Introduction

The word autophagy, a Greek word for self-eating, refers to the catabolic processes conserved in eukaryotic cells by which the cell recycles its own constituents [1]. To date, three major types of autophagy have been described: macroautophagy, microautophagy, and chaperone-mediated autophagy (CMA). Autophagy is involved in the adaptation to starvation, in cell differentiation and development, as well as in the degradation of aberrant structures, the turnover of organelles, tumor suppression, innate and adaptive immunity, life span extension, and cell death [2, 3].

The discovery of ATG (Autophagy-related) genes in yeast has been a milestone in our understanding of how the autophagosome is formed [1, 4]. This is a double-membrane-bound vacuole that sequesters cytoplasmic components either unselectively or selectively via autophagy adaptor proteins [5]. Autophagy also responds selectively toward bacteria and viruses that invade the cytoplasm [6].

Fifteen ATG proteins constitute the core machinery of autophagosome formation [4, 7]. In mammalian cells, these ATG are hierarchically recruited to form a phagophore or isolation membrane that subsequently elongates to form the autophagosome. Recent studies have shown that ATG assembly takes place at the point of contact between the endoplasmic reticulum membrane and the outer membrane of mitochondria [8•]. However, other membranes, such as the endosomes, the Golgi apparatus, and the plasma membrane, also contribute to the biogenesis of the autophagosome [9, 10]. The coordination between these different membranes during autophagosome formation is not yet well understood. Several functional modules involving ATG proteins can be identified during the phases of autophagosome formation, from the initiation step to the elongation/closure step (Fig. 1). Five functional modules have been defined: The first module is the ULK1 complex (ULK1 is the mammalian homolog of yeast Atg1), the second module is the phosphatidylinositol 3-kinase (PIK3) complex I (which contains PIK3C3/Vps34), and Beclin 1 (Beclin 1 or BECN1 being the mammalian homolog of the yeast Atg6). These two complexes (modules) are involved in the initiation of autophagy. Phosphatidylinositol 3-phosphate (PtdIns3P), which is produced by the enzymatic activity of PIK3C3, recruits WIPI1/2 (homologs of yeast Atg18) at the phagophore, and DFCP1 at the ER site of autophagosome formation known as the omegasome [11]. These two PtdIns3P-binding proteins characterize the third functional module. One of the functions of WIPI2 is to control the transport of the multispanning membrane ATG9 from the phagophore to a peripheral endosome/golgi localization [12]. The ATG9 protein makes up the fourth functional module. In yeast, the fusion of Atg9-containing vesicles is important in the very early stages of autophagy [13]. In mammalian cells, the trafficking of ATG9 to the phagophore is also an early event that occurs soon after autophagy induction [14]. The last functional module consists of the two ubiquitin-like conjugation systems: ATG12–ATG5 and the LC3–phosphatidylethanolamine (PE) (LC3 being the mammalian homolog of yeast Atg8), which are involved in the elongation and closure of the autophagosomal membrane [15, 16].

Fig. 1
figure 1

ATG proteins involved in the early steps of mammalian autophagy. Energy depletion, amino acid, glucose and serum starvation all activate the ULK1 complex via two main pathways: (i) via AMPK (AMP-activated protein kinase), which activates ULK1 by phosphorylation, and (ii) via the inhibition of the mTOR (mammalian target of rapamycin) complex, which inhibits ULK1 by phosphorylation. The different modules that constitute the ATG core machinery involved in autophagosome formation are delimited by the gray dotted lines. Narrow arrows indicate the molecular connection between the modules. Thick arrows indicate the autophagic pathway (from the phagophore to the autolysosome). Some of the cross-molecular regulations between modules are listed below. Stimulation of the ULK1 complex induces the class III phosphatidylinositol-3-kinase (PIK3C3) complex, consisting of the core structure (e.g., Beclin 1, Vps15/PIK3R4 and Vps34/PIK3C3), and two regulators, Atg14L, AMBRA1, mainly also via two pathways. In the first pathway, ULK1 phosphorylates Beclin-1, whereas in the second, ULK1 phosphorylates AMBRA1. In turn, AMBRA1 contributes to stabilizing ULK1 by stimulating its ubiquitination. This activation allows phosphatidylinositol-3-phosphate (PI3P) to be generated by the kinase VPS34 on the phagophore to promote the recruitment of the WD repeat domain of PI3P-interacting proteins, such as WIPI1 and 2. The elongation and the closure of the autophagosomal membrane are dependent on two ubiquitin-like conjugation systems. The first results in the formation of the Atg12–Atg5–Atg16L1 complex, which acts as an E3-like enzyme for the second system, which in turn requires the ubiquitin-like protein microtubule–associated protein 1 light chain 3 (LC3) and generates a lipidated (PE)-conjugated form of LC3 (LC3-II). After autophagy has been induced, the proper targeting of the Atg5-Atg12–Atg16L1 complex to the phagophore depends on the ULK1 complex via a direct interaction with FIP200. Moreover, the ULK1 protein contains an LC3-interacting region (LIR motif). The early stages of phagophore formation depend on the vesicular transport of the transmembrane Atg9. The shuttling of Atg9 in and out the phagophore depends on WIPI proteins (such as WIPI2), which control the trafficking from the phagophore to a perinuclear region, and on ULK1 to recruit Atg9 vesicles to the phagophore once autophagy has been induced. Proteins that can be bypassed during the various forms of non-canonical autophagy are boxed in red

Full size image

Recently, non-canonical pathways have been discovered that lead to autophagosomal degradation via variants of the canonical pathway [17•]. These alternative pathways underscore the complexity of the molecular aspects of autophagy, as well as the possibility that autophagy may be modulated in pathological circumstances. Here, we discuss the differences between canonical and non-canonical forms of autophagy in terms of their use of certain ATG and the membrane structures employed, and the consequences that result for physiological homeostasis. In this review we will not consider the recruitment of ATG to single vacuolar membranes, such as the phagosomal membrane emanating from the plasma membrane, or the non-canonical function of ATG. Readers interested in these topics can refer to recent reviews [18–20].

The Canonical Autophagic Pathway

Before we can attempt to define non-canonical autophagy, we first need to describe canonical autophagy. Autophagy is defined as a mechanism that depends on the hierarchically ordered activity of ATG proteins recruited at the phagophore to form an autophagosome (usually a 300-nm to 1-μm diameter vacuole) that will ultimately fuse with the lysosomal compartment. In metazoan cells, the autophagosome may receive input from the endocytic pathway to form an amphisome before merging with the lysosomal compartment [21, 22].

Most of our knowledge about the formation of autophagosomes is based on results obtained after autophagy has been stimulated by incubating cells in a nutrient-free medium [23]. Here, we will be referring to such studies unless stated otherwise. The regulation of basal autophagy is an emerging field—and many questions remain to be solved [24].

The Core ATG Machinery and the Formation of Autophagosomes

Autophagy is initiated by the activation of the ULK1 complex, which contains the serine/threonine kinase ULK1 or -2, ATG13, a 200-kD focal adhesion kinase family interacting protein (FIP200), and ATG101. Once autophagy has been induced, this complex localizes at the site of phagophore formation to regulate the nucleation machinery [25]. Phagophore nucleation is highly dependent on the production of PtdIns3P by PIK3C3. PIK3C3 forms the core PIK3C3 complex I with its adaptor, Vps15/PIK3R4, Beclin 1, and ATG14 [26]. Several proteins, such as AMBRA1 or Bcl-2, can either activate or inhibit the production of PI3P by PIK3C3 complex I [25]. ULK1 activates PIK3C3 complex I by phosphorylating Beclin 1 and AMBRA1 [27, 28•]. In turn, AMBRA1 interacts with the E3-ligase TRAF6 to induce the ubiquitination of ULK1, thus increasing its stability and functional efficiency [29•]. ULK1 can also modulate the activity of other modules of the ATG core machinery by controlling the vesicular transport of ATG9 [12], and by interacting with ATG8 homologs via an LC3-interacting region (LIR motif) [30]. Another component of the ULK1 complex, FIP200, interacts with ATG16L1 [31•, 32•], which is an element in the ubiquitin-like cassette of the ATG core machinery. The production of PI3P in the phagophore membrane allows the recruitment of the WD repeat domain PI3P-interacting proteins WIPI1 and WIPI2 to occur [33]. Both these proteins contribute to the expansion and the closure of the vesicle in concert with two ubiquitin-like conjugation systems, resulting in the ATG12–ATG5–ATG16L complex, and the formation of the phosphatidylethanolamine (PE) conjugate of microtubule-associated protein light chain 3 (LC3; the mammalian ortholog of Atg8 in yeast) [15, 16]. The SNARE protein, VAMP7, and its partner SNAREs (syntaxin7, syntaxin8, and Vti1b), have been found to regulate the homotypic fusion of ATG16L1-positive vesicles after internalization from the plasma membrane [34]. It has been suggested that these vesicles are precursor elements of the phagophore [35]. The transmembrane protein ATG9 is also involved in the nucleation of the phagophore membrane by cycling between different compartments and the phagophore [14].

Maturation of Autophagosomes

Autophagosome maturation and fusion with the lysosome occurs in the vicinity of the centrosome and depends on several lysosomal membrane proteins, such as the small GTPase Rab 7, and the transmembrane lysosome-associated membrane protein 2 (LAMP2) [36]. These fusion events are also dependent on SNAREs. VAMP3 contributes to the fusion of multivesicular bodies with autophagosomes to form amphisomes [37]. Recently, syntaxin 17 has been shown to be targeted to autophagosomes, where it fuses with endosomes/lysosomes [38]. Syntaxin 17 also plays a role during the early stages of autophagosome formation [8•]. These findings suggest that this SNARE is involved in several different steps of the autophagic pathway. The autophagy cargoes are then degraded by the acid hydrolases that are present in the lysosomal lumen.

Non-Canonical Autophagy

During non-canonical autophagy, the formation of the double-membrane-bound autophagosome does not require the hierarchical intervention of all the ATG proteins, whereas canonical autophagy does. Furthermore, the double membrane does not necessarily elongate from a single source [17•].

Beclin 1-Independent and PIK3C3-Independent Autophagy

Various different Beclin 1–PIK3C3 complexes function either by promoting the induction of autophagy (in concert with ATG14), or by regulating the maturation of autophagosomes and endosomal trafficking (in concert with UVRAG and Rubicon) [25, 39]. However, the pathway that leads to autophagic degradation via Beclin 1–PIK3C3 in response to specific stimuli is not obligatory under all circumstances [17•, 40]. Non-canonical, Beclin 1-independent autophagy has been reported after cells have been treated with pro-apoptotic compounds (Table 1). The following examples were extracted from (Table 1) and are discussed in Ref. [17•]: Beclin 1-independent autophagy mediated by the neurotoxin 1-methyl-4-phenylpyridinium is associated with neuronal cell death [41•], and resveratrol-mediated autophagy is reported to be positively correlated with a type of human tumor cell death that is independent of Beclin 1, but dependent on the ubiquitin-like conjugation system [42]. Furthermore, Z18, a compound that targets the BH3-binding groove of Bcl-XL/Bcl-2, has been shown to induce Beclin 1-independent autophagosome formation in HeLa cells [43], and other pro-apoptotic compounds, such as staurosporine, MK801, and etoposide, have been found to induce Beclin-1 independent autophagy in primary cortical neurons [44]. These studies suggest that it might become feasible to use pro-death compounds that induce non-canonical autophagy for therapeutic purposes in cancer when the functions of canonical autophagy proteins are compromised. However, Beclin 1-independent autophagy has also been observed in settings unrelated to cell death, such as differentiation [45], bacterial toxin uptake [46], and during viral infection [47]. Interestingly, Beclin 1-independent autophagy is not synonymous with pathways that exclude the PIK3C3–WIPI–ATG5-LC3 route. Exposure of human tumor cells to arsenic trioxide [48] or gossypol promotes autophagy that is PIK3C3-dependent, but Beclin 1-independent, and which involves WIPI-1 under some circumstances [49]. Recently, resveratrol has been found to promote WIPI-1-dependent LC3 lipidation in the absence of induced phagophore formation, indicating that different membrane sites may be used during non-canonical autophagosome formation [50]. In fact, WIPI-1 specifically localizes to both the plasma membrane and the ER in response to the induction of autophagy, indicating that the WIPI–ATG5–LC3 pathway can function at several different membrane sites.

Table 1 Inducers of non-canonical autophagy
Full size table

Some recent studies suggest that autophagy can be independent of Vps34 and Vps15 (Table 2). In Vps34−/− sensory neurons, autophagosomes and LC3-II production are observed [51]. In the same mouse model, autophagy has been reported in T-lymphocytes [52]. However, in other genetic mouse models in which Vsp34 is ablated, autophagy has been shown to be absent or only minimally observed [53, 54]. A recent report shows that Vps15-deficient mouse tissues are capable of forming LC3-positive autophagosomes [55]. However, it is still possible that PI3P is required in these situations. This would suggest that this lipid is produced either by the degradation of PtdIns(3,4)P2/PtdIns(3,4,5)P3, or by the activity of class III phosphatidylinositol 2-kinase (PIK3C2) [56]. This latter possibility would explain why autophagy is sometimes insensitive to classical PIK3 inhibitors (3-methyladenine and wortmannin) [57]. In studies of autophagy induced by glucose starvation, it has also been shown recently that PtdIns3P is not required to initiate autophagy, which correlates with the absence of WIPI puncta on the phagophore [58].

Table 2 Detection of non-canonical autophagy in vitro and in vivo by the use of knock-out mouse models for ATGs
Full size table

Further Non-Canonical Routes of Autophagosome Formation

A form of autophagy that bypasses the canonical ULK1 initiation step has recently been reported to occur in response to ammonia or glucose deprivation [59•]. Interestingly, Gammoh et al. [31•] confirmed this non-canonical autophagy and explained, by identifying a FIP200-binding domain in ATG16L1 (FBD), how we could distinguish between ULK1-dependent and -independent autophagy processes. Unlike amino acid starvation-induced autophagy (ULK-dependent autophagy), glucose deprivation-induced autophagy (ULK-independent autophagy) is independent of the FBD. In the wake of this finding, various studies have shown that autophagosome formation does not require AMPK activity, and cannot be inhibited by mTOR (for example, see [60]). These findings demonstrate that forms of non-canonical autophagy can also bypass the AMPK–mTOR–ULK route of initiation, indicating that autophagic sequestration can employ a variety of routes of entry in addition to the evolutionarily-conserved routes initially identified.

Although the formation of LC3-II is not observed in Atg5−/− cells, transition electron microscopy reveals the presence of autophagosomes and amphisomes (pre-autolysosomal vacuoles that are formed by autophagosome–endosome fusion) in Atg5−/− cells treated with etoposide over a prolonged period of time [61•]. Based on the unusual lamination of the membrane forming the phagophores and autophagosomes in this context, the trans-Golgi network (TGN) seems to be the membrane source for this form of alternative autophagy, which is initiated by ULK1 and Vps34 complexes in a manner similar to canonical autophagy [17•]. However, the elongation of the initial autophagosomal membrane does not require ATG9 or the ATG proteins of the ubiquitin-like conjugation system (ATG7, ATG5, and LC3). Instead, the monomeric GTPase Rab9, which is involved in vesicular trafficking between the TGN and late endosomes, is required to elongate the initial autophagosomal membrane through fusion events with Rab9-positive vesicles. The resulting non-canonical, double-membraned autophagosome can mature and fuse with the lysosomal compartment to deliver cargo for degradation.

As previously described in Ref. [17•], this form of autophagy has been observed in various cell types and embryonic tissues, and plays a role in the removal of mitochondria during erythrocyte maturation in vivo [61•]. However, in erythroblasts mitochondria are also degraded by autophagy in a manner dependent on ATG5 and ATG7 [62]. These results suggest that both canonical autophagy and non-canonical autophagy may contribute to the elimination of mitochondria during erythroblast differentiation. Further studies are needed to clarify the physiological relevance of the alternative form of autophagy. Partners of Rab proteins, such as the SNARE proteins, have recently been shown to promote early fusion steps in the biogenesis of autophagosomes [34, 63]. However, important functions of the LC3 family proteins in various different aspects of autophagy (such as the completion of autophagosome formation [64], the recruitment of selective cargoes via the interaction of LC3 with LIR-containing proteins [65], and the regulation of the maturation step via the interaction of LC3 with the Rab-GTPase-activating protein OATL1 [66] on the outer membrane of autophagosomes) remain to be elucidated. Nor is it yet clear whether all these functions are conserved in this alternative form of autophagy.

Recently, a form of autophagy that is independent of Atg3 and Atg7 has been reported in the programmed reduction of cell size during intestinal cell death in Drosophila [67•]. This non-canonical autophagy involves Atg8 and Uba1, the E1 enzyme used in ubiquitylation. However, Uba1 is not a substitute for the E1 activity of Atg7 in the conjugation of Atg8 to PE (Atg3 functions as the E2-conjugating enzyme in the formation of Atg8-PE). As the authors propose, Uba1 could function at a different stage of the autophagy pathway [67•]. The mechanism by which Atg8 is recruited to the autophagosomal membrane remains to be elucidated. This study provides important evidence that non-canonical forms of autophagy exist in different phyla, suggesting that non-canonical forms of autophagy have not in fact emerged recently in the course of evolution.

Autophagosomes Formed from Multiple Isolation Membranes

Cells can respond to the invasion of the intracellular milieu by pathogens via a form of selective autophagy known as xenophagy, which is intended to prevent this invasion. In some cases, the enveloping autophagosome does not undergo conventional biogenesis.

The group A Streptococcus (GAS) bacteria are pathogens that cause a wide range of infectious diseases in human beings. GAS bacteria invade non-phagocytic cells via the endocytic pathway, gaining access to the cytoplasm by secreting the toxin streptolysin O to disrupt the endosome membrane. Once they enter the cytoplasm, GAS bacteria are trapped in an autophagosome-like compartment [68]. The formation of the autophagosome-like vacuole containing GAS, depends on the autophagic machinery, i.e., on ATG proteins, and on the recruitment of these ATG proteins in a canonically hierarchical sequence. In contrast to the biogenesis of autophagosomes, GAS are surrounded by several isolation membranes. The fusion of these membranes generates a 10-μm, autophagosome-like vacuole that sequesters chains of GAS bacteria. ATG5 is transiently associated with these isolation membranes before LC3 is recruited, something that is also known to occur in canonical autophagy. The completion of the GAS-containing, autophagosome-like vacuole requires the recruitment and activity of the monomeric GTPase, Rab7 [69], which regulates several steps in endosomal trafficking. Interestingly, Rab7 is recruited to the surface of autophagososmes and allows them to fuse with the endosome and lysosomal compartments [36]. Rab7 could therefore play a role in the homotypic membrane fusion required for the autophagosome-like vacuole to be formed. However, Rab7 could also have other functions, such as generating and stabilizing precursor membranes.

Conclusion

Non-canonical autophagy pathways and structures have the same function as canonical autophagy in sequestering some of the cytoplasm and compartmentalizing pathogens. In many cases, material sequestered by non-canonical autophagy is ultimately degraded in the lysosomal compartment. Non-canonical autophagy, which requires only a subset of ATG proteins, has been observed in various different settings. The first thing we need to understand is how some ATG modules can be bypassed to form a functional autophagosome. In this regard, it is interesting to note that the identification of a new ATG16L1 binding site ([YW]X(2,6)[ED]X(2,6)[YWF]X2L) in several proteins, such as NOD2 or TLR2, which is known to play a role in bacterial autophagy [70•], could now explain how ATG16L1 triggers LC3 lipidation in ectopic membrane localizations by bypassing the upstream autophagic complexes [71]. This Atg16L1-binding motif, which recognizes the C-terminal WD-repeat domain in ATG16L1, might be a signature of the non-canonical autophagic pathway.

Alternatively, PtdIns3P generation seems to be essential for autophagosomes to be generated, whereas autophagy can occur in the absence of Beclin 1 and also of PIK3C3 and Vps15. Further, the PtdIns3P effector WIPI1/2 is necessary to initiate LC3 lipidation in non-canonical autophagy in many settings [17•]. Thus, it can be surmised that PIK3C3-independent PtdIns3P -providing routes might be involved in the formation of the autophagosome [56, 72]. One important question is whether non-canonical forms of autophagy have specific functions in cell physiology or pathological situations. From the literature available about Beclin 1-independent autophagy, it is difficult to identify a single function corresponding to this form of non-canonical autophagy. For example, Beclin-1 independent autophagy has been reported in various different contexts, such as cell survival, death, and proliferation, and immune cell development and proliferation [17•, 40]. Beclin 1-independent autophagy is the most commonly reported non-canonical form of autophagy. Beclin 1 is a key autophagy protein with a large interactome [73, 74], and is involved in both the formation and maturation of autophagosomes, depending on the interacting partners [25, 39]. Moreover, Beclin 1 is targeted by many viruses in attempts to block autophagy at different stages. Beclin 1-independent autophagy could, therefore, be an evolutionary adaptation that prevents the blockade of autophagy by an invading virus, which would otherwise compromise cell survival. Another intriguing matter the relationship between non-canonical autophagy and selective forms of autophagy, such as mitophagy, that specifically sequesters mitochondria [75]. ATG5- and ATG7-independent autophagy is known to be involved in the clearance of mitochondria during erythroid maturation [61•], and determining the role of the different ATG proteins in the various forms of selective autophagy may provide a way to identify selective molecular signatures of autophagy that lie beyond the selective recognition of cargo [76]. In addition, an important question is whether both non-canonical autophagy and canonical autophagy can be activated in a coordinated manner in the same cell in response to a stressful situation. Being able to distinguish between the non-canonical autophagic pathway and the canonical autophagic pathway by means of specific markers would make a significant contribution to the molecular understanding of how autophagy is regulated and how it functions.