Atg9A trafficking through the recycling endosomes is required for autophagosome formation

Macroautophagy, hereafter called autophagy, is a process induced by starvation during which intracellular material is degraded and recycled for cellular homeostasis. In addition to such a function, autophagy plays numerous important physiological roles, such as in damaged organelle clearance, providing cellular immunity against invading pathogens, and tumor suppression (Kawabata and Yoshimori, 2016). In the process of autophagy, a double-membrane structure, called an autophagosome, is formed to sequester and deliver cytosolic proteins and organelles into the lysosomes. The formation of an autophagosome depends on a series of autophagy-related (Atg) proteins that are highly conserved from yeast to mammals and are required for autophagosome formation (Nakatogawa et al., 2009; Shibutani and Yoshimori, 2014).

Mammalian Atg9A, the yeast Atg9 homolog, is the sole multi-spanning transmembrane protein among the core Atg proteins. Although the transmembrane domain of Atg9A is well conserved between yeast and mammals, its length and amino acid sequence outside of this domain are not well conserved (Noda et al., 2000; Young et al., 2006). In mammalian cells, Atg9A mainly localizes to the trans-Golgi network (TGN) and also the endosomal system, and is reported to cycle between them though vesicle trafficking. Upon autophagy induction, a portion of Atg9A transiently localizes to the autophagic membranes (Orsi et al., 2012; Popovic and Dikic, 2014; Young et al., 2006). Recent studies have demonstrated that membrane trafficking pathways concerned with clathrin and/or adaptor proteins (Guo et al., 2012; Popovic and Dikic, 2014), Rab GTPases (Carlos Martín Zoppino et al., 2010; Itoh et al., 2008; Longatti et al., 2012) and the retromer complex (Zavodszky et al., 2014) are important for both proper Atg9A localization and autophagosome formation, indirectly suggesting that proper Atg9A trafficking is necessary for autophagy. However, the approach of manipulating general trafficking machinery used in the above studies means that it remains possible that defects in transport of molecules other than Atg9A itself causes autophagic inhibition. Therefore, we decided to pursue the significance of Atg9A trafficking in autophagy more directly by searching its intrinsic amino acid sequences for motifs required for its trafficking.

Transmembrane protein trafficking between the TGN, endosomes and the plasma membrane generally requires sorting motifs that are located within the cytoplasmic region of the protein (Bonifacino and Traub, 2003; Rohn et al., 2000). Such sorting motifs are usually recognized by adaptor protein (AP) complexes like AP-1 and AP-2, which sort and pack the transmembrane proteins into a transport vesicle as cargoes (Bonifacino and Traub, 2003). Hence, we attempted to investigate whether Atg9A has such sorting motifs, and succeeded to identify them in the N-terminal cytosolic stretch of Atg9A, which binds with AP-2. Mutation of these motifs brought about a trafficking defect of Atg9A, whereby Atg9A accumulated at the recycling endosomes, along with an autophagy defect. This combination of defects were also observed upon the knockdown of TRAPPC8, a specific subunit of the TRAPPIII complex. These findings prove that Atg9A trafficking from the recycling endosomes is a crucial step for the execution of both canonical and selective autophagy.

Atg9A is a membrane-spanning protein and its N- and C-terminal domains are extruded to the cytoplasm (Young et al., 2006). To explore the function of these cytoplasmic regions, we made a series of truncation mutants of Atg9A tagged with GFP at the C-terminus (Fig. S1A), and then stably expressed them in Atg9A-knockout (KO) mouse embryonic fibroblast (MEF) cells. Compared to the wild-type (WT)-MEFs, Atg9A-KO MEFs showed a significantly lower number of autophagosomes, as visualized by immunostaining of LC3B proteins (also known as MAP1LC3 proteins; hereafter referred to collectively as LC3), an autophagosome marker (Kabeya et al., 2000; Saitoh et al., 2009), under nutrient-starved conditions (Fig. 1A). The deficiency was completely rescued by the stable expression of Atg9A-WT in Atg9A-KO MEFs (Fig. 1A). Whereas expression of C-terminal deletion mutants in Atg9A-KO MEFs completely restored the autophagic activity (Fig. S1B), cells expressing N-terminal deletion mutants showed defects in LC3 puncta formation. Cells expressing the ΔN(1-36) mutant of Atg9 had significantly fewer numbers of LC3 puncta under nutrient-starved conditions (Figs 1A and 2A). These results were also confirmed by another autophagy assay called the autophagic flux assay (Tanida et al., 2005). The lipidated form of LC3 (LC3-II) is degraded in autolysosomes, thus the level of LC3-II significantly increases in autophagy-competent cells upon treatment with Bafilomycin A1, which inhibits autophagosome and lysosome fusion. In contrast to the Atg9A-WT cells, accumulation of LC3-II was significantly blocked in Atg9A-KO MEFs expressing the ΔN(1-36) mutant (Fig. 1B). Thus, the N-terminal region (amino acids 1–36) in Atg9A is important for the role of Atg9A in autophagosome formation.

Next, we examined the localization of Atg9A mutant proteins. As previously reported, full-length Atg9A–GFP showed colocalization with the Golgi marker GM130 (also known as GOLGA2) (Fig. 1C) (Young et al., 2006). Deleting the C-terminal stretch did not affect Atg9A localization (Fig. S1C); by contrast, the N-terminal truncated mutant ΔN(1-12), and more severely with ΔN(1-36), did not show Golgi localization (Fig. 1C). Further deletion of the N- or C- terminal domain meant the protein failed to be expressed (data not shown). These results suggest that the N-terminal region (1–36) in Atg9A has a role in both autophagy and the correct localization of the protein.

We further characterized the N-terminal region of Atg9A. We noticed that the N-terminal region (1–36) of Atg9A has three putative sorting motifs as follows (Fig. 2A): (1) a di-acidic motif (residues 5–7) (DxE, where x represents any amino acid), (2) a tyrosine-based motif (residues 8–11) (YxxΦ, where Φ represents a bulky hydrophobic group), and (3) a di-leucine-based sorting motif (residues 22–27) [(D/E)xxxL(L/I/M/V) or D/ExxLL]. The di-acidic motif is recognized by Sec24, which functions in endoplasmic reticulum (ER) export (Nishimura and Balch, 1997), and both the tyrosine-based and di-leucine-based sorting motifs are recognized by the adaptor protein (AP), which functions in post-Golgi trafficking (Bonifacino and Traub, 2003), and are also recognized by Sec24 (Barlowe, 2003). Each sorting motif was replaced with alanine residues and expressed in Atg9A-KO MEFs, and their effects were then observed. The strongest defect in autophagosome formation was found when both the tyrosine-based and di-leucine-based sorting motifs (ΔYQRL+ΔLLV) were mutated (Fig. 2B). Consistent with this result, in the ΔYQRL+ΔLLV mutant, the LC3-II flux was significantly blocked (Fig. 2C), and degradation of p62 (also known as SQSTM1), a selective cargo protein of autophagy, was severely defected (Fig. S2A).

We next assessed whether the phenotypes of the ΔYQRL+ΔLLV mutant could be rescued by ectopically adding a sorting motif (+YQRL) or four alanine residues (+4Ala), as a control, at its N-terminus (Fig. 2A). The defects in autophagosome formation of the ΔYQRL+ΔLLV mutant was recovered by the external addition of YQRL motif, but not with the four alanine residues (Fig. 2D). These results demonstrate that the sorting motifs of Atg9A are required for its function in autophagy.

Next, we observed the localization of the Atg9A mutants to see the function of these sorting motifs in its trafficking. Whereas mutation in the di-acidic motif did not affect the localization, mutation in either the tyrosine-based motif or di-leucine-based sorting motif decreased Atg9A localization on the Golgi, and instead the protein became dispersed within the cytosol as dot-like structures. Double mutation (ΔYQRL+ΔLLV) diminished the Golgi localization the most under nutrient-rich conditions (Fig. 3A). The decreased Golgi localization of the ΔYQRL+ΔLLV mutant was recovered upon the addition of the YQRL motif at N-terminus of the mutant, but not with four alanine residues (Fig. 3B). These results show that the Golgi localization of Atg9A depends on the sorting motifs.

It could have been possible that localization change of the Atg9A mutants was caused by a defect in exiting from the ER because the tyrosine-based and di-leucine-based sorting motifs also work as an ER export signal (Barlowe, 2003). However, this was not the case because the Golgi form of N-glycan modification on Atg9A was observed in all mutants (Fig. S3). Therefore, it would be expected that these Atg9A mutants would be localized at a post-Golgi compartment, and we next aimed to assess the intracellular localization where the trafficking of the Atg9A mutants halted. They did not accumulate at the autophagosome formation sites, because they did not colocalize with the WIPI2 and Atg16L1 markers for the autophagosome formation site (Fig. S2B). Instead, a noticeable amount of the Atg9A accumulated at the juxtanuclear region when the tyrosine-based motif and/or di-leucine-based sorting motif were impaired under nutrient-starved conditions (Fig. 3A). These accumulated structures colocalized well with a recycling endosome marker (Alexa-Fluor-568–transferrin), which was not seen in the WT (Fig. 4A; Fig. S4A). We quantified this phenotype and found that the ΔYQRL+ΔLLV mutant showed 54.3% colocalization, whereas only 7.9% was found in WT (Fig. S4C). The ΔYQRL+ΔLLV mutant no longer accumulated on the recycling endosomes when the sorting motif was ectopically added to the N-terminus of this mutant (Fig. 4A, Fig. S4B). These results indicate that Atg9A depends on the sorting motifs to be transported from the recycling endosomes.

The above results prompted us to further explore the adaptor proteins(s) that interact with the identified sorting motifs. So far there are two past studies about the adaptor proteins potentially involved in Atg9A trafficking, AP-1 and AP-2 (Guo et al., 2012; Popovic and Dikic, 2014). We therefore examined whether either of them was associated with the sorting motif by pulldown experiments. We found that we could successfully pulldown the AP-2 subunit adaptin α together with Atg9A–GFP, but not the AP-1 subunit adaptin γ, at least in our experimental system (ATG9A KO MEFs expressing functional Atg9A–GFP) (Fig. 4B). Importantly, the ΔYQRL+ΔLLV mutant Atg9A–GFP pulled down significantly less AP-2 than wild-type Atg9A (Fig. 4B). Thus, AP-2 is most likely to be the adaptor protein, at least in the sorting step where the sorting motif is involved.

We further extended our study to strengthen the importance of Atg9A trafficking in autophagy by independent means. Recently, we have reported that the autophagy defect caused by the trs85 mutant in yeast is due to a defect in yeast Atg9 trafficking from the endosome to the Golgi (Shirahama-Noda et al., 2013). Trs85 is a specific subunit of the TRAPPIII complex, a guanine-nucleotide exchange factor for Ypt1 (a Rab1 homolog), and functions as a vesicle-tethering complex, mediating general retrograde trafficking from endosomes to the Golgi (Lynch-Day et al., 2010; Shirahama-Noda et al., 2013). Interestingly, it has been reported that knockdown of TRAPPC8, a mammalian homolog of Trs85, blocks both retrograde transport of Shiga toxin from the endosome to the ER (Kampmann et al., 2013) and autophagy (Behrends et al., 2010). Therefore, we examined the possibility that TRAPPC8 knockdown mediated by small interfering RNA (siRNA) affects Atg9A trafficking. Although Atg9A was only minimally colocalized with the recycling endosome marker (Alexa-Fluor-568–transferrin) in control siRNA-transfected cells, it was clearly colocalized with this marker in TRAPPC8-siRNA-transfected cells (Fig. 5A), indicating that Atg9A was trapped on the recycling endosomes upon TRAPPC8 knockdown. We could also reproduce the results from the previous report showing that knockdown of TRAPPC8 severely blocked autophagosome formation (Fig. 5B; Behrends et al., 2010). These observations support the idea that Atg9A trafficking through the recycling endosomes is required for autophagosome formation.

We finally asked whether the sorting motifs in Atg9A are involved in selective autophagy as well as starvation-induced nonselective autophagy. Autophagy targets intracellular-invading Salmonella and suppresses their growth inside cells (Birmingham et al., 2006). In this process, Atg9A is recruited to the intracellular Salmonella that are positive for ubiquitin before eventually dissociating from them (Fujita et al., 2013; Kageyama et al., 2011). First, we assessed whether the sorting motifs are required for the recruitment of Atg9A to the bacteria. To this purpose, we calculated the percentage of Atg9A recruitment to ubiquitin-positive Salmonella. We used FAK-family-interacting protein of 200 kDa (FIP200)-KO MEFs (FIP200 is also known as RB1CC1), in which Atg9A is targeted to the Salmonella as normal but cannot dissociate from them; thus, Atg9A-positive Salmonella accumulate in FIP200-KO MEFs compared to WT MEFs (Kageyama et al., 2011). The percentage of the ΔYQRL+ΔLLV mutant recruitment to Salmonella in these MEFs was significantly lower than that of Atg9A-WT (Fig. 6A), demonstrating that the sorting motifs were required for the recruitment of Atg9A to the Salmonella.

Next, we tested whether the sorting motifs were indispensable for the bacterial growth arrest. The number of Salmonella within the host cells was counted at the indicated time points. Atg9A-KO MEFs expressing Atg9A-WT suppressed Salmonella growth; however, Atg9A-KO MEFs expressing ΔYQRL+ΔLLV allowed Salmonella to grow (Fig. 6B), indicating that the sorting motifs are required for the function of Atg9A in selective autophagy. Taken together, we conclude that the sorting motifs of Atg9A are indispensable for selective autophagy as well as canonical autophagy.

In this study, we have succeeded in identifying the intrinsic Atg9A sorting motif (tyrosine-based and di-leucine-based sorting motif) at its N-terminal cytoplasmic region, which binds to AP-2. This motif is crucial for its proper localization and for autophagy. In contrast to previous studies showing that general membrane trafficking defects somehow lead to an autophagy defect, which was analyzed by several methods such as knockdown of membrane trafficking machineries (Guo et al., 2012; Itoh et al., 2008; Longatti et al., 2012; Popovic and Dikic, 2014; Takahashi et al., 2011; Webber and Tooze, 2010), drugs (Köchl et al., 2006; Puri et al., 2013) and temperature block (Puri et al., 2013) methods, our findings directly clarify that Atg9A trafficking itself is prerequisite for autophagy. Our results do not necessarily exclude the possibility that some other cargoes of general membrane trafficking are also required for autophagy. However so far, there are only two transmembrane proteins that have been shown to be required for autophagy, Atg9A and VMP1, and VMP1 is an ER-resident protein (Ropolo et al., 2007). So at this stage, we propose that Atg9A is the sole known cargo protein under the control of general membrane trafficking that is essential for autophagy.

The Atg9A mutant lacking both the tyrosine-based and di-leucine-based sorting motifs exhibited a decreased Golgi localization, and abnormal accumulation on the recycling endosomes under nutrient-starved conditions (Fig. 4A). The adaptor protein AP-2 would be expected to directly bind to this motif and direct Atg9A to the transport vesicle destined to the next destination, the Golgi and possibly, successively, to the autophagosome (Fig. 7A). This also suggests that nutrient starvation conditions lead to the rerouting of Atg9A trafficking so that it passes through the recycling endosomes.

The importance of exit of Atg9A from the recycling endosomes was also supported by the other experiment, knockdown of TRAPPC8. Resulting accumulation of Atg9A on the recycling endosomes indicates that TRAPPC8 is involved in the retrograde trafficking of Atg9A from the recycling endosomes to the Golgi. This idea is consistent with a recent paper from the laboratory of Sharon Tooze reporting that TRAPPIII complex can bind with TBC1D14 at the recycling endosomes, where Rab1 is activated, and can mediate membrane trafficking from endosomes to the Golgi (Lamb et al., 2015). They also showed that TBC1D14 overexpression causes tubulation of the recycling endosomes and defects in the recycling pathway, suggesting that interaction between TBC1D14 and TRAPPIII complex is required for membrane trafficking from the recycling endosomes to the Golgi. Collectively, Atg9A trafficking from recycling endosomes is important for autophagosome formation [Fig. 7B, pathway (ii)]. This is in line with our previous report showing that in yeast the defect in endosome-to-Golgi trafficking leads to failure of yeast Atg9 arrival to autophagosome or PAS (preautophagosomal structure). Endosome-to-golgi trafficking, and possibly a shuttling between the endosome and the Golgi, could serve as a reservoir for yeast Atg9 protein, with a portion of the yeast Atg9 being delivered to PAS and autophagosome from there (Shirahama-Noda et al., 2013).

It has been suggested that the recycling endosome membrane is one of the membrane sources for autophagosome formation (Knævelsrud et al., 2013; Longatti et al., 2012; Orsi et al., 2012); thus, it might be possible that Atg9A could mediate the delivery of core nucleation factor(s) from the recycling endosomes to the autophagosome. However, a more plausible possibility revealed in this study is that delivery of Atg9A itself from the recycling endosomes to the Golgi, rather than some specific membrane lipid, is the crucial role of membrane trafficking during autophagy. However it is still possible that Atg9A recruits some unknown factors from the recycling endosomes to the autophagosome, and we need to decipher these possibilities in future study. We note that there is also a possibility that Atg9A reaches the forming autophagosome directly from the recycling endosomes [Fig. 7B, pathway (i)]. Canonical and selective autophagy share some common mechanisms and differences; however, the whole picture has not yet been clarified. Finding that Atg9A trafficking is indispensable for both types of autophagy is an important first step to fully understand the mechanisms. Our results establish that Atg9A trafficking itself is the reason that membrane trafficking is required in autophagy, and future study will give us more detailed understanding regarding the role of Atg9A.

pENTR1A vector was purchased from Invitrogen. The pMRX-IRES-puro vector was kindly provided by Toshio Kitamura (University of Tokyo, Tokyo, Japan) (Morita et al., 2000). The DNA fragment containing attR1, ccdB, CmR and attR2 was PCR amplified from pAd/CMV/V5-DEST (Invitrogen) and digested with BglII-NotI. The fragment was subcloned into the BamHI-NotI sites of pMRX-IRES-puro (pMRX-IRES-puro-DEST). pEGFP-Atg9A was generously provided by Toshihiko Yamada (Hokkaido University, Hokkaido, Japan). pDsRed-KDEL was kindly provided by Eiji Morita (Hirosaki University, Hokkaido, Japan). EGFP was PCR amplified from pEGFP-N1 (Invitrogen) and digested with BglII-NotI. The fragment was subcloned into the BamHI-NotI sites of pENTR1A (pENTR1A-EGFP). Atg9A was PCR amplified from pEGFP-Atg9A and digested with EcoRI-KpnI. The fragment was subcloned into the EcoRI-KpnI sites of pENTR1A-EGFP (pENTR1A-Atg9A-EGFP). Various Atg9A mutants were created from pENTR1A-Atg9A-EGFP with the QuikChange technique, including ΔN(1-12), ΔN(1-36), ΔC(799-839), ΔC(715-839), ΔC(672-839), ΔC(630-839), ΔC(585-839), ΔDTE, ΔYQRL, ΔLLV, ΔDTE+ΔYQRL and ΔYQRL+ΔLLV, and RXR (RXR construct was used in Fig. S3). pENTR1A-Atg9A-EGFP and pENTR1A-Atg9A-EGFP including various Atg9A mutation sequences were transferred into the pMRX-IRES-puro-DEST vector using an LR reaction. Recombinant retroviruses were prepared as previously described (Saitoh et al., 2003). Plat-E cells were generously provided by Toshio Kitamura (Morita et al., 2000).

The following antibodies were used: mouse monoclonal anti-GFP (1:2000, cat. no. 11814460001, Roche), mouse monoclonal anti-GM130 (1:1000, cat. no. 610822, BD), mouse monoclonal anti-p230 (1:1000, cat. no. 611280, BD), mouse monoclonal anti-TfR (1:500, cat. no. 13-6800, Invitrogen), mouse monoclonal anti-poly-ubiquitin (1:2000, cat. no. BML-PW8810, clone FK2; BIOMOL), mouse monoclonal anti-GAPDH (1:2000, cat. no. Mab314, Chemicon, Merck Millipore), mouse monoclonal anti-γ-adaptin (1:3000, cat. no. 610386, BD), rabbit polyclonal anti-LC3 (1:1000 for immunofluorescence, 1:2000 for western blotting, cat. no. PM036, MBL), rabbit polyclonal anti-p62 (1:1000, cat. no. PM045, MBL), rabbit polyclonal anti-EEA1 (1:100, cat. no. 2411S, Cell Signaling Technology), mouse monoclonal anti-α-adaptin (1:3000, cat. no. 610502, BD), rabbit polyclonal anti-Atg9 (1:500; cat no. PD042, MBL), rabbit polyclonal anti-Atg16L1 (1:500, cat. no. TMD-PH-AT16L, Cosmobio), rabbit polyclonal anti-WIPI-2 (1:1000, cat. no. SAB4200400, Sigma). Bafilomycin A1 was purchased from Sigma-Aldrich. Alexa-Fluor-568–transferrin was purchased from Invitrogen. Wheat germ agglutinin (WGA)-conjugated Alexa Fluor 568 was purchased from Invitrogen. Endo Hf was purchased from New England Biolabs. PNGaseF was purchased from New England Biolabs.

Samples were fixed with 4% paraformaldehyde for 20 min, quenched with 50 mM NH₄Cl or phosphate-buffered saline (PBS), permeabilized with 50 μg/ml digitonin in PBS, blocked with 0.1% gelatin in PBS and then incubated with the indicated primary antibodies. After washing with PBS, the samples were incubated with secondary antibodies and mounted with Slow Fade Gold (Invitrogen). The microscopy images were taken using a FV1000 confocal laser-scanning microscope system equipped with a 60×1.40 NA oil immersion objective lens. Fluorochromes associated with the secondary antibodies were conjugated to Alexa Fluor 568 or 647. Image acquisition software Fluoview (Olympus) was used. The images were adjusted using Photoshop CS4 software (Adobe). The number of LC3 particles was analyzed by G-Count (G-Angstrom).

MEFs, HeLa Kyoto and Plat-E cells were cultured in Dulbecco's modified Eagle's medium (DMEM D6429; Sigma-Aldrich) containing 10% fetal bovine serum (FBS), 5 U/ml penicillin, and 50 U/ml streptomycin at 37°C with 5% CO₂ (denoted growth medium). For nutrient starvation, cells were cultured in EBSS (Sigma-Aldrich). Transient transfection was carried out using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's protocol.

Cells were lysed in ice-cold RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% w/v Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate) containing protease inhibitor cocktail (Roche) and 1 mM phenylmethylsulfonyl fluoride. Lysates were mixed with 5× SDS sample buffer and heated at 60°C for 5 min. Samples were analyzed by SDS-PAGE and transferred to polyvinylidene difluoride membrane. The membranes were blocked with 1% skim milk in 0.1% Tween 20 in Tris-buffered saline (TBS) and incubated with primary antibodies. Immunoreactive bands were detected using horseradish peroxidase (HRP)-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories, West Grove, PA), Luminata Forte (Merck Millopore) and a chemiluminescence detector (LAS-3000, Fujifilm).

Salmonella enterica serovar Typhimurium SR-11 x3181 harboring the pBR-mCherry plasmid were grown overnight at 37°C, and then subcultured at 1:33 for 3 h in LB without antibiotics. The bacterial inocula were prepared by pelleting at 10,000 g for 2 min, and were then added to host cells at a multiplicity of infection (MOI) of 100–200 at 37°C with 5% CO₂.

siRNA duplex oligomers were designed as follows: 5′-UCGAAGUAUUCCGCGUACG-3′ (luciferase), 5′-GCACAUUGCUUUAUAAACA-3′ (TRAPPC8). A total of 20 nM siRNA was introduced to cells using RNAiMAX (Invitrogen) according to the manufacturer's instructions.

MEFs stably expressing GFP, Atg9A-WT–GFP or Atg9A-ΔYQRL+ΔLLV–GFP were lysed in buffer containing 20 mM CHAPS (Dojindo), 125 mM NaCl, 50 mM Tris-HCl pH 7.5 and complete protease inhibitor cocktail (Roche). Lysates were cleared by centrifugation at 20,400 g for 10 min. The supernatant was incubated with GFP-Trap_M beads (Chromotek, gtm-20) at 4°C for 3 h. The beads were washed three times with lysis buffer and eluted with 1× sample buffer (2% SDS, 100 mM DTT, 60 mM Tris-HCl pH 6.8, 10% glycerol, 0.001% Bromophenol Blue). After boiling for 5 min, the samples were subjected to western blotting. Quantification of band intensity was performed using the IMAGEJ software (National Institutes of Health, Bethesda, MD, USA).

The authors would like to thank T. Yamada for pEGFP-Atg9A vector; E. Morita for pDsRed-KDEL; T. Kitamura for PLAT-E cells and pMX-puro vector; and all of Yoshimori-Lab member for helpful discussion.