TRAPPIII is responsible for vesicular transport from early endosomes to Golgi, facilitating Atg9 cycling in autophagy

Autophagy is an evolutionarily conserved intracellular degradation system that contributes to a wide range of physiological phenomena (Mizushima and Komatsu, 2011). However, the formation of the autophagosome, a spherical membrane structure that sequesters the degradation substrates, remains poorly understood (Yoshimori and Noda, 2008). The autophagosome is surrounded by two membrane layers, suggesting that it forms by membrane dynamics that differ from the mechanisms generally responsible for vesicle budding. Formation of the autophagosome is mediated by the Atg proteins, which include two ubiquitin-like proteins (Nakatogawa et al., 2009). In addition to the Atg proteins, numerous other proteins are involved, either directly or indirectly, in autophagy. However, it remains unclear how other membrane-trafficking pathways are related to autophagy (Longatti and Tooze, 2009). Several yeast mutations that cause defects in intracellular vesicular trafficking also affect autophagy. For example, mutation of Tlg2, a t-SNARE localized in endosomes and the Golgi, disrupts autophagy, but does not abolish it (Abeliovich et al., 1999). Similarly, early and late sec mutants, as well as mutants in the conserved oligomeric Golgi (COG) complex, exhibit defective autophagy (Geng et al., 2010; Hamasaki et al., 2003; Ishihara et al., 2001; Reggiori et al., 2004b; Yen et al., 2010).

Atg9 is a multiple-transmembrane protein that is essential for autophagosome formation (Noda et al., 2000; Young et al., 2006); it is recycled between pre-autophagosomal structure/phagophore assembly sites (PAS), which are involved in autophagosome formation, and other small structures called peripheral reservoirs (Reggiori et al., 2004a). Mutation of Tlg2 or COG complexes inhibits this recycling, suggesting that these factors are involved in Atg9 dynamics (Ohashi and Munro, 2010; Yen et al., 2010). In mammalian cells, Atg9 is transiently associated with some endosomes (Kageyama et al., 2011; Orsi et al., 2012). A recent study characterized Atg9-containing vesicles with respect to physical characteristics such as their size and the number of Atg9 molecules contained in each vesicle (Yamamoto et al., 2012); however, the identity and significance of this vesicular-transport pathway have not been clearly elucidated.

In this study, we screened a mutant collection containing knockdowns of all essential genes in Saccharomyces cerevisiae, by measuring the autophagic competency of each mutant. Among the factors that we identified as contributing significantly to autophagy, we focused on TRAPPIII, which has been reported to be specifically involved in autophagy at the PAS, where autophagosome formation takes place (Lynch-Day et al., 2010). In contrast to the previous report, our results demonstrated that TRAPPIII plays a role in general vesicular trafficking at the Golgi, rather than in an autophagy-specific process. These findings explain how general vesicular traffic contributes to autophagy.

In the yeast Saccharomyces cerevisiae, an alkaline phosphatase assay is widely used to quantitatively assess autophagy (Noda and Klionsky, 2008; Noda et al., 1995). To obtain a comprehensive understanding of the mechanisms underlying regulation of autophagy, we extended the alkaline phosphatase assay system by adapting the procedure to 96-well microtiter plates, enabling large-scale quantitative estimation of autophagic competency and the effects of every non-essential gene mutation in yeast (S.K. et al., unpublished results). Here, we applied this system to the essential gene knockdown collection, including about 900 mutants, in which the mRNA for each gene is destabilized by insertion of a marker gene into the 3′-untranslated region (DAmP method) (Schuldiner et al., 2005). From this collection, we identified a group of genes, including TRS20 and BET5, whose knockdown resulted in markedly low autophagic activity (Fig. 1A,B). Trs20 and Bet5 are core subunits of TRAPP complexes, which function at tethering step in several vesicular-transport pathways (Barrowman et al., 2010). The three TRAPP complexes – TRAPPI, TRAPPII and TRAPPIII – have been reported to function in endoplasmic reticulum (ER)-to-Golgi transport, intra-Golgi transport and autophagy, respectively (Lynch-Day et al., 2010). Each TRAPP complex serves as a guanine-nucleotide exchange factor for the Rab GTPase Ypt1 (Lynch-Day et al., 2010), which was also highly ranked as a candidate gene in our screen (Fig. 1A). The only subunit that is specific to TRAPPIII, Trs85, was also highly ranked in our screen of mutations of nonessential genes (S.K. et al., unpublished results). Therefore, based on the results of our unbiased genome-wide screen, we focused on TRAPPIII in our subsequent investigations.

Atg9 is recycled between PAS and as-yet-unidentified peripheral reservoirs, previously characterized as multiple punctate compartments distributed throughout the cytosol (He and Klionsky, 2007; Noda et al., 2000). Although Atg9 recycling has previously been reported to be independent of TRAPPIII (Lynch-Day et al., 2010), we re-examined this issue using a more elaborate experimental system that exploits ubiquitylation-mediated degradation of the IAA protein induced by addition of auxin (NAA) to the medium (Nishimura et al., 2009). In the presence of auxin, Trs85 tagged with IAA is not detected because it is ubiquitylated and degraded, whereas in the absence of auxin, TRS85-IAA is stably expressed from its own promoter and is not degraded (Fig. 2A). Defect in API processing observed in the absence of TRS85 due to degradation was recovered after removal of auxin (Fig. 2B). We observed mCherry-Atg8 dot formation in these conditions, and found that it is severely defected in the presence of auxin, but recovered after removal of auxin (Fig. 2C). Consistent with the results of a previous paper (Lynch-Day et al., 2010), puncta of Atg9-3×GFP were dispersed through the cytoplasm even in the absence of Trs85, a pattern that was indistinguishable from that observed in the presence of Trs85 (Fig. 2D, left top and bottom panels). When we knocked out ATG1, a protein kinase required for movement of Atg9 from the PAS to peripheral pools, Atg9-3×GFP accumulated at the PAS, which was observed as a bright punctate signal (Reggiori et al., 2004a) (Fig. 2D, right bottom panel). By contrast, in the absence of Trs85 in atg1 mutants, Atg9-3×GFP was dispersed throughout the cells (Fig. 2D, right top panel). Thus, Atg9 trafficking to the PAS, and resultant PAS formation, is dependent on TRAPPIII. Recently, TRAPPIII was reported to be associated with the vesicle where Atg9 resides (Kakuta et al., 2012). Consistent with this, subsets of Trs85-3×mCherry-positive puncta and Atg9-3×GFP-positive puncta were also double-positive (12.2% and 5.23%, respectively, in over 300 cells) (supplementary material Fig. S1A). Furthermore, subsets of GFP-Ypt1 positive puncta and a part of Atg9-3×mCherry were also double positive (10.8% and 2.73%, respectively, in over 400 cells) (supplementary material Fig. S1B). Therefore, at least some Atg9-positive puncta correlate with the vesicles in which TRAPPIII and/or Ypt1 resides.

In contrast to the situation in nutrient-rich conditions, TRAPPIII (trs85) and atg1 double-mutant cells did not exhibit defective Atg9 transport to the PAS under starvation conditions (Fig. 3A,B). On the basis of this observation, we hypothesized that under starvation conditions, another pathway bypasses the TRAPPIII-dependent pathway for transport to the PAS. The Cvt pathway is a selective autophagic process in which API is transported to the vacuole and processed to its mature form (Scott et al., 1996). TRAPPIII (trs85) null mutants did not exhibit defects in the Cvt pathway under starvation conditions, whereas they were severely defective under nutrient-rich conditions, consistent with the existence of a starvation-specific bypass pathway (Meiling-Wesse et al., 2005; Nazarko et al., 2005).

In an effort to identify the bypass pathway, we first examined Retromer, a protein complex that functions in retrograde transport from the late endosomes to the Golgi (Seaman, 2005). To date, Retromer has not been considered to be crucial for autophagy (Kametaka et al., 1998). Atg9 trafficking is affected in several double mutants in endosomal trafficking pathways, although the precise role of these pathways is still ambiguous (Ohashi and Munro, 2010). Although mutation in an essential subunit of Retromer, Vps17, did not affect the distribution of Atg9-3×GFP under nutrient-rich or starvation conditions (not shown), a combination of vps17 and a TRAPPIII mutation (trs85) increased the number of Atg9 puncta, even under starvation conditions (Fig. 3A,B). This observation suggests that a Retromer-dependent process bypasses the TRAPPIII-dependent pathway when cells are starved.

On the basis of the possibility that the TRAPPIII-derived defect is bypassed by a Retromer-dependent pathway, we hypothesized that TRAPPIII has some unknown connection to endosome-to-Golgi trafficking. Therefore, we asked whether other retrograde endosome-to-Golgi transport processes play bypass roles similar to that of the Retromer-dependent pathway. Golgi-associated retrograde protein (GARP) is a complex that functions as a tethering factor in endosome-to-Golgi transport, and Vps51 is its essential subunit (Bonifacino and Hierro, 2010). Similar to the case of Retromer, the vps51 trs85 double mutant exhibited a severe defect in Atg9 trafficking under starvation conditions in the atg1 background (Fig. 3A,B). In this vps51 trs85 atg1 triple mutant, we further observed that Atg9 failed to reach the PAS: Atg9-3×GFP was scarcely associated with API puncta, which represent the PAS, in sharp contrast to the significant overlap between Atg9-3×GFP and API puncta in the atg1 single mutant (Fig. 4). This increase was not due to elevation of the amount of Atg9-3×GFP protein because it was not changed by these mutations (supplementary material Fig. S2).

Consistent with these results, we observed a marked reduction in API processing after 1 hour of starvation in the vps17 trs85 double mutant, relative to each single mutant (Fig. 5A–C). Likewise, the vps51 trs85 double mutant exhibited a severe defect after 3 hours of starvation (Fig. 5A–C), although mutation of the GARP subunit Vps51 was previously shown to only slightly disrupt API processing (Reggiori et al., 2003). By contrast, combination of the GARP mutation (vps51) with a Retromer mutation (vps17) did not result in a marked defect even under starvation conditions (Fig. 5B,C). These observations are consistent with a model in which Retromer and GARP function in the same bypass pathway from the late endosome to the Golgi.

To further understand the relationship between TRAPPIII and the other endosomal pathways, we next examined the sorting nexin complex, which functions in a retrograde pathway that is distinct from the Retromer-mediated process (Hettema et al., 2003). Mutation of the sorting nexin subunit Atg24 (also called Snx4) disrupts API processing under nutrient-rich conditions (Nice et al., 2002). When the sorting nexin mutation atg24 was introduced into the TRAPPIII (trs85) mutant, the phenotype was similar to that of each single mutant, unlike the combinations of trs85 with mutations in GARP and Retromer subunits (Fig. 5A–C). By contrast, combination of atg24 with a mutation in a GARP (vps51) or Retromer subunit (vps17) exacerbated the API processing defects relative to each single mutant (Fig. 5B,C). This result indicates that TRAPPIII functions in the same pathway as sorting nexin, and leads to the novel idea that TRAPPIII is involved in endosome-to-Golgi retrograde trafficking.

On the basis of this finding, we asked whether TRAPPIII function might not be specific to autophagy, but rather associated with general vesicular traffic. To answer this question, we assessed the localization of a general cargo, GFP-tagged Snc1, a v-SNARE protein involved in the fusion of secretory vesicles to the plasma membrane. Snc1 recycles between the plasma membrane, endosomes, and the Golgi complex by a combination of endocytosis, retrograde transport and the secretory pathway (Lewis et al., 2000). Compared with wild-type cells, the TRAPPIII (trs85) mutant exhibited altered distribution of GFP-Snc1 (Fig. 6A,B). Furthermore, when the TRAPPIII (trs85) mutation was combined with a GARP (vps51) mutation, marked changes in the localization of GFP-Snc1 were observed (Fig. 6A,B): no GFP signals were detected at the plasma membrane, and most signals were dispersed throughout the cytoplasm, implying that the GFP-Snc1 protein was packaged in small transport vesicles (Fig. 6A,B). Ric1 is a guanine-nucleotide exchange factor for Ypt6 that functions in a GARP-dependent pathway (Siniossoglou et al., 2000). Consistent with this, the phenotype of the ric1 trs85 double mutant was quite similar to that of the vps51 trs85 mutant (supplementary material Fig. S3A). We also noticed that GARP (vps51) and TRAPPIII (trs85) double mutant, and ric1 and trs85 double mutant, exhibited severe growth defects, growing only at lower temperatures and at a slower overall rate (Fig. 6C and not shown). Furthermore, combination of the TRAPPIII (trs85) and Retromer (vps17) mutations resulted in a phenotype that was milder overall than that of vps51 trs85, but still additive, whereas the TRAPPIII (trs85) and sorting nexin (atg24) double mutant did not exhibit an additive growth defect (Fig. 6C). Next, we investigated another cargo of the endosome-to-Golgi pathway, Vps10, which is a receptor for the vacuolar protease carboxypeptidase Y. Vps10 mostly localized to the Golgi and endosome, and exhibited a typical patchy pattern (supplementary material Fig. S3B) (Shi et al., 2011). However, the pattern of Vps10 was more dispersed in the trs85 vps51 double mutant, appearing as numerous small particles, and was also dispersed to some extent in each single mutant (supplementary material Fig. S3B). Taken together, these results indicate that TRAPPIII is involved in general retrograde transport from the early endosomes to the Golgi.

Because other TRAPP family members, TRAPPI and TRAPPII, act in tethering step at the destinations of their respective pathways (Barrowman et al., 2010), we predicted that TRAPPIII would also function at its destination, the Golgi. To investigate this idea, we determined whether Trs85-GFP expressed at endogenous levels would colocalize with the Golgi marker Sec7-mStrawberry in wild-type cells. In cells under starvation conditions, 80.9% of Trs85-GFP signals were also positive for Sec7-mStrawberry (Fig. 7A), whereas only 0.58% of Trs85-GFP signals overlapped with API-mStrawberry, which represents the PAS in wild-type cells (Fig. 7B); more than 200 cells were counted for each quantification of colocalization. Together, these results support a role for TRAPPIII in tethering to the Golgi retrograde vesicles that originate from endosomes.

Accordingly, we investigated the localization of Ypt1 in relation to the Golgi and Atg8 in wild-type cells (supplementary material Fig. S4A,B). In a count of more than 60 cells, 77.7% of GFP-Ypt1-positive dots were also positive for Sec7-mStrawberry, whereas only 3.7% of GFP-Ypt1-positive dots were positive for 2×mCherry-Atg8. Furthermore, a recent report showed that Atg11 also acts together with Ypt1 and TRAPPIII in autophagy (Lipatova et. al, 2012). However, in a count of more than 500 cells, only 0.95% of GFP-Atg11-positive puncta were also positive for Sec7-mStrawberry (supplementary material Fig. S4C). Together, these results suggest that association of TRAPPIII and Ypt1 to PAS is limited to a small population.

In this study, we revealed an important novel feature of membrane trafficking, namely, a role for TRAPPIII in endosome-to-Golgi retrograde transport. The identity of the tethering factor at this step has been a ‘missing piece’ in the study of membrane traffic. We also reinterpreted the observation that TRAPPIII-dependent Atg9 movement is crucial for autophagy, in terms of the nutrient response. On the basis of these findings, we propose a model (below) describing how general membrane traffic is involved in the regulation of autophagy.

Combinations of mutations in TRAPPIII and GARP subunits produced additive defects, whereas combinations of mutations in TRAPPIII and sorting nexin subunits did not result in more severe phenotypes. These findings indicate that TRAPPIII receives vesicles from early endosomes at the Golgi (Fig. 8). Our results showing that TRAPPIII functions in retrograde transport from endosomes to the Golgi are consistent with the previously reported roles of TRAPPI and TRAPPII in ER-to-Golgi and intra-Golgi trafficking, respectively. Thus, all TRAPP family members can now be considered to function at the Golgi. Furthermore, our results suggest that Ypt1 functions at the Golgi together with all members of the TRAPP family.

These findings provide important insights into the dynamics of Atg9 (Fig. 8). Under nutrient-rich conditions, early endosome-to-Golgi transport is crucial for the Cvt pathway. Because COG-dependent Golgi function and exit from the Golgi are important for autophagy (van der Vaart et al., 2010; Yen et al., 2010), shuttling between the Golgi and early endosomes also appears to be crucial. The essential cargo molecule for autophagy is Atg9, although additional transmembrane cargo proteins, such as Atg27, could also participate in this process. Atg9-GFP-positive puncta in peripheral reservoirs exhibit marked movement (Sekito et al., 2009), and only a small minority of these structures (8–12%) colocalize with Golgi and endosomal markers (Mari et al., 2010). The other labeled structures were probably a mixture of vesicles providing anterograde and retrograde transport between the Golgi and endosomes. During autophagosome formation, some Atg9 will depart from these pathways and arrive at the PAS. In TRAPPIII mutant cells, Atg9 was trapped in the vesicle and did not reach the PAS. As long as autophagy proceeds normally, Atg9 is recycled back to the peripheral pool via the PAS and the autophagosome and/or vacuoles, but in mutant cells (e.g. atg1), it accumulates abnormally in the PAS.

Another finding of this study is that the TRAPPIII-dependent pathway is bypassed under starvation conditions (Fig. 8). Previously, the existence of this alternative pathway might have obscured the role of TRAPPIII in this process. Here, we propose a model in which TRAPPIII and GARP function as two independent complexes at the Golgi. Although the vps51 mutant exhibited defective API processing under nutrient-rich conditions, this defect was not as pronounced as that of the trs85 mutant, and autophagic activity was not affected under starvation conditions. Therefore, the GARP-dependent pathway must represent an alternative, rather than primary, pathway in autophagy. However, Snc1 recycling mainly uses the GARP-dependent pathway, and in this case the TRAPPIII-dependent pathway appears to serve as the backup. In general, starvation alters vesicular trafficking, and the early endosome-to-late endosome pathway is enhanced by starvation, possibly due to the need for increased degradation in the vacuole (Hamasaki et al., 2005; Jones et al., 2012). As a result, some Atg9 in early endosomes will be relocated to late endosomes, and the late endosome-to-Golgi pathway becomes another reservoir of Atg9. Our model also explains why Tlg2, a t-SNARE localized on the Golgi and early endosomes, is required only for the Cvt pathway (Abeliovich et al., 1999). When mutations affecting vesicles at their origin (Retromer, sorting nexin) and destination (TRAPPIII, GARP) were combined, the phenotypes were milder (see Fig. 5B,C; vps51 trs85 and atg24 vps51 double mutants) than that of the trs85 vps51 double mutant. This implies that the apparatuses at the origin and destination do not need to strictly match. Also of note, even in the context of the strongest phenotype (i.e. that of the trs85 vps51 mutant), some limited API processing still occurred. This might reflect the existence of another bypass pathway, such as the Fab1-dependent retrograde pathway (Efe et al., 2007). Alternatively, even without the reservoir sources, Atg9 expressed during starvation could support a small amount of autophagic activity.

Our model might appear to conflict with previous reports that TRAPPIII functions at the PAS in yeast and in forming autophagosome in mammalian cells (Lynch-Day et al., 2010; Kakuta et al., 2012; Lipatova et al., 2012; Huang et al., 2011; Wang et al., 2013). Our model does not necessarily exclude these possibilities, but our results suggest that the direct contribution of TRAPPIII to autophagosome formation is not as large as proposed in those studies. Future analysis of TRAPP-dependent Atg9 trafficking in mammalian cells will determine whether our model is universally applicable.

In preparing for sudden demand for autophagy, pooling the transmembrane protein Atg9 as a cargo of vesicular transport may be more favorable than statically storing the protein in a specific organelle, such as the Golgi. How Atg9 is transferred from these reservoirs to the PAS is still unclear. Our model and those proposed in other studies predict that Atg9 is delivered to the PAS via the Golgi through a canonical mechanism such as the secretory pathway (Geng et al., 2010; Nair et al., 2011). This important issue should be addressed in future studies.

Cells were incubated at 30°C unless otherwise indicated in YPD (1% Bacto yeast extract, 2% Bacto peptone and 2% dextrose) or SD (0.17% yeast nitrogen base without amino acids or ammonium sulfate, 0.5% ammonium sulfate, 2% dextrose, and appropriate amino acids) until an OD₆₀₀ value of 0.8–1.2 was reached. For starvation experiments, cells were washed in deionized water, transferred to SD-N (0.17% yeast nitrogen base without amino acids or ammonium sulfate, 2% dextrose), and incubated at the indicated temperature. For auxin-based degradation of Trs85, cells were grown overnight in YPD containing 200 µM 1-naphthaleneacetic acid (NAA; Sigma N0640) and washed twice with fresh YPD to remove the NAA. The cells were then resuspended in YPD at an OD₆₀₀ value of 0.5, and incubated at 30°C for the indicated times.

Yeast strains used in this study are listed in Table 1. PCR was used to create and confirm all gene disruptions, tagged constructs and other modifications (Goldstein et al., 1999; Janke et al., 2004; Longtine et al., 1998; Nakatogawa et al., 2012). The PHO8 locus in the knockout collection and the DAmP haploid collection (Open Bio System) (Schuldiner et al., 2005) was replaced with pho8Δ60 using a synthetic genetic array method (Tong and Boone, 2007). Briefly, the TNY509 parental strain harboring pho8Δ60 (Noda and Klionsky, 2008; Tong and Boone, 2007) was crossed with each mutant strain on YPD in a rectangular plate (OmniTray, Nunc) with a 96-pinned replicator (VP408, V&P Scientific). Haploid cells containing the pho8Δ60 allele and each mutant allele were selected using auxotrophic and drug-resistance markers and sequential replica plating (Tong and Boone, 2007). The mStrawberry sequence was PCR-amplified from pmStrawberry (Shu et al., 2006) and used to replace the yeGFP sequence of pYM25 to generate pKN14; the resulting plasmid was used for C-terminal tagging with mStrawberry. A partially digested PvuII fragment of pGFP-Snc1 (306) (Lewis et al., 2000) was subcloned into pRS305 digested with PvuII to generate pKN24. The resulting plasmid was linearized with EcoRV and integrated at the chromosomal LEU2 locus. A BamHI-SacI fragment encoding 3×GFP from the pAtg9-3×GFP (306) plasmid (Geng et al., 2010) was subcloned into pRS303 digested with the same enzymes to generate pKN15 (pRS303-3×GFP). Sequence encoding the C-terminal region of ATG9 was PCR amplified using primers 5′-CGGGATCCCCTCTTCCGACGTCAGAC-3′ and 5′-CCGCTCGAGGCACCATTTCTGGTCACATAC-3′. The amplified fragment was digested with BamHI and XhoI, and then subcloned into pKN15 digested with the same enzymes to generate pKN17 (pRS303-Atg9-3×GFP). pKN17 was linearized with BglII and integrated at the ATG9 locus. pNHK53 and pMK43 (National BioResource Project, Japan) harbored OsTIR1 and AtIAA17, respectively (Nishimura et al., 2009). pNHK53 was cut with StuI and integrated at the URA3 locus of a wild-type strain. AtIAA17 in pMK43 was PCR-amplified, cut with HindIII and XhoI, and subcloned into pYM24 digested with HindIII and SalI to add 3×HA to the C terminus of IAA17. The resultant plasmid, pKN19, was used as the template for C-AID tagging of TRS85. pRS314-2×mCherry-Atg8 was a gift from the Ohsumi lab. pKN38 to express GFP-Ypt1 under the ADH1 promoter was constructed as follows. The YPT1 gene was amplified from the wild-type genomic DNA by PCR with the primer pair of 5′-GGG GAT CCA ATA TGA ATA GCG AGT ACG-3′GG GA5′GG GAT CCA ATA TGA ATA GCG AGT ACG-3G, and digested with BamHI and XhoI. The resulting fragment was introduced into pBP73-A digested with the same enzymes to generate pKN37, which has a GFP-YPT1 fusion gene under the ADH1 promoter. The plasmid was then digested with SacI and XhoI and the fragment containing GFP-YPT1 with the ADH-promoter was subcloned into pRS304 digested with the same enzymes. The resulting plasmid pKN38 was linearized at the unique EcoRI site within the YPT1 gene and integrated at the YPT1 locus of a wild-type strain harboring chromosomally tagged Sec7-mStrawberry.

YPD plates were inoculated with cells from each pool using a 96-pin replicator and incubated for 16–24 hours at 30°C. Subsequently, the cells were suspended in 200 µl of SD-N medium in 96-well plates and incubated for 4 hours at 30°C. The plates were centrifuged, and the supernatant was discarded. To each well was added 50 µl of ice-cold lysis buffer (10 mM Tris-HCl at pH 9.0, 10 mM MgSO₄ and 10 µM ZnSO₄,) and ∼10 µl of 0.6 mm zirconia/silica beads (Biomedical Science). The plates were sealed with parafilm and mixed vigorously on a rotary shaker at 2500 rpm for 10 minutes at 4°C. After a brief centrifugation, 150 µl of ice-cold lysis buffer was added, and the plates were centrifuged for 15 minutes at 490 g at 4°C. Protein levels were determined from 50 µl aliquots of the supernatants using the bicinchoninic acid method, and enzymatic activity was measured in 50 µl aliquots as described in a previous report (Noda and Klionsky, 2008) with slight modifications.

Cells were observed Leica AF6500 fluorescent imaging system mounted on a DIM6000 B microscope (HCX PL APO 63× /1.40–0.60 oil-immersion objective lens, mercury lamp) under the control of LAS-AF software (Leica Microsystems) or on a Yokogawa CSU-X spinning-disk confocal system (Yokogawa Electric Corp., Japan) mounted on an Olympus IX71 microscope (100× NA 1.4 PlanApo objective lens) equipped with an Andor iXon CCD camera (Andor Technology, UK) under the control of the Andor IQ software. Images were processed using Adobe Photoshop. Atg9 puncta were counted using the G-Count software (G-Angstrom, Japan). Leica TCS SP8 confocal system mounted on a DMI 6000 CS microscope (HCX PL APO 100× /1.4 Oil STED objective lens, Leica Microsystems) was used to observe the cells in supplementary material Fig. S2.

Yeast cells were harvested, suspended in 100 µl of 0.2 M NaOH and 1% (v/v) 2-mercaptoethanol, and incubated on ice for 10 minutes. One milliliter of chilled acetone was added to the suspension, and the samples were incubated for 10 minutes on ice. After centrifugation at 16,000 g for 10 minutes, cell pellets were washed once with chilled acetone and dissolved in Laemmli sample buffer using a water-bath sonicator and boiled for 5 minutes. Aliquots (0.2 OD₆₀₀ units) were resolved using SDS-PAGE, and proteins were detected with the appropriate antibodies [anti-aminopeptidase I (Noda et al., 2000), anti-HA (12CA5), anti-Atg9 (a gift from the Ohsumi lab)]. Band intensities were measured using ImageJ software. Lysates to detect Atg9 were prepared by glass bead lysis. Cells were suspended in lysis buffer (20 mM PIPES, pH 6.8, 200 mM sorbitol, 5 mM EDTA) containing Complete protease inhibitor cocktail (Roche) and 4 mM PMSF, and disrupted by six rounds of vortexing for 30 seconds with 30 second intervals on ice. Triton X-100 was added to the lysate at the final concentration of 0.5%. The lysates were incubated on ice for 5 minutes and centrifuged at 5000 g for 10 minutes at 4°C. The supernatants were incubated at 65°C for 10 minutes in Laemmli sample buffer before performing SDS-PAGE followed by immunoblotting with anti-Atg9 (a gift from the Ohsumi lab) and anti-PGK (A-6457, Molecular Probes).

The authors would like to thank Dr Daniel J. Klionsky (University of Michigan), Dr Scott D. Emr (Cornell University), Dr Charles Boone (University of Toronto), Dr Yoshinori Ohsumi, Dr Hayashi Yamamoto (Tokyo Institute of Technology), and Dr Roger Tsien (University of California, San Diego) for various strains and plasmids; the National BioResource Project Yeast (Japan) for plasmids; and Mr Yuuma Ito for technical assistance.