Cargo Recognition in Clathrin-Mediated Endocytosis

The best-characterized endosomal sorting process is the internalization of select transmembrane proteins in clathrin-coated vesicles. The process is triggered by the recognition of signals in their cytosolic domains.

Abstract

The endosomal system is expansive and complex, characterized by swift morphological transitions, dynamic remodeling of membrane constituents, and intracellular positioning changes. To properly navigate this ever-altering membrane labyrinth, transmembrane protein cargoes typically require specific sorting signals that are decoded by components of protein coats. The best-characterized sorting process within the endosomal system is the rapid internalization of select transmembrane proteins within clathrin-coated vesicles. Endocytic signals consist of linear motifs, conformational determinants, or covalent modifications in the cytosolic domains of transmembrane cargo. These signals are interpreted by a diverse set of clathrin-associated sorting proteins (CLASPs) that translocate from the cytosol to the inner face of the plasma membrane. Signal recognition by CLASPs is highly cooperative, involving additional interactions with phospholipids, Arf GTPases, other CLASPs, and clathrin, and is regulated by large conformational changes and covalent modifications. Related sorting events occur at other endosomal sorting stations.

The internalization of a subset of plasma membrane proteins by clathrin-mediated endocytosis is one the best-characterized sorting processes that takes place in the endomembrane system of eukaryotic cells (Kirchhausen 2014). Selection of transmembrane proteins (referred to as “cargo”) for internalization by clathrin-mediated endocytosis involves recognition of endocytic signals in the cytosolic domains of the proteins by adaptors located in the inner layer of clathrin coats. Signal–adaptor interactions lead to concentration of the transmembrane proteins within clathrin-coated pits that eventually bud into the cytoplasm as clathrin-coated vesicles (Kirchhausen 2014). Transmembrane proteins that have endocytic signals are thus rapidly delivered to endosomes, whereas those that lack signals remain at the plasma membrane. This article summarizes recent progress in the elucidation of the mechanisms of signal recognition in clathrin-mediated endocytosis, with additional reference to related intracellular sorting events. Further information on this topic can be found in previous reviews (Bonifacino and Traub 2003; Traub 2009; Kelly and Owen 2011).

DIVERSITY OF ENDOCYTIC SIGNALS

The notion that endocytic receptors possess internalization signals was initially derived from morphologic, biochemical, and genetic studies of lipoprotein receptors, well before there was any knowledge of receptor structure (Brown and Goldstein 1979). With the advent of gene cloning, the amino acid sequences of various endocytic receptors became known in steady succession. Initial sequence comparisons, however, did not reveal any conserved groups of amino acids that could constitute a common endocytic signal. The identification of endocytic signals would end up requiring extensive molecular dissection of the receptor sequences using a combination of mutational and functional analyses. This effort led to the current understanding of endocytic signals as a highly diverse set of structural features in the cytosolic domains of transmembrane proteins, which can be grouped into three functionally analogous but structurally distinct classes: (1) linear motifs, (2) conformational determinants, and (3) covalent modifications.

Linear motifs are short arrays of invariant and variant amino acids, including “tyrosine-based” YXXØ (Collawn et al. 1990; Jadot et al. 1992) and [FY]XNPX[YF] motifs (Chen et al. 1990; Collawn et al. 1991), and “dileucine-based” [DE]XXXL[LI] motifs (Table 1) (Letourneur and Klausner 1992; Pond et al. 1995). In this notation, amino acids are represented in single-letter code, X indicates any amino acid, Ø indicates an amino acid with a bulky hydrophobic side chain, and the brackets mean that either amino acid is allowed at that position. The invariant amino acids are the most critical elements, although the variant amino acids influence the strength and fine specificity of the signals. The function of the signals can also be affected by flanking sequences (Ohno et al. 1998), phosphorylation of amino acids at or near the motif (Shiratori et al. 1997; Pitcher et al. 1999), their spacing from the transmembrane domain (Rohrer et al. 1996), and palmitoylation of nearby cysteine residues (Schweizer et al. 1996). Linear motifs that function as signals are generally found within unstructured regions of the cytosolic domains.

Table 1.

Endocytic signals and adaptors

Signals or domains	Adaptors	Adaptor subunits or domains	References
YXXØ	AP-2	µ2	Collawn et al. 1990; Jadot et al. 1992; Ohno et al. 1995; Owen and Evans 1998
[DE]XXXL[LI]	AP-2	α-σ2	Letourneur and Klausner 1992; Pond et al. 1995; Chaudhuri et al. 2007; Doray et al. 2007; Kelly et al. 2008
Acidic clusters	AP-2?	α?	Voorhees et al. 1995; Lindwasser et al. 2008
[YF]XNPX[YF]	ARH; Dab2; Idol; SNX17, 27, and 31	PTB domain	Chen et al. 1990; Collawn et al. 1991; He et al. 2002; Mishra et al. 2002; Stockinger et al. 2002; Zelcer et al. 2009; Ghai et al. 2013
NPFX(1,2)D	Sla1p	SLA1 homology domain	Tan et al. 1996; Howard et al. 2002
Ubiquitin	Eps15, Epsins 1 and 2	UIM domain	Polo et al. 2002; Shih et al. 2002
GPCR phosphorylation	β-arrestins 1 and 2	Amino terminus	Ferguson et al. 1996; Goodman et al. 1996
Synaptotagmin I C2A (C2B) domain	Stonin 2	μHD domain	Martina et al. 2001; Walther et al. 2001
Mid2p cytosolic domain	Syp1p	μHD domain	Reider et al. 2009
Alk8 cytosolic domain	Fcho1	μHD domain	Umasankar et al. 2012
VAMP 7 longin domain	Hrb, AP-3	Carboxy-terminal unstructured domain	Pryor et al. 2008; Kent et al. 2012
VAMP 2, VAMP 3, VAMP 8 SNARE motifs	CALM	ANTH domain	Miller et al. 2011

ARH, autosomal recessive hypercholesterolemia; PTB, phosphotyrosine-binding; UIM, ubiquitin-interacting motifs; GPCR, G-protein-coupled receptor; CALM, clathrin assembly lymphoid myeloid leukemia.

Not all signals, however, are linear sequences or fit a canonical motif. There are now many examples of folded domains that contain information for endocytosis (Table 1) (Pryor et al. 2008; Yu et al. 2010; Miller et al. 2011). This information consists of conformational arrays of amino acids on the surface of the folded domains. Unlike linear motifs, which are common to many proteins, each conformational array described to date appears to be unique for a specific cargo. Finally, covalent modifications such as phosphorylation of hydroxyl amino acids (Ferguson et al. 1996; Di Fiore and von Zastrow 2014) or polyubiquitination on the ε-amino group of lysine residues (Hicke and Riezman 1996; Piper et al. 2014) in the cytosolic domains can also function as endocytic signals (Table 1). In these cases, the modifying groups do not modulate the activity of underlying linear or conformational signals, but themselves act as recognition determinants. Multiple or overlapping signals can occur within the same cytosolic domain (for example, see Johnson and Kornfeld 1992; Doray et al. 2008; Goh et al. 2010; Prabhu et al. 2012). Similar types of signal participate in sorting events that take place at intracellular compartments, such as the trans-Golgi network (TGN) and endosomes.

MULTIPLICITY OF ADAPTORS

The recognition of such a wide diversity of endocytic and intracellular sorting signals obviously necessitates the existence of multiple adaptors. Indeed, many proteins located in the inner layer of protein coats—including proteins that were initially categorized as “accessory”—are now known to function as sorting adaptors (Fig. 1; Table 1). Depending on the identity of the scaffolding protein that forms the outer layer, coats are classified as clathrin coats or nonclathrin coats. Coats involved in rapid internalization from the plasma membrane contain clathrin as their main constituent and a set of adaptors known as “clathrin-associated sorting proteins” (CLASPs) (Fig. 1A). Clathrin coats containing different sets of CLASPs, as well as nonclathrin coats, mediate intracellular sorting events. CLASPs are recruited to membranes primarily via interactions with specific phosphoinositide lipids, small GTPases of the Arf family, and/or other CLASPs. Clathrin then binds to the CLASP armature and polymerizes into an overlying polyhedral scaffold. Concomitantly, CLASPs engage sorting signals in the cytosolic domains of transmembrane cargo, leading to cargo capture and stabilization of the coats. Both CLASP–clathrin (Dell'Angelica et al. 1998; Drake and Traub 2001) and CLASP–CLASP interactions (Brett et al. 2002) involve linear motifs (analogous to, but distinct from, cargo sorting signals) binding to folded domains, highlighting the general role of this binding mode in the assembly and function of clathrin coats. Most interactions among components of clathrin coats are of moderate to low affinity (typically in the 1–100 µM range), making this mechanism of sorting a highly cooperative and dynamic process.

Figure 1.

The endocytic cargo–adaptor interaction network. (A) Schematic representation of selected sorting signal-recognition partner relationships for endocytic trafficking. Both protein–protein and protein–lipid (PtdInsP) interactions among the cargo-selective machinery are highlighted with connection lines. The solid lines indicate documented physical interactions, whereas dashed lines connote either interactions possible based on properties of other domain relatives (IDOL) or known (AP-1), but still of unclear functional necessity. Adaptors, CLASPs, and regulators (Idol) discussed explicitly are double circled. (B) Representative modular domain architecture classes of selected endocytic proteins. Heterotetrameric examples are AP-1, AP-2, and AP-3; dimeric examples are Fcho1, Syp1p, and eps15; and monomeric CLASPs include ARH, Dab2, Numb, CALM, AP180, epsin, and β-arrestin. Tertiary-structured domains are indicated by geometric shapes, and intrinsically disordered protein segments by a line.

From a structural standpoint, CLASPs can be categorized as (1) oligomeric (tetrameric or dimeric), and (2) monomeric (Fig. 1B). The main endocytic adaptor is the clathrin-associated, heterotetrameric adaptor protein 2 (AP-2) complex. This complex is composed of two large “adaptin” subunits (α and β2), one medium-sized subunit (μ2), and one small subunit (σ2). AP-2 is a member of a family of homologous complexes that also includes AP-1 (γ-β1-μ1-σ1), AP-3 (δ-β3-μ3-σ3), AP-4 (ε-β4-μ4-σ4), AP-5 (ζ-β5-μ5-σ5), and COPI-F (γ-COP-β-COP-δ-COP-ζ-COP) (corresponding subunit composition in parentheses) (Robinson 2004; Hirst et al. 2011). All of these complexes are components of clathrin or nonclathrin coats that mediate sorting in intracellular compartments. They comprise a large globular “core” consisting of the amino-terminal “trunk” domains of the large subunits plus the entire medium and small subunits. The carboxy-terminal portions of the large subunits extend from the core as two long projections, each comprising a long disordered “hinge” sequence and a globular “ear” or “appendage” domain. The core mediates recruitment to membranes and sorting-signal recognition, whereas the hinge-appendage extensions interact with clathrin, other adaptors, and various accessory proteins. Dimeric and monomeric CLASPs and related adaptors consist of a single polypeptide chain in which the same functions are distributed among several globular domains joined by disordered linkers, giving them a “beads-on-a-string” appearance.

RECOGNITION OF LINEAR MOTIFS BY AP-2

AP-2 recognizes YXXØ and [DE]XXXL[LI] signals through binding to two distinct sites on the core domain. The binding site for YXXØ motifs is located on the carboxy-terminal domain of μ2 (μ2-C) (Ohno et al. 1995), an immunoglobulin-like β-sandwich fold having hydrophobic pockets for the Y and Ø residues (Fig. 2A) (Owen and Evans 1998). The Y pocket cannot accommodate phosphotyrosine, explaining why tyrosine phosphorylation prevents binding of YXXØ motifs to μ2 (Boll et al. 1996; Ohno et al. 1996) and endocytosis (Shiratori et al. 1997). The μ2-C domain has also been shown to bind the folded DEP domain of the signaling adaptor Disheveled at a surface near the YXXØ-binding site (Yu et al. 2010). The [DE]XXXL[LI]-binding site is on the α-σ2 hemicomplex (Chaudhuri et al. 2007; Doray et al. 2007), with the acidic residue of the motif interacting with basic residues on both α and σ2, and the two hydrophobic residues fitting into adjacent hydrophobic pockets on σ2 (Fig. 2B) (Kelly et al. 2008; Mattera et al. 2011). A phosphoserine residue at position −5 from the first leucine (considered position 0) substitutes for the acidic residue at position −4 in the noncanonical dileucine signal from CD4, exemplifying how phosphorylation can also positively regulate AP-2 recognition and endocytosis (Fig. 2D) (Pitcher et al. 1999; Kelly et al. 2008).

Figure 2.

AP-1 and AP-2 cargo-binding surfaces. (A,B) Molecular details of the YXXØ and [DE]XXXL[LI] interaction surfaces on AP-2. The TGN38-derived YQRL signal cocrystalized with the AP-2 µ2-C domain (PDB ID: 1BXX) (Owen and Evans 1998) and the CD4 dileucine signal bound to the AP-2 heterotetramer core (PDB ID: 2JKR) (Kelly et al. 2008). Amino- and carboxy-terminal ends of the peptide signals (yellow) in stick representation are shown with oxygen colored red and nitrogen colored blue. The binding pockets for key anchor residues are outlined with white dashed ovals. (C) Molecular surface of the ternary AP-1 µ1 subunit•HIV-1 Nef•MHC-1 complex that stabilizes the noncanonical YXXØ signal in MHC-1 in the absence of an effective hydrophobic Ø residue. (D) Proximity of posttranslationally phosphorylated (blue) or ubiquitinated (red) residues identified within or adjacent to different primary-sequence-based sorting signals in the indicated proteins, including the protease-activated receptor 1 (PAR1), the cation-independent mannose 6-phosphate receptor (CI-MPR), and amyloid precursor protein (APP). Sorting signals are shown in color-coded bold type and the number of amino acid residues preceding or following the individual signals from the transmembrane (Tm) domain are indicated. All modifications listed in the PhosphoSitePlus mass spectrometry database (Hornbeck et al. 2004).

Strikingly, the AP-2 core occurs in two distinct conformations: a “locked” conformation in which the binding sites for both YXXØ and [DE]XXXL[LI] signals are occluded by regions of β2 (Collins et al. 2002), and an “open” conformation in which both sites are accessible (Fig. 3A) (Jackson et al. 2010). The change from the locked to the open conformation involves translocation of the μ2-C domain to an orthogonal face and displacement of the amino terminus of β2 from the α–σ2 interface. In the open conformation, the YXXØ- and [DE]XXXL[LI]-binding sites are coplanar with four electropositive patches that bind phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P₂) (Jackson et al. 2010), a phosphoinositide that is highly enriched in the plasma membrane (Di Paolo and De Camilli 2006). This coplanarity ensures coupling of signal recognition to membrane recruitment (Jackson et al. 2010). On the basis of these observations, it has been proposed that the AP-2 core exists in equilibrium between two conformers: the locked form that predominates in the cytosol and the open form that is mainly associated with membranes. The generation of PtdIns(4,5)P₂ and availability of signal-bearing cargo at the plasma membrane shift the equilibrium toward the open form. AP-2 phosphorylation events also appear to contribute to the shift. For example, phosphorylation of a threonine residue (Thr156) in the linker that connects the amino- and carboxy-terminal domains of μ2 by the adaptor-associated kinase 1 (AAK1) enhances binding of AP-2 to endocytic signals and promotes receptor internalization (Olusanya et al. 2001; Conner and Schmid 2002; Ricotta et al. 2002). The μ2 linker is disordered in the locked form but folds into a four-turn α helix that binds to β2 in the open form (Jackson et al. 2010). Thr156 phosphorylation could thus cause stabilization of this μ2–β2 interaction and, by extension, favor the open conformation of the core. In addition, β2 residue Tyr6 undergoes phosphorylation on activation of the epidermal growth factor (EGF) receptor kinase (Huang et al. 2003). In the locked form of the AP-2 core, β2 Tyr6 occupies the σ2 hydrophobic pocket for the [LI] residue in [DE]XXXL[LI] motifs (Kelly et al. 2008). Phosphorylation of this residue could unblock the dileucine-binding site, also favoring conversion to the open form of the AP-2 core. Because internalization of the activated EGF receptor is partly dependent on a dileucine-based signal (Huang et al. 2003), this phosphorylation event illustrates how a signaling receptor can promote its own endocytosis through activation of AP-2.

Figure 3.

Allosteric-regulated exposure of the AP-1 and AP-2 cargo-binding surfaces. (A,B) Space-filling surface representations of AP-1 and AP-2 heterotetramers highlight the conformational transitions on conversion from the AP-2 “locked” state (PDB ID: 2VGL) (Collins et al. 2002) to the open state (PBD ID: 2XA7) (Jackson et al. 2010). A view of the membrane-attached face is depicted and the four subunits colored to reflect structural and functional conservation. Opening chain movements are indicated (colored arrows). Cocrystalized YXXØ and [DE]XXXL[LI] peptide signals (rendered in yellow stick representation) are delimited with white dashed circles. The relative position of the four spatially discrete PtdIns(4,5)P₂ binding sites that only become coplanar on transition to the open conformation in AP-2 are circled and numbered. Similar bottom-up views of the AP-1 heteroterameric core in the closed (PDB ID: 1W63) (Heldwein et al. 2004) and open (PDB ID: 4HMY) (Ren et al. 2013) conformations, with an Arf1 GTPase contacting each large chain trunk domain. The location of the PtdIns(4)P-binding site on the γ subunit (Heldwein et al. 2004) is circled in the open conformation. (C) A composite molecular model of the AP-1 µ1•Nef•MHC-I ternary complex (PDB ID: 4EN2) (Jia et al. 2012), rendered with the AP-2 open conformation backbone model, shown in similar orientation to A and B. Positioning of the MHC-I peptide (yellow sticks) is indicated with a white dashed oval.

RECOGNITION OF LINEAR MOTIFS BY AP-1

Studies on signal recognition by AP-1 have revealed striking similarities but also important differences in comparison with AP-2. AP-1 is a component of clathrin coats associated with the TGN and endosomes, and has been implicated in many pathways, including bidirectional transport between the TGN and endosomes (Hirst et al. 2012), export from the TGN to the plasma membrane (Guo et al. 2013), and polarized sorting from the TGN or endosomes to the basolateral domain of epithelial cells (Folsch et al. 1999; Carvajal-Gonzalez et al. 2012) and the somatodendritic domain of neurons (Dwyer et al. 2001; Margeta et al. 2009; Farias et al. 2012). AP-1 has been additionally shown to complement the role of AP-2 in endocytosis of synaptic vesicle proteins at the neuronal presynaptic terminal (Kim and Ryan 2009). The function of AP-1 in these pathways also involves recognition of linear motifs by mechanisms that are similar to those of AP-2. YXXØ and [DE]XXXL[LI] motifs bind to conserved sites on the μ1 (Ohno et al. 1995; Carvajal-Gonzalez et al. 2012; Farias et al. 2012) and γ-σ1 subunits (Janvier et al. 2003; Doray et al. 2007; Mattera et al. 2011), respectively. Moreover, the AP-1 core also occurs in equilibrium between locked (Heldwein et al. 2004) and open conformations (Fig. 3B) (Ren et al. 2013). On the other hand, the regulation of membrane recruitment and conformational activation are quite different. AP-1 binds phosphatidylinositol 4-phosphate (PtdIns(4)P), a phosphoinositide that is enriched in the TGN and endosomes, through a site on the γ subunit (Wang et al. 2003; Heldwein et al. 2004). However, this binding is too weak for efficient recruitment to membranes. Instead, the main determinant of AP-1 recruitment to membranes is binding to the GTP-bound, activated form of Arf1 (Stamnes and Rothman 1993; Traub et al. 1993). Other Arf-family members, including Arfrp1 (also known as Arl3) (Guo et al. 2013), may also play roles in AP-1 membrane recruitment. Exchange of GTP for GDP on Arf1 exposes an N-myristoylated α helix that inserts into membranes and rearranges the switch I and II regions such that they can bind AP-1. A recent study has identified two binding sites for Arf1 on the trunk domains of the γ and β1 subunits of AP-1 that contribute to recruitment of this complex to membranes (Ren et al. 2013). Strikingly, Arf1 not only promotes AP-1 membrane recruitment but also drives the conformational opening of the core that allows binding of YXXØ and [DE]XXXL[LI] motifs (Ren et al. 2013). This conformational change appears to involve pivoting on a third Arf1-binding site located on a different face of the γ trunk (Ren et al. 2013). The cooperativity of these interactions is further evidenced by the observation that both YXXØ and [DE]XXXL[LI] promote binding to each other, as well as binding of Arf1 to AP-1 (Lee et al. 2008a). The amino acids that bind Arf1 are perfectly conserved in the β2 subunit and partially conserved in the α subunit of AP-2, but the role of Arf-family members in AP-2 membrane recruitment is controversial (Krauss et al. 2003; Paleotti et al. 2005). YXXØ-, [DE]XXXL[LI]-, and Arf-binding sites are also well conserved in AP-3 (Mattera et al. 2011; Mardones et al. 2013; Ren et al. 2013), an endosome-associated adaptor involved in cargo sorting to lysosomes and lysosome-related organelles, indicating that this complex might function by a similar mechanism.

Nef: A COOPERATIVE MODIFIER OF SIGNAL–ADAPTOR INTERACTIONS

The mechanism of action of the Nef protein of human immunodeficiency virus 1 (HIV-1) exemplifies another means of regulating signal recognition by AP-1 and AP-2. Expression of Nef early in the infection cycle of HIV-1 causes down-regulation of CD4 and class I molecules of the major histocompatibility complex (MHC-I) from the surface of the host cells (i.e., T lymphocytes and macrophages) (Tokarev and Guatelli 2011). Remarkably, whereas Nef-induced CD4 down-regulation involves enhanced AP-2-dependent internalization (Jin et al. 2005; Chaudhuri et al. 2007), MHC-I down-regulation occurs by AP-1-dependent delivery of intracellular MHC-I from the TGN to endosomes (Roeth et al. 2004; Lubben et al. 2007). In both cases, the proteins are eventually transported to lysosomes for degradation. These down-regulation events involve cooperative formation of tripartite CD4•Nef•AP-2 and MHC-I•Nef•AP-1 complexes. In the CD4•Nef•AP-2 complex, a Nef carboxy-terminal flexible loop contains a [DE]XXXL[LI] motif that bind to the canonical dileucine-binding site on α-σ2 (Chaudhuri et al. 2007) and other residues that bind to a basic patch on the trunk domain of α (Chaudhuri et al. 2009). In between the dileucine and diacidic motifs there are additional hydrophobic residues that also bind to AP-2 likely via σ2 (Fig. 2B) (Jin et al. 2012). The Nef loop thus exemplifies how a long sequence containing both canonical and noncanonical elements can interact with AP-2. The dileucine motif in the CD4 tail cannot interact with the canonical dileucine-binding site on α-σ2 because the serine at position −5 from the first leucine in CD4 (Fig. 2D) is normally not phosphorylated and the dileucine-binding site on α-σ2 is occupied by the Nef loop. Instead, the dileucine motif in the CD4 tail binds to a hydrophobic patch on another region of Nef (Grzesiek et al. 1996). The cooperativity in the assembly of the tripartite complex (Chaudhuri et al. 2009) predicts the occurrence of another interaction of CD4 with AP-2 that remains to be identified. The recent resolution of the structure of the MHC-I cytosolic domain in a ternary complex with Nef and the μ1 subunit of AP-1 has revealed a different type of cooperative assembly (Jia et al. 2012). In this complex, a noncanonical tyrosine residue in the MHC-I tail occupies the binding pocket for the tyrosine residue of YXXØ signals on μ1 (Figs. 2C and 3C). The rest of the MHC-I tail binds into a groove at the Nef–μ1 interface (Jia et al. 2012). These observations raise the intriguing possibility that other cellular proteins could promote interactions of cytosolic domains with adaptors in a manner similar to Nef.

ALTERNATIVE TYROSINE-BASED SIGNAL RECOGNITION BY PTB-DOMAIN-TYPE CLASPS

Although the [FY]XNPX[YF] signal delineated in the low-density lipoprotein (LDL) receptor (FDNPVY) first showed the importance of an anchor tyrosine to cue internalization (Davis et al. 1986), this signal is actually recognized quite differently from YXXØ signals. Substitution of the carboxy-terminal tyrosine at position 0 with phenylalanine (or even tryptophan) (Davis et al. 1986) has little effect on internalization, whereas a cysteine or alanine change diminishes uptake to that of a cytosolically truncated receptor (Davis et al. 1986; Chen et al. 1990). Several transmembrane cargoes bearing [FY]XNPX[YF] signals naturally have a phenylalanine at this position (like amnionless [Pedersen et al. 2010], Opo [Bogdanovic et al. 2012], P-selectin [Ghai et al. 2013], ROMK [Fang et al. 2009], Sanpodo [Tong et al. 2010], and TAT-1 [Nilsson et al. 2011]). Instead of AP-2, these [FY]XNPX[YF] signals are decoded by monomeric CLASPs with a phosphotyrosine-binding (PTB) domain, such as Dab2, ARH (alternatively designated LDLRAP 1), and Numb (Fig. 1) (McMahon and Boucrot 2011; Reider and Wendland 2011). The modular PTB domain in these CLASPs is similar to that originally identified in signal transduction proteins, but differs in a preference for unphosphorylated over phosphorylated tyrosine. The recent resolution of the structure of the liganded ARH PTB domain (Dvir et al. 2012) explains clearly why either tyrosine or phenylalanine can be accommodated at position 0, in stark contrast to YXXØ signals, where phenylalanine cannot replace tyrosine (Fig. 4). To engage the [YF] side chain at position 0, ARH displays a largely hydrophobic-lined acceptor cavity (Dvir et al. 2012). This is dissimilar to signal transduction PTB domains, which have an overall basic character to the larger phosphotyrosine-accommodating site (Farooq and Zhou 2004; Uhlik et al. 2005). In ARH, the tyrosine hydroxyl hydrogen bonds to a single, noncovalent water molecule to permit packing into the nonpolar surface (Fig. 4), which would be unnecessary for a phenylalanine.

Figure 4.

PTB domains and μHDs in cargo signal recognition. (A) Combined ribbon and space-filling surface representation of the ARH PTB domain bound to the LDL receptor SINFDNPVYQKT sorting-signal peptide shown in stick representation (PDB ID: 3SO6) (Dvir et al. 2012). The location of a water molecule hydrogen bonded to the position 0 tyrosine is indicated by a red sphere within the marked PTB-domain tyrosine-binding site. (B) Combined ribbon and surface representation of the SNX17 FERM-like domain complexed with the TYGVFTNAAYDPT signal from P-selection (PDB ID: 4GXB) (Ghai et al. 2013). Shown in the same relative orientation as A with the positioning of the F1–F3 subdomains indicated and the YXXØ and [FY]XNPX[YF] interaction surfaces highlighted. (C–E) Combined ribbon and surface representations of the AP-2 µ2 subunit bound to the YQRL signal from TGN38 (PDB ID:1BXX) (Owen and Evans 1998), a homology model of the human Stonin 2 mHD generated by Phyre² (Kelley and Sternberg 2009), and the Saccharomyces cerevisiae Syp1p μHD (PDB ID: 3G9H) (Reider et al. 2009). The relative positioning of the YXXØ interaction surface in subdomain A on μ2, and the biochemically defined synaptotagmin 1 C2A-binding site on subdomain B of stonin 2 (Jung et al. 2007) are indicated. Three side chains on the stonin 2 µHD (Lys783, Tyr784, Glu785) implicated in C2A binding are highlighted.

A chemical distinction between tyrosine and phenylalanine at position 0 is that only the former can be posttranslationally phosphorylated. Because endocytic PTB domains cannot well accommodate the bulky, charged phosphotyrosine, this reversible modification can potentially govern sorting efficiency into nascent clathrin-coated structures, as with the YXXØ signal. There is some evidence for this for the LDL receptor-related protein 1 (LRP1), which binds to different partners depending on the phosphotyrosine status (Betts et al. 2008; Guttman et al. 2009). In Drosophila, the YTNPAF signal in the tetraspanin Sanpodo is recognized by Numb (Tong et al. 2010), whereas the FDNPVY signal in the scavenger receptor Draper is recognized by the PTB CLASP Ced-6 (Awasaki et al. 2006; Jha et al. 2102), and tyrosine phosphorylation regulates its uptake into cells (Fujita et al. 2012). Intriguingly, discovery-mode mass spectrometry indicates that Tyr807, the LDL receptor tyrosine 0, is phosphorylated in >125 separate studies (Hornbeck et al. 2004). Because the anchor Tyr20 of the transferrin receptor YXXØ (YTRF) signal is similarly phosphorylated in >1000 mass spectrometry analyses, focal tyrosine phosphorylation during signal relay from the cell surface might, through exclusion, underlie the formation of compositionally and/or kinetically distinct coated vesicles (Fig. 2D) (Cao et al. 1998; Mundell et al. 2006; Puthenveedu and von Zastrow 2006; Liu et al. 2010).

The structural role of all the other conserved side chains in the [FY]XNPX[YF] signal is also known. The leading −5 position tyrosine/phenylalanine packs into a complementary hydrophobic site generated principally by the so-called “selectivity” α2 helix, and usually involving a projecting PTB-domain aromatic side chain that bisects the interaction surfaces for the −5 tyrosine/phenylalanine and the −3 asparagine and −2 proline (Fig. 4). There is also additional binding energy derived from an antiparallel β augmentation, hydrogen bonding to the canonical β5 strand of the PTB domain. Yet, because these involve only main-chain contacts, the β5 strand participates rather “nonselectively” in cargo recognition (Dvir et al. 2012). At the carboxy-terminal end of the short signal β strand, the central −3 asparagine makes an intramolecular bond with tyrosine 0 to redirect the peptide main-chain configuration into a type I β turn, strongly facilitated by the following −2 proline. Overall, this structures the sorting signal to optimally engage the cognate interface of the PTB domain (Fig. 4).

MODULATION OF [FY]XNPX[YF] SIGNAL RECOGNITION BY UBIQUITINATION

Another interesting mode of modulation of the trafficking behavior of the LDL receptor FDNPVY signal is by targeted ubiquitination by an E3 ubiquitin ligase designated IDOL (for inducible degrader of the LDL receptor) (Zelcer et al. 2009). The recognition site for IDOL overlaps the FDNPVY signal; this is because substrate recognition in this particular E3 ligase stems from an atypical FERM (band 4.1/ezrin/radixin/moesin) domain that precedes the catalytic carboxy-terminal RING domain (Calkin et al. 2011; Sorrentino et al. 2011). Prior structural work on FERM domains from actin-regulating proteins like ezrin and talin revealed a tripartite fold, with the so-called F3 subdomain structurally analogous to the PTB domain (Pearson et al. 2000). The talin F3 subdomain binds to the NPXY signal in β integrins (Wegener et al. 2007), using, in part, the same basic structural arrangement as seen in ARH. Likewise, IDOL contacts physically the LDL receptor FDNPVY sorting signal, and signals in the related VLDL and ApoER2 receptors (Hong et al. 2010). The core components are also phylogenetically conserved in insects (Hong et al. 2010; Calkin et al. 2011). The IDOL ubiquitin ligase targets sites that are adjacent to the FDNPVY signal: two (Lys809) or 11 (Cys819) residues following Tyr807 (Fig. 2D) (Zelcer et al. 2009). The functional consequence of this ubiquitination is redirection of the LDL receptor to lysosomes for degradation at the expense of efficient recycling (Zelcer et al. 2009). Remarkably, initial results suggest that the IDOL-mediated polyubiquitin signal attached to the LDL receptor reroutes the protein from clathrin-dependent to clathrin-independent internalization, clearly inactivating the FDNPVY signal (Scotti et al. 2013).

OPERATION OF [FY]XNPX[YF] SIGNALS ON ENDOSOMES

Functionally analogous to YXXØ signals, [FY]XNPX[YF] signals can direct endosomal sorting events. This can happen in at least two ways: indirectly through interaction of PTB-domain-containing CLASPs with AP-1, and directly through recognition by FERM-like domain-containing sorting nexin (SNX) proteins. In one process in polarized epithelial cells, the respective adaptors, ARH and a variant of AP-1 having the μ1B subunit isoform (AP-1B), cooperate for differential sorting to the basolateral membrane from endosomes (Kang and Folsch 2011). Yet the capability of ARH to mesh with both AP-1B and AP-2 depends on the sole ARH β-subunit interaction motif (He et al. 2002; Mishra et al. 2005; Edeling et al. 2006; Schmid et al. 2006); mammalian β1 and β2 are ∼85% identical and functionally interchangeable (Keyel et al. 2008; Li et al. 2010). Consequently, the surface on the β1 appendage that ARH binds to is fully conserved in β2. ARH also has a broadly basic surface perpendicular to the [FY]XNPX[YF]-binding site that could accommodate different phosphoinositides rather than being tailored for a stereospecific isomer.

Another example of cooperation between a PTB-domain CLASP and AP-1 is provided by Numb. This protein (and the redundant paralog Numb-like) is involved in Notch signaling (Berdnik et al. 2002), cell adhesion, migration (Wang et al. 2009), and morphogenesis (Berdnik et al. 2002; Bogdanovic et al. 2012). Numb can also function on endosomes (Smith et al. 2004; Nilsson et al. 2008; McGill et al. 2009; Cotton et al. 2013; Couturier et al. 2013), and clearly has endosome-located binding partners, although biochemical interactions revolve principally around AP-2 and associated proteins (Krieger et al. 2012). The plasma membrane- and endosome-associated chordate Numb isoforms are different splice variants (Dho et al. 1999) and regulated by posttranslational phosphorylation (Smith et al. 2007; Sorensen and Conner 2008). In Drosophila, Numb cooperates with AP-1 to control Sanpodo trafficking on endosomes (Cotton et al. 2013; Couturier et al. 2013) but, in this case, basolateral recycling is prevented (Cotton et al. 2013), contrary to ARH.

A different mechanism is the recognition of [FY]XNPX[YF] signals by SNX proteins, specifically by the paralogs SNX17, SNX27, and SNX31 (Ghai et al. 2011, 2013; Bottcher et al. 2012). SNXs are a diverse class of endocytic proteins unified by the presence of a phox-homology (PX) domain selective for endosomal phosphatidylinositol 3-phosphate (PtdIns(3)P) (Cullen 2008; Teasdale and Collins 2012). SNX17, SNX27, and SNX31 stand apart from the other family members by uniquely containing an atypical FERM domain after the PX fold. The LDL receptor signal directly engages the SNX17 FERM domain (Stockinger et al. 2002; Burden et al. 2004), as do the [FY]XNPX[YF] signals in P-selectin (Florian et al. 2001; Knauth et al. 2005; Ghai et al. 2013), APP (Lee et al. 2008b), β integrins (Bottcher et al. 2012; Steinberg et al. 2012), and LRP1 (van Kerkhof et al. 2005; Donoso et al. 2009) in a phosphorylation-inhibited manner (Betts et al. 2008). The structure of the SNX17 FERM domain complexed with the P-selectin sorting signal (Ghai et al. 2013) confirms both that the F3 subdomain resembles the PTB-domain fold and that peptide binding conforms to conventional structural rules (Fig. 4). Compartmental recognition of [FY]XNPX[YF] signals is assured by the preceding PX domain, which favors endosomal PtdIns(3)P (Knauth et al. 2005; van Kerkhof et al. 2005; Ghai et al. 2011). Because of the proximity of the acceptor lysine, the IDOL-ubiquitinated FDNPVY signal in the LDL receptor may not be recognized by SNX17, explaining the diminished recycling. Strikingly, several different endocytic signals have internal or proximate lysine residues known to be ubiquitinated (Fig. 2D). This suggests that ubiquitin-dependent masking of linear sequences might be another common regulatory mechanism.

If the bulk of PTB-like domains engage [FY]XNPX[YF] signals with common structural features and a conserved binding mode, how is specificity attained? The ARH and SNX17 structures highlight another general principle for signal recognition: directly flanking residues at the amino- or carboxy-terminal end of the core sorting signal allow tailored recognition of a certain tertiary-structured binding partner. For instance, P-selectin contains a bona fide μ2-binding YXXØ-type YGVF signal interleaved with the FTNAA[YF] signal (Owen et al. 2001). In the SNX17 crystal, this peptide sequence also packs against the FERM domain F3 subdomain (Fig. 4) (Ghai et al. 2013), explaining why this could not simultaneously be recognized by AP-1 or AP-2. The β1-integrin chain similarly uses a sequence tract preceding NPKY to optimally engage SNX17 (Bottcher et al. 2012). Such extended contacts facilitate optimal conformational selection by allowing domains to discriminate between a large array of potentially similar unstructured binding peptides and select between subtly different signals. Alternatively, restricted expression patterns or temporal induction can provide specificity, as with IDOL being induced by LXR transcription factor activation (Zelcer et al. 2009).

DISTANT RELATIONS: μ-HOMOLOGY DOMAIN CLASPS

Because of the pivotal role of the μ subunit in AP-1–AP-4 cargo selection, the fact that two other protein classes deposited at clathrin assembly centers, the stonins and muniscins, also contain a μ-homology domain (μHD) (Fig. 4) is highly suggestive of further cargo recognition capabilities. The most closely related μHD is found in stonins (Andrews et al. 1996; Martina et al. 2001; Walther et al. 2001); the human stonin 2 µHD is ∼30% identical to the μ2-C domain of AP-2. In contrast, the muniscin μHD is <10% identical to μ2 accounting for the original designation FCH domain “only” (Fcho), as the initial sequence comparisons failed to identify the structural homology (Katoh 2004; Henne et al. 2007). The μHD of the yeast muniscin Syp1p is ∼14% identical to the stonin 2 domain, but neither has a similarly positioned longin-like μ-N domain, which coassembles with the β chain within the adaptor core. The stonin μHD is preceded by a phylogenetically conserved tract of ∼140 residues known as the stonin-homology domain (Martina et al. 2001) currently of unknown structure or function.

The stonins are ubiquitously expressed in chordate tissues but seem most critical in neurons; mutant alleles or silencing produces neurological phenotypes in flies, worms, and mice (Stimson et al. 2001; Estes et al. 2003; Mullen et al. 2012; Kononenko et al. 2013). At the synapse, tight exo-endocytic coupling maintains a pool of neurotransmitter-filled synaptic vesicles. Here, transmembrane cargo recognition is critical as synaptic vesicles have a highly defined protein composition (Takamori et al. 2006), and there is a remarkably conserved packaging stoichiometry for some transmembrane proteins within reforming synaptic vesicles (Takamori et al. 2006; Mutch et al. 2011). Synaptotagmin 1 is a vesicle-associated, Ca²⁺-responsive bilayer fusion regulator and therefore necessarily packaged into each regenerating synaptic vesicle following exocytosis. Biochemical and functional evidence indicates that the C2 domains of synaptotagmin 1 bind physically to the stonin 2 µHD (Martina et al. 2001; Walther et al. 2001; Diril et al. 2006; Jung et al. 2007). In neurons, the two proteins colocalize in puncta, and proper subcellular localization of SNT-1 depends on UNC-41 (the respective nematode orthologs) but not vice versa (Mullen et al. 2012). Because synaptotagmin (snt-1)-null worms phenocopy stonin (stn-1/unc-41) mutants (Mullen et al. 2012), a principal task of stonins is likely to promote uptake of synaptotagmin at nerve terminals.

Still, there is current controversy over precisely which presynaptic cargoes stonin 2 sorts. Worms with both snt-1 and unc-41 genes disrupted have a more severe phenotype than with either single mutation alone (Mullen et al. 2012). And forced SNT-1 overexpression does not rescue the unc-41 phenotype, unlike in Drosophila (Fergestad and Broadie 2001). Together, this suggests stonins do more than just sort synaptotagmin. A systematic CLASP RNAi screen of cultured neurons suggests that stonin 2 is actually the major adaptor for synaptic vesicle cargoes (Willox and Royle 2012), implying that it recognizes a bigger set of transmembrane proteins. But, of the synaptic vesicle constituents, only synaptotagmin 1 coimmunoprecipitates robustly with stonin 2 (Kononenko et al. 2013). And the subcellular phenotype of stonin 2-nullizygous mice disputes the notion of a general-purpose CLASP; synaptotagmin but not VAMP 2 selectively disperses over the presynaptic surface (Kononenko et al. 2013). A future challenge is to rationalize these discrepant findings.

For the muniscins, cargo partners are documented (Reider et al. 2009; Umasankar et al. 2012), but the μHDs are otherwise different in their interaction networks. This is because these μHDs are not solely engaged in cargo binding. Both the yeast and the chordate muniscin μHDs establish direct contacts with several early-arriving pioneer components of the clathrin coat (Reider et al. 2009; Henne et al. 2010; Mulkearns and Cooper 2012; Umasankar et al. 2012). The crystal structure of the Syp1p μHD leaves no doubt about the structural homology with μ2 (Fig. 4). Still, the precise interaction surface on the μHD remains to be mapped. Therefore, it is not unambiguously resolved whether cargo and endocytic proteins use the same interaction surface(s) to associate with muniscins at the plasma membrane. Certainly for the stonin 2 µHD•C2A/B interaction, both partners are folded domains, distinguishing this contact from the short peptide-based, extended conformation interactions described above.

Elucidation of the molecular basis for selection of endocytic SNAREs has provided new structural insights into recognition of folded sorting signals. Because the immediate fate of donor transport vesicles is fusion with an acceptor compartment, proper packaging of SNAREs is vital. One mechanism to assure selective SNARE capture uses a disordered region of a dedicated CLASP to engage a broad molecular surface of a particular SNARE. The antithesis of the standard mode of engagement, this is how Hrb binds to VAMP 7 (Pryor et al. 2008) and, analogously, how the usually disordered hinge of the AP-3 δ subunit structures around the globular regulatory VAMP 7 longin domain to assure packaging of only a four-member cis-SNARE helical bundle (Fig. 5A) (Kent et al. 2012). The “signal” here is a radial groove along the longin-domain surface accessed by either Hrb, the δ hinge, or intramolecularly by VAMP 7 itself in the autoinhibited conformation that precludes CLASP binding. In contrast, the α-helical-solenoid fold of the CALM (and the neuronal paralog AP180) ANTH domain has evolved to selectively recognize the uncomplexed vesicle SNAREs VAMP 2, 3, and 8 (Koo et al. 2011; Miller et al. 2011). In this case, the single six-turn amphipathic α-helical tract of VAMP 8 that physically contacts the ANTH domain (Fig. 5B) also participates in SNARE helical bundling, making the two interactions mutually exclusive.

Figure 5.

Decoding folded endocytic sorting signals. (A) Schematic representation of the overall architecture of VAMP 7 with the amino-terminal longin domain, the vesicle-SNARE (v-SNARE) helical segment, and the transmembrane (Tm) domain indicated. A combined ribbon and surface representation of the globular VAMP 7 longin-domain “signal” bound by the extended AP-3 δ-subunit hinge polypeptide (blue; in stick representation) is also shown (PDB ID: 4AFI) (Kent et al. 2012). Notice the insertion of several hydrophobic side chains from the δ hinge into complementary acceptor cavities arrayed around the regulatory longin-domain circumference. (B) Schematic of the architecture of VAMP 8, which lacks the autoinhibitory longin domain. A representation of the amino-terminal portion (residues 15–38) of the α-helical v-SNARE domain of VAMP 8 complexed with the α-solenoid folded ANTH domain from CALM (PDB ID: 3ZYM) (Miller et al. 2011). The structured VAMP 8 amphipathic α helix orients seven aliphatic side chains to insert into a vertical groove along the ANTH surface (white dashed line).

CONCLUDING REMARKS AND FUTURE PERSPECTIVES

We are long past a primitive understanding of sorting signals governing coat-mediated endocytic trafficking. With elegant high-resolution structures of verified peptide sorting signals bound to their cognate interaction domains, the field has come of age. What is clear is that there is a rich variety of interaction possibilities to drive distinct sorting processes. The core primary-sequence sorting signals are four-to-six residues long, similar to the majority of eukaryotic linear motifs (ELMs) (Davey et al. 2012). Short linear sequences facilitate convergent evolution. But why is not only a single, optimized master signal present within all cargoes to be internalized from the cell surface? Certain pathogenic bacteria, for example, encode effector proteins injected into host cells, which become tyrosine phosphorylated and then bind to numerous distinct cellular SH2 domain-containing signaling regulators (Selbach et al. 2009). In general, these phosphopeptide sequences are more similar within bacteria than to the multiple endogenous SH2-binding proteins they have evolved to subvert. In fact, SH2 domains can be artificially tailored to bind phosphotyrosine signals considerably tighter (Kaneko et al. 2012). Similarly, interaction partners for clathrin-mediated endocytosis have evolved to permit relatively weak, transient but selective associations to enable efficient trafficking. It is evident then that cargo recognition folds are not optimally evolved to bind sorting signals with highest affinity. Hence, there is significant variation within and between operative signals, which have been favored to allow the flexibility and appropriate assembled dwell times necessary for normal trafficking operations. Other critical aspects of an expanded sorting-signal code are that it avoids direct competition between cargoes for uptake (Warren et al. 1998), permits discrimination between monomeric and complexed SNAREs, and allows synchronization of multiple traffic flows.

Moreover, if sorting depended solely on bimolecular cargo–coat interactions, the process would be highly error prone owing to promiscuous binding errors with related but inappropriate sequences (Ladbury and Arold 2012). And if the same signal operates at different sorting stations, it is implicit that information necessary to recruit the appropriate coat/adaptor is not the signal itself. Indeed, in AP-1 and AP-2, the cargo-recognition surfaces only become available once membrane association is established. One mechanism to increase accuracy is by temporal integration of several low-affinity contacts through spatially separate contact surfaces to guard against random noise, in the form of nonspecific univalent engagement by related (or unrelated) proteins. This robust recognition only under certain circumstances typifies selective binding of conditional peripheral membrane proteins (Moravcevic et al. 2012), and is often termed “coincidence detection” (Carlton and Cullen 2005) or a “dual-key strategy” (Itoh and De Camilli 2004). Coat assembly is therefore a probabilistic process, with increased likelihood of successful assembly with the more synchronous linked interactions at a nascent bud site. By binding to multiple adaptors or CLASPs, cargoes participate directly in the network, and thus modulate the coat assembly process (Mettlen et al. 2010). A corollary of this densely wired redundant network is that gene silencing or RNAi of numerous individual CLASPs does not have a dramatic penetrant effect on endocytosis (Garcia et al. 2001; Kang-Decker et al. 2001; Morris et al. 2002; Kamikura and Cooper 2003; Holmes et al. 2007; Koh et al. 2007; Wang et al. 2008; Chen et al. 2009; Mullen et al. 2012; Pozzi et al. 2012; Scotland et al. 2012; Suzuki et al. 2012; Umasankar et al. 2012; Kononenko et al. 2013; Tsushima et al. 2013). Rather, subtle or tissue-specific defects are manifest linked to improper signal relay by the cargo affected. This concept is also concordant with pathogenic microorganisms requiring different elements of the endocytic machinery for successful entry and infection (Eto et al. 2008; Bhattacharyya et al. 2011; Bonazzi et al. 2011; Fukumatsu et al. 2012).

It is unlikely that we currently have a complete catalog of operational sorting signals. Ongoing functional analysis reveals that within specific cargo, alterations in relative residue spacing and even tolerated loss of invariant anchor residues can occur (Kozik et al. 2010; Gephart et al. 2011; Ortega et al. 2012; Wang et al. 2012). Other noncanonical neighboring side chains compensate under these circumstances. A general screen for cytosolic sequences able to promote efficient internalization of a surface reporter highlights several novel operational sequences, yet are still heavily dependent on AP-2 (Kozik et al. 2010). First, designated “RYR,” appears initially to be an altered YXXØ-type sequence, superficially analogous to the YEQGL sequence in the P2X4 purinergic receptor (Royle et al. 2005). However, although the tyrosine in the RYR is the most vital side chain, it does not appear to bind to the YXXØ site on μ2-C subdomain A (Kozik et al. 2010). μ2 also binds to a basic stretch (devoid of aromatic residues) of the δ subunit of GABA_A receptors (Gonzalez et al. 2012), so there is likely another contact surface for cargo recognition, as is evident on the μ4 subunit of AP-4 (Burgos et al. 2010). Then there are the completely novel operational sorting signals: NMDA receptors that incorporate a GluN3A subunit use a YWL sorting signal for internalization (Chowdhury et al. 2013). In this signal, the leading tyrosine can be mutated; it is not indispensable. Remarkably, although a Y971F substitution is defective, a phosphomimetic Y971E permits better AP-2 binding and internalization (Chowdhury et al. 2013). This, it seems, is a sorting signal stimulated by tyrosine phosphorylation. The signal may be related to the SWF tripeptide sequence at the end of the novel “WPK” signal, which is AP-2 and clathrin dependent (Kozik et al. 2010). In both, alanine substitution of the W[LF] pair strongly inhibits internalization (Kozik et al. 2010; Chowdhury et al. 2013). Ordered packing of unstructured regions of a CLASP over a cargo surface could also be more widespread than currently appreciated, and may allow proteins with large intrinsically disordered regions together with a folded modular domain to display cargo selectivity for more than one signal type, just like the heterotetrameric adaptors.