Life of a clathrin coat: insights from clathrin and AP structures

Key Points

• Membrane sorting between secretory and endocytic organelles is predominantly controlled by small carrier vesicles or tubules that have specific protein coats on their cytoplasmic surfaces. The most widely studied of the various membrane-transport intermediates are clathrin-coated vesicles.

• By positioning crystal structures of domains of clathrin and auxilin (auxilin links clathrin to the 'unfoldase' Hsc70) in cryo-electron-microscopy maps of clathrin cages that were assembled in the presence of auxilin, molecular explanations have been obtained for how clathrin-coat disassembly might proceed.

• Many clathrin adaptors, which are defined as proteins that link polymerized clathrin to membranes and their embedded proteins, have different sequences and structures, but have a common architecture that consists of folded domains that are linked by extended, structureless linkers.

• Structural and biochemical studies on the heterotetrameric adaptor protein complexes (APs), which are the most abundant of all the clathrin adaptors, show how they carry out their numerous and often phosphorylation-regulated functions. These functions include binding protein cargo, phospholipids and clathrin, and coordinating the recruitment of many of the other proteins that are involved in assembling and subsequently disassembling clathrin-coated vesicles.

Abstract

Membrane sorting between secretory and endocytic organelles is predominantly controlled by small carrier vesicles or tubules that have specific protein coats on their cytoplasmic surfaces. Clathrin–clathrin-adaptor coats function in many steps of intracellular transport and are the most extensively studied of all transport-vesicle coats. In recent years, the determination of structures of clathrin assemblies by electron microscopy, of domains of clathrin and of its adaptors has improved our understanding of the molecular mechanisms of clathrin-coated-vesicle assembly and disassembly.

Main

The transfer of macromolecules in the secretory and endocytic pathways occurs by the trafficking of membrane-bound vesicles between donor and acceptor compartments. During the formation of transport vesicles, the protein and lipid cargoes that are to be transported become concentrated in a patch on the donor membrane. Once a protein coat has stably assembled on this patch^1,2, the membrane becomes deformed and is eventually 'pinched off' to form a vesicle that is then targeted to, and ultimately fuses with, an acceptor membrane.

Transport vesicles are classified according to the components of the protein coat that surrounds them during their genesis and early life. One of the most common and probably best-characterized classes of coated vesicle is that comprising three-layered clathrin-coated vesicles (CCVs). CCVs are so-called because the main component of the coat is clathrin, which forms a polymeric mechanical scaffold on the vesicle surface. The inner, membrane layer with its embedded cargo is linked to the outer, clathrin layer by a middle layer that consists of various clathrin-adaptor molecules and other proteins that have accessory/regulatory roles in CCV assembly (Box 1). A clathrin adaptor is a protein that can connect the purely mechanical clathrin scaffold and a component of the membrane, be it a phospholipid, a transmembrane protein or both simultaneously. At least 20 clathrin adaptors have been identified³. They share a common design principle in that they are composed of compact folded domains and long unstructured regions through which they bind to the N-terminal, 330-residue β-propeller terminal domains (TDs) of clathrin and also to each other. These interactions are almost always mediated by short, linear motifs that interact with a folded domain, which results in a moderate-to-weak strength of binding (K_d = 1–100 μM) with a relatively short half life (t_1/2 = 1–10 s). Such interactions are frequently found between proteins that are involved in transport-vesicle coat formation⁴, as well as in other systems, including signal-transduction pathways, for which dynamic interactions (that is, binding partners that can be exchanged easily) are required.

The first clathrin adaptors to be identified were the protein-cargo-binding and phospholipid-binding heterotetrameric adaptor protein complexes (APs) AP1 and AP2, and, to this day, they serve as functional paradigms for all clathrin–membrane/cargo adaptors. This review focuses on how the high-resolution structures of AP1, AP2 and domains of clathrin, in combination with electron microscopy (EM) studies on clathrin assemblies, have brought us to a point where we can understand their functions in CCV formation at a molecular level. A large amount of structural and biochemical work has been carried out on other clathrin–membrane adaptors. These studies have shed light on the mechanical processes of membrane deformation (involving epsins and amphiphysins) and the mechanisms by which specific cargoes are recruited into CCVs (involving clathrin-associated sorting proteins (CLASPs) such as ARH and Dab2, as well as the arrestins). However, these studies do not fall within the scope of this review (for reviews of these topics, see Refs 3,5 and references therein).

Early structures from EM studies gave us views both of clathrin cages^6,7,8,9,10 and of the individual trimeric building blocks (known as triskelia) from which they are built, as well as of APs. The resolution of these studies was 50–100 Å and revealed no more than 'fuzzy outlines' of these molecules that could not be interpreted in terms of a molecular mechanism for CCV formation. Since then, many biochemical and functional studies have been carried out (reviewed in Refs 11, 12) and high-resolution X-ray crystallography structures of several clathrin domains have been described^13,14,15. Now, the structures of whole clathrin cages at resolutions of up to ∼8 Å in the presence of various interacting proteins have been determined^16,17,18,19, into which high-resolution structures of the appropriate proteins can be fitted. High-resolution structures of all of the domains of AP1 and AP2 (with the exception of the AP1 β1-appendage), often in complex with their appropriate peptide ligands, have also been recently elucidated (see later). From these structures, models of whole, intact APs have been built.

We are now in a position to explain much of the existing wealth of functional data in terms of a molecular mechanism for CCV formation. The high-resolution structures also allow structurally non-disruptive point mutations to be designed, which can be used to test functional models in vitro. The ability to remove a wild-type protein efficiently using small interfering RNA (siRNA) in cells that are stably transfected with a gene encoding a mutated version of the protein of interest is making it possible to test the effect of these mutations in vivo. The structures are also predictive, allowing us to propose models for the spatial, temporal and regulatory aspects of this crucial step in membrane trafficking, and to formulate general principles that might apply to the formation of all transport vesicles.

Clathrin

A clathrin coat is a three-dimensional (3D) array of triskelia. Each triskelion is made of three 1,675-residue (approximately 190-kDa) clathrin heavy chains (CHCs) and three 25–29-kDa clathrin light chains (CLCs), and has an approximately three-fold rotational symmetry (Figs 1,2). Purified clathrin can spontaneously assemble into cages or, in the presence of adaptor proteins, into coats in vitro. Clathrin coats come in different shapes and sizes²⁰. The smallest symmetrical form is the hexagonal barrel and studying this barrel has enabled structural biologists to determine the structure of assembled clathrin using cryo-EM and single-particle analysis.

Figure 1: Clathrin-cage reconstructions.

a | Clathrin-cage reconstructions have increased in resolution over time. The first images of clathrin barrels were obtained using negative-stain electron microscopy (EM) in 1969 (Ref. 6). Subsequently, using cryo-EM reconstructions, the resolution increased from approximately 50 Å in 1986 to 8 Å in 2004 (Refs 8, 9, 16, 17). The far left panel is reproduced with permission from Ref. 6 © (1969) The Rockefeller University Press and the far right panel is modified with permission from Ref. 17 © (2004) Macmillan Magazines Ltd. b | An ∼8-Å reconstruction of a clathrin barrel with the light chains highlighted in yellow. Reproduced with permission from Ref. 17 © (2004) Macmillan Magazines Ltd. c | An ∼8-Å reconstruction of a clathrin barrel with one cluster of terminal domains highlighted in red. Modified with permission from Ref. 17 © (2004) Macmilan Magazines Ltd. d | A 12-Å reconstruction of a clathrin barrel that was formed in the presence of a 39-kDa fragment of auxilin. The auxilin J domain is highlighted in purple. Modified with permission from Ref. 18 © (2004) Macmillan Magazines Ltd.

Figure 2: The architecture of clathrin.

a | A clathrin barrel with a single triskelion highlighted in blue. Modified with permission from Ref. 97 © (2005) Blackwell Publishing and from Ref. 17 © (2005) Macmillan Magazines Ltd. b | A schematic representation of a clathrin triskelion, which highlights the various domains using different colours (see the key). The clathrin-heavy-chain-repeat modules (CHCRs) that are involved in each domain are listed. c | A Cα trace of a single clathrin-heavy-chain molecule with the domains coloured as in part b. A potential HSC70 (70-kDa heat-shock cognate protein)-binding motif (QLMLT) near the C terminus is labelled. This structure was created using coordinates from the Protein Data Bank accession file 1XI4. Beneath this trace are linear representations of auxilin and HSC70 that illustrate their domain structure, the interactions these domains make and the position of a clathrin-box motif in auxilin. The domains of auxilin are phosphatase-and-tensin-homologue domain (PTEN), clathrin-binding domain (CBD) and J domain (JD). The domains of bovine HSC70 are nucleotide-binding domain (NBD), substrate-binding domain (SBD) and C-terminal domain (CTD). Interactions are shown using arrows — biochemically defined interactions are shown using solid lines, whereas proposed interactions are shown using dashed lines.

CCVs were first visualized by negative-stain EM⁶, and clathrin triskelia were first visualized using rotary shadowing⁷. These studies were followed by an analysis²¹ of the assembly and packing of cage fragments, which again used negative-stain EM. However, the first 3D map of a clathrin cage, both in the presence and absence of adaptor proteins, was obtained in 1986 using cryo-EM^8,9 (Fig. 1a). This was one of the first examples of the use of cryo-EM methods to determine protein structure, and was a significant milestone in the development of the technique. It was not until 1998 that the resolution was improved to 21 Å (Ref. 16), which allowed the main parts of the triskelia to be traced and the TDs to be roughly positioned (Fig. 1a). Fotin and colleagues have now extended the resolution even further by obtaining two 3D maps of assembled clathrin using cryo-EM^17,18. The first is a 7.9–11-Å-resolution map of clathrin alone, both with and without light chains¹⁷ (Fig. 1b,c), and the second is a 12-Å map of clathrin bound to auxilin¹⁸, an integral coat protein that is required by the chaperone HSC70 (70-kDa heat-shock cognate protein) during its uncoating of CCVs (Fig. 1d). The resolution of all of these maps is significantly higher than has previously been published and many new insights into the structure and function of assembled clathrin have been exposed. In particular, the nature and location of the CLCs has been confirmed (yellow in Fig. 1b), the shape adopted by the TDs (red in Fig.1c) and linker has been defined more fully, and a whole new level of detail has become available on the nature of the cage vertices, including the trimerization domain and the area beneath it.

Model of clathrin density to residue 1,630. The 7.9-Å-resolution EM map of clathrin that was determined by Fotin et al. was obtained using clathrin enriched in the form of hexagonal barrel cages through the addition of substoichiometric amounts of APs. They used existing knowledge of the repeating units of stacked α-helical hairpins that are present in clathrin¹³, together with EM data, to try to model the structures of the sections of clathrin that had not been determined by X-ray crystallography. The resulting structural model showed that the α-helical hairpins do indeed repeat throughout a clathrin molecule (Fig. 2c), with the only deviations being at the N and C termini. At the N terminus, the increased resolution allowed a more accurate fit of the TD than had been previously possible²² and the map confirmed the maintenance of the helical hairpin repeating structure throughout the linker region that joins the TD to the ankle segment of the triskelia (Fig. 2c).

The study by Fotin et al. also provided a difference map of clathrin with and without CLCs, which confirmed the nature and location of CLCs on CHCs. The density corresponded to a single α-helix, 71 residues in length, that runs along the outer edge of the cage (yellow in Fig. 1b). A model of such a helix that corresponded to the CLC sequence was successfully docked into the electron-density map using information from mutagenesis studies, and confirmed earlier modelling studies that were carried out by Chen et al. before these structural data were available. These earlier studies had revealed two contacts (CHC-K1326–CLC-W108 and CHC-K1514–CLC-W130)²³. In both cases, docking that included one interaction showed that the model was in the correct position for the second interaction, which confirmed the accuracy of the fit. The role of CLCs in vivo is still unclear. It has been shown^24,25 that, as well as modulating clathrin assembly, they bind Huntingtin-interacting protein-1 (HIP1) and HIP1-related protein (HIP1R). In view of this, it seems probable that they function in a regulatory capacity. It is hoped that this new structural information will direct future mutagenesis studies that are designed to study this potential role further.

Perhaps the most interesting feature of the new EM map is the extra detail that is now available on the architecture of the clathrin vertices. Figure 3 shows the different features by focusing on one vertex. Travelling towards the centre of a cage, starting at a corner vertex, the first structure encountered is a so-called hub that is composed of three CHC proximal legs (blue in Figs 2,3a). These three legs are mainly held together by the 'helical-tripod' trimerization domain (purple in Figs 2,3a), which is made from the C-terminal 100 residues of each clathrin molecule. This domain follows each proximal leg (blue in Figs 2,3a) and is found directly beneath the place where the three proximal legs meet. As we travel down towards the centre of the cage, we encounter a triangular structure that is composed of three distal leg segments (green in Figs 2,3a) that come from three separate neighbouring triskelia, and below them are three ankle segments (yellow–orange in Figs 2,3a) from three further triskelia. Beneath these, there are three TDs (red in Figs 2,3a), which each come from three separate triskelia that are not connected to the distal leg segments or the ankle segments above them. These TDs curve round towards each other and almost create an enclosure underneath the distal and trimerization domains. Travelling further into the centre of the cage, if it were fully present, there would be the AP layer and, finally, if this was a native coated vesicle, there would be the membrane and cargo.

Figure 3: Molecular details of clathrin interactions.

a | Layers of clathrin molecules can be seen by taking a slice through a clathrin coat. The various domains of the clathrin molecule are coloured as shown in Fig. 2b,c. Moving from the outside towards the membrane (top to bottom), the outer layer is composed of proximal domains (blue) and trimerization domains (purple). Three trimerization domains, each a long single helix, come together to form a 'helical tripod' (purple). The next layer is formed primarily by distal domains (green), and the inner layer is composed of linker and ankle domains (yellow–orange) and terminal domains (TDs; red). In each layer, the clathrin domains involved come from different clathrin triskelia. The grey sphere indicates the approximate position of the auxilin J domain that was proposed by Fotin and co-workers¹⁸. b | A cross-section through a clathrin barrel. This shows the array of TDs (red) that extend from the scaffold, which is made up of the various leg domains (blue), towards the membrane to form a multivalent structure that can bind the extended, motif-rich portions of clathrin adaptors. The structures shown in parts a and b were created using coordinates from the Protein Data Bank accession file 1XI4. c | The interactions between a TD and two peptide motifs — a W-box motif (the carbon atoms of this motif are coloured cyan) and a clathrin-box motif (the carbon atoms of this motif are coloured green) — that are found in the extended regions of clathrin adaptors. The enlargements show the molecular details of these interactions. Specificity is the result of side chains fitting into chemically compatible pockets, with further strength coming from hydrogen bonds that are formed between the TD and the backbones of the peptides. Reproduced with permission from Ref. 3 © (2004) Annual Reviews.

The high-resolution EM map goes a long way towards defining the molecular architecture of the 'hub/trimerization assemblies'. Up to residue 1,597 of the CHC sequence, the density fits with the helical hairpin model. The final hairpins stack against each other forming the top of the trimerization domain (blue in Fig. 3a). From residues 1,598 to 1,630, the density moves into the cage as a 50-Å-long rod-like structure, probably an α-helix (purple in Fig. 3a). Three rods combine to form a tripod shape beneath the hub and end close to the triangle of distal domains that lies beneath the vertex. Finally, density that probably arises from the C-terminal residues 1,631–1,675 merges with the density of the ankle regions (yellow–orange in Fig. 3a) that lie below the distal domains.

Maps of clathrin with auxilin bound. When a fragment of auxilin, including its J domain, was bound to clathrin, the whole shape of the cage altered, and became more square and less well ordered (Fig. 1d). Density that related to this 39-kDa fragment of auxilin was clearly seen in the space beneath each clathrin vertex, which is created by a hub and a tripod assembly that is surrounded by triskelion ankles and TDs (Fig. 3a). Docking a five-helix model for auxilin (residues 800–910), in which three of the helices correspond to the known structure of the auxilin J domain²⁶, into this density accounted for approximately a third of the observed electron density that was not accounted for by clathrin. This is consistent with this fragment representing about a third of the molecular weight of the 39-kDa auxilin fragment that was bound to the cages. The fact that other parts of this fragment can bind to clathrin and therefore possess visible electron density is consistent with pulldown data from Scheele and colleagues^27,28, which showed that auxilin has several binding sites for clathrin that are characterized by a range of motifs including DLL, DPW, WDW and NWQ. The motifs between residues 547–715 are believed to bind mainly to TDs, whereas motifs between residues 716–910 are believed to bind mainly to the distal and proximal legs of clathrin, and some of these regions of clathrin border the volume occupied by the auxilin electron density (Fig. 3a).

HSC70 is recruited to clathrin coats through the auxilin J domain, which also stimulates the ATPase activity of the HSC70 chaperone. It has always been intriguing how HSC70, which usually carries out its unfolding/disassembly function by binding to unfolded polypeptides, might function on clathrin, which presumably remains folded during CCV uncoating. Now we can see that the location of the auxilin J domain means that HSC70 could be recruited into the space beneath the vertex where a large number of inter-triskelion contacts are made. Crucially, the trimerization domains and potentially structureless C termini of clathrin meet with the distal domains and ankle segments of other triskelia at this position. The map from Fotin et al. indicates that the C terminus of a CHC merges with the density of the ankle segments. Interestingly, the C terminus is proline rich and is therefore probably unfolded. Fotin et al. also noted a possible HSC70-binding motif in the proline-rich domain — QLMLT (Fig. 2c) — which is similar to a known HSC70-binding peptide motif (FYQLALT)²⁹. The authors propose that auxilin introduces a change to the ankle-crossing contacts, which could release the C-terminal segments from their interaction with the ankles, priming these segments for HSC70 binding. HSC70 binding to these C-terminal segments could destabilize the numerous interactions that link several individual triskelia together and, in this way, promote triskelion release from the cage. This would initiate the cascade of triskelion dissociation from the clathrin coat that results in the uncoating of the CCV — that is, the 'death' of the coat. The predicted location of HSC70 at the vertex is consistent with early studies, which indicated that up to three HSC70 molecules could bind to each clathrin vertex³⁰.

The model proposed by Fotin et al. for HSC70 binding and function is an attractive one because of the undeniable logic of achieving uncoating by disrupting a site where so many contacts are present in such a small space. However, a recent, albeit much lower resolution, 28-Å map of HSC70 bound to clathrin cages might pose a challenge to this model. Heymann et al.¹⁹ obtained a map of clathrin cages that were assembled in the presence of a recombinant fragment containing the auxilin J domain joined to part of the AP180 clathrin-binding domain. The proposed location of HSC70 that was derived from this map differs from the location predicted by Fotin and colleagues. However, it should be noted that the samples used had different components and that both publications agree that the J domains, to which HSC70 binds, lie close to the clathrin TDs. What is clear is that each map poses interesting questions as to the mechanism of action of HSC70, which should be investigated using biophysical techniques as well as by striving for a map of HSC70 bound to clathrin at a much higher resolution.

The results of Fotin et al.¹⁸ are consistent with the proposal of Smith et al.³¹ that the N-terminal PTEN (phosphatase and tensin homologue) domain of auxilin might contact the membrane (Fig. 2c). The structure of Fotin et al. contains only residues 547–910 of auxilin, whereas the structure of Smith et al. contains the intact protein. The key difference between the two is the presence in the work of Smith et al. of density close to where the membrane would be in a native coated vesicle. Placing the PTEN domain innermost on the membrane would allow a clathrin-box motif at residue 495, which was identified by Smith et al., to bind nearby TDs. The 547–910 segment, which includes the J domain at its C terminus, would then be able to interact at several points along its length with the proximal and distal leg regions that surround the auxilin density in the Fotin et al. map. Much of the auxilin fragment between residue 547 and the J domain is unstructured in solution^28,32, but the Fotin et al. map indicates that, when it is bound to clathrin, this part of auxilin is either fully folded or, in some way, condensed to occupy space beneath the clathrin vertex. This raises the possibility that auxilin alters its conformation on binding to clathrin, as has indeed also been shown for the CLCs²³. However, biophysical data that support these ideas are not yet available.

Taking a view in the opposite direction — that is, outwards from where the vesicle membrane would be — reveals a 'forest' of TDs (coloured red in Fig. 3b) that project inwards towards the membrane. In native CCVs, this would be the site where the TDs would meet the 'string-like' unfolded regions of clathrin adaptors that extend out from the membrane and contain clathrin-binding motifs. The underside of a clathrin coat therefore functions as a planar multivalent scaffold for the attachment of clathrin to the membrane through numerous clathrin adaptors. The molecular details of the interactions between TDs and the clathrin-box and W-box peptide motifs that are found in the extended regions of clathrin adaptors have been determined by protein crystallography and confirmed by binding studies using structure-directed mutagenesis (Fig. 3c).

Recently, an entirely novel function has been proposed³³ for clathrin in chromosome separation during cell division. This was suggested on the basis of an unequal partitioning of chromosomes during mitosis that was caused by the siRNA-mediated depletion of clathrin, as confirmed by immunofluorescence and immunoEM imaging of clathrin that was associated with mitotic spindles. However, as there are no biochemical data to indicate exactly how clathrin functions in this new role, the clathrin structures described here cannot shed light on the molecular mechanisms involved in this process.

Adaptor protein complexes (APs)

The history of the clathrin adaptor complexes can be traced back almost 30 years to the discovery of 'assembly polypeptides' or APs in extracts of clathrin vesicles that promoted the assembly of purified clathrin triskelia into clathrin coats³⁴. Although purified clathrin triskelia could be assembled into clathrin coats in the absence of these assembly polypeptides at low pH, the polypeptides were absolutely required under physiological conditions. Further characterization showed that the assembly polypeptides mainly consisted of two distinct heterotetrameric complexes — AP1 and AP2 — and a third protein that is now generally referred to as AP180 (Ref. 35). The first structural characterization of assembly polypeptides was by deep-etch EM¹⁰ and, in agreement with proteolysis studies^36,37, indicated that APs are composed of a central brick-like core that is joined to two smaller globular appendages by flexible linkers. The location of the assembly polypeptides between the outer clathrin layer and the inner membrane layer of CCVs was established by observing the differences between EM cage reconstructions that were made in the presence and absence of APs^8,9 and through the inspection of clathrin-stripped coats using quick-freeze, deep-etch EM¹⁰. Linking clathrin to the membrane — that is, functioning as an adaptor between the two — is the defining function of the APs, so much so that 'AP' has come to mean 'adaptor protein' rather than the original 'assembly polypeptide'.

Organization of adaptor protein complexes (APs). APs are heterotetrameric proteins of which there are four (AP1–4). All APs have two large subunits of 100–130 kDa (α and β2 in AP2; the α-subunit is homologous to the γ-, δ- and ɛ-subunits in AP1, AP3 and AP4, respectively; β2 is homologous to β1, β3 and β4 in AP1, AP3 and AP4, respectively). They also contain a medium-sized subunit of ∼50 kDa (μ1–4 in AP1–4, respectively) and a small subunit of ∼20 kDa (σ1–4 in AP1–4, respectively). In the case of AP1 and AP2, the β-, μ- and σ-subunits share a high degree of homology (∼40–85% sequence identity), whereas α and γ are more diverse (∼25% identity). Many studies have shown that AP2 is the predominant clathrin adaptor at the plasma membrane, whereas, by immunoEM and immunofluorescent microscopy, AP1, -3 and -4 have been shown to localize both to the endosomes and the TGN³⁸. And AP3 might, in fact, be able to function in a clathrin-independent manner in some circumstances³⁹. All four APs bind directly to both YXXΦ motifs (in which X represents any amino acid and Φ represents a large hydrophobic residue) and [D/E]XXXL[L/I] 'acidic dileucine' motifs, at least one of which is contained in many CCV cargo proteins. The heterotetrameric arrangement of the APs is mirrored by that of the β, γ, δ and ζ subcomplex of heptameric coatomer protein-I (COPI). COPI forms a coat around vesicles that mediate both retrograde transport from the Golgi back to the endoplasmic reticulum and intra-Golgi transport (Box 2).

Core structure. In recent years, the compact molecular architecture that is common to the AP1 and AP2 cores — the N-terminal two thirds (trunk domains) of the large subunits together with a medium and a small subunit — has been described by protein X-ray crystallography^40,41 (Box 2). The trunk domains (γ and β1 for AP1, and α and β2 for AP2) share a common superhelical fold that includes a tight ∼90° turn, so that each domain provides two sides of a rectangle. These rectangles correspond to the brick-like structure that is seen by EM (Fig. 4a). Even more striking is the similarity in the folds of the N-terminal domains of μ1 (AP1) and μ2 (AP2) (collectively known as N-μ1/2) and the small subunits σ1 (AP1) and σ2 (AP2) (σ1/2). Both N-μ1/2 and σ1/2 fold into a central β-sheet that is flanked by helices on either side, and each domain fits into the 90° bend in β1/2 and γ/α, respectively (Fig. 4b,c). Intriguingly, this fold is found in other proteins in eukaryotes that also have functions that involve membranes (Box 3).

Figure 4: AP1 and AP2 at high resolution.

a | An original electron-microscopy image of an adaptor protein complex (AP). Modified with permission from Ref. 10 © (1988) The Rockefeller University Press. b,c | Models of AP1 and AP2 (coloured by subunit as indicated) were made using structures that were determined by X-ray crystallography (Protein Data Bank (PDB) accession files 1W63, 1GYU, 1GW5, 1B9K and 1E42). The linkers and the β1-appendage of AP1 (shaded grey) were modelled, because their structures are unknown. The position of a phosphoinositide mimic that was bound to AP2 is shown in spacefill representation. These AP1 and AP2 structures are in an inactive, closed state, because their μ-subunit YXXΦ-binding sites (X represents any amino acid and Φ represents a large hydrophobic residue) are blocked by their β-subunits. d | The YXXΦ-binding site on μ1 of AP1. A YXXΦ peptide is overlaid in white, and it can be seen that residues from the β1-subunit block the YXXΦ-binding site in the inactive, closed state of AP1. e | The YXXΦ-binding site on μ2 of AP2, again with a YXXΦ peptide overlaid in white and similarly blocked by residues from β2. f | Details of the binding of a YXXΦ peptide to the isolated C-terminal subdomain of μ2. Some of the residues that are important for binding are indicated. Parts d, e and f of this figure were created using coordinates from the PDB accession files 1W63, 1BXX and 1GW5.

Core structure reveals an inactive conformation. The C-terminal domains of the μ-subunits, collectively known as C-μ1/2, have elongated and banana-shaped β-sheet structures that sit on one face of the core structures and complement a shallow depression that is created by the other four subunits in each AP. The contact regions between C-μ1/2 and the β-trunks highlight important implications for function. The YXXΦ-binding sites on the surface of C-μ1/2 comprise hydrophobic pockets for each of the two essential bulky residues of this motif⁴² and a free β-strand that can form hydrogen bonds with the backbone of the motif. However, in the core structures, these sites are blocked by residues from the corresponding β-subunit. A valine (Val365) sits perfectly in the Φ pocket, while the side chain of Tyr405 blocks the space where the peptide backbone residues of the YX motif would be located (Fig. 4d–f). In other words, these AP1 and AP2 core structures are in an inactive, closed state that necessitates a change in conformation to expose the YXXΦ-binding site in C-μ1/2.

Significantly, the core structure indicates that the linker between N-μ1/2 and C-μ1/2, which contains the main site for phosphorylation (μ1Thr154/μ2Thr156)⁴³ that is essential for the endocytosis of tranferrin⁴⁴, is surface exposed and flexible. This is consistent with it being a target for protein kinases. In the case of AP2, the kinase responsible is the recently identified α-adaptin-associated kinase-1 (AAK1)⁴⁵. Several lines of evidence indicate that AAK1 is an important regulator of AP function, possibly by promoting the conformational change in AP2 that is needed for it to bind to endocytic cargo. AAK1 colocalizes to endocytic clathrin-coated pits in vivo⁴⁵ through its interaction with the α-appendage, and its kinase activity is stimulated by clathrin^46,47. Furthermore, phosphorylation of μ2Thr156 enhances the affinity of AP2 for endocytic signals^48,49,50, which is consistent with a role for AAK1 phosphorylation in promoting a conformational change in AP2, such that the binding site for endocytic cargo becomes exposed.

AP1 binding to YXXΦ is also accelerated by the phosphorylation of μ1 (Ref. 51) and a candidate for the responsible kinase is GAK (cyclin-G-associated kinase, which is also known as auxilin-2). GAK is a serine/threonine kinase that is related to AAK1 (Refs 52, 53). The precise molecular nature of the conformational change that is required for binding to YXXΦ is unclear, but it requires that the μ-subunits are at least partially dislodged from their binding sites, so that they can 'scan' the nearby membrane for motif-containing cargo.

Phosphoinositide binding and membrane recruitment. The AP core structures also revealed the mechanism of AP recruitment to membranes through the recognition of phosphoinositides. Phosphoinositides are important determinants of membrane identity⁵⁴ and, as such, are important signals for recruiting APs to different membrane compartments. As a result, the control of specific phosphoinositide levels by a wide range of phosphoinositide-headgroup kinases and phosphatases is an important temporal and spatial switch for dictating the affinity of the different membranes of the endocytic pathway for various APs (and other clathrin adaptors)^55,56,57, and therefore for dictating the vesicle-trafficking events that take place at these membranes. For example, AP2 is targeted to the plasma membrane mainly by an α-subunit-mediated interaction with phosphatidylinositol-4,5-bisphosphate (PtdIns(4,5)P₂) or phosphatidylinositol-3,4,5-trisphosphate (PtdIns(3,4,5)P₃)^58,59, which are phosphoinositides that are enriched in regions of the plasma membrane. Therefore the enzymes that generate them — PtdIns4-position and PtdIns5-position kinases that generate PtdIns(4,5)P₂, and the clathrin-activated phosphatidylinositol 3-kinase C2α (Ref. 60) that generates PtdIns(3,4,5)P₃ from PtdIns(4,5)P₂ — will function to recruit AP2 to nascent plasma-membrane CCV coats and thereby drive CCV formation. Conversely, synaptojanin, which hydrolyses the phosphates on PtdIns(4,5)P₂ headgroups, will promote the dissociation of AP2 from the membrane and, consequently, the disassembly or 'death' of a CCV.

The structure of AP2 revealed the exact binding site on the α-subunit for PtdIns(4,5)P₂ (actually, for a phosphoinositide-headgroup mimic, D-myo-inositol-1,2,3,4,5,6-hexakisphosphate; InsP₆) (Fig. 4c). This site is at the N terminus of the α-subunit and is formed by mainly basic residues. The mode of binding is such that two phosphates of InsP₆ are supported by the tips of interacting side chains. Mutation of this site prevents the recruitment of AP2 to PtdIns(4,5)P₂-containing membranes, even if these membranes contain YXXΦ-containing cargo⁵⁰. Unexpectedly, the structure of AP2 revealed a second distinct binding site for InsP₆ (Fig. 4c). A basic patch on the surface of C-μ2 was shown to interact with two phosphates of InsP₆ through a number of lysine residues. This second site has now also been shown to be important for localizing AP2 to clathrin-coated pits in vivo⁶¹ and in helping to drive the conformational change that allows both the PtdIns(4,5)P₂-binding and the YXXΦ-binding sites to engage their ligands simultaneously^40,41,50.

AP1 recruitment to Golgi membranes also involves phosphoinositides, but, in this case, the supposedly enriched Golgi-membrane marker phosphatidylinositol-4-phosphate (PtdIns4P)^54,62 is the probable target, although other data indicate that AP1 binds most strongly to PtdIns(4,5)P₂ (Ref. 63). AP1 also binds to the GTP-bound form of ARF1, a member of the ADP-ribosylation factor (ARF) family of small GTPases that are found on many internal membranes throughout the cell⁶⁴. Data indicate that both interactions are equally important, which indicates a bipartite recruitment mechanism.

A comparison of the structures of AP1 in the absence of phosphoinositides and AP2 in the presence of a phosphoinositide headgroup reveals a potential phosphoinositide-binding site in the γ-subunit of AP1 that is similar to the one in the α-subunit of AP2. Importantly, this site has a lower concentration of basic charge with fewer conserved basic residues than AP2. This is consistent with a preference for PtdIns4P, which has fewer phosphates than PtdIns(4,5)P₂ or PtdIns(3,4,5)P₃ (Ref. 41). Mutation of a crucial arginine residue in this region, which disturbs the usual perinuclear targeting of AP1, has been proposed to function by reducing the specificity of AP1 for PtdIns4P⁴¹. Furthermore, the PtdIns4P pool at the Golgi membranes is dependent on Golgi-resident phosphatidylinositol 4-kinase-IIα. Using siRNA against this kinase reduced the levels of PtdIns4P at the TGN and caused AP1 to become cytosolic, an effect that could be reversed by the addition of PtdIns4P⁶⁵.

The activation of ARF1 involves the exchange of GDP for GTP, which is catalysed by guanine nucleotide-exchange factors (GEFs). These, in turn, cause the covalently bound myristoyl group of ARF1 to become exposed and able to insert into membranes. By analogy with other small G-protein/effector systems, it is expected that, as a result of conformational changes that are transmitted through the switch-1 and -2 regions of ARF1, the GTP-bound form can probably bind to the large subunits of AP1 (Refs 66, 67), which causes the complex to be recruited onto membranes. Whether this model is correct will only be established when the structure of an AP1–ARF•GTP complex is determined. Inhibition of ARF activation by the GEF inhibitor brefeldin A results in AP1 dissociation from the membrane fraction^68,69. It should be noted that the membrane localization of AP2 is not brefeldin-A sensitive, which rules out ARF1–5 as potential direct membrane co-recruitment factors along with PtdIns(4,5)P₂ and/or PtdIns(3,4,5)P₃. However, ARF6 remains a possible candidate⁷⁰ because it is mostly localized to the plasma membrane and its GEFs are not sensitive to brefeldin A, although it could also function in an indirect manner through phosphoinositide metabolism^71,72.

A final uncertainty in AP1 function is where exactly in the cell AP1 is found (the TGN, endosomes or a TGN–endosome intermediate compartment) and in what pathway(s) it consequently functions (reviewed in Ref. 73). The solution to this problem might come from using structural/biophysical studies to define the true phosphoinositide-binding partner of AP1, because phosphoinositides are mainly localized to a single organelle or organelle subdomain⁵⁴.

Appendage domains. Structural and biochemical studies have localized the binding sites of a plethora of accessory/regulatory proteins and CLASPs to the appendage domains of APs (see, for example, Refs 3, 5, 74). Appendage-binding partners display different arrangements of short linear binding motifs that are found in different numbers and combinations in the unfolded regions that link their small folded domains³². High-resolution structures of both AP2 appendages (α and β2)^{74,75,76,77,78} and the γ-appendage of AP1^79,80 have been determined (Figs 4,5). The β1/2 (β1 and β2 have ∼80% sequence identity) and α-appendages all share a similar bilobal structure — an N-terminal eight- or nine-stranded β-sandwich subdomain and a C-terminal 'platform' subdomain. The structure of the platform subdomain includes a five- or six-stranded β-sheet that is flanked by two helices on one side and a third on the opposite side. By contrast, the γ-appendage has a structure similar only to that of the N-terminal β-sandwich subdomain of the other appendages.

Figure 5: AP complexes: binding sites in the appendages and the motifs bound.

a | A DPF/DPW motif in the Eps15 peptide GSDPFK is shown in complex with the α-appendage of adaptor protein complex-2 (AP2) (Protein Data Bank (PDB) accession file 1KYF). b | An FXDXF motif in the amphiphysin peptide SFFEDNFVPE is shown in complex with the α-appendage of AP2 (PDB accession file 1KY7). c | A WXXF motif in the synaptojanin peptide NPKGWVTFEEEE is shown in complex with the α-appendage of AP2 (PDB accession file 1W80). d | The structure of the β2-appendage of AP2 is shown (PDB accession file 1E42), with a question mark highlighting the place where an ∼16-residue peptide sequence that is found in certain clathrin-associated sorting proteins (CLASPs) is thought to bind. e | For ΨXXΨ motif binding to the AP1 γ-appendage or to the GGA (Golgi-localized, γ-ear-containing, ADP-ribosylation factor (ARF)-binding) appendages, the GGA1 appendage is shown in complex with the p56 peptide DDDDFGGFEAAETFD (PDB accession file 1OM9). The affinities of α-appendage binding to DPF, FXDXF (L. M. Traub, personal communication) and WXXF motifs, β2-appendage binding to a peptide of the CLASP ARH (autosomal recessive hypercholesterolaemia protein), and γ/GGA-appendage binding to ΨXXΨ motifs are shown beneath the structures⁸³. In the motifs, X represents any amino acid and Ψ represents an aromatic amino acid.

Structures of the different appendages in complex with the various peptide motifs they bind are depicted in Fig. 5. The molecular details provided by these structures explain how the specificity of the different sites on the various appendages for different motifs is achieved. Most of the interactions are mediated by mainly hydrophobic residues that are separated by a set number of other residues and that fit into chemically compatible pockets on the appendage surfaces. Further strength/specificity comes from hydrogen-bonding interactions between the peptide backbone of the motif and the appendage surface. The interaction of a single motif is relatively weak (K_d in the 10–100-mM range) and consequently will have a short (<10 s) half-life. The number of copies of a motif for a given site on an adaptor appendage domain will therefore determine the apparent strength of interaction between an accessory/regulatory protein and that appendage site.

The AP2 α-appendage engages DPF/DPW and FXDXF motifs through an overlapping site in the platform subdomain⁷⁷, whereas WXXF motifs interact through the sandwich subdomain^74,78 (Fig. 5a–c). The affinity of a protein for a given appendage must also take into account the fact that some proteins, such as synaptojanin, possess both an FXDXF and a WXXF motif and can therefore bind to both subdomain sites simultaneously. This will increase the apparent strength of binding by an avidity or chelate effect^74,81. The β-appendages can probably engage DX[FW] motifs through their platform subdomains, because, in structure, they resemble the platform domain of the α-appendage that binds DX[FW] and FXDXF motifs^77,82. However, in β-appendages, this site participates in the relatively tight binding (K_d 1–2 mM) of an ∼16-residue peptide sequence that is found in the CLASPs ARH and the arrestins⁸³ (Fig. 5d). So, if DX[FW] motifs can bind to this site on β-appendages (the expected K_d is 100 μM; Ref. 75), it is probable that they would be competed out by CLASP sequences, which would ensure that CLASPs are preferentially included in forming CCVs.

The AP1 γ-appendage and the highly homologous GGA (Golgi-localized, γ-ear-containing, ARF-binding) appendages bind to ΨXXΨ motifs⁸⁴ (in which Ψ represents an aromatic amino acid), but only the molecular details of the GGA interaction and its ligand are known at the structural level^85,86 (Fig. 5e). A question of interest, therefore, is how do two highly related ΨXXΨ motifs in NECAP (adaptin-ear-binding coat-associated protein) — the sequences WVQF and WGDF — show such remarkably high specificity for opposite faces of the sandwich subdomains of the α- and γ-appendages, respectively⁷⁸. Acidic residues that lie downstream of WVQF and upstream of WGDF might provide further specificity, so it might be that it is more complicated to predict the affinity of a single motif for a single appendage site than was originally thought.

Finally, it should be remembered that many appendage-binding proteins also bind to other components of CCVs, including other accessory/regulatory proteins, clathrin, phosphoinositides, dynamin and cytoskeletal elements, so they need to be considered as part of the dynamic networks of interacting proteins. So, there are many complex factors that govern the steady-state levels and the temporal regulation of accessory/regulatory-protein and CLASP recruitment to the various stages of the life of a CCV. The level of any given accessory/regulatory protein in any given CCV cannot be accurately predicted through motif hunting or 'sequence gazing', and so a systems-biology approach might be required to enable us to understand this problem.

Future directions

The recent high-resolution reconstructions of clathrin cages provide a way to understand clathrin function at an atomic level, and the goal of obtaining higher resolution structures of assembled clathrin bound to many other adaptor proteins now seems more achievable. In order to extend our knowledge of clathrin-coat 'death' to the same level as our understanding of its 'birth' and 'life', it is clear that a sub-15-Å map of HSC70 bound to clathrin cages is urgently required. However, the way is now paved for informed biochemical and cell-biological studies to test the possible mechanisms of CCV uncoating and to continue to investigate CLC function. However, in order to truly understand the mechanisms by which clathrin achieves its dynamic assembly and, in particular, disassembly, it will be necessary to determine the structure of clathrin cages in as many of its different conformational states bound to as many different partners as possible.

The situation with regard to the APs is similar, in that all the evidence that relates to the function of APs as cargo adaptors in CCV formation indicates a molecular mechanism that involves conformational changes. At present, the exact nature of these changes can only be proposed. A concerted effort is needed to determine the structure of APs in a number of different states bound to various ligands, so that we gain a series of snapshots of different stages of AP function. These would, in turn, indicate the biochemical/biophysical experiments that need to be carried out to confirm the functional models that are proposed on the basis of new structures. Will the current structural techniques, with their various technical limitations and the fact that they provide only a series of 'static snapshots', be able to deliver accurate residue-level-resolution functional models? Or will new time-resolved and single-molecule structure-determination techniques be needed? We can only wait and ultimately, we hope, see.

Box 1 | The lifecycle of a clathrin coat

Clathrin-coated vesicles, like other transport intermediates, originate from a patch of membrane that aquires a protein coat, the components of which are recruited from the cytosol. This coat selects cargo for incorporation into the vesicle and helps its mechanical formation. A clathrin-coated pit forms, and the vesicle then buds from its parent membrane, the protein coat dissociates and the vesicle is transported to, tethered to, and finally fuses with, its target membrane.

Clathrin forms the mechanical scaffold part of the coat but is unable to interact directly with any components of the membrane. The two are linked by many different clathrin adaptors, which bind to clathrin and membrane phospholipids and/or protein cargo. The adaptors are recruited by interactions with short-lived membrane species such as phosphoinositides and small G-proteins of the ARF (ADP-ribosylation factor) family. The adaptors, in turn, recruit clathrin trimers, which assemble into a lattice that can then recruit more clathrin adaptors to the edges of the forming coat, and so drive and accelerate coat formation. The formation of the coat is a continuous rather than stepwise process, and a clathrin coat should not be considered as a static structure but as a dynamic network in which there is a constant exchange of interacting partners. Following vesicle scission, the intrinsic GTPase activity of ARFs and the action of phosphoinositide-headgroup-hydrolysing enzymes weakens the affinity of adaptors for membranes, while the chaperone HSC70 (70-kDa heat-shock cognate protein), through auxilin-family members, disrupts the clathrin scaffold by removing individual triskelia. The balance of assembly and disassembly is now in favour of disassembly and the coat falls apart completely with its components returning to the cytosol. However, the components might actually never be far from the membrane owing to the possible interactions they can have with the moieties that are left behind on the parent membrane, ready for reuse in the 'birth' of the next clathrin coat. Clathrin-coat formation/disassembly and composition is also regulated by cycles of phosphorylation–dephosphorylation of many of its components (see figure). AP, adaptor protein complex; CLASPs, clathrin-associated sorting proteins; Pi, inorganic phosphate; PtdIns(4,5)P₂, phosphatidylinositol-4,5-bisphosphate.

Box 2 | COPI and the APs

The family of heterotetrameric adaptor protein complexes (APs) includes the clathrin adaptors AP1 and AP2, as well as AP3 and AP4. This family also includes part of the coatomer protein-I (COPI) coat. COPI-coated vesicles mediate both retrograde transport from the Golgi to the endoplasmic reticulum and intra-Golgi transport. The cytosolic precursor of the COPI coat, which is a heptameric coatomer complex, can be thought of as consisting of two subcomplexes. The first consists of the β-, γ-, δ- and ζ-subunits, which were proposed to be structurally homologous to the heterotetrameric AP subunits⁸⁷ and to have originated from a common ancestor⁸⁸. These relationships were confirmed by the recent determination of the structure of the C-terminal 265 amino acids of γ-COP, which showed that it has the same bilobed structure as both the α- and the β2-appendages of AP2 and that it is similarly connected to the rest of the protein by an unstructured linker^89,90 (see figure). Furthermore, the relationship between COPI and the APs extends from structure to function, because the γ-COP appendage binds to a COPI-coated-vesicle accessory protein — an ARFGAP (ADP-ribosylation factor GTPase-activating protein) — in a similar manner and on a site that is analogous to the platform site in the α- and β2-appendages that recruits many accessory proteins for clathrin-coated-vesicle formation^90,91.

The relationship between the COPI and clathrin–AP systems might extend beyond the fact that the β, γ, δ and ζ subcomplex of COPI resembles the APs, because the α, β′ and e COPI subcomplex, similar to clathrin itself, is believed to contain β-propeller domains and stacked helical hairpins⁹² (see figure). Indeed, the presence of β-propellers and stacked helical hairpin domains in the components of various transport-vesicle coats and of nuclear pore complexes might indicate that these components are ultimately derived from a simple membrane-coating/bending module that contained minimal copies of these two types of domain, was present in 'protoeukaryotes', and was responsible for the formation of the internal membrane systems of modern eukaryotes⁹³.

Box 3 | Conserved σ2 folds in other proteins

The σ-subunits and the N terminus of the μ-subunits of the adaptor protein complexes (APs) AP1 and AP2 fold into a five-stranded β-sheet that is sandwiched by helices on either side (see figure; σ2/N-μ2). This fold is also found in the N-terminal longin domain of the membrane-fusion SNARE proteins (soluble N-ethylmaleimide-sensitive fusion protein (NSF) attachment protein (SNAP) receptor proteins) Sec22b (Ref. 94), Ykt6 (Ref. 95) and Vamp7 (see figure, Sec22b). This raises the intriguing possibility of a link between proteins that are involved in the formation of a vesicle coat and those that are involved in the fusion of a vesicle with its target membrane.

More recently, the same σ/N-μ2 fold has been found in the SR (signal-recognition-particle receptor)-β-interacting SRX domain of SRα (Ref. 96; see figure, SRα–SRβ•GTP). SRα and SRβ together form the SR in eukaryotes, which is involved in targeting nascent polypeptides that are emerging from ribosomes to the channels that are responsible for inserting transmembrane proteins into membranes and transporting secreted and lumenal proteins into the endoplasmic reticulum. It is interesting to note that SRβ is a member of the small Ras-like GTPase superfamily, which includes the ARFs (ADP-ribosylation factors) and ARLs (ARF-like proteins) that are involved in the temporal and spatial regulation of various stages of the lifecycle of vesicles (see figure, ARF1•GTP). This raises the possibility that the σ/N-μ2/longin/SRα structure might be a general small-GTPase-binding structure, and therefore that interactions between these two types of structure are the mechanical link between the structural components of vesicles (coats and SNAREs) and their regulatory machinery (ARFs and ARLs) — that is, that the σ/N-μ2/longin/SRα fold is a small-GTPase effector domain.

The structures shown in the figure were created using the Protein Data Bank accession files 1GW5, 1IFQ, 1NRJ and 1J2J, and GTP is shown using a spacefill representation.