Autophagosome formation in mammalian cells
- First Online:
- Cite this article as:
- Burman, C. & Ktistakis, N.T. Semin Immunopathol (2010) 32: 397. doi:10.1007/s00281-010-0222-z
- 744 Views
Autophagy is a fundamental intracellular trafficking pathway conserved from yeast to mammals. It is generally thought to play a pro-survival role, and it can be up regulated in response to both external and intracellular factors, including amino acid starvation, growth factor withdrawal, low cellular energy levels, endoplasmic reticulum (ER) stress, hypoxia, oxidative stress, pathogen infection, and organelle damage. During autophagy initiation a portion of the cytosol is surrounded by a flat membrane sheet known as the isolation membrane or phagophore. The isolation membrane then elongates and seals itself to form an autophagosome. The autophagosome fuses with normal endocytic traffic to mature into a late autophagosome, before fusing with lysosomes. The molecular machinery that enables formation of an autophagosome in response to the various autophagy stimuli is almost completely identified in yeast and—thanks to the observed conservation—is also being rapidly elucidated in higher eukaryotes including mammals. What are less clear and currently under intense investigation are the mechanism by which these various autophagy components co-ordinate in order to generate autophagosomes. In this review, we will discuss briefly the fundamental importance of autophagy in various pathophysiological states and we will then review in detail the various players in early autophagy. Our main thesis will be that a conserved group of heteromeric protein complexes and a relatively simple signalling lipid are responsible for the formation of autophagosomes in mammalian cells.
KeywordsAutophagyPhosphoinositideSignallingEndoplasmic reticulumPI 3-kinase
Autophagy and disease
Basal autophagy plays an important role in intracellular quality control, this is especially important in non-dividing cells such as neurones and myocytes. Mice with a neural specific deficiency of the autophagy gene, Atg5, display inclusion bodies and diffuse ubiquitinated proteins in their neurones, and have abnormal motor functions . Moreover, mice lacking Atg7 specifically in the central nervous system, also displayed a build up of inclusion bodies and ubiquitinated proteins in their neurones, and exhibited both behavioural and motor defects . Autophagy also plays an important role in the removal of disease-related mutant proteins that build up in neurodegenerative diseases such as Huntingtin’s disease, spinocerebellar ataxia and Parkinson’s disease (reviewed in ) .
Autophagy is also used by the cell to deliver microorganisms to lysosomes for their destruction (known as Xenophagy). Interestingly, pathogen-containing autophagosomes have been found to be larger than classical autophagosomes, suggesting that the autophagic process can adapt to engulf microbes that are larger than cellular organelles . Increasing amounts of evidence also suggests a role for autophagy in delivery of microbial antigenic material to the adaptive and innate immune systems. For example, the autophagic machinery is used for the major histocompatibility complex class II presentation of some endogenously synthesised viral antigens . In terms of the innate immune system, autophagy is also required for the delivery of viral RNA to the endosomal toll-like receptor TLR7, and subsequent activation of type I interferon signalling . However, some bacteria such as Brucella abortus, Porohyromonas gingivalis, and Staphylococcus aureas can impair the fusion of the autophagosome with the lysosome. They then reside in the autophagosome and utilise nutrients from sequestered cytoplasmic material for their own replication . It is also common for pathogens to block the signalling pathways that regulate autophagy, or block the membrane trafficking events of autophagy in order to evade autophagy-mediated lysosomal elimination. For example, in herpes simplex virus encephalitis Beclin1, a protein essential for autophagy (see below), is inhibited by a viral neurovirulence protein ( ; reviewed in ).
Autophagy can act as a tumour suppressor by removing damaged organelles and growth factors, and by reducing chromosomal instability. Autophagy malfunction is prevalent in many cancers. In fact, in 40–75% of human breast, ovarian, and prostate cancers the Beclin1 gene is monallelically deleted. It is thought that the autophagy proteins Beclin1 and Atg5 act as ‘guardians’ of the cellular genome. Specifically, it has been demonstrated that immortalised epithelial cells with loss of Beclin1 or Atg5 display increased DNA damage, gene amplification and aneuploidy, in parallel with increased tumourigenicity . Paradoxically, the cytoprotective role played by autophagy can be counter-productive in cancer, as it can help cancer cells resist anti-cancer therapy and survive in conditions of low nutrient supply (reviewed in ).
PI3P, a simple signalling lipid with a dynamic regulation and a significant number of effectors
The phosphoinositide PI3P has been found to play important roles in both nutrient sensing and autophagogenesis. Phosphoinositides (PIs) are formed by the phosphorylation of phosphatidylinositol on its inositol ring. With the exception of the 2′ and 6′ positions all free OH groups on the inositol ring can be phosphorylated. The unique function of each type of PI can be attributed to the distinct arrangement of phosphate groups on the inositol ring. The enzymes responsible for phosphorylating the 3′-OH position of PI are known as phosphoinositide 3-kinases (PI3 kinases), and three classes of these kinases exist in cells [15, 16]. Class I PI3 kinases commonly phosphorylate PI [4, 5] P2 to produce PI [3–5] P2 (often referred to as PIP3). Class II PI3 kinases can use PI, PI  P and PI [4, 5] P2 as substrates, although they exhibit a preference for PI. Class III kinases also phosphorylate PI to produce PI3P, these enzymes are orthologues of the yeast vesicular protein-sorting protein Vps34. It is thought that the class III enzyme is responsible for most of the PI3P synthesis within cells, whereas the class II PI3 kinase is involved in specialised signal-dependent production of PI3P . PI3P can also be further phosphorylated by 4’-kinases and 5′-kinases to generate PI [3, 4] P2 and PI [3, 5] P2, respectively . PI3P is maintained at a cellular concentration of approximately 200 μM. PI3P can be dephosphorylated by phosphatases including PTEN (phosphatase and tensin homologue; ), myotubularin proteins such as Jumpy [20, 21] and MTMR3 , and the yeast phosphatase, Sac1p .
PI3P binding can affect the localisation, conformation and/or activity of PI3P-binding proteins, and two types of PI3P-binding domains have been reported. One is the FYVE domain (whose name reflects the first four proteins found to contain it: Fab1p, YOTB, Vac1p, early endosome antigen 1; [24, 25]). FYVE domains are known to occur in 37 human proteins, they contain a characteristic basic motif [(R/K) (R/KHHCR], which binds to the inositol head group of PI3P with high affinity (Kd = 50 nM; [16, 26]). The intracellular location of PI3P is often studied using probes constructed of double FYVE domains. The second PI3P-binding domain is the 120 amino acid Phox homology (PX) domains. PI3P binding is thought to occur via a pair of highly conserved basic motifs consisting of [RR(Y/F)]. PX domains have been found in NADPH oxidase subunits, a PI3-kinase, sorting nexins, a SNARE, as wells as some phospholipids and protein kinases . As well as playing important roles in autophagogenesis and nutrient-sensing PI3P binding proteins also play important roles in endocytic trafficking, retrograde trafficking from endosomes to Golgi and phagocytosis .
Autophagy induction/nutrient sensing
Amino acid stimulation causes TORC1 to translocate to lysosomes where the Rag guanosine-5'-triphosphate (GTP)ases reside . The GTP/guanosine diphosphate (GDP) loading of Rag GTPases is affected by amino acids, and GTP loading of the Rag GTPases promotes the interaction of the Rag heterodimers with TORC1 . A trimeric complex of proteins comprised of MP1, p14 and p18, known as the Ragulator is also localised to lysosomes and co-immunoprecipitation experiments revealed that it interacts with both Rag GTPases and TORC1 . The Ragulator complex was found to be essential for TORC1 activation by amino acids in both mammalian and Drosophila cells. Mice lacking p14 or p18 have growth defects and die around embryonic day 7.5-8 [32, 33], and humans with a mutation that reduces p14 expression exhibit strong growth retardation . The Ragulator localises the Rag GTPases to lysosomes, as in cells lacking members of the Ragulator the Rag GTPases localised to small puncta throughout the cytoplasm rather than to lysosomes. Furthermore, in cells lacking components of the Ragulator, TORC1 also failed to translocate to lysosomes upon amino acid stimulation. Forced lysosomal targeting of TORC1 eliminated the amino acid sensitivity of TORC1 (i.e. S6 kinase was phosphorylated in the absence of amino acids; ). It has previously been reported that the Rag GTPases mediate the intracellular trafficking of TORC1 to late endosomal compartments containing one of its activators, the small GTPase-activating protein Rheb [31, 35, 36]. In the above study, TORC1 and Rheb were both forced to localise to the plasma membrane, this also rendered TORC1 insensitive to amino acid stimulation, indicating that its co-localisation with Rheb is enough to activate TORC1 . In summary, amino acid stimulation leads to the translocation of TORC1 to lysosomal membranes, at the lysosome Rag GTPases (localised to lysosomes by Ragulator) serve as docking sites for TORC1, TORC1 can then encounter and become activated by Rheb.
Vps34 has been shown to be a positive regulator of TORC1 following amino acid stimulation. Interestingly, the late endosomal/lysosomal compartment that is thought to be important for amino acid sensing by TORC1 (see above) also contains Vps34, and can be regulated by Vps34 and PI3P [27, 30, 37–41]. Amino acid addition to starved cells leads to an increase in Vps34 activity, and Vps34 can be inhibited by amino acid deprivation [42–44]. It has been observed that TORC1 signalling is increased upon over expression of Vps34. Conversely, inhibition of Vps34 by inhibiting antibodies or siRNA leads to inhibition of TORC1 signalling, as does sequestration of PI3P by over expressing FYVE domains [42–44]. Since Vps34 is not inhibited by the TORC1 inhibitor rapamycin, Vps34 must therefore lie upstream of TORC1.
Growth factors such as insulin can also signal to TORC1 activation, and inhibition of autophagy. Growth factors bind to and activate their specific receptor tyrosine kinases, this leads to the activation of class I P1 3-kinases and the subsequent production of PIP3. An increased amount of PIP3 leads to the membrane recruitment and activation of the serine/threonine kinase Akt, which can bind to PIP3 via its PH domain. Akt can then inhibit the TSC1/TSC2 (tuberous sclerosis) complex. This complex acts as a GTPase activating protein (GAP) for the Rheb GTPase, resulting in GDP-loading of Rheb. Therefore, Akt mediated inhibition of TSC1/TSC2 leads to GTP-loading of Rheb, which activates TORC1 kinase activity. Levels of PIP3 can be regulated by the phosphatase PTEN (phosphatase and tensin homologue deleted on chromosome ten), which plays an important role in controlling PI 3-kinase-mediated TORC1 activation (reviewed in ).
Low energy levels (ATP) can also signal to autophagy. Energy levels are primarily sensed by AMPK kinase, which is activated by increased ratios of AMP to ATP. When AMP is bound to the regulatory subunit of AMPK the serine/threonine kinase LBK1 is then able to phosphorylate and activate the catalytic subunit of AMPK. Under low energy conditions, activated AMPK phosphorylates TSC2, which activates the GAP activity of TSC1/TSC2. TSC1/TSC2 then acts as a GAP towards Rheb, promoting its GDP loading and inactivation of TORC1 . Moreover, AMPK can also phosphorylate Raptor, a member of the TORC1 complex, which disrupts the binding of TOR to Raptor  (reviewed in ).
The pre-autophagosomal structure
The fundamental process of autophagy is conserved from yeast to mammals. Studies utilising yeast genetics have so far identified 31 Atg [4, 7]. The 31 Atg genes can be subdivided into five subgroups: the Atg1 kinase and its regulators, the autophagy-specific class III PI 3-kinase complex, the Atg12 conjugation system, the Atg8 conjugation system, and a final subgroup including the Atg18-Atg2 conjugate and also Atg9 which interact with each other.
A clear difference between yeast and mammalian autophagy is the presence of a pre-autophagosomal (PAS) in yeast (reviewed in ). This structure appears as a dot proximal to the vacuole under nutrient-rich conditions, and labels with Atg8, a known marker of membrane dynamics during autophagy . Time-lapse microscopy revealed that autophagosomes originate from the PAS . Moreover, upon starvation, Atg8 translocates from the PAS to autophagosomes [50, 51]. The PAS is thought to be required for assembly of the Atg proteins prior to autophagosome formation. It has been suggested that the Atg proteins localise at the PAS in a hierarchical manner [48, 52]. For example, localisation of the Atg proteins at the PAS is severely disrupted in the absence of Atg17, suggesting that Atg17 acts as a scaffold for the other Atg proteins at the PAS. Many of the yeast Atg proteins have mammalian orthologues, and these are often localised to the isolation membrane.
A protein complex composed of Atg1 (serine/threonine kinase), Atg13 (scaffold protein), Atg17, Atg29 and Atg31 locates to the PAS and functions in the initial step of autophagosome formation. Active TORC1 can phosphorylate Atg13 at multiple serine residues, this can destabilise the Atg1 complex, leading to the subsequent inactivation of Atg1, and the inhibition of autophagy. Upon starvation and TORC1 inactivation Atg13 is dephosphorylated, and Atg1 and Atg13 associate. This results in translocation to pre-autophagosomal structures, activation of Atg1 and induction of autophagy. Importantly, it has been shown that the interaction of yeast Atg1, Atg13 and Atg17 is enhanced by nutrient starvation [53, 54]. Atg17, Atg29 and Atg31 form a ternary complex, which associates with the Atg1-Atg13 complex in starvation conditions. Several Atg proteins have been found to be phosphorylated by Atg1, although the functional consequence of this phosphorylation remains elusive .
Orthologues of the Atg1 complex proteins also exist in mammals, namely Ulk1 (Atg1), mAtg13 (Atg13) and FIP200 (Atg17; ; Fig. 2). A fourth small novel protein, Atg101, also exists in this complex in mammals [56, 57]. It has been found that both mAtg13 and FIP200 can enhance Ulk1 kinase activity [58, 59]. One major difference between the yeast and mammalian complexes is that the mammalian complex forms a stable complex irrespective of nutrient conditions . However, TORC1 is only incorporated into the mammalian complex in nutrient rich conditions . It is thought that TORC1 phosphorylates both ULK1 and mAtg13, and that this leads to inhibition of the complex translocation to autophagosomal structures. Conversely, under starvation conditions (when TORC1 signalling is suppressed and it is no longer present in the ULK1-Atg13-FIP200 complex), Ulk1 can phosphorylate and activate Atg13 and FIP200, the complex localises to isolation membranes and autophagy ensues [45, 58, 61, 62].
Atg9 is the only integral membrane protein among all the Atg proteins. It is located both at the PAS, and at dot structures moving around the cytoplasm. The N-terminus of Atg9 interacts with Atg11 and Atg17, this interaction targets Atg9 to the PAS, and is essential for autophagy [63, 64]. It is thought that Atg9 cycles between the cytoplasmic pool and the PAS, and Atg9 brings supplies membranes from the cytoplasmic pool to the PAS (reviewed in ). Mammalian Atg9 spans the membrane six times, and in nutrient replete conditions is located on the trans-Golgi network and late endosomes. Upon starvation it translocates and partially overlaps with autophagosomes. This is dependent on both ULK1 and Vps34 .
Ubiquitin-like conjugation systems
Two ubiquitin-like protein conjugation systems exist among the Atg protein subgroups, and both are conserved from yeast to mammals (Fig. 2). The two systems both localise to autophagic membranes, including the PAS. Atg8 is synthesised as a precursor, with additional amino acids at the carboxyl-terminus. These additional residues are cleaved by the protease Atg4, to leave an exposed a glycine residue. This glycine residue later becomes conjugated to phosphatidylethanolamine (PE) in reactions involving Atg7 and Atg3. The synthesis and lipidation of Atg8 is increased in autophagy-inducing conditions . The PE-conjugated form of Atg8 (also known as LC3-II in mammals) acts as a marker for researchers to study the progression of autophagy in both yeast and mammalian cell types. Moreover, yeast Atg8 mutants exhibit significant defects in autophagosome formation. Interestingly, Atg8-PE can form oligomers, it is known that this form of Atg8 can cause liposome aggregation and hemifusion (fusion between outer leafs of membranes), and that this may explain the role played by Atg8 in expansion of autophagic membranes .
Atg12 becomes conjugated to Atg5 via reactions catalysed by Atg7 and Atg10. The Atg12–Atg5 conjugate further interacts non-covalently with Atg16. This can then oligomerise to form a much larger (350 Kda in yeast and 800 Kda in mammals) complex . In yeast, it is known that Atg16 is required for the localisation of the Atg12-Atg5 conjugate at the PAS . This complex also specifically localises to the outer surface of the isolation membrane. Studies using GFP-labelled Atg5 found that Atg5 leaves the isolation membrane immediately before or after completion of the autophagosome . Interestingly, it has been shown that yeast Atg12-Atg5 stimulates the formation of Atg8-PE in vitro, and that Atg12-Atg5 can interact with Atg3 to enhance its activity . Atg16 not required for this effect in vitro, but is required for Atg8-PE formation in vivo . It has been speculated that Atg16 directs the spatial location of Atg8 lipidation. Interestingly, LC3-II also seems to be essential for formation of the Atg12–Atg5–Atg16 complex. Since in Atg3-deficient mice, where no LC3-II is detected, Atg12–Atg5 conjugation is dramatically reduced .
PI3P involvement in autophagy
As mentioned above, Vps34 acts as a positive regulator of TORC1 during amino acid stimulation, and this leads to inhibition of autophagy. Paradoxically, it is also well established that PI3P synthesised by Vps34 plays an essential role in autophagy induction in both yeast and mammals (Fig. 2). This phenomenon may occur via spatially distinct Vps34 complexes that respond differently to amino acid levels. A wealth of evidence supports a role played by Vps34 in autophagy. For example, deletion of the gene encoding Vps34 in yeast inhibits autophagy. Moreover, similar effects were seen when the wild type allele of Vps34 was replaced with a lipid kinase domain mutant of Vps34 [40, 71, 72]. Drosophila loss of function and kinase dead Vps34 also exhibit severe defects in starvation-induced autophagy . Experiments using siRNA against mammalian Vps34 also show that it is required for autophagy in mammalian cells [37, 74–76]. Additionally, treatment with wortmannin or 3-methyladenine, both of which inhibit Vps34 activity, blocks autophagy in a variety of different cell types [27, 75–81]. Studies using double FYVE domain containing probes to track PI3P during autophagy have found that PI3P is localised on autophagosomes [37, 73, 82], and that it is also transported to the yeast vacuole via autophagosomes [40, 72].
Our recent work [37, 83] indicated a possible connection between ER and omegasomes (Figs. 2 and 3.), and recently two studies have confirmed a physical association between the ER and early autophagic membranes [84, 85]. Hayashi-Nishino et al. utilised a mutant form of Atg4 that causes defects in autophagosome formation, resulting in the accumulation of unclosed autophagosomal membranes and isolation membranes. In both mutant Atg4 expressing NIH3T3 cells, and control cells, ER was found to adhere to both the outer and inner surfaces of the cup-shaped isolation membranes. However, the ratio of ER–isolation membrane (ER–IM) complexes to total autophagic structures was increased in Atg4 mutant cells compared to control cells. This association disappeared when the isolation membranes matured into autophagosomes. The authors also expressed GFP-DFCP1 in both control and Atg4 mutant cells, and upon culture in nutrient-free conditions GFP-DFCP1 translocated to punctate structures. Immunoelectron microscopy found that immunogold particles against GFP-DFCP1 strongly labelled ER–IM complexes. The authors concluded that the ER–IM complex is related to the omegasome. Furthermore, 3D electron tomography found that a spherical ER structure was present on both the inside and outside of the isolation membrane. Importantly, the spherical ER structures are considered subdomains of the ER from which the ER–IM complexes originate. This study found that a single narrow membrane extension from the isolation membrane connects to the ER (; reviewed in) ). In parallel experiments also utilising 3D tomography of NRK cells, Yla-Anttila et al. also observed that the isolation membrane is lined by ER on both sides. In contrast to Hayashi-Nishino et al., they found that the isolation membrane was connected to the ER in multiple locations, with as many as fourteen connections between one isolation membrane and the ER ; reviewed in ). Although the localisation of PI3P was not investigated in these studies, it is clear that a modified subdomain of the ER provides a platform for the formation of autophagosomes.
Distinct Vps34-containing complexes
The situation in yeast
As well as playing very important roles in autophagy, PI3P and its synthesising enzyme Vps34 are also involved in endocytic/phagocytic trafficking. Given this multi-functionality, it is not surprising that several distinct Vps34-containing complexes exist in cells. Vps34 was originally discovered in yeast, and studies in yeast have identified two Vps34-containing complexes. Complex 1 is composed of Vps34, and its regulatory protein Vps15, as well as two accessory proteins Vps30 and Atg14, this complex is essential for autophagy. Complex 2 comprises Vps34, Vps15 and Vps30, but Atg14 is replaced by Vps38, and this complex is essential for the endosomal Vps pathway, but dispensable for autophagy. The two complexes are characterised by the mutually exclusive expression of Atg14 or Vps38. It has been shown in yeast that deletion of Vps30, Vps15 or Vps34 leads to both an inhibition of the sorting of the lysosomal hydrolase CPY (carboxypeptidase Y), as well as an inhibition of autophagy . Vps15 is a putative serine/threonine kinase, and Vps34 activity is abolished in yeast expressing ΔVps15 or kinase-dead Vps15 . However, it has been shown that Vps15 does not phosphorylate Vps34 . Consistent with its presence in Complex 1, autophagy is severely disrupted in ΔAtg14 yeast. Conversely, Vps38-deficient yeast is defective in both vacuolar protein sorting and retrograde transport [71, 89].
Orthologues of yeast Vps15, Vps30 (known as Beclin1 in mammals), Vps34, Vps38 (UV radiation resistance-associated gene (UVRAG) in mammals) and Atg14 exist in mammals and other higher organisms. As aforementioned, Vps34 is essential for autophagy in mammalian cells. Similar to the situation in yeast, Vps15 is required for optimal Vps34 activity in mammalian cells . Additionally, studies using Drosophila revealed that autophagy is dramatically impaired in ΔVps15 flies . Vps34 and Vps15 form a membrane-associated complex, and the membrane-targeting of Vps34 requires Vps15. Co-immunoprecipitation experiments revealed that Vps15 enhances the ability of Beclin1 and UVRAG to interact with Vps34. Moreover, these two proteins could only enhance the activity of Vps34 in the presence of Vps15 .This is despite the fact that Beclin1 can directly bind to Vps34 . Interestingly, expression of Beclin1/UVRAG was also found to enhance Vps34-Vps15 binding . Additional roles played by Vps15 include binding to Rab5 and Rab7, which helps to localise Vps34 to early and late endosomes .
Beclin1 was first identified as a Bcl2 interacting protein . Co-immunoprecipitation experiments revealed that Vps34 and Beclin1 interact in mammalian cells [92, 95]. In fact, it has been estimated that 50% of mammalian Vps34 is in a complex with Beclin1. Conversion to the mature form of cathepsin D is often used to study trafficking from the trans-Golgi network to lysosomes. In order to be converted to its mature form, procathepsin D is trafficked from the trans-Golgi network to late endosomes and lysosomes, and this requires active Vps34. However, in mammalian cells, and in contrast to the situation in yeast, Beclin1 is not required for cathepsin D maturation [92, 95]. Similar to yeast, Beclin1 is also required for autophagy in mammalian cells, as assessed by reduced LC3-II conversion in cells treated with Beclin1 siRNA . Moreover, over expression of Beclin1 in MCF7 breast cancer cells, which do not endogenously express detectable levels of Beclin1, leads to an induction of autophagy in these cells . Therefore, Beclin1 is essential for the role played by Vps34 in autophagy, but dispensable for the role played by Vps34 in endocytic trafficking and lysosomal sorting. Several other proteins have been found to interact with Beclin1 in mammalian cells, and these include Bcl2, UVRAG, Atg14, Rubicon, Bif1, activating molecule in Beclin1-regulated autophagy (Ambra1), and Rab5.
As in yeast, mammalian Beclin1 is present in two complexes, and both complexes also contain Vps34 and Vps15. It is thought that both complexes also contain a Beclin1-interacting protein, Ambra1 [2, 97]. The two complexes are also characterised by the mutually exclusive presence of Atg14 or UVRAG. The complex containing Atg14 is required for autophagy, whereas the complex containing UVRAG is essential for endocytosis . Immunoprecipitation experiments revealed that both Atg14 and UVRAG bind to Vps34 via its N-terminal C2 domain . Experiments using NIH3T3 cells (mouse fibroblasts) co-expressing either Vps34-GFP with HA-UVRAG, or Vps34-GFP with HA-Atg14, revealed that in nutrient replete conditions almost all Vps34-GFP co-localised with HA-UVRAG, whereas in starvation conditions 30% of the Vps34-GFP dots co-localised with HA-Atg14 .
Bcl2 and BclXL can both bind to Beclin1, and both act to inhibit autophagy. Beclin1 contains a BH3 (Bcl2-homology-3) domain, that allows it to bind to Bcl2 family members such as Bcl2 and BclXL . Bcl2 decreases Beclin1 interaction with Vps34, and this also decreases Beclin1-associated Vps34 kinase activity [98–101]. Upon nutrient starvation the amount of Beclin1 that co-immunoprecipitates with Bcl2/BclXL decreases and autophagy ensues. Only Bcl2 that resides on the ER, and not mitochondrial Bcl2, can inhibit autophagy [99, 101]. This is interesting in light of growing evidence implicating an important role played by the ER in isolation membrane formation and expansion into autophagosomes (see above). It seems that the interactions of Bcl2/BclXL with Beclin1 may represent an on/off mechanism for autophagy induction. The interaction between Bcl2/BclXL and Beclin1 can be modulated by another BH3 domain containing protein, Bad. Bad is activated upon nutrient starvation, and can compete with Beclin1 for binding to Bcl2/BclXL. Increased co-immunoprecipitation of Bad with Bcl2/BclXL can be observed upon nutrient starvation. Moreover, over expression of Bad can induce autophagy in nutrient-rich conditions [100, 102].
C-Jun N-terminal kinase1 (JNK1) has also been found to play a role in modulating Bcl2-Beclin1 interactions . JNK1 becomes activated during nutrient starvation, and JNK1 can phosphorylate Bcl2 at serine and threonine residues. When Bcl2 is phosphorylated by JNK1 it can no longer bind to Beclin1, and autophagy ensues. Autophagy is inhibited by JNK1 inhibition or expression of a Bcl2 mutant that is resistant to phosphorylation by JNK1. Moreover, constitutively active JNK1 can stimulate Bcl2 phosphorylation and autophagy in non-starved cells. However, it has recently been reported that JNK activation does not occur during nutrient starvation and autophagy, although JNK was found to be activated during autophagic cell death . In this study, JNK inhibition had no effect on autophagy-induced GFP-LC3 puncta, but did affect autophagic cell death. The authors suggest that JNK activation actually occurs downstream of autophagy, since inhibition of autophagy by 3-methyladenine treatment and siRNA against Atg5 inhibited the phosphorylation of JNK.
Phosphorylation of the BH3 domain of Beclin1 by the death-inducing kinase, DAP-kinase, can promote dissociation of Beclin1 from Bcl2/BclXL, which leads to induction of autophagy [99, 105, 106]. Recently, a protein named nutrient-deprivation autophagy factor-1 (NAF-1) was found to regulate Bcl2-Beclin1 interaction at the ER [107, 108]. NAF-1 does not have a BH3 domain, yet co-immunoprecipitation experiments reveal that NAF-1 can interact with Bcl2, and that NAF-1 binding to Bcl2 can be displaced by the BH3-only protein Bik. Experiments using shRNA against NAF-1 found that NAF-1 knockdown enhanced autophagy in starved cells, and reduced the interaction between Bcl2 and Beclin1. Hence, NAF-1 acts to stabilise the Bcl2-Beclin1 interaction, and therefore inhibit autophagy.
UVRAG is thought to be an orthologue of yeast Vps38 [74, 109]. Co-immunoprecipitation experiments revealed that Beclin1 and UVRAG interact. Moreover, UVRAG can increase the interaction between Beclin1 and Vps34 as well as increase Vps34 activity . Early studies implied that UVRAG played a role in autophagosome formation in a Beclin1-dependent manner. However, further studies demonstrated that siRNA against UVRAG had no effect on autophagy . The same study found that UVRAG localised to early and late endosomes, and that co-expression of GFP-Vps34 with HA-UVRAG localised GFP-Vps34 to HA-UVRAG positive structures. Importantly, UVRAG did not localise with any markers of autophagy . However, UVRAG has been reported to play a role in endocytic trafficking and autophagosome maturation. It interacts with the class C Vps complex (C-Vps), which is a key component of endosomal fusion machinery . Specifically, UVRAG can recruit C-Vps to the autophagosome and also stimulate Rab7 GTPase activity, and this is required for autophagosome maturation [111, 112]. Several studies have shown that UVRAG co-immunoprecipitates with both Vps34 and Beclin1, implying that these proteins exist in a complex that is similar to complex II in yeast. This would suggest that Beclin1 is present in an endocytosis-specific complex with UVRAG, but this is despite the fact that another report has shown that Beclin1 is not required for mammalian endocytosis . Autophagosome maturation occurs by fusion of the autophagosome with different endosomal populations . Therefore, since UVRAG is thought to play a role in endocytosis, it may regulate the maturation of autophagosomes.
Mammalian Atg14 is present in a complex with Vps34, Vps15 and Beclin1, and like yeast complex I this complex is essential for autophagy. It co-localises with both Atg5 and Atg16L on isolation membranes  and is thought to play a role in early autophagogenesis. Atg14 is essential for autophagy, as starvation-induced GFP-LC3 puncta formation was markedly reduced in Hela cells transfected with siRNA against Atg14. Moreover, electron microscopy revealed that autophagosomes are virtually absent in Atg14 knockdown cells . Additional studies have also identified Atg14 as being essential for autophagy [113, 114]. Furthermore, although Atg14 localises to starvation-induced puncta, these only partially co-localise with LC3, and Atg14 is not present on larger LC3 puncta . Hence, it is possible that Atg14 may not be present on complete autophagosomes. The formation of starvation-induced Atg14 puncta was unaffected by wortmannin treatment, suggesting that Vps34 activity is not required for Atg14 puncta formation. It is known that Vps34 functions upstream of the Atg12-Atg5-Atg16L complex in autophagy, therefore Atg14 must also localise to autophagic membranes before the Atg5-Atg12-Atg16L complex. An Atg14 mutant which is unable to bind to Vps34 and Beclin1, still accumulated to starvation-induced punctate structures, and these co-localised with Atg16L and partially with LC3. Hence, Atg14 can localise to isolation membranes or its precursor structures independently of Vps34 and Beclin1. Further, the inhibition of autophagy observed in Atg14 knockdown Hela cells could be rescued by wild-type Atg14 expression, but not by expression of the Atg14 mutant unable to bind Vps34 and Beclin1 . Therefore, the role played by Atg14 in autophagy requires it to be in a complex with Vps34 and Beclin1. It has also been shown that Atg14 can increase Vps34 kinase activity by 2.5-fold, and this stimulation of Vps34 kinase activity only occurred in the presence of Vps15 and Beclin1 . Hence, ATG14 may act to recruit Vps34 to the isolation membrane, where it can stimulate its activity.
Recently, two independent studies identified Beclin1-binding proteins [113, 114]. These were Vps15, Vps34, UVRAG, Atg14, and a novel protein called Rubicon (RUN domain and cysteine-rich domain containing Beclin-1-interacting protein). Immunoprecipitation experiments revealed that Atg14 is present in a complex with Beclin1, Vps34 and Vps15, but not with UVRAG or Rubicon. In contrast, Rubicon co-precipitated with UVRAG, Beclin1, Vps34, and Vps15, but not with Atg14. Furthermore, gel filtration experiments revealed that only a subpopulation of UVRAG complexes also contain Rubicon. In summary, three types of Beclin1-Vps34-Vps15 complexes were found to exist: an Atg14 containing complex, an UVRAG complex, and a Rubicon-UVRAG complex . Knockdown of Rubicon in both A549 cells (human epithelial carcinoma;  and NIH 3 T3 cells (mouse fibroblasts; ) led to an increase in autophagic activity. Conversely, over expression of Rubicon inhibits autophagy. GFP-tagged Rubicon localised to late endosomes/lysosomes and to starvation-induced Atg16L and LC3 positive dots. GFP-UVRAG also localised to late endosomes/lysosomes. Therefore, the Rubicon–UVRAG–Beclin1–Vps34–Vps15 complex localises to late endosomes/lysosomes. Upon Rubicon knockdown, the lysosomal degradation of the endocytosed EGF receptor was accelerated. In contrast, over expression resulted in inhibition of the degradation of the endocytosed EGF receptor. In Rubicon knockdown cells, more autolysosomes accumulated than autophagosomes. Whereas in cells over expressing Rubicon, the turnover of LC3-II was inhibited, indicating that autophagosome maturation was impaired. These results would suggest that Rubicon plays a role in negatively regulating autophagosome maturation [113, 114]. Interestingly, over expression of Rubicon was found to inhibit Vps34 kinase activity. Although it has been shown that Rubicon does not bind to PI3P, Rubicon-positive structures were enriched with PI3P. These structures were unaffected by wortmannin treatment, suggesting that the maintenance of these structures was not dependent on PI3P .
Bax-interacting factor 1 (Bif-1) is a member of the endophilin protein family. It can interact with Beclin1 indirectly via UVRAG, to regulate the activity of Vps34 and progression of autophagy in mammalian cells [115, 116]. Loss of Bif-1 suppresses Vps34 activation and autophagosome formation in both Hela cells (human epithelial cervical cancer) and mouse embryonic fibroblasts (MEFs; ). Upon nutrient starvation, a proportion of cellular Bif-1 translocates to foci in the cytoplasm, and these foci co-localise with the autophagy markers LC3 and Atg5. The co-localisation of Bif-1 with Atg5 implies that Bif-1 is involved in the early stages of autophagosome biogenesis, as Atg5 is known to be present on the isolation membrane but not on completed autophagosomes. Bif-1, like all endophilins, contains an N-terminal N-BAR (Bin-Amphiphysin-Rvs) domain and also has intrinsic membrane curvature-inducing activity, which could be important in the biogenesis of isolation membranes. Loss of Bif-1 suppresses starvation-induced autophagy as well as starvation-induced activation of Bax . Bif-1 could therefore induce autophagy by promoting Bax binding to Bcl2, thereby releasing Beclin1 from negative regulation by Bcl2.
A further protein that has been shown to interact with Beclin1 is Ambra1. Both yeast two-hybrid experiments and co-immunoprecipitation assays revealed that Beclin1 and Ambra1 interact . siRNA against Ambra1 led to a decrease in autophagy, as assessed by decreased LC3-I to LC3-II conversion. Conversely, over expression of Ambra1 led to an increase in both basal and rapamycin-induced autophagy. Furthermore, mutant forms of Ambra1 that were unable to bind to Beclin1, had no detectable effect on autophagy. Therefore, the effect of Ambra1 on autophagy is dependent on its interaction with Beclin1. It was also observed that Beclin1-mediated induction of autophagy was reduced after Ambra1 downregulation. Moreover, Ambra1 downregulation was found to lead to a reduced interaction between Beclin1 and Vps34, indicating a role played by Ambra1 in facilitating the interaction between Beclin1 and Vps34. Importantly, conversion of LC3-1 to LC3-II was also reduced in Ambra1 mutant embryos, suggesting a bone fide in vivo effect on autophagy by Ambra1 .
Rab5 has also been found to immunoprecipitate with Beclin1, and this only occurs in the presence of Vps34 . Earlier studies also found that Rab5 can interact with Vps15 . Aggregation of mutant huntingtin can be regulated by autophagy . A role for Rab5 in autophagy was hinted at when over expression of dominant-negative Rab5 (DN-Rab5) increased the aggregation of a mutant huntingtin exon 1 fragment with 74 polyQ (Q74) repeats . The same study also confirmed an in vivo role played by Rab5 in huntingtin disease by using a Drosophila melanogaster huntintin disease model, whereby a huntingtin fragment with 120 polyQ repeats is expressed in fly photoreceptors. Degeneration of these huntingtin disease fly photoreceptors was rescued when these flies were crossed with flies transgenic for Rab5-EGFP over expression. Furthermore, over expression of DN-Rab5 in wild-type MEFs caused an increase in Q74 aggregation. Importantly, over expression of DN-Rab5 or constitutively-active (CA-Rab5) had no effect on Q74 aggregation in autophagy-incompetent Atg5−/− MEF’s. The authors then went on to further define the role played by Rab5 in autophagy. They found that inhibition of Rab5 decreased the proportion of COS-7 cells (African green monkey) with over 20 LC3 positive autophagosomes, whereas over expression of wild-type or CA-Rab5 had the opposite effect. Inhibition by 3-methyladenine or RNAi knockdown of Vps34 resulted in the appearance of Atg5 positive structures that co-localised with Beclin1, but not with LC3, indicating that they are early autophagic structures. Rab5 co-localised with these Atg5 positive structures, and inhibition of Rab5 led to the appearance of similar Atg5 positive structures. Hence, Rab5 could play a role in early autophagogenesis, possibly by activating Vps34. The accumulation of Atg5 positive/LC3 negative structures by either Vps34 or Rab5 inhibition could be due to a block in the progression of Atg5 positive membranes to the formation of autophagosomes . Interestingly in Caenorhabditis elegans, Rab5 is necessary for the formation of ER tubules . It may be that Rab5 acts to recruit Vps34 to omegasomes (see above), which are in close contiguity to the ER.
Calcium and Vps34
The paradoxical roles played by Vps34 in both autophagy induction and amino acid signalling to TORC1 (and therefore inhibition of autophagy) both seem to rely on elevated intracellular calcium levels. For example, elevation of cytosolic calcium levels induces autophagy , whereas chelation of intracellular calcium is inhibitory . Work in our laboratory also indicates that calcium is an early requirement for the autophagic response, since chelation of intracellular calcium inhibits the translocation of GFP-DFCP1 to omegasomes upon starvation (Chandra and Ktistakis, in preparation). However, evidence also suggests that elevation of cytosolic calcium levels could play a role in signalling from amino acid stimulation to Vps34 activation and increased TORC1 signalling , although another study suggests that this may not be via a direct effect . Interestingly, cytosolic calcium elevated by amino acid addition appears to come from the extracellular media, whereas the calcium that is required for autophagy comes from internal stores.
Effectors of PI3P during autophagy
A family of PI3P-binding proteins that play a role in autophagy have been identified in yeast , the family is comprised of Atg18, Atg21 and Ygr223c. They are all WD40 repeat containing proteins, thought to be putative scaffold proteins involved in regulating the assembly of multi-protein complexes. These proteins can bind to PI3P and PI(3,5)P2 by their conserved FRRGT motif . All three proteins were found to localise to endosomes, and PI3P synthesised by complex II (not involved in autophagy) was essential for this localisation. Atg18 was found to be essential for autophagy and the Cvt pathway . Atg21 was found to be essential for the Cvt pathway, but not for autophagy . Whereas, Ygr223c was found to play a role in a specific kind of autophagy, whereby portions of the nucleus are sequestered and degraded (micronucleophagy; ). Atg18 is also located at the PAS, and this is dependent on Atg14. Moreover, autophagy was severely reduced in yeast expressing a mutant form of Atg18 that cannot bind to PI3P. When this mutant form of Atg18 was fused to a mammalian FYVE domain, full autophagic activity was restored. It is known that Atg18 forms a complex with Atg2 at the PAS, and the formation of this complex was found to not require PI3P. However, the localisation of this complex at the PAS was found to be dependent on PI3P .
There are two mammalian orthologues of Atg18 known as WIPI1 and WIPI2 . The WIPI family of proteins has four members, all of which contain WD40 repeats. WIPI1 (also known as WIPI49) was found to bind to 3′-phosphorylated phosphoinositides , and to co-localise with LC3 on autophagosomes upon starvation. It has also been discovered that WIPI1 localises to isolation membranes . WIPI2 has recently been characterised , in nutrient-rich conditions endogenous WIPI2 was found to have a diffuse localisation in HEK293 cells, upon starvation WIPI2 translocated to punctate structures that partially co-localised with LC3. WIPI2 was not found to be present on mature autophagosomes or autolysosomes, this suggests that WIPI2 dissociates from autophagosomes before they mature. WIPI2 knockdown by siRNA was found to reduce GFP-LC3 lipidation and the number of GFP-LC3 puncta in HEK293 cells. Furthermore, WIPI2 knockdown was shown to reduce the endogenous LC3-II/LC3-I ratio in multiple cell types. Overexpression of Atg16L disrupts autophagosome formation and leads to the accumulation of isolation membranes, endogenous WIPI2 was found to be present on these accumulating isolation membranes. Overexpression of a proteolytic activity-deficient mutant of Atg4B prevents LC3 lipidation and inhibits autophagosome formation, also leading to the accumulation of isolation membranes. The number of WIPI2 positive structures did not increase upon starvation in mutant Atg4B expressing cells, but the number of WIPI2 positive puncta did increase in control cells. The authors concluded that WIPI2 is present on early autophagic membranes where it acts to positively regulate LC3 lipidation before dissociating.
To investigate the lipid-binding properties of endogenous WIPI2, HEK293 cells were lysed and phopsholipid–protein overlay assays were performed, WIPI2 was found to bind to PI3P and to a lesser extent to PI(3,5)P2. Wortmannin treatment did not affect the amount of WIPI2 in membrane fractions, but did decrease the number of starvation-induced WIPI2 puncta. Therefore, upon autophagy induction WIPI2 translocates from a membrane associated population to puncta in a PI3P-dependent manner .This study also utilised HEK293 cells stably expressing GFP-DFCP1 (see above), when autophagy is activated GFP-DFCP1 translocates to omegasomes, and this is dependent on PI3P .WIPI2 was found to be present on 90% of starvation-induced GFP-DFCP1 puncta . Moreover, the early autophagic markers Atg16 and ULK1 were also found to be present on WIPI2 positive omegasomes. Interestingly, when siRNA was used against WIPI2 in HEK293 GFP-DFCP1 cells, omegasomes accumulated in fed cells, whereas WIPI2 depletion had no effect on the number or shape of omegasomes in starved cells. WIPI2-depleted cells showed an accumulation of DFCP1 positive puncta but these were LC3 negative, whereas cells expressing WIPI2 form DFCP1 positive puncta that are also positive for LC3. The authors hypothesised that WIPI2 is required for the recruitment of LC3 to the forming omegasome.
Alfy, novel a FYVE domain-containing PI3P binding protein has been identified in humans, with putative orthologues in Drosophila, C. elegans, Dictostelium, Arabidopsis, and Schizosaccharomyces pombe . Alfy is known to be ubiquitously expressed, as expressed sequence tags have been identified from a variety of tissues including liver, kidney, testis, uterus, brain, breast, prostate, intestine, lung, pharynx, nervous tissues, placenta, ovary, marrow, head/neck, colon, hypothalamus and stomach. Alfy was found to accumulate on cytoplasmic structures upon amino acid starvation, as assessed by staining HeLa (human cervical cancer) with an anti-Alfy antibody. Vinblastine is a compound which destabilises microtubules and inhibits the fusion of autophagosomes with lysosomes, thereby leading to an accumulation of autophagosomes. Treatment with vinblastine led to an increase in the proportion of HeLa cells with Alfy positive puncta. Moreover, Alfy was found to co-localise with the autophagy markers Atg5 and LC3. Treatment with the PI3-kinase inhibitors wortmannin or 3-methyladenine had no effect on proportion of cells with punctate Alfy staining. This would suggest that Alfy is recruited to punctate structures independent of PI3P production. However, Alfy was found to bind to PI3P containing-liposomes in vitro, and this required its FYVE domain. Additionally, using a 2xFYVE domain probe it was observed that Alfy positive structures co-localised with PI3P in HeLa cells. Therefore, although PI3P production is not required for the formation of starvation-induced Alfy positive structures, these structures may contain PI3P or be located in close proximity to PI3P containing structures. It was also found that inhibiting proteasomal degradation caused an increase in the number of Alfy positive puncta, and that Alfy co-localised with ubiquitin. In addition, electron microscopy could detect similar structures inside autophagosomes. This led the authors to hypothesise that Alfy might recognise protein aggregates targeted for degradation and act as a scaffold for components of the autophagic machinery .
A further effector of PI3P during autophagy is DFCP1 (see above; ). It is highly conserved in mammals as well as in some other organisms including chicken, Xenopus, Zebrafish, honeybee, mosquito and in several species of Drosophila, but not in D. melanogaster or in yeast. As it is absent in yeast and D. melanogaster, we would suggest that its—as yet unknown—function must not be essential.
PI3P signal termination
Most pathways regulated by the phosphoinositides depend not only on the generation of PI3P by PI3-kinases, but also on the consumption of the lipid signal by PI3-phosphatases. Conditions where the phosphoinositide signal persists beyond a physiologically appropriate duration can result in diseases. A PI3P phosphatase called Jumpy has recently been identified in humans . Jumpy is a member of the myotubularins, a family of 16 lipid phosphatases. Jumpy is thought to inhibit autophagy, since knockdown of Jumpy expression caused an increase in basal levels of autophagy in several different cell types. Moreover, Jumpy knockdown stimulated both autophagosome formation and maturation. Specifically, Jumpy knockdown increased both the total number of autophagic organelles and the number of autolysosomes. Hence, Jumpy knockdown stimulates both the formation of autophagosomes and their maturation into autolysosomes. Jumpy was found to co-localise with Atg16, Atg12 and LC3 on isolation membranes, and also with LC3 on autophagosomes . As mentioned above, WIPI1 can bind to PI3P and co-localise with LC3 on autophagosomes . Knockdown of Jumpy was found to increase the number of starvation-induced WIPI1 puncta. Therefore, Jumpy can act to prevent PI3P-dependent WIPI1 recruitment to autophagic membranes. It has also been found that a catalytically inactive Jumpy mutant that has lost the ability to negatively regulate autophagy is found in a congenital disease known as centronuclear myopathy .
Recently, a further myotubularin protein, MTMR3 has been functionally characterised . Both over expression of an inactivated dominant negative mutant of MTMR3 and siRNA against MTMR3 increased the number of GFP-LC3 and GFP-Atg5 puncta under nutrient-rich conditions. This MTMR3 mutant was also found to co-localise with both GFP-LC3 and GFP-Atg5 puncta. Moreover, in MTMR3 dominant negative mutant over expressing cells, wortmannin treatment impaired GFP-Atg5 puncta formation, indicating that the effect of the mutant MTMR3 on GFP-Atg5 puncta was PI3P-dependent. The authors used a GST-2xFYVE probe to observe the cellular distribution of PI3P. In dominant negative MTMR3 expressing cells the number of GST-2xFYVE puncta was increased under nutrient-rich conditions, compared to control cells. In addition, the localisation of the MTMR3 mutant overlapped with both GST-2xFYVE and GFP-LC3, indicating that the amount of PI3P is focally increased in areas of mutant MTMR3 accumulation on autophagosomes. Both over expression of dominant negative MTMR3 and siRNA against MTMR3 also caused an increase in GFP-WIPI1 puncta in nutrient-rich conditions, an effect that was not observed in control cells or cells over expressing wild-type MTMR3. A similar effect was seen when the localisation of GFP-DFCP1 was studied, whereby the number of GFP-DFCP1 puncta was increased in nutrient-rich conditions in cells over expressing dominant negative MTMR3. Since the localisation of DFCP1 is dependent on its PI3P binding properties (see above), it seems that over expression of inactive MTMR3 leads to increased PI3P levels at omegasomes. Furthermore, MTMR3 co-localised with GFP-DFCP1 at omegasomes. Interestingly, electron microscopy revealed that in cells over expressing wild-type MTMR3 the number of autophagosomes or autolysosomes over 500 nm in diameter was lower compared to control cells, although autophagosomes or autolysomes under 500 nm in diameter were frequently observed in cells over expressing wild-type MTMR3. Interestingly, nearly all the structures less than 500 nm in cells over expressing wild-type MTMR3 were positive for LC3, indicating that they were autophagosomes. The authors conclude that the local levels of PI3P (as controlled by MTMR3) affect both autophagy initiation and autophagosome size .
Proposed mechanism for induction and progression of autophagy: a working model
Work in our laboratory is supported by the Biotechnology and Biological Sciences Research Council (BBSRC).