated from activated T-cells as a binding partner of the Wiskott Aldrich Syndrome Protein (GenBank TM accession number AF001629). PX domains are found in proteins with quite distinct functions, including the sorting nexins, phosphatidylinositol 3-kinases, and NADPH oxidase subunits (45,61). Some PX domain-containing proteins appear to regulate the subcellular localization of other proteins or protein complexes (62,63). PX domains contain proline-rich sequences that may themselves interact with SH3 or WW domains (45). Since SH3PX1 interacts with MDC9 and MDC15 via a fragment containing its SH3 domain (residues 1-160, data not shown), its coiled-coil domain and the PX domain should both be free to interact with other cytoplasmic binding partners. Thus SH3PX1 and endophilin I may function as adaptor proteins that could regulate the assembly of cytoplasmic protein complexes that include MDC9 and/or MDC15. Furthermore, any one MDC9 or MDC15 molecule may be in contact with endophilin I and SH3PX1 simultaneously since each of these SH3 domain proteins binds a separate SH3 ligand domain in MDC9 and MDC15.
MDC9 and MDC15 are both made as larger precursors that are processed by a furin-type pro-protein convertase in the trans-Golgi network (25,40,42). This removes the N-terminal pro-domain, which has a role in protein folding and keeps the protease inactive via a cysteine-switch mechanism (25,42,55). Removal of the pro-domain is therefore thought to be a prerequisite for protease activity. The observation that both endophilin I and SH3PX1 preferentially bind to the pro-form of MDC9 and MDC15, rather than the mature form, suggests that this interaction occurs in a secretory pathway compartment prior to the medial Golgi. Endophilin I and SH3PX1 may thus have a role in regulating transit of MDC9 and MDC15 through the Golgi apparatus or in some other aspect of the maturation or sorting of MDC9 and MDC15.
This study is the first to demonstrate an interaction between a recombinantly expressed metalloprotease disintegrin and SH3 domain-containing cytoplasmic proteins in cells. One possible consequence of an interaction between MDC9 or MDC15 and endophilin I and/or SH3PX1 could be the regulation of transport or subcellular localization. This in turn may affect the rate of pro-domain removal by limiting access to a proprotein convertase such as furin in the trans-Golgi network or may regulate access to their substrates. Either possibility, or a combination of both, could provide a means of regulating the function of MDC9 and/or MDC15. Alternatively, the interaction with cytoplasmic proteins may be important for outside-in signal transduction, for example in cell-cell interactions, or in inside-out regulation of the proteolytic activity or the integrinbinding ability of these two metalloprotease disintegrins, although this would presumably require an interaction with the mature form of the protein. The fact that endophilin I and SH3PX1 interact with MDC9 and MDC15 indicates that these two ADAMs may have a common mode of regulation or that they have overlapping functions mediated by a common intermediate. Future studies will be directed toward evaluating the role of endophilin I and SH3PX1 in regulating the function of MDC9 or MDC15 in cell-cell interactions or in protein ectodomain processing.