A structure of the COPI coat and the role of coat proteins in membrane vesicle assembly

See allHide authors and affiliations

Science  10 Jul 2015:
Vol. 349, Issue 6244, pp. 195-198
DOI: 10.1126/science.aab1121

A coat of many components

The formation of coated trafficking vesicles is among the most fundamental of cellular processes. COP1 transport vesicles are involved in retrograde membrane trafficking in the Golgi apparatus and endoplasmic reticulum. Dodonova et al. applied cryo–electron tomography to determine the structure of the COPI coat in its fully assembled form on budded vesicles (see the Perspective by Noble and Stagg). They combined structural data with cross-linking mass spectrometry to generate a complete molecular model. The model suggests a mechanism of coat assembly in which coat proteins cluster via flexible interactions instead of forming a protein cage on the membrane.

Science, this issue p. 195; see also p. 142


Transport of material within cells is mediated by trafficking vesicles that bud from one cellular compartment and fuse with another. Formation of a trafficking vesicle is driven by membrane coats that localize cargo and polymerize into cages to bend the membrane. Although extensive structural information is available for components of these coats, the heterogeneity of trafficking vesicles has prevented an understanding of how complete membrane coats assemble on the membrane. We combined cryo–electron tomography, subtomogram averaging, and cross-linking mass spectrometry to derive a complete model of the assembled coat protein complex I (COPI) coat involved in traffic between the Golgi and the endoplasmic reticulum. The highly interconnected COPI coat structure contradicted the current “adaptor-and-cage” understanding of coated vesicle formation.

Coat protein complexes I and II (COPI and COPII) and clathrin-coated vesicles mediate transport between different compartments of the cell. COPI mediates retrograde transport, both from the Golgi to the endoplasmic reticulum and within the Golgi (1, 2). Coated vesicles are generally formed in a similar manner (35). In most cases, exchange of guanosine diphosphate for guanosine triphosphate within a small guanosine triphosphatase (GTPase) induces a conformational change that exposes an N-terminal amphipathic helix that inserts into the membrane. The GTPase recruits adaptor proteins, which in turn recruit coat proteins that polymerize to form the outer coat or “cage.” This modular approach can allow different adaptor proteins to be used, depending on the cargo being packaged. In the COPI system, where transport can occur independently of the presence of cargo, the adaptor and outer coat are combined in a soluble heteroheptameric complex (called coatomer) that is recruited to the membrane en bloc (6). After a coated bud has formed, vesicles are released by a scission reaction before uncoating and fusing with the target membrane.

The three archetypal vesicle coats share aspects of their structural organization. The outer-coat components—clathrin, Sec31 (for COPII), and α- and β′-COP (for COPI)—consist of N-terminal β propellers followed by extended α solenoids (7). This “protocoatomer” motif, also found in the nuclear pore, is thought to represent an ancestral module involved in membrane bending (7). Interactions between α solenoids contribute to polymerization of the clathrin and COPII coats to form symmetrical cages whose structures have been solved by cryo–electron microscopy and single-particle analysis (8, 9). In COPI, β′-COP (together with part of α-COP) can be crystallized to form a triskelion-like structure (10) that has been proposed to form a polygonal cage. Further similarities between coats are found in the adaptor components: The AP1 and AP2 clathrin adaptors are homologs of the tetrameric subcomplex of COPI, γ-ζ-β-δ-COP (1113). These observations make it reasonable to predict that the assembled COPI coat can also be functionally subdivided into adaptor and cage.

Structural biology of heterogeneous membrane-containing systems is challenging. The structures of clathrin and adaptors have not been resolved together on membranes (14). Low-resolution structural data have been obtained describing the COPII coat assembled on membranes: The adaptor subunits coat the membrane and are linked flexibly to the outer subunits that form a cage around the membrane (15). The structure of the COPI coat on budded coated vesicles has been described at low resolution (16). The building block of the coat is a triad of coatomers that are linked together with contacts of variable valence. However, the arrangement of the proteins within the assembled coat remains unclear.

To determine a complete molecular model of the COPI coat, COPI-coated vesicles were produced in vitro by incubating giant unilamellar vesicles with coatomer, the GTPase Arf1, the guanine exchange factor ARNO, and guanosine 5′-O-(3′-thiotriphosphate) (16). The sample was vitrified by plunge-freezing and imaged by cryo–electron tomography (Fig. 1A). Eighty-two tomograms were collected under conditions optimized for high-resolution structure determination (17). To determine the structure of the repeating building block of the coat (the triad), the data set was split into two halves, each of which was independently subjected to subtomogram averaging (17). Structures from the two half data sets were compared to determine the resolution and averaged to give a final structure with a resolution of 13 Å (Fig. 1B and fig. S1B) (18). A total of 39,558 asymmetric units of the assembled coat from 1265 vesicles and near-complete buds contributed to the final structure.

Fig. 1 COPI coat structure.

(A) Electron micrograph of in vitro–formed COPI-coated vesicles. Scale bar, 100 nm. (B) EM reconstruction of the triad at 13 Å resolution, colored from green to blue according to the radial distance from the membrane (red). (C) Representations of complete COPI-coated vesicles. The membrane is shown in gray. Densities of intertriad linkages are colored pink (I), orange (II), yellow (III), and red (IV) (fig. S1).

Subtomogram averaging also determines the positions and orientations of the triads on each vesicle. As previously observed, the triads adopted preferred relative orientations (16). Following the approach described in (16), we generated reconstructions of the linkages between triads at resolutions between 18 to 23 Å (fig. S1, C to G). By combining these with the structure of the triad, we could visualize the overall arrangement of the coat on individual vesicles (Fig. 1C).

To interpret the reconstruction, we assigned the positions of the individual protein components within the density. Crystal structures are available for ε-COP bound to the C-terminal domain of α-COP (10, 19), β′-COP bound to a part of α-COP (10), and for a complex Arf1-γ-ζ-COP (fig. S2) (12). At the resolution obtained, global fitting of the crystal structures into the electron microscopy (EM) density allowed us to identify the possible positions of each structure within the map (Fig. 2 and fig. S3) (18). Note that α- and β′-COP are structural homologs (20), as are γ-ζ-COP and β-δ-COP (11). In these cases, as expected, two positions were identified (Fig. 2) that could be assigned to the correct homolog on the basis of characteristic structural features and labeling experiments (fig. S4) (18). Three regions of EM density were not occupied by the core domains of the α-β′-ε-COP or γ-ζ-β-δ-COP subcomplexes (fig. S3C). These densities most likely correspond to the appendage (ear) domains of γ-COP (21) and β-COP and the μ-homology domain of δ-COP.

Fig. 2 Global fitting of COPI crystal structures.

(A) Fitting results for β′-α-COP (PDB ID 3MKQ; β′-COP in light blue, α-COP in dark blue). An additional α helix adjacent to the second β-propeller domain of β′-COP, but absent in α-COP, is marked with black arrow. (B) Fitting results for Arf1-γ-ζ-COP (PDB ID 3TJZ; Arf1 in pink, γ-COP in light green, ζ-COP in yellow). Left panels: Histograms of cross-correlation values resulting from 10,000 fitting experiments (Chimera global search). Green arrows mark the highest-scoring fitted positions illustrated in the right panels. Additional density observed when the β-COP N terminus is nanogold-labeled is shown in red (fig. S4).

To further validate the arrangement of the core domains determined by fitting and labeling, we applied chemical cross-linking to in vitro–assembled COPI vesicles, followed by digestion and mass spectrometry. Twenty-six pairs of lysine residues were sufficiently close together within the assembled structure to be cross-linked (22) (table S1). We then generated multiple alternative structural models for the arrangement of core and appendage domains and measured the separation of the cross-linked lysine pairs in each model (figs. S5 and S6A) (18). The model most consistent with the cross-linking data matched the arrangement of core domains determined from fitting and labeling and allowed the appendage domains and δ μ-homology domain to be assigned to the unoccupied densities.

We generated a final structural model (Fig. 3 and movie S1) by placing available crystal structures or homology models of the coat subunits at the positions determined above and performing flexible fitting into the EM density (fig. S6B) (18). The structure is consistent with available biochemical data: The curved triad structure positions the Arf1 molecules and cargo binding sites proximal to the membrane (Fig. 3D and fig. S7). Arf1 is accessible for ArfGAP binding (fig. S7D), and binding sites for tethering proteins are on the outside of the coat (fig. S8, A to D). The arrangement of COPI components within the structure suggests that it cannot be functionally subdivided into outer coat and adaptor.

Fig. 3 Molecular architecture of the COPI coat.

Color scheme: membrane, gray; Arf1, pink; γ-COP, light green; β-COP, dark green; ζ-COP, yellow; δ-COP, orange; β′-COP, light blue; α-COP, dark blue; ε-COP, cyan. (A) Final triad model after flexible fitting of crystal structures (colored) into EM density (transparent). (B) Crystal structures illustrated as colored isosurfaces filtered to 13 Å resolution. (C) Asymmetric unit of the coat: one coatomer and two Arf1 molecules. (D) Membrane-associated domains form a curved surface. Arf1 dimers (32) are not observed (fig. S7). (E) β′- and α-COP form an archlike dimer. The N-terminal β-propeller domains, containing cargo-binding sites (orange), contact the membrane. (F) γ-ζ-β-δ-COP forms a hyper-open archlike dimer. The Arf1 N-termini (illustrative cyan helices) contact the membrane. γ-ζ-COP approaches the membrane closely, whereas its homolog β-δ-COP is further away, which is consistent with the role of γ-COP in binding the cytoplasmic tails of p24 proteins, a function that β-COP lacks (33). (G) Structure model of an open-form clathrin adaptor for comparison [based on PDB ID 4HMY and the γ-Arf1 interface (PDB ID 3TJZ)]. (H) Illustrative model of a complete COPI-coated vesicle showing structures (color) and EM density (transparent). (I) Equivalent surface model as in (D).

β′- and α-COP interact via the interface described in their x-ray structure [Protein Data Bank identification number (PDB ID) 3MKQ], but they do not assemble triskelions, as previously proposed (10, 23). Instead, the α solenoids of β′- and α-COP form an arch over the γ-ζ-β-δ-COP subcomplex, orienting their N-terminal β propellers such that the K(X)KXX cargo-motif binding sites (K, Lys; X, any amino acid) (24, 25) are optimally positioned against the membrane (Fig. 3E). β′- and α-COP do not form a cage or lattice as in COPII and clathrin coats; instead, they are linked to one another via the γ-ζ-β-δ-COP subcomplexes, forming an interconnected assembly.

The clathrin adaptors AP2 and AP1 have each been crystallized in a closed form (thought to represent the cytoplasmic conformation) and an open form in which cargo binding sites are accessible (thought to represent the membrane-associated conformation) (2629). We found the adaptorlike γ-ζ-β-δ-COP subcomplex of COPI (Fig. 3F) to be more extended than either of these conformations; it appears as a “hyper-open” form (compare Fig. 3, F and G). Biochemical experiments point to conformational opening of coatomer by ligand binding (30, 31), suggesting that the transition to the hyper-open form occurs upon membrane association. We speculate that, upon membrane binding, clathrin APs may also transition to a form that is more open than predicted from the available crystal structures. In the hyper-open form, the β- and γ-COP α solenoids form an arch that is bound to the membrane at each end via an interaction of their respective trunk domains with Arf1. This arrangement mimics the arrangement of the α-β′-COP subcomplex (Fig. 3, E and F): Both outer-coat–like and adaptorlike subcomplexes form extended α-solenoid arches linked at each end to the membrane.

The extended α solenoids in the coats of COPI, COPII, and clathrin oligomerize very differently (fig. S9), but in all cases they function as extended spacers that distribute cargo-binding or membrane-bending domains over the curved membrane surface while leaving the surface accessible. We suggest that this, rather than cage formation, is the ancestral function of protocoatomers.

The triads are connected by flexibly attached domains. One set of interactions is formed by the μ-homology domain of δ-COP (fig. S8) and another by ε-COP and the C-terminal domain of α-COP. In some positions, ε-COP forms a homodimeric interface observed in a crystal form (PDB ID 3MKR) (10); in other positions, it bridges the C-terminal and core domains of α-COP (fig. S8).

The structural model of the assembled coat is consistent with a model in which local membrane curvature is induced partly by membrane scaffolding by the curved triad structure and partly by insertion of six Arf1 amphipathic helices (18) (fig. S7 and movie S2). Linking triads together via flexibly attached domains would propagate this local curvature over larger membrane areas to form buds (fig. S9).

Supplementary Materials

Materials and Methods

Figs. S1 to S9

Tables S1 and S2

References (3468)

Movies S1 and S2

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank K. Bacia for providing Arf1 protein; M. Schorb and J. Kosinski for technical assistance; and F. Schur, W. Wan, B. Brügger, D. Devos, M. Kozlov, and D. Owen for discussions. This work was technically supported by EMBL IT services and was funded by the Deutsche Forschungsgemeinschaft within SFB638 (A16) to J.A.G.B. and F.W. EM maps and structural models are deposited in the Electron Microscopy Data Bank (accession codes EMD-2985, EMD-2986, EMD-2987, EMD-2988, and EMD-2989) and the Protein Data Bank (PDB IDs 5A1U, 5A1V, 5A1W, 5A1X, and 5A1Y).
View Abstract

Stay Connected to Science

Navigate This Article