Clathrin-mediated endocytosis is an essential cellular mechanism by which all eukaryotic cells regulate their plasma membrane composition to control processes ranging from cell signaling to adhesion, migration and morphogenesis. The formation of endocytic vesicles and tubules involves extensive protein-mediated remodeling of the plasma membrane that is organized in space and time by protein–protein and protein–phospholipid interactions. Recent studies combining high-resolution imaging with genetic manipulations of the endocytic machinery and with theoretical approaches have led to novel multifaceted phenomenological data of the temporal and spatial organization of the endocytic reaction. This gave rise to various – often conflicting – models as to how endocytic proteins and their association with lipids regulate the endocytic protein choreography to reshape the plasma membrane. In this Review, we discuss these findings in light of the hypothesis that endocytic membrane remodeling may be determined by an interplay between protein–protein interactions, the ability of proteins to generate and sense membrane curvature, and the ability of lipids to stabilize and reinforce the generated membrane shape through adopting their lateral distribution to the local membrane curvature.
Endocytosis is a cellular process that involves the formation of small membrane vesicles or tubules to deliver cargos, such as ligand-bound receptors, ion channels, transporters or other types of plasma membrane molecules, into the eukaryotic cell interior (Brodsky, 2012; Kaksonen and Roux, 2018; McMahon and Boucrot, 2011). Clathrin-mediated endocytosis (CME) arguably is a primary endocytic mechanism and of central importance for nutrient uptake, cell signaling, adhesion, migration and morphogenesis, as well as for neurotransmission, among many other processes. The formation of endocytic vesicles during CME is accompanied by extensive remodeling of the plasma membrane that is organized in space and time by endocytic proteins and their interactions with each other and with specific membrane lipids (Daumke et al., 2014; Kaksonen and Roux, 2018; McMahon and Boucrot, 2011). During recent years we have acquired extensive knowledge about the CME machinery, which, apart from clathrin, comprises more than 50 additional cytosolic proteins that have roles in clathrin recruitment and assembly, cargo selection, membrane bending, membrane fission and vesicle uncoating to eventually allow fusion of the nascent endocytic vesicle with endosomes (Box 1).
The CME machinery comprises more than 50 additional cytosolic proteins, in addition to clathrin, that have roles in clathrin recruitment and assembly, cargo selection, membrane bending, membrane fission and vesicle uncoating to eventually allow fusion of the nascent endocytic vesicle with endosomes (extensively reviewed in Kaksonen and Roux, 2018; Kelly and Owen, 2011; McMahon and Boucrot, 2011; Merrifield and Kaksonen, 2014). In typical CCPs of 150 nm diameter these proteins are present in copy numbers that range from ∼180 for AP-2 or CALM to ∼30–40 molecules for PI3KC2α and SNX9/18 (Borner et al., 2012; Schoneberg et al., 2017) and 30 molecules for dynamin (Grassart et al., 2014).
Cargo selection and clathrin recruitment are mediated by adaptors, such as the heterotetrameric AP-2 complex (comprising α, β, µ and σ subunits) and a plethora of cargo-selective adaptors for sorting specific cargos [e.g. epidermal growth factor receptor substrate 15 (Eps15) and Eps15-interacting 1 and 2 (epsin 1 and 2) for ubiquitylated cargos, arrestins for G protein-coupled receptors, Dab2 and ARH for LDLR family receptors, CALM or AP180 for VAMPs/ synaptobrevins, or stonin 2 for synaptotagmins (reviewed in Kaempf and Maritzen, 2017; Kelly and Owen, 2011; Maritzen et al., 2010; Reider and Wendland, 2011; Traub, 2009)]. In many cell types, the maturation of CCPs is accompanied and possibly facilitated by the assembly of a network of branched actin filaments mediated by neuronal Wiskott–Aldrich syndrome (N-WASP) and the actin-related protein (ARP) 2/3 complex (Doyon et al., 2011; Kaksonen and Roux, 2018; Yoshida et al., 2018). Recent data have suggested that N-WASP is activated at the base of CCPs by the F-BAR protein FCHSD2 (Almeida-Souza et al., 2018). Filamentous (F)-actin might aid CCP maturation (Almeida-Souza et al., 2018) and/or fission of the invagination neck of late-stage CCPs that is mediated by BAR-domain-containing proteins, such as endophilins, amphiphysins and SNX9 and SNX18, and the BAR domain protein-associated GTPase dynamin (Doyon et al., 2011; Kaksonen and Roux, 2018).
In spite of important advances in the characterization of these factors, key questions regarding the endocytic process remain. For example, different models have been proposed regarding the initiation of clathrin-coated pit (CCP) formation and the onset of membrane curvature acquisition. Moreover, knowledge regarding the nanoscale distribution of endocytic proteins within the bilayer plane during vesicle formation and the mechanisms that guide their orchestrated assembly and disassembly is only now beginning to emerge. Finally, our understanding of the nature and dynamics of membrane phospholipids (Box 2) and their associations with endocytic proteins during endocytic membrane remodeling remains limited.
Endocytic protein recruitment is governed by specific membrane phospholipids, in particular by phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2] and phosphatidylinositol 3,4-bisphosphate [PI(3,4)P2] (Posor et al., 2014). While CCP nucleation strictly depends on PI(4,5)P2 (Di Paolo and De Camilli, 2006; Kaksonen and Roux, 2018), membrane remodeling during CCP maturation into a CCV is accompanied by the partial conversion of the identity of the internalized membrane from PI(4,5)P2, a characteristic lipid of the plasma membrane that is progressively hydrolyzed during the late stages of the endocytic reaction by the 5-phosphatases OCRL and synaptojanin, to PI(3,4)P2, which is generated by local phosphorylation of PI(4)P underneath the clathrin-coated membrane area by phosphatidylinositol 3-kinase C2a (PI3KC2α) (Kaksonen and Roux, 2018; Posor et al., 2014). Arrival of PI3KC2α follows that of clathrin with some delay to trigger the PI(3,4)P2-dependent recruitment of PX-BAR domain proteins SNX9 and SNX18. Lack of either PI3KC2α, or SNX9 or SNX18 stalls CME at the level of deeply invaginated CCPs that fail to form a narrow membrane neck, which provides a suitable substrate for endocytic vesicle scission by dynamin (Lo et al., 2017; Posor et al., 2013; Schoneberg et al., 2017). A similar CCP maturation phenotype is observed in cells depleted of the N-WASP-associated F-BAR domain protein FCHSD2, another effector of PI(3,4)P2 (Almeida-Souza et al., 2018). As CCPs are scissioned from the plasma membrane, PI(3,4)P2 is, likely, hydrolyzed to PI(3)P, a signature lipid of the endosomal system (He et al., 2017; Kaksonen and Roux, 2018; Posor et al., 2014).
Here, we review the existing data on four mutually complementing aspects of the process of endocytic vesicle generation in CME: (1) the timecourse of shape transformations of the endocytic site, (2) the ability of endocytic proteins to generate membrane curvature, (3) the spatial organization of endocytic proteins in the plane of the endocytic site, and (4) the timecourse of endocytic protein arrival at and departure from the endocytic site. We will formulate a model of CME that accounts for all four sets of data, and, hopefully, will bring us closer to understanding the molecular mechanism of the endocytic process.
Timecourse of shape transformations of endocytic plasma membrane sites
Formation of clathrin-coated endocytic vesicles (CCVs) involves major transformations of the local shape and topology of the plasma membrane. In general, formation of a CCV involves two processes: (1) the assembly of a clathrin coat on the membrane surface, and, (2) the shaping of the resulting clathrin-coated structure (CCS) into a CCV. The latter reaction starts with the invagination of a CCS (Fig. 1A) towards the cytoplasm and the formation of a shallow CCP that adopts the shape of a spherical membrane segment characterized by homogeneous and isotropic curvature (Fig. 1B). Shallow pits undergo progressive invagination into deep dome-like shapes, which are smoothly connected to the plasma membrane by a funnel-like rim (Fig. 1C). Further invagination leads to formation of a spherical bud, while the rim transforms into a distinct hourglass-like membrane neck (Fig. 1D). Eventually, the neck undergoes fission resulting in formation of a CCV (Fig. 1E).
A topic of heated discussion in the literature is the temporal relationship between these two processes, that is, between the timing of the clathrin coat assembly that can be visualized as CCS area growth and the dynamics of CCS shaping (curvature acquisition) into a CCV. EM images of CCSs taken in the 1980s showed that CCPs coexist with, and are linked to, the edges of larger flat clathrin lattices or ‘plaques’ (Heuser, 1980, 1989; Heuser and Kirchhausen, 1985; Larkin et al., 1986). Clathrin plaques were often seen to be surrounded by smaller clathrin domes that had emerged out of the otherwise flat lattices (Maupin and Pollard, 1983), and scission events leading to the formation of coated vesicles tended to cluster at the edges of plaques (reviewed in Lampe et al., 2016). The relative dynamics of pit and plaque formation, however, was not resolved at that time. Recent studies using state-of-the-art imaging techniques have led to conflicting suggestions regarding the origin of curved pits and flat plaques and the timecourse of a possible transformation of plaques into pits. According to one scenario, referred to as the constant curvature model, the essence of CCP maturation is the increase of its area through self-assembly of identical clathrin elements characterized by a curved shape in the membrane plane. The curvature of a CCP remains constant throughout the whole maturation process and equal to the curvature of a single clathrin element (Cocucci et al., 2012; Kirchhausen, 2009; Saffarian et al., 2009). In contrast, the constant area model postulates that CCS transformation into a curved CCP and, further, a spherical CCV proceeds upon conservation of the in-plane area of the clathrin array, while its curvature gradually increases through the structural rearrangements of the clathrin cage (Avinoam et al., 2015; see Box 3 for details). The essential difference between the underlying principles of the constant area and constant curvature models consists in the coupling between the degree of CCS indentation into the cytoplasm and the growth of its area. The constant curvature model suggests that the two processes proceed in parallel, that is that they are strongly coupled to each other, while the constant area model proposes that the degree of CCS indentation that results from the curvature acquisition is fully decoupled from CCS area growth. A number of recent studies have provided further indications in favor of the latter model by showing that CCS indentation into the cytoplasm is largely decoupled from CCS area growth (Bucher et al., 2018; Scott et al., 2018). Application of correlative light and electron microscopy (CLEM) to BSC-1 cells, which lack long-lived clathrin-coated plaques, in combination with mathematical analysis led to the conclusion that CCSs first grow flat and start developing curvature later on, once they have accumulated about 70% of their final clathrin content (Bucher et al., 2018). Beginning from this stage, while completing the area growth, a CCS acquires the consecutive shapes of pits, domes and buds (Bucher et al., 2018). Another study of CCP dynamics used polarized total internal reflection fluorescence (pol-TIRF) microscopy combined with electron, atomic force and super-resolution optical microscopy, and measurements of the time evolution of CCSs in SKMEL-2 cells that express endogenously tagged clathrin light chain and dynamin 2 (Scott et al., 2018). That work revealed a variable delay between the beginning of CCS assembly and the onset of curvature acquisition. For approximately half of the CCSs, an acquisition of curvature was detected already at the very beginning of CCS assembly. In the other half of the events, CCSs started to grow prior to the detection of membrane bending. Within these latter group of events, there was a small subset in which CCSs started to bend after having reached their final areas, and a larger group for which the curvature acquisition overlapped in time with the completion of area growth (Scott et al., 2018). Delayed membrane bending after the onset but prior to the completion of clathrin assembly was also described in a recent study that combined high-speed atomic force microscopy with fluorescence confocal microscopy (Yoshida et al., 2018).
The constant curvature model of CME (Cocucci et al., 2012; Kirchhausen, 2009; Saffarian et al., 2009) sees a CCP as composed of effective units, with each unit consisting of a clathrin triskelion with an underlying membrane element and having a shape of a spherical segment that is characterized by a certain curvature. These units self-assemble within the surrounding membrane side-by-side, similar to mosaic pieces, into a CCS. Within this scenario, in the course of the CCS area growth, that is, as additional clathrin triskelia are added, the structure progressively invaginates into the cytoplasm with its shape evolving from a shallow to a dome-like CCP and further to a spherical clathrin-coated bud. Importantly, the curvature of all these structures remains constant and equal to that of a single clathrin element (Kirchhausen, 2009; Saffarian et al., 2009).
Recently, the constant curvature model was challenged by the application of a correlative fluorescence microscopy and electron tomography method, which provides three-dimensional information about the shapes of clathrin-coated pits (Avinoam et al., 2015). According to this study, the coated membrane area in CCPs and, hence, the total number of clathrin molecules, do not appreciably change throughout the entire CCP maturation and budding process, beginning from slightly invaginated pits through completely spherical CCVs. Conservation of the area in the course of invagination means that the curvature of the structures must change all the way from zero at the onset of CCP shape acquisition up to its maximal value at the final stage of CCV formation. This scenario is referred to as the constant area model.
The lack of a firm temporal coupling between the processes of CCS area growth and shaping suggests that curvature acquisition by a growing CCS is a process that requires clathrin coat assembly, but is not a direct consequence of it. Hence, another process must mediate the link between CCS area growth and curvature acquisition. An example of such a process could be the accumulation of curvature-mediating proteins, such as epsins and CALM (or its close neuronal paralog AP180, also known as SNAP91) (see below; Chen et al., 1998; Ford et al., 2002; Koo et al., 2015; Maritzen et al., 2012; Messa et al., 2014; Miller et al., 2015; Scott et al., 2018; Zhang et al., 1998) by the emerging clathrin coat. Heterogeneity between cells or cell types may be one reason for the observed variations in the delay between the onset of curvature acquisition and the timing of clathrin coat self-assembly.
Membrane curvature generation by endocytic proteins
The architecture of an invaginating CCS has two parts with fundamentally distinct shapes: the dome-like part, which has the shape of a spherical segment, whose solid angle increases in the course of pit transformation into a dome-like and, then, a spherical bud (Fig. 1). In addition, there is the membrane neck, which connects the CCP to the plasma membrane and has a funnel- or hourglass-like shape depending on the stage of CCP maturation (Fig. 1). The spherical dome has an overall concave shape that is isotropically curved, that is has equal curvatures in all directions along the membrane surface (Fig. 1). In contrast, the neck part displays a saddle-like configuration that is anisotropically curved: the neck membrane is concave along the direction of the CCP axis, and convex in the perpendicular direction (Fig. 1). In geometrical terms, the isotropically curved dome-like part of a CCP is characterized by two equal principal curvatures, which are conventionally defined to be negative, whereas the two principal curvatures of the neck have opposite signs, one being positive and the second being negative (see, for example, Spivak, 1970).
Among the ∼50 different proteins involved in CCV formation (Kaksonen and Roux, 2018; McMahon and Boucrot, 2011; Merrifield and Kaksonen, 2014), there is a relatively limited subset of proteins that are deemed responsible for shaping the CCS membrane. We will refer to these proteins as the structural proteins of endocytosis. In the following section, we will first discuss the basic mechanisms by which proteins generate and/or stabilize membrane curvature, and then describe the endocytic structural proteins involved in sculpting the isotropically curved dome-like part of an invaginating CCS and its anisotropically bent neck.
Membrane curvature generation
The ability of proteins to generate membrane curvature has been extensively explored both experimentally and theoretically over the past decade (for recent reviews, see Jarsch et al., 2016; McMahon and Boucrot, 2015; Simunovic et al., 2016). Less emphasis has been put on the isotropic versus anisotropic character of the local membrane shape created by each of the specific curvature-generating proteins. Here, we will stress the latter issue since it is of importance for our discussion of the shaping of the dome- and neck-like parts of CCSs. There are two major mechanisms by which proteins can generate local membrane bending. One is the hydrophobic insertion (or wedging) mechanism, based on shallow embedding in one of the membrane leaflets of a hydrophobic or amphipathic protein domain (Zimmerberg and Kozlov, 2006), such as N-terminal amphipathic helices (Ford et al., 2002; Kozlov et al., 2014) or short hydrophobic loops (McMahon et al., 2010). Such a mode of protein insertion results in splaying the region of lipid polar heads of the lipid monolayer with respect to the remainder of the monolayer interior, which generates a strong tendency of the membrane to locally bulge in the direction of the insertion. Importantly, embedding of an amphipathic helix that is a few nanometers large generates an essentially isotropic curvature around the insertion (Campelo et al., 2008; Kozlov et al., 2014). This is a consequence of the behavior of lipid monolayers as two-dimensional fluids. Owing to this membrane property, the possible curvature anisotropy that has developed in the closest proximity of the inserted domain disappears at a nanometer-scale distance from the insertion, such that the membrane deformation around the insertion is, largely, isotropic (Campelo et al., 2008). Hence, proteins generating membrane curvature through the hydrophobic insertion mechanism are particularly suited to shape the dome-like part of CCSs characterized by an isotropically curved shape.
The other established mechanism of local curvature generation referred to as the scaffolding mechanism, underlies the action of rigid hydrophilic protein domains with an intrinsically bent shape (McMahon and Gallop, 2005; Zimmerberg and Kozlov, 2006). Binding of such protein domains to the polar face of the membrane imposes the curvature of the protein onto the membrane patch underneath. For the scaffolding mechanism to be effective, the intrinsic rigidity of the protein domain and its affinity to the membrane surface must be substantially larger than the rigidity of the membrane with respect to bending (Zimmerberg and Kozlov, 2006). The most common protein domains producing bent membranes by the scaffolding mechanism are the ∼10 nm-long dimers of bin-amphiphysin-rvs (BAR) domains (for a review, see Daumke et al., 2014; Frost et al., 2009; McMahon and Boucrot, 2015) that most often have a crescent-like shape due to the way the two monomers are bound to each other. Because of its strongly elongated shape, a BAR domain dimer is predicted to bend the membrane underneath in the direction of the protein axis but not in other directions in the membrane plane (Schweitzer and Kozlov, 2015). It has been computationally predicted that this curvature anisotropy must propagate along the membrane to distances of up to tens of nanometers around the protein (Schweitzer and Kozlov, 2015). Hence, such scaffolding proteins can be expected to act in shaping the CCP neck.
In principle, protein scaffolds that generate nearly isotropic membrane curvature may also exist. As discussed below, a hypothetical example of such a scaffold is the clathrin triskelion (den Otter and Briels, 2011; Kirchhausen, 2009; Merrifield and Kaksonen, 2014; Saffarian et al., 2009).
The ability of a protein to produce membrane curvature guarantees that its affinity to membranes depends on the membrane curvature, a property referred to as curvature sensitivity (Antonny, 2011; Campelo and Kozlov, 2014). The mechanism by which a protein senses membrane curvature is intimately related to the mechanism of curvature generation. Protein domains that curve membranes through the hydrophobic insertion mechanism enable relaxation of the intra-membrane stresses generated as a result of membrane bending by external forces (Campelo and Kozlov, 2014). A specific microscopic mechanism of such relaxation was proposed to underlie the protein-mediated rescue of the hydrophobic structural defects that develop in the membrane as a result of bending (Vanni et al., 2013). Stress relaxation makes protein binding more favorable. As a consequence, stronger membrane bending by external forces creates larger intra-membrane stresses, resulting in an enhanced membrane-binding affinity of a membrane-inserting domain, which represents the essence of its curvature-sensing ability (Campelo and Kozlov, 2014). Different from the hydrophobic insertions, the protein domains, which bend membranes by scaffolding, directly sense the curvature of the membrane surface to which they bind rather than intra-membrane stress. The largest affinity of a scaffolding protein must be to a membrane whose curvature matches best the intrinsic curvature of the protein itself, which is the optimal curvature for the protein to associate with. Binding of a protein scaffold to membranes whose curvatures are larger or smaller than that of the protein is expected to be weaker than binding to a membrane with optimal curvature.
Endocytic proteins involved in membrane shaping
The endocytic machinery exerts both the hydrophobic insertion and the scaffolding mechanisms of membrane shaping. For some of the proteins that are potentially involved in sculpting CCSs, the ability to generate and sense membrane curvature has been directly demonstrated and quantitatively characterized. Endocytic factors proven to bend membranes through the hydrophobic insertion mechanism are the AP180 (ANTH) or epsin (ENTH) N-terminal homology domain-containing proteins CALM and epsins 1 and 2 (Ford et al., 2002; Messa et al., 2014; Miller et al., 2015). Depletion of CALM, an abundant clathrin-associated sorting adaptor for VAMP/synaptobrevin SNARE sorting (Koo et al., 2015; Maritzen et al., 2012; Miller et al., 2011), causes a twofold decrease in the curvature of CCPs and CCVs, indicating that CALM can generate membrane curvature by itself (Miller et al., 2015). In vitro studies showed that curvature generation by CALM depends on an N-terminal amphipathic helix within its ANTH domain, which on its own is extremely efficient at extensively tubulating 200 nm-diameter liposomes containing phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2] (Miller et al., 2015). The resulting tubules have a pearl-on-a-chain-like appearance with a pearl diameter of ∼40 to 50 nm, similar to that of the smallest endocytic CCVs found in neuronal synapses (Miller et al., 2015). A similar function may be executed by the CALM-related protein AP180 in presynaptic neurons (Koo et al., 2015; Zhang et al., 1998). Consistent with this, it has been shown that loss of CALM and/or AP180 in flies, worms and mice results in abnormally sized and shaped synaptic vesicles formed by clathrin-mediated budding at synapses (Koo et al., 2015; Zhang et al., 1998). Epsins, similar to CALM, are early-acting clathrin- and AP-2-associated endocytic proteins that contain a PI(4,5)P2-binding membrane-targeting ENTH domain (Chen et al., 1998; Ford et al., 2002). Upon interaction with PI(4,5)P2, the ENTH domain forms an additional amphipathic α-helix that generates local membrane curvature. Importantly, while, at first, this curvature generation by the ENTH domain leads to conversion of the nearly flat liposomal membranes into anisotropically bent thin tubules (Ford et al., 2002), at longer time scales, these tubules transform into separate spherical vesicles with isotropic curvature (Boucrot et al., 2012), as expected for the hydrophobic insertions. Hence, the initial tubules likely represent the kinetically trapped intermediates rather than the equilibrium membrane shapes (Boucrot et al., 2012). Collectively, these data support the idea that CALM and epsins use the hydrophobic insertion mechanisms to shape the isotropically curved dome-like part of invaginating CCPs.
Among the structural proteins that generate curvature merely by scaffolding are endocytic F-BAR domain proteins, such as F-BAR domain only protein 1 or 2 (FCHo1 or FCHo2), and possibly F-box protein 7 (FBP17, also known as FNBP1) and syndapin (Henne et al., 2010, 2007), as well as the structurally unrelated GTPase dynamin (Ford et al., 2011; Faelber et al., 2011; Ferguson and De Camilli, 2012; Sochacki et al., 2017). The dimeric F-BAR module displays the shape of a shallow arc whose concave face binds the membrane surface through attractive interaction between positively charged residues and the negatively charged polar headgroups of acidic phospholipids (Henne et al., 2007; see also Box 2). F-BAR domains can polymerize on the membrane surface into a helical coat, in which the dimers are held together by lateral and tip-to-tip interactions (Frost et al., 2008). These coats convert membranes into tubules with variable diameters between 70 nm (Frost et al., 2008) and up to 130 nm (Henne et al., 2007). Since the F-BAR dimers can be described as shallow crescents characterized by a small or almost vanishing curvature, the function of F-BAR proteins in CME in general and that of FCHo1/2, in particular, remains a matter of debate (Hollopeter et al., 2014; Ma et al., 2016). Initial experimental data are consistent with the idea that FCHo1/2 contributes to slight membrane bending at the rim of shallow early-stage CCPs during the initiating steps of CCP formation (Henne et al., 2010). Interestingly, the subunits of dynamin also show intrinsically curved arc-like shapes; they bind the membrane through electrostatic interactions (Faelber et al., 2011; Ford et al., 2011) and polymerize on the membrane surface into helical structures, which constricts the membrane into 50 nm tubules (see, for example, Hinshaw, 1999). Importantly, upon GTP hydrolysis, dynamin drives fission of membrane necks and tubules through mechanisms that remain uncertain (Antonny et al., 2016).
Taken together, F-BAR domain proteins and dynamin use a scaffolding mechanism that generates anisotropic membrane curvature and are, therefore, suitable for shaping the rims or necks of CCPs.
Finally, dimers of N-BAR endocytic proteins, such as amphiphysin and endophilin, possess both amphipathic N-terminal helices and elongated crescent-like scaffolding domains. The radii of the latter are ∼10 nm (Gallop et al., 2006; Peter et al., 2004) such that their curvatures are significantly larger than those of F-BAR domains (Frost et al., 2009; McMahon and Boucrot, 2015). These proteins have the potential to curve membranes by both the hydrophobic insertion and scaffolding mechanisms (Gallop et al., 2006; Peter et al., 2004). When added to liposomes, N-BAR proteins generate membrane tubules that are a few tens of nanometers (i.e. 20 to 40 nm) thin (Peter et al., 2004; Gallop et al., 2006), and a fraction of these is, over time, converted into small spherical vesicles (Boucrot et al., 2012). Vesiculation appears to depend largely on the hydrophobic insertions as it is more pronounced in the case of endophilin, whose BAR domain dimer harbors an additional amphipathic helix as compared to amphiphysin (Boucrot et al., 2012). Because of these dual properties, endocytic N-BAR domain proteins could, in principle, participate in shaping both the dome-like parts and the necks of CCSs. In the case of the N-BAR domain protein endophilin, its dimers have been shown to form helical polymeric coats through interactions between the membrane-inserted N-terminal helices of adjacent dimers (Mim et al., 2012). These coats can convert liposomes into tubules of 25 nm and 32 nm diameters. Based on the diameter of these endophilin-coated tubules, endophilin dimers would be predicted to localize to the neck of CCPs in cells. Clearly, the interplay between the N-terminal helices and the crescent-like scaffolds of N-BAR proteins in the process of endocytic membrane remodeling needs further exploration.
An important open question is whether the isotropic curvature of the dome-like parts of CCSs is also generated or stabilized by protein scaffolds, in addition to hydrophobic insertion of epsin1/2 and CALM. One possibility is that clathrin triskelia perform this function (Dannhauser and Ungewickell, 2012; Hinrichsen et al., 2006), as the 110° pucker angle at the triskelion apex matches the characteristic values of the isotropic membrane curvature of endocytic vesicle domes (den Otter and Briels, 2011; Kelly et al., 2014; Kirchhausen, 2009; Lampe et al., 2016; Saffarian et al., 2009). However, a problem is that, in addition to the curved configurations, CCSs also can adopt a nearly flat geometry (den Otter and Briels, 2011), in which the pucker angle is 90°. Moreover, for geometrical reasons, the transition between flat and curved configurations of clathrin triskelia requires a substantial rearrangement of the clathrin cage that alters the ratio between the number of hexagons and pentagons, which might represent a serious kinetic challenge (den Otter and Briels, 2011; Lampe et al., 2016). Hence, the question remains as to whether clathrin subunits are able to generate membrane curvature by themselves or whether the polymerized clathrin cage merely serves as a platform for the recruitment of other proteins, which generate the isotropic membrane curvature that, in turn, induces changes in clathrin topology to match the membrane shape underneath (for recent discussion, see Bucher et al., 2018; Scott et al., 2018).
Spatial organization of endocytic proteins within the plane of the endocytic site
As discussed above, the structural proteins responsible for endocytic membrane remodeling can be subdivided into those producing the isotropic membrane shape that characterizes the dome-like part of an invaginating CCS and proteins that generate anisotropic shapes of the membrane neck connecting the CCS to the surrounding flat membrane. Based on this essential difference between the subsets of endocytic structural proteins, it seems logical to assume that these proteins are differently distributed over the CCP surface. For instance, one would expect that dome-sculpting proteins occupy the central part of a nascent CCP, whereas neck-shaping proteins segregate along the CCP periphery. In addition to their membrane-shaping properties, other factors may contribute to an uneven distribution of endocytic proteins along the CCP surface. For example, endocytic proteins that sort and recruit cargo are expected to localize to the central part of the CCS structure, while proteins such as dynamin that are involved in fission of the membrane neck at the final stage of CCV formation may be concentrated along the CCP periphery.
Recent progress in super-resolution imaging has indeed made it possible to directly study the endocytic protein distribution along the CCP membrane (Sochacki et al., 2017). Use of large-scale correlative super-resolution light and transmission electron microscopy has enabled the mapping of the distribution of 19 key proteins involved in CME with nanometer-scale precision and provided a comprehensive molecular architecture of endocytic structures (Sochacki et al., 2017). The most remarkable feature of these data is a clear separation of endocytic proteins into three main groups according to their localization with respect to the edge of the clathrin lattice, which covers the central dome-like part of CCPs (Fig. 2A,B). The inside of the clathrin-coated zone is occupied by a group of seven proteins – epsin (1/2), NECAP (1/2), HIP1R, CALM, and the transmembrane cargos transferrin receptor and VAMP/synaptobrevin 2 – which are distributed all over the clathrin lattice, similar to clathrin light chains. In the course of CCP maturation these proteins remain in the center of the structure and change their concentrations in accordance with that of clathrin (Sochacki et al., 2017). The important structural factors, such as EPS15, FCHo2, amphiphysin, endophilin, syndapin, SNX9 and dynamin, localize almost exclusively at the edge of the clathrin lattice, that is at the neck region of CCPs, similar to what has been observed for SNX9 by 3D-dSTORM microscopy (Schöneberg et al., 2017). Interestingly, the concentrations of early-acting factors, such as FCHo2 and its binding partner EPS15, decreased as CCPs matured into a spherical bud with a corresponding narrowing of the membrane neck. In contrast, the amounts of dynamin and the associated N-BAR protein amphiphysin at the neck increased with CCP maturation towards a spherical bud (Sochacki et al., 2017). Finally, several proteins, such as AP-2, DAB2, stonin2, arrestin, and intersectin, have been found both in the central and the peripheral zones of CCS (Sochacki et al., 2017).
Of note, the existence of proteins that are localized to the periphery of the major protein coat appears to be a general feature that, in addition to clathrin coats, also characterizes the caveolin–cavin coat, which is bordered by EHD2 complexes (Ludwig et al., 2013; Yeow et al., 2017) and the Gag-containing shells of budding HIV viruses, which are bounded by ESCRT-I complexes (Sundquist and Krausslich, 2012). In the two latter systems, the peripheral protein complexes appear to spatially limit coat polymerization (Sundquist and Krausslich, 2012; Yeow et al., 2017), in addition to other possible functions. A challenge for future research will be to understand whether the segregation of endocytic proteins into central and peripheral nanodomains is a general feature of all protein coats in cells and to determine whether the peripherally localized proteins execute functions that are common for different vesicle coats.
Timecourse of endocytic protein recruitment to sites of endocytosis in mammalian cells
The entire process of endocytic vesicle formation from its initiation to vesicle separation from the plasma membrane and uncoating can be completed within 60 to 120 s in most types of eukaryotic cells (Cocucci et al., 2012; Kadlecova et al., 2017; Nández et al., 2014; Taylor et al., 2011). The specific timecourse of the assembly and shape transformations of CCSs driven by endocytic proteins and lipids (Box 2) and the characteristic spatially modulated distribution of these proteins in the membrane plane imply a certain time order in which endocytic proteins are recruited to the endocytic site. Indeed, several recent studies have explored the dynamics of endocytic protein recruitment in mammalian cells (Bucher et al., 2018; Henne et al., 2010; Miller et al., 2015; Scott et al., 2018; Sochacki et al., 2017; Taylor et al., 2011). These works have demonstrated that, similar to what was shown in the seminal findings by Drubin and coworkers for yeast cells (Kaksonen et al., 2005), mammalian endocytic proteins can be grouped into distinct modules according to their recruitment kinetics (Taylor et al., 2011). Extensive measurements of the recruitment kinetics of 34 mammalian endocytic proteins by using dual-color total internal reflection fluorescence microscopy (TIRF) with respect to the timing of CCV scission (Taylor et al., 2011) have defined four major endocytic protein groups that may comprise functional modules in cells (Fig. 2C). The earliest group of proteins that accumulate at the endocytic site comprises the peripherally located proteins FCHo1/2 and ESP15 and their binding partners intersectin 1/2 and AP-2, with the latter found both at the periphery and in the center of the coat (Sochacki et al., 2017). The concentrations of these proteins reaches their maximal level ∼20 s before scission and decay towards the scission reaction (Taylor et al., 2011). According to a more detailed scenario (Henne et al., 2010), within this group of proteins, FCHo1/2 appear first and recruit Eps15 and the intersectins, which then cooperatively engage and activate the adaptor complex AP-2 (Hollopeter et al., 2014; Ma et al., 2016) through a large-scale conformational transition that aligns its lipid- and cargo-binding sites with the plasma membrane (Jackson et al., 2010).
The second group of proteins that is recruited with some delay comprises the endocytic factors found in the central part of CCSs, such as clathrin, CALM (or AP180 in neurons), epsin and NECAP1/2. These proteins attain their maximal presence close to the time point of the scission event and disappear right after it (Taylor et al., 2011).
Proteins of the third group arrive and reach their maximal concentration at the endocytic site relatively late, just a few seconds prior to scission, and disappear at or right after membrane scission. These comprise the N-BAR domain proteins endophilin and amphiphysin, as well as dynamin (Taylor et al., 2011). These factors predominantly localize to the coat periphery (Sochacki et al., 2017) and, at the time point of their maximal recruitment, the concentrations of the early-recruited peripheral proteins, such as FCHo2 and EPS15, are already low (Sochacki et al., 2017; Taylor et al., 2011). As a consequence, the proteins of the first and third groups, practically, do not co-exist at the CCS edge.
The fourth group of proteins, which nucleate and promote the polymerization of actin filaments, such as N-WASP, ARP2/3, and actin itself reach their maximal concentration close to fission (Taylor et al., 2011). According to a recent proposal, actin polymerization is initiated primarily at the base of endocytic CCPs by the N-WASP-associated F-BAR domain protein FCHSD2 to facilitate CCP maturation (Almeida-Souza et al., 2018). In addition, F-actin may aid membrane fission (Ferguson et al., 2009) and propel the newly formed endocytic vesicles away from the endocytic site (Nández et al., 2014). As the exact role and importance of actin may vary depending on cell type and conditions (e.g. membrane tension), we refer the reader to excellent recent reviews on the topic (Hinze and Boucrot, 2018; Kaksonen and Roux, 2018).
Finally, the fifth group of proteins that is recruited to CCSs comprises factors that are linked to vesicle uncoating, such as the J-domain protein auxilin, a co-factor for Hsc70-mediated ATP-driven clathrin disassembly, and inositol phosphate phosphatases, including oculocerebrorenal syndrome of Lowe (OCRL) and synaptojanin (Di Paolo and De Camilli, 2006; He et al., 2017; Kaksonen and Roux, 2018; Nández et al., 2014; Posor et al., 2015).
A speculative model for endocytic vesicle formation
How precisely endocytic membrane sites are progressively reshaped has remained largely enigmatic. Based on the recent data regarding the timing and spatiotemporal nanoscale distribution of endocytic proteins described above, we propose the following speculative model for endocytic vesicle formation that integrates information from cellular and in vitro experiments, as well as theoretical modeling.
According to the time sequence of endocytic protein recruitment to sites of CME (Henne et al., 2007; Taylor et al., 2011), the endocytic process is proposed to be initiated by the self-assembly of FCHo1/2 proteins (Henne et al., 2007) on the surface of the plasma membrane through their lateral and end-to-end interactions (Frost et al., 2008) (but see Cocucci et al., 2012 for a different view). In in vitro experiments using liposomes in the presence of high concentrations of FCHo2 that are sufficient to cover the entire available membrane surface (Frost et al., 2008; Henne et al., 2007), FCHo2 forms helical coats molding membranes into tubules of ∼130 nm diameter (Henne et al., 2010). It is reasonable to assume that at lower intracellular FCHo concentrations, arc-like FCHo coats are initially formed on the membrane surface (Fig. 3A), which can cover only a small portion of the plasma membrane and will be referred to below as FCHo arcs. The shapes of the FCHo arcs and of the membrane underneath is determined by the interplay between the bending rigidity of the membrane and the overall rigidity of the FCHo arc, with the latter depending on the arc thickness (the amount of proteins forming the arc). For a small FCHo polymer that forms an arc, the arc rigidity must be smaller than or comparable to that of the membrane. In this case, the FCHo arc is not expected to be able to effectively bend the membrane into a cylindrical shape, because of the internal membrane resistance to curving. As a result, the FCHo arc, likely, lies nearly flat on the membrane surface or may adopt a shape of a segment of an almost flat truncated cone with an internal edge diameter of 130 nm (Fig. 3A). The corresponding shape of the membrane in the vicinity of the arc must be flat or slightly curved (Fig. 3A). The internal edge of the FCHo arc recruits and activates AP-2 proteins, which become concentrated near the internal border of the FCHo arc. Activated AP-2 (Hollopeter et al., 2014) in turn binds clathrin and activates its polymerization. FCHo arcs may further polymerize into complete rings that direct clathrin polymerization into an almost flat cage that covers the membrane patch inside the ring (Fig. 3B), in accordance with Sochacki et al. (2017). The expected diameter of such clathrin arrays should, therefore, roughly correspond to the preferred internal diameter of the FCHo ring (∼130 nm), similar to what has been observed in recent experimental data (Bucher et al., 2018). The clathrin array formed within the FCHo ring then recruits epsin and CALM to the central part of the CCS (Sochacki et al., 2017), which insert their amphipathic helices into the clathrin-coated membrane to generate local isotropic curvature (Fig. 3C). The generation of membrane curvature via positive feedback facilitates the recruitment of additional curvature-sensitive epsin and CALM molecules, leading to the progressive invagination of the membrane within the FCHo ring into a dome-like shape, and further towards a closed spherical shape. This transformation requires additional polymerization of clathrin within the ring, which should be made possible by the fast exchange of clathrin triskelia between the clathrin cage and the surrounding cytoplasm (Avinoam et al., 2015; Wu et al., 2001).
In the course of epsin- and CALM-driven shaping of the central clathrin-coated part of the endocytic site, the FCHo ring remains at the periphery of the structure and, thereby, covers the forming membrane neck that connects the nascent bud to the surrounding membrane. The neck becomes gradually narrower during the evolution of the central part towards a closed spherical shape (Fig. 3D). Thinning of the neck makes it progressively unfavorable for FCHo proteins to bind to the membrane as it prefers the radius of membrane curvature to be ∼130 nm. We predict that the progressive incompatibility between neck diameter and curvature preference of the FCHo polymer drives its dissociation from the membrane neck. At the same time, as the neck becomes narrower, the binding of protein scaffolds that are more curved, such the PX-BAR and N-BAR domain proteins, SNX9/18, amphiphysin and endophilin, is favored (Fig. 3E). Eventually, these factors, possibly in association with dynamin, replace FCHo on the neck, thereby facilitating further constriction of the neck region to eventually give rise to dynamin-mediated scission and the release of a free CCV.
Predictions from the hypothetical model and outlook
Several predictions arise from this hypothetical model for endocytic vesicle formation. FCHo1/2 and associated factors, such as Eps1, Eps15R and intersectin 1, direct clathrin polymerization to the inside of the predicted FCHo arc, so that clathrin polymerization is initiated along the internal edge of the FCHo arc and propagates towards the center. Therefore, FCHo polymers may serve to spatially limit the polymerization of clathrin to an area with a calibrated diameter of ∼130 nm. With the recent advent of novel super-resolution and correlative microscopy imaging methods (Lehmann et al., 2015; Liu et al., 2018; Schöneberg et al., 2017; Sochacki et al., 2017), testing this prediction indeed seems experimentally feasible. The effective purpose of limiting clathrin polymerization to the inside of the predicted FCHo ring could be to allow the formation of a membrane neck that is free of a clathrin coat, which facilitates the binding and boosts the activity of fission proteins, such as endophilin and dynamin.
However, the FCHo–Eps15–intersectin complex also stabilizes clathrin coats by recruiting and activating AP-2. Our model therefore predicts that, depending on the cell type and the expression levels of other factors that can recruit clathrin or AP-2, loss of FCHo may result in an excessive growth of clathrin arrays, which leads to enhanced clathrin plaque formation, as reported previously (Ma et al., 2016; Mulkearns and Cooper, 2012), and/or to reduction of the stability of productive CCPs, as a consequence of a partial inactivation of AP-2. The latter effect can be expected, in particular, in cells where clathrin plaque formation, for reasons that remain to be fully understood, does not occur (e.g. BSC-1 cells) (Cocucci et al., 2012; Henne et al., 2010). The budding of endocytic vesicles from the inside of large clathrin-coated arrays is energetically costly owing to the presence of a clathrin coat at the neck region that would need to be removed prior to vesicle scission. Hence, loss of FCHo in either of the above scenarios would be expected to hamper the efficiency of endocytic vesicle formation in CME.
The above model further predicts a specific and crucial endocytic function for epsins and CALM or AP180 in generating curvature inside FCHo rings to enable CCS invagination. Therefore, loss of either of these endocytic structural proteins (Boucrot et al., 2012; Henne et al., 2010; Koo et al., 2015; Miller et al., 2015) would be expected to result in shallow, possibly less stable CCSs in cells.
Our final prediction is that progressive CCS invagination and the resulting constriction of the CCP neck that is derived from the rim area of flat or shallow CCPs will cause FCHo proteins to dissociate from endocytic structures and to be replaced at the CCP necks by BAR-domain proteins, such as SNX9, SNX18, amphiphysin and endophilin, which preferentially bind to highly curved membrane. Indeed, FCHo has been shown to dissociate from CCPs during late stages prior to membrane fission (Henne et al., 2010; Taylor et al., 2011). However, whether FCHo rings are retained under conditions of impaired curvature acquisition (e.g. owing to depletion of epsins and/or CALM or AP180) or neck constriction (e.g. loss of SNX9 or SNX18) has not been experimentally tested so far.
Moreover, we suggest that similar principles to those outlined here for membrane remodeling during CME may also govern other related membrane-trafficking processes, such as the formation of vesicles and tubules during the different pathways of clathrin-independent endocytosis, caveolar membrane invagination, constitutive secretion, and the formation and budding of vesicles and tubules at endosomes and the Golgi complex (Antonny, 2011; Bonifacino and Glick, 2004; Faini et al., 2013; Kozlov et al., 2014). The recent development and application of genome engineering together with quantitative super-resolution and correlative light and electron microscopy approaches should allow us to test the above predictions and address these exciting questions in the near future.
We wish to thank Dr Dmytro Puchkov (Electron microscopy facility of the FMP, Berlin) for providing the electron micrographs shown in Fig. 1.
Work in the authors' laboratories was supported by grants from the German Funding Agency Deutsche Forschungsgemeinschaft (DFG) (SFB958/A07 to V.H.), the Israel Science Foundation (grant 1066/15 to M.M.K.), and European Union consortium InCeM (to M.M.K.).
The authors declare no competing or financial interests.