Structural elucidation of how ARF small GTPases induce membrane tubulation for vesicle fission

Significance

The formation of intracellular transport vesicles involves the recruitment of coat proteins to compartmental membranes to initiate vesicle budding followed by vesicle fission that releases buds as transport vesicles. The ARF small GTPases play central roles in these mechanistic steps for multiple intracellular pathways. A molecular understanding of how ARF proteins act in coat recruitment and vesicle budding is being achieved through structural studies, but a similar level of understanding of its role in vesicle fission has remained elusive. Our study addresses this outstanding issue.

Abstract

ADP-Ribosylation Factor (ARF) small GTPases have been found to act in vesicle fission through a direct ability to tubulate membrane. We have pursued cryoelectron microscopy (EM) to reveal at 3.9 Å resolution how ARF6 assembles into a protein lattice on tubulated membrane. Molecular dynamics simulation studies confirm and extend the cryo-EM findings. The ARF6 lattice exhibits features that are distinct from those formed by other membrane-bending proteins. We identify protein contacts critical for lattice assembly and how membrane insertion results in constricted tubules. The lattice structure also enables docking by GTPase-activating proteins (GAP) to achieve vesiculation. We have also modeled ARF1 onto the ARF6 lattice, and then pursued vesicle reconstitution by the Coat Protein I (COPI) complex to further confirm that the ARF lattice acts in vesicle fission. By elucidating how an ARF protein tubulates membrane at the structural level, we have advanced the molecular understanding of how this class of transport factors promote the fission stage of vesicle formation.

ADP-Ribosylation Factor (ARF) small GTPases act in intracellular vesicular transport, a fundamental cellular process that is critical for the proper localization of proteins and lipids within the cell. Early studies identified a regulatory role for ARF proteins, which involves their recruitment of coat complexes onto membrane compartments to initiate the formation of vesicular carriers (1, 2).

This role has been best characterized for ARF1 and an ARF-like protein known as Sar1. At the Golgi, ARF1 recruits the Coat Protein I (COPI) complex to form vesicular carriers for transport from the Golgi to the endoplasmic reticulum (ER) (3, 4). ARF1 also recruits adaptor proteins (AP) (5–7), which act in bidirectional transport between the Golgi and endosomal compartments and also transport among endosomal compartments (8). At the ER, Sar1 recruits the COPII complex for transport from the ER to the Golgi (9).

ARF6 has been found to act in endocytosis (10) and endocytic recycling (11). At the recycling endosome, ARF6 recruits a clathrin complex that uses ACAP1 (ArfGAP with Coiled-Coil, Ankyrin Repeat, and PH Domain 1) and Akt as coadaptors for recycling to the plasma membrane (12, 13). However, in contrast to the case of COPI regulated by ARF1 and COPII regulated by Sar1, for which vesicle formation has been reconstituted to enable mechanistic details to be elucidated (4, 9), a similar achievement for the ARF6-regulated coat has not been attained.

A molecular understanding of how ARF proteins recruit coat complexes to membrane compartments is being achieved through structural studies. Cocrystal structures of ARF1 in complex with subunits of coatomer (the core components of the COPI complex) (14), and Sar1 in complex with Sec23/24 (the inner components of the COPII complex) (15), have revealed how these prototypic ARF proteins interact with their cognate coats for membrane recruitment.

Besides having a regulatory role, ARF proteins have also been found to play effector roles in vesicle formation. Vesicle formation occurs in two general stages, vesicle budding, which involves positive membrane curvature, and then vesicle fission, which involves negative membrane curvature. Structural studies through high-resolution cryoelectron microscopy (EM) are providing a detailed molecular understanding of how ARF1 assembles with coatomer on COPI vesicles (16–18). These structural elucidations also shed insight into the budding stage of COPI vesicle formation, as this process involves the generation of positive membrane curvature. Similarly, cryo-EM studies are providing molecular insights into how Sar1 assembles with the COPII components (19–21), and how ARF1 assembles with AP1 (22) and AP3 (23), for bud formation.

Whereas these findings reveal how ARF proteins cooperate with coat complexes to achieve membrane bending, a surprising revelation has been that ARF proteins alone can also bend membrane. This ability was initially found by incubating Sar1 with liposomes generated from defined lipids, which resulted in liposome tubulation (24). Vesicle reconstitution studies have pinpointed this effect of Sar1 to act in COPII vesicle fission (24). ARF1 has also been found to induce liposome tubulation, and reconstitution studies have similarly revealed that this action of ARF1 acts in COPI vesicle fission (25). However, in contrast to their roles in coat recruitment and vesicle budding, a molecular understanding of how ARF proteins act in vesicle fission remains to be achieved. We address this gap in knowledge by elucidating at the structural level how the ARF protein is assembled into a lattice structure to explain its ability to tubulate membrane for vesicle fission.

Results

ARF6 Lattice on Tubular Membrane.

The GTP-bound forms of ARF1 and ARF6 have been found previously to induce liposome tubulation (26). We first coexpressed ARF6 and N-myristoyltransferase in bacteria to generate myristoylated ARF6 (SI Appendix, Fig. S1 A and B). We then confirmed that loading with GTP enabled ARF6 to induce liposome tubulation (SI Appendix, Fig. S1C). The GTP-bound form of ARF1 also induced tubulation (SI Appendix, Fig. S1D), with most tubules having diameters similar to that of ARF6 tubules (in the range of ~30 nm; SI Appendix, Fig. S1E). Diffraction patterns suggested that both ARF proteins assemble with helical symmetry (SI Appendix, Fig. S1 F and G). However, ARF6 had a higher efficiency in tube formation and a clearer diffraction pattern, and thus more conducive to determination of high-resolution cryo-EM structures. Therefore, we pursued further studies on ARF6 tubules.

We next vitrified samples that contain ARF6-induced tubules followed by imaging through cryo-EM. From the raw electron micrographs, we could readily observe the outer surface of tubulated liposomes to be coated with ARF6 (SI Appendix, Fig. S2A).

The power spectra of the ARF6 tubules indicated that they belong to a 5-start helical assembly with a helical rise of 28.0 Å and twist of 27.3° (SI Appendix, Fig. S2A). Combining the iterative helical real-space reconstruction approach (27) with refinement using RELION (28) (SI Appendix, Fig. S2A), we determined the 3D density map of the ARF6 assembly on tubulated membrane to 3.9 Å resolution (SI Appendix, Fig. S2 B and C and Table S1 and Movie S1), which revealed that ARF6 forms a right-handed helical tubule with outer and inner diameters of 31.4 nm and 12.0 nm, respectively (Fig. 1A and SI Appendix, Fig. S2D).

Fig. 1.

Cryo-EM reconstruction and atomic model of ARF6 coated tubule. (A) Cryo-EM map of the reconstructed ARF6-coated tubule. A single helical rung (red dashed lines) and a helical asymmetric unit (orange) are highlighted. (B) Model of ARF6 tetramer on membrane with each subunit shown in a different color: blue, cyan, green, and purple. The bound GTP molecules are shown as sticks. (C) Representative cryo-EM densities of side chains of α-helix 1 (K26-L35) (Left), GTP molecule (Center), and a strand from β-sheet 2 (F47-T53) (Right). (D) Tetramer organization. Centroid of each subunit is represented by an orange ball. Dotted lines that connect four adjacent centroids form a parallelogram. (E) ARF6 tetramers in helical arrays. Centroid of each tetramer is represented by yellow ball. Dotted lines that connect four adjacent centroids form a rhombus.

We next sought to fit previously solved crystal structures of ARF6 onto the density map. Initial rigid body fitting by Chimera (29) showed the cross correlation between the crystal structure and the experimental density to be 0.63 for ARF6 in its GTPγS-bound state [Protein Data Bank (PDB) ID: 2J5X] (30), and 0.48 in its GDP-bound state (PDB ID: 1E0S) (31). Thus, as ARF6-GTP provided the better fit, its atomic coordinates were then refined to fit on to the density map more precisely (Fig. 1 A and B). This enabled many bulky side chains to be resolved (Fig. 1C). The lattice structure was further verified by the detection of densities that correspond to GTP binding by ARF6 (Fig. 1C).

The solved structure further revealed that four ARF6 molecules assemble into a tetramer with a two-fold symmetry. When the centroids of these four ARF6 molecules within the tetramer are connected, they form a parallelogram configuration (Fig. 1D and Movie S2). The lengths of the long sides and short sides of the parallelogram are 4.1 nm and 3.1 nm, respectively. This parallelogram-shaped tetramer is the asymmetric unit of helical packing (Fig. 1 A and E). Each tetramer is oriented with the N terminus of ARF6 facing toward the membrane and the GTP-binding pocket facing the opposite direction (Movie S2). Whereas connecting the centroids inside the tetramer forms a parallelogram, connecting the centroids of four tetramers grouped from two adjacent helical rows forms a rhombus. These tetramers interlock with each other to reinforce the overall architecture of the lattice (Fig. 1E).

Protein Interfaces Critical for the ARF6 Lattice.

The solved structure predicted multiple protein interfaces critical for the ARF6 lattice assembly (Fig. 2 A and B and SI Appendix, Table S2). These interfaces can be divided into three types: i) interactions in forming the tetramer (interfaces 1, 2, and 3), ii) interactions between tetramers in the same helical row (interface 4), and iii) interactions between tetramers in adjacent helical rows (interface 5) that crosslink the helical rows (Fig. 2 A and B).

Fig. 2.

Protein interactions critical for the ARF6 lattice assembly. (A) Organization of ARF6 tetramers. Upper panel: black dashed box outlines four subunits forming a tetramer, and red dashed box outlines two adjacent tetramers. Lower panel: close-up of the tetramer assembly with secondary structures and the major protein interfaces shown. (B) Close-up of the major protein interfaces in the tetramer. Interface 1 is formed by helix α1 and switch I interacting with their counterpart in an adjacent ARF6 molecule in an antiparallel manner. Switch I is highlighted in orange. Interface 2 is formed by helix α5 and interswitch from adjacent molecules forming a symmetric interaction. Interface 3 is formed by switch I, sheet β6, and helix α5 between two adjacent molecules forming interactions. Interface 4 involves tetramer interactions within the same helical row, highlighting the role of two pairs of electrostatic interactions (D68-R145 and K69-D146). Switch II is highlighted in pink. Interface 5 involves tetramers on different helical rows interacting with each other, highlighting the role of salt bridges (E102-R110, R105-D109) forming symmetric interactions. (C) Snapshots of the major protein interfaces of the membrane-bound ARF6 tetramer at the final frame of the molecular dynamics simulations. Salt bridges are indicated by yellow dashed lines. (D) Interaction energy for each residue pair in the protein interfaces, with salt bridges and hydrogen bonds highlighted by green and blue labels, respectively. (E) EM analysis examining the effect of ARF6 mutations on liposome tubulation. Specific residues mutated are M1, Y31A/L35A/V39A; M2, Y54A/Y163A; M3, V45A/F47A/W149A/Y150A/V151A /W168A; M4, K69E/R145E; M5, R105E/R110E. Representative images are shown on Left. (Scale bar, 200 nm.) Quantitation of tubulation is shown on Right: +++ (strong), ++ (moderate), + (weak), or ± (almost none). (F) Endosomal membrane binding by ARF6 mutants. Centrifugation was performed after incubation of membrane with ARF6 proteins as indicated, with pellet (P) containing membrane-bound fraction and supernatant (S) containing soluble fraction. A representative result is shown above. The mean with SD from 3 independent experiments is shown below. Statistical analysis was performed using the Student’s t test, comparing the wild-type and the different mutants, ns P > 0.05. (G) The TfR recycling assay was performed on HeLa cells to assess the effect of ARF6 mutations on transport from the recycling endosome to the plasma membrane (PM). Representative images of endosomal TfR (red) colocalizing with a recycling endosome marker, cellubrevin (Cbv, green), at times indicated are shown on Left. Quantitation is shown on Right, comparing control with different ARF6 conditions, n = 3. A representative experiment (which examines 10 cells) is shown; ns, P > 0.05, ****P < 0.0001.

Interface 1 occurs through an antiparallel interaction between two ARF6 molecules in forming a dimer, and involves helix α1 and its connecting loop, namely Switch I (residues 36 to 47) between helix α1 and sheet β2 of one subunit interacting with that of another subunit. Interactions in this interface are mediated by hydrogen bonds (S38-T40) and hydrophobic contacts (Y31, L35, and V39) (Fig. 2B).

Interface 2 is formed by interaction between two opposing ARF6 molecules that have a two-fold symmetry and involves the tip of Interswitch I (residues 48 to 64) and helix α5. Hydrophobic residues Y163 and Y54 from these two subunits form a symmetric interaction (Fig. 2B). Additional salt bridges (K32-E164) between helix α1 and helix α5 from these neighboring subunits further stabilize the interface (Fig. 2B).

Interface 3 is formed by two ARF6 interacting through Switch I, β6 sheet, and α5 helix. Hydrophobic residues V45 and F47 of one subunit interact with residues W149, Y150, V151, and W168 of an adjacent subunit. Additional salt bridge (H76-R145) and hydrogen bonds (Q67-Q136 and V49-N172) from these neighboring ARF6 molecules further stabilize the interface (Fig. 2B).

Interface 4 is formed by two adjacent ARF6 tetramers along the same helical row interacting with each other through short contacts. This involves the loop in Switch II (residues 65 to 82, between sheet β3 and helix α2) and the loop between helix α4 and sheet β6 of one subunit interacting with those of another subunit. This interface is likely supported by electrostatic interactions (D68-R145 and K69-D146) (Fig. 2B).

Interface 5 is formed by ARF6 tetramers in adjacent helical rows interacting with each other with a two-fold symmetry. This involves α3 and η3 helices of a tetramer residing in one helical row interacting with another in a tetramer residing in an adjacent row (Fig. 2B). Salt bridges (E102-R110, R105-D109) form symmetric interactions between tetramers that reside in adjacent helical rows.

As the tetramer is the basic unit of the ARF6 lattice structure, we next pursued molecular simulation studies to further characterize the three major interfaces predicted to be critical for tetramer assembly (Fig. 2 C and D). Simulation was initiated by placing the ARF6 tetramer on top of the membrane (Movie S3). For Interface 1, the hydrophobic contacts (L35 and V39) and the hydrogen bond (S38-T40) remain stable throughout the simulation. The interaction energies of hydrogen bonds between two pairs of S38 and T40 were similar, consistent with Interface 1 having antiparallel interactions. For Interface 2, simulation highlighted particularly the importance of the salt bridge (K32-E164), as it had the highest calculated interaction energy among the predicted interactions in this interface. For Interface 3, simulation confirmed the importance of salt bridges (H76-R145) and particularly the hydrophobic interaction between F47 and W149, as they had the highest calculated interaction energies.

Functional Mutagenesis Studies Targeting the Protein Interfaces.

We then pursued functional mutagenesis studies to confirm the importance of the predicted five major interfaces. For interface 1, we targeted residues Y31, L35, and V39 (Fig. 2B). For interface 2, we targeted residues Y163 and Y54 (Fig. 2B). For interface 3, we targeted residues V45, F47, W149, Y150, V151, and W168 (Fig. 2B). These residues were mutated to alanines to eliminate hydrophobic interactions, with the mutants named M1, M2, and M3, respectively (SI Appendix, Table S3). To target interface 4, we mutated K69 and R145 (Fig. 2B). For interface 5, we mutated R105 and R110 (Fig. 2B). These residues were mutated to glutamates to eliminate charge interactions, with the mutants named M4 and M5 (SI Appendix, Table S3).

We initially performed circular dichroism (CD) spectroscopy, which revealed that the mutants maintained similar conformations as that of the wild-type form (SI Appendix, Fig. S2E). We then performed liposome studies to confirm that targeting against the interfaces reduced the ability of ARF6 to induce membrane tubulation (Fig. 2E). These effects were not due to reduced membrane binding by the mutants, as they all bound to endosomal membrane similarly as wild-type ARF6 (Fig. 2F). As ARF6 acts in endocytic recycling, we also confirmed that this transport was inhibited by the mutations. Tracking the transport of the transferrin receptor (TfR) from the recycling endosome to the plasma membrane, as previously described (13), we found that the inhibition of TfR recycling induced by siRNA against ARF6 was rescued by wild-type ARF6, but not the mutants (Fig. 2G). We also confirmed that the ARF6 mutants were expressed at levels similar to that of the wild-type (SI Appendix, Fig. S3A). Moreover, the interface mutations did not affect ARF6 localization to the recycling endosome (SI Appendix, Fig. S3B).

ARF6 Lattice Is Distinct from Those Formed by Other Proteins.

We next compared the ARF6 lattice to those previously solved for other membrane-bending proteins. These included proteins that contain the BAR domain, such as endophilin (32), ACAP1 (33), and SNX1 (34), as well as dynamin (35), which does not contain a BAR domain. The result revealed distinct features in the ARF6 lattice structure. ARF6 assembles as a 5-start helical array. In contrast, the BAR domain of endophilin (N-BAR), as well as dynamin in its GTP-bound state, are 2-start helical assemblies, while SNX1 and the BAR-PH tandem domain of ACAP1 (BAR-PH) assemble as a 1-start helical array (SI Appendix, Fig. S3C). Another distinction was that the asymmetric unit of the ARF6 assembly is a tetramer, while those of endophilin, SNX1, and dynamin are dimers and that of ACAP1 is a dimer of dimer (SI Appendix, Fig. S3C). These differences result in the ARF6 lattice having an overall architecture that could be readily distinguished from lattices formed by the other proteins, as highlighted in the cross-section views of the different coated tubules (SI Appendix, Fig. S3D). Thus, the ARF6 lattice represents a different type of protein interaction and assembly pattern that mediates membrane bending.

Positioning of the Amphipathic Helix in the ARF6 Lattice.

ARF proteins have been found to contact the membrane through two structural elements, an N-terminal myristoyl chain (MYR) and an amphipathic helix (AH) located also near the N terminus (1, 2, 26, 36). In the solved lattice structure, the myristoyl chain merges imperceptibly with the lipids of the membrane. However, the structure of the amphipathic helix could be readily discerned.

We performed a low-pass filter of the cryo-EM map to 7 Å. At this resolution, additional densities connecting N11 were displayed clearly which arise from the N-terminal helix (Fig. 3A). Docking the helix onto the ARF6 lattice suggested that it inserts substantially into the membrane (Fig. 3B). As the helix had been predicted to be amphipathic (26) with the hydrophobic face consisting of V4, L5, I8, and F9, and the hydrophilic face consisting of K3, S6, and K7, our confirmation of how the helix is inserted into the membrane further supported the validity of the solved lattice structure assembled on membrane (Fig. 3B).

Fig. 3.

Interaction of the ARF6 lattice with membrane tubule. (A) Cross-sectional view of reconstructed ARF6-coated tubules. Black dashed box outlines a tetramer cluster. (B) Close-up views of the ARF6 lattice interacting with the membrane. Upper panels show a tetramer with its subunits (in four colors) interacting with the membrane, either in the same orientation as the cross-sectional view or rotated by 30°. Lower panels show the ribbon model of the amphipathic helix with charged face on one side and hydrophobic face on the opposite side. Key residues on both faces are labeled. (C) Endosomal membrane binding by ARF6 mutants. Specific residues mutated are M6, F9W; M7, F9A; M8, F9E; M9, V4E/L5E/I8E. Centrifugation was performed after incubation of membrane with ARF6 proteins as indicated, with pellet (P) containing membrane-bound fraction and supernatant (S) containing soluble fraction. A representative result is shown above. Quantitation is shown below, n = 3; *P < 0.05, ns P > 0.05, Student’s t test. (D) EM analysis examining the effect of ARF6 mutations on liposome tubulation. Representative images are shown. (Scale bar, 200 nm.) Quantitation of tubulation ability is shown as a table: +++ (strong), ++ (moderate), + (weak), or ± (almost none), n = 3. (E) The TfR recycling assay examining the effect of ARF6 mutations on transport from the recycling endosome to the plasma membrane. Representative images of endosomal TfR (red) colocalizing with Cbv (green), at times indicated are shown on Left. Quantitation is shown on Right, comparing control with different ARF6 conditions, n = 3. A representative experiment (which examines 10 cells) is shown; ns, P > 0.05, ****P < 0.0001, Student’s t test. (F) The GAP and ANK domains of ACAP1 can be docked onto the ARF6 lattice structure without structural collision. (G) Incubation of ACAP1 with ARF6 liposomal tubules results in vesiculation, while the use of an ACAP1 catalytic-dead mutant (R448Q) has no effect, n = 2. Representative EM images are shown.

To further characterize how the tetramer engages the membrane, we next pursued molecular dynamics simulation studies (Movie S3). The results confirmed that ARF6 tetramer associates firmly with the lipid membrane through the N-terminal myristoyl chain (MYR) and amphipathic helix (AH) (SI Appendix, Fig. S4A), with the latter involving the insertion of hydrophobic residues V4, L5, I8, and F9 into the membrane. Simulation also provided a temporal description of how the MYR, AH, as well as other parts of the tetramer engage the membrane (SI Appendix, Fig. S4B). The time-evolved separation profile of molecular dynamics confirmed that the MYR and AH regions insert fully into the membrane (SI Appendix, Fig. S4C). Moreover, the hydrophobic face of AH penetrates the membrane more deeply than the hydrophilic face (SI Appendix, Fig. S4A). The insertion mode is further supported by the reduction in RMS fluctuation (RMSF) upon membrane binding (SI Appendix, Fig. S4D).

We next pursued functional mutagenesis to validate our structural predictions. Focusing on the aromatic residue F9 in the AH, we found that a conservative mutation to tryptophan (named M6; see SI Appendix, Table S3) largely retained membrane binding by ARF6, while mutation to alanine (referred to as M7; SI Appendix, Table S3) or to glutamate (referred to as M8; SI Appendix, Table S3) moderately inhibited this binding (Fig. 3C). When multiple residues that line the hydrophobic face of the AH were mutated to glutamates (V4E, L5E, and I8E; referred to as M9; see SI Appendix, Table S3), membrane binding was reduced more drastically (Fig. 3C). In contrast to the progressive effect of the mutations on membrane binding, we found that, aside from mutating to a conserved residue (M6), all other mutations (M7, M8, M9) led to marked reduction in liposome tubulation (Fig. 3D). Consistent with this finding, we found that M6 did not have a significant effect on TfR recycling, while the other mutations (M7, M8, M9) inhibited this recycling to a similar extent (Fig. 3E). Thus, individual residues in the AH were more critical for membrane bending than membrane binding by ARF6. We also confirmed that the ARF6 mutants were expressed at levels similar to that of the wild-type (SI Appendix, Fig. S3A). Moreover, other than the most severe mutation that inhibited membrane recruitment substantially (Fig. 3C), the other AH mutations did not significantly affect ARF6 localization to the recycling endosome (SI Appendix, Fig. S3B).

We also compared membrane insertion by ARF6 in the lattice structure to that by Sar1 in the context of its assembly with COPII on tubulated membrane (20). Although the depth of insertion was similar, with the AH of ARF6 and Sar1 both inserting about 13 Å into the membrane, the density of membrane insertion was quite different. Whereas each ARF6 insertion covered about 10.4 nm² of the membrane surface area (SI Appendix, Fig. S5A), each Sar1 when complexed with COPII occupied about 114.5 nm² in surface area (SI Appendix, Fig. S5B). Thus, for each Sar1 insertion in the COPII tubule, roughly eleven ARF6 insertions cover the same surface area in the ARF6 tubule (SI Appendix, Fig. S5C). These observations suggested why the ARF6 tubule is narrower than the COPII tubule (30 nm versus 60 nm). They are also consistent with the ARF6 lattice gaining insight into vesicle fission, while the COPII lattice informs on coat organization during vesicle budding.

Interaction of ARF GAPs with the ARF6 Lattice.

We next considered that the ARF GAP activity has also been found to act in vesicle fission for the better characterized coats (37, 38). Thus, we examined whether the solved ARF6 lattice structure would allow docking by the GAP. An ARF6–GAP complex has been solved previously, which involves ARF6 bound to ASAP3 (PDB ID: 3LVQ) (39). The overall structure of ARF6, when complexed with ASAP3, was similar to that of ARF6 assembly in the lattice structure, with the RMSD between the two being 0.6 Å for 164 C_α atoms. After superimposing the structure of the ARF6–ASAP3 complex onto the ARF6 lattice structure, ASAP3 was predicted to be located on the surface of lattice assembly with its GAP domain positioned toward the active site of ARF6, with no structural collisions being predicted (SI Appendix, Fig. S5D). These results suggested that the ARF6 lattice structure would allow the subsequent action of the GAP activity in promoting vesicle fission.

We also performed a similar docking for ACAP1, the GAP that acts on ARF6 in endocytic recycling (12), and found that its GAP domain would also be positioned on the ARF6 lattice to allow GTP hydrolysis (Fig. 3F). As functional confirmation, we incubated ACAP1 with the ARF6-induced liposomal tubules, and vesiculation was observed, which was not seen when a catalytic-dead form of ACAP1 was used instead (Fig. 3G). Thus, the results not only confirmed that the lattice structure would allow the GAP to act on ARF6, but also predicted that vesicle fission would involve the sequential actions of ARF-induced constriction of the bud neck followed by GAP-induced neck scission.

Modeling ARF1 onto the ARF6 Lattice.

We then sought to confirm more directly that the ARF lattice structure acts in vesicle fission by pursuing vesicle reconstitution studies. A practical consideration was that this approach has not been established for the ARF6-regulated coat. However, it has been established for the ARF1-regulated COPI complex (40, 41). Thus, we next examined whether ARF1 could be modeled onto the ARF6 lattice structure.

The feasibility of this modeling was suggested by the high degree of sequence homology between ARF1 and ARF6 (SI Appendix, Fig. S6A), with the RMSD between the crystal structure of ARF1 in its GTP-bound state (42) (PDB ID 1O3Y) and the cryo-EM structure of ARF6 in its GTP-bound state (this study) being 1.6 Å for 163 C_α atom and 0.7 Å for 157 pruned C_α atom (SI Appendix, Fig. S6B). Further supporting the robustness of this modeling, we found that it predicted that the tetramer would also be the asymmetric unit of an ARF1 lattice on membrane tubules (SI Appendix, Fig. S6C). Furthermore, like the case of ACAP1 inducing the vesiculation of ARF6 tubules, the incubation of ARFGAP1 with ARF1 tubules also led to vesiculation (SI Appendix, Fig. S6D).

Conservation in structure also extended to each of the predicted protein interfaces (SI Appendix, Fig. S6E). For ARF1, interface 1 was predicted to be formed by the antiparallel (close to C2 symmetry) interaction of hydrogen bonds (I42-T44) and hydrophobic contacts (Y35, L39, and V43) (SI Appendix, Fig. S6E). Interface 2 would require conserved hydrophobic residues Y167 and Y58 from adjacent subunits interacting with each other, with salt bridges (K36-E168) further stabilizing this interface (SI Appendix, Fig. S6E). For interface 3, hydrophobic residues I49, F51, and V53 of one subunit, were predicted to interact with W153, Y154, I155, and W172 of an adjacent subunit (SI Appendix, Fig. S6E). The same interactions between the other paired subunits would further contribute to stabilize the tetramer organization. For interface 4, two adjacent ARF1 tetramers would interact with each other along the same helical row through short contacts, likely formed by electrostatic interactions (D72-R149 and K73-H150, SI Appendix, Fig. S6E). Interface 5 was predicted to involve contacts between adjacent helical rows, in which salt bridges (R109-E113) would form symmetric interactions between tetramers on different helical rows (SI Appendix, Fig. S6E).

ARF1 in the Lattice Structure Versus Its Positioning in COPI Vesicles.

As ARF1 interacts with coatomer for earlier stages of COPI vesicle formation that include coat recruitment and vesicle budding, we next examined whether the residues predicted to be critical for the ARF1 lattice structure, which is predicted to act in vesicle fission, would also be critical for ARF1 interacting with coatomer for the earlier stages of vesicle formation. Prior cryo-EM studies have elucidated in molecular detail how ARF1 interacts with coatomer on reconstituted COPI vesicles (17, 18, 43). In one structure of the ARF1–coatomer complexes (PDB ID, 5A1U), there are six copies of ARF1 per triad, three at the center of the triad and three toward the edge of the triad. In this configuration, the ARF1 proteins are not close enough (~9.0 Å apart) to interact directly in mediating COPI trimer formation. Instead, trimer formation is conducted by the COP I subunits β’ and γ. We then superimposed our model of Arf1 tetramer onto the cryo-EM structure of the COPI coat on the reconstituted vesicles (PDB ID, 5A1U). When an ARF1 subunit in the tetramer was aligned with an ARF1 subunit in the ARF1–coatomer complex, severe structural collision was observed between another ARF1 subunit in the tetramer and γ-COP in the ARF1–coatomer complex (SI Appendix, Fig. S7A), predicting that the positioning of ARF1 in the lattice structure would be quite different than that in the COPI coat on vesicles. Another previously solved structure (PDB ID 5NZR) achieved a higher resolution of the ARF1–coatomer complex, having 9 Å resolution as compared to 13 Å resolution achieved by PDB ID 5A1U. Thus, we also superimposed ARF1 in the lattice structure onto this higher-resolution structure of the ARF1–coatomer complex, which further confirmed that mutations that target the lattice structure were unlikely to affect how ARF1 interacts with coatomer for COPI trimer formation (SI Appendix, Fig. S7B).

We next compared in detail how ARF1 interacts with the coatomer subunits in the better-resolved ARF1–coatomer complex (PDB ID, 5NZR), which are β-COP, δ-COP, β’-COP, and γ-COP, versus how ARF1 is positioned in the lattice structure (SI Appendix, Fig. S8 A–F).

In Interface 1 of the ARF1 lattice, although residues I42 and T44 were in the vicinity of the two interfaces predicted for the ARF1–coatomer complex, which involves interaction between ARF1 and δ-COP (SI Appendix, Fig. S8B) and ARF1 and β’-COP (SI Appendix, Fig. S8E), they were not predicted to be directly involved in these two ARF1–coatomer interfaces. Instead, the interface between ARF1 and δ-COP was predicted to involve mainly L37 of ARF1 with E158/K162 of δ-COP (potential hydrogen bond with main chain of L37), and E54 of ARF1 with Q151 of δ-COP (potential hydrogen bond) (SI Appendix, Fig. S8B), while the interface between ARF1 and β’-COP was predicted to involve mainly T161 of ARF1 with G173 of β’-COP, E41 of ARF1 with K219 of β’-COP (SI Appendix, Fig. S8E).

In Interface 2 of the ARF1 lattice, residues Y167 and E168 were close to the interface between ARF1 and δ-COP in the ARF1–coatomer complex (SI Appendix, Fig. S8B), but again were unlikely to participate in this interface, as this interface of the ARF1–coatomer complex was predicted to involve instead a different set of paired residues, as noted above.

In Interface 3 of the ARF1 lattice, residues I49 and F51 were predicted to be critical. However, the interface between ARF1 and γ-COP was predicted to involve residues F51, L77, H80, and Y81 of ARF1 and residues F71, T74, I104, and K75 of γ-COP, respectively (SI Appendix, Fig. S8 D and F), while the interface between ARF1 and β-COP likely involved residues F51, L77, R73, and H80 of ARF1 interacting with residues L82, M75, H111, and I114 of β-COP, respectively (SI Appendix, Fig. S8A). Thus, because multiple other residues, besides F51, were predicted to participate in Interface 3 of the ARF1 lattice and in the interface between ARF1 and γ-COP/β-COP in the ARF1–coatomer complex, targeting against F51 seemed unlikely to destabilize these interfaces.

In Interface 4 of the ARF1 lattice, residues K73 and R149 were predicted to play critical roles. Whereas R149 was not within the vicinity of any of the ARF1–coatomer interfaces, K73 was situated in closer proximity to the interfaces between ARF1 and β-COP (SI Appendix, Fig. S8A, shown as R73 because the yeast ARF1 was used in the previous cryo-EM structure of ARF1–coatomer interactions) and between ARF1 and γ-COP (SI Appendix, Fig. S8F, shown as R73 for reason mentioned above) in the ARF1–coatomer complex. However, because K73 was situated at the edge of these interfaces, targeting against this residue also seemed unlikely to destabilize the ARF1–coatomer interfaces.

In Interface 5 of the ARF1 lattice, residues R109 and D114 were predicted to play critical roles. However, they were not within the vicinity of any of the ARF1–coatomer interfaces. There was another interaction interface between ARF1 and γ-COP predicted for the ARF1–coatomer complex (SI Appendix, Fig. S8C), but none of the ARF1 residues critical for lattice assembly was within the vicinity of this ARF1–coatomer interface. Thus, the results altogether further supported that ARF1 is positioned quite differently in the lattice structure as compared to its position in the ARF1–coatomer complex.

The ARF lattice structure revealed another notable insight. ARF1 dimerization has been found to be needed for COPI vesicle fission with the Y35 residue playing a critical role (25). The equivalent residue in ARF6 is the Y31 residue. In the ARF6 lattice structure, the distance between the two Y31 residues was 1.5 nm. As the cutoff distance for a significant interaction was predicted to be 1.2 nm, the Y31 residue was unlikely to be directly involved in forming Interface 1. This prediction was further supported by simulation studies that revealed the interaction energy for an Y31–Y31 pairing to be negligible (SI Appendix, Fig. S4E). A similar analysis on ARF1 Y35 predicted that it should also not be involved directly in forming Interface 1, as the interaction energy for an Y35–Y35 pairing was also negligible (SI Appendix, Fig. S4F). We further found that ARF6 Y31 was predicted to be located in a hydrophobic environment, making hydrophobic interaction with V39 (3.8 Å). Such a hydrophobic environment would stabilize L35 and V39, both of which were predicted to make direct contacts to form interface 1. As such, the results altogether predicted that ARF1 Y35 would be involved indirectly in ARF1 dimerization by stabilizing the interaction between the nearby hydrophobic residues in explaining how it is critical for COPI vesicle fission.

Elucidating the Role of the ARF1 Lattice in COPI Vesicle Formation.

We next pursued mutagenesis studies to confirm functionally that residues which are critical for the protein interfaces of the ARF1 lattice would also be critical for COPI vesicle fission. We targeted the key residues in Interfaces 1 to 5 mentioned above, with the resulting mutants named ARF1 M1-5 (SI Appendix, Table S3). To target the amphipathic helix, we mutated the F13 residue (corresponding to F9 of ARF6) on the hydrophobic face to glutamate (F13E), and this mutant was named ARF1 M6.

We initially performed the COPI transport assay in cells, which involved tracking a model COPI cargo, known as VSVG-KDELR, in its redistribution from the Golgi to the ER, as previously described (44). Whereas siRNA against ARF1 inhibited COPI transport, and the reexpression of the wild-type ARF1 rescued this defect, reexpression of the ARF1 mutants could not (Fig. 4 A and B). We also confirmed that the mutants were expressed at levels similar to that of the wild-type (SI Appendix, Fig. S6F).

Fig. 4.

The role of the ARF1 lattice structure in COPI transport. Quantitations are expressed as mean with SD, with number of experiments indicated, and statistics using the Student’s t test, ns, P > 0.05, *P < 0.05, ***P < 0.001. (A) COPI transport assay examining the effect of ARF1 mutations was performed on HeLa cells, n = 3. Specific residues mutated are M1, I42A/T44A; M2, Y167A/E168R; M3, I49A/F51A; M4, K73E/R149E; M5, R109E/D114R. Representative images of VSVG-KDELR (red) colocalizing with a Golgi marker (giantin, green) at times indicated are shown on Left. Quantitation of a representative experiment (which examines 10 cells) is shown on Right, comparing control with different ARF1 conditions. (B) COPI transport assay examining the effect of a mutation that targets the amphipathic helix of ARF1 (M6, F13E), n = 3. Representative images of VSVG-KDELR (red) colocalizing with giantin (green) at times indicated are shown on Left. Quantitation of a representative experiment (which examines 10 cells) is shown on Right, comparing control with different ARF1 conditions. (C) Reconstituting COPI vesicles from Golgi membrane to examine the effect of ARF1 mutations on vesicle formation. Quantitation compares the wild-type and different mutants, n = 5. (D) Representative EM images from the COPI vesicle reconstitution system showing arrest in the two stages of vesicle fission induced by the ARF1 mutations. Upper panels show arrest in late fission, as reflected by buds with markedly constricted necks. Lower panels show arrest in early fission, as reflected by the formation of tubules. (Scale bar, 50 nm.) (E) Quantitation of COPI buds with constricted necks induced by using different ARF1 mutants in the vesicle reconstitution system, n = 3. Values indicate the percentage of buds over all Golgi protrusions (buds and tubules) observed. (F) Quantitation of COPI tubules induced by using different ARF1 mutants in the vesicle reconstitution system, n = 3. Values indicate the percentage of tubules over all Golgi protrusions (buds and tubules) observed. (G) Reconstituting COPI vesicles from Golgi membrane to examine the effect of ARF1 mutations on vesicle formation. The effect of ARF1 mutants that target against two interfaces (as indicated) was examined, n = 4.

We next pursued the reconstitution of COPI vesicle formation using Golgi membrane, as previously described (38). In recent years, this reconstitution has been refined to enable COPI vesicle fission to be subdivided into two stages, with the inhibition of early fission that prevents the constriction of the COPI bud neck resulting in the generation of COPI tubules, and the inhibition of late fission resulting in the accumulation of COPI buds with constricted necks (45). We initially confirmed that the ARF1 mutations which target each of the protein interfaces in the lattice structure, as well as the amphipathic helix, inhibited COPI vesicle formation (Fig. 4C). We then performed EM examination and found that targeting the protein interfaces or the amphipathic helix resulted in mixed accumulations of tubules and buds with constricted neck (Fig. 4D), predicting that the ARF1 lattice would participate in both the early and the late stage of COPI vesicle fission.

We next performed quantitation and found that the mutations had relatively uniform effect in accumulating COPI buds with constricted necks (Fig. 4E), suggesting that the different protein interfaces and the amphipathic helix all contribute similarly to late fission. In contrast, the effect of the mutations in accumulating COPI tubules was more varied, with M2 having less effect and M4 having more effect (Fig. 4F), predicting that the protein interface targeted by M2 would be less critical for early fission, while the interface targeted by M4 would be more critical for early fission.

The ARF1 mutations all reduced the efficiency of COPI vesicle formation, suggesting that the explicit symmetry of the lattice assembly is needed for vesicle fission. However, because the reduction was partial, we examined whether efficient vesicle fission would also require the full set of multivalent interactions critical for ARF lattice assembly. We generated two additional mutants that targeted more than one interface in the lattice structure, and found that the expression of these mutants resulted in a further reduction in vesicle formation (Fig. 4G). Thus, multivalent interactions in the lattice structure are also critical for vesicle fission.

ARF1 Mutants Retain Interactions with COPI-Related Factors.

As the results from the COPI vesicle reconstitution studies revealed that ARF1 mutations only affected vesicle fission, an implication of this finding was that they should not disrupt ARF1 interactions with factors that are needed for the earlier stages of COPI vesicle formation. Thus, we next pursued further studies to confirm this prediction. Membrane recruitment of coatomer requires its interaction with ARF1. We found that the ARF1 mutations that targeted the lattice interfaces did not affect the ability of ARF1 to recruit coatomer onto the Golgi membrane (SI Appendix, Fig. S9A). The mutations also did not affect the ability of ARF1 to be recruited to Golgi membrane (SI Appendix, Fig. S9B), or to localize to the Golgi in cells (SI Appendix, Fig. S9C).

We next considered that, besides ARF1 and coatomer, which are sufficient to reconstitute vesicle formation using liposomes, COPI vesicle formation from Golgi membrane is more complex, requiring ARFGAP1 and BARS additionally (41, 46). Interrogating the interaction between ARF1 and ARFGAP1, we found that it also was not affected by the ARF1 mutations (SI Appendix, Fig. S9D). We also found that most ARF1 mutants still supported the GAP activity of ARFGAP1 (SI Appendix, Fig. S9E). The lone exception was mutant 1 that showed partial ability. This finding could be explained by a residue altered in mutant 1 being involved in switch 1, which would be predicted to affect the GAP activity. We next found that ARF1 did not interact with BARS, either in solution (SI Appendix, Fig. S9F) or on membrane (SI Appendix, Fig. S9G), and thus ruling out that this interaction could be a relevant target of the mutations. We further noted that a previous study had identified point mutations in ARF1 that disrupt its interaction with coatomer (14). Examining one such mutation (L77E), we first confirmed that it inhibited COPI vesicle formation (SI Appendix, Fig. S9H). We then found that it still enabled ARF1 to induce liposome tubulation (SI Appendix, Fig. S9I). Thus, this set of results further supported the specificity by which the mutations that target the ARF1 lattice structure inhibit COPI vesicle fission.

Discussion

We have elucidated in molecular detail how ARF6 assembles into a lattice structure on membrane to induce tubulation. This structure possesses features that are distinct from those formed by other membrane-bending proteins. Whereas endophilin (32) and dynamin (35) assemble as 2-start helical arrays, and SNX1 (34) and ACAP1 (33) assemble as 1-start helical arrays, we find that ARF6 assembles as 5-start arrays. Another notable distinction is that ARF6 forms tetramers as the basic building block of its lattice structure, while dimers are involved for lattices formed by endophilin (32), SNX1 (34), and dynamin (35), and a dimer of dimer is involved in the ACAP1 lattice (33). While the ARF6 lattice exhibits an organization distinct from lattices formed by other membrane-bending proteins, it shares nevertheless overall similarity in explaining how membrane tubulation is induced, which involves the protein lattice forming helical rows around tubulated membrane.

Previous studies on vesicle formation by two of the best-characterized coat complexes have revealed that the ability of ARF1 and Sar1 to induce liposome tubulation reflects their roles in COPI and COPII vesicle fission, respectively (24, 25). A low-resolution view of how Sar1 is assembled on membrane to induce tubulation has been achieved, which revealed the formation of an ordered lattice on membrane, with Sar1 dimers being the basic unit (47). Our study now achieves a high-resolution (atomic level) view of how an ARF protein is organized on tubulated membrane, which reveals that ARF tetramer is the basic unit that organizes into helical rows for lattice formation. We have also identified key protein interfaces critical for the ARF lattice assembly, as well as revealing that the dense insertion of its amphipathic helix into membrane explains why the ARF lattice induces highly constricted tubules.

Key evidence that this structure explains how ARF promotes vesicle fission comes from functional studies that involve mutating key residues predicted to be critical for lattice assembly and then finding that they inhibit COPI vesicle fission. We have also ruled out that the mutations could also be disrupting ARF1 interacting with other COPI-related factors in explaining how vesicle formation is inhibited. We first compared the positioning of ARF1 in the lattice structure to that of ARF1 in complex with coatomer on COPI vesicles, which had been previously elucidated by cryo-EM (18). The results predicted that the mutations targeting the major protein interfaces of the ARF1 lattice are unlikely to have a significant impact on the ability of ARF1 to assemble with coatomer for vesicle formation. As further support, we find that the mutations allow vesicle formation to proceed through the budding stage, which would require ARF1 interacting with coatomer. We also performed additional studies for further confirmation, which included the interrogation of ARF1 interacting with coatomer, ARF1 interacting with its regulators for localization to the Golgi, and ARF1 interacting with ARFGAP1 and BARS. These results further supported the specificity by which the ARF1 mutations disrupt the ARF lattice structure in explaining how COPI vesicle fission is inhibited.

We also followed up on the previous elucidation that the ARF GAP activity is needed for both COPI and COPII vesicle fission (37, 38), and find that the ARF lattice structure allows docking by ARF GAPs. As functional confirmation, we observe the vesiculation of ARF tubules when a cognate GAP is incubated with these tubules. We have also compared how ARF6 inserts into the membrane versus that by Sar1 in the context of COPII assembly on tubulated membrane. Whereas the depth of membrane insertion is similar, the density of insertion is quite different, with calculations predicting that the same membrane surface area would have eleven ARF6 insertions for every Sar1 insertion. This finding not only suggests why the ARF tubule is narrower than the COPII tubule, but also further supports that the ARF lattice structure acts in vesicle fission, while the COPII coat on tubulated membrane informs on coat organization during vesicle budding.

Vesicle formation involves three major sequential stages: i) coat recruitment, ii) vesicle budding, and iii) vesicle fission. Studies over the years have revealed that ARF proteins act in all three stages. For the first two stages, structural studies have achieved a molecular understanding of how ARF acts (14, 16–18). Our structural elucidation of an ARF-induced tubular lattice has now achieved a molecular understanding of how ARF acts in the third stage (summarized in SI Appendix, Fig. S10).

Studies on dynamin, which acts in clathrin-mediated endocytosis, have been leading the way in the molecular understanding of how vesicle fission occurs (48, 49). Dynamin has one domain that binds GTP and another that possesses GAP activity. This arrangement has similarity to an ARF protein in complex with its GAP. GTP-binding by dynamin induces the constriction of the clathrin bud neck, followed by GTP hydrolysis that results in neck scission to complete the process of vesicle formation. Thus, as the ARF lattice that induces tubulated membrane allows docking by the GAP for vesiculation, the noted parallel between how dynamin and ARF/GAP act has the general implication that mechanisms underlying the fission stage of vesicle formation are likely to be fundamentally conserved.

Materials and Methods

Details on Materials and Methods for structural studies, which include Purification of ARF6 and ACAP1, EM, Helical reconstruction, Model building and refinement, and Molecular dynamics simulation are described in SI Appendix. Details on Materials and Methods for in vitro functional studies, which include Liposome deformation assays, ARF membrane recruitment assays, In vitro reconstitution of COPI vesicle formation, GST pulldown assay, GTPase activity assay are also described in SI Appendix. SI Appendix also contains details on Materials and Methods for cell-based studies, which include cell culturing and transfections, Transport assays, ARF localization studies, and Quantitative colocalization studies.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Overall ARF6 assembly on tubulated membrane.

Movie S2.

Tetramer is the asymmetric unit of ARF6 helical packing.

Movie S3.

Time frames of the molecular simulation for ARF6 tetramer interaction with lipid membrane.

Data, Materials, and Software Availability

3D EM map, atomic structure models data have been deposited in Electron Microscopy Data Bank (https://www.ebi.ac.uk/emdb/) with the accession code EMD-33414 (50). The corresponding coordinates of ARF6 helical array on the tubule have been deposited in the Protein Data Bank (https://www.rcsb.org) with the accession code 7XRD (51). All study data are included in the article and/or supporting information.