Autophagy was discovered in the late 1950s when scientists using the first electron microscopes saw membrane-bound structures in cells that contained cytoplasmic organelles, including mitochondria. Pursuant to further morphological characterization it was recognized that these vesicles, now called autophagosomes, are found in all eukaryotic cells and undergo changes in morphology from a double-membraned vesicle with recognizable content, i.e. sequestered organelles, to a uniformly dense core autolysosome. Genetic screens in the yeast Saccharomyces cerevisiae in the 1990s provided a molecule framework for the next era of discovery during which the interest in, and research into, autophagy has rapidly expanded into many areas of human biology and disease. A relatively small cohort of approximately 36 proteins, called Atgs (autophagy-related proteins), orchestrate the formation of the autophagosome, and these are now being studied and functionally characterized. Although the function of these proteins is being elucidated, the underlying molecular mechanisms of how autophagosomes form are still not completely understood. Recent advances have, however, provided a significant advance in both our understanding of the molecular control of the Atg proteins and the source of the membranes. A consensus view is emerging from these advances that the endoplasmic reticulum is the nucleation site for the autophagosome, and that contributions from other compartments (Golgi, endosomes and plasma membrane) are required. In the present chapter, I review the data from the pre-molecular decades, and discuss the most recent publications to give an overview of the current view of where, and how, autophagosomes form in mammalian cells.
In the 1950s after the development of EM (electron microscopy), cell biologists using newly developed techniques and EM recognized that there was an unusual type of lysosome in some cells which contained mitochondria and cytosolic components , subsequently called cytolysosomes  or autophagosomes . Later, some of the vesicles were confirmed as bona fide lysosomes; use of cytochemical techniques confirmed they contained lysosomal hydrolases. However, what intrigued these scientists (and continues to intrigue us) are the membrane dynamics and topological issues relating to the formation of the double membrane, and the mechanism of internalization of cytosolic components. Both of these issues reinforce the notion that the origin and formation of these vesicles is unique.
The current view of the autophagy pathway is shown in Figure 1. The pathway is initiated at the PAS (pre-autophagosomal structure, also known as phagophore assembly site). It is worth noting from the outset that it is not clear from either morphological or biochemical techniques what the PAS is. A double-membrane structure, seen by EM in cells as an open cup-like structure, forms from the PAS, which is called a phagophore (or isolation membrane). This structure closes to form an autophagosome, which after its closure and formation has a double membrane. One face of this double membrane (which was originally cytosolic) is now within the lumen of the autophagosome. The autophagosome then fuses with endosomes and lysosomes, and acquires degradative enzymes, including proteases and lipases, becoming an autolysosome.
Although the static images collected by EM provide detailed morphological data, they did not reveal the dynamics of the process, in particular where and how rapidly the autophagosomes form, although early EM data did suggest autophagosomes were relatively short-lived . Live-cell imaging techniques [5–8] revealed the rapid kinetics of autophagosome formation, in particular after amino-acid starvation. Autophagosomes typically form within 5–10 min of starvation, which suggests that the formation occurs from pre-assembled structures or by efficient vesicular trafficking from existing compartments and membranes. Thus the following questions arise: what is the PAS, what are phagophores and where do the membranes come from? Finally, how do they become autophagosomes?
Definition of the PAS and phagophore
A key observation, again using EM techniques, was that the phagophore membranes were electron-dense, and more osmophilic than most other cellular membranes . Further morphological advances and high-quality cytochemical studies led to several different hypotheses about the origin of the autophagic membranes, including the possibility that the membrane originated from the ER (endoplasmic reticulum) [10–12] (and references therein) and the GERL (Golgi–ER–lysosome) . Essentially, an electron-dense reaction product produced by the enzymes resident in organelles, such as the ER, Golgi and lysosomes, can be detected by conventional EM in the positive organelles, helping to understand the origin or composition of the compartment. These morphological approaches were expanded by the application of subcellular fractionation techniques, especially powerful when combined with morphology, which aimed to understand what the intermediates in the pathway were, and most crucially, what the composition of the phagophore, or isolation membrane and amphisome were [12,14]. In addition, using freeze-fracture microscopy the phagophore membrane appeared to be protein-poor, suggesting that the membrane was a unique composition . However, technical limitations remained with different laboratories reporting different amounts of enzyme labelling and activity for resident enzymes, or different membrane markers in subcellular fractionation, so, despite the use of these state-of-the-art approaches, no universal consensus was obtained about the origin of the PAS or phagophore.
The molecular machinery
There are now more than 36 known Atg proteins in yeast, and among these, there is a set of proteins that are required for macroautophagy in yeast and mammals. Macroautophagy, also known as autophagy, is the best studied type of autophagy, among the numerous types that include: selective autophagy (mitophagy and xenophagy); microautophagy; and the mammalian-specific chaperone-mediated autophagy. (Macro)autophagy is thought to be a non-selective process, although some of the cargo-selection machinery used in selective autophagy is also required (see Chapter 5). In the case of the yeast proteins there is typically a single gene and protein, whereas in mammals there is some gene duplication. Both the yeast and mammalian proteins can be grouped into functional categories, and arranged in a hierarchy (Figure 2). Briefly, in yeast, the Atg1 serine–threonine kinase complex, including Atg1, Atg13, Atg17, Atg29 and Atg31, is thought to be the most upstream regulator of autophagy, and downstream of TOR (target of rapamycin). In mammals, this complex is the ULK (uncoordinated-51-like kinase) complex comprising ULK1/2, FIP200 (focal adhesion kinase family-interacting protein 200 kDa) and Atg13, and is also regulated by mTORC1 [mammalian (also known as mechanistic) TOR complex 1] and AMPK (AMP-activated protein kinase) (see Chapter 1). The second conserved kinase complex is a lipid kinase complex containing Vps34, and is also known as the class III PI3K (phosphoinositide 3-kinase) complex in mammals. The autophagy-specific PI3K complex consists of a catalytic subunit, Vps34, the regulatory subunit p150 (homologue of yeast Vps15), Beclin1 (in yeast, Vps30) and Atg14. Using inhibitors of the PI3Ks, it was shown that PI3P (phosphatidylinositol 3-phosphate) is essential for the initiation of autophagy. It is not known what role PI3P has in the organization of the phagophore or autophagosome membrane. However, as is also the case in other intracellular trafficking steps, the effectors of PI3P are thought to have important functions . Effectors of PI3P produced during autophagy are proposed to be the WIPI (WD-repeat protein interacting with phosphoinositides) family (WIPI1–4), which are the homologues of yeast Atg18 and Atg21. An additional effector of the pool of PI3P produced in the ER is DFCP1 (double FYVE-domain-containing protein 1) , which is found only in mammals and is present on so-called omegasomes as described below (also see Chapter 2).
Additional protein complexes required for autophagy include two ubiquitin-like conjugation systems that are conserved between species. These systems share one Atg protein (Atg7) and are also inter-dependent. The first, as described in Chapter 4, is the Atg12–Atg5–Atg16 complex, in which the Atg12 molecule is the ubiquitin-like protein. Atg5 is covalently linked to Atg12 while Atg16 associates with the Atg12–Atg5 conjugate to form a large complex. The second ubiquitin-like conjugation system involves Atg8 that is covalently modified by the lipid PE (phosphatidylethanolamine). Mammalian cells have several Atg8-like molecules, called LC3 (light-chain 3), GABARAP (γ-aminobutyric acid receptor-associated protein) and GATE-16 (Golgi-associated ATPase enhancer of 16 kDa), of which the first two have multiple family members (see Chapter 5). Importantly, Atg12–Atg5–Atg16 provides an E3-like activity to promote the lipidation of LC3 by Atg3 . The best studied Atg8 protein in mammalian cells is LC3B, which when lipidated associates with autophagosomal membranes, and the GFP-tagged protein (GFP–LC3B) has been widely used as a marker for these membranes. Lastly, Atg9 is a transmembrane protein that spans the membrane six times, and has its N- and C-terminal domains in the cytosol. Importantly, it is the only transmembrane Atg protein and as such understanding its localization should provide important information about the membranes contributing to autophagy.
The function of Atg proteins: hierarchal analysis
There are two hierarchies to consider: the one that transmits the induction signal, and the other that occurs on the membrane. Using biochemical approaches, and in particular studying signature phosphorylation events, which report activating phosphorylations, it has been shown that the ULK and Beclin1 complexes are the most upstream of the Atg proteins, directly downstream of nutrient and energy sensors (Figure 2). These kinases (through phosphorylation events) alter the activities of their substrates and directly transmit the signal; however, the substrates of the ULK1/2 protein kinase complex remain uncharacterized. The ubiquitination-like modifications also occur in a biochemically defined sequence, and could also be considered non-enzymatic transmitters of the signal. Atg7 acts as an E1 in both the upstream Atg12–Atg5 conjugate and the LC3–PE conjugation. The transmission of the signal is now described elegantly by the structural analysis demonstrating Atg12 binds Atg3, the E2 of the second conjugation step for LC3–PE formation .
The membrane-mediated hierarchal analysis has been informed by the analysis of the recruitment on to the PAS or phagophore, first performed in yeast, and then in mammals . This analysis confirmed the primacy of the ULK complex in the membrane-recruitment hierarchy, followed by the Beclin1 complex. The WIPI proteins and the Atg12–Atg5–Atg16 complex are then recruited downstream of the ULK and Beclin1 complexes. LC3–PE, and the other family members, are generally agreed to be the last Atg proteins to be recruited to the membrane. Intriguingly, the concept of sequential recruitment has been strengthened by recent papers showing that FIP200, a member of the ULK complex, interacts with Atg16 [21,22]. In addition, ULK1 has also been shown to have an LIR (LC3-interacting region) motif and bind LC3 family members [23,24], suggesting in fact that there may be a reinforcement cycle occurring between the Atg proteins that are sequentially recruited.
The notable exception in the hierarchy is Atg9. In yeast, Atg9 has been shown to be required for the assembly of the PAS, catalysing the delivery of vesicles from the Atg9 compartment [25,26] and is required for expansion of the forming and closing autophagosome. In mammals, in fed cells, Atg9 traffics between the Golgi and endosomes, and in starved cells additionally from the equivalent Atg9 compartment (Figure 1) . An important difference between the yeast and mammalian pathway is the dynamics of Atg9 from the expanding autophagosome: in yeast, Atg9 appears to be delivered to the autophagosome membrane [25,26], although in mammals it appears to only participate in kiss-and-run fusion events, and is not incorporated into the expanding phagophore or autophagosome .
The origin and source of the phagophore
Live-cell imaging, which can be used to visualize the dynamics of the forming autophagosome during amino-acid starvation, has also suggested that these form simultaneously at multiple sites in cells. This implies that PAS, defined as the primary site or origin of the membranes, may also be widely distributed throughout the cell [5–8]. Nonetheless, although a consensus view is emerging of what phagophores are, many questions remain unanswered regarding the origin and source of the membrane for the PAS and phagophore. One significant issue is the lack of a unique marker to distinguish the PAS from the phagophore, and both from later stages of the pathway. As far as we know, the PAS and the phagophore, and the phagophore and the autophagosome, have overlapping compositions. In addition, the relationship between the phagophore and the omegasome is not yet clear, but is of immediate interest to the field. Live-cell imaging, advanced by immunogold labelling and 3D electron tomography, has shown DFCP1-positive domains (the omegasomes) form from ER membranes which are intimately and perhaps physically linked with phagophores [29,30]. DFCP1 is a unique marker for the omegasome, but is not required for autophagy .
Recently, several membranes have been proposed as sites for the nucleation of the phagophore: plasma-membrane-derived vesicles, mitochondria, Golgi and ER (for review see ). Importantly, plasma membrane, mitochondria and ER are all distributed widely throughout cells. Regarding the unexpected sources of membrane: the endocytosed plasma membrane, released by fission of plasma membrane domains which is dependent on the coat protein AP-2 and dynamin, has been shown to provide Atg16-positive vesicles which coalesce through a SNARE (soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptor)-mediated fusion reaction into phagophores [32,33]. Similarly, it was shown that vesicle buds form from mitochondria outer membranes, dependent on mitofusin2, a protein that mediates ER and mitochondrial connections and potentially the production of curvature-initiating lipids. These vesicles are initially Atg5-positive and become LC3-positive . However, in both of these cases, the recruitment of early Atg proteins (ULK complex, Beclin1 complex and WIPI proteins) has not yet been confirmed. The importance of these proteins at early stages has been shown by immunofluorescence and cryo-immunoelectron microscopy in cells expressing a catalytic mutant of Atg4 (which prevents phagophore closure and causes an accumulation of the phagophores by sequestering LC3 and its family members) showing ULK1, WIPI2 and Atg12–Atg5–Atg16 accumulates on phagophores [18,29,35]. Additionally, in mammalian cells, siRNA of ULK1, WIPI2 and its putative interactor Atg2 causes accumulation of Atg9 on phagophores [28,36].
In contrast with plasma membrane and mitochondria, Golgi membranes, implicated in the early cytochemical studies, have been shown in numerous recent studies to be important for autophagosome formation. In particular, these act as a source of Atg9 vesicles and other vesicles whose trafficking is controlled by multi-subunit vesicle-target membrane tethers such as TRAPP (transport protein particle) complexes or the exocyst complex (for details see ).
With regard to the ER as a formation site, several laboratories have shown that phagophores emerge from omegasomes, which are subdomains of the ER [17,29,30]. The ULK1/2 and Beclin1 complex members are the first Atg proteins to be recruited to the subdomains on the ER membrane [37,38]. VMP1 (vacuolar membrane protein 1) is an ER-resident transmembrane protein, and again although not an Atg protein, is required for autophagy under particular conditions, and is recruited to these early phagophores on the ER . In addition, these ER subdomains, which contain the Atg proteins listed above, plus DFCP1 and VMP1, have been shown recently to contain MAM (mitochondria-associated ER membrane) proteins, found in close contact with mitochondria, and regulated by the SNARE protein syntaxin 17  (Figure 3). These data have combined to present a strong case for the ER, or subdomains of the ER, being required for formation of the phagophore, at least during amino-acid starvation. More research remains to be carried out in order to understand how the phagophores can form, and detach from, ER subdomains, and to elucidate what is the composition of the inner and outer membrane of the autophagosome.
Our understanding of the complex biology underlying the essential process of autophagy is rapidly increasing and becoming ever more supported by molecular details. Our knowledge of both the signals sensing alterations in the growth conditions of cells, and the machinery which responds to the signals have clearly demonstrated that autophagy is a tightly regulated process. The most recent results strongly implicate the ER, or subdomains of the ER, in the formation of the phagophore, but questions remain such as: is the ER the PAS, or do separate PAS structures exist which associates with the expanding ER? Further questions regarding, for example, the mechanism of expansion of the ER-derived phagophore into an autophagosome also requires more study. However, there has been significant progress over the last decades and this progress will eventually be translated into a complete understanding of the pathway.
• The PAS is the site of the origin of the autophagy pathway.
• Phagophores originate from sites on the ER, in proximity to mitochondria.
• Membrane for the phagophore growth is supplied by the Golgi, plasma membrane and endosomes.
- © The Authors Journal compilation © 2013 Biochemical Society