Essential role of PACSIN2/syndapin-II in caveolae membrane sculpting

Bin-amphiphysin-Rvs167 (BAR) domain-containing superfamily members, including extended Fes-CIP4 homology (EFC)/FCH-BAR (F-BAR) domain-containing subfamily members, such as FCHo2, FBP17, Toca-1, CIP4, PACSINs/syndapins and others, sculpt membranes according to the curvature of their three-dimensional structures (Henne et al., 2010; Peter et al., 2004; Shimada et al., 2007; Shimada et al., 2010). The positively charged surface of the domain binds to the negatively charged inner surface of the plasma membrane, thereby bending the membrane according to its protein structures. The various BAR domain surface curvatures are thought to correlate with the diverse invaginations and protrusions of cells. The BAR domain proteins engaged in the formation of clathrin-coated pits, filopodia and spines have been well characterized, in contrast to those participating in forming other subcellular structures (Doherty and McMahon, 2009; Itoh and De Camilli, 2006; Suetsugu et al., 2010; Takenawa and Suetsugu, 2007).

PACSINs (protein kinase C and casein kinase substrate in neurons proteins) and syndapins, which include three paralogs (Modregger et al., 2000), consist of an F-BAR domain and an SH3 domain. The SH3 domain binds to dynamin and the actin-nucleating protein N-WASP (Itoh et al., 2005; Qualmann and Kelly, 2000; Simpson et al., 1999; Tsujita et al., 2006). PACSIN1/syndapin-I is implicated in synaptic vesicle recycling in the brain (Qualmann et al., 1999), and PACSIN3 expression is specific to muscles (Modregger et al., 2000). Although PACSIN2/syndapin-II is expressed ubiquitously (Modregger et al., 2000), its functions remain to be elucidated.

The membrane tubulation activity of the F-BAR domain is autoinhibited in full-length PACSIN1 (Wang et al., 2009). We found that this autoinhibition is also conserved in PACSIN2, since full-length PACSIN2 exhibited weaker membrane tubulation ability than the F-BAR domain fragment (Shimada et al., 2010). The crystal structure of full-length PACSIN1 suggested that the autoinhibition is mediated by an interaction between the SH3 and F-BAR domains (Rao et al., 2010).

The F-BAR domain of PACSIN2 possesses a deeper concave surface than those of FBP17 and CIP4 (Shimada et al., 2010; Wang et al., 2009). Correspondingly, the membrane tubules induced by the PACSIN2 F-BAR domain have smaller diameters (~50 nm) than those induced by the F-BAR domains of FCHo2, FBP17, CIP4 and Toca-1 (Shimada et al., 2010; Wang et al., 2009). Caveolae are flask-shaped invaginations with diameters of approximately 50 nm to 100 nm (Mundy et al., 2002; Palade and Bruns, 1968; Parton and Simons, 2007; Rothberg et al., 1992; Yamada, 1955). This diameter size might correspond to that of the tubules induced by the PACSIN2 F-BAR domain.

Dynamin is also localized at the neck of caveolae (Henley et al., 1998; Oh et al., 1998), and is involved in the fission of caveolae during their endocytosis. The recently identified cavin family proteins, including polymerase I and transcript release factor (PTRF)/cavin-1, reportedly promote caveola biogenesis (Hill et al., 2008). Caveolin-1 is suggested to induce the formation of membrane tubules in the absence of PTRF (Verma et al., 2010).

In the present study, we analyzed the function of PACSIN2-induced tubulation in caveolae. We found that the cellular tubules induced by the F-BAR domain of PACSIN2 contained caveolin-1, but less PTRF. Surprisingly, caveolin-1 interacted directly with the F-BAR domain of PACSIN2 to release its autoinhibition, thus allowing membrane tubulation. Abnormally-shaped caveolin-1-containing plasma membrane invaginations were also observed in PACSIN2-knockdown cells. The cellular tubulation observed in cells expressing the F-BAR domain or with PACSIN2 knockdown might be caused by the lack of recruitment of dynamin-2 to caveolae.

We examined the localization of endogenous PACSIN2, by labeling HeLa cells with anti-PACSIN2 and anti-caveolin-1 antibodies (Fig. 1A,B). A proportion of the endogenous PACSIN2 or caveolin-1 puncta (44±13% or 48±14%, respectively) also exhibited caveolin-1 or PACSIN2 staining. Importantly, exogenous green fluorescent protein (GFP)–PACSIN2 and caveolin-1–DsRed exhibited 90±5% colocalization, when these two proteins were coexpressed in HeLa cells (Fig. 1C). Similar colocalization of endogenous PACSIN2 and endogenous caveolin-1 was observed in other cells, such as Hs578T, SK-BR-3 and BT549 cells (supplementary material Fig. S1).

To further demonstrate the localization of PACSIN2 at caveolae, we examined non-transfected HeLa cells by electron microscopy. A plasma membrane sheet was prepared from HeLa cells and was labeled with anti-caveolin-1 and anti-PACSIN2 antibodies, followed by immunogold. The PACSIN2 label was found in close proximity to the caveolin-1 label (Fig. 1D,E). The PACSIN2 label was also observed at structures lacking the caveolin-1 label, which were smaller than caveolae (Fig. 1E). Thin-section electron microscopy revealed anti-PACSIN2 antibody staining in the necks of the flask-like invaginations typical of caveolae (Fig. 1F). Staining signals were also detected on the relatively flat plasma membrane and in the cytosol (Fig. 1F). No signals were identified in the absence of antibody (supplementary material Fig. S2). These data suggest that PACSIN2 functions in caveolae at the plasma membrane.

The canonical F-BAR-domain-containing proteins, such as FBP17 and Toca-1, induce plasma membrane invagination when overexpressed in cells. These invaginations are often observed as the mesh-like localization of the overexpressed F-BAR domain proteins. The overexpressed F-BAR domain fragment of PACSIN2 exhibited a similar mesh-like localization in HeLa cells (Fig. 2). We confirmed that this mesh-like structure corresponded to the plasma membrane invaginations, by labeling the cells with membrane-staining FM dye and the culture medium with sulforhodamine (supplementary material Fig. S3).

We then examined the localization of various membrane marker proteins to the PACSIN2 F-BAR-induced mesh-like structures, and found that caveolin-1 colocalized with PACSIN2. In cells with low expression of the PACSIN2 F-BAR domain (Fig. 2A, left cell in large panel), the F-BAR domain and caveolin-1 colocalized along the tubules induced by F-BAR domain expression. However, caveolin-1 accumulated into several larger assemblies, which formed at the convergence of the PACSIN2-induced invaginations in cells with high F-BAR-domain expression (Fig. 2A, right cell in large panel). This caveolin-1 accumulation was not observed in cells expressing PACSIN2 F-BAR domains with tubulation-defective mutations (M124T/M125T) (Shimada et al., 2010) or in those expressing FBP17 (supplementary material Fig. S4A,B). Importantly, the clathrin distribution was not altered by expression of the PACSIN2 F-BAR domain (supplementary material Fig. S4C). PTRF was only partially localized at tubules induced by the PACSIN2 F-BAR domain (supplementary material Fig. S4D).

Interestingly, the overexpressed PACSIN1 F-BAR domain had a weak effect on caveolin-1 localization, whereas the overexpressed PACSIN3 F-BAR domain had a similar effect on caveolin-1 localization as the PACSIN2 F-BAR domain (Fig. 2B,C).

The substantial colocalization of PACSIN2 with caveolin-1 upon their coexpression suggested a direct interaction between these two proteins. PACSIN2 consists of the F-BAR domain and the SH3 domain (Fig. 3A). Caveolin-1 consists of two cytoplasmic regions connected by a region embedded in the membrane (Fig. 3A). We prepared several purified fragments of caveolin-1, and analyzed their binding to purified full-length PACSIN2 in a pull-down assay. The N-terminal cytoplasmic region of caveolin-1 (residues 1–100 or 61–100) interacted with PACSIN2 (Fig. 3B). However, the association was not observed for the C-terminal cytoplasmic region of caveolin-1 (Fig. 3B). We then mapped the binding region of PACSIN2 to caveolin-1. Similarly to the finding that the PACSIN2 F-BAR domain alone was sufficient for colocalization with caveolin-1 (Fig. 2), the F-BAR domain alone was sufficient for binding to caveolin-1 (Fig. 3C). The binding of caveolin-1 to the F-BAR domain was stronger than that to full-length PACSIN2 (Fig. 3C). The PACSIN2 C-terminal region, including the SH3 domain (residues 301–486), did not interact with caveolin-1 in a pull-down assay with purified proteins (supplementary material Fig. S5A). The association of PACSIN2 and caveolin-1 in cells was confirmed by immunoprecipitation analysis with an anti-caveolin-1 antibody (supplementary material Fig. S5B).

We then examined the binding of the PACSIN1 and PACSIN3 F-BAR domains to caveolin-1. Consistent with the weak effect of the PACSIN1 F-BAR domain expression on caveolin-1 localization, the PACSIN1 F-BAR domain had weak affinity to caveolin-1. By contrast, the PACSIN3 F-BAR domain bound well to caveolin-1 (Fig. 3D).

Because the F-BAR domain binds to the membrane, we examined the binding surface of the PACSIN2 F-BAR domain for caveolin-1 with those of the F-BAR domains bearing either the R50D or M124T/M125T mutation, on the membrane-binding concave surface, and the F-BAR domain bearing the R245E mutation, on the convex surface (Fig. 3E). The binding was not altered by the R50D or M124T/M125T mutation, but it was weakened by the R245E mutation (Fig. 3C and supplementary material Fig. S5C). Therefore, the PACSIN2 F-BAR domain appears to bind to both caveolin-1 and the membrane simultaneously, for the induction of caveolin-1-localized tubules. Interestingly, the amino acid residues conserved between PACSIN2 and PACSIN3, but not conserved in PACSIN1, were mapped on the convex surface of PACSIN2 (Fig. 3E and supplementary material Fig. S6).

We then addressed the effect of caveolin-1 fragment binding to the F-BAR domain in the liposome association of the PACSIN2 F-BAR domain. The caveolin-1 1–100 amino acid (aa) fragment is known to form oligomers, which precipitated upon centrifugation (supplementary material Fig. S5D) (Sargiacomo et al., 1995). Therefore, the binding of the caveolin-1 fragment to liposomes could not be examined (supplementary material Fig. S5D). However, in the presence of the PACSIN2 F-BAR domain, but in the absence of liposomes, the caveolin-1 fragment did not precipitate after centrifugation, confirming the physical interaction between the PACSIN2 F-BAR domain and the caveolin-1 fragment. The presence of the caveolin-1 1–100 aa fragment did not affect the liposome binding of the PACSIN2 F-BAR domain (supplementary material Fig. S5D). The membrane tubulation induced by the F-BAR domain was not affected by the presence of the caveolin-1 fragment (data not shown).

To confirm the caveolin-1 binding to PACSIN2 on the tubulated membrane, we expressed the caveolin-1 fragment with PACSIN2 in cells. The caveolin-1 fragment was found on the PACSIN2-induced tubular structures, indicating the interaction of PACSIN2 with the caveolin-1 fragment on the tubulated membrane (supplementary material Fig. S7). Consistently, the cellular tubulation induced by the expression of the PACSIN2 F-BAR domain was decreased by the knockdown of caveolin-1, suggesting that caveolin-1 enhances the membrane localization of PACSIN2 F-BAR domain in cells (supplementary material Fig. S8).

We next addressed the autoinhibition of PACSIN2 through the intramolecular interaction between its SH3 and F-BAR domains, by pull-down assay. We found that this interaction was decreased remarkably by incubation with the caveolin-1 fragment (aa 1–100) (Fig. 3F). The SH3 domain of PACSIN2 is also known to bind to dynamin, and the binding to dynamin is also suggested to affect the intramolecular interaction. The intramolecular interaction was further decreased by incubation with the dynamin-derived PxxP peptide and the caveolin-1 fragment (Fig. 3F).

We also examined the tubulation induced by full-length PACSIN2 in vitro. We found that the tubulation was markedly strengthened in the presence of the caveolin-1 fragment (Fig. 3G,H). The tubulation occurred more efficiently in the presence of both the caveolin-1 fragment and PxxP peptide (Fig. 3G,H). The caveolin-1 fragment alone, or the PxxP peptide alone, had no effect on liposome morphology (Fig. 3G,H). Therefore, the membrane interaction might be stimulated by the release of the SH3 domain from the F-BAR domain, thus allowing membrane tubulation in the presence of caveolin-1.

We then examined whether this mechanism functioned in cells. We expressed dynamin-2 proline-rich peptide or the caveolin-1 fragment with PACSIN2 in cells, and the enhancement of PACSIN2-induced tubulation was observed, supporting the disruption of the intramolecular interaction by these two fragments (supplementary material Fig. S7). The overexpression of full-length caveolin-1 with PACSIN2 also enhanced PACSIN2-induced tubulation (supplementary material Fig. S7).

We also co-expressed the wild-type or dominant-negative mutant of dynamin-2 (K44A) with the wild-type PACSIN2 in HeLa cells, and quantified the tubulation formation. PACSIN2-expressing cells with the dynamin-2 K44A mutant exhibited more tubulation than the cells expressing PACSIN2 alone (supplementary material Fig. S9). The PACSIN2-expressing cells with wild-type dynamin-2 displayed less tubulation than the cells expressing PACSIN2 alone (supplementary material Fig. S9). The co-expression of another SH3 binding protein, N-WASP, minimally affected the tubulation (supplementary material Fig. S9). These results suggested that the tubules induced by PACSIN2 could be activated by K44A dynamin-2 or dynamin-2 proline-rich peptide, but were antagonized by wild-type dynamin-2, presumably as a result of its membrane scission activity.

The results mentioned above indicated that PACSIN2 mediates membrane tubulation upon binding to caveolin-1, thus sculpting the membranes of caveolae. Therefore, we performed RNA interference (RNAi) of PACSIN2 and analyzed the morphology of the caveolin-1-associated membranes by electron microscopy. HeLa cells treated with small interfering RNA (siRNA) to knock down PACSIN2 contained significantly less PACSIN2 (Fig. 4A). The amounts of caveola-related proteins, including dynamin-2, caveolin-1 and PTRF/cavin-1, were not affected by RNAi of PACSIN2 (Fig. 4A). We then analyzed the morphology of caveolae in cells treated with PACSIN2 siRNA. These cells were labeled with an anti-caevolin-1 antibody and then with a colloidal gold-labeled secondary antibody. Portions of the plasma membrane with caveolin-1 in PACSIN2 siRNA cells showed non-flask-shaped invaginations (Fig. 4B,C). To confirm this morphological defect, the plasma membrane associated with caveolin-1 was traced (Fig. 4D,E). Because PACSIN2 was localized at the neck of caveolae (Fig. 1F), the neck width of the caveolin-associated invaginations was measured (Fig. 4F). The average diameter at the neck was 62±2 nm in control siRNA cells, whereas it was 79±3 nm in PACSIN2 siRNA cells (values are mean ± s.e.m.; P<0.05 by Student's t-test), clearly indicating the function of PACSIN2 at the caveolae neck. The depths of the caveolin-associated invaginations were also measured (Fig. 4G). The depths were 106±6 nm in control cells and 189±9 nm in PACSIN2 siRNA cells. Although the change in the depth of caveolae was not very large, for observations by light microscopy, the increase in the depth might suggest the failure of proteins, such as dynamin, to interact with caveolae.

Finally, we addressed the possible defects in dynamin recruitment to caveolae in PACSIN2 siRNA cells. The caveolae associated with dynamin-2 are believed to execute endocytosis, and thus the quantification of dynamin-2 localization at caveolin-1 spots would be difficult. Therefore, we examined the localization of the dynamin-2 K44A mutant at caveolin-1 spots, in control and PACSIN2 siRNA cells. Many caveolin-1 spots associated with dynamin K44A were observed in the control siRNA cells, whereas very few were detected in the PACSIN2 siRNA cells (Fig. 5A,B).

To confirm the defect in dynamin-2 recruitment upon RNAi of PACSIN2, the cells were briefly treated with an inhibitor of dynamin, dynasore, to prevent the scission of caveolae into vesicles. The cells were then stained with anti-caveolin-1 and anti-dynamin-2 antibodies. The number of caveolin-1 spots colocalized with dynamin-2 significantly decreased upon PACSIN2 RNAi (Fig. 5C–E). The defect in dynamin-2 recruitment to caveolae in PACSIN2 siRNA cells was then examined by electron microscopy. The plasma membrane sheets of these dynasore-treated cells were prepared and then stained with these antibodies. The caveolin-1 signals in control siRNA cells colocalized with the dynamin-2 signals in the plasma membrane sheets. By contrast, these signals were not colocalized in the plasma membrane sheets of the PACSIN2 siRNA cells; the dynamin signals were observed at non-caveolin-1-associated structures in these cells (Fig. 5F,G). The distribution of caveolin-1 signals on the plasma membrane was abnormal in the PACSIN2 siRNA cells, as observed by thin-section electron micrography (Fig. 4C, Fig. 5G).

We then examined whether PACSIN2 is involved in caveola-mediated endocytosis. We examined cholera toxin B (CTxB) incorporation into control and PACSIN2 siRNA cells, as well as PACSIN2 siRNA cells expressing wild-type or mutants of PACSIN2. CTxB is known to bind to GM1 gangliosides, which are enriched at caveolae. The incorporated CTxB accumulated at the center of the cells, whereas the unincorporated CTxB remained at the cell periphery or at the plasma membrane. The CTxB incorporation was observed in control cells, and it was significantly decreased in cells treated with PACSIN2 siRNA (supplementary material Fig. S10). The expression of the F-BAR domain fragment of PACSIN2 induced tubular membrane invaginations with CTxB localization, supporting the idea that the F-BAR-domain-induced tubules are related to caveolae. The incorporation was restored by the expression of wild-type PACSIN2, but not by the expression of either the R245E or M124T/M125T mutant of PACSIN2 (supplementary material Fig. S10).

These results provide several lines of evidence indicating that PACSIN2 mediates the shape formation of caveolae. First, the caveolin-1-associated plasma membrane was not flask-shaped in PACSIN2 siRNA cells (Fig. 4). Second, the presence of PACSIN2 at the neck of caveolae appeared to correspond to the membrane tubulating ability in vivo (Fig. 1). Therefore, the membrane-sculpting activity of the PACSIN2 F-BAR domain seems to be responsible for caveola morphology.

PACSIN2 appears to recruit dynamin-2 for caveola fission, because fewer caveolin-1 spots containing dynamin-2 were detected in PACSIN2 siRNA cells, and the PACSIN2-induced tubules were antagonized by dynamin-2 (Fig. 5 and supplementary material Fig. S9). Elongated tubules that contained caveolin-1 were observed in the PACSIN2 siRNA cells (Fig. 4). Because PACSIN2 was knocked down, the elongated tubules were probably formed by proteins other than PACSIN2 in the absence of a sufficient amount of dynamin. Caveolin-1 reportedly binds directly to dynamin (Yao et al., 2005). Thus, an insufficient amount of dynamin-2 might induce deeper invaginations without membrane scission.

The PACSIN2 F-BAR domain binds to negatively charged phosphatidylserine at the plasma membrane. Phosphatidylinositol-4,5-bisphosphate [PtdIns(4,5)P₂] is also negatively charged, and facilitates the membrane binding of the PACSIN2 F-BAR domain (Dharmalingam et al., 2009). Interestingly, PtdIns(4,5)P₂ is enriched in caveolae (Fujita et al., 2009). In addition, the Eps15-homology-domain containing (EHD) protein, which also possesses membrane-deformation ability, reportedly binds to the Asn-Pro-Phe (NPF) motif of PACSIN2 (Braun et al., 2005; Xu et al., 2004). The EHD proteins are highly conserved eukaryotic ATPases that are implicated in caveolae formation, clathrin-independent endocytosis and recycling from endosomes, and they have membrane-tubulation ability (Daumke et al., 2007; Verma et al., 2010). An EHD protein bound to caveolae might cooperate with PACSIN2 in caveola biogenesis or caveola-mediated endocytosis. In addition, EHD proteins are present in PTRF/cavin-1-rich fractions (Aboulaich et al., 2004). Thus, the EHD proteins might function as bridges between cavins and PACSIN2.

Cavins are believed to facilitate the membrane shape formation of caveolae, by suppressing the formation of caveolin-1-localized long tubules (Hansen et al., 2009; Hill et al., 2008; McMahon et al., 2009; Verma et al., 2010). The limited participation of PTRF in the tubulation induced by the PACSIN2 F-BAR domain appears to be consistent with this inhibition of tubulation by PTRF (supplementary material Fig. S4E).

PACSIN2 was first identified as a casein kinase and protein kinase C (PKC) substrate (Plomann et al., 1998), and PKC is enriched at caveolae (Smart et al., 1995). This might also provide a key to regulate the caveola invagination mediated by PACSIN2. The detailed time course of the recruitment of these proteins should be examined in the future, to clarify the regulation of caveola formation.

Our present results reveal that the PACSIN2 F-BAR domain protein induces membrane tubulation for caveola sculpting and fission. In addition, caveolin-1 was found to regulate the sculpting ability of PACSIN2. Caveolae constitute platforms for various receptors, channels, and signal transduction. PACSIN2 therefore appears to be a fundamental protein that is involved in cellular homeostasis and disease (e.g. in generation of tumors), through the regulation of caveolae. Interestingly, the PACSIN3 F-BAR domain was associated with caveolin, whereas the PACSIN1 F-BAR domain was not. The different affinities of the PACSIN/Syndapin paralogs to caveolin-1 might reflect the tissue-specific variations of caveolae function.

Glutathione S-transferase (GST)–PACSIN2 (mouse), GST–PACSIN2 F-BAR domain (EFCL; aa 1–339 and EFCS aa 1–306), GST–SH3 domain (aa 301–486 or aa 387–486), GST–PACSIN1 F-BAR domain (mouse) (aa 1–306), GST–PACSIN3 F-BAR domain (mouse) (aa 1–306) and GST–caveolin-1 fragments (mouse) (amino acids 1–100, 61–100, and 137–179) were expressed in Escherichia coli, as described previously (Shimada et al., 2010). GST was removed by PreScission Protease (GE Healthcare). Dynamin-1 proline-rich peptide, RSPTSSPTPQRRAPAVPPARPG, was from Bex (Tokyo, Japan). Pull-down assays with the indicated fragments were performed in buffer [20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM phenylmethylsulfonyl fluoride (PMSF), 10% glycerol, and 5 mM ethylenediaminetetraacetic acid (EDTA)] containing 2% Triton X-100 and the indicated protein concentrations. The purified proteins and glutathione Sepharose beads were mixed in the buffer for 1 hr at 4°C. After washing, the bound proteins were visualized by western blotting or Coomassie Brilliant Blue (CBB) staining after sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). EFCL was used as the PACSIN2 F-BAR domain for all of the assays, except for the comparison with PACSIN1–PACSIN3 (Fig. 3D).

Liposomes were prepared from total bovine brain lipids (Folch fraction 1; Avanti Polar Lipids) (Michelsen et al., 1995). Lipids were dried under nitrogen gas in glass test tubes and were resuspended in XB (10 mM HEPES, pH 7.9, 100 mM KCl, 2 mM MgCl₂, 0.2 mM CaCl₂ and 5 mM ethylene glycol tetraacetic acid/EGTA), containing 100 mM sucrose, by mixing with a vortexer, and then hydrated at 37°C for 1 hour. This preparation yielded a mixture of unilamellar liposomes with various diameters (0.1–2 μm) and large, multilamellar vesicles, as determined by thin-section electron microscopy. The majority of the liposomes were unilamellar. Liposome tubulation was examined by negative staining, as described previously (Shimada et al., 2010). Proteins without GST were used in tubulation assays. The co-sedimentation assay was performed as follows. The proteins were incubated with liposomes (1 mg/ml) in 50 μl XB for 20 minutes at room temperature (RT), and were then centrifuged at 78,000 g for 30 minutes at 25°C in a TLA-100 rotor (Beckman). Supernatants and pellets were subjected to SDS-PAGE.

HeLa and Hs578T cells were cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal calf serum (FCS), penicillin, and streptomycin. SKBR-3 and BT549 cells were cultured in Roswell Park Memorial Institute (RPMI) medium supplemented with 10% FCS, penicillin, and streptomycin.

Subconfluent HeLa cells in a 15 cm dish were washed with PBS, and suspended in buffer containing 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM PMSF, 10% glycerol, 2% Triton X-100, 5 mM NaF, 5 mM EDTA, 1 μg/ml aprotinin and 1 μg/ml leupeptin. The resulting cell lysate was sonicated and centrifuged at 15,000 r.p.m. at 4°C. The supernatant was incubated with 1 μg of anti-caveolin-1 (7C8) antibody or control IgG. After incubation for 1 hour at 4°C, Dynabeads protein A (Invitrogen) were added and further incubated for 1 hour at 4°C. The beads were then washed and subjected to western blot analysis.

Ggreen fluorescent protein (GFP), Venus and mCherry labelled PACSIN1, PACSIN2 and PACSIN3 proteins were prepared by subcloning the mouse Pacsin1, Pacsin2 and Pacsin3 cDNA into the pEGFP-C1, pVenus-C1, and pmCherry-C1 vectors (Clontech), in which the GFP in pEGFP-C1 was replaced with Venus and mCherry, respectively (Nagai et al., 2002; Shaner et al., 2004). Venus is a GFP mutant with brighter fluorescence. Venus was labeled as GFP, for simplicity. The PACSIN2–GFP and PACSIN2–mCherry proteins were respectively expressed by the pEGFP-N3 vector (Clontech) and the pmCherry-N3 vector, in which the GFP in pEGFP-N3 was replaced with mCherry. The caveolin-1–GFP was expressed by subcloning cDNA encoding mouse caveolin-1 into the pEGFP-N3 or pDsRed-Monomer-N1 vector. The wild-type and K44A dynamin-2 constructs were gifts from Mark A. McNiven (Orth et al., 2002). The wild-type, actin-polymerization-defective ΔVCA, and the partially active phospho-mimic Y253E were described previously (Sasaki et al., 2000; Suetsugu et al., 2002). The proline-rich region of dynamin-2 (aa 741–870) was subcloned into the pCMV-Tag3B vector (Stratagene) with a Myc tag. The siRNAs to knockdown caveolin-1 and PACSIN2 were Stealth Select RNAi™ siRNAs, Invitrogen catalog nos. 1299003 and 1299001, respectively, containing three oligo siRNAs that were mixed and transfected simultaneously. The control RNA for siRNA was Stealth RNAi Negative Control Duplexes (Invitrogen). Transfection was performed with the Lipofectamine LTX and PLUS reagents (Invitrogen), according to the manufacturer's protocols. Dynasore (Sigma, D7693) was applied to DMEM + 10% FCS at 40 μM for 5 minutes before fixation.

HeLa cells were seeded on glass-bottom dishes (Asahi Glass, 3911-035), transfected with the vector expressing the GFP-tagged F-BAR domain of PACSIN2, and cultured overnight. Sulforhodamine (Invitrogen, S359) was then added to the medium at a 0.1 mM final concentration. After 10 minutes, cells were visualized by confocal microscopy (FV1000D, Olympus) at 25°C. FM-dye labeling was performed basically as described previously (Terebiznik et al., 2002). HeLa cells were seeded on glass-bottom dishes, transfected, and cultured overnight. The cells were then chilled on ice for 15 minutes, washed with HBSS supplemented with 10 mM glucose, and incubated with 10 μM FM4-64 on ice for 1 minute. After the medium was removed, the cells were washed twice with ice-cold HBSS supplemented with 10 mM glucose for 5 minutes, and then the cells were observed by confocal microscopy at 25°C.

The cells were fixed in 3.7% formaldehyde in PBS for 5 minutes, followed by permeabilization with TBS supplemented with 1% BSA and 0.1% Triton X-100 for 5 minutes and blocking with TBS supplemented with 1% BSA for 2 hours. The anti-PACSIN2 antibody was affinity purified from the serum of rabbits immunized with the F-BAR domain protein of PACSIN2. The rabbit monoclonal anti-caveolin-1 antibody was purchased from Cell Signaling Technology. The mouse monoclonal anti-Myc (clone PL14) and the anti-caveolin-1 (7C8) antibodies were purchased from MBL International, and Santa Cruz Biotechnology, respectively. The anti-clathrin heavy chain (clone 23) and anti-dynamin-2 antibodies were obtained from BD Transduction Laboratories. After washing, cells were stained with Alexa Fluor 488 and/or Alexa Fluor 568 labeled secondary antibodies and Alexa Fluor 633 phalloidin. Fluorescence images were obtained by confocal microscopy (Olympus Fluoview 1000D) at room temperature. A 100× oil-immersion objective (NA 1.45; Olympus) was used.

The spots of certain protein staining that were recognized as particles of higher signals than the surrounding areas in each image were counted. Subsequently, if the spots contained other protein staining signals that were higher than the area surrounding the spots, then the spots were considered to be colocalization spots of the two proteins. This analysis was performed manually, with the aid of the Cell Counter Plug-in and Channel Tools of the ImageJ program (NIH).

HeLa cells were labeled with Alexa-Fluor-555-labeled Cholera toxin B (CTxB) (Invitrogen) on ice, as described previously (del Pozo et al., 2005). Briefly, cells were chilled on ice for 15 minutes, labeled with 1 μg/ml CTxB on ice for 15 minutes, washed twice with ice-cold PBS, and incubated with DMEM + 10% FCS for 60 minutes. The cells were then fixed with 4% paraformaldehyde in PBS. The localization of CTxB close to the coverslip, including the basal plasma membrane, was then visualized by confocal microscopy. The cells with CTxB at the cell periphery without significant CTxB accumulation at the cell center were considered as cells without CTxB incorporation.

Cells on plastic coverslips (Thermanox, Nalge Nunc International) were fixed in 2% paraformaldehyde and 2% glutaraldehyde for 5 minutes at room temperature. The cells were permeabilized with 0.1% saponin and 1% bovine serum albumin (BSA) in Tris-buffered saline for 20 minutes. The cells were blocked overnight with 1% BSA in Tris-buffered saline and then incubated with the anti-PACSIN2 antibody and a Nanogold rabbit Fab fragment (Nanoprobes). The stained cells were postfixed with 1% glutaraldehyde in phosphate-buffered saline and blocked, and then the gold signal was enhanced by Goldenhance (Nanoprobes). After the gold-enhancement reaction, the cells were washed, stained with 1% osmium tetroxide and 1% potassium ferrocyanide, washed, dehydrated in 50%, 60%, 70%, 80%, 90%, 95.5% and 100% ethanol, placed in propylene oxide, and embedded in the epoxy resin Quetol-812, according to the manufacturer's instructions (Nissin EM). After polymerization of the resin, 80 nm sections were prepared and counterstained for 10 minutes in uranyl acetate and for 2 minutes in lead citrate. Dried sections were examined by transmission electron microscopy.

Plasma membrane sheets were prepared and observed as described previously (Vinten et al., 2001), and were stained with an anti-PACSIN2 rabbit polyclonal antibody and an anti-caveolin-1 (7C8) mouse monoclonal antibody, or an anti-caveolin-1 rabbit monoclonal antibody and anti-dynamin-2 mouse monoclonal antibody, followed by 5-nm- or 10-nm-gold-conjugated antibodies.

All statistical analyses were performed using Microsoft Excel. Significance was assessed by the Student's t-test. All of the images are representative of at least three independent experiments.

We thank Mark A. McNiven (Mayo Clinic, Rochester, MN) for the dynamin-2 constructs. We are grateful to Toyoshi Fujimoto for critical reading of the manuscript. We thank Tadaomi Takenawa, Kazuya Tsujita and Tsukasa Oikawa (Kobe University, Kobe, Japan) for the caveolin-1 and FBP17 expression vectors. This work was supported, in part, by a Grant-in-Aid from the Ministry of Education, Culture, Sports, Science and Technology of Japan and by Grants-in-Aid from the Japan Science and Technology Corporation (J.S.T.), the Uehara Memorial Foundation, the Mochida Memorial Foundation for Medical and Pharmaceutical Research and the Inamori Foundation.