Generation of high curvature membranes mediated by direct endophilin bilayer interactions

A major pathway of synaptic vesicle retrieval after exocytosis is clathrin-mediated endocytosis (Heuser and Reese, 1973). This specialized version of a general endocytic pathway is necessary for the maintenance of the synaptic vesicle pool, and perturbation of this process results in rapid depletion of synaptic vesicles (Koenig and Ikeda, 1989; Shupliakov et al., 1997; Gad et al., 2000). The 100-kD GTPase, dynamin, has been implicated in the “fission” stage of clathrin-mediated endocytosis (Koenig and Ikeda, 1989), and disruption of dynamin function results in an accumulation of electron-dense “collars” at the tubular neck of deeply invaginated clathrin-coated buds (Koenig and Ikeda, 1989; Takei et al., 1995). These collars are enriched with dynamin, as well as two of the major SH3 domain–containing dynamin binding partners at the synapse: endophilin 1 (endophilin) and amphiphysin 1 (amphiphysin) (David et al., 1996; de Heuvel et al., 1997; Ringstad et al., 1997). Recombinant dynamin 1 and amphiphysin have been shown to deform artificial lipid bilayers into narrow tubules, both independently as well as cooperatively in a complex, likely reflecting a role for these proteins in membrane binding and deformation at the neck of clathrin-coated pits (Takei et al., 1998, 1999).

Endophilin has been implicated in many stages of clathrin-mediated synaptic vesicle endocytosis, from early events generating membrane curvature, to later stages such as vesicle fission and uncoating (Ringstad et al., 1999; Gad et al., 2000). Selective depletion of endophilin from rat brain cytosol inhibited the generation of synaptic-like microvesicles (SLMVs)^* in broken PC12 cells, and markedly reduced the formation of dynamin-coated tubules on synaptic membranes in the presence of ATP/GTPγS (Ringstad et al., 1999; Schmidt et al., 1999). Therefore, we tested whether endophilin plays a direct role in the generation of membrane curvature in a similar fashion to dynamin and amphiphysin. Furthermore, we tested whether this property was shared by other proteins with homology to endophilin.

Using a liposome-based assay previously shown to support generation of coated intermediates of clathrin-mediated endocytosis (Takei et al., 1998, 1999), purified recombinant endophilin efficiently and robustly deformed lipid bilayers into tubules with outer diameters ranging between 20–100 nm (Fig. 1 A). The tubule surfaces were decorated by tightly packed thin transverse striations, reminiscent of the coat observed on tubules generated by recombinant amphiphysin (Fig. 1 A, see inset for a comparison of the endophilin, amphiphysin, and dynamin coats on tubules) (Takei et al., 1999). The recombinant pleckstrin homology (PH) domain of PLCδ, a known phospholipid binding domain, did not deform the lipid bilayers into tubes, thereby distinguishing this activity from nonspecific changes due to protein membrane interactions (Fig. 1 A). Endophilin binding to liposomes appeared to saturate at a molar ratio of under 1:200 (protein to lipid) in a liposome sedimentation assay (Fig. 1 B), and displayed a nonlinear loss of binding with increasing salt (Fig. 1 C).

Deletion constructs of endophilin defined amino acids 1–125 as both necessary and sufficient for tubulation (Fig. 1 D). Correspondingly, this 125–amino acid fragment was also seen to cosediment with liposomes, whereas a construct that lacked this fragment stayed in the supernatant (Fig. 1 D). In the presence of a chemical cross-linker, bis(sulfosuccinimidyl) suberate (BS³), the endophilin NH₂-terminal fragment cross-linked into higher order oligomers in a liposome-stimulated manner, whereas in the absence of liposomes, the fragment cross-linked only to a dimer (Fig. 1 E). Similar results were obtained using a second cross-linker, EDC, and with the full length protein (unpublished data). The presence of coated striations on the tubules and the cross-linking data together suggest that endophilin generates bilayer curvature by oligomerizing on the surface of liposomes.

The property of endophilin to deform the lipid bilayer is independent from its reported lysophosphatidic acid (LPA)–acyl transferase (AT) activity (Schmidt et al., 1999). Endophilin tubulated synthetic liposomes devoid of the putative substrates for the LPA-AT reaction, arachidonoyl-CoA and LPA (Schmidt et al., 1999) (Fig. 1 F). Furthermore, endophilin tubulated liposomes in a reaction incubated on ice, minimizing any potential enzymatic contribution (Fig. 1 F, inset). Thus, bilayer deformation may represent a thermodynamically favorable transition mediated by endophilin polymerization at the liposome surface, rather than by the active enzymatic generation of specific membrane microdomains.

Incubation of liposomes, purified clathrin coat proteins, and endophilin-generated tubules associated with clathrin-coated buds, consistent with a potential role for this protein in generating tubular membrane curvature at the neck of the nascent clathrin-coated pit (Fig. 1 G). Similar observations were made for amphiphysin (Takei et al., 1999). Of note, amphiphysin has binding sites for clathrin and the endocytic clathrin adaptor, AP2 (Slepnev et al., 2000), but we have not observed binding to clathrin or to clathrin adaptors by endophilin.

Since endophilin, like amphiphysin, interacts with dynamin via its SH3 domain, we investigated whether dynamin and endophilin could form a complex on liposomes, similar to what has been described for dynamin and amphiphysin (Takei et al., 1999). The coat generated by affinity-purified dynamin 1 on bilayer tubules was represented by rings with an average spacing of 132 ± 5 Å (Fig. 2 A), in agreement with the previously reported ring spacing determined by electron diffraction (Sweitzer and Hinshaw, 1998). Liposomes incubated with both recombinant endophilin and purified dynamin created a different ring morphology (Fig. 2 B, and lower inset). The rings were thicker, and had an average spacing of 202 ± 15 Å, a difference of >50% compared with the rings formed by dynamin alone (Fig. 2, A and B). This is similar to the ring morphology generated by the dynamin–amphiphysin complex (Takei et al., 1999) and by brain cytosol with ATP/GTPγS (Takei et al., 1995, 1998). Such a finding supports the presence of a multiprotein complex in the thick electron-dense rings observed at the neck of clathrin-coated pits on synaptic membranes, and explains the colocalization of endophilin with dynamin on coated tubules formed by incubating synaptic membranes with brain cytosol and ATP/GTPγS (Fig. 2 B, upper inset) (Ringstad et al., 1997, 1999). Both the NH₂-terminal lipid binding domain of endophilin and its COOH-terminal dynamin-binding SH3 domain were necessary for the formation of the complex with dynamin on the membranes, since neither domain alone could reproduce the phenomenon (Fig. 2 C).

We were intrigued by the similarities in the membrane deformation properties of endophilin and amphiphysin (Fig. 3 D), and in the tubular coats which they form (Fig. 1 A, inset). The two proteins have an overall similar domain structure, each with a phylogenetically highly conserved NH₂-terminal region comprising predicted α-helices and coiled-coils, and a COOH-terminal SH3 domain which binds dynamin and synaptojanin (Fig. 3 B). Furthermore, the membrane-tubulating properties of both proteins map to their respective NH₂-termini (Takei et al., 1999; and this study). We explored whether endophilin and amphiphysin share any primary sequence similarity within this region. A BLAST alignment of full-length endophilin 1 and amphiphysin 1 (both rat) identified the highest scoring region to be a 29 amino acid stretch (41% identity and 75% similarity) within the first 35 amino acids of endophilin and the first 41 amino acids of amphiphysin (Fig. 3 A). Deleting most of this region of homology abolished binding to liposomes (Fig. 3 C) and tubulation (unpublished data) for both endophilin and amphiphysin. Secondary structure algorithms (nnpredict and PSIPRED) for this 29 amino acid stretch predict α-helices and random coils. When plotted on a helical wheel, a similar amphipathic pattern emerged for both proteins, with a hydrophobic patch and an opposite hydrophilic face consisting of several basic residues (Fig. 3 A). This feature may allow this domain to interact with phospholipid headgroups via the hydrophilic face, and then to partially embed into the bilayer via the hydrophobic patch in a manner which favors membrane deformation. A point mutation of a conserved hydrophobic residue to an acidic residue (phenylalanine to glutamate) within this putative hydrophobic patch in endophilin abolished both lipid binding and tubulation, whereas a conservative hydrophobic to hydrophobic point mutation (phenylalanine to tryptophan) of the same residue preserved both lipid binding and tubulation (Fig. 3, E and F).

Given the similar morphology of amphiphysin- and endophilin-coated tubules both in the presence and absence of dynamin, we compared their individual effects on dynamin-mediated membrane fragmentation. Dynamin-coated membrane tubules readily fragment into small vesicles in the presence of GTP (Sweitzer and Hinshaw, 1998; Takei et al., 1999). As reported previously, amphiphysin supports this fragmentation (Takei et al., 1999). Accordingly, after an incubation of liposomes with dynamin-GTP, or dynamin with amphiphysin-GTP, a large number of small vesicles were seen as evidenced by electron microscopy (Fig. 4, A and B). Any tubules which remained were generally much shorter. In contrast, when liposomes were incubated with dynamin and endophilin-GTP, fragmentation was inhibited, with a large number of long tubules persisting (Fig. 4, A and B). This result suggests that the presence of endophilin stabilizes the lipid tubule against the structural changes in dynamin which lead to vesiculation in the presence of GTP. Both endophilin and amphiphysin inhibited phosphate release by dynamin-GTP in the presence and absence of liposomes (Fig. 4 C). These results may reflect unique dynamics of dynamin oligomers in coassembly with endophilin versus amphiphysin: one which allows for a mechanochemical transduction productive for vesiculation, and another which hinders this. Therefore, despite important similarities, the dynamin–endophilin interaction is qualitatively different from the dynamin–amphiphysin interaction on lipid bilayers. This difference may have significant ramifications for the in vivo dynamics of these complexes at the neck of clathrin-coated pits.

We searched the database for other proteins which have sequence homology to the NH₂ terminus of endophilin 1 to determine whether they also deform lipid bilayers. The search revealed the recently named endophilin B family of proteins (Huttner and Schmidt, 2000). This protein family shares a common domain structure with endophilin, including a COOH-terminal SH3 domain (EMBL/GenBank/DDBJ accession nos. AF263364 and AF263293) (Fig. 5 A). Antibodies were raised against the recombinant endophilin B1 protein. By Western blot analysis, this antibody recognized a 41-kD band mainly in heart, brain, spleen, and lung, with lower expression in liver and testis (Fig. 5 C). Additional higher molecular weight bands were also detected, specifically in brain, possibly representing other splice isoforms or posttranslational modifications (EMBL/GenBank/DDBJ accession no. AF263364) (Fig. 5 C). By immunofluorescence of brainstem frozen sections, endophilin B was detected in a synaptic-like pattern as shown by colocalization with synaptic markers such as amphiphysin (which also shows diffuse cytosolic staining) (Fig. 5 D). In addition, endophilin B staining was detected on intracellular membranes in the perikaryon-dendritic region of neurons (Fig. 5 D). Further analysis of this staining showed partial colocalization with the Golgi complex marker GM-130 (Fig. 5 E). In CHO cells, endophilin B localized to a perinuclear compartment strongly positive for the Golgi complex marker GM-130, in addition to a more diffuse reticular staining pattern throughout the cell (Fig. 5 F). Brefeldin A caused dispersion of the perinuclear staining by endophilin B, consistent with localization to the Golgi complex (unpublished data). There was no colocalization to endosomal compartments labeled with transferrin (unpublished data).

The purified recombinant human endophilin B1 protein tubulated liposomes in a manner morphologically indistinguishable from endophilin A1 and amphiphysin (Fig. 5 G). The amphipathic region of homology between endophilin (now referred to as endophilin A1) and amphiphysin is also conserved in endophilin B1, and deletion of this region in endophilin B1 also blocks tubulation (unpublished data). Thus, endophilin B1 may be involved in tubulovesicular dynamics of the Golgi complex and potentially of other intracellular membrane compartments, underscoring a diverse role for proteins which share this membrane tubulating property.

Collectively, our results provide evidence that endophilin directly binds and deforms lipid bilayers into narrow tubules within the size range of the tubular neck of clathrin-coated pits. This activity of endophilin is a protein-mediated phenomenon which likely occurs by oligomerization at the surface of the bilayer. Endophilin tubulates membranes at 4°C, and also tubulates synthetic bilayers devoid of the putative substrates of its reported LPA-AT activity, showing that formation of high curvature membranes by endophilin can be independent of lipid-modifying enzymatic activity. We did observe that endophilin binds and tubulates liposomes made from a brain lipid extract more efficiently than synthetic liposomes, thus indicating some role for the composition of the bilayer (unpublished data). However, elucidation of this finding requires further experimentation. We hypothesize that the tubulating activity of endophilin reflects one possible function for this protein in clathrin-mediated endocytosis. Along with dynamin and amphiphysin, endophilin may be part of a protein complex governing aspects of membrane deformation and fission.

Our findings provide a rationale for the observed requirement for endophilin in the tubulation of synaptic membranes by brain cytosol in the presence of GTPγS (Ringstad et al., 1999). These observations may also partially explain the arrest of clathrin-mediated synaptic vesicle retrieval at shallow stages of invagination upon perturbation of endophilin function in lamprey (Ringstad et al., 1999). Endophilin membrane binding and protein–protein interactions may serve to localize and functionally orient the protein at the clathrin-coated pit. Amphiphysin and endophilin tubulate lipid bilayers both independently and as a complex with dynamin; however, their distinct effects on dynamin in vitro suggest a potential differential modulatory role on dynamin function in vivo which needs to be explored further.

We find that endophilin B1, a member of a newly discovered family of proteins, localizes to both synaptic and Golgi/intracellular membrane compartments, and displays a membrane tubulating activity similar to that of endophilin A1 and amphiphysin. All three proteins share a short NH₂-terminal amphipathic region which we show to be required for membrane tubulation. This may have implications for a shared function for this motif in tubulo-vesicular dynamics in cellular trafficking events. Recently, RICH (RhoGAP interacting with CIP4 homologues), a novel RhoGAP containing protein, was found in a yeast two-hybrid screen for CIP 4 (CDC-42–interacting protein 4) interacting proteins (Richnau and Aspenstrom, 2001). RICH has an NH₂-terminal domain with homology to endophilin A1, and the amphipathic motif of endophilin A1 is also conserved. A splice isoform of RICH which only contains the endophilin homology domain (RICH 1B) localizes to intracellular punctate compartments (Richnau and Aspenstrom, 2001). It will be interesting to see whether RICH also has a membrane-interacting/tubulating function, further expanding the diversity of proteins with this motif in membrane dynamics.

Liposomes composed of brain lipid extract (type 1, Folch fraction 1; Sigma-Aldrich) were made as described (Takei et al., 1999). Synthetic liposomes were made using HPLC-purified DOPC (40%), DOPS (40%), and DOPE (20%) from Avanti Polar Lipids, Inc.

Dynamin was affinity purified from rat brain cytosol using amphiphysin 1–SH3 domain as described (Owen et al., 1998). This material, as assessed by SDS-PAGE, yielded ∼90–95% dynamin, and 5–10% synaptojanin. All data shown use this material. Synaptojanin was immunodepleted from this affinity-purified material using a monoclonal antibody to yield a more highly purified dynamin preparation, and similar results were obtained.

Recombinant rat endophilin A1, human amphiphysin 1, human endophilin B1, and PLCδ-PH were cloned in pGEX (Amersham Pharmacia Biotech), and purified as glutathione S-transferase (GST) fusion proteins according to standard methods (Amersham Pharmacia Biotech). For endophilin, amphiphysin, and endophilin B1, the GST tag was subsequently cleaved by PreScission Protease (Amersham Pharmacia Biotech).

Endophilin deletion constructs were prepared by PCR to yield the following fragments: amino acids 1–125, 1–261,126–352, and 34–352. The constructs were subcloned and purified as above. Point mutants were generated using mutant primers and plasmid extension by PCR.

Clathrin-coat fraction was purified from bovine calf brain as described (Takei et al., 1999).

Liposomes (0.1 mg/ml final) were incubated at 37°C for 10–20 min in buffer A (25 mM Hepes-KOH, pH 7.4, 25 mM KCl, 2.5 mM Mg²⁺ acetate, 150 mM K-glutamate) with various proteins and nucleotides at the following final concentrations: dynamin 0.1 mg/ml, amphiphsyin 0.1 mg/ml, endophilin 0.1 mg/ml, endophilin deletion constructs 0.1–0.2 mg/ml, coat proteins 0.5 mg/ml, PLCδ-PH 0.1 mg/ml, GTP 1 mM, and GTPγS 0.5 mM (Takei et al., 1999). At the end of the incubation, aliquots were adsorbed onto 200–400 μM formvar- and carbon-coated copper EM grids for 3–5 min at room temperature, washed in 0.1 M Hepes, pH 7.4, stained in 1–2% uranyl acetate, blotted, and then allowed to air dry. For incubations at 4°C, all of the above was performed on ice.

Average ring spacing was determined by counting the number of rings in 10 350-nm tubules, and then dividing by the tubule length. To quantify the number of tubules seen under various conditions, 6 13-μm² areas of the grid were scored. Each independent tubule and each branch of a tubule was counted as a tubule.

Liposome sedimentation was performed using 100 μg sucrose-loaded liposomes incubated with 5–10 μg protein (1–25 μg for saturable binding) in 400–500 μl buffer A or in Hepes-KCl, pH 7.4 (to determine salt sensitivity) for 10–20 min at 37°C. Liposomes were sedimented at 100,000 g in a Beckman Coulter TLA 100.3 rotor for 20 min, the supernatant was thoroughly removed, and sedimented liposomes were solubilized in 2% SDS. To monitor recovery, liposomes were labeled with 0.5% NBD–phosphatidyl choline and absorbance was measured at 460 nm. In some cases, proteins in the pellet and supernatant were concentrated by chloroform precipitation and methanol extraction. Samples were subjected to SDS-PAGE and analyzed by either Coomassie staining or by slot blot analysis using affinity-purified polyclonal endophilin antibodies and ¹²⁵I. Band intensities were quantified by optic densitometry or phosphorimaging.

For the crosslinking assay, 4 μg of the endophilin 125 amino acid fragment (with or without 8 μg liposomes) was preincubated in Hepes, pH 7.4, 100 mM KCl for 20 min at 37°C, then an equal volume of 2× buffer with the heterobifunctional chemical cross-linker bis(sulfosuccinimidyl) suberate (BS³) was added to achieve final concentrations of 0, 1, 5, and 10 mM cross-linker. The mixture was incubated for an additional 30 min at 37°C, and the samples were then prepared for SDS-PAGE analysis.

Immunogold labeling of endophilin was performed by standard agarose-embedding/labeling protocols as described (Ringstad et al., 1999) followed by thin sectioning. The GTPase activity of dynamin was determined by phosphate release as described (Takei et al., 1999). Western blotting and immunofluorescence was performed using standard procedures as described (Ringstad et al., 1997).

We are most grateful to Drs. Graham Warren and Vinzenz Unger for many helpful discussions and critical reading of the manuscript. We also wish to thank Drs. Gilbert Di Paolo, Lorenzo Pellegrini, Markus Wenk, Vladimir Slepnev, and Gianluca Cestra for technical advice and helpful discussions.

This work was supported in part by National Institutes of Health grants NS36251 and CA46128 to P. De Camilli. K. Farsad and S.R. Floyd are supported in part by an Medical Scientist Training Program fellowship from the National Institutes of Health. N. Ringstad was supported in part by a predoctoral training grant (T32GM07527) from the National Institutes of Health to the Department of Cellular and Molecular Physiology, Yale University School of Medicine.