The structure of COPI vesicles and regulation of vesicle turnover

COPI‐coated vesicles mediate transport between Golgi stacks and retrograde transport from the Golgi to the endoplasmic reticulum. The COPI coat exists as a stable heptameric complex in the cytosol termed coatomer and is recruited en bloc to the membrane for vesicle formation. Recruitment of COPI onto membranes is mediated by the Arf family of small GTPases, which, in their GTP‐bound state, bind both membrane and coatomer. Arf GTPases also influence cargo selection, vesicle scission and vesicle uncoating. Guanine nucleotide exchange factors (GEFs) and GTPase‐activating proteins (GAPs) regulate nucleotide binding by Arf GTPases. To understand the mechanism of COPI‐coated vesicle trafficking, it is necessary to characterize the interplay between coatomer and Arf GTPases and their effectors. It is also necessary to understand interactions between coatomer and cargo, cargo adaptors/receptors and tethers facilitating binding to the target membrane. Here, we summarize current knowledge of COPI coat protein structure; we describe how structural and biochemical studies contributed to this knowledge; we review mechanistic insights into COPI vesicle biogenesis and disassembly; and we discuss the potential to answer open questions in the field.

cisternae. Arf1 (red) recruits coatomer (green) to the membrane en bloc, leading to budding and scission of coated vesicles which subsequently uncoat. (B) A schematic model of COPI recruitment to the membrane and its dissociation. A membrane-associated GEF promotes Arf1 recruitment to the membrane and exchange of GDP for GTP to stabilize Arf1membrane association. Next, COPI is recruited to the membrane by the GTP-bound form of Arf1. Dimerized Arf1 might be responsible for vesicle scission. Dissociation of the COPI coat from the membrane occurs upon GTP hydrolysis within Arf1 assisted by an ArfGAP. (C) The structure of the complete COPI triad (PDB 5A1U). The COPI triad consists of three copies of coatomer. Coatomer can be conceptually subdivided into an adaptor subcomplex (green) and an outer-coat subcomplex (blue); however, they are obligate partners-the entire COPI complex is recruited to the membrane en bloc. COPI components are shown as maps simulated from PDB coordinates (upper row-Side view, parallel to the membrane; lower row-View from the cytosolic side, perpendicular to the membrane). EM density from EMDB-2985 is shown in light gray to provide context. Each of the coats mediates transport between distinct sets of cellular compartments and has different protein subunits, but there are similarities both in their overall domain and structural architectures and in the general principles of coated vesicle assembly [19,20]. Vesicle formation is initiated in many cases by the action of small GTP-binding proteins (an exception is clathrinmediated endocytosis at the plasma membrane, where the AP2 adaptor protein of clathrin-coated vesicles is recruited to the membrane by direct interaction with lipids) [19,21,22]. The small GTP-binding proteins that regulate the formation of COPI-and COPII-coated vesicles are typically localized to the cytoplasm in their GDP-bound state. Interaction with a guanine nucleotide exchange factor (GEF) converts the protein to its GTP-bound state, stabilizing membrane association and recruitment of coat proteins [23][24][25].
COPI vesicles are coated with the heteroheptameric COPI complex termed coatomer. Members of the Arf (ADP-ribosylation factor) family of GTP-binding proteins, canonically Arf1, are responsible for COPI coat protein recruitment to membranes to initiate coated vesicle biogenesis [24,26,27]. Arf1 also plays roles in cargo sorting, vesicle scission and uncoating of the vesicle, which must occur before the vesicle can fuse with its destination membrane to deliver its cargo [26,[28][29][30][31][32][33][34]. Arf1 does not have intrinsic GTPase activity, but the role of Arf1 in both cargo sorting and vesicle uncoating depends on GTP hydrolysis. Arf1 is therefore also regulated by Arf GTPase-activating proteins (ArfGAPs). Thus, Arf1 and the regulators of its GTPase cycle (Arf-GEFs and ArfGAPs) are important regulators of the COPI-coated vesicle life cycle (Fig. 1B).
Multiple tethering factors help orchestrate COPI vesicle binding to the target organelles (Golgi or ER) and can facilitate the recruitment of SNARE proteins. Finally, SNARE proteins mediate the fusion of the COPI vesicle with the target membrane to deliver its membrane-associated and lumenal cargoes [35,36].
In this review, we describe the structure of the COPI coat and how it interacts with the regulatory GTPase Arf and the effector proteins, ArfGEFs and ArfGAPs. We discuss these structures and interactions in the context of the mechanism of COPI vesicle formation. We highlight open questions in the field of COPI vesicle biogenesis and consider opportunities for additional insight.
The structure of the COPI coat The COPI coat, or coatomer, is a multimeric complex consisting of seven subunit proteins: a-COP, b-COP, b 0 -COP, d-COP, e-COP, c-COP and ξ-COP, which are stably associated as a highly flexible complex in the cytosol and are recruited to membranes as a heteroheptameric unit during vesicle biogenesis [10,11,37,38]. Due to the flexible nature of cytosolic coatomer, no high-resolution structural information is available on the intact complex, but the structures of individual components have been solved using x-ray crystallography [39][40][41][42][43][44][45][46]. In contrast to the lack of structural insight into cytosolic coatomer, studies of the assembled, membrane-associated COPI coat using cryoelectron tomography have provided significant insight into its structure [46][47][48]. The structure of the in vitro assembled COPI coat was resolved by electron microscopy and subtomogram averaging to 9 A. This resolution is sufficient to enable the positioning of each protein domain within the electron density [46].
Two subcomplexes can be separately dissociated from assembled COPI-coated vesicles in vitro using mild chemical treatment. The a-COP:b 0 -COP:e-COP subcomplex has been termed the outer-coat subcomplex, B-subcomplex or cage-like subcomplex (Fig. 1C). a-COP and b 0 -COP have a domain architecture consisting of one or two N-terminal b-propeller domains followed by an a-solenoid domain that is shared with clathrin heavy chain, COPII Sec 13/31, components of the nuclear pore (reviewed in ref. [20]) and other membrane-associated eukaryotic structures (reviewed in ref. [49]). This fold, termed the 'protocoatomer' architecture, is believed to have been present in the first eukaryotic common ancestor [50,51]. The b-COP:d-COP:c-COP:ξ-COP subcomplex has sequence and structural similarities to the adaptor complex of clathrin, AP2, and has been termed the F-subcomplex or the adaptor-like subcomplex (Fig. 1C) [52][53][54][55]. A heterotetrameric protein complex called TSET has structural similarity to both the adaptor-like subcomplex of COPI and APs 1-5 and represents a shared evolutionary link between these adaptor complexes [56]. The structural similarities between the proteins that are responsible for forming vesicles and nonendosymbiotic cellular substructures support the organelle paralogy hypothesis, which posits that organellar complexity arose through gene duplication and co-evolution of a core set of vesicle formation and fusion proteins [49][50][51]. Unlike the clathrin and COPII coats, which are assembled in two stages-with the adaptor subcomplex binding first and proximally to the membrane followed by binding of the cage subcomplex as an outer layer-coatomer is recruited en bloc from the cytosol in a single step, and both subcomplexes contact the membrane and form an interconnected assembly rather than two separate protein layers [38,46,47]. When assembled onto a membrane, three copies of coatomer assembled to form a trimer or triad with three-fold symmetry that makes up the basic unit for the assembly of the coat (Fig. 1C) [57]. The triads further oligomerise via four different intertriad linkages [46][47][48]57]. Linkages I and II represent trimeric interactions with threefold symmetry formed between the COPI triads while linkages III and IV are formed by interactions between four COPI triads to generate two-fold pseudosymmetry (Fig. 2). The interfaces of linkages I and IV are composed of interactions between a-COP and e-COP, while linkage II is mediated by contacts by the l-homology domain (MHD) of d-COP. By linking COPI triads together, curvature induced by the triads can be propagated over larger areas of the membrane to allow vesicle formation [46,47]. Linkage via d-COP and e-COP cannot be essential for vesicle formation because e-COP is dispensable for COPI function, as is the MHD of d-COP [58][59][60]. Even the combined deletion of both these components is not lethal in yeast [60].
There are several possible explanations for the functional inconsistency between the dispensability of e-COP and the MHD of d-COP in vivo and their position mediating contacts at the triad interfaces. Additional factors, such as tethering factors or other protein components of COPI vesicles, might compensate in vivo when e-COP is depleted or the MHD of d-COP is deleted. Alternatively, linkages comprised of interactions between other COPI components may replace the linkage types observed with wild-type coatomer. These ideas can be tested by biochemical experiments that assess the in vitro assembly of coated vesicles lacking e-COP and the MHD of d-COP or   with mutations in the regions that mediate contacts combined with corresponding in vivo structural data. Another possibility is that regions of COPI distal from the triad interfaces may form long-range interactions to tie the triads together. There are several flexible and disordered regions of COPI subunits that are not well resolved in either the triad or linkage structures that could serve to bring the COPI triads together during coat assembly. It will be important to characterize the position of the disordered domains of coat components in the COPI triad and the interactions formed by these domains within the assembled COPI coat in order to understand the functional roles of the disordered regions of coatomer in vesicle assembly. The architectures of linkages observed in in vitro assembled vesicles correspond well to those of the linkages observed for the in situ determined structure of the COPI coat [48]. The linkages between COPI triads have only been resolved to resolutions between 15 and 17 A for linkages I, II and III and around 30 A for the lower abundance linkage III for the in vitro structure. The lower resolution of these regions relative to the resolution of the triad itself may be due to increased flexibility in these regions or structural heterogeneity in the specific contacts formed between COPI subunits even within a given linkage. Determination of the structure of the triad linkages to higher resolution, both in vitro and in situ, will be important for obtaining a better understanding of the key contacts required to mediate the assembly of the COPI coat.
Thus far, structural studies of the solution state of coatomer-the form present in the cytoplasm prior to membrane binding-have been limited to crystal structures of individual subunits or complexes of two subunits and a relatively low-resolution structure of soluble coatomer determined by negative stain EM [61]. These studies do not reveal how the overall organization of the soluble coatomer complex differs from the membrane-assembled state. The clathrin adaptor complex, AP2, undergoes significant conformational changes between its cytosolic form, which adopts a closed state, and its membrane-associated form, which adopts a much more open state [62,63]. Biochemical studies suggest that there are conformational differences between the cytosolic and membrane-associated states of coatomer [46,47,52,[64][65][66]. It is therefore an important open question whether and how the conformation of coatomer changes upon membrane association and what role conformational changes play in the regulation of membrane binding and coat assembly. Significant flexibility in the soluble coatomer complex may preclude the determination of high-resolution structures of the entire complex, but structural studies of complexes with selected domains removed may yet provide information about functionally important conformational changes that occur upon membrane binding.

Regulation of COPI vesicle formation by Arf family GTPases
Overview of the function of the Arf GTPases In addition to the seven COPI coatomer proteins, small GTPases from the ADP-ribosylation factor (Arf) family are also stoichiometric components of the COPI coat. Arf proteins are responsible for COPI coat protein recruitment to membranes to initiate coated vesicle biogenesis and also are involved in cargo sorting, vesicle scission and vesicle uncoating [29,31,34,67,68]. There are six mammalian Arf proteins (humans have five: Arf1, Arf3, Arf4, Arf5 and Arf6), which are separated into three classes on the basis of sequence homology [69]. Class I consists of Arf1, Arf2 and Arf3 and is found in all eukaryotes, while classes II (Arf4 and Arf5) and III (Arf6) are less conserved [70]. In yeast, simultaneous deletion of Arf1 and Arf2 results in a lack of viability suggesting an overlapping function of these two proteins not provided by Arf3 [71]. In humans, all five Arf proteins are ubiquitously expressed, with Arf1, Arf3, Arf4 and Arf5 present on Golgi membranes and Arf6 at the plasma membrane [72]. Arf1 and Arf4 play the most significant role in COPI vesicle formation, as simultaneous depletion of this pair of Arf proteins results in significant disruptions to Golgi morphology and COPI protein localization [73][74][75][76][77].
In the context of the assembled COPI coat, two Arf1 proteins bind to each COPI leaf at nonequivalent positions, with a total of six Arf1 proteins bound to each triad. One copy of Arf1 (bArf) is bound at the periphery of the COPI leaf where it interacts primarily with b-COP and the N-terminal portion of d-COP while the other copy of Arf1 (cArf) binds at a central position within the leaf where it interacts with c-COP (Fig. 1C) [46,47]. Both copies of Arf1 are positioned proximally to membrane but are sufficiently exposed to the cytosol to recruit ArfGAP proteins.
Arf proteins and their effector molecules, ArfGEFs and ArfGAPs, play multiple roles in the cycle of COPI-coated vesicle assembly and disassembly. Exchange of GDP for GTP within Arf is required for recruitment of the COPI coat to the assembling vesicle [67]. GTP hydrolysis by Arf is required for the efficient uptake of some classes of cargo into COPI vesicles  [29,78]. Vesicle scission is independent of GTP hydrolysis by Arf1 but depends on Arf1 dimerization [10,19,32,57,68,79,80]. Disassembly of the COPI coat is triggered by hydrolysis of the GTP bound to Arf1. Each of these GTP-dependent activities depends in turn on the activity of Arf1 GTPase-activating proteins (ArfGAPs), due to the low inherent GTPase activity of Arf1.

Activation of Arf GTPases by ArfGEFs
Arf GTPases have overall structural similarity to the Ras superfamily of small GTP-binding proteins, which are characterized by a GTP-binding domain containing nucleotide-sensitive regions called the switch 1 and switch 2 loops (Fig. 3A) separated by the interswitch domain [81,82]. Conversion of Arf-GDP to its activated form, Arf-GTP, requires interaction between Arf and a Sec7 catalytic domain found in all ArfGEFs (Fig. 1B) [83,84]. The ArfGEF that is responsible for mediating activation of Arf1 and initiating COPIcoated vesicle formation at the cis-Golgi is called GBF1 in mammals, with yeast homologs Gea1p and Gea2p [73,74,76]. GBF1 is localized to the cis-Golgi and the ER-Golgi intermediate compartment (ERGIC) and is targeted to Golgi membranes through a domain  that binds phosphoinositide lipids [85]. Membrane association of GBF1 and Arf1 is essential for GBF1 to successfully displace GDP from Arf1 [86]. GBF1 and Arf1 can each form transient associations with the membrane, but the interaction between the two proteins mutually stabilizes the association of both proteins with the membrane [87]. After nucleotide exchange, Arf1 remains membrane bound while GBF1 dissociates from the membrane. Displacement of the GDP nucleotide bound to Arf1 is accomplished through interactions between Arf1 and the Sec7 domain of the ArfGEF. In the GDP-bound state of Arf1, the interswitch domain of Arf1 is retracted and forms a binding pocket for the myristoylated N-terminal amphipathic helix of Arf1. Recent cryo-EM structures of the full-length Gea2p (a yeast homolog of GBF1) have revealed that the Sec7 domain can adopt two conformational states: one that binds Arf1-GDP (closed state) and one that binds to nucleotide-free Arf1 (open state) [88]. A hydrophobic groove on the surface of the Sec7 domain of Gea2p binds to switch 1 and 2 of Arf1-GDP (Fig. 3B) [26,88,89]. A conserved glutamate residue on the Sec7 domain of GBF1 interacts with the nucleotide-binding side of Arf1 to promote dissociation of GDP coincident with the transition of the Sec7 domain into the open state, stabilizing a nucleotide-free complex [26,83,85,86,90,91]. A conserved linker region adjacent to the Sec7 domain appears to be important for controlling the interconversion between the two conformational states of Gea2p [88]. These interactions cause a conformation change in the Arf1 interswitch region to expose the myristoylated N-terminal helix, which can then insert into and stabilize the interaction with the membrane (Fig. 3A) [45,81,82,[92][93][94]. These structures help explain the observed mutual stabilization of interaction with the membrane that occurs for GBF1 and Arf1 [87,88]. Subsequent binding of GTP to Arf1 leads to dissociation from the Sec7 domain. GTP-bound Arf1 is able to interact with coatomer, recruiting coatomer to the membrane to initiate the formation of COPI-coated vesicles [45,95,96]. The membrane-inserted myristoylated amphipathic helix from Arf1 contributes to the generation of positive membrane curvature during vesicle budding [97][98][99]. GBF1 may also promote the recruitment of coatomer as it has been found to have a direct interaction with the c-COP subunit of coatomer [100]. Whether GBF1 forms additional interactions with other subunits of coatomer and whether these interactions are functionally significant for the formation of COPI vesicles is not yet clear. Characterization of how GBF1 interacts with coatomer and Arf1 in the context of the assembling COPI coat will be important to allow an understanding of how GBF1 modulates vesicle assembly.

Structure and function of ArfGAPs
The interaction between ArfGAP and Arf is mediated by ArfGAP binding to switches 1 and 2 in Arf [101,102]. Arf GTPases lack an inherent GTP hydrolysis activity because the active site for GTP hydrolysis is incomplete, lacking a catalytic arginine-the arginine finger. ArfGAP proteins complete the active site by contributing this arginine, and together with a glutamine residue in Arf stabilizes the transition state of the GTP hydrolysis reaction, thus catalyzing hydrolysis [102].
Two types of GAP proteins function as GTPase activators in the COPI system that are differentiated on the basis of their noncatalytic domains. ArfGAP1 (with the yeast homolog Gcs1p) was the first protein identified with GAP activity for Arf, and it contains domains that respond to increasing membrane curvature with an increase in GTPase stimulatory activity [103][104][105][106]. ArfGAP1 also has a C-terminal coatomer-binding site that interacts with the MHD of d-COP [107,108]. ArfGAP2 and ArfGAP3 (with the yeast homolog Glo3p), by contrast, do not contain domains that respond to membrane curvature, and their activity is instead stimulated by interaction with coatomer mediated through contacts with c-COP [40,[109][110][111][112].
ArfGAP1 and ArfGAP2/3 proteins have distinct but overlapping activities. ArfGAP1 is evenly localized throughout the Golgi, while ArfGAP2/3 is enriched at the cis-Golgi, with a gradient of localization from cis to trans [113]. In yeast, either Gcs1p or Glo3p is required for COPI-mediated transport to occur [110,113]. However, some evidence suggests that Arf-GAP2/3 may contribute more to COPI vesicular transport. Genetic work in yeast has shown that Glo3p deletion results in much stronger Golgi-to-ER retrograde transport defects and in vitro assembled vesicles coated with yeast COPI were found to associate with Glo3p but not Gcs1p [109,112,114]. In human cells, ArfGAP2/3 is required for the assembly of the COPI coat, while ArfGAP1 is not [115].
When either ArfGAP1 or ArfGAP2 was incubated with in vitro assembled COPI-coated vesicles, only the structure of ArfGAP2 bound to the COPI coat could be determined, consistent with biochemical data that suggested that very little ArfGAP1 binds stably to the COPI coat [46,112,114,116]. ArfGAP2 binds adjacent to the centrally-located Arf1 protein (cArf), with no extra electron density detected near the peripherally located Arf1 protein (bArf) (Fig. 3C), suggesting that the central and peripheral Arf1 molecules are differently regulated and may have distinct functional roles. When bound to the central Arf1 protein, the catalytic domain of ArfGAP2 is located proximally to the nucleotide-binding site, consistent with the required interaction between the arginine finger of ArfGAP2 and the catalytic site in Arf1. Structures of the noncatalytic domains of ArfGAP2 have not been resolved. These domains play important roles in mediating the binding of ArfGAP2 to COPI, and revealing their position and structure is essential to understand how ArfGAP2 distinguishes bArf and cArf.

The roles of Arf in COPI vesicle biogenesis
In addition to the function of Arf GTPases in the recruitment of coatomer to the membrane to initiate vesicle assembly, Arf has also been implicated in other stages of the COPI vesicle cycle. An Arf1 mutant that is incapable of dimerization retains its ability to recruit coatomer to the membrane and initiate COPI vesicle budding, but scission does not occur [32]. Chemical crosslinking of this mutant Arf to stabilize the dimer state restores vesicle release. These observations imply that a monomeric form of Arf1 recruits COPI triad to the membrane, while an Arf1 dimer is required for COPI vesicle scission [32,97]. Scission has been shown to be independent of GTP hydrolysis [117]. A recent study demonstrated that Arf1 dimers segregate to donor membranes during in vitro vesicle formation, with only Arf1 monomers incorporated into released vesicles [68]. Arf1 can therefore occupy (at least) three different environments during the life cycle of a COPI vesicle: dimeric Arf that remains on the donor membrane, bArf located at the periphery of the COPI triad and cArf located at the center of the COPI triad. Each of these different Arf subtypes may be differentially regulated. Recruitment of certain types of cargo depends on GTP hydrolysis, as does vesicle uncoating, but these processes likely occur at different stages of vesicle transport [118]. Characterization of the timing of GTP hydrolysis during COPI vesicle formation will be important to allow an understanding of how Arf regulates COPI vesicle biogenesis.

Cargo recognition
In addition to the components that are required for vesicle formation, COPI-coated vesicles contain several other protein components including Golgi-and ERresident cargo proteins, constitutively cycling proteins such as the p24 family of proteins, and cargo receptors and adaptors. Incorporation of each of these classes of COPI-coated vesicle components is mediated by distinct interactions with various coatomer proteins and different recognition motifs.
For ER-resident cargo proteins that directly interact with coatomer, sites in the N-terminal b-propeller of a-COP and b 0 -COP bind to dilysine motifs present in the cargo protein, K(X)KXX, where X represents any amino acid [12,43,44,[119][120][121]. Cargo-binding sites have also been proposed in b-COP and d-COP, which recognize arginine-based signaling motifs with the recognition sequence RXR [122]. Recent work has also identified binding sites in d-COP and ξ-COP that recognize Φ-(K/R)-X-L-X-(K/R) motifs in combination with the C-terminal carboxylic acid (where Φ represents an aromatic amino acid) that are present in several Golgi-localized glycosyltransferases [123].
p24 proteins are a family of proteins that constitutively cycle in COPI-coated vesicles and are important for COPI vesicle biogenesis [124]. p24 proteins have two recognition motifs in their cytoplasmic tails: a dibasic motif similar to the recognition motif that mediates the transport of ER-resident proteins and a diphenylalanine motif, which together form the recognition signal FFXX[KRH][KRH]B(X) n . This motif binds to sites in c-COP [125]. Uptake of p24 proteins into COPI vesicles does not depend on GTP hydrolysis, and in fact, p24 proteins may reduce GTP hydrolysis activity by inhibiting ArfGAP activity, possibly promoting vesicle formation by preventing hydrolysis of GTP from occurring before vesicle assembly and scission is complete [125,126].
Coatomer interacts with K63-linked polyubiquitin of a ubiquitinated yeast SNARE Snc1 [127]. A more recent study demonstrated that ubiquitin binding by the N-terminal b-propeller of b 0 -COP is required for the correct localization of multiple SNAREs suggesting that ubiquitin recognition is a more general mechanism of cargo recruitment into COPI vesicles [128].
COPI-coated vesicles also transport cargo via interactions with a variety of cargo receptors and adaptors [129]. Cargo receptors transport lumenal cargo that cannot interact directly with coat components on the cytosolic face of the lipid bilayer and also assist in the transport of membrane proteins with recognition sequences distinct from those recognized by the coatomer proteins. The KDEL receptor mediates retrograde transport of ER-resident cargo with the recognition signal K/HDEL in a pH-dependent manner [130,131]. Cargo adaptors are soluble cytoplasmic proteins that mediate the transport of other proteins through direct interaction with both their cargo and the COPI coat proteins. Other cargo receptors and cargo adaptors include the Rer1 complex, which transports ERresident proteins that lack either the K(X)KXX or the K/HDEL signals and the VPS74/Golph3 protein that coordinates the transport of Golgi-resident glycosylases by binding the motif F/L-L/V/I-X-X-R/K [132][133][134][135][136][137].
Although recognition motifs in cargo proteins and binding sites in COPI proteins have been identified for many COPI vesicle components, how each of these components interacts with the COPI coat is not well understood. It will be important to characterize how cargo proteins and cargo receptors interface with the COPI coat, including whether and to what extent the presence of specific cargo proteins/receptors modulates the structure of the assembling COPI coat. COPI vesicle assembly and transport may proceed independently of the levels of cargo loading or the presence of specific cargo. It appears more likely that COPI vesicle assembly is at least modulated in response to the loading of particular cargo [138]. Investigating these possibilities will provide new insight into the functional roles of COPI vesicles.

Tethering factors
Mature COPI vesicles have to be transported to the target membrane. This process is directed by tethering factors-target membrane-associated proteins that mediate the binding of vesicles to the target membrane. COPI vesicles make use of both Golgi-and ER-associated tethering factors (Fig. 4A). They can be grouped into two main classes: coiled-coil tethers and multisubunit tethering complexes [35,36].
Coiled-coil tethers typically reside in the Golgi membrane (golgins), are characterized by pronounced coiled-coil motifs and are long (several nanometers in length) and flexible [139]. These Golgi-resident tethers anchor COPI vesicles to different Golgi cisternae [140]. Stretches of 20-50 amino acids at the N-termini of these tethers are responsible for their vesicle tethering activity [139]. The coiled-coil tether, RAB6-interacting golgin (GORAB) was also reported to scaffold COPI vesicle assembly in the trans-Golgi [141].
Multisubunit tethering complexes comprise a diverse group of proteins that are nevertheless functionally similar to each other. They include the Dsl1 complex and the conserved oligomeric Golgi (COG) complex [36]. The yeast Dsl1 complex (with the mammalian analogue ZW10) is one of the most studied COPI tethering factors. The ER-resident Dsl1 complex consists of three proteins: Tip20, Sec39 and Dsl1. Tip20 and Sec39 resemble two legs bound by Dsl1, which acts as a flexible joint. Structures of all three proteins are available [142,143]. The Dsl1 complex is thought to interact with the COPI coat through a flexible domain of Dsl1 called 'lasso' (Fig. 4B). The 'lasso' domain contains tryptophan-containing motifs that act as binding sites for a-COP and the d-COP MHD domain [144]. a-COP and the d-COP MHD are involved in the formation of the intertriad linkages in the COPI coat. Interestingly, a-COP has its own lasso-like domain with a tryptophan-containing motif. The ability of a-COP to form oligomers in vitro and its role in stabilizing COPI linkages in vivo hints at a model where Dsl1-based COPI tethering promotes further COPI vesicle uncoating through competition between a-COP-a-COP and Dsl1-a-COP connections [145]. The distal ends of the two legs of the Dsl1 complex interact with N-terminal Habc domains of SNAREs Sec20 and Use1. Tip20 interacts with Sec20 while Sec39 binds Use1 [146]. Recent structural study suggests that the Dsl1 complex is anchored to the ER membrane through these interactions. Binding of Sec20 and Use1 to the Dsl1 complex results in a conformational change that forces their SNARE motifs into the orientation required for subsequent SNARE complex assembly [146].
COG is a hetero-octameric, evolutionary-conserved complex that also can tether COPI vesicles [147]. Unlike Dsl1, it is localized in cis-and medial-Golgi and mediates the tethering of COPI vesicles to the Golgi itself. COG interacts with c-COP [148]. Evidence shows that COG interacts with SNARE machinery suggesting that it may also be involved in vesicle fusion [149][150][151].
As noted above COPI tethering factors may promote COPI coat disassembly and play a role in vesicle fusion by facilitating SNARE complex assembly. However, neither the possible link between tethering and uncoating nor the regulation of SNARE interactions during COPI fusion is well understood.

Open questions and future perspectives
The structure of COPI coat proteins and the mechanism of COPI-coated vesicle biogenesis have been the subject of extensive research efforts; however, many open questions remain (Box 1). Studies of other coat proteins, in particular AP2, suggest that the regulation of coat assembly is linked to structural changes between the cytosolic and membrane-bound forms of the coat. To reveal these structural changes and allow their interpretation, it is therefore essential to obtain a structural understanding of the cytosolic form of coatomer.
Coat assembly not only involves interactions between folded protein domains, which are well resolved in the available structures of the coat but also involves longer-range interactions with short motifs within otherwise disordered regions of the protein subunits. To reveal the nature of these interactions and how they might contribute to regulated coat assembly, higher-resolution structures of the COPI triad will be needed.
Studies investigating the architecture of COPI coats showed that COPI comprises 4 types of linkages between the triads. Although these structures have only been resolved to low resolution, we can identify some subunits involved in intertriad interactions: namely, a-COP and e-COP in linkages I and IV, and d-COP in linkage II. a-COP and d-COP were also reported to be a potential binding target for Dsl1 complex that is responsible for the tethering of COPIcoated vesicles to the ER. Interestingly, simultaneous deletion of both e-COP and the MHD of d-COP is not lethal suggesting that other proteins might be involved in the linkage formation [60]. Obtaining higher-resolution structures of COPI linkages in vitro may be able to resolve these discrepancies.
Understanding the lifecycle of the COPI coat will require additional structural data on the coat in complex with ArfGEFs and ArfGAPs as a basis for further mechanistic and biochemical insight into the COPI vesicle life cycle. Structural and biochemical work has revealed that there are at least three distinct environments for Arf during the lifecycle of a COPI vesicle. Although the dimeric membrane-associated state of Arf is not incorporated into mature COPI vesicles, it is required to allow vesicle scission. In the assembled coat, Arf occupies two distinct niches where it interacts with different subsets of COPI proteins and possibly also with different ArfGAPs. The significant difference between these Arf-binding sites raises the possibility that the two molecules of Arf may not undergo GTP hydrolysis simultaneously, a hypothesis that could explain the seemingly contradictory biochemical studies that implicate GTP hydrolysis in cargo loading during vesicle formation but also in vesicle uncoating. Understanding the timing of GTP hydrolysis during COPI transport and obtaining a detailed view of intermediate states of COPI vesicle assembly and disassembly will help to answer these important functional questions.
Finally, COPI coat components interact with cargo proteins, cargo receptors and tethering factors. There is limited information available on how these interactions take place in the context of the assembling, assembled or disassembled coat. Here again, obtaining structural data on the coat in complex with tethers, specific cargoes and cargo receptors, can lay the groundwork for functional and mechanistic studies. Many of these questions can be best addressed through a combination of in vitro structural studies with studies of COPI-coated vesicles in situ. In situ structural studies allow the study of different stages of the COPI vesicle life cycle in the native context and have the potential to reveal populations of COPI vesicles containing particular cargo or cargo receptors/ adaptors or bound to different tethering complexes.