Atg18 function in autophagy is regulated by specific sites within its β-propeller

Eukaryotes utilize two catabolic pathways to dispose unwanted cellular components: The ubiquitin-proteasome pathway and autophagy. The proteasome is exclusively involved in protein degradation while autophagy permits the elimination of large protein complexes and entire organelles or microorganisms, allowing the turnover of all cellular components (Nakatogawa et al., 2009; Ravid and Hochstrasser, 2008). Autophagy is characterized by the formation of double-membrane vesicles called autophagosomes, which sequester and deliver cytoplasmic structures into the mammalian lysosomes or the yeast and plant vacuoles (Klionsky, 2007). The resulting degradation products are transported back in the cytoplasm and used for either the synthesis of new macromolecules or as an energy source. Induction of autophagy often occurs during stress conditions such as starvation but this pathway also plays a key role in numerous physiological and pathological situations including development and tissue remodelling, ageing, immunity, neurodegeneration and cancer (Mizushima et al., 2008).

Sixteen autophagy-related (Atg) proteins compose the conserved core machinery essential for double-membrane vesicle formation. In yeast, these Atg proteins are recruited to a single perivacuolar site, called the phagophore assembly site or pre-autophagosomal structure (PAS) (Suzuki et al., 2007), which appears to be present in mammals as well (Itakura and Mizushima, 2010). According to the current model, Atg proteins first mediate the biogenesis of a small cup-shaped cisterna known as the phagophore or isolation membrane, and then its expansion into an autophagosome through the acquisition of additional lipid bilayers (Nakatogawa et al., 2009). An important event during autophagosome biogenesis is the generation of phosphatidylinositol-3-phosphate (PtdIns3P) at the PAS by the autophagy-specific phosphatidylinositol-3 kinase complex I (Kihara et al., 2001). Although it has been shown that PtdIns3P is a key regulator of autophagy, the precise function of this lipid is poorly understood (Kihara et al., 2001). One hypothesis is that PtdIns3P is necessary for the recruitment of a subset of the Atg proteins.

One of these proteins is yeast Atg18, which is part of the core machinery and is essential for autophagy (Barth et al., 2001; Guan et al., 2001). The main structural feature of Atg18 is that its seven WD40 repeats, which are stretches of ∼40 amino acids ending with the residues tryptophan and aspartate, fold into a seven-bladed β-propeller (Barth et al., 2001; Dove et al., 2004). Its predicted structure is very similar to the recently published crystal structure of Kluyveromyces lactis Hsv2, a homolog of Atg18 (Baskaran et al., 2012; Krick et al., 2012; Watanabe et al., 2012). WD40 domain-containing proteins often act as scaffolds, which promote and/or coordinate the assembly of protein complexes by creating a stable platform for simultaneous and reversible protein-protein interactions (Chen et al., 2004; Paoli, 2001; Smith et al., 1999). Atg18 is also able to bind both PtdIns3P and phosphatidylinositol-3,5-biphosphate [PtdIns(3,5)P₂] through a conserved phenylalanine-arginine-arginine-glycine (FRRG) motif within its β-propeller (Dove et al., 2004; Krick et al., 2006). Interaction of Atg18 with these phosphoinositides is essential for its localization to the PAS, endosomes and vacuole (Krick et al., 2008; Krick et al., 2006; Nair et al., 2010; Obara et al., 2008b; Strømhaug et al., 2004). While nothing is known about the role of Atg18 at the endosomes, this protein is part of a large complex at the vacuolar membrane that regulates PtdIns(3,5)P₂ levels (Efe et al., 2007; Jin et al., 2008; Michell and Dove, 2009). The localization of Atg18 to the PAS also depends upon Atg2 and vice versa (Guan et al., 2001; Obara et al., 2008b; Suzuki et al., 2007), and it has been proposed that these two proteins constitutively form a complex (Obara et al., 2008b). The Atg18 ability to interact with Atg2 does not depend on its PtdIns3P-binding capacity, whereas the binding of Atg18 to PtdIns3P seems necessary for the appropriate targeting of the Atg18-Atg2 complex to the PAS (Obara et al., 2008b). The presence of Atg18 on three different localizations probably requires a tight control of its recruitment and function. The molecular principles of this regulation are unknown.

In order to understand Atg18 regulation in autophagy and gain insights into the principles controlling the different cellular functions of this protein, we have studied how Atg18 is recruited to the PAS. We have identified the Atg2-binding site of Atg18 and discovered that this sequence is located in a stretch of amino acids connecting β-sheets between WD repeats 2 and 3 of the β-propeller. We have also found that PtdIns3P and Atg2 are the two determinants that mediate the specific recruitment of Atg18 to the PAS. In absence of one of these interactions, Atg18 remains cytosolic and autophagosome biogenesis is blocked at an early stage.

Atg18 is a 500 amino acids protein that contains seven WD40 repeats, which are predicted to fold into a seven-bladed β-propeller (Dove et al., 2004) (Fig. 1A). Previous studies have indicated that Atg18 requires Atg2 for its recruitment to the PAS, and that these two proteins are able to form a complex of ∼500 kDa (Obara et al., 2008b). To study how the function of Atg18 is regulated at the PAS, we decided to identify the Atg2-binding region in Atg18. We thus first tested the ability of Atg18 to interact with Atg2 using the yeast two-hybrid (Y2H) assay. As shown in Fig. 1A, no growth was observed in cells harboring an empty vector or exclusively expressing Atg2. In contrast, cells carrying Atg2 and Atg18 were able to grow, confirming that Atg18 interacts with Atg2 (Fig. 1A).

Crystallographic studies of other WD40 domain-containing proteins have previously shown that the amino acids from the loops that interconnect the blades of the β-propeller are often at the interacting face between the protein and its binding partners (Paoli, 2001). To determine whether the loops within the Atg18 β-propeller mediate the interaction to Atg2, we decided to create point mutant versions of Atg18 modifying the six loops and tested them by Y2H. Since protein-protein interactions often occur through charged or polar amino acids, all these types of amino acids present in the six loops were replaced by alanines (Fig. 1B). When the six loops were individually mutated, the association between Atg18 and Atg2 was not affected (Fig. 1C). Importantly, the mutation of the two arginines in loop 5 that play a critical role in the binding of Atg18 to specific phosphoinositides (Dove et al., 2004) also did not interfere with the Atg18-Atg2 interaction (Fig. 1C). During the realization of our study, three different crystallographic studies have revealed the structure of K. lactis Hsv2 (Baskaran et al., 2012; Krick et al., 2012; Watanabe et al., 2012). These works have uncovered the presence of a large loop connecting the last two β-strands of blade 6, which was lacking in our initial structural model of Atg18. As a result, our predicted location of loop 6 was incorrect and thus the mutated residues are located in the sixth blade.

There are numerous documented cases, where WD40 domain-containing proteins associate with their binding partners through amino acid residues present in different parts of the β-propeller (Chen et al., 2004; Cheng et al., 2004; Paoli, 2001; Pashkova et al., 2010). Accordingly, we constructed a number of Atg18 mutants that combine several loop mutants and tested their ability to bind Atg2. We found that the binding of Atg18 to Atg2 was perturbed when loops 1 and 2 were simultaneously mutated (Fig. 1D), indicating that that the charged amino acids in these sequences could mediate the interaction between the two proteins.

To verify the results obtained with the Y2H assay in the appropriate physiological context, we performed a Protein A (PA) affinity isolation. We first generated a plasmid expressing a functional 13×myc-tagged Atg18 fusion protein under the control of the endogenous promoter (supplementary material Fig. S1), before inserting the different loop mutations. To test the binding capacity of the resulting chimera, the plasmids were transformed into either the atg18Δ strain, in which Atg2 was endogenously tagged with PA or just the atg18Δ knockout, which served as negative control. Analysis of the cell extracts confirmed that all the Atg18 point mutants have similar expression levels as the wild type protein, showing that the mutant proteins are not degraded due to a potential misfolding (Fig. 2A). Mutations in loop 1 led to the appearance of an additional lower molecular mass Atg18 band and the cause of this phenotype is currently under investigation. In accordance with the Y2H results and previous work (Obara et al., 2008b), we were able to co-isolate wild type Atg18 and Atg2-PA confirming the interaction between these two proteins (Fig. 2A). No wild type Atg18-13×myc was detected in the affinity eluate, when the pull-down was performed using the negative control (Fig. 2A, lane 7). Atg18(L1) and Atg18(L5) were also co-isolated with Atg2-PA in comparable amounts to wild type Atg18. Crucially, almost no Atg18(L2) and Atg18(L1,2) were detected after pull-down with Atg2-PA, showing that these two mutant proteins are unable to interact with Atg2. These data show the key role played by the amino acids in loop 2 of the Atg18 β-propeller in binding Atg2.

To show that the mutated amino acids in loops 1 and 2 do not affect the folding of the Atg18 β-propeller, we determined the lipid-binding capacity of each Atg18 loop mutant by conducting a phospholipid-binding assay. Native cell extracts from strains overexpressing the myc-tagged ATG18 loop mutants were incubated on phospholipid strips and proteins detected by western blotting. We found that the Atg18(L1), Atg18(L2) and Atg18(L1,2) mutants were able to bind PtdIns(3,5)P₂ to the same extent as wild type Atg18 (Fig. 2B). Consistently with previous data (Dove et al., 2004; Krick et al., 2006), the Atg18(L5) mutant carrying the mutated FRRG motif did not bind phosphoinositides (Fig. 2B). Dove and co-workers have reported that recombinant Atg18 predominately binds PtdIns(3,5)P₂ but also PtdIns3P with a much lower affinity (Dove et al., 2004). In agreement with this observation, we also found that in our experimental set-up, the binding of the Atg18 variants to PtdIns(3,5)P₂ is favored while what associated to PtdIns3P is below detection levels. These data show that mutations in the Atg2-binding domain do not affect the phosphoinositide binding capacity of Atg18, and consequently the β-propeller is correctly folded.

To investigate the relevance of the interaction between Atg18 and Atg2 in autophagy, we generated plasmids expressing the untagged Atg18(L1), Atg18(L2), Atg18(L1,2) and Atg18(L5) mutants under the control of the endogenous promoter. These constructs were co-transformed with a plasmid carrying the GFP-Atg8 fusion protein into atg18Δ mutant cells to perform the GFP-Atg8 processing assay. This is a well-established method to monitor bulk autophagy in yeast by measuring the accumulation of free GFP in the vacuole over time (Cheong and Klionsky, 2008). When this assay was performed in wild type cells, a band of 25 kDa corresponding to free GFP appeared under starvation conditions indicating normal autophagy (Fig. 3A). In contrast, no GFP-Atg8 cleavage was observed in the atg18Δ mutant due to the complete block of autophagy (Barth et al., 2001; Guan et al., 2001). As shown in Fig. 3A, wild type Atg18 and Atg18(L1) were able to complement this defect, indicating that loop 1 is not required for bulk autophagy. In contrast, a complete block of the pathway was observed when this mutation was combined with that in loop 2. The relevance of loop 2 in autophagy was also underscored by the analysis of Atg18(L2), which showed a severe defect in the progression of autophagy (Fig. 3A). Comparable results were obtained when we examined the effects of the loop mutations on a selective type of autophagy, the biosynthetic cytosol-to-vacuole targeting (Cvt) pathway (Lynch-Day and Klionsky, 2010) (supplementary material Fig. S2). Based on these results and the one obtained with the pull-down experiment, we decided to focus on the Atg18(L2) mutant rather than on Atg18(L1,2) to study the role of the binding between Atg18 and Atg2 in autophagy because it appears to be specific for this interaction.

Next, we explored whether Atg18 loop 2 was also important for the function of this protein at the vacuole by analyzing the morphology of this organelle upon osmotic stress. As reported atg18Δ cells transformed with an empty plasmid displayed enlarged single-lobed vacuoles compared to cells transformed with a plasmid expressing Atg18 used as the wild type control (Fig. 3B) (Dove et al., 2004; Krick et al., 2008). As expected cells expressing the phosphoinositides-binding mutant Atg18(L5) also displayed enlarged vacuoles. In contrast, the morphology of the vacuoles in the strain carrying the Atg18(L2) construct was almost identical to that of the wild type. Based on this result, we concluded that the mutations in the loop 2 of Atg18 do not impair the vacuolar function of this protein and thus this loop very likely only mediates interaction with Atg2. At present, we cannot exclude, however, that the Atg18 function on endosomes is not affected.

To unveil the step of autophagosome biogenesis in which the interaction between Atg18 and Atg2 is required, we scrutinized the formation of the PAS by fluorescence microscopy using CFP-tagged Atg8 as a protein marker for this structure (Suzuki et al., 2001). The atg18Δ strains carrying both genomically integrated CFP-Atg8 and 13×myc-tagged wild type Atg18, Atg18(L2) or Atg18(L5) were assessed in both growing and starvation conditions to determine whether the PAS is formed. In accordance with the literature (Suzuki et al., 2007), the recruitment of CFP-Atg8 to this structure seen as a perivacuolar puncta was not affected by the deletion of ATG18 in the presence or absence of nutrients (Fig. 4A). CFP-Atg8 also localized to the PAS in cells expressing wild type Atg18 or the loop mutants in all growing conditions (Fig. 4A). In nutrient-rich conditions no significant differences were detected between the different strains upon quantification of the number of cells displaying a CFP-Atg8-positive punctate structure (Fig. 4B). Under autophagy conditions, in contrast, expression of the loop mutants led to a clear increase in the number of CFP-Atg8-positive cells compared to wild type Atg18 (Fig. 4B). A similar result was also obtained in the atg18Δ knockout. Analysis of cells expressing the Atg18(L2) or the Atg18(L5) mutant at an ultrastructural level by electron microscopy furthermore revealed that no autophagosomal structures/autophagosomes are being formed in these cells (supplementary material Fig. S3). This result indicates that although the interaction between Atg18 and Atg2 is not required for the PAS formation, it is essential at an early stage of autophagosome biogenesis.

Next, we asked whether the interaction between Atg2 and Atg18 is required for Atg2 recruitment to the PAS as proposed (Obara et al., 2008b; Suzuki et al., 2007). We examined the subcellular distribution of Atg2-GFP in presence of wild type Atg18 or the different loop mutants. As shown in Fig. 5 and as expected (Shintani et al., 2001; Suzuki et al., 2007), Atg2-GFP localized to a single perivacuolar puncta representing the PAS in cells expressing wild type Atg18 in presence and absence of nitrogen. It has been shown that this distribution depends on Atg18 (Obara et al., 2008b; Suzuki et al., 2007). Unexpectedly, we detected Atg2-GFP at the PAS in the atg18Δ knockout but also in the atg18Δ strain expressing Atg18(L2) (Fig. 5), indicating that Atg2 recruitment to this site can also occur independently from its interaction with Atg18. While a minor decrease in the number of Atg2-GFP positive structures was observed in cells expressing Atg18(L5), a substantial amount of protein was still found at the PAS. Quantification of the Atg2-GFP fluorescence signal intensity revealed no significant differences in the amount of this chimera localizing to the PAS between the atg18Δ knockout or the atg18Δ strain expressing wild type Atg18, Atg18(L2) or Atg18(L5) (not shown) further supporting the notion that Atg2 association with the PAS does not require Atg18.

Atg18 and Atg21 share a high degree of sequence homology and they are partially redundant in nutrient-rich conditions (Nair et al., 2010). Although Atg21 has a function in Atg8 association with the PAS rather than recruiting Atg2 to the same location (Meiling-Wesse et al., 2004; Strømhaug et al., 2004), we examined whether the Atg2-GFP localization observed in our atg18Δ background was due to Atg21 presence. As shown in Fig. 5, Atg2-GFP localization was not altered upon deletion of ATG21 in the atg18Δ knockout strain in nutrient-rich and starvation conditions, indicating that Atg2 recruitment to the PAS can also occur in absence of Atg18 and Atg21.

We subsequently explored whether Atg18 binding to Atg2 is required for its association to the PAS. As Atg18 localizes to the PAS, endosomes and the vacuole, we labeled the PAS with the mCherry-V5 (mCheV5)-Atg8 chimera to specifically investigate the subpopulation of this protein at this structure. The atg18Δ strains carrying both genomically integrated mCheV5-Atg8 and GFP-tagged versions of Atg18 or the different loop mutants were imaged before and after nitrogen starvation (Fig. 6; supplementary material Fig. S4). Consistent with previous results (Guan et al., 2001; Krick et al., 2008; Obara et al., 2008b), Atg18-GFP was found at the vacuolar membrane and punctuate structures, one of them colocalizing with mCheV5-Atg8 (Fig. 6A). Wild type Atg18-GFP and mCheV5-Atg8 were colocalizing in 50–60% of the cells in both growing and starvation conditions (Fig. 6B). This colocalization was almost completely abolished in cells expressing Atg18(L2)-GFP or Atg18 (L5)-GFP to the same extent as in the atg2Δ mutant (Fig. 6B). These data show that the interaction between Atg18 and Atg2 is required for the recruitment of Atg18 to the PAS. They also confirm that Atg18 binding to the PtdIns3P is crucial for Atg18 association with this structure (Krick et al., 2006; Nair et al., 2010; Obara et al., 2008b).

To determine whether Atg18 binding to Atg2 and phosphoinositides is required for the specific formation of the Atg2-Atg18 complex on the PAS membranes, we turn to the bimolecular fluorescence complementation (BiFC) approach (Sung and Huh, 2007). This assay allows studying protein-protein interactions in vivo, and is based on the formation of a fluorescent complex by the C- and N-terminal fragments of Venus, a variant of the yellow fluorescent protein, which are fused to two proteins of interest. An interaction between the two proteins of interest brings them together leading to the reconstitution of the fluorescent protein Venus, which can be visualized by fluorescence microscopy. We created strains expressing solely or in combination Atg2 and Atg18 endogenously tagged with the N-terminal fragment of Venus (VN) and the C-terminal fragment of Venus (VC), respectively. After confirming that the fusion proteins are functional (supplementary material Fig. S5), cells were imaged before and after nitrogen starvation. In both the wild type strain and cells expressing only one of the fusion proteins, no fluorescence signal was detected (Fig. 7A, not shown). In the strain carrying both Atg2-VN and Atg18-VC, in contrast, a strong BiFC signal concentrating to a single perivacuolar punctuate structure was observed in presence and absence of nitrogen. Atg2 accumulates at the PAS in atg3Δ cells but not in atg13Δ mutant (Suzuki et al., 2007). When we repeated the experiment in these mutant backgrounds, we observed an increase in the percentage of cells positive for the perivacuolar punctuate BiFC signal in atg3Δ cells, and a complete loss in the atg13Δ strain demonstrating that the visualized puncta are PAS (Fig. 7A,B). The same experiment was also performed with the Atg2-binding mutants Atg18(L2) and Atg18(L5) fused to the VC tag. The frequency of cells displaying the BiFC signal was dramatically reduced in cells expressing Atg2-VN and these two chimeras (Fig. 7C,D). This result shows that the interaction between Atg2 and Atg18 at the PAS requires the specific targeting of Atg18 to this site, which is mediated through the dual recognition of Atg2 and PtdIns3P. In agreement with this hypothesis, when we preformed the BiFC assay for Atg2-VN and Atg18-VC in atg14Δ cells (Kihara et al., 2001), which do not generate PtdIns3P at the PAS, the BiFC signal was strongly reduced, especially in starvation conditions (Fig. 7C,D).

To confirm that Atg2 and Atg18 predominantly interact at the PAS and their binding requires the generation of PtdIns3P at this location, we performed an in vivo pull-down experiment in the atg14Δ background where both Atg2 and Atg18 are cytoplasmic (Shintani et al., 2001). As shown in Fig. 7E and in accordance with our previous results, we were able to co-isolate Atg2-PA and Atg18 but not Atg18(L2) in wild type cells (Fig. 7E, lanes 2 and 3). In contrast, in the atg14Δ knockout almost no wild type Atg18 was detected after pull-down (Fig. 7E, lanes 4 and 5). The data confirm our observation that the binding between Atg2 and Atg18 principally occurs at the PAS and requires the generation of PtDIns3P.

Cells expressing either Atg18(L2) or Atg18(L5) are still able to sustain minimal levels of autophagy (Fig. 3) and based on our data this could be due to the fact that these two Atg18 mutants are still capable of binding one of the determinants present at the PAS. To prove this hypothesis, we combined the mutations in loop 2 with those of loop 5, creating an Atg18 mutant unable to bind both Atg2 and PtdIns3P, and assessed autophagy in atg18Δ cells expressing this construct. As shown in supplementary material Fig. S6, this strain displayed a complete autophagy block. We conclude that the β-propeller plays a key role in mediating the specific association of Atg18 to the PAS by binding two determinants on this structure, PtsIns3P and Atg2.

The main conclusion of our work is that Atg18 β-propeller has a phosphoinositide-binding site that is combined with an organelle-specific binding site, which results in a multiplied affinity allowing the specific recruitment of Atg18 to either the PAS and possibly endosomes and the vacuole. Our findings thus highlight a novel way for localizing multi-purpose adaptor proteins onto different membranes by utilizing the versatile nature of the β-propeller as a platform.

Atg18 localizes to the PAS, vacuole and endosomes (Dove et al., 2004; Guan et al., 2001; Krick et al., 2008) and consequently the recruitment of this protein to these organelles has to be tightly regulated to correctly carry out its function. Atg18 association to membranes depends on the presence of either PtdIns3P and/or PtdIns(3,5)P₂. These phosphoinositides are enriched at the PAS, vacuole and endosomes (Gillooly et al., 2000; Obara et al., 2008a), implying that the temporal and spatial regulation of Atg18 localization must depend at least on one other organelle-specific factor.

The computational prediction of the Atg18 structure indicates that this protein folds into a seven-bladed β-propeller, each propeller unit composed of four-stranded antiparallel β-sheets, which are interconnected through six loops (Fig. 8A). During the submission and revision of the present paper, three reports describing the structure of Hsv2, a yeast homologue of Atg18 (see below), have been published and confirm this prediction (Baskaran et al., 2012; Krick et al., 2012; Watanabe et al., 2012). Loop 5 contains the FRRG motif that based on the Hsv2 model participates most probably in two different binding pockets that bind the phosphorylated lipid headgroups on the target membrane (Baskaran et al., 2012; Dove et al., 2004; Krick et al., 2006). Our data now reveal that one or more amino acids situated in loop 2 mediate Atg18 interaction with Atg2, a conclusion also reached by one of the recent structural works (Watanabe et al., 2012). Both loop 2 and 5 are positioned on the top of the β-propeller and protrude from its surface but they are on opposite sides of the barrel (Fig. 8A) (Baskaran et al., 2012; Krick et al., 2012; Watanabe et al., 2012). With loop 5 docking the barrel on membrane horizontally, loop 2 will be exposed towards the cytoplasm (Baskaran et al., 2012). It is thus very well possible that Atg18 β-propeller has the ability to simultaneously bind a phosphoinositide and Atg2, supporting the notion that it could act as a scaffold for the assembly of protein-lipid complexes. While to the best of our knowledge this is the first report of a β-propeller mediating the formation of protein-lipid complexes, other WD40 domain-containing proteins use the loops exposed on the top surface of their β-propeller to interact with multiple binding partners at the same time. For example, Ski8 plays an essential role in the assembly of a multi-protein complex, which also comprises Ski2 and Ski3, involved in the exosome-dependent mRNA decay, but also in the meiotic recombination by interacting with Spo11. The residues presents in the loops exposed on the top surface of Ski8 β-propeller were found to be important for all these different interactions (Cheng et al., 2004).

Our pull-down experiments show that the residues in loop 2 are predominantly mediating the binding of Atg18 to Atg2. A couple of evidences, however, indicate that specific residues in loop 1 could also be involved in the association between these two proteins. First, the interaction between Atg18 and Atg2 detected using the Y2H assay, where proteins are highly overexpressed, could only be abolished when the mutations in loops 1 and 2 were combined. Second, while the Atg18(L2) mutant is still able to sustain minimal levels of autophagy, the Atg18(L1,2) mutant displays a complete block of this pathway. Loop 1, however, appears to have no major roles in Atg18 binding to both Atg2 and phosphoinositides. Because of its proximity to loop 2 (Fig. 8A), we cannot exclude that few of the amino acids in loop 1 participate in the binding to Atg2. Alternatively, loop 1 could be used to regulate the Atg18-Atg2 interaction or other functions of Atg18. This hypothesis is evoked by the observation that the mutations in loop1 lead to the appearance of a lower molecular form of Atg18 (Fig. 2A), indicating that this protein undergoes a post-translational modification. The nature of this post-translational modification is currently under investigation.

Comparison of the amino acid sequence of loop 2 of the Atg18 β-propeller with the same region of Atg21 and Hsv2, two yeast proteins highly homologous to Atg18 that also bind phosphoinositides (Krick et al., 2006; Strømhaug et al., 2004), shows almost no conservation (Fig. 8B). In agreement with this observation, we have not been able to detect an interaction between Atg2 and Atg21 or Hsv2 (not shown). Atg18 is evolutionary related to the mammalian WD40 repeat protein Interacting with PhosphoInositides (WIPI) protein family, which compromises four proteins: WIPI1, WIPI2, WIPI3 and WIPI4 (Jeffries et al., 2004; Proikas-Cezanne et al., 2004). All WIPI proteins are suggested to also fold into a seven-bladed β-propeller with an open-velcro configuration, harbouring critical arginine residues for specific phosphoinositide binding (Proikas-Cezanne et al., 2007; Proikas-Cezanne et al., 2004). WIPI1, WIPI2 and WIPI4 have been implicated in autophagy (Lu et al., 2011; Polson et al., 2010; Proikas-Cezanne and Robenek, 2011; Proikas-Cezanne et al., 2004), inciting a debate about which one of them is the functional Atg18 ortholog. Recently, two mammalian Atg2 homologs, Atg2A and Atg2B, have been identified and both are required for autophagy (Velikkakath et al., 2012). Interestingly, human WIPI4 interacts with Atg2A and Atg2B as well as C. elegans EPG-6/WIPI4 with ATG-2 (Behrends et al., 2010; Lu et al., 2011). These observations suggest that WIPI4/EPG-6 and yeast Atg18 overlap in their role in autophagy by carrying out the functional interconnections with Atg2. The amino acid sequence alignment of loop 2 of the Atg18 β-propeller with that of various WIPI4 proteins from different species supports this idea, because several amino acids are well conserved except for C. elegans EPG-6 (Fig. 8C). The ATG-2-binding site of EPG-6 has been mapped to the fifth and sixth blades of the EPG-6 β-propeller, and this could explain the lack of amino acid conservation in loop 2 of EPG-6 β-propeller (Lu et al., 2011). This binding region, however, was identified by sequential deletion of the blades and therefore it cannot be excluded that this type of approach leads to a complete disruption of the β-propeller structure making the interpretation of the result difficult.

Similarly to the Atg18 mutant unable to bind phosphoinositides, the one blocking the interaction between Atg18 and Atg2, i.e. Atg18(L2), severely impairs the progression of non-selective and selective autophagy by affecting autophagosome biogenesis at an early stage. These defects are caused by the inability of these two mutant proteins to be recruited to the PAS. The two binding capacities of Atg18 β-propeller, however, appear to not be reciprocally regulated but rather independent. That is, mutation of the FRRG motif within loop 5 does not affect the interaction between Atg18 and Atg2, and conversely the Atg2-binding mutant of Atg18 is capable of binding phosphoinositides. The fact that cells expressing either Atg18(L2) or Atg18(L5) are still able to sustain minimal levels of autophagy supports this notion as Atg18 is still capable of binding one of the determinants present at the PAS. Indeed, when we combined the two sets of mutations, creating an Atg18 mutant unable to bind both Atg2 and PtdIns3P, we observed a complete autophagy block.

Based on our observations, we propose the following model for Atg18 recruitment to the PAS (Fig. 8D). Upon induction of a double-membrane vesicle formation, one of the first events during the organisation of the PAS is the association of the phosphatidylinositol-3 kinase complex I to it (Suzuki et al., 2007), which presumably starts synthesizing PtdIns3P. As reported (Shintani et al., 2001; Suzuki et al., 2007), the phosphatidylinositol-3 kinase complex I and/or PtdIns3P are required for Atg2 recruitment to the PAS. So far, we have not been able to determine whether Atg2 binds PtdIns3P. Therefore, we do not know whether this protein directly or indirectly interacts with this lipid to associate with the PAS. Nevertheless, the simultaneous presence of PtdIns3P and Atg2 at this location allows the Atg18 β-propeller to bind to this structure with high affinity, and subsequently forms the Atg18-Atg2 complex. Our fluorescence microscopy data support this model, because they show that Atg2 can be recruited to the PAS independently from Atg18. Furthermore, the BiFC experiments indicate that the Atg2-Atg18 interaction occurs at this site.

Our results are in part contradictory with the published literature, which indicate that Atg18 and Atg2 form a cytoplasmic complex that is recruited as a unit to the PAS and Atg2 fails to associate with this structure in absence of ATG18 (Obara et al., 2008b; Suzuki et al., 2007). While we observed that Atg18 recruitment to the PAS requires Atg2 as reported (Obara et al., 2008b; Suzuki et al., 2007), in our hands Atg2 localizes to the PAS independently from Atg18. The fact that Atg2 was not detected at the PAS in previous studies (Guan et al., 2001; Suzuki et al., 2007) could be due to a difference in the microscope sensitivity or strain background. Nonetheless, we also found that Atg2 association with the PAS requires Atg1, Atg9 and Atg14 [not shown (Shintani et al., 2001; Suzuki et al., 2007; Wang et al., 2001)]. The assumption that Atg2 and Atg18 form a cytoplasmic complex is based on data showing that upon gel filtration of solubilized cell extracts, Atg2 is not detected as a monomer but rather in a complex of ∼500 kDa, and part of Atg18 is in the same fraction (Obara et al., 2008b). We fractionated proteins and protein complexes present in solubilized cell extracts from a wild type strain expressing endogenously PA-tagged Atg2 on a continuous 10–50% glycerol gradient (supplementary material Fig. S7A,B). We did not observe an evident difference in the distribution of Atg2 over the gradient in presence or absence of Atg18. Interestingly, the fact that most of Atg2 and Atg18 appeared to be not associated, i.e. not cofractionating on the gradient, supports our model where these two proteins preferentially bind each other at the PAS. To substantiate this observation, we also overexpressed Atg2 with or without Atg18. Overexpression of both proteins resulted in very similar fractionation profiles as with the endogenous proteins (supplementary material Fig. S7C,D). The absence of Atg18 did not influence the fractionation profile of overexpressed Atg2, indicating that the apparent high molecular mass of Atg2 is due to some structural characteristics of this protein and/or self-interaction. While we cannot exclude that Atg2 and Atg18 can also form a cytoplasmic complex under certain circumstances, we currently ascribe the observed differences to a more sensitive strain background or experimental conditions. Alternatively, a yet unidentified factor, which depends on Atg18, mediates the functional Atg2-Atg18 interaction and this is reflected by our experimental setup at least in some circumstances.

Alternative models describing the Atg18 recruitment to the PAS can also be contemplated. Therefore, additional studies are necessary to fully understand the regulation of this event and the unique property of the Atg18 β-propeller to form lipid-protein complexes. This information will also be crucial to understand the function of Atg18 in autophagy.

The S. cerevisiae strains used in this study are listed in supplementary material Table S1. For gene disruptions, coding regions were replaced with genes expressing auxotrophic markers using PCR primers containing ∼60 bases of identity to the regions flanking the open reading frame. Gene knockouts were verified by examining prApe1 processing by western blot analysis using a polyclonal antibody against Ape1 (Mari et al., 2010) and/or PCR analysis of the deleted gene locus.

Chromosomal tagging of the ATG2 gene at the 3′ end was done by PCR-based integration of the PA or the GFP tag using pFA6a-PA-TRP1 and pFa6a-GFP(S65T)-TRP1 as template plasmids, respectively (Longtine et al., 1998). For construction of the strain expressing Atg2 under the control of the GAL1 promoter, the ATG2 gene was chromosomally tagged at the 5′ end with the GAL1 promoter and the HA tag by PCR-based integration using the pFA6a-HIS3MX6-PGAL1-3HA plasmid as a template (Longtine et al., 1998). Chromosomal taggings were verified by western blot analysis using antibodies against goat IgG for the detection of PA (Invitrogen Life Science, Carlsbad, CA), or monoclonal antibodies recognizing either GFP (Roche, Basel, Switzerland) and HA (Covance, Princeton, NJ).

For the BiFC assay, the ATG2 and ATG18 genes were chromosomally tagged at the 3′ end by PCR-based integration of the VN or VC fragments using pFA6a-VN-HIS3MX6 or pFa6a-VC-TRP1 as template plasmids (Sung and Huh, 2007). Correct integration of the tags was verified by PCR.

Yeast cells were grown in rich (YPD; 1% yeast extract, 2% peptone, 2% glucose) or synthetic minimal media (SM; 0.67% yeast nitrogen base, amino acids and vitamins as needed) containing 2% glucose. Starvation experiments were conducted in synthetic media lacking nitrogen (SD-N; 0.17% yeast nitrogen base without amino acids, 2% glucose). For galactose-induced overexpression of proteins, cells were grown overnight in SM medium containing 2% glucose, diluted and regrown to an exponential phase in SM medium containing 2% raffinose. Protein overexpression was then induced by transferring cells in SM medium containing 2% galactose overnight.

For the construction of the Y2H plasmids, DNA fragments encoding ATG2 and ATG18 were generated by PCR using S. cerevisiae genomic DNA as a template and cloned as a XmaI-SalI and EcoRI-SalI fragment, respectively, into both pGAD-C1 and pGBDU-C1 vectors (James et al., 1996). The C-terminal truncation of ATG18 (1–377) was generated by PCR using the same 5′ primer as for the cloning of the full-length gene and a 3′ primer for the specific site of truncation, which introduced a stop codon followed by a SalI restriction site. The point mutations in ATG18 designed to create the loop mutants were introduced by PCR using unique restriction sites in close proximity of the nucleotide stretch coding for each loop. Combinations of Atg18 loop mutants were made by PCR by combining primers used to create the different loop mutants and already constructed Atg18 loop mutant plasmids as templates. The correct introduction of the point mutations was verified by DNA sequencing.

Plasmids expressing untagged Atg18 loop mutants under the control of the endogenous promoter of Atg18 were constructed by PCR using the Y2H plasmids of the different loop mutants as templates and cloned as NheI-NdeI fragments into the pCvt18(415) plasmid, which carries the wild type ATG18 gene (Guan et al., 2001).

The promAtg18GFP416 plasmid was generated by amplifying the promoter (700 bp) and the ATG18 gene from genomic DNA by PCR and cloning it as an XhoI-BclI fragment in a pRS416 vector (Sikorski and Hieter, 1989) digested with XhoI-BamHI and containing a (gly-Ala)₃-linker and GFP, which were inserted as a BamHI-SacII fragment at the 3′ end of the gene. The GFP-fusion proteins of the Atg18 loop mutants were then generated by PCR using the Y2H plasmids of the different loop mutants as templates and cloned as Tth111I-BsiWI fragments into the Atg18-GFP plasmid. To create the plasmids expressing the different ATG18 mutants tagged with 13×myc, the GFP gene was replaced with the sequence coding for the 13×myc tag followed by the ADH1 terminator obtained by PCR from the pFA6a-13×myc plasmid (Longtine et al., 1998). To create the vectors integrating the various Atg18-GFP and Atg18-13×myc constructs into the genome, the backbone of the expression plasmids was replaced with that of the pRS405 vector (Sikorski and Hieter, 1989) using XhoI and SacI. Correct integration of the different constructs was verified by western blot analysis using polyclonal antibodies against myc (Santa Cruz Biotechnology, Santa Cruz, CA) or monoclonal antibodies against GFP. Plasmids expressing the different ATG18 loop mutants under the control of the GAL1 promoter were generated by PCR amplification of the sequences coding for the different Atg18 loop mutants, plus the 13×myc tag and the ADH1 terminator from the plasmids described above. The PCR fragments were cloned into the pRS416 vector (Sikorski and Hieter, 1989) using HindIII and KpnI before inserting the GAL1 promoter using XhoI and HindIII.

The pCuGFPATG8414 and pCuGFPATG8416 plasmids expressing GFP-Atg8 under the control of the CUP1 promoter have been described elsewhere (Kim et al., 2002). To create the integrative pCFPATG8406 plasmid that leads to the expression of the CFP-Atg8 fusion protein from the authentic ATG8 promoter, the backbone of the pRS314 ECFP-AUT7 plasmid (Suzuki et al., 2001) was exchanged for that of the pRS406 vector (Sikorski and Hieter, 1989) using XhoI and SacII. The integrative pCumCheV5ATG8406 plasmid, which expresses mCherry-V5-Atg8 from the CUP1 promoter, was generated by replacing the vector backbone of the pCumCheV5ATG8415 plasmid (Mari et al., 2010) with that of the pRS406 vector using KpnI and SacI.

To create the promAtg18VC405, promAtg18(L2)VC405 and promAtg18(L5)VC405 plasmids, the VC fragment was amplified by PCR using the pFa6a-VC-TRP1 plasmid as a template and used to replace the myc tag coding sequence in the integrative plasmids expressing wild type Atg18-13×myc, Atg18(L2)-13×myc and Atg18(L5)-13×myc using PacI and SacI. Correct integration of the constructs into yeast was then verified by PCR analysis.

The plasmids pGAD-C1 and pGBDU-C1 containing ATG2 and ATG18 or its mutated and truncated forms were transformed into the PJ69-4A test strain and grown on 2% glucose-containing SM medium lacking leucine and uracil (James et al., 1996). Colonies were then spotted on 2% glucose-containing SM medium lacking histidine, leucine and uracil.

Cells were grown in YPD medium or nitrogen starved in SD-N medium for 3 h. Vacuoles were stained with the FM4-64 dye (Invitrogen Life Science) as previously described (Vida and Emr, 1995). Fluorescence signals were captured with a DeltaVision RT fluorescence microscope (Applied Precision, Issaquah, WA) equipped with a CoolSNAP HQ camera (Photometrix, Tucson, AZ). Images were generated by collecting a stack of 18 pictures with focal planes 0.20 µm apart, and by successively deconvolving and analyzing them with the SoftWoRx software (Applied Precision). A single focal plane is shown at each time. The percentage of cells positive for a CFP-Atg8, Atg2-GFP or BiFC-positive puncta was determined by analysing at least 100 cells from two independent experiments. To determine the degree of colocalization between the fusion proteins Atg18-GFP and mCheV5-Atg8, the number of mCheV5-Atg8 puncta positive for the Atg18-GFP signal was counted in at least 100 cells from two independent experiments.

Phospholipid-binding assays with native cell extracts has previously been described (Proikas-Cezanne et al., 2007) and adapted to yeast as follows. Native yeast cell extracts were generated from 200 OD₆₀₀ equivalents of frozen yeast cells by vortexing them three times for 30 s in the binding buffer [750 mM aminocaproic acid, 50 mM Bis-Tris, 0.5 mM EDTA, pH 7.0, supplemented with protease and phosphatase inhibitor cocktails (Roche)] in presence of glass beads. The soluble fraction was obtained by removing the glass beads and cell debris through centrifugation at 14,000 rpm for 15 min at 4°C. Before use, the membrane-immobilized phospholipids (Echelon Biosciences, Salt-Lake City, UT) were rinsed first in TBS buffer (Tris-buffered saline, 20 mM Tris-HC1, 0.15 M sodium chloride, pH 7.5) and then in 0.1% Tween 20 in TBS buffer before blocking them in 0.1% Tween 20/3% BSA in TBS buffer for 1 h at room temperature. Membranes were subsequently incubated with cell extracts at 4°C for 2 days before detecting bound proteins by western blotting using anti-myc antibodies.

A total of 100 OD₆₀₀ equivalents of exponentially growing cells were collected and resuspended in 300 µl of lysis buffer. Cells were broken using glass beads and vortexing, and centrifuged at 13,000 rpm for 10 min at 4°C. The resulting cell extract was subjected to high-speed centrifugation at 45,000 rpm for 1 h at 4°C, before being separated on a continuous 10–15% glycerol gradient by centrifugation at 33,000 rpm for 18 h. Fractions (861 µl) were collected and precipitated using trichloroacetic acid. Proteins were resolved by SDS-PAGE and analyzed by western blotting using antibodies against myc and goat IgG. Molecular mass protein standards were used to calibrate the gradient (GE Healthcare).

The protein extraction, PA affinity purifications, western blot analyses and the GFP-Atg8 processing assay were carried out as previously described (Cheong and Klionsky, 2008; Reggiori et al., 2003). Detection and quantification of western blotting were performed using an Odyssey system (Li-Cor Biosciences, Lincoln, NE). Processing of the electron microscopy samples and the counting of autophagic bodies has already been illustrated (van der Vaart et al., 2010).

We thank Daniel Klionsky for reagents and Janice Griffith for technical assistance.