Atg21 Is Required for Effective Recruitment of Atg8 to the Preautophagosomal Structure during the Cvt Pathway*

Atg21 and Atg18 are homologue yeast proteins. Whereas Atg18 is essential for the Cvt pathway and autophagy, a lack of Atg21 only blocks the Cvt pathway. Our proteinase protection experiments now demonstrate that growing atg21Δ cells fail to form proaminopeptidase I-containing Cvt vesicles. Quantitative measurement of autophagy in starving atg21Δ cells showed only 35% of the wild-type rate. This suggests that Atg21 plays a nonessential role in improving the fidelity of autophagy. The intracellular localization of Atg21 is unique among the Atg proteins. In cells containing multiple vacuoles, Atg21-yellow fluorescent protein clearly localizes to the vertices of the vacuole junctions. Cells with a single vacuole show most of the protein at few perivacuolar punctae. This distribution pattern is reminiscent to the Vps class C(HOPS) (homotypic fusion and vacuolar protein sorting) protein complex. In growing cells, Atg21 is required for effective recruitment of Atg8 to the preautophagosomal structure. Consistently, the covalent linkage of Atg8 to the lipid phosphatidylethanolamine is significantly retarded. Lipidated Atg8 is supposed to act during the elongation of autophagosome precursors. However, despite the reduced autophagic rate and the retardation of Atg8 lipidation, electron microscopy of starved atg21Δ ypt7Δ double mutant cells demonstrates the formation of normally sized autophagosomes with an average diameter of 450 nm.

Autophagy starts at the so-called preautophagosomal structure (PAS), 1 where numerous autophagy proteins colocalize (17,18). Out of this structure, double membrane-layered autophagosomes filled with parts of the cytosol are formed (1,19,20). The outer membrane of the autophagosomes then fuses with the vacuole, and still membrane-bound autophagic bodies are released into the vacuolar lumen, where they are broken down.
In S. cerevisiae, proaminopeptidase I (pAPI), a resident vacuolar proteinase, has been shown to be selectively targeted to the vacuole via autophagy using the receptor-like Atg19 protein (21,22). Once in the vacuole pAPI is proteolytically matured. Under nutrient-rich conditions, when autophagy is not induced, pAPI uses the Cvt pathway as an alternate route to the vacuole. The Cvt pathway and autophagy are morphologically very similar, but the Cvt transport intermediates, the double-layered Cvt vesicles, are smaller in size compared with autophagosomes and do not include cytosolic material. Autophagy and the Cvt pathway both start at the PAS and share most of the components needed for the biogenesis of the transport vesicles. At the PAS, a set of commonly used proteins has been identified: (i) a protein complex containing the phosphatidylinositol 3-kinase Vps34; (ii) the integral membrane protein Atg9; (iii) a covalently linked protein conjugate of Atg5 and Atg12, generated by the action of a ubiquitin-like protein conjugation system; (iv) Atg8 covalently linked to phosphatidylethanolamine (Atg8-PE), which is generated by a second ubiquitin-like conjugation system; and (v) a protein complex containing Atg13 and the serine/threonine protein kinase Atg1, which is involved in switching between the Cvt pathway and autophagy.
Besides these shared components, an increasing number of proteins such as Vac8 (23), the VFT (Vps fifty-three) complex (24), Atg20 (25), and Atg27 (26) have been identified, which specifically affect only the Cvt pathway. Since all of the specific components, besides the pAPI-receptor Atg19, affect biogenesis of the Cvt vesicles, the main mechanistic differences between the Cvt pathway and autophagy are expected during biogenesis of their transport vesicles. Getting more insight into the differences between the Cvt pathway and autophagy should therefore also improve our understanding of the sophisticated mechanisms used during the formation of these unconventional double membrane-layered vesicles.
Previously, we identified the yeast Atg21 as a component essential for maturation of pAPI under conditions where the Cvt pathway is active. Starvation induction of autophagy bypassed the defect in pAPI maturation in atg21⌬ cells, pointing to a specific function of Atg21 in the Cvt pathway (27). S. cerevisiae cells contain two further proteins homologous to Atg21. One of these proteins, Atg18, has been shown to be essential for formation of both autophagosomes and Cvt vesicles (28,29). The third Atg21 yeast homologue, Ygr223, has not been tested so far for its implication in either autophagy or the Cvt pathway. Atg21 is ϳ20% identical with Atg18, but whereas Atg18 affects both autophagy and the Cvt pathway, Atg21 is only essential for the Cvt pathway. This makes the two proteins ideal objects for studying the mechanistic features of both pathways.
Extending our initial studies on the function of Atg21, we here report localization of Atg21-YFP to some perivacuolar punctae. When multiple vacuoles are present in a cell, Atg21-YFP is recruited to the vertices of the vacuole junctions. Atg21 shares this localization pattern with components of the Vps class C(HOPS) (homotypic fusion and vacuolar protein sorting) complex. Our findings show that in contrast to ypt7⌬ cells, proaminopeptidase I is not membrane-enclosed in Cvt vesicles in growing atg21⌬ ypt7⌬ cells. Furthermore, a lack of Atg21 prevents normal recruitment of Atg8 to the preautophagosomal structure and consistently reduced its lipidation significantly.

Strains and Plasmids
For a list of strains, see Table I. YPT7 deletion strains were constructed using the plasmid pBSKSϩ ypt7::HIS3 (D. Gallwitz), which was digested with XhoI and PacI and transformed into the respective cells. Transformants were selected on plates lacking histidine and chromosomal replacement of the YPT7 gene confirmed by Southern blotting (not shown).
Plasmids cen-ATG21-HA, 2-ATG21-HA, and 2-ATG18-HA-The ATG21-HA plasmid with its native promoter was constructed using a PCR fragment created from the chromosomally hemagglutinin (HA)-tagged strain, YHB5 (27). The PCR fragment was generated using the primers pPL-HA(HindIII)w (GCGGGCaagcttAATATGGTTCAATTTATGGG) and MUTdown (TCCAGTTTAAACGAGCTCGAATTCCTATTAGC) and the chromosomal DNA from YHB5. cen-ATG21-HA and 2-ATG21-HA resulted from ligating the 1.9-kb PCR fragment into the HindIII and SmaI sites of the plasmids pRS316 and pRS426, respectively. cen-ATG18-HA was created similarly. A PCR product was generated using the chromosomal DNA from YHB1 (28) and the primers pFR-HA(HinDIII)w (GCGGGCaagcttTAACACATAGCGAGCTATGA) and MUTdown from above. The 2.4-kb fragment was cut and inserted into the HindIII and SmaI sites of pRS315. 2-ATG18-HA was generated by ligating a SalI/NotI fragment from cen-ATG18-HA into pRS425.
Atg21-YFP-The enhanced YFP fluorescent protein was excised from pEYFP (BD Clontech) with XmaI and EcoRI and inserted into the pRS316 vector (New England Biolabs) at the same sites. This vector (pkmw5) was cut with KspI and AgeI, and a PCR fragment, containing ATG21 with its native promoter and the added KspI and AgeI incision sites, was inserted, yielding the plasmid pkmw16. The PCR fragment was constructed with the primer KSPI-MAI1 (AGGGccgcggGT-TCAATTTATGGGAAATATAC) and AGEI-MAI1 (agggAccggtgATGTA-AATTTATTATTTTTAGtcag) and the plasmid cen-ATG21-HA.
Ape1-CFP-The enhanced CFP fluorescent protein was excised from pECFP (BD Clontech) with XmaI and EcoRI and inserted into the pRS313 vector (New England Biolabs) at the same sites. This vector (pkmw2) was cut with KspI and XmaI, and a PCR fragment, containing APE1 with its native promoter and the added KspI and XmaI incision sites, was inserted, yielding the plasmid pkmw17. The PCR fragment was constructed with the primer KSPI-APEI (AGGGCCGCGGC-TACTTTAGGGTATAGGTTG) and XMAI-APEI (AGGGCCCGGGA-CAACTCGCCGAATTCATCG) and the plasmid pRN1 (30).

Antibodies
Atg8 antibodies were raised through the coupling of the peptide MKSTFKSEYPFEKC on the keyhole limpet hemocyanin and its immunization into a rabbit (Eurogentec, Seraing, Belgium). Polyclonal rabbit anti-green fluorescent protein (GFP), mouse anti-3-phosphoglycerate kinase, mouse anti-vacuolar ATPase (100-kDa subunit), mouse anti-Pep12, and mouse anti-Dpm1 antibodies were from Molecular Probes (Leiden, The Netherlands). Mouse antibodies against HA were from Babco (Richmond, CA). Rabbit anti-proaminpeptidase I antibodies were described in Ref. 27. Rabbit antibodies against Vps33 were a gift from W. Wickner. Vac8 antiserum was from D. S. Goldfarb. The secondary antibody horseradish peroxidase-conjugated goat anti-mouse was from Dianova (Hamburg, Germany), and horseradish peroxidase-conjugated goat anti-rabbit was from Medac (Hamburg, Germany).

Cell Fractionation
The culture was grown in YPD to the late stationary phase. 80 A 600 units were harvested, washed once with water, and incubated at 30°C for 15 min in 20 mM dithiothreitol containing 1 M Tris-HCl buffer, pH 9.4, while shaking at 220 rpm. Cells were resuspended in 1 M sorbitol, 50 mM sodium phosphate buffer, pH 7.4, containing 50 g/ml oxalyti- case, spheroplasted at 30°C for 30 min and then hypotonically lysed in ice-cold PS200 buffer (200 mM sorbitol, 20 mM potassium-PIPES, pH 6.8, with 5 mM MgCl 2 ) containing 1 mg/ml leupeptin, 1 mg/ml chymostatin, 1 mg/ml antipain, 1 mg/ml pepstatin, 1 mM phenylmethylsulfonyl fluoride, and Complete® protease inhibitor mix (Roche Applied Science). Cell debris was removed by centrifugation at 1000 ϫ g, and the supernatant was transferred to a fresh tube three times. 300 l of supernatant was taken (total), and the proteins were precipitated on ice with trichloroacetic acid. 700 l was transferred to a fresh tube and centrifuged for 20 min at 4°C and 10,000 ϫ g. 300 l of supernatant was kept as S 13 , and 400 l of the supernatant was centrifuged for 1 h at 4°C and 100,000 ϫ g. The pellet fraction (P 13 ) was resuspended in Laemmli buffer with 1% ␤-mercaptoethanol. After the high speed centrifugation, 300 l of supernatant (S 100 ) was precipitated with trichloroacetic acid, and the pellet (P 100 ) was resuspended in Laemmli buffer with 1% ␤-mercaptoethanol. The trichloroacetic acid-precipitated proteins were centrifuged at 10,000 ϫ g for 10 min and washed twice with ice-cold acetone. The dried pellets were resuspended in Laemmli buffer with 1% ␤-mercaptoethanol.

Proteinase K Protection Experiment
Proteinase protection was done according to Ref. 21 with the following modifications. 40 A 600 units of stationary cells were harvested, washed twice with water, and incubated for 20 min in 20 mM dithiothreitol containing 0.1 M Tris/HCl buffer, pH 9.4. The cells were then resuspended in 1 M sorbitol, 50 mM sodium phosphate buffer, pH 7.4, containing 50 g/ml oxalyticase and spheroplasted at 30°C for 30 min. The spheroplasts were hypotonically lysed by resuspending in PS200 (20 mM potassium-PIPES, 200 mM sorbitol, pH 6.8, with 5 mM MgCl 2 ). The lysis solution was repeatedly precleared, and the supernatant was divided into three 300-l aliquots. 300 l of PS200, PS200 with 100 g/ml proteinase K, and PS200 with 100 g/ml proteinase K, and 0.4% Triton X-100 were given to each aliquot, respectively. After incubating for 15 min on ice, the digestion was halted through trichloroacetic acid precipitation. The pellets were dissolved in Laemmli buffer.

Optiprep Gradient
80 A 600 units of stationary cells were harvested, spheroplasted, and lysed as described above. After preclearing, the lysate was divided into five fractions and centrifuged at 100,000 ϫ g in a TLA 100.2 rotor (Beckman) at 4°C for 1 h. One fraction was taken as a control for pellet and supernatant. The pellets from the remaining four fractions were combined in 500 l of PS200 buffer plus 5 mM MgCl 2 , and 100 l was removed and centrifuged a second time at 100,000 ϫ g. The pellet and supernatant fractions were retained as a second control. The remaining 400 l was layered upon an Optiprep gradient and centrifuged for 16 h at 4°C and 140,000 ϫ g in an SW41Ti rotor. Thirteen fractions were collected from top to bottom, trichloroacetic acid-precipitated, and taken up in Laemmli buffer. The four control aliquots and the 13 sample fractions were processed for immunoblots.

Glycerol Gradient
The glycerol gradient was performed according to Ref. 35 with the following modifications. 50 A 600 units of stationary cells were resuspended in 800 l of 0.1 M potassium dihydrogen phosphate, pH 7.0, containing 1 mM phenylmethylsulfonyl fluoride and Complete® protease inhibitor mix (Roche Applied Science). Glass beads were then added, and the cells were beaten for 30 min at 4°C. The mixture was centrifuged at 10,000 ϫ g for 15 min at 4°C, and the supernatant was applied on the top of the glycerol step gradient, consisting of 450 l from 20, 30, 40, and 50% glycerol in 20 mM PIPES, pH 6.8. The gradient was centrifuged at 20,000 ϫ g for 4 h at 15°C (TLS-55 rotor) in a Beckman ultracentrifuge TL-100. 10 fractions of 200 l were collected and precipitated with trichloroacetic acid. The pellets were dissolved in 50 l of Laemmli buffer and processed for immunoblots.

Quantitative Measurement of Autophagy
The pho8⌬::LEU2 deletion strains were generated using the pho8::LEU2 deletion plasmid pGF10 (36) and then transformed with the Pho8⌬60 expression plasmid pCC5 (37). Enzymatic activity was measured as described (38) with the following modifications. Logarithmically grown cells were washed with water and resuspended in SD(ϪN) medium. One A 600 unit of cells was harvested at each time point and washed once with water. The cells were then suspended in 0.2 ml of assay buffer (250 mM Tris/HCl, pH 9.0, 10 mM MgSO 4 , 10 M ZnSO 4 ) and disrupted by vortexing with glass beads. After centrifugation, 50 l of the supernatant were added to 0.5 ml of assay buffer and 50 l of 55 mM potassium naphthylphosphate. After incubation for 15 min at 30°C, 0.5 ml of 2 M glycine/NaOH (pH 11.0) was added to stop the reaction. Fluorescence intensity was measured with excitation at 345 nm and emission at 472 nm. Protein concentration was determined with the BCA method (Pierce).

Microscopic Procedures
Electron microscopy after permanganate fixation and epon embedding was done as described (39). A Zeiss EM 900 transmission electron microscope was used. The mean area of autophagosome profiles was estimated by point counting from photographs taken at ϫ 12,000 magnification.
Fluorescence and light microscopy was done using a Zeiss Axioscope2 microscope with an Axiocam digital camera.

The Homologous Proteins Atg18 and Atg21 Play Distinct
Roles in Autophagy and the Cvt Pathway-Atg21 has two yeast homologues, Atg18 (ϳ20% identity) and the protein of so far unknown function, Ygr223c (ϳ14% identity). We here analyzed Ygr223c for defects in autophagy or the Cvt pathway. As shown in Fig. 1A, lanes 3 and 4, complete maturation of pAPI in starved and nonstarved ygr223c⌬ cells does not suggest defects in the autophagic or Cvt pathway. Also, starved ygr223c⌬ atg21⌬ double mutant cells did not show an additive defect in the autophagic pathway (Fig. 1A, lane 1).
We next generated 2-m plasmids expressing HA-tagged versions of Atg18 and Atg21 under control of their respective endogenous promotors and confirmed their expression and biological activity in rescuing pAPI maturation (Fig. 1B, lanes  2-5). With ϳ20% identity, Atg18 and Atg21 show significant homologies over their entire length, but overexpression of Atg18-HA in atg21⌬ and of Atg21-HA in atg18⌬ cells did not rescue the defects in the Cvt pathway or autophagy (Fig. 1B). This suggests that both proteins play distinct, nonredundant roles.
For a quantitative measurement of the autophagic rate, Noda et al. (38) introduced a system based on expression of a truncated version of the alkaline phosphatase Pho8. This truncated enzymatically inactive Pho8⌬60 lacks the endoplasmic reticulum targeting sequence and therefore stays in the cytoplasm. After autophagic transport to the vacuole, proteolytic processing of Pho8⌬60 generates enzymatically active phosphatase, whose activity can be measured. After chromosomal deletion of PHO8, we expressed Pho8⌬60 in atg21⌬ cells and measured the generation of phosphatase activity during starvation in nitrogen-free medium. Compared with wild-type cells, the autophagic rate in atg21⌬ cells was reduced to 35% ( Fig.  2A). This indicates that autophagy proceeds in the absence of Atg21, but at a reduced rate. Consistently, maturation of pAPI is not complete in atg21⌬ cells after only a short time of nitro-gen limitation (Fig. 2B, lanes 4 -6).
Atg21 Is Required for Formation of Proaminopeptidase I-containing Cvt Vesicles-The Rab GTPase Ypt7p is essential for fusion of Cvt vesicles and autophagosomes with the vacuolar membrane (40). Cells lacking Ypt7p therefore accumulate the respective vesicular intermediates in their cytosol. Consistently, proaminopeptidase I accumulates in ypt7⌬ cells in a membrane-protected form, not accessible to exogenously added proteinase after cell lysis. This opens an easy way to determine at which step Atg21 functions during the Cvt pathway. If biogenesis of proaminopeptidase I-containing Cvt vesicles proceeds normally in atg21⌬ ypt7⌬ cells, pAPI should accumulate in a proteinase-protected form, whereas defects in formation of these vesicles would result in accumulation of pAPI in a proteinase-sensitive form.
We generated an atg21⌬ ypt7⌬ strain and converted the nonstarved cells to spheroplasts, lysed them hypotonically without affecting the integrity of Cvt vesicles, and probed accessibility of proaminopeptidase I to proteinase K in the presence and absence of the detergent Triton X-100. As expected, in lysed spheroplasts of nonstarved ypt7⌬ control cells, proaminopeptidase I was proteinase-protected, indicating its enclosure in Cvt vesicles (Fig. 3A, lane 5), and proteinase-accessible in the presence of Triton X-100 (Fig. 3A, lane 6). An aliquot of the lysed spheroplasts was further centrifuged at 10,000 ϫ g, yielding a P 10 pellet and a S 10 supernatant fraction. The lack of cytosolic 3-phosphoglycerate kinase (PGK) in the P 10 pellet fraction (Fig. 3A, lane 2) excludes the possibility that unlysed whole cells mimic this phenotype. In nonstarved atg21⌬ ypt7⌬ cells, proaminopeptidase I was proteinase-accessible even in the absence of detergent (Fig. 3A, lane 17). This demonstrates that in nonstarved atg21⌬ ypt7⌬ cells proaminopeptidase I is not enclosed in Cvt vesicles (i.e. Atg21 is required for formation of proaminopeptidase I-containing Cvt vesicles). Proteinase accessibility of proaminopeptidase I in nonstarved atg21⌬ cells (Fig. 3C, lane 17) further confirmed this finding. As controls, we included atg1⌬ and atg3⌬ ypt7⌬ cells, respectively. Atg1 and Atg3 have both been shown to be essential for formation of Cvt vesicles (41). In starved atg21⌬ ypt7⌬ cells, pAPI accumulated in proteinase-protected form (Fig. 3B, lane 17), consistent with the formation of proaminopeptidase I-containing autophagosomes in the absence of Atg21.
Prior to its uptake into Cvt vesicles, pAPI oligomerizes into dodecamers (19). We further analyzed crude extracts from stationary atg21⌬ cells in a glycerol density gradient to check this oligomerization. Wild type-like sedimentation of pAPI (Fig. 3D) suggested normal oligomerization of pAPI in atg21⌬ cells and argues against a role of Atg21 in this early step of the Cvt pathway. The appearance of some mature aminopeptidase I in this experiment is due to limited proteolysis during gradient centrifugation.
Atg21 Is Needed for Effective Recruitment of Atg8 to the PAS-The PAS is believed to act as a donor compartment in formation of autophagosomes and Cvt vesicles (19,20). In fluorescence microscopy, it appears as a perivacuolar dot at which multiple autophagy proteins such as Atg9, Atg19, and Atg8 colocalize (17,18). We next investigated whether the defect in formation of proaminopeptidase I-containing Cvt vesicles in atg21⌬ cells is reflected by defects in the organization of the PAS. We therefore expressed GFP-Atg9, GFP-Atg19, and GFP-Atg8 and determined in fluorescence microscopy the percentage of cells showing accumulation of the fusion proteins at the PAS, indicated by formation of a perivacuolar dot. Compared with wild-type cells, both logarithmically growing and starved atg21⌬ cells showed no significant difference in PAS localization of GFP-Atg9 and GFP-Atg19 (Fig. 4, A and B). GFP-Atg8,

FIG. 2. Atg21-deleted cells show a reduced rate of autophagy.
A, the rate of autophagy was measured quantitatively in atg21⌬ pho8⌬ cells expressing Pho8⌬60 from a plasmid. In wild-type cells, enzymatically inactive Pho8⌬60 is transported to the vacuole via autophagy, where it is proteolytically activated. Aliquots were taken, and the enzymatic activity was measured at the indicated times after shifting the cells to the starvation medium SD(ϪN). The enzymatic activity of the wild type was set to 100%. atg15⌬ cells are known to be defective in autophagy. wt, wild type. B, the processing of pAPI was observed after shifting the cells to the starvation medium, 1% potassium acetate. Aliquots were taken at the specified times and processed for immunoblotting with antibodies to proaminopeptidase I. wt, wild type. atg1⌬ cells are defective in both autophagy and the Cvt pathway. mAPI, mature aminopeptidase I. however, exhibited in logarithmically growing atg21⌬ cells a significant defect in its recruitment to the PAS and was dispersed throughout the cytosol (Fig. 4C).
Formation of Atg8-PE Is Reduced in atg21⌬ Cells-Atg8 is activated by a ubiquitin-like protein conjugation system involving Atg7, Atg3, and Atg4 and is then finally covalently coupled to the membrane lipid phosphatidylethanolamine (PE) (4). Compared with the free form of Atg8, the lipid-coupled

FIG. 3. In nonstarved (A) atg21⌬ ypt7⌬ cells, pAPI is proteinase-accessible, whereas in atg21⌬ ypt7⌬ cells starved for 4 h in SD(؊N) (B) pAPI is protected from proteinases.
The cells were spheroplasted, hypotonically lysed, and divided into five aliquots. In lanes 4, 10, and 16, the aliquot was treated with buffer (B); in lanes 5, 11, and 17, the aliquot was treated with proteinase K (K); and in lanes 6, 12, and 18, the aliquot was treated with proteinase K and Triton X-100 (K ϩ T). The fourth fraction was centrifuged at 10,000 ϫ g and separated into pellet (P 10 ) and supernatant (S 10 ) fractions. The samples were then separated by SDS-PAGE, electroblotted onto polyvinylidene difluoride membranes, and probed with antibodies against proaminopeptidase I (upper lane). Subsequent immunoblotting with antibodies against cytosolic 3-phosphoglycerate kinase (PGK) then confirmed the absence of contaminating intact cells. As controls, ypt7⌬ cells known to accumulate proteinase-protected pAPI and atg3⌬ ypt7⌬ cells, known to be defective in formation of Cvt vesicles and autophagosomes, are included. C, in nonstarved atg21⌬ cells, pAPI is proteinase-accessible. The cells were treated as described in A. atg1⌬ cells are defective in formation of Cvt vesicles. Aminopeptidase I is a resident vacuolar hydrolase. Consistently, its proform is not readily degraded by exogenously added proteinase but is converted into a pseudomature form (API*). D, in stationary atg21⌬ cells, proaminopeptidase I is dodecameric. Stationary wild-type and atg21⌬ cells were lysed in phosphate buffer with glass beads. After centrifugation at 10,000 ϫ g, the supernatant was applied to the top of a glycerol density gradient (20 -50%), and the gradient was centrifuged at 20,000 ϫ g for 4 h at 15°C (TLS-55 rotor) in a Beckman ultracentrifuge TL-100. The gradient was divided into 10 fractions from top to bottom and precipitated with trichloroacetic acid. After SDS-PAGE and electroblotting, the polyvinylidene difluoride membrane was subsequently probed with antibodies to pAPI, fatty acid synthase (FAS) (molecular mass of 2400 kDa), and 3-phosphoglycerate kinase (molecular mass of 45 kDa). mAPI, mature aminopeptidase I. Atg8-PE species shows a higher motility in urea-containing SDS-PAGE (42). We used this feature to analyze Atg8 lipidation in cells lacking Atg21. As shown in Fig. 5A, lanes 1-3, significant amounts of free Atg8 are present in atg21⌬ cells, whereas in wild-type cells (lanes 7-9), especially after starvation induction of autophagy, almost all Atg8 is converted into its lipidated Atg8-PE form. As a control, atg3⌬ cells were included; Atg3 is the E2-like enzyme required for lipidation of Atg8. The reduced ability of atg21-deficient cells to lipidate Atg8 was further confirmed in a cell fractionation experiment. Nonstarved cells were converted to spheroplasts and then lysed hypotonically. The total lysate (T) was then first separated in a low speed (13,000 ϫ g) pellet P 13 and supernatant S 13 fraction. Additional high speed centrifugation (100,000 ϫ g) of the 13,000 ϫ g supernatant then yielded a P 100 pellet and a S 100 supernatant. We used ypt7-deleted cells for this experiment, since a defect in vacuolar fusion of the Cvt vesicles leads to their accumulation. In nonstarved ypt7⌬ cells, Atg8-PE was found in both the low and high speed pellet fractions (Fig. 5B,  lanes 3 and 5). In nonstarved atg21⌬ ypt7⌬ cells, however, significantly reduced levels of Atg8-PE were detectable in the pellet fractions (Fig. 5B, lanes 8 and 10).
To further determine whether the reduced levels of Atg8-PE in atg21⌬ cells are due to insufficient amounts of Atg8, we checked starvation induction of Atg8. As shown in Fig. 5C, no significant differences in Atg8 levels were detectable compared with wild-type cells. One might further speculate that the Cvt targeting defect in atg21⌬ cells could be overcome by overexpressing Atg8. Expression of Atg8 from a 2-m plasmid, however, did not rescue the proaminopeptidase I maturation defect in nonstarved atg21⌬ cells (Fig. 5D).
Atg8 is thought to be involved in the expansion step of autophagosomes (43). We therefore speculated that autophagosomes in atg21⌬ cells might be smaller in size compared with wild-type cells, due to the lack of sufficient amounts of Atg8-PE. This would offer an easy explanation for the reduced autophagic rate seen in atg21⌬ cells. To check this hypothesis, we starved ypt7⌬ and atg21⌬ ypt7⌬ cells for 4 h in nitrogen-free SD(ϪN) medium and then prepared the cells for electron microscopy and estimated the mean area of the accumulating autophagosomes in electron microscopy (Fig. 5E). Because the autophagosomes are not perfectly round, determination of the area instead of diameter appeared more accurate. Surprisingly, autophagosomes formed in starved atg21⌬ ypt7⌬ cells with an average area of 0.1570 m 2 (S.D. ϭ 0.095; 28 autophagosomes) were equally sized as autophagosomes in ypt7⌬ cells (0.1611 m 2 ; S.D. ϭ 0.064; 29 autophagosomes). Assuming that autophagosomes are spherical, this corresponds to a diameter of 450 nm, which is well in agreement with published results on the size of yeast autophagosomes (44). These results do not appear to support a role for Atg21 in the expansion step during autophagosome formation.
Atg21 Localizes to Vertices of Vacuolar Junctions and to Perivacuolar Punctae-Indirect immunofluorescence microscopy visualized a chromosomally integrated Atg21-HA (HA tag fused to the carboxyl terminus of Atg21) expressed with its own promotor to diffuse dots, which appeared to cluster around the vacuole; additionally, some cytosolic staining was present (Fig.  6B) (27). To monitor Atg21 in living cells, we now generated a centromeric plasmid expressing a Atg21-YFP fusion protein with the endogenous ATG21 promotor. Formation of mature aminopeptidase I in nonstarved atg21⌬ cells suggested biological activity of the fusion protein (Fig. 6A). Compared with indirect immunofluorescence microscopy, direct visualization of Atg21-YFP gave significantly more insights into the localization, due to the lack of fixation and spheroplasting. Besides some cytosolic staining, Atg21-YFP was found at the vacuolar membrane and most prominent at some distinct punctae at the  (17,18), the PAS is not visible in each cell; here, typical pictures were selected to illustrate this. In fluorescence microscopy, vacuolar transport of the individual GFP fusion proteins can be monitored, since vacuolar proteolysis liberates a rather proteolysisresistant free GFP inside the vacuole. GFP-Atg9 does not reach the vacuole via autophagy; therefore, no GFP is detectable in these vacuoles (A). Atg19 and Atg8 both localize to the PAS and then are selectively targeted to the vacuole inside autophagosomes. Thus, vacuoles of wild-type cells expressing GFP-Atg19 (B) and GFP-Atg8 (C) contain GFP. The significantly reduced vacuolar level of GFP in starved atg21⌬ cells expressing GFP-Atg19 or GFP-Atg8 further confirms the reduced autophagic rate of these cells. Bar, 10 m. vacuolar membrane (Fig. 6C). Some of the perivacuolar punctae seemed to be in proximity to the nucleus. We therefore further checked localization of Atg21-YFP with respect to Nvj1-CFP (Fig. 6D). By interacting with Vac8, Nvj1 forms nuclear vacuolar junctions (33) and therefore localizes to these contact sites. Some of the Atg21 perivacuolar punctae were clearly not linked to the nuclear vacuolar junctions; others, however, appeared adjacent to the junctions (Fig. 6D). Evaluation of a larger number of cells revealed that only in half of the cells the Atg21-YFP punctae localized in proximity to the nuclear vacuolar junctions. Considering the existence of several Atg21-YFP punctae per cell, this might not be of statistical significance.
We noticed that in cells containing multiple vacuoles, Atg21-YFP seemed enriched at the contact sites between the individual vacuoles (Fig. 6E). To quantitate this finding, we counted how many cells with multiple vacuoles showed a Atg21-YFP localization at the vertices of vacuolar junctions and at their boundary membrane. Most interestingly, 69% of the cells exhibited a vertex localization, and a further 28% exhibited a localization at the boundary membrane (Fig. 6E). Only 3% of the cells showed no localization of the perivacuolar Atg21-YFP punctae at the vacuolar contact sites. This localization pattern of Atg21 is reminiscent of the Vps class C(HOPS) complex (45).
Does Atg21 Colocalize with the Preautophagosomal Structure?-The PAS is believed to be the origin of autophagosome formation (20). The PAS has been defined in fluorescence microscopy as a punctate, perivacuolar structure, where many of the autophagy proteins colocalize (17,18). We wanted to check whether Atg21 localizes to the PAS. We generated a pro-API-CFP fusion protein, a marker of the PAS, and after coexpression with Atg21-YFP checked for colocalization in fluorescence microscopy. As shown in Fig. 6F (arrows), most of the Atg21-YFP punctae did not colocalize with pro-API-CFP, demonstrating that Atg21 localizes to a compartment distinct from the PAS. Few of the Atg21-YFP punctae, however, colocalize with the PAS (arrowheads in Fig. 6F). For further work, it should therefore be taken into account that a subpopulation of Atg21 might exist at the PAS. Cells starved for 0, 2, and 4 h in SD(ϪN) were lysed, and crude extracts were separated on a 6 M urea SDS-PAGE. In lanes 1 and 2, significant amounts of free Atg8 are present in atg21⌬ cells. In the wildtype (lanes 7-9), Atg8 is almost completely lipidated with PE after induction of autophagy. As a control, atg3⌬ cells defective in lipidation of Atg8 are included. B, cell fractionation with nonstarved ypt7⌬ and atg21⌬ ypt7⌬ cells. Spheroplasted cells were hypotonically lysed; the total lysate was then separated in a 13,000 ϫ g pellet (P 13 ) and a supernatant fraction (S 13 ). The S 13 supernatant fraction was further centrifuged at 100,000 ϫ g, yielding a P 100 and an S 100 fraction. In ypt7⌬ Atg8-PE is within the P 13 and P 100 fractions (lanes 3 and 5), whereas in atg21⌬ ypt7⌬ cells, significantly reduced levels of Atg8-PE were found in the P 13 and P 100 fractions (lanes 8 and 10). C, starvation induction of Atg8. Cells were grown to log phase and shifted to SD(ϪN) starvation medium. Samples were taken after 0, 1, 2, 3, and 4 h. No significant difference in the Atg8 levels in atg21⌬ cells compared with wild-type cells is detectable. D, overexpression of Atg8 did not overcome the proaminopeptidase I maturation defect in atg21⌬ cells. E, although the lipidation of Atg8 in atg21⌬ is retarded, the autophagosomes created are of normal size as seen in atg21⌬ ypt7⌬ and ypt7⌬ cells. The cells were fixed with permanganate, embedded in epon, and then processed for electron microscopy. V, vacuole; N, nucleus; A, autophagosomes; bar, 0.6 m.
Atg21-HA Is Partly Associated to Membranes-Our previous work (27) demonstrated the peripheral association of Atg21 with so far unknown membranes. To learn more about the nature of these unknown membranes, we performed an Optiprep density gradient centrifugation. Atg21 is only weakly membrane-associated and is rapidly released to the soluble fraction. 2 After gentle lysis of spheroplasted cells expressing Atg21-HA with its own promotor from the chromosome, we separated the lysate in a pellet and supernatant fraction. After immunoblotting the samples, quantification using an ECL imager demonstrated that 68% of Atg21-HA was in the pellet fraction (Fig. 7A, lane P) and only 32% in the supernatant (Fig.  7A, lane S). Resuspension of the pellet fraction with the buffer appropriate for density gradient fractionation can also easily lead to release of Atg21 from membranes. We therefore further analyzed an aliquot of the sample layered on top of the density gradient in a separate centrifugation step (not shown) to confirm that Atg21-HA was still membrane-bound in this sample. In the Optiprep gradient, most of Atg21-HA sedimented in fractions 3 and 4 (Fig. 7B), with some protein appearing in the denser part of the gradient. A vacuolar membrane marker (100-kDa subunit of vacuolar ATPase) was found at the top of the gradient (Fig. 7B, fractions 1 and 2), showing a sedimentation pattern different from Atg21-HA. This finding seems to be in conflict with the visualization of some Atg21-YFP at the vacuolar membrane in fluorescence microscopy (Fig. 6, C-E). Most likely, this part of Atg21 is only very weakly attached to the vacuolar membrane and released even during our gentle cell lysis (Fig. 7A, lanes S and P). Atg21 further did not cosediment with the prevacuolar marker Pep12 (46) and the endoplasmic reticulum marker Dpm1. Also, the vacuolar membrane protein Vac8, which together with Nvj1 forms nucleus-vacuole junctions, did not cosediment with Atg21. The Sec1 homolog Vps33, a component of the Vps class C-HOPS complex sedimented to a position similar to Atg21-HA (not shown). Further detailed work is necessary to clarify whether Atg21 colocalizes with this complex or not.
Does the Localization of Atg21 Depend on the Presence of Other Atg Proteins?-The localization of Atg21-YFP to perivacuolar punctae and to the vertices of vacuole junctions distinguishes it from other known autophagy proteins. Nevertheless, we checked whether Atg21-YFP is mislocalized in mutant cells defective in autophagy. No obvious mislocalization was observed in growing cells lacking Atg1, Atg8, Atg4, Atg9, Atg15, Atg18, or Atg19 (not shown). DISCUSSION The Cvt pathway and autophagy are morphologically very similar and consistently share most of their components. The identification of a growing number of proteins acting in only one of these pathways, however, illustrates that the size of their double membrane-layered transport vesicles is not the only difference between these pathways. Besides the pro-API receptor-like protein Atg19 (21,22), all proteins specific for one pathway directly interfere with the biogenesis of the respective transport vesicles. This points to significant mechanistic differences between the biogenesis of Cvt vesicles and autophagosomes. We previously identified atg21⌬ cells for their defect in maturation of pro-API under nutrient-rich conditions, where the Cvt pathway is active (27). Starvation-induction of autophagy reverted the pro-API maturation defect. Atg21 is a very promising candidate for studies aiming to learn more about the mechanistic features and differences of autophagy and the Cvt pathway, since it shares homologies with two further yeast proteins, Atg18 and Ygr223c. As shown here, Ygr223c has no obvious function in the Cvt pathway or autophagy (Fig. 1A), whereas Atg18 has been shown to function in both biogenesis of autophagosomes and Cvt vesicles (28,29). Despite the sequence homologies of Atg18 and Atg21, overexpression of either protein can not substitute for the lack of the other (Fig.  1B), which suggests that Atg18 and Atg21 have distinct and nonredundant functions.
The localization of the peripherally membrane-associated Atg21 is unique among the known Atg proteins and further draws attention to this protein. In addition to a small cytosolic pool, we detected the Atg21-YFP fusion protein at the vacuolar membrane and at few distinct perivacuolar punctae. In cells having multiple vacuoles, Atg21-YFP clearly accumulated at the vacuolar junctions, most prominently at their vertices (Fig.  6E). Also, components of the Vps class C(HOPS) complex, such as Vps33, accumulate at the vertices of vacuolar contact sites (45). Further work will be needed to determine whether Atg21 interacts with components of this protein complex. The Vps class C(HOPS) complex is involved in Golgi-to-vacuole protein transport, homotypic vacuole fusion (45,47), and fusion of the autophagosomal outer membrane with the vacuole (48). Accordingly, mutants with defects in the Vps class C complex accumulate autophagosomes or Cvt vesicles in their cytoplasm; consequently, pAPI accumulates in a proteinase-protected form in these cells. As shown in Fig. 3, A and C, growing atg21⌬ and atg21⌬ ypt7⌬ cells, in contrast, accumulate proaminopeptidase I in a proteinase-sensitive form. This clearly demonstrates that in the absence of Atg21, pAPI is not trapped inside Cvt vesicles. The role of Atg21 during the Cvt pathway is in this respect different from the role of the Vps class C complex.
The proaminopeptidase I maturation defect in growing atg21⌬ cells can rapidly be overcome by starvation induction of autophagy (Fig. 2B). Our quantitative analysis of the autophagic rate ( Fig. 2A) also uncovered a function of Atg21 during autophagy. This function is not essential for autophagy to occur, but it improves the overall efficiency of autophagy. Consistent with our finding, the lack of other components, whose function is only essential for the Cvt pathway but not for autophagy such as Vac8 or the VFT complex, also leads to a reduction of the autophagic rate (23,24). These components seem to have a helper function, which improves the fidelity of the autophagic process.
Since the PAS is believed to be the origin of vesicle formation in the Cvt pathway and autophagy, we further checked the organization of this structure in the absence of Atg21. As indicated by the localization of GFP-Atg9 (Fig. 4A), atg21⌬ cells contain a PAS-like structure. However, most interestingly, growing atg21⌬ cells almost completely failed to recruit a fluorescent GFP-Atg8 fusion protein to the PAS, a defect that is further substantiated by a significant defect in lipidation of Atg8 (Fig. 5, A and B). The lipidation defect of Atg8 is not due to the lack of sufficient amounts of Atg8, since compared with wild-type cells, atg21⌬ cells did not show significantly reduced Atg8 levels, and also overexpression of Atg8 did not rescue the Cvt defect in these cells (Fig. 5, C and D). Lipidated Atg8 has been proposed to act during the membrane elongation step of autophagosome formation (43). However, the autophagosomes formed in the absence of Atg21 were normally sized (Fig. 5E). One possible explanation might be the existence of a control step during formation of autophagosomes, which senses the size and the amount of available lipidated Atg8. Thus, only normally sized autophagosomes are formed.
At the moment, it is not clear how Atg21 influences the lipidation of Atg8. Probably, a small amount of Atg21 is located directly at the PAS. Alternatively, consistent with its localization at the vacuolar membrane and structures near or at the vacuole, Atg21 might act in retrograde trafficking from the vacuole to the late endosome or the PAS. While this manuscript was under revision in agreement with the latter idea, a function of the Atg21 homologue Atg18 in retrograde vacuolar trafficking was presented (49).
Since Atg21 is peripherally membrane-associated, we expect the existence of interacting proteins involved in membrane attachment. To identify those partners, we checked several mutants defective in autophagy or the Cvt pathway for correct localization of Atg21-YFP. Unfortunately, this analysis was hindered by the very faint fluorescence signal observed in strains of the genetic background, which was used in generation of the commercially available yeast deletion strains. Nevertheless, mislocalization of Atg21-YFP to the cytosol seems to occur in cells lacking Vps34. 3 Vps34 is the yeast phosphatidyl-3 K. Meiling-Wesse, unpublished observations. inositol-3 kinase (50), and it is present in two distinct protein complexes (51). One of the Vps34 complexes acts in vacuolar protein sorting to the vacuole, and the other is located at the PAS and plays an essential role in biogenesis of Cvt vesicles and autophagosomes. Mislocalization of Atg21 in vps34⌬ cells might hypothetically suggest binding of Atg21 to phosphatidylinositol 3-phosphate. Due to the faint fluorescence intensity of Atg21-YFP in this background, this otherwise highly interesting finding needs further confirmation.
Atg21 was here detected in density gradient centrifugation (Fig. 7B) at a position similar to Atg18 (29). This prompted us to look for a direct interaction between Atg18 and Atg21. For this experiment, we generated a centromeric plasmid expressing Atg21-HA under control of its own promotor (not shown). We expressed this plasmid in cells with a chromosomally integrated Atg18-GFP fusion protein and performed immunoprecipitation with antibodies against HA or GFP and checked in Western blots for coimmunoprecipitation. Under the conditions used, we could not detect a coimmunoprecipitation of Atg21 with Atg18. 2