1 Introduction

Normal cell growth and development requires a balance between protein synthesis and protein degradation. Eukaryotic cells have two mechanisms for degrading proteins, namely, the ubiquitin-proteasome system that is responsible for degradation of short-lived proteins, and autophagy that acts to degrade long-lived proteins and organelles [1]. Autophagy occurs in basal conditions to perform homeostatic functions and is upregulated upon nutrient deprivation or cellular stress to provide amino acids and maintain energy levels. Autophagy is therefore essential for the survival of yeast during starvation and in higher eukaryotes it is required shortly after birth when the trans-placental food supply is lost (reviewed by Ref. [2]). In addition, autophagy is found to be involved in lifespan extension, cellular development, and has been implicated in cancer and neurodegenerative diseases such as Huntington's, Alzheimer's and Parkinson's diseases [2].

Autophagy begins by the elongation of a membrane cisternae to enclose cytoplasmic material into a double-membrane vesicle called the autophagosome (Fig. 1 ). The autophagosome fuses with the vacuole in yeast and plants or the lysosome in mammalian cells to enable degradation of the sequestered contents by vacuolar/lysosomal hydrolases. Yeast genetic screens have identified more than 30 autophagy related (Atg) proteins that are required for autophagy [3]. Many of these Atg proteins are found to transiently localise to a distinct site that is juxta-posed to the vacuolar membrane termed the pre-autophagosomal structure or phagophore assembly site (PAS). Correct recruitment of Atg proteins to this site is required for normal autophagy and thus it is a potential site for organisation of the forming autophagosome [4, 5]. Orthologues of many of the Atg proteins have been identified in higher eukaryotes including Caenorhabditis elegans, Drosophila, mice and humans (for review see [6]).

The Atg proteins can be categorised into several distinct complexes that collectively make up the core machinery required for autophagy. This core machinery comprises a protein kinase complex and a phosphatidylinositol 3-kinase (PI3K) complex, two ubiquitin-like conjugation systems, as well as the multi-spanning membrane protein Atg9.

The protein kinase complex includes the serine/threonine protein kinase Atg1 that forms a complex with a number of Atg proteins including Atg13 and Atg17. In yeast the formation of a functional complex of Atg1, Atg13, Atg17, Atg29 and Atg31 is required for autophagy and is negatively regulated by the protein kinase target of rapamycin (TOR). Atg1 orthologues in Drosophila and mammals are also found to regulate autophagy downstream of TOR signaling (reviewed by Ref. [7]).

The PI3K complex consists of the PI3K, Vps34, its regulatory subunit, Vps15, and Vps30/Atg6 and Atg14 [8]. This complex appears to play an essential role in the induction of autophagy. It is one of the first components recruited to the forming autophagosome and is required for localisation of autophagy proteins to the PAS. In mammalian cells, hVps34, the Vps15 homologue, p150, the Atg6 orthologue, Beclin 1, and human Atg14/Barkor are also essential for autophagy (reviewed by Ref. [9]). The function of phosphatidylinositol-3-phosphate (PI3P) in the early stage of autophagy is not clear, although it has been suggested to be required for increasing the size of the sequestering membrane as well as recruiting PI3P-binding proteins [10].

Two ubiquitin-like conjugation pathways are required for vesicle expansion, namely the Atg12–Atg5 and the Atg8–PE conjugation systems. Although these proteins have no clear sequence homology to ubiquitin, crystallisation studies revealed a conserved ubiquitin fold in both Atg8 and Atg12. In the first pathway, the ubiquitin-like protein Atg12 is conjugated to Atg5 by the action of E1- and E2-like enzymes, Atg7 and Atg10, respectively (reviewed by Ref. [11]). The Atg12–Atg5 conjugate can then associate with Atg16 and the complex is then recruited to the elongating membrane. The Atg12–Atg5–Atg16 complex is responsible for recruiting a second ubiquitin-like protein Atg8, to the membrane [12]. Atg8 is conjugated not to another protein, but to an abundant membrane lipid phosphatidylethanolamine (PE). This conjugation is catalysed by the E1-like enzyme Atg7 and E2-like enzyme Atg3. Furthermore, the Atg12–Atg5-conjugate is found to act as an E3-like enzyme in vitro facilitating the conjugation of Atg8 to PE [13]. Atg8 has been suggested to be required for elongation of the autophagosomal membrane; in vitro experiments have revealed that Atg8 can mediate the tethering and hemifusion of liposomes containing Atg8–PE [14]. These conjugation pathways are well conserved in mammalian cells where lipidation of one of the six mammalian homologues of Atg8, LC3, and in particular the GFP-tagged version, GFP–LC3, is widely used to monitor autophagy.

Most of the autophagy machinery is also utilised for the biosynthetic cytoplasm-to-vacuole targeting (Cvt) pathway in yeast [15, 16]. The Cvt pathway involves selective sequestration of the vacuolar hydrolases α-mannosidase (Ams1) and a precursor form of aminopeptidase I (prApe1) into double-membrane Cvt vesicles that exclude bulk cytoplasm following binding to cargo receptor such as Atg19, a cargo receptor for prApe1 and Ams1 [17]. The Cvt pathway has not yet been identified in higher eukaryotes. Examples of other types of selective autophagy are also known in both yeast and mammalian cells, such as mitophagy and pexophagy, which involve selective sequestration of mitochondria and perioxisomes, respectively, and mammalian cells, for instance, protein aggregates and invading bacteria can be selectively incorporated into autophagosomes and targeted for degradation (for review see [18]).

2 Regulation of yeast Atg9 trafficking

Although work in Saccharomyces cerevisiae is beginning to unravel the molecular mechanisms of autophagy, a number of questions remain unanswered including the nature of the PAS and the origin of the autophagosome membrane. The autophagosome membrane is proposed to originate from either a pre-existing organelle or be formed de novo by localised lipid synthesis. Various organelles have been suggested to supply membrane including the ER and the Golgi complex, although characterisation of the autophagosome membrane isolated from mammalian cells was found to lack known protein markers from these suggested ‘donor’ organelle membranes [19]. Atg9 has been the focus of many experiments aimed at answering this question, as it is the only conserved multi-spanning membrane protein that is essential for autophagy and therefore the best candidate to mark the source of the membrane.

Atg9 has six highly conserved transmembrane domains and cytosolic NH₂- and COOH-terminal domains that are divergent in both length and sequence. It was originally identified in yeast genetic screens for defects in both the Cvt and autophagic pathways. Atg9 deficient cells do not mature prApe1, are unable to accumulate autophagic vesicles in the vacuole in the presence of the proteinase B inhibitor PMSF and display a reduced survival rate during nitrogen starvation [20, 21]. Using live-cell imaging, fluorescently tagged Atg9 is observed to localise to the PAS as well as to several dispersed punctate structures that are scattered throughout the yeast cytoplasm [20, 22]. This is in contrast to most other Atg proteins, which are found to reside only at the PAS, and are otherwise diffuse in the cytosol [5]. The peripheral pool of Atg9 partially co-localises with protein markers of mitochondria and it has been speculated that Atg9 resides either in the mitochondria or within vesicles in very close apposition to the mitochondria [23]. Atg9 also co-fractionates with protein markers of the Golgi complex by density gradient analysis suggesting that a population of Atg9 also localises to this compartment [24]. Reggiori et al. [22] have shown that Atg9 cycles between the PAS and peripheral pool, which is facilitated by the interaction of Atg9 with multiple binding partners.

3 Anterograde transport of Atg9

Under nutrient-rich conditions, the PAS localised population of Atg9 depends upon interaction with the peripheral membrane protein Atg11 via its N-terminal domain (Fig. 2 A), and this interaction is facilitated by actin [25]. Perturbation of the actin cytoskeleton with the drug Latrunculin A disrupts Atg11 localisation to the PAS and therefore inhibits Atg9 trafficking to this compartment [26]. Furthermore, Atg9 binds directly to the actin related protein, Arp2, and is found to be immobile in the periphery of the cell in arp2-1 mutants [27]. Localisation of Atg9 to the PAS is required for Cvt vesicle formation; indeed, atg9Δ cells are unable to process prApe1 into its mature form. In the absence of the Cvt complex (Atg19 bound to the cargo proteins prApe1 and Ams1) trafficking of Atg9 to the PAS is defective, which suggests that trafficking of Atg9 is initiated by the presence of the cargo complex. Atg9 cycling was also found to require the peripheral membrane protein Atg23 and the type I membrane protein Atg27. These proteins localise to the PAS and peripheral pools forming a cycling complex with Atg9 [28]. There are so far no known mammalian homologues of Atg11, Atg23 or Atg27.

During autophagy, the localisation of Atg9 to the PAS is independent of Atg11 and actin and instead depends upon physical interaction with Atg17 (Fig. 2B). Atg17 acts as a scaffold for PAS organisation and this process requires the protein kinase Atg1 [29]. Although Atg23 and Atg27 are required for efficient trafficking of Atg9 during autophagy as in the Cvt pathway, they are not essential as autophagosomes still form in atg23 and atg27 mutants although the generated autophagosomes are smaller [30, 31]. In contrast in atg9Δ cells no autophagosomes form resulting in a reduced survival rate and total protein turnover during starvation [20, 21].

Recent data suggests that yeast Atg9 is able to self-interact during nutrient-rich and starvation conditions to form a multimeric Atg9-containing complex that is required for trafficking of Atg9 to the PAS in both the Cvt pathway and autophagy [32]. Atg9 is observed to localise to a cup-like structure that is speculated to represent the growing autophagosome or phagophore, the formation of which is dependent upon the ability of Atg9 to self interact. A loss-of-self interaction Atg9 mutant (Atg9D766–770) results in fragmentation of this structure, apparent loss of membrane at the PAS and the formation of abnormal autophagosomes. This has led to speculation that Atg9 trafficking to the PAS may participate in supplying membrane, as well as mediating tethering and fusion of small membranes to the forming autophagosome [32].

4 Retrograde transport of Atg9

Atg9 is not present on autophagic vesicles that accumulate in the vacuole upon addition of PMSF, and thus it is assumed that Atg9 is absent from the completed autophagosomes, which fuse with the vacuole. The retrieval of Atg9 from the PAS and/or the closed autophagosome to the peripheral pool involves the Atg1–Atg13 complex, the Atg9 binding partners Atg2 and Atg18, which form a complex, and the PI3K complex [22]. This retrieval pathway is required for both the Cvt pathway and autophagy. Indeed, in the absence of any of these proteins Atg9 accumulates at the PAS and autophagy is defective.

The proposed model involves the independent association of Atg9, Atg1 and the PI3K complex with the PAS that leads to the membrane recruitment of the PI3P-binding protein, Atg18, and the peripheral membrane protein, Atg2. The interaction and association of these proteins with Atg9 at the PAS results in the formation of a functional complex that allows Atg9 retrieval to the peripheral pool [22]. It is not clear whether Atg9 cycles to and from the PAS continuously during the formation of an autophagosome or whether only one cycle of Atg9 transport is carried out. This question may be important in determining whether Atg9 is required for the delivery of lipids to the forming autophagosome. Whereas one cycle of Atg9 could deliver a fixed amount of lipid to the forming autophagosome, the same number of Atg9 molecules that are continuously cycling would be able to deliver more lipid. A role for Atg9 in the initiation of autophagosome formation may require only one cycle of Atg9 traffic to the PAS, whereas the delivery of lipids and thus expansion or propagation of autophagosomes is likely to require continuous cycling of Atg9 to the PAS. A caveat of this model is that Atg9 retrieval from the autophagosome would require a loss of lipids from the forming and/or completed autophagosome. The autophagic machinery may be able to compensate for this recruiting more lipids in the case of anterograde traffic to the PAS and fewer lipids in the retrograde pathway back to the peripheral pool. This would also explain why different machinery is required for trafficking of Atg9 to the PAS during either the Cvt pathway or bulk autophagy, as the Cvt vesicles are much smaller in size than autophagosomes. Therefore, understanding the dynamics of Atg9 traffic may provide further insight into the role for Atg9 in autophagosome biogenesis.

Overall, the yeast model and data provide a clear picture of the network of Atg proteins controlling Atg9 localisation. This information together with the data regarding the role of actin has provided a detailed model for trafficking of Atg9 in the Cvt pathway and autophagy. It is worth noting again, however, that the cytoplasmic domains of Atg9 are variable in higher eukaryotes, which suggests that there maybe different mechanisms for the regulation of Atg9 trafficking amongst species.

5 Mammalian Atg9

Atg9 homologues exist in all species so far examined, and although its function is still unknown, it is required for autophagy in all species studied. Deletion of the C. elegans ORF T22H9.2, the orthologue of yeast Atg9, is found to shorten the life span of worms and increases the embryonic lethality of Cup-5 embryos that require autophagy for survival [33]. Furthermore, siRNA-mediated knockdown of Drosophila melanogaster Atg9 in S2 cells results in an increase in vesicular stomatitis virus (VSV) infection due to a defect in autophagy, which would normally limit VSV replication [34]. Finally, knockout of Atg9 in plants results in accelerated leaf senescence again due to a defect in autophagy [35].

In mammalian cells, there are two functional orthologues of Atg9, namely AtgL1 and Atg9L2. Whereas AtgL1 (mAtg9 hereafter) is ubiquitously expressed, Atg9L2 is only found expressed in the placenta and pituitary gland [36]. siRNA-mediated knockdown of mAtg9 was found to inhibit LC3 lipidation and protein degradation in HEK293 cells [36, 37] suggesting that mAtg9 is required for the autophagic pathway in mammalian cells as well. Most recently, mAtg9 knock out mice have been generated, and mAtg9 has been shown to be required for survival immediately after birth, an identical phenotype to Atg5 knock out mice [38].

In nutrient-rich conditions, and similar to yeast where Atg9 is observed to traffic between a single PAS and peripheral structures [22], mAtg9 has a dual localisation being found in a juxta-nuclear region as well as in a dispersed peripheral pool [37]. The juxta-nuclear pool of mAtg9 has been shown to co-localise with protein markers of the trans-Golgi network (TGN), while the peripheral pool partially co-localises with the endosomal protein markers Rab5, Rab7 and Rab9. The difference between the localisation of yeast Atg9 and mAtg9 appears so far to be restricted to the identity of the two compartments or involved in its cycling. During nutrient starvation, loss of mAtg9 from the juxta-nuclear pool coincides with an increase in the peripheral pool and co-localisation with protein markers of autophagosomes including GFP–LC3 [37]. The localisation of mAtg9 with either GFP–LC3 alone or GFP–LC3 and Rab7 together suggests that mAtg9 is found on both immature autophagosomes (GFP–LC3-positive) and degradative autophagosomes (GFP–LC3 and Rab7 positive). Subcellular fractionation and isolation of mAtg9 and LC3-positive autophagosomes from rat hepatocytes supports this observation (A. Young and S.A. Tooze, unpublished data). This is in contrast to yeast, where mAtg9 is not found on completed autophagosomes. However, following re-addition of nutrients and in the presence of cyclohexamide, a protein synthesis inhibitor, mAtg9 can cycle back from the peripheral pool to the juxta-nuclear pool suggesting that a retrieval pathway from forming autophagosomes exists. Further work is required to determine if mAtg9 is found on autolysosomes, and if a small amount of mAtg9 is degraded during autophagy.

Starvation dependent trafficking of mAtg9 requires the Atg1 orthologue, Unc-51 like kinase (ULK1), and mammalian Atg13 [37, 39]. As in yeast, ULK1 and mAtg13 are in a complex with the functional orthologue of Atg17, FIP200 (reviewed by Ref. [40]). However, the requirement for FIP200 in mAtg9 trafficking remains to be tested. Inhibitors of the PI3K complex were also found to inhibit dispersal of mAtg9 during starvation suggesting that trafficking of mAtg9 is also dependent upon this regulatory complex [37]. Although mammalian homologues of Atg18 and Atg2 exist, they are yet to be characterised in terms of mAtg9 trafficking.

Analysis of mAtg9 on isolated autophagosomes by electron microscopy confirmed the presence of mAtg9 on the membranes of autophagosomes, as well as on internal sequestered membranes (A. Young and S.A. Tooze, unpublished data). Recent data using isolated membrane fractions from HEK293 cells show that GFP-mAtg9 has an asymmetric distribution on these membranes in contrast to GFP-tagged LC3, Atg16 and Atg5 that appear to have a more uniform distribution [41]. It cannot be excluded that, due to the hydrophobic nature of mAtg9, these structures are generated by the isolation procedure. However, while the function of the mAtg9 in these concentrated regions on the autophagosome membrane is not known, it has been speculated that mAtg at these regions may represent docking sites for fusion of incoming vesicles with forming autophagosomes. Furthermore, preliminary experiments using exogenous proteins suggested a co-localisation of mAtg9 and Bif-1 [42], a member of the endophilin family, which can bind to membranes and induce curvature via a BAR domain, has led to speculation that these two proteins may interact. Fusion of Bif-1 positive vesicles with mAtg9 positive compartments leads to the formation of a cup-like structure that is similar to that observed in yeast [42], possibly supporting the idea that mAtg9 may deliver small vesicles to forming autophagosomes and recruit machinery to remodel the membrane.

The picture emerging from these data suggests the dynamic localisation of mAtg9 is important for the formation of autophagosomes, and as such may be subject to regulatory controls. While from the yeast data we understand some of the Atg molecules involved in the control of its cycling, in mammalian cells with the exception of Bif-1 there have been no other reports of proteins which interact with mAtg9. Recently, using the C-terminal cytoplasmic domain of mAtg9 we have found that p38IP (p38 interacting protein) binds to mAtg9 and is required for its trafficking during starvation [43] (Fig. 3 ).

6 Regulation of mAtg9 trafficking by p38IP

p38IP, which is also known as family with sequence similarity 48 (FAM48), contains a nuclear localisation sequence, a PEST domain and two serine rich domains [44]. HA-tagged p38IP localises both to the nucleus and a cytoplasmic pool that partially co-localises with mAtg9. This is in agreement with the finding that endogenous p38IP localises to the nucleus and can be seen in a cytoplasmic pool in HeLa cells [45]. Starvation-dependent mAtg9 trafficking was found to require p38IP as siRNA-mediated depletion of p38IP resulted in the loss of mAtg9 dispersal that is normally observed following starvation and a retention of mAtg9 at the juxta-nuclear pool [43]. The small population of p38IP and mAtg9 that co-localise, in both nutrient-rich and starvation conditions, suggests the interaction of mAtg9 with p38IP may be dynamic and may be better visualised by live-cell imaging. p38IP was identified in a screen for neural tube closure and gastrulation defects. Importantly, p38IP was found to bind to p38α MAPK and is required for its activation during mouse development [44]. Expression of mutant p38IP, which contains a premature stop codon resulting in a truncated protein at 317 amino acids, in the mouse eye results in a reduction in the levels of phosphorylated p38α and its downstream substrates CREB and ATF2. This p38IP mutant was found to have an increased incidence of spina bifida and exencephaly, whereas a splicing mutant that results in truncation at 133 amino acids had severe gastrulation defects. It is not known if these phenotypes are due to defects in the activation of p38α and the proposed downregulation of E-cadherin. However, inhibition of p38 activity by SB 203580 compound, an inhibitor of MAPKinases did not significantly alter the incidence of spina bifida or gastrulation observed [46], which suggests that these defects may be due to another function of p38IP that is independent of the activity of p38 MAPK. Similar neural tube closure defects are observed upon functional deficiency of Ambra1, a Beclin1 binding partner that is involved in autophagy and neuronal development [47]. These common phenotypes suggest a role for autophagy in the p38IP mouse mutant phenotype.

As p38IP was identified as a p38 MAPK interacting partner we decided to investigate the role of p38 in autophagy and regulation of mAtg9 trafficking. Depletion of p38 MAPK results in dispersal of mAtg9 in full medium and leads to an upregulation in LC3 lipidation suggesting that p38 is a negative regulator of autophagy. Furthermore, the function of p38 in autophagy was found to be dependent upon p38IP, and this data supports the hypothesis that p38 regulates the interaction between Atg9 and p38IP [43]. p38 has previously been implicated in autophagy regulation, although the ability of p38 to negatively or positively regulate autophagy appears to be dependent upon the cell type and stimulus used. For example, whereas we have shown that activation of p38 with either anisomycin or UV irradiation leads to an inhibition in the autophagic pathway [43], accumulation of mutant glial fibrillary acidic protein (GFAP) in astrocytes is found to induce autophagy via activation of p38 and inhibition of the TOR pathway [48]. It would be interesting to determine the relationship between Atg9 trafficking and p38 in these conditions.

Interestingly, new data on the regulation of the TOR pathway by p38 MAPK supports our hypothesis concerning p38 regulation of autophagy. In A549 cells, p38 activation by anisomycin has been shown to activate mTOR and S6K [49]. It was under these conditions that we observed an inhibition of Atg9 trafficking and autophagy. Although we did not analyse TOR phosphorylation directly we were also able to detect an increase in S6K phosphorylation, detected by immunoblotting with phospho-specific S6K antibodies, following activation of p38 with anisomycin (J. Webber, S.A. Tooze, unpublished results). p38α-dependent activation of TORC1 is blocked following siRNA knockdown of the upstream p38 kinases MKK3 and MKK6 [49]. Interestingly, in contrast, we find that whereas p38α depletion increases GFP-LC3 and LC3 lipidation [43], siRNA-mediated knockdown of MKK6 inhibits GFP-LC3 lipidation (Ref. [50] and J. Webber, S.A. Tooze, unpublished results). This suggests that in our cell model, either MKK3 can over-compensate for loss of MKK6, or alternatively there are potentially two separate mechanisms for regulation of autophagy by p38, one via TOR, and the second by inhibition of mAtg9 trafficking and LC3 lipidation.

However, as mentioned above, in certain circumstances p38α can positively regulate autophagy in an mTOR-dependent manner. Upon accumulation of mutant glial fibrillary acidic protein (GFAP) aggregates in Alexander disease, a genetic disorder affecting the central nervous system caused by mutations in GFAP, p38 MAPK is activated, and mTOR is inhibited [48]. In these conditions autophagy is upregulated in order to reduce GFAP levels. This is opposite to that observed following activation of p38 with anisomycin or UV irradiation, which may suggest that the role of p38 in autophagy is subject to cell type and stimuli specificity. It would be important in this instance to examine the role of p38 in other protein aggregate diseases such as Huntington's and Alzheimer's disease.

Recently p38IP was identified as a component of the human SAGA (Spt-Ada-Gcn5-Acetylase) complex, which together with RNA polymerase II is required for transcriptional regulation of stress-responsive genes [45, 51]. The SAGA complex is found in yeast, and is comprised of multiple subunits including Spt20. p38IP is suggested to be a homologue of yeast Spt20, sharing five conserved regions within the N-terminal domain. Conservation is also observed in these regions with putative Spt20 homologues across species including Drosophila and mice. Yeast Spt20 interacts with the yeast p38 homologue Hog1, a protein required for osmotic stress in yeast, although it is not known which region of the protein is required for this interaction [52]. It is not known whether yeast Spt20 is also involved in autophagy, however, Hog1 has been shown to regulate autophagy during high osmolarity in S. cerevisiae: loss of Hog1 results in an inhibition of autophagy during incubation in hypo- or hyper-osmotic conditions [53].

In contrast, binding of p38IP to p38 MAPK has been mapped to the C-terminal domain of p38IP [44], which is not conserved in Spt20. Spt20 is also found to have a role in osmotic stress, indeed during conditions of high osmolarity, deletion of Spt20 resulted in a strong inhibition in cell growth and a reduction in gene expression of osmotic stress-responsive genes to a similar extent to that observed upon deletion of the MAPK kinase PBS2, the mammalian homologue of which is MKK4 [52]. This is in contrast to the finding that p38IP is not recruited to the promoter for stress-induced genes that are associated with activation of the p38 MAPK pathway following stimulation with Na-arsenite suggesting that the involvement of p38IP in the MAPK pathway may be specific to the stimulus used to activate p38 [45].

7 Function of mammalian Atg9

Although a number of Atg9 binding partners have been identified for yeast Atg9, and more recently we have shown that mAtg9 can interact with p38IP, we still do not know the function of Atg9 in autophagy. An important study in which quantitative fluorescence microscopy was used to follow the kinetics of Atg protein recruitment to the PAS in yeast [54] has led to the suggestion that Atg9 may be involved in the regulation of autophagosome size. The intensity of Atg9 increases at the PAS where it acts in the initial recruitment of other Atg proteins. It then appears to remain at relatively constant levels during starvation-induced autophagy. As stated above, Atg9 has also been speculated to be a carrier of lipids to forming autophagosomes. In atg27Δ cells (Atg27 is required for correct Atg9 trafficking), Atg9 trafficking is inhibited upon starvation and fewer autophagosomes of normal size are formed [54]. The atg27 phenotype may therefore be due to the loss of delivery of Atg9 and its associated lipids to forming autophagosomes [28, 55]. Likewise in rat primary hepatocytes depleted of mAtg9 under conditions where there was no detectable LC3-II, using electron microscopy smaller autophagosomes were found compared to control siRNA treated cells (A. Young, S.A. Tooze, unpublished results). The difference between this result and the findings in yeast that no autophagosomes form in atg9D cells may be due to the efficiency of siRNA-mediated knockdown of mAtg9. However, it is possible that the autophagosomes found in the Atg9 depleted hepatocytes by electron microscopy are formed using the alternative macroautophagy pathway (see below).

Alternatively the relationship between Atg8 and Atg9 may be important for regulation of autophagosome size. Deletion of Atg8 results in smaller autophagosomes than those seen in wild-type cells [56]. As Atg9 has been shown to act upstream of Atg8 lipidation in yeast, loss of Atg9 may lead to a reduction in autophagosome size due to a loss of Atg8 recruitment to the PAS. As Atg8 can facilitate homotypic fusion in vitro, the amount of lipidated Atg8 may therefore regulate the expansion of the autophagosomal membrane. In rat hepatocytes, as mentioned above, depletion of mAtg9 is also associated with an inhibition in LC3 lipidation [37] which may account for the smaller autophagosomes observed. In cells depleted of p38IP, mAtg9 remains at the trans-Golgi network after nutrient starvation and LC3 lipidation and p62 degradation are inhibited [43] suggesting that p38IP is required for efficient autophagy in mammalian cells. It is possible that regulation of size of the autophagosome via interaction of mAtg9 with p38IP would enable control of the flux through the autophagic pathway and a fine tuning of autophagy.

Our recent data demonstrates that activation of p38 MAPK can inhibit mAtg9 trafficking during starvation resulting in a block in the autophagic pathway [43] supporting our hypothesis that Atg9 may also act as a control point that can be targeted by upstream signaling events. Furthermore, we find that p38α can negatively regulate the interaction between mAtg9 and p38IP and thus we speculate that it is the disruption of mAtg9 trafficking that results in the inhibition of autophagy. This is the first evidence that suggests mAtg9 may have a role in regulation of autophagy. As autophagic activity decreases during prolonged starvation [57] control of degradation would prevent cells from undergoing cell death. The magnitude of the autophagic response must therefore be subject to regulation, which we propose could be through mAtg9 and controlled by p38α.

We have shown that knockdown of p38α in HEK293 cells leads to an upregulation in LC3 lipidation and Atg9 trafficking in full medium suggesting that p38 is a negative regulator of basal autophagy. To function as a negative regulator p38 should sense the nutrient status of the cell. Casas-Terradellas et al. [58] have shown that p38 is activated by amino acids, and this may provide a mechanism by which the phosphorylation status of p38 can be controlled by the availability of nutrients. Following phosphorylation of p38 with anisomycin, a loss of the membrane-associated pool of p38IP was observed along with a corresponding decrease in the interaction between mAtg9 and p38IP. Replenishment of the intracellular amino acids pool upon autophagic degradation of cellular components may feedback to inhibit Atg9 trafficking by favouring the p38α–p38IP interaction and disrupting the Atg9–p38IP interaction and thereby inhibiting autophagy. Thus re-activation of p38α may provide an ideal feedback mechanism to prevent autophagic cell death. Indeed, inhibition of p38 by SB compound in colorectal cancer cells leads to an upregulation in the LC3 homologue, GABARAP, and eventual autophagic cell death [59].

Although Atg9 is an essential protein for starvation-induced autophagy in yeast and mammalian cells, an alternative form of autophagy has been recently proposed in mammalian cells that is independent of Atg5, Atg7 and mAtg9. This alternative non-canonical autophagy is induced in Atg5−/− and Atg7−/− MEFs (mouse embryonic fibroblasts) following treatment with etoposide, which induces cell stress via inhibition of topoisomerase II [60]. In Atg5−/− and Atg7−/− MEFs autophagosomes were observed by electron microscopy to form following etoposide treatment. The generation of these autophagosomes requires Rab9 and it is speculated that they form following fusion of phagophores with vesicles derived from the TGN and late endosomes. This alternative autophagy was shown to be independent of mAtg9 which is puzzling as mAtg9 is the only autophagy related protein so far identified to reside at both the TGN and late endosomes, and it co-localises with Rab9 [37]. Furthermore, mAtg9 trafficking is dependent upon ULK1, which is also required for etoposide-induced autophagy. It is therefore a conundrum that this process does not require mAtg9 and it remains to be explained how and where the autophagosomes originate. It will be interesting to determine if p38 MAPK is able to regulate this process as we speculate that its mechanism of action is via regulation of mAtg9 trafficking we would predict that this mAtg9-independent autophagy would also be independent of p38α.

Most recently, a novel function for mAtg9 has been described in the regulation of innate immunity following double-stranded DNA (dsDNA) stimulation [38]. A multi-spanning membrane protein, stimulator of IFN genes (STING), is a regulator of IFN production assembling into a complex with TANK-binding kinase 1 (TBK1) in the cytoplasm following dsDNA stimulation. The formation of STING-TBK1 structures is negatively regulated by mAtg9, as shown by an enhanced co-localisation of STING and TBK1 in mAtg9-knockout MEFS. It will be interesting to test if this function is dependent upon p38IP and p38 MAPK. p38 MAPK has previously been shown to be involved in IFN production although to the best of our knowledge there is no evidence for regulation of p38 MAPK by dsDNA [61]. This is the first report that mAtg9 has a function other than in autophagy. It is not known if this process is related to autophagy although other Atg proteins do not regulate STING localisation. Furthermore, Atg7 and Atg16 are not required for IFN production stimulated by dsDNA suggesting that this response is independent of autophagy. Interestingly, the Atg12–Atg5 conjugate acts as a suppressor of innate immune responses and can associate with RIG-1 and IPS-1 which are required for recognition of RNA virus infection and function upstream of TBK1 and enables upregulation of IFN. RIG-1 is also required for dsDNA recognition leading to TBK dependent IFN production [62]. Although the Atg12–Atg5 conjugate have not been shown to suppress the immune response following dsDNA stimulation it is tempting to speculate that this may also be regulated by the Atg12–Atg5 conjugate.

8 Concluding remarks

Atg9 trafficking has been well characterised in yeast and although no direct evidence for a role of Atg9 trafficking in regulation of autophagosome size exists a model is emerging. Whether this function is direct or via an interaction with Atg8 lipidation remains to be explored. It would be interesting to determine the size of autophagosomes that are observed to form independently of Atg7, Atg5 and Atg9 in the same system. In contrast, mAtg9 has not been extensively studied although our recent data implicates it as an important protein for regulation of autophagy through action of the p38 MAPK. It remains to be seen whether this is true in all cases of autophagy or if it is specific to starvation-induced autophagy. Furthermore, p38 MAPK appears to be able to regulate the TOR pathway and may be able to provide a negative feedback mechanism to limit the extent of autophagy that occurs. As p38 is also known to be an important protein in the control of apoptosis (reviewed by Ref. [63]), this may provide the link between the two cell death pathways. Understanding further novel interacting partners of mAtg9 will enable us to gain insight into the function of mAtg9 in autophagy and in other pathways.