Membrane dynamics in autophagosome biogenesis

Macroautophagy, hereafter referred to as autophagy, is the best characterized form of autophagy and has been implicated in a diverse set of physiological and pathophysiological conditions. Autophagy was for a long time considered to be a non-selective cell survival process, involving random sequestration of cytoplasmic material into forming autophagosomes, but it is now evident that autophagy also plays an essential quality control function by selective removal of damaged or dysfunctional organelles, as well as protein aggregates and pathogens (Stolz et al., 2014). The pathway initiates with the nucleation of a double-membraned structure (the phagophore, also called the isolation membrane), which expands to engulf cytoplasm into a double-membraned vesicle (autophagosome). Fusion of the autophagosome with lysosomes leads to degradation of the sequestered material and the recycling of degradation products into new macromolecules (Klionsky and Codogno, 2013; Lamb et al., 2013) (Fig. 1).

A large number of autophagy-related (Atg) proteins have been identified and found to be essential for both the non-selective and selective types of autophagy (see Box 1) (Klionsky et al., 2011). Recent progress in the field has made it clear that Atg proteins act together with general membrane trafficking components, such as coat proteins, soluble N-ethylmaleimide-sensitive factor (NSF) attachment protein receptor (SNARE) proteins and RAB proteins. Membrane shaping and remodeling are required at several steps of the autophagy pathway (for general principles on the generation of membrane curvature, see Box 2) and these processes appear to be tightly regulated by lipid modifying enzymes, membrane sculpting and remodeling proteins, lipid-binding effector proteins and protein kinases and phosphatases. This Commentary will focus on recent progress in our understanding of the early membrane modeling events in phagophore biogenesis, such as those involved in the nucleation process and the role of phosphatidylinositol (PtdIns) 3-phosphate (PtdIns3P) and other lipids, as well as later steps of phagophore expansion and closure, including tethering and fusion reactions to obtain the membrane composition and curvature of the functional autophagosome. Finally, we briefly discuss the membrane dynamics and modeling events occurring in selective autophagy. For detailed reviews of the complex hierarchy of Atg proteins and their regulation upon induction of autophagy, see Lamb et al., 2013; Mizushima et al., 2011 and Box 1.

The phagophore is a small cup-shaped membrane precursor formed upon induction of autophagy, and its origin and identity have been a matter of debate for decades (Tooze, 2013). Recent studies of Atg proteins in yeast and mammalian cells have contributed substantially to our understanding of the mechanisms involved in phagophore formation, but the constitutive lipid species and membrane modeling proteins involved in determining phagophore shape and size are still mostly unknown.

It is generally agreed that phagophores are formed de novo by nucleation on a preexisting membrane (for recent reviews, see Abada and Elazar, 2014; Hamasaki et al., 2013b; Lamb et al., 2013; Shibutani and Yoshimori, 2014). In yeast, the phagophore membrane has been found to originate from a single origin near the vacuole, called the pre-autophagosomal structure (PAS) (Suzuki and Ohsumi, 2010). In higher eukaryotes, phagophore nucleation might occur at different locations in the cell, but it is now largely accepted that, at least upon starvation, phagophores nucleate at an intricate membranous structure, termed the omegasome (due to its resemblance to the Greek letter omega in electron microscopy pictures). Omegasomes were originally identified as endoplasmic reticulum (ER)-associated spots labeled by the PtdIns3P-binding double-FYVE-containing protein 1 (DFCP1, also known as ZFYVE1) (Axe et al., 2008). The class III PtdIns3-kinase (PI3KC3, also known as PIK3C3; Vps34 in yeast) is required for omegasome formation (Axe et al., 2008) and forms an autophagy-specific complex with p150 (also known as PIK3R4; Vps15 in yeast), beclin 1 (Atg6 in yeast) and ATG14L (also known as BARKOR; Atg14 in yeast) (Itakura et al., 2008; Kihara et al., 2001; Sun et al., 2008) (see also Box 1).

DFCP1-positive omegasomes appear to concentrate at or near the connected mitochondria-associated ER membrane (Hamasaki et al., 2013a), in line with a study showing that the outer mitochondrial membrane provides lipids for the phagophore (Hailey et al., 2010). It is, however, clear that the phagophore also receives input from other membrane sources. The ER exit sites (ERES) (Graef et al., 2013; Zoppino et al., 2010), the ER–Golgi intermediate compartment (ERGIC) (Ge et al., 2013), the Golgi (Bodemann et al., 2011; Geng et al., 2010; van der Vaart et al., 2010), the plasma membrane (Ravikumar et al., 2010) and recycling endosomes (Knævelsrud et al., 2013; Longatti et al., 2012; Puri et al., 2013) have all been implicated as membrane sources for the phagophore (Fig. 2E). It has been difficult to differentiate between membranes contributing to phagophore nucleation and those contributing to phagophore expansion, but the recent elucidation of the time- and function-dependent hierarchy among the core autophagy proteins involved in phagophore nucleation has provided markers for the different steps of phagophore formation (Karanasios et al., 2013; Koyama-Honda et al., 2013), which might allow elucidation of the membrane source hierarchy. However, as most investigations have been performed in cells undergoing starvation-induced autophagy, it cannot be ruled out that, depending on the autophagy-inducing signal and the nature of the cargo to be sequestered, different membrane sources contribute to phagophore nucleation and/or expansion, as discussed below.

Several of the factors known to be crucial for phagophore nucleation have membrane-interacting domains that are thought to be important for their targeting to the appropriate membrane and/or their ability to mediate membrane shaping (see Table 1). Based on recent live-cell imaging and genetic and biochemical studies, we propose that the following events lead to phagophore nucleation upon induction of autophagy. Firstly, the mammalian ULK (UNC51-like) kinases and their associated complex partners [ATG13, FIP200 (also known as RB1CC1) and ATG101] are recruited to the membrane at the site of phagophore nucleation (for a recent review, see Wirth et al., 2013) (Fig. 2A). This might be mediated by the C-terminal domain of ULK (and its yeast ortholog Atg1), which is called the early autophagy targeting/tethering (EAT) domain, as it appears to be essential for recruitment to the site of phagophore nucleation (Chan et al., 2009; Ragusa et al., 2012). Moreover, a cluster of positive amino acids in the N-terminus of ATG13 has been found to interact with acidic phospholipids and to be important for the translocation of ATG13 to omegasomes (Karanasios et al., 2013). ULK1 has been found to be targeted to pre-existing structures containing the ER-resident multi-membrane spanning protein vacuole membrane protein 1 (VMP1) (Koyama-Honda et al., 2013). Secondly, PI3KC3 is recruited to phagophore nucleation sites to generate PtdIns3P (Fig. 2A). A recent study found that the N-terminal domain of yeast Atg13 forms a HORMA domain that is required for autophagy and for recruitment of the PI3KC3 subunit Atg14 to the PAS (Jao et al., 2013). ATG14L contains an ER-binding motif in its N-terminal domain, which appears to be essential for its function in autophagy and for the recruitment of the other PI3KC3 subunits (see Box 1) to the sites of phagophore formation (Matsunaga et al., 2010). An interaction between VMP1 and beclin 1 (Molejon et al., 2013) might further stabilize the association of the PI3KC3 complex with the ER membrane. Finally, the ATG12–ATG5–ATG16L1 complex (see Box 1) is recruited to the phagophore (Fig. 2A,B). ATG5 was found to be recruited to VMP1-positive structures concomitant with ULK1 (Koyama-Honda et al., 2013), which is rather surprising as the ATG12–ATG5–ATG16L1 complex has been placed downstream of the ULK1 and PI3KC3 complexes in the functional hierarchy of autophagosome formation. The synchronized recruitment of ATG5 with ULK1 might be mediated by a recently described interaction between the ULK complex subunit FIP200 and ATG16L1 (Gammoh et al., 2013; Nishimura et al., 2013). The VMP1–beclin 1 interaction was also found to facilitate the association of ATG16L1 and LC3 (also known as microtubule-associated protein 1 light chain 3) with the autophagosomal membranes (Molejon et al., 2013). As VMP1 expression is known to trigger autophagy (Ropolo et al., 2007) and because it seems to be the most ‘temporally’ upstream factor identified to date, it is tempting to hypothesize that VMP1 acts as a recruitment platform that determines the site of phagophore formation by recruitment of the early core ATG proteins. Further characterization of the VMP1-positive structures is required to understand why phagophores or omegasomes form from such spots, and why not all VMP1 spots appear to be involved in phagophore formation. The beclin-1-interacting transmembrane ER membrane protein complex subunit 6 (EMC6) might further contribute to the generation of the unique geometry of the phagophore, as it localizes to omegasomes and its depletion causes the accumulation of autophagosomal precursor structures and impaired autophagy (Li et al., 2013).

Although the initial recruitment of the early core autophagy components to sites of phagophore nucleation can act independently of PI3K activity (Itakura and Mizushima, 2010), PtdIns3P production appears to be important for their stable membrane association and further phagophore growth, as almost all ULK1, ATG5 and ATG16L1 punctae disappeared upon inhibition of PI3K activity (Karanasios et al., 2013; Koyama-Honda et al., 2013). PtdIns3P is an inverted cone-shaped membrane lipid that, when clustered, can be seen to create a cytosol-facing bud in the membrane, which in turn can create a platform for the binding of membrane-active proteins. In line with this, ATG14L has been found to sense membrane curvature through an amphipathic α-helix located in its C-terminal BATS domain (Fan et al., 2011), which might facilitate its recruitment to already curved structures and/or be important for further stabilization of phagophore curvature. Interestingly, the C-terminal EAT domain of Atg1/ULK1 also appears to sense membrane curvature, as it binds to liposomes in a geometry-dependent manner (Ragusa et al., 2012). Recently, ULK1 was found to activate the PI3KC3 complex by phosphorylation of beclin 1 (Russell et al., 2013), and the ULK complex subunit ATG13 was found to bind to PtdIns3P (Karanasios et al., 2013), indicating that the ULK1 complex plays an important role in PtdIns3P production and stabilization at the phagophore. Furthermore, the crystal structure of yeast Atg17, which forms an Atg1-interacting complex with Atg29–Atg31, showed that it forms a crescent-shaped dimer reminiscent of curvature-sensing and inducing Bin-Amphiphysin-Rvs (BAR) domains, suggesting that Atg17 contributes to the stabilization of membrane curvature and vesicle tethering at the PAS (Ragusa et al., 2012). It is not known whether ULK proteins or mammalian ATG17 have similar functions, but it is interesting to note that the beclin-1-interacting protein AMBRA1 (activating molecule in beclin-1-related autophagy 1) mediates ULK1 dimerization (Nazio et al., 2013) and that ULK1-mediated phosphorylation of AMBRA1 is important for proper localization of the PI3KC3 complex to sites of phagophore nucleation (Di Bartolomeo et al., 2010). Taken together, these studies strongly support a model whereby the PI3KC3 and ULK complexes are recruited to specific membrane spots in the ER membrane where they act in a synergistic manner to stimulate PtdIns3P production and facilitate the elongation and initial membrane curvature of the phagophore membrane.

In addition to a possible role for PtdIns3P as a cone-shaped lipid important for initial phagophore curvature, PtdIns3P facilitates the recruitment of autophagy proteins to the membrane (Fig. 2B). An important group of PtdIns3P-binding proteins are the PROPPINs (β-propellers that bind polyphosphoinositides), including the mammalian WD repeat domain phosphoinositide-interacting (WIPI) proteins and their yeast homolog Atg18 (Bakula et al., 2013), as well as yeast Atg21, which is involved in the cytoplasm-to-vacuole (Cvt) pathway (Krick et al., 2008) (for details on WIPI proteins, see Box 3).

A recent study shows that WIPI2 functions to recruit ATG16L1 to sites of phagophore formation (Dooley et al., 2014). A WIPI2-binding site in ATG16L1 that is different from the sites involved in binding to ATG5 and FIP200 has been identified. Interestingly, this WIPI2-binding site is absent from ATG16L2, an isoform of ATG16L1 that is not involved in autophagy, although it forms a complex with ATG5–ATG12 (Ishibashi et al., 2011). The ATG12–ATG5–ATG16L1 complex mediates the transfer of activated LC3 or GABARAP from the E2-like enzyme ATG3 to phosphatidylethanolamine (PtdEtn) and appears to determine the site of LC3 or GABARAP lipidation (Fujita et al., 2008b) (see Box 1; Fig. 2C). An amphipathic N-terminal helix in ATG3 was found to have curvature-sensing properties and to be necessary for its activity, suggesting that ATG3 might detect lipid-packing defects at the rim of the growing phagophore (Nath et al., 2014). This function would localize the lipidation reaction of LC3 or GABARAP to the highly curved surface at the edge of the growing phagophore where the addition of incoming vesicles is thought to occur (Fig. 2D). Therefore, the sequence of events leading to the production of lipidated LC3 or GABARAP at the phagophore could be explained by the following order of events: the ULK and PI3KC3 complexes are recruited to sites of phagophore nucleation, leading to PtdIns3P production and WIPI2 recruitment (as described above), followed by recruitment of the ATG12–ATG5–ATG16L1 complex, which again facilitates the transfer of LC3 or GABARAP from ATG3 to PtdEtn at the rim of the phagophore membrane. LC3/GABARAP-containing membrane might also be produced by ATG16L1 at other locations in the cell and transported in vesicular form to yield additional membrane for phagophore growth. One such source of membrane is recycling endosomes, which have been found to contain ATG16L1 and to contribute to the production of LC3/GABARAP-containing membrane by a process dependent on the PX-BAR protein sorting nexin 18 (SNX18) (Knævelsrud et al., 2013). In this context, ATG16L1 is likely to be recruited by other means than binding to WIPI2, as WIPI2 is not found on endosomes and the process was shown to be dependent on phosphatidylinositol 4,5-bisphosphate [PtdIns(4,5)P₂] rather than PtdIns3P.

PtdIns3P levels at the phagophore appear to be tightly regulated by phosphoinositide 3-phosphatases. Both MTMR14 (myotubularin-related protein 14) and MTMR3 were identified as negative regulators of autophagy that act by preventing recruitment of WIPI1 (WIPI2 not investigated) to phagophores (Taguchi-Atarashi et al., 2010; Vergne et al., 2009). Interestingly, smaller autophagosomes were seen in cells with overexpression of MTMR3 (Taguchi-Atarashi et al., 2010), indicating that PtdIns3P levels at the phagophore regulate autophagosome size. A likely explanation for this was provided by studies in yeast, showing that PtdIns3P hydrolysis (by the MTMR-related phosphatase Ymr1) is required for the dissociation of core Atg proteins before closure of the phagophore membrane to form an autophagosome (Cebollero et al., 2012).

The roles of phospholipids and sphingolipids as signaling molecules for the regulation of autophagy are subjects of great current interest (for reviews, see Dall'Armi et al., 2013; Knævelsrud and Simonsen, 2012), but the influence of membrane lipid species on the structure and/or function of phagophores and autophagosomes has not been thoroughly studied. Owing to technical difficulties in the biochemical isolation of autophagic membranes in pure form, the exact lipid composition of these membranes is not known. As the interplay between constituent membrane lipids and proteins is vital for the behavior and biological function of a membrane (Sprong et al., 2001), future research into autophagy ‘lipidomics’ will likely provide novel insights into the processes of phagophore formation and growth. Knowledge about the membrane lipids that build up the autophagosomes is crucial for the interpretation of in vitro reconstitution experiments with purified proteins and liposomes that are increasingly being performed.

A role for the PI3KC3 complex in the initial phagophore formation is very convincing, but the class II PI3K (PI3KC2, also known as PIK3C2) was also found to regulate autophagosome biogenesis by contributing to PtdIns3P generation at autophagosome nucleation sites (Devereaux et al., 2013). Apart from PtdIns3P, no specific lipid has been identified that marks the phagophore nucleation site or autophagic membranes in general. However, other phosphoinositides might come into play during phagophore elongation, as shown for phosphatidylinositol 4-phosphate (PtdIns4P) (Wang et al., 2012), PtdIns(4,5)P₂ (Knævelsrud et al., 2013) and phosphatidylinositol 3,5-bisphosphate [PtdIns(3,5)P₂] (Ho et al., 2012). Although these lipid species are mainly viewed as recruitment platforms for autophagy-related proteins in this context, they might, in addition, influence the structure and geometry of the membrane when clustered, owing to their inverted cone shape.

Cone-shaped membrane lipids have a propensity to form negative membrane curvature, such as that required for the sharp negative curvature of the lumenal membrane leaflet at the edges of the growing phagophore. One such lipid, phosphatidic acid, might have a role in autophagy, as the phosphatidic-acid-producing enzyme phospholipase D1 was found to localize, in a PtdIns3P-dependent manner, to autophagosome-related structures upon starvation (Dall'Armi et al., 2010). Other specific lipids, such as sphingosine-1-phosphate and ceramide, are also suggested to have roles in autophagy (Dall'Armi et al., 2013; Knævelsrud and Simonsen, 2012), but it is not yet known whether these lipids are utilized for the recruitment of other factors or whether they have a role in the generation of membrane structure.

PtdEtn is the co-substrate for the lipidation reaction of LC3 or GABARAP (or the yeast homolog Atg8) and is required in the membrane at the nucleation site and elsewhere (Mizushima et al., 2011). PtdEtn is produced from phosphatidylserine (PtdSer) in mitochondria, a fact that has been used as the basis of the argument that mitochondria might supply membrane for autophagosome biogenesis (Hailey et al., 2010). However, PtdEtn is also shuttled from mitochondria to ER, through mitochondria–ER contact sites, and can be delivered efficiently from there to other organelles (Rowland and Voeltz, 2012). The plasma membrane and recycling endosomes also contain PtdEtn, and the latter organelle is, in addition, rich in PtdSer (Gagescu et al., 2000; Taguchi, 2013). PtdSer is highly exposed on the cytosolic side of the recycling endosome membrane, possibly by the action of a flippase (Uchida et al., 2011), and has been described as being important for endosomal trafficking in yeast and higher eukaryotes (Uchida et al., 2011; Xu et al., 2013). It is not yet known whether PtdSer is involved in the biogenesis of autophagosomes.

Autophagosomal membranes appear to be rich in unsaturated lipids (Reunanen et al., 1985), and a recent study indicated that the nature of the hydrocarbon chains of membrane lipids influences autophagy (Ogasawara et al., 2014). Inhibition of a stearoyl-CoA desaturase resulted in the suppression of autophagy at an early stage, indicating that unsaturated fatty acids are required for autophagosome formation. Genetic depletion of the corresponding enzyme in Drosophila also affects autophagy (Köhler et al., 2009). These findings point to the possibility that membrane fluidity is an important factor in the intricate dynamics and curvature of the autophagic membrane.

The size of autophagic cargoes varies from small precursor aminopeptidase I (prApe1) particles of the Cvt pathway in yeast (∼150 nm) (Jin and Klionsky, 2014) to large bacteria and mitochondria (up to several micrometers) in selective autophagy (Stolz et al., 2014). The process of phagophore elongation and closure must therefore be flexible and regulated in a cargo-specific manner.

Elucidating the localization and trafficking of ATG9, the only transmembrane protein that is recruited to the omegasome and that is required for autophagy (and that of yeast Atg9 to the PAS), might provide insight into the process of phagophore elongation (Mari and Reggiori, 2007; Webber and Tooze, 2010; Zavodszky et al., 2013). The majority of yeast Atg9 exists on highly dynamic Golgi-derived vesicles with a diameter of 30–60 nm that appear to be generated upon induction of autophagy (Mari et al., 2010; Yamamoto et al., 2012). These vesicles then accumulate at the PAS in an Atg1-dependent manner, where Atg1-mediated phosphorylation of Atg9 further facilitates the recruitment of Atg8 and Atg18 and subsequent phagophore expansion (Papinski et al., 2014). Induction of autophagy also leads to the dispersal of mammalian ATG9 from the Golgi into tubo-vesicular compartments, including recycling endosomes, in an ULK1-dependent manner (Orsi et al., 2012; Young et al., 2006). Moreover, the N-BAR-containing protein Bif-1 (Bax-interacting factor 1, also known as endophilin B1) (Takahashi et al., 2011) and binding of its interaction partner UVRAG (ultraviolet irradiation resistance-associated gene) (Takahashi et al., 2007) to PtdIns3P appear to regulate the transport of ATG9 from the Golgi to the phagophore (He et al., 2013). ATG9 was recently found to traffic through the plasma membrane into recycling endosomes, where it colocalizes with ATG16L1 (Puri et al., 2013). Vesicular or tubular trafficking from recycling endosomes might feed ATG9- and ATG16L1-positive membrane onto the growing phagophore in a process that appears to be regulated by the PX-BAR protein SNX18 and by the RAB11 effector protein TBC1D14 in an opposite manner (Knævelsrud et al., 2013; Longatti et al., 2012). Interestingly, yeast Atg9 was found to be embedded in the outer membrane of the autophagosome, and, after fusion with the vacuole, new Atg9-positive vesicles bud off from the vacuolar membrane (Yamamoto et al., 2012). This is in contrast to mammalian ATG9, which associates only transiently with the forming phagophore (Orsi et al., 2012), but these differences could also be explained by methodological and/or model system differences. The exact function of ATG9 is not known, but being a multi-spanning integral membrane protein, it is possible that it directly or indirectly participates in the formation of phagophore curvature, for example, by wedging of the membrane. It is also possible that ATG9 could function as a lipid transfer protein.

In addition to ATG9 and ATG16L1, recycling-endosome-derived membrane might contain membrane-anchored LC3 or GABARAP proteins (Knævelsrud et al., 2013) As several transport pathways converge at recycling endosomes, such trafficking would efficiently supply phagophores with membrane of the correct composition. LC3-containing membranes formed at other cellular locations could also be sources for phagophore elongation, and, in a recent study, the ERGIC membrane was found to stimulate LC3 lipidation both in vitro and in vivo (Ge et al., 2013). It is tempting to speculate that phagophore-bound proteins that have a LC3-interacting region (LIR) could mediate the docking of such incoming LC3-containing membranes. The ULK1 complex subunits ULK1, ATG13 and FIP200 all have an LIR motif that preferentially binds to GABARAP family proteins in vitro (Alemu et al., 2012). Although the ULK1 complex functions in early phagophore nucleation, recent studies have found that the LIR motifs of ULK1 and Atg1 are not required for this process but rather for their recruitment into autophagosomes and their colocalization with WIPI2- and LC3B-positive structures (Alemu et al., 2012; Kraft et al., 2012; Nakatogawa et al., 2012). Thus, the interaction of Atg1/ULK1 with Atg8 homolog proteins might facilitate the recruitment of membrane to the forming phagophore.

What are the mechanisms involved in the tethering and fusion of incoming membrane to the phagophore, and how is the curvature of the phagophore obtained? As discussed above, the class III PI3K and ULK complexes might contribute to the formation and stabilization of phagophore curvature, through production of PtdIns3P, but also as membrane-binding proteins with curvature-generating properties. Recent structural and functional analysis of the yeast Atg1 complex (also containing Atg17, Atg29 and Atg31) indicates that it forms a crescent-shaped dimer, where the Atg1 EAT domain is able to bind to and tether highly curved 20–30 nm vesicles, corresponding to the size of Atg9-positive vesicles (Ragusa et al., 2012). This, together with the findings that an interaction between Atg17 and Atg9 is required for Atg9 localization to the PAS (Sekito et al., 2009), that Atg17–At29–Atg31 can regulate the activity of the Atg1 EAT domain (Ragusa et al., 2012) and that Atg1-mediated phosphorylation of Atg9 is required for phagophore expansion (Papinski et al., 2014), indicate that the Atg1 complex is involved in the recruitment and tethering of Atg9 vesicles to the phagophore. Moreover, Atg11, a scaffold protein required for the recruitment of Atg9 to the PAS, was found to tether Atg9 vesicles (Lipatova et al., 2012), and Atg9 itself might facilitate transport of the TRAPP (transport protein particle) vesicle tethering complex to the PAS (Kakuta et al., 2012). The yeast SNARE proteins Sso1, Sso2, Sec9 and Tlg2 have all been implicated in the early steps of autophagy (Nair et al., 2011) and might mediate the fusion of these tethered Atg9 vesicles. In mammalian cells, small ATG16L1 vesicles accumulate upon depletion of the endosomal SNARE molecules VAMP7, syntaxin 7, syntaxin 8 and VT1B (also known as VTI1B or vesicle transport through interaction with t-SNAREs homolog 1B) (Moreau et al., 2011). The SNARE protein syntaxin 17 (STX17) appears to play a dual role in autophagy, being an upstream effector for ATG14L that is required for phagophore formation at the ER–mitochondria contact sites (Hamasaki et al., 2013a), but that is also important for the fusion of autophagosomes with lysosomes through the association of a C-terminal hairpin structure of two tandem transmembrane domains in STX17 with the outer membrane of completed autophagosomes (Itakura et al., 2012; Jiang et al., 2014). The latter is supported by the interaction of STX17 with SNAP-29 and the endosomal/lysosomal SNARE VAMP8 (Itakura et al., 2012), as well as the homotypic fusion and vacuole protein sorting (HOPS) tethering complex (Jiang et al., 2014).

Several members of the RAB (Ras-related protein) family of small GTPases have been implicated in autophagy. RAB proteins cycle between an active GTP-bound state and an inactive GDP-bound state and act by recruiting effector proteins to mediate trafficking between different compartments (Stenmark, 2009). RAB1 regulates vesicular transport between the ER and Golgi, and was recently found to be important for omegasome formation (Mochizuki et al., 2013). The yeast TRAPPIII complex acts as an autophagy-specific guanine nucleotide exchange factor (GEF) for Ypt1 (the yeast homolog of RAB1) (Lynch-Day et al., 2010). TRAPPIII is recruited to the PAS by Atg17, and the activated Ypt1 stimulates the recruitment of Atg1 to the PAS (Wang et al., 2013), suggesting that TRAPPIII and Ypt1/RAB1 might facilitate the tethering of Atg9 vesicles. Moreover, the conserved oligomeric Golgi (COG) complex, another Ypt1 effector, localizes to the PAS and is required for proper localization of Atg8 and Atg9 (Yen et al., 2010). Interestingly, the TRAPPIII complex interacts with a subunit of the coat protein complex (COP)II, suggesting that COPII vesicles might provide membrane for phagophore formation and/or elongation (Tan et al., 2013). It is not known whether TRAPP complexes are involved in autophagy in mammalian cells, but as RAB1 and functional ER exit sites (ERES) have been implicated in autophagosome biogenesis (Graef et al., 2013; Zoppino et al., 2010) it is likely that ERES-derived COPII-coated vesicles also provide membrane to the growing phagophore in a TRAPP-dependent manner. The Golgi-resident protein RAB33B interacts with ATG16L1 and modulates LC3 lipidation (Itoh et al., 2008), but has also been implicated in the fusion of autophagosomes with lysosomes (Itoh et al., 2011). Thus, further studies are needed to fully understand RAB33B-mediated membrane fusion events in autophagy. The endosomal RAB proteins RAB4 (Talaber et al., 2014), RAB5 (Ravikumar et al., 2008), RAB7 (Gutierrez et al., 2004; Jäger et al., 2004; Pankiv et al., 2010; Tabata et al., 2010) and RAB11 (Fader et al., 2008; Knævelsrud et al., 2013; Longatti et al., 2012; Puri et al., 2013) have all been shown to play a role in autophagy. However, as endosomes have a dual role in autophagy, because they fuse with autophagosomes after their closure and, at the same time, provide membrane input to the forming phagophore, it is difficult to interpret the function of the different endosomal RAB proteins in autophagy (for recent reviews, see Szatmári and Sass, 2014; Tooze et al., 2014).

No typical ‘coat’ structure is visible in electron microscopy images, and it is possible that hitherto unknown curvature-generating proteins act to remodel and stabilize the autophagic membrane into the correct dimension. It is conceivable that Atg8 homolog proteins themselves participate in forming the geometry of the autophagosome. LC3 lipidation and accumulation at the membrane surface might create a crowding effect that leads to membrane bending (LC3-II can indeed be envisioned as a membrane lipid with a very large head group). In such a scenario, curvature might be the result of membrane tension caused by the unbalanced distribution of LC3-II between the concave and the convex side of the double bilayer. Moreover, as PtdEtn, the target of LC3 lipidation, is a cone-shaped lipid, it might further contribute to the membrane curvature and possibly to the regulation of autophagosome size. Indeed, in experiments where the level of Atg8 was manipulated, the autophagosome size was found to be affected (Xie et al., 2008). LC3-mediated recruitment of cargo-bound autophagy receptors to the concave surface of the phagophores might also influence the curvature of the expanding phagophore. Several such cargo-specific receptors have been identified that all possess LIR sequences that bind to LC3 and GABARAP proteins (Rogov et al., 2014). LC3-II molecules that are trapped inside the autophagosome together with receptors and cargo are eventually degraded, whereas LC3-II on the outer convex membrane is recycled to the cytosol by the action of the cysteine protease ATG4 (Kabeya et al., 2004). Little is known about the specific roles of the individual Atg8 homolog proteins in autophagy, but one study found that LC3 is important for phagophore elongation, whereas GABARAPL2 functions at a later stage of the process (Weidberg et al., 2010). LC3 family proteins have also been implicated in the fusion reaction that is required to seal the membrane (Fujita et al., 2008a; Weidberg et al., 2010).

In vitro studies with liposomes and purified autophagy proteins have suggested additional functions of the essential ATG12–ATG5–ATG16L1 complex. Yeast Atg5 was shown to have affinity for negatively charged lipids, and to mediate the tethering of membrane vesicles independently of Atg8 (Romanov et al., 2012). Such a mechanism might be at work in the early fusion of incoming vesicles, as suggested for Atg9-containing vesicles (Mari et al., 2010; Yamamoto et al., 2012). Another study, also using yeast proteins, showed that once lipidated Atg8 is formed, it can anchor the Atg12–Atg5–Atg16 complex to the membrane through a specific Atg8–Atg12 interaction, leading to the assembly of a ‘coat-like’ scaffold at the convex side of the phagophore (Kaufmann et al., 2014). At the concave side, binding of cargo or cargo adaptors through Atg8–LIR-sequence interactions would outcompete the binding of Atg8 to Atg12, providing an explanation as to why Atg16–Atg12–Atg5 is exclusively found on the outer phagophore membrane (Mizushima et al., 2011). It remains to be determined whether these interesting observations with purified proteins indeed reflect mechanisms that operate in the cell.

When it comes to selective autophagy, it is likely that the cargo and its bound autophagy receptors contribute to phagophore expansion and curvature. A recent study found that the Cvt vesicle cargo prApe1 facilitates the interaction of the cargo receptor Atg19 with Atg8 in the membrane (Sawa-Makarska et al., 2014). Atg19 contains multiple Atg8-binding motifs in its C-terminus, whereas its paralog Atg34 contains only one Atg8-binding motif. Using an in vitro reconstituted system with giant unilamellar vesicles (GUVs) and prApe1 pro-peptide beads, it was found that the Atg8-positive GUV membrane closely wrapped around the beads in the presence of Atg19 and that this correlated with the density of Atg8 and the Atg8-binding activity of Atg19. A chimeric Atg19–Atg34 protein with one Atg8-binding site was unable to support prApe1 processing through the Cvt pathway, although binding to the prApe1 cargo was intact. In contrast, prApe1 processing was nearly normal upon rapamycin-induced bulk autophagy, leading the authors to suggest a model in which close membrane–cargo apposition (termed ‘exclusive autophagy’) requires the binding of Atg19 to several Atg8 molecules, whereas one Atg8-binding site in Atg34 (or the Atg19–Atg34 chimera) is sufficient for selective but not exclusive sequestration of prApe1.

A similar principle is likely to apply to other forms of selective autophagy. It is interesting to note that although the binding affinity between the LIR domain of the ubiquitin-binding autophagy receptor p62 (also known as SQSTM1) and LC3 or GABARAP is rather low (Maruyama et al., 2014), the avidity of the interaction is high, owing to PB1-domain-mediated self-polymerization of p62 (Bjørkøy et al., 2005). This likely contributes to the exclusive p62-mediated sequestration of cargo (e.g. protein aggregates). The p62-interacting adaptor protein ALFY (also known as WDFY3) has an LIR motif with a greater than tenfold higher affinity for GABARAP proteins than that of the p62 LIR (Lystad et al., 2014). Binding of ALFY to GABARAP facilitates the recruitment of LC3B to ALFY and p62-positive structures, further indicating that a high-affinity LIR–Atg8 protein interaction is required for exclusive autophagy. In line with this, interaction of the ubiquitin-binding autophagy receptor NDP52 (also known as CALCOCO2) with LC3C was found to be required for the recruitment of other Atg8 family members to cytosolic Salmonella Typhimurium (von Muhlinen et al., 2012). Taken together, these studies suggest that an initial interaction between an Atg8 protein and a LIR-containing receptor forms a platform for further phagophore expansion and cargo sequestration. Several key questions remain to be addressed, including whether the phagophore membrane of phagophores involved in selective autophagy also nucleates from ER-associated omegasomes.

Little, if anything, is known about the mechanisms involved in the closure of the phagophore membrane. The phagophore is a double-membraned structure, which infers that its closure involves the fusion of a narrow opening, different from most other membrane fusion events. The topology of the phagophore is similar to that of multivesicular bodies (MVB) formed upon invagination of the early endosome membrane, viral budding and cytokinesis, which all rely on the endosomal sorting complex required for transport (ESCRT) (Rusten et al., 2012). The ESCRTs and their associated proteins facilitate membrane budding away from the cytosol and subsequent cleavage of the bud neck (Hurley and Hanson, 2010). Several studies have shown that depletion of ESCRT subunits, as well as of their regulatory ATPase Vps4, causes an accumulation of autophagosomes (Filimonenko et al., 2007; Lee et al., 2007; Rusten et al., 2007), but it is not clear whether this is due to ESCRTs being required for autophagosome closure or autophagosome–endosome fusion. It is interesting to note that several ESCRT subunits are PtdIns3P-binding proteins. PtdIns3P turnover by the PtdIns3P phosphatases (MTMRs, Ymr1) was found to mediate the dissociation of the early core ATGs from the phagophore membrane and to be required for the final closure of the phagophore to form an autophagosome. A tight regulation of PtdIns3P levels at the phagophore also appears to regulate autophagosome size (Taguchi-Atarashi et al., 2010). Thus, one can speculate that ESCRTs are recruited to closing phagophores through interactions with PtdIns3P, and that a tight regulation of PtdIns3P levels in the highly curved membrane at the remaining opening controls the final phagophore closure. In line with such a model, loss of Atg2, which forms a complex with the PtdIns3P-binding protein Atg18 was found to inhibit autophagosome closure (Velikkakath et al., 2012). Several ESCRT subunits interact with ubiquitin (Shields and Piper, 2011), and a recent study found that interaction of the ESCRT-III subunit vacuolar protein sorting 2.1 (VPS2.1) with the deubiquitylating enzyme AMSH1 is important for autophagy in Arabidopsis. Both AMSH1 and VPS2.1 mutants accumulate autophagosome markers and have fewer autophagic bodies in the vacuole (Katsiarimpa et al., 2013). It is not known whether ubiquitin is recycled before autophagosome closure, as is the case in the endocytic pathway, and future studies are needed to address this. Consistent with such a model, the ubiquitin-like proteins LC3 and Atg8, being conjugated to both sides of the phagophore membrane, are cleaved off the outer membrane by the activity of the Atg4 protease (Kabeya et al., 2000).

Our understanding of the membrane modeling events involved in autophagosome biogenesis has increased immensely during the last few years. However, the constitutive lipid species and membrane modeling proteins that are involved in determining phagophore curvature and autophagosome size are still mostly unknown and several fundamental questions remain to be answered. Where is the membrane coming from? What are the mechanisms involved in the regulation and execution of membrane tethering and fusion? How is the final closure mediated and regulated? Investigations of phagophore membrane properties and biogenesis in the near future are likely to reveal novel principles in membrane biology.

We thank Carina V. S. Knudsen for help with the figures.