Autophagosomes form in eukaryotic cells in response to starvation or to other stress conditions brought about by the unwanted presence in the cytosol of pathogens, damaged organelles or aggregated protein assemblies. The uniqueness of autophagosomes is that they form de novo and that they are the only double-membraned vesicles known in cells, having arisen from flat membrane sheets which have expanded and self-closed. The various steps describing their formation as well as most of the protein and lipid components involved have been identified. Furthermore, the hierarchical relationships among the components are well documented, and the mechanistic rationale for some of these hierarchies has been revealed. In the present review, we try to provide a current view of the process of autophagosome formation in mammalian cells, emphasizing along the way gaps in our knowledge that need additional work.
Autophagy is conserved in all eukaryotic cells from yeast to humans and depends on a common set of genes first identified and characterized in Saccharomyces cerevisiae [1–3]. The autophagic pathway is activated primarily in response to nutrient limitation (most notably nitrogen in yeast and amino acids in mammalian cells), and it eventually provides nutrients from self-digestion [autophagy=self (auto) eating (phagy)]. This is accomplished by the formation in the cytosol of novel membrane precursors that expand and close upon themselves creating a double-membraned vesicle, the autophagosomes . The process of expansion and self-closure traps cytosolic components in the lumen of the double-membraned vesicle, and it provides the cell with a useful mechanism to isolate some cytosolic components from the rest of the cytosol. Eventually, and after a series of maturation steps involving elements of the endosomal system, autophagosomes fuse with lysosomes and the luminal content is then digested to provide nutrients. Although autophagy is a nutrient-generating pathway, it is also used for the elimination of toxic materials (damaged organelles and aggregated proteins) from the cytosol, whereas in many settings autophagosomes can be specifically employed for the elimination of invading pathogens [5,6]. It is quite likely that additional cellular processes will be shown to depend on autophagy, taking advantage of the fact that this is a unique trafficking pathway that can transform cytosolic material into vesicular cargo.
The earliest morphological studies of autophagosome formation relied on EM, and, in a remarkable feat of analysis, the main steps of this pathway were correctly inferred [7–9]. In recent years, many live imaging studies relying on overexpression of autophagy-related proteins have complemented these early studies. The goal has been to identify the origin of the autophagosomal membrane, and to provide a dynamic view of the pathway. For both of these aims, a clear consensus is lacking and it is possible that autophagosomes form in connection to multiple membrane sources depending on physiological state and perhaps the nature of the stimulus, be it the type of starvation or the component that needs to be eliminated [10–13]. In the present review, we focus on a single pathway of autophagosome formation which has been described in complementary experiments from several laboratories. In this pathway, autophagosomes form in close physical proximity to the endoplasmic reticulum (ER), and they involve an omegasome intermediate.
Steps of autophagosome formation
The signature organelles of autophagy are double-membraned vesicles originally identified and described by Christian de Duve and colleagues by EM (reviewed in [8,9]). Their appearance suggested lipid-rich and protein-poor structures, frequently engulfing entire organelles. More detailed EM work soon identified an elongated membrane precursor termed the phagophore or isolation membrane that appears to expand and close into the mature double-membraned vesicle. With the concomitant isolation of yeast mutants blocked at distinct steps of autophagosome formation, it was soon possible to divide the process into a series of separate steps. Much recent work combining genetic and biochemical approaches has further clarified the steps of this pathway into a description widely accepted in the field [4,13,14]:
Initiation. A fall in nutrient levels (primarily amino acids) is sensed by mammalian (or mechanistic) target of rapamycin complex 1 (mTORC1), a kinase that connects and regulates almost all anabolic and catabolic cellular pathways . Inactivation of mTORC1 then activates the Unc-51-like kinase 1 (ULK1) protein complex, which translocates to ER membranes and activates a second protein complex containing the VPS34 protein [16,17].
Nucleation. VPS34 phosphorylates phosphatidylinositol to generate phosphatidylinositol 3-phosphate (PI3P) on specialized membrane regions physically connected to the ER that are termed omegasomes [18,19].
Expansion. A small membrane sheet formed in the previous stage within omegasomes begins to expand until it reaches a maximal size before beginning the process of self-closure.
Fusion. The edges of the curved membrane sheet fuse to form a vesicle of two membranes (four bilayers). This is a complete early autophagosome. It eventually becomes acidic before fusing with the lysosomes to deliver its cargo.
Many variations on this basic pathway have been described experimentally [20–23]. Autophagosome formation can take place at the plasma membrane, or using elements of the retromer complex, whereas some autophagosomes are evident in the absence of ULK1 complex or vacuolar protein sorting 34 (VPS34) complex components. A form of autophagy that does not rely on canonical autophagosomes has also been described in animals deficient in the autophagosome expansion/lipidation machinery. There are, in addition, cellular pathways relying on some elements of the autophagic machinery without involving autophagosome formation. Because of this variability, it is difficult to describe the molecular details of the autophagic pathway in the most general terms. It is, however, possible to describe one of the pathways (as outlined in the previous paragraph) in quite extensive detail, and, most importantly, provide mechanistic explanations of how the different steps fit together. This is done in the next two sections (Figure 1).
Pathway of autophagosome formation
Machinery of autophagosome initiation
Signalling to VPS34
In the short term, the signal sensed by mTORC1 to induce autophagy is amino acid availability; other signals such as glucose or serum levels can also lead to autophagy induction, but at much lower rates (minutes compared with hours) [16,24]. mTORC1 senses amino acids from the surface of lysosomes, although the primary sensor may be within the lysosomal lumen, via a cascade of complexes involving the Rag GTPases and the Ragulator complex . Amino acid withdrawal causes mTORC1 to fall off the lysosomal surface within minutes, whereas amino acid resupply either via autophagy or by external supplementation reverses this translocation quickly. Inactivation of mTORC1 leads directly to dephosphorylation of ULK1, an important autophagy-specific kinase, which in turn directly phosphorylates ATG13 and 200 kDa focal adhesion kinase family-interacting protein (FIP200), two other proteins of the ULK1 complex . This heterotetrameric ULK1 complex (ULK1, FIP200, ATG13 and ATG101), with phosphorylated sites on FIP200 and ATG13, is the most upstream activated complex that initiates autophagy. The second complex to be activated during autophagy contains the VPS34 lipid kinase and its associated proteins VPS15 (a stoichiometric VPS34 binder), BECLIN1 and ATG14. Signalling from ULK1 to VPS34 involves direct phosphorylation of BECLIN1 by ULK1, but additional signals are very likely . For example, BECLIN1 availability for interacting with VPS34 is regulated by a number of kinases such as AMP-activated protein kinase (AMPK), c-Jun N-terminal kinase 1 (JNK1) and death-associated protein kinase (DAPK), whereas VPS34 itself is subject to direct phosphorylation by cyclin-dependent kinase 1 (CDK1) and protein kinase D (PKD) [27–29]. All of these phosphorylations are important for autophagy induction, but it is unclear whether there are hierarchical relationships.
Protein components and PI3P
Activation of VPS34 and its translocation to membranes via an ER-targeting domain on ATG14 initiates the formation of PI3P on ER-related membranes, which have been termed omegasomes due to the Ω shape they form together with the ER [30,31]. The exact geometry of the ER–omegasome connection is still a matter of investigation. Two independent studies examining this question by electron tomography described a similar set of connections between two ER cisternae and an early autophagic structure sandwiched in between that probably corresponds to the omegasome [32,33]. In these tomograms, thin ER strands connect the inner and outer cisternae. Another study examining this question in cells deficient for downstream autophagy components (which thus accumulate early intermediates) has provided a more detailed view of the omegasome–ER connection . According to this study, thin tubular clusters emanating from the ER and perhaps corresponding to the omegasome connect the ER to the forming autophagosome. A similar view has emerged in studies using cryo-soft X-ray tomography to describe early autophagosomal structures in unstained biological samples . Again, a cluster of omegasomes was seen in distinct regions of the ER. A unifying theme in these ultrastructural studies is that the PI3P-enriched membranes are tubular and distinct from the rest of the ER, consistent with the idea that the omegasome is a modified ER domain. Of note, however, none of these approaches was able to detect the PI3P lipid directly.
A mechanistic rationale for the necessity to initiate autophagosome formation from PI3P-enriched membranes was lacking until very recently. Work from Sharon Tooze and colleagues has provided an answer to this question by demonstrating that WIPI2, an essential PI3P effector for autophagy , binds to ATG16L1, a protein that is part of the ATG12-5-16 elongation machinery that mediates the conjugation of light chain 3 (LC3) with phosphatidylethanolamine (PE) . Thus a PI3P effector links omegasomes with lipidation on LC3, the main protein of the growing autophagosome. This provides the most compelling explanation as to why autophagosomes form within PI3P-enriched membranes.
How it all fits together
Perhaps because early mechanistic studies of autophagy relied on yeast genetic data organized in a hierarchical pathway, autophagosome formation has also been described morphologically as a series of hierarchical steps in both yeast and mammalian cells [38,39]. In recent years, it has also become clear that this pathway exhibits remarkable temporal characteristics. Data from independent laboratories working on different cell types fit best with the idea that autophagosomes do not form en bloc (from all components gathered together), but rather rely on a series of transient intermediates that form and collapse as an initial nucleating structure becomes a mature autophagosome [40,41] (Figure 2). The earliest visible structure traceable to the mature particle is a small punctum formed by the ULK1 components . This is frequently seen in association with the ER, and almost never as a free-standing structure away from cellular membranes. In other words, autophagosomes always form in association with a pre-existing membrane. Although in principle close proximity to the ER is the only requirement, it seems that mitochondria may also be sporadically involved, especially the ER–mitochondria junction . The initial ULK1 structure very rapidly becomes positive for PI3P as shown by the translocation of double-FYVE-containing protein 1 (DFCP1), a major omegasome component , to it. Later in this process, LC3 begins to accumulate within this mixed ULK1/omegasome membrane, and we now know that this is enabled by the lipidation of LC3 on the PI3P-enriched membrane . This expansion continues until a time when two events appear to take place: the ULK1 complex dissociates from the structure, and the autophagosomal membrane begins to bud off. It is tempting to speculate that these two events are causally connected, but this has not been addressed experimentally. A little later, any sign of PI3P diminishes and a fully formed autophagosome is visible that can freely move in the cell, no longer in association with any cellular membrane. This early autophagosome undergoes one more intrinsic maturation step, namely it becomes acidic, before fusing with the endolysosomal system to deliver its cargo [19,43].
Wave theory of autophagosome formation
This description of the steps leading to formation of one autophagosome from one nucleating structure has omitted several important considerations. One is discussed at length below and involves a dimension that is not visible by light microscopy, but is likely to be of critical importance. The second involves membranes and organelles that, although not part of the finished structure, interact continuously with forming autophagosomes. For example, an essential autophagy protein, Atg9, is found in large vesicles originating from the trans-Golgi network. These vesicles have been shown to interact with the growing omegasomes as autophagosomes begin to form inside without being incorporated into them [44–46]. Similarly, vesicles of late endosomal origin containing VPS34 have also been found to transiently associate with expanding omegasomes . In addition, vesicle clusters marking ER exit sites frequently co-localize with early autophagy structures, but without being absorbed into them . It is therefore important to emphasize that the growing autophagosomal structures described above are in constant communication with a plethora of (stable) cellular membranes, perhaps obtaining important protein and lipid components as they grow.
The reason forming autophagosomes rely on a series of transient intermediates is not clear. For the omegasome–autophagosome transition, we have suggested that first forming an omegasome may provide economy in resources and speed in reaching a starvation-induced steady state . It is possible that similar reasons of efficiency hold for the rest of the intermediates. Alternatively, dividing the process of formation of a large complicated structure into discrete intermediates may allow a lower error frequency, as has been shown theoretically for large virus assembly .
Missing dimension: challenge for future studies
A tremendous amount of progress has been made in determining crystal structures of various autophagy-specific proteins and assemblies [50,51]. These studies are beginning to provide a framework for understanding aspects of autophagosome formation in molecular terms. As noted elsewhere , it seems that the majority of these structures do not correspond to known motifs, but rather seem to be specific to the proteins involved. In addition, a major functional output of these structures seems to be to provide scaffolding platforms for the assembly of various intermediates. In terms of relative progress, much more is known about the conjugation machinery that is involved in the covalent conjugation of LC3 with PE, and less about the two complexes regulating early events, ULK1 and VPS34. At the level of the whole complex, neither ULK1 nor VPS34 has been solved. At the level of individual subunits, a crystal structure of the catalytic domain of Drosophila VPS34 and parts of BECLIN1 have been described, as well as structures for various parts of the ULK1 complex from yeast [50,52]. Finally, good structural information is available for one of the PI3P effectors, the yeast HSV2, which is a close homologue of ATG18 (corresponding to the WIPI family in humans) [53,54]. It is very likely that in the next 2–4 years structural information for most of the autophagy proteins/complexes will be available that will allow us to merge microscopical studies on the pathway of autophagosome formation with its description in atomic terms.
There is one level of understanding of the autophagic pathway that falls between light microscopy and X-ray crystallography, and, owing to the dynamic nature of the event, cannot be adequately described by EM. Drawing a parallel with another field of cell biology, namely vesicular transport, this is an area where vesicles not visible by light microscopy (of the order of 50 nm in diameter) are important contributors to a process but cannot be precisely followed. In vesicular transport, this has led to long-standing controversies on the nature of the transport process itself, and on the intermediates involved (compare, for example, these essays spanning 20 years and revolving around a still unsettled question on the nature of Golgi transport [55–57]). For autophagy, there may be two aspects where this can become a major issue. On the one hand, it is likely that the earliest structures formed during autophagosome biogenesis may be too small to be seen by light microscopy and it will not be possible to look at these structures by EM since their provenance cannot be known a priori based on light microscopy. Such structures may be those involving ATG9 vesicles . Although ATG9 is required at the earliest step, it is not clear how this is justified because many ATG9 vesicles are too small to be seen as single particles by live imaging. On the other hand, the process of phagophore expansion almost certainly relies on vesicular transport and, in this aspect, it is invisible by light microscopy, even when combined with EM (again the question of the provenance of structures seen by EM is a problem). On the basis of current technology, two approaches may help with this potential bottleneck. Modelling of the process of autophagosome formation based on a mixture of observation and theory may help uncover requirements that can then be more precisely addressed experimentally. In a complementary way, in vitro reconstitution studies starting from pure components may provide a simpler framework wherein biochemical and EM studies can be combined. Work along both of these fronts (modelling and in vitro reconstitution) is already underway for questions in autophagosome formation [59–62].
Membrane Morphology and Function: A Biochemical Society Focused Meeting held at Hotel del Camerlengo, Fara San Martino, Abruzzo, Italy, 5–8 May 2014. Organized and Edited by Banafshé Larijani [IKERBASQUE, Basque Foundation for Science and Unidad de Biofísica (CSIC-UPV/EHU), University of the Basque Country, Spain] and Marco Falasca (Barts and The London School of Medicine and Dentistry, U.K.)
double-FYVE-containing protein 1
200 kDa focal adhesion kinase family-interacting protein
mammalian (or mechanistic) target of rapamycin complex 1
Unc-51-like kinase 1
vacuolar protein sorting
Our work is supported by the Biotechnology and Biological Sciences Research Council.