Vacuolar protein Tag1 and Atg1–Atg13 regulate autophagy termination during persistent starvation in S. cerevisiae

Macroautophagy (hereafter, autophagy) is a degradation process in which targets are enwrapped by autophagosomes and eventually degraded in the lysosome or vacuole (Mizushima and Komatsu, 2011). Autophagosome formation is executed by a group of Atg proteins, and the mechanism of autophagic progression have been studied intensively for several decades. Triggering autophagy by a stimulus such as nitrogen starvation leads to a series of events that includes phosphorylation of Atg proteins and their recruitment to sites of autophagosome formation (Reggiori and Klionsky, 2013). An important question is whether and how autophagy is terminated. Because autophagy is a self-degradation process, failure to properly terminate autophagy can ultimately lead to cell death (Kroemer and Levine, 2008). Although autophagic cell death is a recognized form of regulated cell death, most autophagy is related to pro-survival processes, suggesting the existence of mechanisms that terminate autophagy at appropriate points. In mammalian cells, autophagy terminates under persistent starvation (Yu et al., 2010). Specifically, ubiquitin-dependent degradation of Ulk1 and Vps34 (also known as PIK3C3) is responsible for termination of autophagy (Liu et al., 2016). In NRK cells, TORC1 is inactivated under serum starvation but eventually re-activated under further serum starvation, leading to termination of autophagy (Yu et al., 2010). Nevertheless, the termination mechanism remains to be elucidated.

In this study, we examined the mechanisms underlying termination of autophagy in the yeast Saccharomyces cerevisiae, and found that Atg1–Atg13 plays an important role in this phenomenon. Furthermore, we identified a novel vacuolar membrane protein, Tag1, that is responsible for this process.

We first followed the autophagic activity of yeast cells under nitrogen-deficient conditions. We measured autophagic activity by means of an alkaline phosphatase (ALP) assay in strains bearing genomic integrations of Pho8Δ60, a vacuolar alkaline phosphatase genetically engineered to be expressed in the cytosol (Araki et al., 2017). In these cells, ALP activity reflects the integrative amount of cytosol transported into the vacuole after onset of autophagy (Araki et al., 2017; Noda et al., 1995). Within 1 h of shifting to a nitrogen starvation medium from a nutrient-rich medium (YEPD), ALP activity increased steeply and was sustained for 2–4 h (Fig. 1A). From 4 to 8 h after the shift, the sustained increase in ALP activity started to decrease, and ultimately stopped around 10 h, whereas the expression level of the Pho8Δ60 protein was maintained until 24 h of starvation (Fig. 1A; Fig. S1A). These observations indicate that autophagy is terminated 8–10 h after a shift to nitrogen starvation conditions, implying the existence of some molecular mechanism responsible for this process.

We asked whether TORC1, the master regulator of autophagy, is involved in autophagy termination (Noda, 2017). Sch9 is phosphorylated at S737 by TORC1. Hence, we monitored TORC1 activity by western blotting using the anti-Sch9 phospho-S737 antibody (Urban et al., 2007). In wild-type cells, the phosphorylated form of Sch9 was observed in nutrient-rich conditions, but disappeared after 0.5–2 h of starvation (Fig. 1B). Sch9 was re-phosphorylated after 12–24 h of starvation (Fig. 1B). In the autophagy-defective Δatg2 mutant, however, re-phosphorylation of Sch9 was significantly reduced after 24 h of starvation (Fig. 1B). Treatment with rapamycin compromised the re-phosphorylation of Sch9 (Fig. 1B). Thus, in yeast, as in mammalian cells, TORC1 is re-activated under persistent starvation in a manner dependent upon autophagy (Yu et al., 2010).

If autophagy termination were dependent on TORC1 re-activation, inhibition of TORC1 activity would cause a defect in this process. However, when cells were cultured in nitrogen-deficient medium in the presence of rapamycin, autophagy termination was not affected (Fig. 1C). Reactivation of TORC1 has been proposed to be regulated by Gtr2, which in its GTP-bound form directly binds to TORC1 and inactivates it (Kira et al., 2014; Péli-Gulli et al., 2015). Conversion of Gtr2 from its GTP-bound form to the GDP-bound form is catalyzed by the GTPase-activating protein (GAP) Lst4–Lst7 heterodimer, the counterpart of mammalian folliculin–FNIP (Péli-Gulli et al., 2015; Petit et al., 2013; Tsun et al., 2013). However, autophagy terminated normally in Δlst7 cells, suggesting that TORC1 reactivation via Gtr2 does not play an important role in autophagy termination (Fig. S1B).

Next, we monitored Atg13, which is directly phosphorylated by TORC1 under nutrient-rich conditions and promptly dephosphorylated under nitrogen starvation to induce autophagy (Kamada et al., 2000). When nitrogen starvation is extended to 12 h, it is re-phosphorylated (Shin and Huh, 2011) (Fig. 1D). In addition, we monitored Atg13 phosphorylation status over a time course; re-phosphorylation of Atg13 started gradually after 8–10 h starvation (Fig. S1C), which correlates with the timing of autophagy termination (Fig. 1A). We asked whether this re-phosphorylation was mediated by TORC1 kinase. To address this issue, we first starved cells for 30 min, after which Atg13 was completely dephosphorylated. The cells were then cultured under starvation conditions with or without rapamycin for an additional 24 h. Atg13 was re-phosphorylated even in the presence of rapamycin (Fig. 1D). To inactivate TORC1 kinase activity in another way, we utilized Gtr1 and Pib2 mutants, which are upstream regulators of TORC1 (Kim and Cunningham, 2015; Michel et al., 2017; Tanigawa and Maeda, 2017; Ukai et al., 2018; Varlakhanova et al., 2017). To ablate both genes without killing the cells, as the double deletion mutant is lethal, GTR1 was deleted and Pib2 expression was conditionally suppressed through the addition of doxycycline using the TETO7^off promoter system with degron (Gnanasundram and Koš, 2015), which inactivates TORC1 and arrests growth (Ukai et al., 2018). Although Pib2 protein levels were reduced in the absence of doxycycline-induced suppression after 24 h of starvation, doxycycline treatment further decreased the level of Pib2 (Fig. S1D). In this strain, TORC1 reactivation was suppressed in the presence or absence of doxycycline, but Atg13 was still re-phosphorylated at 24 h (Fig. S1D). These data indicate that Atg13 re-phosphorylation is not mediated by TORC1 kinase and that autophagy termination is not dependent on TORC1 re-activation.

We next asked which kinase is responsible for re-phosphorylation of Atg13 under persistent starvation. Atg1 is a protein kinase that interacts with Atg13 (Kamada et al., 2000; Yamamoto et al., 2016). Atg13 re-phosphorylation did not occur in Δatg1 cells, reproducing results from a previous report (Shin and Huh, 2011) (Fig. 2A), raising the possibility that autophagy is required. However, normal Atg13 re-phosphorylation was observed in Δatg2 cells, which is also defective in autophagy (Fig. 2A). Furthermore, in Δatg1 Δatg2 double deletion mutants, Atg13 was not re-phosphorylated (Fig. 2A). In Δatg17 cells, Atg13 phosphorylation was defective, implying that the integrity of the Atg1 complex is important for re-phosphorylation (Fig. S2A). Deletion mutants of other core components of the Atg machinery, Δatg3, Δatg9, Δatg14 and Δatg18, did not exhibit a defect in Atg13 re-phosphorylation, although Δatg16 exhibited a mild defect (Fig. S2A). These data strongly suggest that the defect in Atg13 re-phosphorylation in Δatg1 under persistent starvation is not due to a defect in autophagy, but instead dependent on the lack of Atg1 itself. The kinase activity of Atg1 is important for Atg13 re-phosphorylation because expression of the ATG1 kinase-dead mutant did not complement the Atg13 re-phosphorylation defect (Fig. 2B). An in vitro kinase assay revealed that Atg13 was phosphorylated by the purified Atg1 protein, but not by the kinase-dead mutant protein (Fig. 2C). We also confirmed that Atg13 S379 and S428/S429, which are critical moieties for pre-autophagosomal structure, also known as phagophore assembly site (PAS), formation and induction of autophagy (Fujioka et al., 2014; Yamamoto et al., 2016), were phosphorylated by purified Atg1 in vitro (Fig. S2B). Furthermore, mass spectrometry analysis revealed that the abundance of Atg13 phospho-peptides was significantly lower in Δatg1 cells than in wild-type cells after starvation for 24 h (Fig. 2D), while there was only a modest effect in nutrient-rich conditions (Fig. S2C). Taken together, these data indicate that Atg1 kinase re-phosphorylates Atg13 under persistent starvation.

Next, we compared the phosphorylation time course of Atg1 substrates. Atg9 is phosphorylated by Atg1 kinase, and its phosphorylation is important for autophagic activity (Papinski et al., 2014). We utilized Phos-tag PAGE to monitor Atg9 phosphorylation, as previously reported (Yamamoto et al., 2016). After starvation for 2 h, Atg9 was already phosphorylated, but Atg13 re-phosphorylation was barely detectable (Fig. 2E). This data indicates that the timing of phosphorylation by Atg1 kinase differs among the substrates. To further analyze the difference in phosphorylation patterns among Atg1 substrates during starvation, we investigated PP2C phosphatases, which can dephosphorylate Atg13 (Fujioka et al., 2020; Memisoglu et al., 2019). Among seven PP2C genes, we constructed a quadruple deletion mutant (Δptc1-4) of PTC1, PTC2, PTC3 and PTC4, which encode proteins located in the cytosol rather than the mitochondria. In Δptc1-4 cells, Atg13-S379 was phosphorylated even after 2 h of starvation (Fig. 2F). In western blots with anti-Atg13 antibody, an upshifted band of ∼150 kDa appeared, but disappeared upon deletion of ATG13, suggesting that it represents a hyper-phosphorylated form of Atg13 (Fig. 2F, right panel). Additional deletion of ATG1 in Δptc1-4 eliminated phosphorylation of Atg13-S379 and the hyper-phosphorylated form of Atg13, indicating that Atg13 phosphorylation at the earlier time point of starvation was mediated by the Atg1 kinase. Consistent with the positive role of PP2C in these processes, under starvation, Ptc2 formed puncta (Fig. S2D) that were colocalized with PAS marker proteins Atg13 and Ape1 (Fig. S2E). By contrast, the phosphorylation status of Atg9, another Atg1 substrate, was barely affected in Δptc1-4 under starvation (Fig. S2F). Collectively, these findings indicate that PPC2 phosphatases specifically affect phosphorylation of Atg13, which regulates autophagy termination.

To search for other factors responsible for autophagy termination, we surveyed autophagic activity in two kinds of genome-wide yeast mutant collections – one in which each non-essential gene has been individually deleted, and another in which mRNA expression of each essential gene is decreased due to a disruption in the 3′UTR (Schuldiner et al., 2005). A total of 5761 mutant strains were cultured on YEPD agar plate medium, and then shifted for nitrogen starvation, and ALP activity was measured at 0, 4 and 24 h (Fig. 3A,B; Table S1). As we had reported in previous studies, almost all of the core Δatg mutants, which are fully defective in starvation-induced autophagy, were among the strains with the lowest autophagic activities after 4 h of starvation (Kira et al., 2014; Shirahama-Noda et al., 2013). Ksp1 is a protein kinase that negatively regulates autophagy through activation of TORC1, and Δksp1 cells also had high autophagic activity after 24 h of starvation (Table S1) (Umekawa and Klionsky, 2012). Notably, several vacuolar ATPase–deficient mutants (Δvma2, Δvma5, Δvma7, Δvma8, Δvma10, Δvma22, etc.) had higher autophagic activity at 0, 4 and 24 h (Table S1). We performed secondary screening using an ALP assay with more carefully cultured cells in YEPD liquid medium against top 50 highest ALP activities and starved for 24 h. Based on the results of the secondary screen, we decided to focus on one uncharacterized gene, YLR173W, whose mutant had 1.9-fold higher ALP activity than wild type after starvation for 24 h; it had the 8th highest activity overall in the primary screen, and the 7th highest in the secondary screen. We named YLR173W TAG1 (for Termination of AUTOPHAGY) based on the following results. In a time course analysis of Δtag1 cells, ALP activity increased until 8 h of starvation, just like in wild-type cells, but exhibited no decrease after 8 h (Fig. 3C). Thus, the higher ALP activity in Δtag1 cells reflects a defect in termination of autophagy, rather than higher basal activity or accelerated autophagy during earlier time points. To confirm these findings, we examined autophagic bodies, which are inner membrane structures of autophagosomes after delivery into the vacuole; these bodies disintegrate immediately after release into the vacuolar lumen, but are kept intact by treatment with the protease inhibitor PMSF (Takeshige et al., 1992). Using a conditional gene expression system for GFP–Atg8 (McIsaac et al., 2013), we observed autophagic bodies that accumulated after persistent starvation. After starvation for 12 h, estradiol was added to induce GFP–Atg8 expression, and PMSF was simultaneously added to block degradation of autophagic bodies in the vacuole. GFP–Atg8 was indeed induced after starvation for 15 h (Fig. S3A). In wild-type cells, quite a few autophagic bodies accumulated, indicating that autophagy had been terminated (Fig. S3B, Movie 1). However, a number of autophagic bodies clearly accumulated in Δtag1 cells (Fig. S3B, Movie 2). Therefore, we concluded that TAG1 is responsible for autophagy termination.

TAG1 encodes a 608-amino-acid protein with a single hydrophobic stretch appropriate for a transmembrane region [amino acids (aa) 54–75] (Fig. 4A). In a previous proteomics analysis of isolated yeast vacuoles, Tag1 was identified as a potential vacuolar-localized protein (Wiederhold et al., 2009). Consistent with this, N-terminally GFP-tagged Tag1 under the control of its own promoter was localized along the vacuolar membrane (Fig. 4B). Two facts confirm that Tag1 is a vacuolar type II membrane protein with the N-terminal region oriented toward the cytosol and the C-terminus inside the lumen. First, previous phospho-proteomic analysis revealed phosphorylation sites only in the N-terminal region (S5, T16, T24, S27) (Fig. 4A) (Albuquerque et al., 2008; Holt et al., 2009; Swaney et al., 2013). Second, there is a possible adaptor protein complex 3 (AP-3) recognition motif ([D/E]XXXL[L/V]) that could act as a vacuolar targeting signal in the N-terminal region (aa 15–20, ETQPLL) (Cowles et al., 1997). We replaced the di-leucine with di-alanine (aa 15–20, ETQPAA), resulting in a slight defect in vacuolar localization (Fig. 4C) (Darsow et al., 1998). In combination with deletion of Δvps4, the di-leucine mutant of GFP–Tag1 was not delivered to the vacuole, but instead accumulated in a perivacuolar structure stained with FM4-64, a feature of the class E compartment (Fig. 4C), whereas wild-type GFP–Tag1 was delivered normally even in Δvps4, implying that the di-leucine mutant bypassed the vacuole via the CPY pathway. These observations indicate that the ETQPLL motif is a functional AP-3 recognition motif on the cytoplasmic side.

N-terminally 3HA-tagged Tag1 expressed from its native promoter could be detected as a 116-kDa band on immunoblots (Fig. 5A). Endoglycosidase H treatment resulted in a mobility shift to 81 kDa, indicating that Tag1 is modified by N-linked sugar chains in the luminal domain (Fig. S4A). In addition to the 116 kDa band, we also detected bands at 21 kDa and 50 kDa (Fig. 5A). These bands were not observed in cells lacking Pep4, the vacuolar proteinase A, which processes various vacuolar enzymes to activate them (Fig. 5A) (Hecht et al., 2014). Therefore, these bands must be generated by cleavage, either directly by Pep4 or by other vacuolar proteases activated by Pep4. Based on the molecular masses, the cleavage is predicted to occur at the luminal domain, close to the transmembrane domain.

To narrow down the cleavage sites, several C-terminal deletion constructs at or around the predicted sites were generated, tagged with FLAG at N-terminus and with HA at C-terminus, and expressed in cells with or without the major vacuolar proteases Pep4 and Prb1 (Hecht et al., 2014) (Fig. S4B). Even the shortest C-terminal deletion mutant, truncated to 80 aa, yielded the cleaved form in the presence of Pep4 and Prb1, but not in absence of Pep4 and Prb1 (Fig. S4B). A construct that lacks the 23 amino acid residues following the transmembrane region (Δ77–90) was also cleaved (Fig. S4C). These observations suggest that the cleavage occurs at the luminal region adjacent to the transmembrane domain in a sequence-independent manner.

To determine the significance of Tag1 processing, we replaced the endogenous Tag1 with a variety of Tag1 luminal-domain deletion mutants (Fig. 5B). Although the full-length Tag1 construct fully complemented the autophagy termination defect in Δtag1 cells (Fig. S4D), all of the other luminal deletion mutants failed to complement Tag1 function (Fig. 5B,C), despite vacuolar localization of these mutants being maintained (Fig. 5D; Fig. S5G). These data imply that most luminal regions are required for its function. Nevertheless, the cleaved 21-kDa form was generated in cells expressing all of these deletion constructs (Fig. S4E), indicating that the cleaved 21-kDa form alone could not support Tag1 function.

To determine the importance of the other portion of Tag1, we replaced each domain of Tag1 with Pho8, an alkaline phosphatase and vacuolar membrane protein with the same orientation as Tag1 (Huang et al., 2015; Klionsky and Emr, 1989) (Fig. 5E). These constructs, tagged with GFP, were expressed in place of the native TAG1 gene. All constructs localized along the vacuolar membrane under nutrient-rich conditions (Fig. 5G). Wild-type GFP-Tag1 fully complemented Tag1 function, but replacement of the luminal domain with Pho8 in which the catalytic center was mutated, in order to avoid interference with the ALP assay (Kira et al., 2016), prevented complementation, as expected from the results described above (Fig. 5F). On the other hand, the variant in which the transmembrane domain was replaced with Pho8 could complement Tag1 function (Fig. 5F). Strikingly, complementation also occurred when the cytosolic domain was replaced with Pho8 (Fig. 5F). These data indicate that both the cytosolic and transmembrane domain are dispensable for Tag1 function, or at least interchangeable with an unrelated vacuolar membrane protein.

These results raised two possibilities, (1) that the luminal domain could function even as a cleaved soluble form in the vacuolar lumen, or (2) that the luminal domain must be connected to the transmembrane and cytosolic domains, but these are replaceable with an unrelated protein. To discriminate between these two possibilities, we artificially targeted the Tag1 luminal domain into vacuolar lumen. The vacuolar enzyme carboxypeptidase S is transported into the vacuolar lumen via the multivesicular body pathway, and its C-terminal pro-region is cleaved off by Pep4 (Katzmann et al., 2001; Odorizzi et al., 1998). We made a fusion protein of CPS and the GFP–Tag1 luminal domain (Fig. 5H). The fusion protein was transported into the vacuolar lumen (Fig. 5J), but did not complement the Δtag1 autophagic termination defect (Fig. 5I). Therefore, the Tag1 luminal domain connected to the transmembrane domain in the vacuolar membrane is required for full Tag1 function.

We next noticed that, under starvation, vacuolar membrane GFP–Tag1 signals became less intense, and mostly gathered in a single punctum in the vacuolar membrane, in a time-dependent manner (Fig. S5A–D). Foci formation was not observed for GFP–Pho8, implying that it is not general phenomena (Fig. S5E). These foci were not more abundant in the autophagy-defective mutant Δatg7 (Fig. S5A,B), indicating that autophagy plays a positive role in formation of Tag1 foci. By contrast, the foci do not overlap with the Atg13–mCherry signal, a marker of the PAS, where autophagosome formation takes place (Fig. S5F), and no physical interaction was detected between Tag1 and Atg1 (Fig. S5G). We observed no significant colocalization of GFP–Tag1 foci and various perivacuolar organelle markers (Fig. S5H).

Next, we sought to determine which domain of Tag1 is important for foci formation. To this end, we used truncation mutants and domain-swap mutants with Pho8 (Fig. 5D,G; Fig. S5G,I). Luminal domain truncation mutants did not form foci (Fig. 5D; Fig. S5I). The cytosolic and transmembrane domain-swapped mutants formed foci, whereas the luminal domain-swapped mutant failed to do so (Fig. 5G; Fig. S5J). These data support the idea that the luminal domain is important for foci formation.

We next found that autophagy termination was also defective when Tag1 was overexpressed from the strong GPD promoter (Fig. 6A), suggesting that the amount of Tag1 must be maintained within some proper range in order for the protein to fulfill its function. On the other hand, the level of TAG1 mRNA was elevated under nitrogen starvation (Fig. 6B). Nevertheless, the level of Tag1 protein, both the uncleaved 116-kDa form and the cleaved 21-kDa fragment, was not increased during starvation (Fig. 5A). When protein synthesis was inhibited by cycloheximide treatment, the Tag1 protein level decreased during starvation (Fig. 6C). These observations prompted us to hypothesize that the Tag1 protein level is maintained by balancing protein synthesis and degradation during starvation.

When GFP–Tag1 was expressed in Δpep4 cells, the GFP signal was detected in the vacuolar lumen during starvation (Fig. 6D). In mutants of the vacuolar lipase Atg15, which is responsible for disintegration of vacuolar luminal vesicles such as autophagosomes and multivesicular bodies (Epple et al., 2001), GFP–Tag1 also accumulated in the lumen (Fig. 6D). However, GFP–Pho8 also exhibited such accumulation in Δpep4 cells (Fig. S5C). The Tag1 lumen-swapped mutant, which was defective in Tag1 foci formation, also showed accumulation of GFP signals in the vacuole in Δpep4 cells (Fig. S5H), indicating that Tag1 foci formation is dispensable for vacuolar accumulation. This vacuolar luminal signal could not be detected in the double mutant of VPS4, which encodes a protein engaged in the endosomal sorting complexes required for transport (ESCRT)-dependent process, and PEP4 (Fig. 6D) (Oku et al., 2017; Zhu et al., 2017). Thus, a fraction of GFP–Tag1, and possibly also other vacuolar membrane proteins, such as Pho8, was incorporated into the vacuolar lumen via ESCRT-dependent microautophagy under nitrogen starvation, as observed for the vacuolar membrane proteins Vph1 and Pho8 during the diauxic shift (Oku et al., 2017; Zhu et al., 2017). This mechanism would be expected to maintain the proper amount of Tag1 in the vacuolar membrane under nitrogen starvation.

Next, we examined the effect of TAG1 gene deletion on Atg proteins. We first monitored the levels of Atg proteins in wild-type yeast cells, and found that most Atg proteins became more abundant during persistent starvation (Fig. S6A). Therefore, the decrease in the Atg protein level does not seem to be the cause of autophagy termination in yeast. Of note, Atg1 expression level increased under starvation (Fig. S6B), despite a previous report showing that the Atg1 level was reduced due to autophagic degradation (Kraft et al., 2012). The increase in Atg1 protein levels during starvation was suppressed by cycloheximide treatment (Fig. S6B), indicating that Atg1 expression is upregulated under starvation. In combination with evidence that Atg1 is degraded by autophagy (Kraft et al., 2012; Nakatogawa et al., 2012b), this implies that the Atg1 protein synthesis level is higher than the rate of degradation by autophagy during starvation.

We also investigated whether Tag1 affects Atg1 kinase activity by using Phos-tag PAGE to monitor Atg1 autophosphorylation and phosphorylation of its substrates Atg9 and Atg29. We confirmed that phosphorylation of Atg9 and Atg29 were dependent on Atg1 (Mao et al., 2013; Papinski et al., 2014) (Fig. S6D; Fig. S6E, right panel). However, we detected no significant difference between wild-type and Δtag1 cells (Fig. S6C–E).

We then investigated whether TAG1 affects Atg13 re-phosphorylation, focusing on phosphorylation sites critical for regulation of autophagy (Fujioka et al., 2014; Yamamoto et al., 2016). Atg13 S379 and S428/S429 dephosphorylation is critical for the interaction between Atg13 and Atg17, as well as for autophagy (Fujioka et al., 2014; Yamamoto et al., 2016). This phosphorylation disappeared upon rapamycin treatment, suggesting that TORC1 is responsible for S379 phosphorylation, at least under nutrient-rich conditions (Yamamoto et al., 2016). Under such conditions, Atg13-S379 was also phosphorylated in Δtag1 cells, but this band disappeared after 30 min of starvation (Fig. 7A). In wild-type cells, Atg13-Se379 re-phosphorylation was observed after 24 h of starvation (Fig. 7A). However, re-phosphorylation of Atg13-S379 was significantly lower in Δtag1 than in the wild type (Fig. 7A), although overall Atg13 re-phosphorylation was not significantly altered in Δtag1 cells (Fig. 7A, upper panel). We also monitored the phosphorylation status of Atg13 S428/S429, another critical phosphorylation site for autophagy regulation (Fujioka et al., 2014). Re-phosphorylation of Atg13 S428/S429 was also significantly reduced in Δtag1 cells (Fig. S6F). Taken together, these observations indicate that Tag1 affects re-phosphorylation of the crucial sites S379 and S428/S429 under persistent starvation.

We next asked whether Atg13-S379 phosphorylation status is sufficient for control of autophagic termination. Cells expressing the dephosphorylation mimic mutant Atg13-S379A terminated autophagy normally, and no additive effect was observed in Δtag1 cells expressing Atg13-S379A (Fig. S6G). Furthermore, the general re-phosphorylation pattern of Atg13 was barely affected in cells expressing Atg13-S379A (Fig. S6H). These data indicate that dephosphorylation of the Atg13-S379 moiety itself is not sufficient for autophagic termination. We also tested the Atg13-S379A, S428A, S429A triple mutant; however, no significant effect was observed (Fig. S6H).

Next, we investigated the localization of Atg proteins in Δtag1 cells. ATG11 was deleted to discriminate Atg assembly at PAS dependent on starvation-induced autophagy from that involved in the CVT pathway (Kawamata et al., 2008; Shintani and Klionsky, 2004). In wild-type cells, perivacuolar puncta of Atg13–GFP and Atg2–GFP (representing the PAS) were induced after 2 h of starvation (Fig. 7B–D). However, the proportion of cells showing Atg13–GFP and Atg2–GFP puncta was significantly decreased after 24 h starvation (Fig. 7B–D). Thus, Atg proteins eventually disassemble from the PAS under persistent starvation, leading to termination of autophagy. In Δtag1 cells, Atg13 puncta were formed in 2 h, as in wild-type cells; however, a significant number of cells contained Atg13–GFP puncta even after 24 h of starvation (Fig. 7B,C). Similar results were obtained with Atg2 (Fig. 7B,D). Thus, Tag1 affects Atg disassembly from the PAS under persistent starvation. Collectively, these data imply that Tag1 affects phosphorylation of Atg13, leading to PAS disassembly and autophagy termination under persistent starvation.

In this study, we sought to determine how autophagy is terminated under nitrogen starvation in yeast. Yeast appear not to employ either of the two mechanisms for autophagy termination proposed for mammalian cells, that is, mTORC1 reactivation or degradation of ATG proteins by the proteasome (Antonioli et al., 2014; Liu et al., 2016; Nazio et al., 2016; Yu et al., 2010). Inactivation of TORC1 at the reactivation phase did not affect autophagy termination (Fig. 1C), and rather than being degraded, ATG proteins were upregulated during the termination period (Fig. S6A–C). Thus, yeast seem to use an unexplored mechanism for autophagy termination involving Tag1. No obvious homolog of Tag1 exists outside of fungi, suggesting that the termination mechanism differs between yeast and mammals.

Re-phosphorylation of Atg13 under persistent starvation was defective in cells lacking Atg1 kinase activity (Shin and Huh, 2011), and was not mediated by TORC1 but rather by Atg1 kinase (Figs 2A,B and 1D; Fig. S1C). This is reasonable, as Atg1 and Atg13 form a complex during starvation. Because PAS formation is suppressed by phosphorylation of critical moieties of Atg13 (Fujioka et al., 2014; Kamada et al., 2009; Yamamoto et al., 2016), autophagy termination could be mediated by re-phosphorylation of this protein. Indeed, re-phosphorylation of these critical moieties was defective in Δatg1 cells (Fig. 2A,D; Fig. S6F). Furthermore, these moieties of Atg13 were phosphorylated by Atg1 in vitro (Fig. S2B). Consistent with this, the PAS scaffold protein complex Atg13–Atg17–Atg29–Atg31 assembles as a liquid droplet, and phosphorylation of Atg13 by Atg1 disassembles this liquid droplet in vitro (Fujioka et al., 2020). We observed that the PAS disappeared under persistent starvation (Fig. 7B–D). We also showed that PP2C phosphatases play a negative role in this regulation of Atg13, but not in regulating the other Atg1 substrate, Atg9 (Fig. 2E; Fig. S2D,E). Therefore, dispersion of PAS mediated re-phosphorylation of Atg13 by Atg1 serves as a ‘safety timer’-like mechanism that prevents overshoot of autophagy.

Another crucial factor is the novel vacuolar protein Tag1. In the Δtag1 mutant, Atg13 re-phosphorylation is defective, leading to suppression of PAS dispersion and autophagy termination (Fig. 7). Thus, there must exist a mechanism that links Tag1 to the Atg1–Atg13 axis. How does Tag1 regulate Atg13 re-phosphorylation? We could not detect any difference in Atg1 kinase activity in Δtag1 cells, nor could we demonstrate any physical interaction between Atg1 and Tag1 (Figs S6C–E, S5E). Another possibility is that Tag1 affects PPC2 phosphatase activity toward Atg13. Atg13 is gradually re-phosphorylated during starvation, even in the presence of PP2C phosphatase (Fig. S1C), possibly because phosphatase activity is decreased. Given that Atg13 re-phosphorylation is affected in the Δtag1 mutant, one hypothesis is that Tag1 suppresses PPC2 phosphatase during starvation; this should be clarified in future work.

The Tag1 luminal domain was crucial, and needed to be connected to the membrane and cytosolic domains (Fig. 5). Therefore, the most plausible mechanism is that Tag1 luminal domain is affected by the outcome of autophagy inside the vacuolar lumen; for example, amino acids produced by protein degradation. This hypothesis is supported by the observation that Tag1 foci formation was defective in the Δatg7 mutant (Fig. S5A–C), indicating that Tag1 responds to the progression of autophagy at least in terms of its localization. This is further supported by the observation that the luminal domain is necessary for foci formation (Fig. 5D,G; Fig. S5G,H). The Tag1 protein amount is determined by the balance between the de novo synthesis of the Tag1 protein and microautophagic degradation (Fig. 6). Hence, inter-luminal information is transmitted to the Atg1–Atg13 main switch through the vacuolar membrane via a proper level of Tag1. One reasonable possibility is that Tag1 acts as a sensor of autophagic degradation products including amino acids. We tried to pursue the relationship between the amino acid pool and Tag1-mediated autophagy termination using a mutant of the Atg22 amino acid permease mutant; however, this approach revealed no relationship (data not shown). It would also be interesting to investigate whether the conformation of Tag1 is affected by amino acid concentrations in vitro.

Our study provides an avenue for exploring the mechanism by which autophagy is terminated, which in turn determines the ultimate destiny of a cell – survival or autophagic cell death. We have investigated whether the Δtag1 mutation affected survival rates under nitrogen starvation, but did not observe a significant phenotype (data not shown). We may still be missing some important piece, such as a back-up system, that would enable us to fully understand the significance of autophagy termination. Nevertheless, this and subsequent studies will facilitate the development of autophagy-related applications for treatment of cancer and neuronal diseases.

Yeast cells were grown in YPD [1% yeast extract (BD Biosciences, Franklin Lakes, NJ, USA), 2% peptone (BD Biosciences), 2% glucose (Wako, Osaka, Japan)] or SCD [0.17% yeast nitrogen base without amino acids and ammonium sulfate (BD Biosciences), 0.5% (NH₄)₂SO₄ (Nacalai Tesque, Kyoto, Japan], 0.5% casamino acids (BD Biosciences), 2% glucose]. For the starvation condition, the cells were cultured in SD-N buffered with MES (0.17% yeast nitrogen base without amino acids and ammonium sulfate (BD Biosciences), 50 mM MES-NaOH pH 6.2 (Wako), 2% glucose]. Cycloheximide (Wako), rapamycin (LC laboratories, Woburn, Massachusetts) and PMSF (Nacalai) were added to the media at final concentrations of 50 µM, 200 ng/ml and 1 mM, respectively.

Yeast strains, plasmids and primers used in this study are listed in Tables S2, S3 and S4, respectively. Gene deletion and tagging were performed using PCR-based methods as previously reported (Janke et al., 2004; Longtine et al., 1998; Nakatogawa et al., 2012a). To make the multiple PP2C deletion mutant, we used the CRISPR-Cas9 method as previously described (Laughery et al., 2015) with slight modifications. Cells were co-transformed with Cas9 and the gRNA expression plasmid, and the 80-mer single-stranded oligonuclotide DNA as a donor for homologous recombination. Plasmids were cloned using the NEBuilder kit (NEB, E2621) and Gibson assembly (Gibson, 2011). To obtain the linearized vector backbone, pRS series vector (Sikorski and Hieter, 1989) was amplified by inverse PCR. Each insert fragment was amplified by PCR with an 18–20-base tail sequence overlapping each end of the vector backbone. After gel extraction, insert and vector DNA fragments were mixed with 2× NEBuilder HiFi DNA Assembly Master Mix (NEB, E2621) or 2× GA premix [320 µl of 5× ISO buffer, 0.64 µl of 10 U/µl T5 exonuclease (NEB, M0363), 5 µl of 2 U/µl Phusion polymerase (NEB, M0530), 40 µl of 40 Taq DNA ligase (NEB, M208), and 435 µl dH₂O] and incubated for 1 h at 50°C. The mixture was subsequently transformed in E. coli. MNY069, which contains a genomic integration of GFP–Tag1, was generated as follows. The hygromycin resistance marker gene (hphNT1) cassette from pFA6a-NT1, proTag1, and GFP DNA fragments were amplified. These three DNA fragments were connected by PCR to generate hphNT1-proTag1-GFP, and this fragment was then integrated into the Tag1 N-terminus. The proTag1-GFP-Tag1-Tag1 terminator fragment was amplified from MNY069 genomic DNA using primers oSK1492 and oSK1127, and the pRS316 vector backbone was amplified by inverse PCR with oSK1063 and 1064; the resultant amplicons were connected using the NEBuilder kit (NEB, E2621) to generate pSK402.

Cells were collected by centrifugation (600 g, 2 min) and observed using a Leica AF6500 fluorescence imaging system mounted on a DIM6000 B microscope (HCX PL APO 100×/1.40-0.70 oil-immersion objective lens, xenon lamp; Leica Microsystems) under the control of the LAS-AF software (Leica Microsystems). Cells expressing Atg13–mCherry were observed using an Olympus SpinSR10 imaging system [UPLS APO 100× oil immersion objective lens, and spinning disk unit (CSU-W1, Yokokawa)]. For FM4-64 staining, 1–2 OD units of cells were collected, resuspended in 100 µl SCD containing 1.64 µM FM4-64 (Sigma-Aldrich), and cultured for 1 h at 30°C. After washing, cells were resuspended in 1 ml SCD for 30 min at 30°C.

Anti-Atg proteins used in this study were kind gifts from Dr Ohsumi Yoshinori (Tokyo Institute of Technology, Tokyo, Japan) and are described in the indicated publications: anti-Atg1 (1:2000; Matsuura et al., 1997); anti-Atg2 (1:5000; Shintani et al., 2001); anti-Atg3 (1:2500; Ichimura et al., 2000); anti-Atg4 (1:2000; Nakatogawa et al., 2012a); anti-Atg7 (1:500; Noda et al., 2011); anti-Atg8 (1:1000; Kirisako et al., 2000); anti-Atg9 (1:1000; Noda et al., 2000); anti-Atg12 (1:1000; Kuma et al., 2002); anti-Atg13 (1:200), anti-Atg13 S379 (1:5000), and S428/S429 (1:5000) (Fujioka et al., 2014; Yamamoto et al., 2016); and anti-Atg18 (1:2000), anti-Vps15 (1:1000) and Vps34 (1:2000) (Kihara et al., 2001). Anti-PGK (1:10,000; Life Technologies, 459250), anti-HA (1:1000; Roche, 3F10), anti-Protein A (1:5000; Sigma Aldrich, P3775) and anti-FLAG (1:1000; Sigma Aldrich, M2) antibodies were used for western blotting. Anti-Sch9 (1:3000) and anti–Sch9 phospho-T717 polyclonal antibodies (1:3000) were generated by immunizing rabbits with the synthesized phosphorylated peptide fragment of Sch9. Secondary antibodies for anti-mouse IgG (1:5000; Southern Biotech, 1031-05), anti-rabbit IgG (1:5000; Cell Signaling Technology, 7074) and anti-goat IgG (1:5000; Cell Signaling Technology, 7077) were used.

The yeast knockout library and the DAmP essential gene knockdown library harboring the Pho8Δ60 allele, constructed by the SGA method, were prepared as described previously (Kira et al., 2014). Cells spots were plated on YPD in rectangular OmniTrays (242811, Thermo Fisher Scientific, Waltham, MA, USA), cultured for 24 h at 30°C, cultured for 0, 4 or 24 h in SD-N with MES, and subjected directly to the ALP assay. Assays were performed as previously reported (Kira et al., 2014).

ALP assays for autophagy activity measurements were performed as previously reported (Noda and Klionsky, 2008).

Ten OD units of cells were suspended in 400 µl of TES buffer (10 mM Tris-HCl pH 7.5, 0.5% SDS and 10 mM EDTA) and lysed with zirconia beads (φ=0.6 mm, Biomedical Science) by vortexing vigorously for 10 min. RNA purification and cDNA synthesis were performed using the RNeasy mini kit (Qiagen) and Script cDNA Synthesis kit (Bio-Rad), respectively. mRNA level (normalized against ACT1) was measured by QuantiTect SYBR Green PCR (Qiagen) on a StepOne Plus Realtime PCR system (Life Technologies) using the ΔΔCt method.

For sample preparation, 1–2 OD units of cells were collected and resuspended in 200 µl of 0.3 M NaOH for 10 min at room temperature. Cells were resuspended in 100 µl of SDS-PAGE sample buffer and boiled for 5 min. For sample preparation for western blotting with antibodies against Atg9, Atg13, Atg13 S379, Sch9, Sch9 T717 and Protein A, 2–3 OD units of cells were transferred to a 1.5-ml screw cap tube to which trichloroacetic acid (Wako) was added at a final concentration of 6%, and incubated on ice for at least 20 min. Cells were washed with 700 µl of ice-cold acetone, dried, resuspended in 200 µl of urea buffer [50 mM Tris-HCl pH 7.5, 6 M urea, 5 mM EDTA, 1% SDS, 1 mM PMSF, 1× Complete (Roche), 1× PhosSTOP (Roche)] and 400 µl of zirconia silica beads (φ=0.6 mm, Biomedical Science, Tokyo), and lysed with FastPrep (MP-BioMedicals). 2-mercaptoethanol (2 μl; Wako) and a tip of Bromophenol Blue were added, and the samples were incubated for 5 min at 65°C. For EndoH treatment, cells were lysed in a bead beater; the crude lysates were treated with or without 500 U of EndoH (NEB, P0702) at 37°C for 3 h, and then subjected to western blot analysis.

Fifty OD units of cells expressing TAP or 4×Protein A-tagged proteins were collected in each tube, resuspended in 400 µl lysis buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 1 mM PMSF, 1× Complete and 1× PhosSTOP) with 400 µl of zirconia silica beads, and lysed with FastPrep. Triton X-100 was added to a final concentration of 1%, and the samples were rotated for 10 min at 4°C. After cell debris was eliminated by centrifugation (20,000 g, 10 min, 4°C), IgG Sepharose 6 Fast Flow (GE) or IgG Dynabeads were added, and the samples were rotated for 2 h at 4°C. Samples were washed three times with the lysis buffer and subjected to downstream analysis.

For radioisotope labeling, Atg13–TAP proteins were immunopurified from atg1Δ cells. After intensive washing steps of the IgG Dynabeads, bound proteins were used as substrates for in vitro phosphorylation experiments. Atg1–3×FLAG (wt) and Atg1^D211A1–3×FLAG (kd) were prepared from yeast cells grown in the presence of rapamycin. Harvested cells were resuspended in lysis buffer supplied with 1 mM PMSF, Complete EDTA-free protease inhibitor cocktail (Roche), 50 mM NaF, 2 mM Na₃VO₄, 10 mM Na₄P₂O₇, and 60 mM β-glycerophosphate. Atg1–3×FLAG (wt) and Atg1^D211A–3×FLAG (kd) were precipitated with anti-FLAG M2 affinity gel (Sigma-Aldrich). The beads were washed three times, followed by elution of the bound proteins by 3×FLAG peptide (Sigma, 1 mg/ml final concentration) in kinase buffer (50 mM Tris-HCl pH 7.5, 75 mM NaCl, 5% glycerol, 10 mM MgCl₂ and 1 mM DTT). Reactions were incubated at 30°C for 30 min in kinase buffer with 10 µM ATP and 10 µCi γ-[³²P]ATP. The reactions were terminated in HU buffer (50 mM Tris-HCl pH 7.5, 6 M urea, 5 mM EDTA and 1% SDS) by heating for 15 min at 65°C, and proteins were resolved by SDS-PAGE and analyzed on a phosphorimager (FLA-300; Fujifilm).

For detection of Atg13 phosphorylation by western blotting, Atg13–GFP proteins were immunopurified from atg1Δ cells using GFP Trap Magnetic agarose (Chromotek). The Atg13-bound beads were incubated with the Atg1–3×FLAG described in the previous section at 30°C for 1 h in kinase buffer containing 1 mM ATP. Reactions were terminated by addition of SDS-PAGE sample buffer and incubation for 10 min at 95°C, and the samples were subjected to western blot analysis.

Atg13-TAP proteins were immunopurified with the IgG sepharose 6 Fast Flow from WT or Δatg1 cells starved for 0 and 24 h. Bound proteins were eluted with Laemmli buffer, followed by SDS-PAGE. The proteins were excised from each gel, destained and digested in the gels with 12.5 ng/μl trypsin (Wako) in 50 mM ammonium bicarbonate overnight at 37°C. The phosphopeptides were enriched with Titansphere Phos-Tio kit (GL Sciences) and desalted with Empore Disk SDB-XC (GL Sciences). Nanoscale liquid chromatography coupled to tandem mass spectrometry (nanoLC-MS/MS) analysis was conducted using a Q Exacitve hybrid quadrupole-orbitrap mass spectrometer (Thermo Fisher Scientific), with Xcalibur software, and coupled to an EASY-nLC 1000 (Thermo Fisher Scientific). The data were processed, searched and LFQ-quantified using the Proteome Discoverer (version 2.4.0.305, Thermo Fisher Scientific), employing the S. cerevisiae UniProt database (version Feb. 21, 2016) containing 6749 entries. The search parameters were as follows: trypsin digestion with two missed cleavage permitted; variable modifications, protein N-terminal acetylation, oxidation of methionine, propionamidation of cysteine and phosphorylation of serine, threonine and tyrosine; peptide charge (2+, 3+ and 4+); peptide mass tolerance for MS data, ±10 ppm; and fragment mass tolerance, ±0.02 Da.

We are grateful to Dr Hayashi Yamamoto, Dr Sho Suzuki, and Dr Yoshinori Ohsumi for kind gifts of antibodies and plasmids, as well as valuable experimental advice. We also thank Dr Kuninori Suzuki for providing plasmids.