Stonin 2: An Adaptor-like Protein That Interacts with Components of the Endocytic Machinery

The internalization of plasma membrane proteins is critical to many important physiological processes such as synaptic vesicle recycling and receptor-mediated endocytosis. These processes are mediated by a complex molecular machinery that is recruited from the cytosol to the inner surface of the plasma membrane, where it assembles into a supramolecular structure known as a protein coat (for reviews see Owen and Luzio 2000; Kirchhausen 2000; Slepnev and De Camilli 2000). The coat selects integral membrane proteins destined for internalization and participates in the formation of endocytic vesicles. The main structural constituent of plasma membrane coats is the protein clathrin. Assembly of clathrin coats onto the plasma membrane is initiated by binding of the heterotetrameric adaptor protein (AP) complex, AP-2, to a member of the synaptotagmin family of integral membrane proteins (Zhang et al. 1994; Haucke et al. 2000). Clathrin triskelions then polymerize onto the membrane-bound AP-2 to form a lattice that serves as a scaffold for the protein coat. Integral membrane proteins containing either tyrosine-based or dileucine-based endocytic signals are concentrated within the assembling coats by interaction of these signals with AP-2 (for review see Bonifacino and Dell'Angelica 1999; Heilker et al. 1999). Various accessory proteins (e.g., amphiphysins, Eps15, intersectins, dynamins, etc.) become associated with the coat by either direct or indirect interaction with clathrin and/or AP-2. These accessory proteins are thought to regulate coat assembly and vesicle budding, as well as interactions of the endocytic machinery with signal transduction pathways and the actin cytoskeleton (Kirchhausen 2000; Owen and Luzio 2000; Slepnev and De Camilli 2000).

Novel components of the machinery are still being identified through biochemical and genetic approaches. A particularly useful approach to identify physiologically critical components of the endocytic machinery has been the study of temperature-sensitive paralytic mutants of D. melanogaster pioneered by David Suzuki and colleagues (Grigliatti et al. 1973). Some of these mutants bear defects in genes involved in synaptic vesicle recycling from the plasma membrane. Inactivation of these genes upon shift to the nonpermissive temperature results in an inability to reform synaptic vesicles after exocytosis. One of these mutant genes, shibire, was found to encode a D. melanogaster homologue of mammalian dynamins (Chen et al. 1991; van der Bliek and Meyerowitz 1991). Another paralytic mutant, stoned, was found to bear mutations in a dicistronic transcript encoding two structurally unrelated proteins, stoned A and stoned B (Andrews et al. 1996). Although stoned A has no homology to any other known protein, stoned B comprises a COOH-terminal domain with homology to the μ2 subunit of AP-2 (Andrews et al. 1996). Combination of hypomorphic shibire and stoned mutations produces synthetic lethality, suggesting that the products of these genes participate in either the same or parallel pathways of synaptic vesicle biogenesis (Petrovich et al. 1993). It remains to be established, however, whether the stoned gene products play a role in endocytosis itself, and whether this role is limited to neurons or general to all cells.

To investigate the possible involvement of stoned gene products in endocytosis in a system that is more amenable to biochemical and cellular analyses, we sought to identify human homologues of these proteins. Here, we report the identification and characterization of two ubiquitously expressed human homologues of the D. melanogaster stoned B protein named stonin 1 and stonin 2. The results of our study suggest that stonin 2 is a component of the general endocytic machinery in higher eukaryotes.

A search of DNA databases for human homologues of D. melanogaster stoned B identified the SALF protein (Upadhyaya et al. 1999) and a related fetal lung protein encoded by a novel cDNA sequence (GenBank/EMBL/DDBJ accession no. AI734225). Primers corresponding to SALF were used in 5′- and 3′-RACE PCR on a Marathon-Ready™ human heart cDNA library (CLONTECH Laboratories, Inc.) to isolate a novel transcript, herein termed Stonin 1 (GenBank/EMBL/DDBJ accession no. AF255310), encoding the NH₂-terminal 711 of the 1,182–amino acid SALF. The complete ORF of the second human cDNA, herein termed Stonin 2 (GenBank/EMBL/DDBJ accession no. AF255309) was similarly obtained from a Marathon-Ready™ human spleen cDNA library (CLONTECH Laboratories, Inc.), using primers designed on the basis of partial genomic DNA sequences (GenBank/EMBL/DDBJ accession nos. AC010582 and AL136040). The identical stonin 1 and stonin 2 full-length sequences were amplified by RT-PCR from mRNA of multiple human cell lines (HeLa, Jurkat, 293T, and Caco-2A). The above PCR products were cloned into the pCR2.1 vector (Invitrogen) and sequenced.

Northern blot analysis was performed on multiple human tissue blots (CLONTECH Laboratories, Inc.) as described previously (Dell'Angelica et al. 1997). Stonin 1 and stonin 2 cDNA probes were obtained by PCR. For stonin 1, two different probes were used in separate experiments: the proline-rich domain probe comprised nucleotides 1–615 and the μ-homology domain probe comprised nucleotides 1187–2509 of the full-length stonin 1 cDNA. The probe for stonin 2 corresponded to the complete coding sequence.

Polyclonal antibodies to human stonin 1 and stonin 2 were generated by immunizing rabbits with the corresponding proline-rich domains fused to a deca-histidine tag. The fusion proteins were recovered from inclusion bodies by extraction in 6 M urea. A monoclonal antibody to human Eps15 and a polyclonal antibody to Xenopus intersectin were gifts from Dr. Pier Paolo Di Fiore (European Institute of Oncology, Milan, Italy) and Dr. Peter McPherson (McGill University, Montreal, Canada), respectively. The following monoclonal antibodies were also used: HA.11; to the hemagglutinin epitope and 9E10; to the Myc epitope (Covance), 100/3; to γ1-adaptin and AP.6; to α-adaptin (Sigma-Aldrich), clones 7.1 and 13.1; to the green fluorescent protein (GFP; Roche Molecular Biochemicals) and H4A3; to lamp-1 (Developmental Studies Hybridoma Bank). The following polyclonal antibodies were also used: anti-GFP antiserum (Medical and Biological Laboratories Co.) and serum anti–caveolin antiserum (Transduction Laboratories).

GFP fusion constructs, comprising the proline-rich domain of human stonin 2, namely GFP–stonin 2_PRD (residues 1–223) and the full-length GFP–stonin 2, were cloned into the pEGFP-C1 (HindIII-SmaI) vector (CLONTECH Laboratories, Inc.). The complete ORF of human stonin 1 was fused to GFP by cloning into pEGFP-C2 (GFP–stonin 1, KpnI-BamHI). Epitope-tagging at the NH₂ terminus of stonin 2 was performed through PCR amplification of the full-length stonin 2 cDNA using 5′-primer containing the nucleotide sequence for the Myc epitope. This product was cloned into the XhoI-XbaI sites of the mammalian expression vector pCI-neo (Promega). The PRD + SHD domains of stonin 2 were subcloned into the XhoI-XmaI sites of the pCI-neo vector (Myc-stonin 2_PRD+SHD; residues 1–359) using the same approach. Two-hybrid constructs in the pGBKT7 (TRP1) vector (CLONTECH Laboratories, Inc.) were prepared by cloning of nucleotide sequences coding for the following amino acid residues: 1–296 of human Eps15 (Eps15–Eps15 homology (EH); NdeI-BamHI); 1–395 of human Eps15R (Eps15R-EH; EcoRI-BamHI); and 1–333 of human intersectin 1 (intersectin 1-EH; NdeI-XmaI). Similarly, the stonin 1 (full-length; XhoI-XmaI), stonin 2 (full-length; NdeI-XmaI), stonin 2_PRD+SHD (NdeI-XmaI), and human epsin 2 (full-length; NdeI-EcoRI) cDNAs were cloned into the pGADT7 (LEU2) vector (CLONTECH Laboratories, Inc.) to obtain the corresponding GAL4 transcription activation domain (GAL4ad) fusion constructs. The constructs (protein names and amino acid numbers indicated in parentheses) glutathione S-transferase (GST)–intersectin 1-EH (intersectin 1, 1–333), GST–synaptotagmin I-C2AB (murine synaptotagmin I, 96–421), GST–synaptotagmin I-C2A (synaptotagmin I, 96–265), GST–synaptotagmin I-C2B (synaptotagmin I, 248–421), GST–synaptotagmin II-C2AB (murine synaptotagmin II, 124–422), GST–synaptotagmin II-C2A (synaptotagmin II, 124–272), and GST–synaptotagmin II-C2B (synaptotagmin II, 249–422) were cloned into the EcoRI-XhoI sites of the pGEX-5X-1 vector (Amersham Pharmacia Biotech). GST–Eps15-EH was generated by cloning of the cDNA sequence corresponding to residues 1–296 of Eps15 into the BamHI-XhoI sites of the pGEX-4T-1 vector (Amersham Pharmacia Biotech). For protein expression and purification, competent E. coli strains DH5α (PGC Scientific) or BL21 (DE3) (Novagen Inc.) cells were transformed with the above constructs. Recombinant GST fusion proteins were purified by using glutathione–Sepharose 4B beads (Amersham Pharmacia Biotech), and amino acid substitutions in the proline-rich domain of stonin 2 (N110A, P111A, F112A, P127A, and F128A) were made using the QuickChange mutagenesis kit (Stratagene), according to the manufacturer's instructions.

The sources and culture conditions for all the human cell lines used in this study are described elsewhere (Dell'Angelica et al. 1997). Human HeLa or M1 cells, grown on glass coverslips, were transfected by using FuGENE-6 (Roche Molecular Biochemicals). Cells were incubated with 7 mM Na-butyrate for 12 h before fixation. At 24 h after transfection, cells were either fixed and processed for immunofluorescence (Dell'Angelica et al. 1997) or used directly for analysis of rhodamine-EGF (500 ng/ml), Dil-LDL (10 ng/μl), or Cy3-transferrin (530 ng/ml) internalization (Benmerah et al. 2000; Cy3-transferrin was a gift from Timothy McGraw, Weill Medical College, New York, NY). The internalization of rhodamine-EGF was performed using the same conditions as for Cy3-transferrin, with the exception that the serum starvation was much longer (12 h) and was done in the presence of 0.5% (vol/vol) serum. Images were acquired on an LSM 410 confocal microscope (Carl Zeiss, Inc.).

Metabolic labeling of cells in culture with [³⁵S]methionine, preparation of Triton X-100, detergent-free extracts of post-nuclear membranes, ultracentrifugation on sucrose gradients, and immunoprecipitation–recapture were carried out as described (Dell'Angelica et al. 1997). For cell fractionation, [³⁵S]methionine-labeled cells were disrupted in lysis buffer (10 mM Tris-HCl, pH 7.4, 250 mM sucrose with a cocktail of protease inhibitors) (Dell'Angelica et al. 1997) by 20 passages through a 25-gauge needle on ice. Samples were centrifuged at 500 g for 10 min at 4°C, and the resulting supernatants (postnuclear supernatant) were subjected to ultracentrifugation at 105,000 g for 1 h at 4°C to obtain cytosolic (supernatant) and membrane (pellet) fractions. For extraction experiments, the membrane fraction was resuspended in lysis buffer and divided into six equal aliquots. Aliquots were then incubated for 45 min at 4°C in lysis buffer containing either 1 M Tris-HCl (pH 7.4), 1 M NaCl, 0.2 M Na₂CO₃ (pH 11.3), 1.6 M urea, 2% (wt/vol) Triton X-100, or left untreated. After the incubation period, samples were centrifuged at 105,000 g for 1 h, and both pellets and supernatants were collected for further analysis. To obtain a fraction enriched in detergent-resistant membranes, cellular extracts were subjected to a discontinuous sucrose gradient as described (Ilangumaran and Hoessli 1998), with the exception that 0.9-ml fractions were collected from the top of the gradient. Then, 0.5 ml of each fraction was diluted 1:1 in immunoprecipitation buffer and used for immunoprecipitation–recapture with the antibodies described in the corresponding figure legend. For in vitro translation of Myc-stonin 2_PRD+SHD, [³⁵S]methionine-labeled proteins were synthesized in vitro using the TNT^® Coupled Reticulocyte Lysate System (Promega). Postribosomal supernatants were then generated by centrifugation at 200,000 g and used for GST pull-down assays.

15 μg of either GST–Eps15-EH or GST–intersectin 1-EH fusion proteins immobilized on glutathione–Sepharose were incubated with in vitro translation products or extracts from transfected HeLa cells (0.8 mg of protein) in binding buffer (25 mM Tris-HCl, pH 7.5, 100 mM NaCl, 5 mM MgCl₂, 1 mM EDTA, protease inhibitors, and 1% [wt/vol] Triton X-100) for 4 h at 4°C. 20 μg of GST–synaptotagmin fusion proteins were incubated as described above, with the exception that the binding buffer used was 50 mM Hepes (pH 7.4), 100 mM NaCl, 1 mM MgCl₂, protease inhibitors, and 1% (wt/vol) Triton X-100. Beads were spun and washed four times with ice-cold binding buffer and once with ice-cold binding buffer lacking detergent. Bound proteins were analyzed either by immunoblotting or immunoprecipitation–recapture followed by SDS-PAGE and fluorography. For all experiments, GST alone and GST–GGA3-GAE (for nonspecific binding) were used as controls.

The S. cerevisiae strain AH109 (CLONTECH Laboratories, Inc.) was transformed by the lithium acetate procedure as described in the instructions for the MATCHMAKER two-hybrid kit (CLONTECH Laboratories, Inc.). For colony growth assays, AH109 transformants were resuspended in water to 0.1 OD₆₀₀/ml, and then 5 μl was spotted on plates lacking leucine, tryptophan, and histidine in the presence or absence of adenine, and containing 10 mM 3-amino-1,2,4-triazole (Fluka Chemie AG) and incubated at 30°C for 4–5 d.

Two distinct human transcripts encoding homologues of the D. melanogaster stoned B protein were identified through searches of GenBank. The complete ORFs for these proteins, herein referred to as stonin 1 and stonin 2, were obtained by a combination of 5′ and 3′ RACE and PCR amplification. Unlike the D. melanogaster stoned transcripts, the human stonin transcripts were monocistronic and did not encode stoned A–related products. In fact, our searches did not turn up homologues of D. melanogaster stoned A in any other organism. The human stonin 1 transcript encodes a protein of 735 amino acids and a predicted molecular mass of ∼83 kD, whereas the stonin 2 transcript encodes a protein of 702 amino acids and a predicted molecular mass of ∼79 kD. Residues 1–711 of stonin 1 are almost identical to residues 1–711 of the 1,182–amino acid product of the SALF transcript (Upadhyaya et al. 1999). This transcript has been proposed to encode a protein with an NH₂-terminal region homologous to the D. melanogaster stoned B protein and a COOH- terminal region homologous to the general transcription factor TFIIAα/β (Upadhyaya et al. 1999). It has been suggested that SALF is a product of two separate but adjacent human genes, which is formed from a pre-mRNA transcript that initiates at an upstream Stoned B–like factor gene and continues into the ALF gene (Han et al. 2001). The stonin 1 transcript reported here differs from the SALF transcript in that it has an inframe termination codon after the stonin 1 sequence and does not contain the TFIIAα/β–homologous sequence. The existence of the stonin 1 transcript was confirmed by RT-PCR from total RNA of multiple cell lines and by PCR amplification from human heart and spleen libraries. Orthologues of both human stonin 1 and stonin 2 were found in the mouse (Martina, J.A., and J.S. Bonifacino, unpublished observations). Single homologues of these proteins exist in D. melanogaster (i.e., stoned B) and C. elegans (i.e., hypothetical protein C27H6.1), but none in yeast and prokaryotes.

The two human stonins, as well as their D. melanogaster and C. elegans counterparts, exhibit a modular structure consisting of an NH₂-terminal proline- and serine-rich domain (herein referred to as proline-rich domain), a central domain specific to the stonins (stonin homology domain), and a COOH-terminal domain homologous to the signal-binding domain of the μ subunits of AP complexes (μ-homology domain) (Fig. 1 A). The proline-rich domains of the human, D. melanogaster, and C. elegans homologues are of variable length and exhibit little or no significant sequence homology to one another, except for the abundance of proline and serine residues. The proline-rich domain of D. melanogaster stoned B displays seven NPF motifs (Andrews et al. 1996; Stimson et al. 1998), which in general have been shown to mediate interactions with Eps15 homology (EH) domains (Salcini et al. 1997). Despite the lack of overall homology in this region, the proline-rich domain of stonin 2, but not that of stonin 1, has two NPF motifs separated by 13 amino acids (Fig. 1 B). A similar arrangement of NPF motifs is found in epsin 1 and epsin 2 (Fig. 1 B) (Chen et al. 1998; Yamabhai et al. 1998; Hussain et al. 1999; Rosenthal et al. 1999). Not only are the NPF motifs in stonin 2 and the epsins similarly spaced but they also flank an intervening sequence containing additional conserved residues. Most notable is the presence of an L residue after the first NPF motif, a feature that has been shown to enhance interaction with certain EH domains (de Beer et al. 2000). Another short stretch of amino acids in the proline-rich domain of stonin 2 is highly similar to a stretch in the proline-rich region of amphiphysin 2 (Fig. 1 B) (Leprince et al. 1997; Ramjaun et al. 1997; Wigge et al. 1997).

Next to the proline-rich domain lies an ∼140–amino acid domain (stonin homology domain) that is conserved in both human stonins, as well as their D. melanogaster and C. elegans homologues (Fig. 1A and Fig. C). This domain has not been described in other proteins and may thus be unique to members of the stonin family. Immediately adjacent to this domain in all stonin homologues is an ∼350–amino acid domain homologous to the YXXØ signal-binding domain of μ2 (Aguilar et al. 1997; Owen and Evans 1998), a subunit of the heterotetrameric AP-2 complex (Fig. 1A and Fig. C). A similar domain is found in the related μ1 (A and B isoforms), μ3 (A and B isoforms), and μ4 subunits of the heterotetrameric AP complexes, AP-1, AP-3, and AP-4, respectively (for review see Hirst and Robinson 1998; Bonifacino and Dell'Angelica 1999). These theoretical analyses thus indicate that the stonins share several structural features with known components of the sorting machinery.

Because mutations in the D. melanogaster stoned locus result in severe neurological dysfunction (Petrovich et al. 1993; Andrews et al. 1996; Fergestad et al. 1999), it was of interest to determine the pattern of tissue and cell expression of the human stonins. Northern blot analysis revealed expression of a 6.5-kb stonin 1 mRNA and a 11.3-kb stonin 2 mRNA in most human tissues examined (Fig. 2 A). To study expression of the corresponding proteins, we raised polyclonal antibodies to their proline-rich domains. These antibodies were used to identify the proteins by immunoprecipitation–recapture from metabolically labeled cells (Bonifacino and Dell'Angelica 1998). The antibodies to stonin 1 and stonin 2 were found to immunoprecipitate major species of ∼81 and ∼88 kD, respectively, in reasonable agreement with their predicted molecular weights (Fig. 2 B). The bands detected by these antibodies were specific, as they were not detected using preimmune sera (PI) or antisera absorbed with the corresponding immunogens (Fig. 2 B). Use of these antibodies allowed us to demonstrate expression of both stonins in neuronal (H4) as well as nonneuronal (RD4, M1, HeLa, and 293T) human cells (Fig. 2 C). Thus, both stonins are ubiquitously expressed in human tissues, suggesting that their role might not be restricted to the nervous system.

Fractionation of metabolically labeled HeLa cells showed that stonin 1 was mostly recovered in a cytosolic fraction (C), with a small amount being associated with membranes (M) (Fig. 3 A). Stonin 2 was present in equal amounts in the cytosolic and membrane fractions (Fig. 3 A). The membrane-bound stonin 1 and stonin 2 could be partially extracted with 1 M Tris or NaCl, and totally extracted with 0.2 M Na₂CO₃, pH 11.3 (Fig. 3 B), indicating that they behaved as peripheral membrane proteins. Interestingly, a large fraction of both membrane-bound stonins remained insoluble upon extraction with 2% (wt/vol) Triton X-100 at 4°C (Fig. 3 B). This insoluble stonin 2, however, was not associated with detergent-resistant lipid domains, as determined by its failure to float with caveolin-containing fractions in sucrose density gradients (Fig. 3 C). Thus, the stonins appear to exist in cytosolic and peripheral membrane pools.

As mentioned above, the stonins contain a COOH-terminal domain homologous to the YXXØ signal-binding domain of the μ subunits of the heterotetrameric AP complexes, AP-1, AP-2, AP-3, and AP-4 (Aguilar et al. 1997; Owen and Evans 1998). However, the stonins lack another domain present in the μ subunits that mediates assembly with the β subunits of the corresponding AP complexes (β-binding domain; Fig. 1 C, BBD) (Aguilar et al. 1997). In agreement with this, we did not observe coprecipitation of stoichiometric amounts of either stonin 1 or stonin 2 with subunits of any of the four AP complexes (data not shown). Analysis of the cytosolic stonins by sedimentation velocity analysis on sucrose gradients (Marks 1998) revealed that both proteins sedimented as if they were much smaller than a typical heterotetrameric AP complex, AP-1 (sedimentation coefficient 7.7S, molecular mass ∼270 kD) (Fig. 3 D). In fact, the size of both stonins corresponded to that of monomers or small oligomers.

To gain insight into the possible function of the human stonins, we sought to identify some of their binding partners. The presence of two NPF motifs in the proline-rich domain of stonin 2 pointed to possible interactions with proteins containing EH domains. Moreover, the existence of two similarly spaced NPF motifs in epsin 1 and epsin 2, as well as the conservation of additional amino acids in the intervening sequence of these proteins, suggested that the EH-containing binding partners could be shared with the epsins. Both epsin 1 and epsin 2 have been shown to bind Eps15 and intersectin 1 via NPF–EH domain interactions (Chen et al. 1998; Yamabhai et al. 1998; Hussain et al. 1999; Rosenthal et al. 1999; Sengar et al. 1999). To examine if the stonins also interact with these proteins, we performed yeast two-hybrid analyses. GAL4 DNA binding domain (GAL4bd) constructs encoding the three NH₂-terminal EH domains from Eps15 or the related Eps15R protein, or the two NH₂-terminal EH domains from intersectin 1 (reviewed in Di Fiore et al. 1997; Mayer 1999), were coexpressed with GAL4ad constructs encoding stonin 1, stonin 2, or mutants of these proteins (Fig. 4 A). Stonin 1 did not interact with any of the EH domain constructs, as expected from the absence of NPF motifs in this protein (Fig. 4 A). Stonin 2, on the other hand, interacted with the EH domains of Eps15 and Eps15R, and more weakly with those of intersectin 1 (Fig. 4 A). Deletion of the μ-homology domain from stonin 2 enhanced growth of the cotransformants (Fig. 4 A, Stn2_PRD+SHD), probably due to increased expression of the truncated protein (data not shown). Mutation of the first NPF motif to either NAA or APA in the context of the stonin 2_PRD+SHD construct decreased interactions with the EH domains of Eps15 and Eps15R and completely abrogated interactions with the EH domains of intersectin 1. Mutation of the second NPF motif to NAA, on the other hand, had little effect on the interactions. Finally, mutation of both NPF motifs to NAA completely abolished interactions with all EH domains (Fig. 4 A). All the positive and negative controls yielded the expected results, thus verifying the specificity of the interactions observed. These experiments demonstrated that stonin 2 has the ability to interact with Eps15, Eps15R, and intersectin 1 via the two NPF motifs in its proline-rich region.

The interaction of stonin 2 with Eps15 was confirmed using in vitro binding assays. An in vitro translated, [³⁵S]methionine-labeled construct comprising the proline-rich and stonin homology domains of stonin 2 tagged with the Myc epitope was found to bind to a GST fusion protein having the three EH domains from Eps15, but not to GST (Fig. 4 B). Mutation of either or both NPF motifs largely abolished binding. This suggested that both NPF motifs are required for binding in vitro in apparent contradiction with the results of two-hybrid analyses, which showed that the first NPF motif was more important for interactions. This discrepancy could be explained by the greater sensitivity of yeast two-hybrid assays to low-affinity interactions. Similar results were obtained by transiently expressing the same stonin 2 constructs in transfected HeLa cells rather than by in vitro translation (Fig. 4 C). Furthermore, GST fused to the EH domains of Eps15 or intersectin 1 were able to bind endogenous stonin 2, but not stonin 1, from a HeLa cell lysate (Fig. 4 D).

Next, we investigated if the endogenous stonin 2, and Eps15 formed a complex in vivo. This was done by immunoprecipitation of HeLa cell lysates with either preimmune sera (PI) or antibodies to stonin 1 or stonin 2, and then blotting for Eps15 (Fig. 5 A). In agreement with the two-hybrid and in vitro binding assays, we could demonstrate association of stonin 2, but not stonin 1, with Eps15 in vivo. A similar experiment could not be performed for intersectin 1 because the antibody to this protein did not work well for immunoblotting. To circumvent this problem, we performed another experiment in which HeLa cells were transfected with plasmids encoding either GFP or GFP–stonin 2. Intersectin 1 was then isolated by immunoprecipitation, and the immunoprecipitates were analyzed by immunoblotting with antibody to GFP (Fig. 5 B). Using this approach, we could demonstrate coprecipitation of intersectin 1 with GFP–stonin 2, but not GFP (Fig. 5 B).

To map the structural determinants for Eps15–stonin 2 interactions in vivo, we transfected HeLa cells with constructs encoding stonin 1, stonin 2, or mutants of stonin 2 fused to GFP (Fig. 5 C). We then immunoprecipitated the GFP fusion proteins and tested for the presence of Eps15 by immunoblotting. Again, we found that Eps15 coprecipitated with the stonin 2, but not stonin 1, fusion protein. This in vivo interaction was mediated by the proline-rich domain of stonin 2 and required the two NPF motifs within this domain (Fig. 5 C). Since Eps15 has multiple DPF motifs in its COOH-terminal region that bind to the α-adaptin subunit of AP-2 (Benmerah et al. 1996), we also tested for the presence of α-adaptin in the anti-GFP–stonin immunoprecipitates. As expected, we found that α-adaptin was present only in the immunoprecipitates that contained Eps15 (Fig. 5 C), suggesting that AP-2 is indirectly associated with stonin 2 via Eps15.

Despite the overall homology of the μ-homology domain of the stonins to the YXXØ-binding domain of AP–μ subunits, residues involved in interactions of μ2 with YXXØ signals (Owen and Evans 1998) (denoted by blue dots in Fig. 1 C) are not conserved in the stonins. In line with this observation, yeast two-hybrid assays revealed that neither human stonin was able to bind a panel of 10 different YXXØ signals (data not shown). μ2 (Haucke et al. 2000) as well as D. melanogaster stoned B (Phillips et al. 2000) have been shown to bind to the C2B domain of synaptotagmin I. We found that 4.9–6.5% of Myc–stonin 2 expressed by transient transfection into HeLa cells bound to GST fusion proteins bearing the cytoplasmic domains of synaptotagmins I and II (Fig. 6 A). This binding was judged to be specific, as only a negligible amount of Myc–stonin 2 was recovered on a GST–GGA3 fusion protein (Fig. 6 A) or GST (data not shown). Additional GST pull-down assays revealed that Myc–stonin 2 interacted mainly with the C2B domain of both synaptotagmins (Fig. 6 B).

Although our antibodies to stonin 2 allowed us to immunoprecipitate the endogenous protein, they were not sensitive enough for immunofluorescence microscopy. This prompted us to examine the localization of GFP–stonin 2 expressed by transfection into cells. We observed a heterogeneous distribution including dispersed punctate and juxtanuclear structures superimposed over a diffuse cytoplasmic pattern (Fig. 7A, Fig. D, and Fig. G). Some of the punctate structures colocalized with AP-2 (Fig. 7, M–O, inset). At the highest expression levels, we observed a coalescence of GFP–stonin 2 in the juxtanuclear area (Fig. 7 M). In these cells, we noticed a depletion of AP-2 from plasma membrane clathrin-coated pits and its colocalization with the juxtanuclear GFP–stonin 2 (Fig. 7N and Fig. O, arrow). This behavior of stonin 2 was similar to that reported for epsin 2 (Rosenthal et al. 1999), consistent with the ability of both proteins to interact directly or indirectly with AP-2. Overexpression of GFP–stonin 2 had no effect on the distribution of two other AP complexes, AP-1 and AP-3 (data not shown), indicating that the effects on AP-2 were specific.

The ability of stonin 2 to interact with several components of the endocytic machinery predicted that overexpression of GFP–stonin 2 could interfere with receptor-mediated endocytosis, as is the case for epsin 2 (Rosenthal et al. 1999). Indeed, we observed that GFP–stonin 2 overexpression inhibited internalization of Cy3-transferrin (Fig. 7, A–C), rhodamine-EGF (D–F), and Dil-LDL (G–I) by their corresponding receptors. Overexpression of GFP, on the other hand, had no effect on internalization of Cy3-transferrin (Fig. 7, J–L) or the other ligands (data not shown). These observations further emphasize the functional connection of stonin 2 to the endocytic machinery.

Here, we report the identification of two ubiquitous human proteins, stonin 1 and stonin 2, which have a modular structure characteristic of components of the endocytic machinery. In all of our protein interaction analyses, we could only demonstrate weak binding of stonin 1 to synaptotagmins I and II. Thus, at present we do not know if the nature of the stonin 1 modules is indicative of a role in endocytosis or other protein sorting processes. We were more successful, however, in identifying several proteins that interact with stonin 2 (i.e., Eps15, Eps15R, intersectin 1, and synaptotagmins I and II). All of these interaction partners have been previously implicated in endocytosis, which makes it likely that stonin 2 is also a component of the endocytic machinery. In support of this notion, overexpression of stonin 2 perturbs the localization of AP-2 to the plasma membrane and inhibits internalization of several endocytic receptors.

The proline-rich domain of stonin 2 contains two NPF motifs that mediate interactions with the EH domains of Eps15, Eps15R, and intersectin 1. This property is shared with other NPF-containing endocytic proteins such as epsin 1 (Chen et al. 1998; Yamabhai et al. 1998), epsin 2 (Yamabhai et al. 1998; Rosenthal et al. 1999), the 170-kD form of synaptojanin (Haffner et al. 1997), nonneuronal members of the AP180/CALM family (Dreyling et al. 1996; Tebar et al. 1999), several SCAMP isoforms (Fernández-Chacón et al. 2000), and Numb (Santolini et al. 2000). Despite the fact that all of these NPF-containing proteins share the same EH-containing partners, interactions between them and Eps15 need not be competitive since Eps15 is a tetramer (Cupers et al. 1997; Tebar et al. 1997) and could thus assemble simultaneously with several NPF-containing proteins. The proline-rich domains of stonin 1 and stonin 2 also contain several PxxP motifs that could potentially interact with SH3 domains of endocytic proteins such as intersectins, amphiphysins, and endophilins (for review see Slepnev and De Camilli 2000), or with other domains that recognize proline-based motifs (Kay et al. 2000).

The μ-homology domains of stonin 1 and stonin 2 are likely to fold into an all β-sheet structure similar to that of the YXXØ-binding domain of the μ2 subunit of AP-2 (Owen and Evans 1998). This similarity notwithstanding, the human stonins do not appear to bind either tyrosine- or dileucine-based endocytic signals in the yeast two-hybrid system (data not shown). The possibility remains that they could recognize a different type of signal. Nonetheless, the stonins do share with μ2 the ability to bind the C2B domain of synaptotagmins I and II, most likely through a conserved interface distinct from the YXXØ-binding site (Haucke et al. 2000). Synaptotagmins have been proposed to function as docking proteins for the AP-2 complex, in what would constitute the first step in the assembly of clathrin coats onto the plasma membrane (Zhang et al. 1994). By analogy, synaptotagmins could also function to recruit stonin 2 to membranes via C2B–μ-homology domain interactions. The NPF motifs of stonin 2 would then serve as ligands for EH-containing proteins such as Eps15, Eps15R, and intersectin 1. These proteins would, in turn, congregate all of their other binding partners (e.g., epsins, dynamins, synaptojanins, endophilins, and syndapins) at the membrane, and assembly of a clathrin coat and vesicle formation would ensue.

The AP-2 complex is similarly able to bind to synaptotagmin I via its α and μ2 subunits (Zhang et al. 1994; Haucke et al. 2000) and to Eps15 and Eps15R via its α subunit (Benmerah et al. 1996; Salcini et al. 1999). It is thus plausible that the role of stonin 2 could be partly redundant or alternative to that of AP-2. Another possibility is that stonin 2 and AP-2 cooperate to stabilize the dynamic network of low-affinity interactions required to assemble a functional endocytic coat. The coprecipitation of endogenous AP-2 with stonin 2, which is likely mediated by Eps15 (Fig. 5), argues in favor of the latter explanation. The fact that mutations in either the α subunit of AP-2 (Dornan et al. 1997; González-Gaitán and Jackle 1997) or stoned B (Andrews et al. 1996; Fergestad et al. 1999) in D. melanogaster result in defective synaptic vesicle recycling further suggests that both AP-2 and stoned B (of which stonin 2 appears to be the most closely related human homologue) play unique and essential roles in endocytosis.

In addition to providing additional linking potential to the assembling coats, stonin 2 could be involved in the selection of specific cargo molecules for incorporation into the budding vesicles. In favor of this hypothesis is the observation that synaptotagmin is mislocalized in D. melanogaster stoned mutants (Fergestad et al. 1999). Moreover, synaptotagmin appears to be degraded more rapidly in stoned mutants (Fergestad et al. 1999). In contrast, the localization and stability of other synaptic vesicle proteins are unaltered in stoned mutants (Fergestad et al. 1999). These observations suggest that stoned B could play a role in the specific retrieval of a subset of synaptic vesicle proteins including synaptotagmin. The dual roles of stonin 2 and stoned B in coat assembly and cargo selection need not be mutually exclusive, as the AP-2 complex is known to participate in both processes.

The neurological defects of D. melanogaster stoned mutants are likely due to perturbation of the high endocytic activity of presynaptic terminals. Although less obvious, other endocytic processes in stonin-deficient organisms may be defective as well. The expression of stonin 2 in most human tissues and cells examined, its ability to interact with several widely-expressed proteins, and the overexpression-induced interference with receptor-mediated endocytosis in nonneuronal cells, all suggest that stonin 2 may play a general role in endocytosis.

We thank Xiaolin Zhu for excellent technical assistance; P.P. Di Fiore, P. McPherson, and T. McGraw for gifts of reagents; and E. Dell'Angelica and M. Marks for critical review of the manuscript.

C.J. Bonangelino was supported by a Pharmacology Research Associate fellowship.