Reconstruction of destruction – in vitro reconstitution methods in autophagy research

Eukaryotic cells have evolved a highly conserved and sophisticated degradation and recycling system for damaged or superfluous organelles, termed macroautophagy (hereafter referred to as autophagy) (Dikic and Elazar, 2018). This system requires cargo selection in a spatiotemporally defined and regulated fashion (Khaminets et al., 2016), and the degradation of cargo with very different sizes and shapes, demanding versatility. The complexity of these requirements is reflected in the complexity of autophagy, which involves the coordinated action of more than 40 dedicated autophagy-related (Atg) proteins (Mercer et al., 2018) that collaborate and are intertwined with a large number of cellular signaling, trafficking, and stress response pathways (Davis et al., 2017).

Autophagy was first characterized at a morphological and ultrastructural level as a pathway that delivers cytoplasm to lysosomes (Ashford and Porter, 1962; De Duve, 1963), and seminal work identified a set of Atg proteins through a genetic screen in yeast (Tsukada and Ohsumi, 1993), followed by related screens that expanded the repertoire of Atg proteins (Harding et al., 1995; Thumm et al., 1994). The identification of human homologs of yeast Atg proteins established that autophagy is conserved in higher eukaryotes (Kabeya et al., 2000; Mizushima et al., 1998). Furthermore, autophagy is involved in the onset of various human diseases, such as cancer and neurodegeneration, which has revealed the importance of the pathway in maintaining cellular homeostasis under normal conditions and in response to cytotoxic stress (Liang et al., 1999; Ravikumar et al., 2002).

Although yeast and human autophagy pathways are conserved, there are significant mechanistic differences, and the human autophagic machinery is much more elaborate compared to its yeast counterpart. In yeast, autophagy is initiated at the phagophore assembly site (PAS) at the vacuole, and involves activation, assembly and recruitment of the Atg1 kinase complex (Fig. 1A) (Cheong et al., 2005; Kabeya et al., 2005, 2009). Following its activation, the Atg1 kinase complex recruits vesicles that contain the membrane protein Atg9 (hereafter Atg9 vesicles) to the PAS (Fig. 1A) (Mari et al., 2010; Rao et al., 2016; Sekito et al., 2009). These vesicles fuse to give rise to a precursor membrane, termed a phagophore (Nair et al., 2011; Yamamoto et al., 2012). The expansion of these membrane cisternae allows cytoplasm to be sequestered and enclosed. The high demand of lipids that are needed for phagophore expansion requires a constant supply with membranes of various origins including the ER, the Golgi, the ER–Golgi intermediate compartment and mitochondria (Graef et al., 2013; Meiling-Wesse et al., 2005; Ohashi and Munro, 2010; Reggiori et al., 2005). Moreover, the early phagophore is enriched in phosphatidylinositol 3-phosphate [PtdIns(3)P] through the action of an autophagy-specific class III phosphatidylinositol 3-kinase (PI3K) complex (Fig. 1B). The recruitment of this lipid kinase to the PAS is mediated by an interaction of Atg13 with the PI3K subunit Atg14 (Jao et al., 2013; Suzuki et al., 2007).

Enrichment of the phagophore with PtdIns(3)P allows PtdIns(3)P-binding proteins, such as the yeast β-propellers that bind phosphoinositides (PROPPIN)-repeat-containing Atg18 and Atg21, to be recruited (Nascimbeni et al., 2017). Atg21 initiates the transition from phagophore initiation to phagophore expansion by recruiting Atg16 (Juris et al., 2015), whereas Atg18, in conjugation with its interaction partner Atg2, stimulates phagophore expansion by establishing ER–phagophore contact sites (Gómez-Sánchez et al., 2018). The expansion of the phagophore membrane is regulated by a ubiquitin (Ub)-like conjugation machinery that catalyzes the covalent attachment of the Ub-like protein Atg8 to phosphatidylethanolamine (PE) within the phagophore (Fig. 1B) (Ichimura et al., 2000). This reaction is initiated by an Atg4-dependent priming of Atg8 through proteolytic cleavage of its C-terminal residue (Kirisako et al., 2000). The conjugation of Atg8 itself involves the sequential activity of the E1-like Atg7 and the E2-like Atg3 enzymes as well as the E3-like ligase complex Atg12–Atg5–Atg16 (Geng and Klionsky, 2008). Apart from its function during phagophore expansion, Atg8 also coordinates selection of cargo through its interaction with cargo receptors such as Atg19 or Atg32 (Fig. 1C). Moreover, Atg8 promotes sealing of phagophores and fusion with the yeast vacuole (Fig. 1D,E) (Abeliovich et al., 2000; Kaufmann et al., 2014; Kondo-Okamoto et al., 2012; Nakatogawa et al., 2007; Noda et al., 2010; Suzuki et al., 2010; Xie et al., 2008). Shortly before fusion of the two organelles occurs, the cytosolic accessible pool of Atg8 is cleaved from the membrane by Atg4 (Fig. 1D) (Abreu et al., 2017; Kirisako et al., 2000; Yu et al., 2012). This ‘uncoating reaction’ is followed by tethering of autophagosomes to the vacuole through the homotypic fusion and vacuole protein sorting (HOPS) complex and SNARE-mediated fusion of the outer autophagosomal and the vacuolar membranes (Fig. 1E) (Nair et al., 2011; Rieder and Emr, 1997). Although autophagy is well understood at a molecular level in yeast, mechanistic insights into the process as well as its spatiotemporal regulation are still limited. The core autophagy machinery is highly conserved and close homologs of the yeast proteins are expressed in human cells (Yang et al., 2017) (see Box 1). Significant mechanistic differences between the pathways in these species reflect regulatory requirements that are needed for cooperation of cells in multicellular organisms (Boya et al., 2018) as well as autophagy-unrelated functions of ATG proteins in higher eukaryotes (Schaaf et al., 2016).

Many insights into the nucleation of autophagosomes, the selection of cargo and its tethering to autophagic membranes as well as autophagosome completion and fusion with lysosomes have been revealed by in vitro reconstitution methods. The advantages of in vitro approaches in characterizing complex biological pathways were first recognized by Günter Blobel, who developed a cell-free system to recapitulate the translocation of proteins across the ER membrane (Blobel and Sabatini, 1971). A similar approach was used in seminal studies by Randy Schekman, which led to the identification of components of the secretory pathway (Rexach and Schekman, 1991). Advances in membrane physics and recombinant protein production made it possible to reconstitute entire biological pathways from purified components in the test tube. Here, we highlight such reconstitution approaches and summarize how they contributed to our current understanding of autophagy in yeast and humans.

The coordinated activity of the Atg1 (Ulk1 in mammals) kinase complex, the PI3K complex and the Atg9 compartment is required to initiate autophagy in yeast and humans. The two subunits Atg17 and Atg11 are major scaffolds for the assembly of two distinct Atg1 kinase complexes that function in selective (Atg11) or non-selective (Atg17) autophagy in yeast (Kabeya et al., 2005, 2009; Kamber et al., 2015; Yorimitsu and Klionsky, 2005). Moreover, both proteins recruit the transmembrane protein Atg9 to the PAS, linking autophagy initiation to nucleation of the phagophore (He et al., 2006; Sekito et al., 2009). First insights into membrane-binding properties of the Atg1 kinase complex came from in vitro reconstitutions of a minimal complex, comprising Atg17, Atg31 and Atg29, as well as fragments of Atg1 and Atg13 (Ragusa et al., 2012). In experiments using Folch liposomes with different diameters as membrane substrate (see Box 2), the C-terminal domain of Atg1 was shown to bind to membranes in a curvature-dependent manner, preferring liposomes with high curvature that corresponds to vesicle diameters of ∼30 nm (Ragusa et al., 2012). Atg9 vesicles that are recruited by the Atg1 kinase complex in vivo possess similar diameters, suggesting that curvature sensing by Atg1 contributes to the selective recruitment of these vesicles.

Furthermore, mutational analysis and immunoprecipitation suggested that both Atg17 and Atg13 bind Atg9 (Sekito et al., 2009; Suzuki et al., 2015), although which of the two proteins directly interacted with Atg9 was unclear. This was addressed by in vitro reconstitutions using artificial Atg9 vesicles and purified subunits of the Atg1 kinase complex: Atg9 vesicles were generated by incorporating recombinant and purified Atg9^core, which contains the highly conserved transmembrane region of Atg9 but lacks the unstructured N- and C-termini, into small unilamellar vesicles (SUVs) (see Box 2). Floatation experiments of these vesicles with purified Atg1 kinase complex subunits (see Box 3) revealed that Atg17 physically interacts with Atg9, suggesting that the binding of Atg9 by the Atg1 kinase complex mainly depends on Atg17 (Rao et al., 2016). Furthermore, the crystal structure of the Atg17–Atg31–Atg29 subcomplex revealed two interesting properties: first, that Atg17 is a highly elongated crescent-shaped molecule that forms tail-to-tail dimers; second, that Atg31 directly binds Atg17 and both Atg31 and Atg29 occupy the central region of the Atg17 crescent (Ragusa et al., 2012). A plausible explanation for this unusual architecture was, again, revealed by reconstitution approaches (Rao et al., 2016). The Atg17 dimer was found to bind two Atg9 vesicles independently and electron microscopy revealed that these vesicles were tightly tethered to each other (Fig. 1A). The Atg17–Atg31–Atg29 subcomplex was, however, not able to tether such vesicles, although the subcomplex retained a limited capacity to bind Atg9. This suggests that the prominent localization of Atg31 and Atg29 that has been observed in the crystal structure inhibits binding of Atg17 to Atg9.

Upon induction of autophagy, the pentameric Atg1 kinase complex assembles at the PAS in vivo (Kabeya et al., 2005). In agreement with this observation, interaction studies with purified components revealed that this complex binds Atg9 and tethers Atg9 vesicles in vitro (Fig. 1A) (Rao et al., 2016). The combination of in vivo experiments, crystal structures and in vitro reconstitution of the Atg1 kinase complex thus provided a detailed model for its function during autophagy initiation. However, the high demand of lipids for nucleation and expansion of autophagosomes cannot be fulfilled by just two Atg9 vesicles, suggesting that the early-stage PAS contains more than one Atg1 kinase complex and multiple tethered Atg9 vesicles. A combinatorial in vitro and in vivo approach revealed that Atg13 has two Atg17-binding sites. One of these sites activates Atg17–Atg31–Atg29 within the same complex whereas the other binding site links two Atg1 kinase complexes together by binding Atg17 of another complex (Yamamoto et al., 2016). This crosslinking allows the formation of larger super-molecular assemblies and thus provides the means to recruit and tether multiple Atg9 vesicles in order to nucleate the phagophore. This model is supported by fluorescence-based quantifications in vivo, which demonstrated that 20 to 60 Atg1 kinase complexes are recruited to the PAS upon induction of autophagy (Geng and Klionsky, 2008; Köfinger et al., 2015; Sweeney and Holzbaur, 2018).

The inhibition of Atg17 by Atg31–Atg29 limits its function to conditions under which autophagy is induced. However, selective cargo, such as the vacuolar aminopeptidase 1 (Ape1), is delivered to the vacuole even under non-induced conditions and independently of Atg17 (Shintani et al., 2002). The initiation complex in selective autophagy consists of Atg1, Atg13 and the scaffolding protein Atg11 (Kim et al., 2001). In vitro reconstitution experiments using Atg11 purified from yeast cell lysates indicated that the major function of Atg11 involves activation of Atg1. By using an engineered Atg1 kinase that metabolized an ATPγS analog instead of conventional ATP, that study revealed that the activation of Atg1 depends not only on Atg11, but requires the formation of a complex of Atg1 with cargo, cargo receptor and Atg11 (Kamber et al., 2015). However, the kinase activity of Atg1 had previously been found to be important for autophagy expansion, but dispensible for the nucleation of autophagosomes (Cheong and Klionsky, 2008). This suggests that the Atg1–Atg11 complex has two functions in autophagy: promoting autophagy initiation and regulating phagophore expansion. The function of Atg11 during nucleation of selective autophagosomes might depend on properties that are similar to those of Atg17, including the recruitment of Atg9 vesicles to the PAS (Backues and Klionsky, 2012).

A key step during the phagophore initiation to expansion transition is the acquisition of Atg8 (LC3 proteins; see Box 1), a process that involves recruitment of the conjugation machinery to the phagophore. Key factors in this are the two PtdIns(3)P-binding proteins WIPI2 (mammals) and Atg21 (yeast), which recruit ATG16L1 (mammals) and Atg16 (yeast) to the phagophore (Fig. 2A). Both WIPI2 and Atg21 thus promote expansion by coupling Atg8/LC3 lipidation to PtdIns(3)P-containing phagophores (Dooley et al., 2014; Juris et al., 2015). In order to identify the membrane source that becomes the substrate for ATG8 conjugation, a cell-free reconstitution approach combined cytosol fractions of wild-type cells with membrane fractionation from ATG5-knockout cell lysates (Ge et al., 2013). Owing to the absence of ATG5, these membranes were devoid of LC3-conjugated PE. Combining them with wild-type cytosol complemented ATG5 and initiated lipidation of LC3. The fraction that was enriched in membranes from the ER–Golgi intermediate compartment was found to be the best substrate for LC3 lipidation, implicating that this compartment donates membrane for autophagosome biogenesis (Ge et al., 2013). By applying a similar approach, coat protein complex II (COPII) has been found to mediate the budding of vesicles from the ER–Golgi intermediate compartment upon autophagy induction, suggesting that such vesicles contribute to the formation of autophagosomes (Ge et al., 2014). Furthermore, several other studies have reported that COPII is required for initiation and expansion of autophagosomes (Fig. 2B) (Ge et al., 2017; Graef et al., 2013; Stadel et al., 2015). Whether COPII vesicles transport membranes from the ER to autophagosomes remains, however, to be investigated.

The most likely region of the phagophore to which COPII and other donor vesicles might fuse is its highly bent membrane at the phagophore rim (Nguyen et al., 2017). This suggests that factors that are involved in this process are targeted to this membrane region by sensing membrane curvature (Fig. 2B,D). The localization of Atg proteins at the phagophore in yeast was analyzed by inducing the assembly of a giant ApeI substrate through overexpression of ApeI. Under these conditions, a large spherical ApeI structure is formed in the cytoplasm at which the phagophore membrane could be visualized. It was shown that subunits of the Atg1 kinase and PI3K complexes, as well as Atg9, colocalize with the edges of the phagophore, whereas Atg1 itself and the Ub-like conjugation system cover the entire membrane (Suzuki et al., 2013). Thus, autophagy initiation complexes are found exclusively at the edges of phagophores, whereas the phagophore expansion machinery is more or less evenly distributed on yeast phagophores. The spatially restricted localization of Atg17 and Atg9 implies that Atg17 tethers Atg9 vesicles not only to each other upon autophagy induction, but also to the phagophore rim during expansion (Fig. 2B).

The localization of the PI3K complex to the rim of phagophores can be explained by curvature-sensing motifs in VPS34 (also known as PIK3C3) and ATG14 that have been identified in vitro using large unilamellar vesicles (LUVs) with different sizes (Fan et al., 2011; Miller et al., 2010). The curvature-sensing property of ATG14 was found to be counteracted by increasing concentrations of PtdInsPs. This suggests that during autophagy initiation, a positive-feedback loop enriches ATG14 at the PtdIns(3)P-containing omegasome, whereas during phagophore expansion, incoming PtdIns from donor vesicles are phosphorylated in a concerted manner at the phagophore rim (Fan et al., 2011) (Fig. 2C).

Interestingly, membrane curvature-sensing is not only restricted to components of the initiation machinery: an in vitro reconstitution of the human Ub-conjugation machinery identified ATG3 as a curvature sensor. As shown for ATG14, ATG3 contains an amphipathic lipid-packing sensor (ALPS) motif, which forms, upon contact with lipid bilayers, an amphipathic helix that partially inserts into membranes (Nath et al., 2014). The penetration of this helix is promoted by lipid-packaging defects in the bilayer of highly curved membranes, explaining why such motifs are curvature sensors (Antonny, 2011; Bigay et al., 2005) (Fig. 2D). The localization of ATG3 at the phagophore rim spatially restricts conjugation of ATG8, which allows incoming membranes during expansion to be decorated with ATG8 in a spatiotemporally concerted manner. Moreover, ATG8 is found on both membranes of the phagophore, but the high curvature of its rim forms an effective diffusion barrier for lipids and thus for lipidated ATG8s as well (Domanov et al., 2011). Targeting ATG3 to the rim permits the ATG8 proteins to reach both faces of the phagophore after conjugation (Fig. 2D).

In contrast to these observations in vitro, yeast Atg3 has been found to be evenly distributed at the entire area of the phagophore in vivo (Ngu et al., 2015). This indicates that curvature sensing by ATG3 can be overruled by other factors such as the ATG12–ATG5–ATG16L1 complex, which is known to recruit ATG3 in vivo (Hanada et al., 2007). ATG16L1 also covers the surface of phagophores, but its localization is confined to the outer phagophore membrane (Mizushima et al., 2003). The restriction of ATG16L1 to one face of the phagophore might be a consequence of the asymmetric distribution of PtdIns(3)P in human cells (Fig. 2C). PtdIns(3)P is enriched in the outer membrane and bound by WIPI2 (Cheng et al., 2014), which also physically interacts with ATG16L1 (Dooley et al., 2014). How the asymmetric distribution of PtdIns(3)P is achieved and maintained remains to be shown. Moreover, the functional consequence of the asymmetry is currently unclear since in yeast, PtdIns(3)P is enriched at the inner and not at the outer membrane (Cheng et al., 2014).

The components and activity of the autophagy-specific conjugation system were revealed in landmark in vitro reconstitution studies from the laboratory of Yoshinori Ohsumi, which used purified proteins to recapitulate the enzymatic conjugation of Atg12 to Atg5 by Atg7 and Atg10 (Mizushima et al., 1998). Subsequently, a second conjugation system that targets Atg8 to membranes through its conjugation to PE was identified (Kirisako et al., 2000). Using LUVs as model membranes, this conjugation could be recapitulated with purified Atg8, Atg7 and Atg3 (Ichimura et al., 2004), and a functional connection of both conjugation reactions was revealed by the observation that Atg12–Atg5 promotes lipidation of Atg8 in vivo. Furthermore, reconstitution experiments of Atg8 lipidation in the presence of Atg12–Atg5 showed that the latter functions as an E3 ligase and promotes the transfer of Atg8 from the E2-like enzyme Atg3 to PE (Hanada et al., 2007).

The observation that, at least in human cells, the ATG12–ATG5–ATG16L1 complex localizes at the outer membrane of the expanding phagophore raises the question as to why the E3 ligase complex is retained on the phagophore after conjugation of ATG8 has taken place. This was addressed by a reconstitution of the entire yeast conjugation machinery on model membranes (Kaufmann et al., 2014). Atomic force microscopy of supported lipid bilayers (SLBs) (see Box 2) to which Atg8 was conjugated in the absence of Atg16 revealed that Atg8 associates with the Atg12–Atg5 conjugate in uniform particles composed of at least two subunits each. In the presence of Atg16, however, Atg8–Atg12–Atg5 became immobile and, instead of single particles, a continuous 12 nm thin membrane coat was detected on SLBs. The meshwork-like appearance of the coat was caused by the 16 nm-long coiled-coil domain of Atg16 that spatially separated two Atg8–Atg12–Atg5 complexes (Fig. 1C). An Atg16 mutant that could not self-assemble into tail-to-tail dimers caused a remarkable autophagy defect: cells that expressed this mutant produced large phagophore structures with an unusually long persistence in the cytoplasm. Moreover, these phagophores did not transport autophagic cargo to the vacuole, implying that the formed membrane structures correspond to immature and non-productive autophagosomes (Kaufmann et al., 2014). In vivo data from human cells supported this model by showing that deletion of all human ATG8 homologs led to a maturation and closure defect of autophagosomes, although autophagy progressed at a reduced level (Nguyen et al., 2016; Tsuboyama et al., 2016).

Taken together, the combination of in vitro reconstitution experiments with in vivo experiments has revealed that the Atg8 conjugation system is an important regulator of phagophore expansion, but is not essential for autophagy. Which membranes promote expansion in response to different environmental conditions, how their transport is coordinated and how incoming membranes merge with the phagophore are questions that remain to be answered.

Electron tomography of autophagic structures in human cells showed that the fully expanded phagophore adopts a spherical shape (Biazik et al., 2017). Sealing of the structure requires that the membrane rims of the remaining spherical pore come into very close proximity to eventually merge. How this process is coordinated and which proteins promote sealing of the phagophore remain unknown. In principle, the process should be similar to the fusion of two small vesicles of high membrane curvature or to the scission of a membrane neck that projects away from the cytoplasm into the lumen of organelles or the extracellular space. Research over the past decades has tried to clarify which of the two diametrically opposed mechanisms serve this purpose.

Initial efforts focused on the fusion model in yeast and human cells. Many fusion reactions in cells are carried out by SNARE proteins, which merge membranes of vesicles with those of organelles or the plasma membrane. To overcome the high energy barrier for membrane fusion, SNAREs from two opposite membranes form a four-strand helical bundle that brings the opposite membranes in such close proximity that fusion takes place (Poirier et al., 1998). However, a recent study reported that, although priming of SNARE proteins is required for their maturation, they are not involved in the formation of autophagosomes (Abada et al., 2017; Reggiori et al., 2004).

The lack of experimental evidence for a SNARE-mediated sealing mechanism has shifted the focus towards Atg proteins with potentially fusogenic properties. First indications for the involvement of yeast Atg8 in this process came from reconstitution experiments of the lipidation reaction on LUVs. Conjugation of Atg8 to LUVs resulted in massive clustering of the vesicles, which was macroscopically visible by an increase in the turbidity of the suspension (Nakatogawa et al., 2007). Further characterization of this process revealed that Atg8 molecules on opposite membranes interact with each other to promote tethering. Atg8-mediated fusion of theses membranes was, however, only observed if membranes with unphysiologically high levels of unsaturated phospholipids were used (Nair et al., 2011). This does not necessarily exclude the participation of Atg8 during phagophore closure since packaging defects of lipids lead to a destabilization of the membrane at the phagophore rim. However, additional experiments are needed to confirm the contribution of Atg8 in sealing of phagophores in vivo (Nair et al., 2011). Reconstitution reactions of human ATG8 homologs on model membranes revealed another potential mechanism by which ATG8 can promote fusion. The two ATG8 family members, light chain 3B (LC3B, also known as MAP1LC3B) and γ-aminobutyric acid receptor-associated protein L2 (GABARAPL2) possess highly basic N-terminal helices that are required and sufficient to tether and fuse membranes in vitro by binding negative charges of opposing phospholipid bilayers (Weidberg et al., 2011). Since lipidation of ATG8 might be restricted to the phagophore rim (Nath et al., 2014), these ATG8 proteins would be positioned at the right place to promote fusion (Fig. 2D). However, their small size would allow them to only bridge distances that are smaller than ∼5 nm. Spontaneous membrane fusion occurs will occur if membranes are as close as 2 nm (Kozlovsky and Kozlov, 2003), suggesting that ATG8 proteins can only catalyze the very last step of sealing at which the remaining phagophore pore is sufficiently constricted. Furthermore, other proteins, such as Atg16, have been found to tether membranes in in vitro reconstituted systems. As for the ATG8 proteins, tethering of model membranes was based on interactions of positively charged amino acids with acidic phospholipids (Romanov et al., 2012). Further studies thus need to clarify how these different tethering events contribute to autophagosome formation.

Recent progress in identifying the biophysical principles of membrane scission reactions indicates that sealing of the autophagosome does not necessarily require fusion, but rather scission of membranes (Knorr et al., 2015; Renard et al., 2018). Although intuitively less obvious, the topology of the phagophore pore is very similar to that of a membrane neck in membrane abscission processes during multivesicular body biogenesis or viral budding. Both scission reactions are carried out by the endosomal sorting complex required for transport (ESCRT) complex III (Wenzel et al., 2018; Wollert et al., 2009). In vitro reconstitutions of this process on giant unilamellar vesicles (GUVs) (see Box 2) using purified ESCRT components have demonstrated that the correct membrane topology is required and sufficient for cutting of the membrane by ESCRT-III (Wollert and Hurley, 2010). In vivo, ESCRT-III appears to cooperate with the AAA-ATPase VPS4 in order to sever not only membrane necks in multivesicular body biogenesis, but also during nuclear envelope assembly and midbody abscission during cell division, as well as for repair of damaged plasma or lysosomal membranes (Adell et al., 2017; Christ et al., 2017; Mierzwa et al., 2017). Owing to the diversity of cellular processes that rely on ESCRT-mediated membrane scission and that share a similar topology of the membrane neck that needs to be cut, a potential role of ESCRTs in autophagy was proposed. However, convincing experimental evidence was missing until a recent study reported that the ESCRT-III subunit-charged multivesicular body protein 2A (CHMP2A) is required for sealing of the phagophore membrane (Takahashi et al., 2018). In related reactions, ESCRT-III cooperates with upstream factors to generate membrane necks that are narrow enough to be cleaved by ESCRT-III. This suggests that – as discussed for a potential ATG8-mediated fusion reaction – the phagophore rim needs to be sufficiently constricted to become a substrate for ESCRT-III. Whether phagophore expansion inherently leads to a sufficiently constricted pore remains a challenging question. Furthermore, a collaboration of Atg8-mediated fusion and ESCRT-III-mediated scission mechanisms remain a possibility. Innovative experiments using physiologically relevant templates such as cup-shaped membranes, might be able to answer this question in the future.

The final step in autophagy requires the outer membrane of the autophagosome to fuse with the limiting membrane of the lysosome. Before this, autophagosomes need to be transported to lysosomes and both compartments need to be tethered to each other. Yeast autophagosomes form in the immediate vicinity of the vacuole, allowing their completion to be coordinated with fusion independently of microtubule-based transport mechanisms (Kirisako et al., 1999). In human cells, however, autophagosomes are formed at various locations (Jahreiss et al., 2008) and therefore need to be transported to the perinuclear region where lysosomes are located (Korolchuk et al., 2011). This process depends on microtubules, the minus-end-directed motor dynein, which mediates retrograde transport of cargo to lysosomes, but also the plus-end-directed kinesin motor KIF5B, which is important for autophagosome positioning and fusion with lysosomes (Cardoso et al., 2009; Kimura et al., 2008; Köchl et al., 2006). Advanced in vitro approaches have been based on fusion of single dynein or kinesin motors to fluorescent beads and their incubation with microtubule-coated coverslips (Svoboda and Block, 1994). Optical traps allowed movement of single motor proteins on microtubules to be recorded and forces in the pico-Newton range to be measured (Simmons, 1996): Dynein, which drives the anterograde transport of autophagosomes, moves with a speed of ∼0.7 µm/s and can exert a force up to 5 pN (King and Schroer, 2000; Toba et al., 2006). Neurons are particularly well suited for the study of transport processes in vivo. Most autophagosomes are produced at the distal axon of neurons and thus need to be transported to the soma, where lysosomes are located, over distances of 500–1000 µm (Maday and Holzbaur, 2014). Based on the measured velocity of dynein, it takes 12–24 min for an autophagosome to be transported from the distal axon to the soma. Compared to myosin-based transport on actin filaments, dynein-mediated trafficking is rather slow (Schott et al., 2002).

Having reached perinuclear lysosomes, autophagosomes need to be recognized as fusion partners. This requires selective markers on their surface to provide identity and one such marker, the SNARE protein syntaxin 17 (STX17), is recruited to autophagosomes after their completion (Itakura et al., 2012; Tsuboyama et al., 2016).

The activity of SNARE proteins is tightly regulated; before fusion occurs, organelles need to be brought in close vicinity. This tethering is mediated by the lysosomal HOPS complex, which binds autophagosomes and lysosomes through its interaction with the small GTPases such as ADP-ribosylation factor-like protein 8 (ARL8) and the Ras-related protein RAB2 in humans and Ypt7 (a RAB7 homolog) in yeast (Reggiori and Ungermann, 2017). The molecular architecture of the yeast HOPS complex was revealed by single-particle electron microscopy (Bröcker et al., 2012). The complex spans 30 nm, possesses Rab-binding sites (in Vsp41 and Vps39) on both ends and its headpiece contains the SNARE-interacting subunit Vps33. In vitro reconstitution experiments with LUVs revealed that the HOPS complex tethers membranes based on its interaction with Rab GTPases (Ho and Stroupe, 2015), but also through a curvature-dependent direct interaction with membrane vesicles (Ho and Stroupe, 2016). Moreover, the cooperation of HOPS, SNAREs, Ypt7 and the SNARE chaperones Sec17 and Sec18 is required to drive robust membrane fusion (Stroupe et al., 2009). The HOPS complex thus not only tethers autophagosomes to lysosomes, it also promotes SNARE assembly (Fig. 3A). In autophagy, the HOPS complex promotes fusion by cooperating with STX17 (Jiang et al., 2014) as well as with SNAP29 and VAMP8 (Fig. 3A) (Itakura et al., 2012). Mechanistic details on how SNAREs drive membrane fusion have been revealed using SNARE-containing LUVs. Examples include the specificity of SNARE proteins (Paumet et al., 2004) and the process of zippering into a four-strand helical bundle (Jakhanwal et al., 2017; Stein et al., 2009), as well as the process of membrane hemi-fusion and fusion (Hernandez et al., 2012).

Even in the presence of tethering factors, the spontaneous assembly of SNAREs is inefficient and controlled by Sec1/Munc18-like (SM) proteins (Gerber et al., 2008). Insights into the mechanism by which SM proteins promote assembly of SNAREs came from in vitro reconstitutions and structural characterizations: the SM protein Vps33 helps to assemble the t-SNARE motifs of SNAREs within the target membrane such that they are in register for interaction with the v-SNARE within the vesicle membrane (Baker et al., 2015). A similar assisted assembly of SNAREs was found in autophagy. Unexpectedly, however, ATG14, the component of PI3K complex that initiates autophagy, promotes this step (Fig. 3A). The structure of the SNARE–ATG14 complex revealed that the coiled-coil domain of ATG14 stabilizes the interaction of STX17 and SNAP29 to prime both SNAREs for zippering with the corresponding partner VAMP8 (Baker et al., 2015). Consequently, reconstitutions on LUVs showed that SNARE-mediated membrane fusion was strongly enhanced in the presence of ATG14 (Diao et al., 2015). Taken together, ATG14 appears to have different functions in autophagy: as a component of the PI3K complex, it promotes induction of autophagy at the ER, whereas its function in autophagosome–lysosome fusion is independent of the other PI3K complex subunits (Fig. 3A).

Before fusion of autophagosomes and lysosomes occurs, Atg proteins – including the cytoplasmic accessible pool of Atg8–PE or its human homologs – are released from the autophagosome. Recycling of Atg8 is achieved through proteolytic cleavage of the amide bond between the C-terminal glycine of Atg8 and the amine function of PE by Atg4 (Kirisako et al., 2000). The recruitment and activity of Atg4 needs to be tightly regulated to prevent premature cleavage of Atg8 (Abreu et al., 2017). This is achieved by Atg1/ULK1-mediated phosphorylation of Atg4 or its human homologs at a conserved serine residue near its catalytic site (Pengo et al., 2017; Sánchez-Wandelmer et al., 2017). The importance of this serine residue was confirmed by reconstituting the deconjugation reactions using SUVs with enzymatically conjugated Atg8 and incubating these with Atg4 or related serine mutants (Sánchez-Wandelmer et al., 2017). The observation that Atg1/ULK1 is present on phagophore membranes during their expansion and released shortly after autophagosome completion strongly implies that deconjugation of Atg8 proteins and fusion of autophagosomes with lysosomes are interdependent and precisely regulated processes (Fig. 1D,E) (Cebollero et al., 2012; Yu et al., 2012). Spatiotemporal insights into Atg4-mediated deconjugation were obtained using fluorescently labeled Atg8 conjugated to GUVs (Kaufmann et al., 2014). Atg4 efficiently recycles Atg8 from model membranes in vitro, even if Atg8 is engaged in a membrane scaffold with Atg12–Atg5–Atg16. Consequently, the scaffold formation, which is predicted to regulate phagophore expansion, and recruitment of Atg4 by Atg8 are mutually exclusive (Fig. 1D) (Kaufmann et al., 2014).

The deconjugation of human ATG8 proteins is more complex, given that six ATG8 homologs are conjugated to autophagosomes and four ATG4 homologs catalyze the priming and recycling reactions (Kabeya et al., 2004). The human ATG4B protein is the most potent isoform involved in priming human ATG8 proteins and efficiently cuts the LC3 family proteins GABARAP and GABARAPL2 (Li et al., 2011; Sou et al., 2006). Although the other ATG4 homologs are considerably slower in priming of human ATG8 proteins, they possess similar catalytic activities to ATG4B for deconjugating ATG8 proteins in vitro (Kauffman et al., 2018). The precise timing of the recycling reaction, as well as the physiological functions of the individual ATG4 homologs in human cells, remain to be characterized. Interestingly, depletion of Atg4 in yeast or ATG4 homologs in humans leads to the formation of considerably smaller autophagosomes and impact on the sealing of the phagophore (Fujita et al., 2008; Hirata et al., 2017; Nair et al., 2012). This suggests that deconjugation of Atg8 or its human homologs not only takes place shortly before, but also during phagophore expansion and sealing.

After autophagosomes and lysosomes have fused and autophagic cargo has been degraded, lysosomes need to be regenerated. The involved process has been termed autophagic lysosome reformation (ALR) and involves the formation of membrane extrusions from autolysosomes. Proto-lysosomes bud from theses tubes and their maturation restores the population of functional lysosomes (Chen and Yu, 2017).

The fusion of the outer membrane of autophagosomes with lysosomes delivers autophagic cargo to the lysosomal lumen where it is degraded. Much attention has thus been devoted to reveal how specific cargo is selected and targeted to phagophores. This involves a family of conserved autophagy receptors, which bind cargo on the one hand and the phagophore on the other (Khaminets et al., 2016). Tethering of cargo to the phagophore depends on a LC3-interacting region (LIR)- or Atg8-family interacting motif (AIM)-mediated interaction of autophagy receptors with Atg8 family proteins (Lamark et al., 2017) (Fig. 3B). The classical LIR motif is characterized by two bulky hydrophobic residues that are separated by two random residues and flanked by acidic amino acids. More recently, a number of motifs that distinguish between different ATG8 family members have been identified, suggesting that binding of cargo receptors is more complex in higher eukaryotes (von Muhlinen et al., 2012; Rogov et al., 2017; Stolz et al., 2017). The interaction of purified yeast and human Atg8 proteins with peptides that contain the various LIR/AIM motifs has been investigated and the underlying structural, and to a certain extend regulatory, principles are well understood (Kim et al., 2016; Wild et al., 2013). Despite the important contributions of these in vitro binding studies, they have some limitations. The LIR motif of tectonin β-propeller repeat-containing protein 2 (TECPR2), for example, binds all human ATG8 proteins with similar affinity in vitro, but TECPR2 selectively interacts with LC3C (MAP1LC3C) in vivo (Stadel et al., 2015). Furthermore, increasing evidence suggests that lipidated and non-lipidated Atg8 proteins adapt distinct structural conformations that impact on their affinity to LIR motifs (Coyle et al., 2002; Ichimura et al., 2004; Kumeta et al., 2010; Nakatogawa et al., 2007), which should be systematically investigated in the future.

Another interesting insight into selectivity of autophagy was the finding that multiple Atg8-binding sites on Atg19 are critical for a tight tethering of the cargo ApeI to membranes in order to avoid uptake of other material (Sawa-Makarska et al., 2014). A related mechanism to achieve selective and exclusive uptake of distinct cargo was revealed by recapitulation of aggrephagy: the receptor p62 (also known as SQSTM1) oligomerizes upon binding to protein aggregates, which facilitates efficient tethering of such aggregates to LC3-conjugated membranes in vitro (Wurzer et al., 2015). However, cargo such as yeast ApeI does not only activate its corresponding cargo receptor Atg19, it also appears to initiate lipidation of Atg8 by recruiting the E3 ligase complex (Fig. 3B). This couples receptor-mediated tethering of cargo to Atg8 conjugation, providing a possible mechanism to spatiotemporally coordinate cargo tethering with expansion of the isolation membrane in a selective and exclusive manner (Fracchiolla et al., 2016).

Taken together, the combination of in vitro and in vivo studies greatly advanced our understanding of how cargo is tethered to membranes during selective autophagy. One of the most important questions that remain to be investigated concerns the two faces of autophagy – how can the cell transform a highly selective pathway that degrades defined cargo in an exclusive manner into a non-selective pathway that degrades bulk cytoplasm?

In vitro reconstitutions of biological processes have proven to be powerful tools to reveal unprecedented insights into fundamental mechanisms in biology. The combination of different model membranes, such as vesicles and supported lipid bilayers (see Box 2), with structural, fluorescent-based, biophysical and biochemical methods (see Box 3) have allowed the dissection of molecular principles that govern autophagy induction, and phagophore expansion and maturation, as well as phagophore fusion with lysosomes. The advantage of in vitro systems to provide a chemically well-defined environment is, however, at the same time its strongest limitation. Often, biophysical properties of isolated components in vitro simply cannot be correlated one-to-one with their functions in vivo. This might reflect fundamental differences of artificial membranes, which are composed of pure phospholipids (see Box 2), and biological membranes, which are much more complex in terms of their lipid composition. Biological membranes contain also a substantial amount of membrane proteins that strongly impact on physical properties, spatial organization and shape of these membranes. Moreover, many experiments with reconstituted systems are ‘out of equilibrium’, relying on physical separation of components to analyze or quantify them independently (see Box 3). Examples include a vast number of binding studies with liposomes and recombinant proteins. In vivo systems are, however, mostly in chemical equilibrium and fine-tuned by variations in expression levels of proteins or physical compartmentalization of membranes into microdomains. A successful strategy to overcome these limitations is a combinatorial approach in which in vitro experiments are complemented by in vivo approaches. Quite often, these combinations create positive-feedback loops in which one method complements the limitations of the other. The resulting synergy will be well suited to tackle the most challenging remaining questions in autophagy in the future. Combined with new biophysical approaches, even more complicated aspects of autophagy can hopefully be reconstituted in vitro. The final goal in ‘reconstructing destruction’ remains to build an autophagosome from purified components in the test tube.

We thank Lisa Dempfle for comments on the manuscript.