Formation Process of Autophagosome Is Traced with Apg8/Aut7p in Yeast

Yeast strains used in this study are shown in Table. To construct the Δapg8 strains, TRP1, LEU2, and HIS3, which was obtained from pJJ288, pJJ252, and pJJ215, respectively, was substituted for the AccI–HpaI fragment containing 98% of the APG8 open reading frame (ORF). TK402, TK404, TK405, TK407, and KVY5 were obtained by selection on appropriate amino acid drop-out medium. Media used in this study were described previously, and SD medium supplemented with 0.5% casamino acid was referred as SD+CA medium (Shirahama et al. 1997).

APG8 was cloned according to the method previously reported (Kametaka et al. 1996). pAPG8317 was generated by cloning 2.6 kbp of XbaI–XbaI fragment, including APG8, into pBluescript II KS⁺. Using pAPG8317 as a template DNA, the SpeI–EcoRI fragment, including the 0.9 kbp of HincII–HincII sequence, was generated by PCR using following primers: APG8F1, 5′-GACTAGTAGGTCTCGCAAGAGAGC-3′; APG8R1, 5′-GGAATTCGAAATCTTGGCTCCGTTG-3′. pTK101 and pTK201 were generated by inserting this SpeI–EcoRI fragment into the yeast centromeric and multicopy vectors, pRS316 and pRS426 (Sikorski and Hieter 1989), respectively. For construction of 3 × HA (hemagglutinin)-tagged APG8 plasmids, a BamHI site was generated at the 5′ terminus of the APG8 ORF by PCR using APG8F1, APG8R1, and the following primers: 5′-GACATGGGATCCAAGTCTACATTTAAGTCTG-3′ and 5′-AGACTTGGATCCCATGTCTCTAGTAATTAT-3′. The resulting fragment was cloned into pBluescript II KS⁺, and a BamHI–BamHI fragment containing a 3 × HA sequence generated by PCR was inserted in the BamHI site at the 5′ terminus of the APG8 ORF. The 3 × HA-tagged APG8 plasmids, pTK110 and pTK108, were generated by cloning this SpeI–EcoRI fragment containing the 3 × HA-tagged APG8 gene into yeast centromeric plasmids, pRS314 and pRS316 (Sikorski and Hieter 1989), respectively.

Alkaline phosphatase (ALP) assay was performed as previously described (Noda et al. 1995; Noda and Ohsumi 1998).

Cells growing in YEPD medium were treated with nocodazole (Sigma Chemical Co.) for 3 h at a final concentration of 10 μg/ml. Cells were transferred to 0.17% yeast nitrogen base without amino acids and ammonium sulfate supplemented with 1 mM PMSF and 10 μg/ml nocodazole, and further incubated for 4.5 h. The accumulation of autophagic bodies was observed under a microscope. For ALP assay, the nocodazole-treated cells in YEPD were starved in SD(−N) medium containing 10 μg/ml nocodazole for 4.5 h.

A peptide including the NH₂-terminal sequence of Apg8p, MKSTFKSEYPFEKC, was synthesized by a Model 433A peptide synthesizer (PE Applied Biosystems). The peptide was conjugated to Keyhole Limpet Hemocyanin (Sigma Chemical Co.) with sulfosuccinimidyl 4-(p-maleimidophenyl)butyrate (Pierce Chemical Co.) according to the method previously reported (Iwai et al. 1988). The resulting conjugates were immunized to a rabbit and anti-Apg8p antiserum was obtained.

Cell lysates were prepared by breaking cells with the glass beads in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, 5 mM EGTA, 1 mM PMSF, and the protease inhibitor cocktail™ (Boehringer Mannheim Corp.). Total protein (20 μg) was subjected to SDS-PAGE and transferred to PVDF (polyvinylidene fluoride) membrane (Millipore Corp.). The resulting membrane was incubated with a 1:10,000 dilution of the anti-Apg8p antibody for 1 h, followed by a 1:10,000 dilution of HRP-conjugated goat anti–rabbit IgG (Jackson ImmunoResearch Laboratories) for 30 min. Signals were detected by the ECL kit (Nycomed Amersham, Inc.), which was used throughout this study. For the analysis of rapamycin treatment, growing cells were treated with 0.2 μg/ml rapamycin (Sigma Chemical Co.) for 2 h at 30°C in YEPD medium. Lysate preparation, followed by Western blotting, were performed as described above.

Cells were suspended in TES solution (10 mM Tris-HCl, pH 7.5, 10 mM EDTA, 0.5% SDS) and treated with acid phenol at 65°C for 1 h. Aqueous phase was collected and RNA was recovered by ethanol precipitation. mRNA was recovered by Oligotex-dT30 < super > (Takara) according to the manufacturer's protocol, separated on 1.2% agarose containing 2.2 M formaldehyde by electrophoresis, and transferred to NYTRAN membrane (Scleicher & Schuell, Inc.). ³²P-labeled probes for APG8 and ACT1 mRNA were prepared from each ORF fragment using Megaprime DNA labeling system (Nycomed Amersham, Inc.) according to the appended protocol. The mRNA-transferred membrane was incubated with each probe at 42°C overnight, and signals were detected by autoradiography.

Subcellular fractionation was performed as previously described by Horazdovsky and Emr 1993. YW5-1B cells were cultured in YEPD medium at 30°C to 1–2 × 10⁷ cells/ml, shifted to SD(−N) medium for 3 h and converted to spheroplasts. Spheroplasting medium was composed of YEP, 1.2 M sorbitol, 20 mM Tris-HCl, pH 7.5, 1% glucose, and 25 μg/10⁸ cells of Zymolyase 100T (Seikagaku Kogyo) for growing cells, or 0.17% yeast nitrogen base without amino acids and ammonium sulfate, 1.2 M sorbitol, 20 mM Tris-HCl, pH 7.5, 1% glucose, and 37.5 μg/10⁸ cells of Zymolyase 100T for starved cells. Spheroplasts were harvested and lysed with a lysis buffer containing 0.2 M sorbitol, 50 mM Tris-HCl, pH 7.5, 1 mM EDTA and EGTA, 1 mM PMSF, and the protease inhibitor cocktail. Lysates (T) were generated by centrifugation at 500 g for 5 min and then spun at 13,000 g for 15 min to separate to pellet (LSP) and supernatant (LSS). LSS was then centrifuged at 100,000 g for 1 h, and pellet (HSP) and supernatant (HSS) were obtained. LSP and HSP were resuspended in equal volume of the lysis buffer to the original lysates. Equal volume of each sample was subjected to SDS-PAGE, transferred to PVDF membranes, and then incubated with a 1:5,000 dilution of the anti-Apg8p antibody.

For the solubilization experiment, lysates were prepared from YW5-1B cells growing or starved for 3 h as described above. Lysates were sonicated and spun at 100,000 g for 1 h to generate pellet. The pellet was suspended in the lysis buffer (without sorbitol) containing 2% Triton X-100, the suspension was chilled on ice for 30 min and centrifuged at 100,000 g for 1 h again to separate supernatant and pellet. Pellet was resuspended in an equal volume of the lysis buffer to the original sample. Proteins recovered in each fraction were precipitated with 10% TCA and resuspended in SDS sample buffer. Equal volume of the samples was subjected to Western blotting as described.

Immunofluorescent staining was performed according to the method previously described by Nishikawa et al. 1994. In this experiment, Δapg8 and Δapg8Δpep4 cells harboring 3 × HA-tagged APG8 plasmid were used. Spheroplasts prepared from the cells fixed with formaldehyde were laid on the multiwell slide glass (Cel-line Associates, Inc.) coated by polylysine (mol wt >300,000; Sigma Chemical Co.), and permeabilized by treatment with 0.5% Triton X-100 in PBS for 10 min. Permeabilized cells were incubated in PBS containing 1% BSA at room temperature for 10 min and treated with a 1:1,000 dilution of anti-HA mAb, 16B12 (Berkeley Antibody Co., Inc.), in the same buffer at room temperature for 1 h. The cells were then incubated with 10 μg/ml of anti-mouse Ig-fluorescein F(ab′)2 fragment (Boehringer Mannheim Corp.) at room temperature for 1 h and observed under a confocal microscope, LSM 510 (Zeiss).

Cells were subjected to rapid freezing and freeze-substitution fixation, and observed as previously reported (Baba et al. 1997). For immunoelectron microscopy, ultrathin sections were collected onto formvar-coated nickel grids and blocked in PBS containing 2% BSA at room temperature for 15 min. Incubations were carried out by floating grids on a 20 μl drop of a 1:1,500 dilution of anti-HA mAb, 16B12, at room temperature for 1.5 h. After washing, the grids were incubated for 1 h with 5- or 10-nm gold-conjugated goat anti-mouse IgG (Bio Cell Lab.). The grids were washed several times in PBS followed by several drops of distilled water and fixed with 1% glutaraldehyde for 3 min. The sections were stained with 4% uranyl acetate for 7 min and examined.

Cells were starved in SD(−N) medium for 4.5 h and converted to spheroplasts. Spheroplasts were lysed in 20 mM Pipes/KOH, pH 6.8, 0.2 M sorbitol, 0.5 mM PMSF, and the protease inhibitor cocktail. Cleared lysates were generated by centrifugation at 500 g for 5 min, and treated with 100 μg/ml of proteinase K with or without 2% Triton X-100 on ice for 30 min. Equal volume of 20% TCA was added to the lysates to stop the reaction. Precipitant was collected by centrifugation, suspended in SDS sample buffer, and subjected to immunoblotting. To detect the signal of API, 1:5,000 diluted anti-API antibody (gift from D. Klionsky, Section of Microbiology, UC Davis, CA) was used.

We reported an apg8-1 mutant defective in the autophagy in yeast (Tsukada and Ohsumi 1993). To identify the APG8, cloning was done as previously described (Kametaka et al. 1996). Subcloning and sequencing revealed that APG8 is identical to the ORF named YBL078c, and encodes a hydrophilic protein of 117 amino acids. During this study, Lang et al. 1998 reported AUT7 as a novel autophagy-responsible gene, which was revealed to be identical to APG8.

We constructed the apg8 null mutant and ascertained morphologically that Apg8p is essential for autophagy, judging from no accumulation of autophagic bodies in the vacuole (data not shown). Furthermore, using the ALP assay system (Noda et al. 1995), we confirmed biochemically that Apg8p is essential for autophagy. The cells used for this assay have a truncated form of Pho8p, Pho8Δ60p, in the cytosol as an inactive precursor form. During starvation, Pho8Δ60p is sequestered to the vacuole, depending upon autophagy, where it is processed by the vacuolar enzymes and acquires phosphatase activity. So, we can monitor the progress of the autophagy by the increase of the ALP activity. Wild-type (TN124), Δapg8 mutant, and the Δapg8 mutant harboring APG8 on a centromeric plasmid were grown in SD+CA medium until 1–2 × 10⁷ cells/ml, and then shifted to SD(−N) medium for three hours. The ALP activity in the cell lysates prepared from each culture was measured (Fig. 1). In wild-type cells, the ALP activity increased in response to starvation, whereas in the Δapg8 cells its elevation was considerably reduced. The autophagic defect in the Δapg8 mutant was recovered by introducing APG8 on a centromeric plasmid. The result clearly shows that deletion of APG8 causes severe defect in autophagy.

A homology search showed that Apg8p and its homologues make a large gene family through yeast to higher eukaryotes (Lang et al. 1998). One of the homologues has been reported as rat MAP1-LC3, which is directly bound to microtubule in vitro (Mann and Hammarback 1994). Furthermore, Lang et al. 1998 proposed that Apg8p is bound to microtubule via another protein, Aut2p, and functions on the delivery of autophagosomes to the vacuole. We have already cloned 13 APG genes, and realized that AUT2 is allelic to APG4, and that Apg8p physically interacts with Apg4p. However, Apg4p has an entirely different function from what they proposed (Kirisako, T., and Y. Ohsumi, manuscript in preparation).

So far, there is no evidence showing that microtubules play a role in the autophagy in yeast. Nocodazole is a microtubule depolymerizing drug, which affects both spindle and cytoplasmic microtubules (Solomon 1991). We asked, morphologically and biochemically, whether autophagy proceeds normally in the presence of this drug. Wild-type cells were cultured until 10⁷ cells/ml in YEPD medium at 30°C. Nocodazole was added to the culture and it was incubated at 30°C for three hours. After this treatment, >70% of the cells were arrested at G2 stage with a large bud. Then the cells were transferred to starvation medium containing 1 mM PMSF and 10 μg/ml nocodazole, and incubated further at 30°C for 4.5 h. As shown in Fig. 2 A, most large budded cells accumulated the autophagic bodies in their vacuole. TN124 cells were treated the same way, and the ALP activity was measured. Nocodazole treatment did not affect the increase of ALP activity (Fig. 2 B). Furthermore, in a tub2 mutant, autophagic bodies were normally accumulated in the vacuole (data not shown). These results indicate that depolymerization of microtubule does not affect the progress of autophagy. Hence, we concluded that microtubule is not necessary for the autophagy in yeast.

We raised an antiserum against Apg8p using a synthetic peptide of the NH₂-terminal 13 amino acids of Apg8p. Using the anti-Apg8p antibody, we carried out Western blotting of the cell lysate prepared from logarithmically growing cells. In wild-type cells, the anti-Apg8p antibody brought out a single band at ∼15 kD (Fig. 3 A, lane 1), which is close to the predicted molecular mass of Apg8p (13.6 kD). Intensity of the band was enhanced in the cells harboring APG8 on a multicopy plasmid (Fig. 3 A, lane 2), whereas no signal was detected in the Δapg8 cells (Fig. 3 A, lane 3). The result revealed that Apg8p is expressed in the wild-type cells at growing phase. Since APG8 is required also for the transport of proAPI from the cytosol to the vacuole (Scott et al. 1996), Apg8p expressed during vegetative growth should play a role in this pathway. Next, we examined Apg8p levels before and after shift to nitrogen starvation. The wild-type cells growing in YEPD medium were shifted to SD(−N) medium for various periods of time, and the lysates prepared from these cell cultures were subjected to immunoblotting with the anti-Apg8p antibody. As shown in Fig. 3 B, the amount of Apg8p started to increase 30 min after shift to starvation and remained at a high level after two hours. Finally, the amount of Apg8p increased about eightfold in response to starvation (Fig. 3 B).

Next, we performed Northern blotting analysis. Total mRNA was prepared from wild-type cells growing in YEPD medium, and the cells shifted to SD(−N) medium for various periods of time. As shown in Fig. 3 C, the amount of APG8 mRNA increased drastically in response to starvation, and reached at a maximum level within 30 min after shift to starvation. After 30 min, the amount gradually decreased, but at six hours, it was still severalfold higher than that of the growing cells. This result indicates that the increase of Apg8p during starvation is regulated at transcription level. Among the 12 APG genes (APG1, 4–10, 12–14, and 16) characterized so far, APG8 is the first gene whose expression is enhanced by starvation.

Autophagy is known to be induced by inactivation of Tor, a phosphatidylinositol kinase homologue (Noda and Ohsumi 1998). We investigated the effect of Tor function on the expression of Apg8p. Wild-type cells were treated with rapamycin in YEPD medium for two hours, and the lysates prepared from the cells were subjected to immunoblotting with the anti-Apg8p antibody. As shown in Fig. 3 D, the expression of Apg8p was enhanced in response to the drug. Since rapamycin specifically inhibits the signal transduction mediated by Tor (for review see Thomas and Hall 1997), the transcription of APG8 should be under the control of Tor-mediated signal transduction.

In addition, we investigated whether other APG genes are necessary for the increase of Apg8p. Lysates were prepared from the other 14 apg mutant cells before and after shift to starvation and subjected to immunoblotting with the anti-Apg8p antibody. In every other apg mutant, Apg8p increased normally in response to starvation as the wild-type cells (data not shown), indicating that no other APG genes are required for the induction of Apg8p.

We carried out subcellular fractionation. Cell lysates were prepared from growing and starved wild-type cells. Cell lysates (T) were centrifuged at 13,000 g for 15 min and the pellet (LSP) fractions were obtained. The resulting supernatant fractions were spun at 100,000 g for one hour to generate supernatant (HSS) and pellet (HSP) fractions, and the distribution of Apg8p was examined by immunoblotting with the anti-Apg8p antibody. Apg8p was recovered mostly in LSP and HSP fractions under both growing and starvation conditions, although in starved cells it was more detectable in HSS fraction than in growing cells (Fig. 4 A). Next, the cell lysates were vigorously sonicated to exclude lumenal proteins out of organelles or other membrane structures, and was spun at 100,000 g for one hour to generate a pellet fraction. Under both growth conditions, Apg8p mostly remained in the pellet fraction, in contrast CPY, a vacuolar lumenal protein, was completely recovered in supernatant fraction (Fig. 4 B). It suggests that Apg8p is bound to membrane or associated with a pelletable large protein complex. The pellet fractions generated from the sonicated lysates were treated with 2% Triton X-100, incubated for 30 min on ice, and then separated to supernatant and pellet fractions by centrifugation at 100,000 g for another one hour. As shown in Fig. 4 C, 2% Triton X-100 efficiently solubilized Apg8p, indicating that most pelletable Apg8p is bound to some membrane structures under both growing and starvation conditions.

To examine intracellular localization of Apg8p, immunofluorescence microscopy was performed. First, we carried out this experiment using the anti-Apg8p antibody, but sufficient signal was not obtained. Then, we constructed a single copy 3 × HA-tagged APG8 plasmid and introduced it into the Δapg8 mutant. We ascertained that the expression level of the 3 × HA-tagged APG8 gene in the resulting transformant, TK114, was similar to that of authentic APG8 (data not shown), and that the transformant recovered the autophagic activity up to 70% of the wild-type cells (Fig. 1). The cells were grown in SD+CA medium until 2 × 10⁷ cells/ml, and diluted fourfold in YEPD medium. The cells were grown for two generations, transferred to SD(−N) medium, and incubated for 0, 0.5, and 3 h. The localization of 3 × HA–Apg8p was examined with anti-HA mAb. In the cells growing in YEPD medium (t = 0), the fluorescence image consisted of many tiny dots dispersed throughout the cytoplasm, which was distinct from well known organelles such as ER, Golgi body, nucleus, vacuole, or plasma membrane (Fig. 5A and Fig. C, Fig. 010 h). In addition, we occasionally observed punctate signals distinct from the tiny dots (Fig. 5 A, arrows). 30 min after the shift to the starvation condition, the staining pattern drastically changed; one or two bright, large, punctate signals appeared in the cytoplasm (Fig. 5 C). These large punctate signals always resided in the cytoplasm, just beside the vacuole (Fig. 5 C). The number of these punctate signals was constantly 1–3/cell during starvation.

We further examined the localization of Apg8p in the background of Δpep4 mutant, in which the autophagic bodies accumulate in the vacuole during starvation (Takeshige et al. 1992). TK116 strain was generated by introducing the 3 × HA-tagged APG8 plasmid into the Δapg8Δpep4. As shown in Fig. 6 A, under growing condition, the fluorescence pattern was observed as tiny dots dispersed in the cytoplasm as TK114 cells. At 30 min after shift to the starvation condition, 1–3 large punctate signals per cell appeared next to the vacuole, also like wild-type cells (Fig. 6 B). At one hour, more punctate signals were observed than wild-type cells, and >50% were detected in the vacuole (Fig. 6 C). The punctate signals in the vacuole gradually increased and completely filled the vacuole at six hours later (Fig. 6, C–E). The signals coincided with autophagic bodies (Fig. 6C and Fig. H, arrows). As autophagic bodies are derived from autophagosomes, at least some population of Apg8p might be localized on or in the autophagosomes and delivered to the vacuole during starvation. Some of the large punctate signals in the cytoplasm (Fig. 5 C, 0.5 and 3 h, and Fig. 6) may represent autophagosomes.

To obtain further information about the intracellular localization of Apg8p during starvation, immunoelectron microscopic analysis was performed using the Δpep4 cells, TK116. The logarithmically growing cells were shifted to SD(−N) medium for one, two, or three hours. The localization of 3 × HA-Apg8p in the starved cells was examined with the anti-HA antibody. In the autophagic bodies, density of the gold particles was significantly higher than that in the cytoplasm (Fig. 7, arrows). This fact was in good agreement with the result obtained from immunofluorescence analysis. Moreover, it was revealed that most of Apg8p was in the lumen, but not on the membrane of autophagic bodies.

Next, we examined localization of Apg8p in the cytoplasm of the starved cell. As expected from the result of immunofluorescence microscopy, gold particles were associated with the autophagosomes (Fig. 8). Most mature autophagosomes with clear unit membranes and uniform intramembrane space were certainly stained, but not heavily (Fig. 8A and Fig. B), and, in some cases, the gold particles were mostly detected in their lumen (Fig. 8 B). As shown in Fig. 8C and Fig. D, some autophagosomes were heavily stained with the gold particles. In these autophagosomes, the gold particles were mostly detected along the double membrane. However, their intramembrane space was not homogenous, but partly swollen (Fig. 8C and Fig. D, arrows) not like the mature autophagosome. These structures may represent nascent autophagosomes or the latest structures in the autophagosome formation.

Besides autophagosomes, we found heavily stained structures with immunogold in the cytoplasm. In Fig. 9 A, the gold particles are located along the surface on a curved membrane sac. Fig. 9 B shows that the gold particles were arranged in a rough circle, and some of them resided along membrane structures, which partly appear as clear double membrane. In Fig. 9 C, a semicircular isolation membrane (large arrow) and its open region (small arrow) were stained with gold particles. The gold particles were apparently more concentrated in the open region rather than on the isolation membrane. In addition, we found that gold particles were concentrated in a limited area close to the vacuole (Fig. 9 D). The electron density in this area was less than that of the cytosol. The image of this Apg8p-enriched area is similar to that of the open region of the semicircular isolation membrane shown in Fig. 9 C. We suppose that all these structures shown in Fig. 9 represent intermediate structures of autophagosome.

To elucidate the step at which Apg8p functions in the autophagic pathway, we used ypt7 mutant as a control. Ypt7p, a Rab family protein, is responsible for the fusion events to the vacuole (Schimmoller and Riezman 1993; Haas et al. 1995). It was reported that in Δypt7, Cvt vesicles become detectable (Kim et al. 1999). We studied by electron microscope, using rapid freezing and freeze-substitution-fixation method, whether Ypt7p is also required for the fusion of autophagosome to the vacuole. As shown in Fig. 10 A, Δypt7 cells under starvation had many fragmented vacuoles, and autophagosomes were much more detectable in the cytoplasm than in wild-type cells. Thus, the depletion of YPT7 causes the accumulation of autophagosomes in the cytoplasm.

The Δapg8 cells were examined also by EM. They were mostly normal with a few large vacuoles, but autophagosomes could be hardly detected in their cytoplasm (Fig. 10 B). However, at low frequency, membrane structures, having enclosed a portion of cytosol, were observed only under starvation conditions. Some were indistinguishable from the autophagosome (Fig. 10 C), but others showed more complicated structures, distinct from typical autophagosome, such as the structure having condensed contents or multivesicular structure (Fig. 10D and Fig. E). Next, Δapg8 cells having the background of Δypt7 were examined. The Δypt7Δapg8 cells did not accumulate autophagosomes in the cytosol, and at low frequency autophagosome-like structures were detected in their cytoplasm, like Δapg8 cells (data not shown). We counted the number of autophagosomes and autophagosome-like structures in Δypt7 cells and Δypt7Δapg8 cells. In the sections of 500 cells, we found 269 autophagosomes in Δypt7 cells, whereas in Δypt7Δapg8 cells, only 16 autophagosome-like structures were detected (5.9% of Δypt7 cells). The result clearly shows that autophagosomes are not accumulated in Δapg8 cells.

We reported that autophagy is not only responsible for nonselective degradation of cytosolic components, but also highly selective proAPI sequestration to the vacuole during starvation (Baba et al. 1997). ProAPI could be a marker as a cargo of the autophagosomes under starvation, just like Cvt vesicles under growing condition (Kim et al. 1999). Cell lysate was prepared from starved Δypt7 cells and analyzed by immunoblotting with anti-API antibody. As shown in Fig. 10 F, when the lysate was treated with proteinase K, ∼50% of proAPI was resistant to the proteinase, but it was completely digested in the presence of 2% Triton X-100, implying that the proteinase K-resistant proAPI resides in the lumen of autophagosomes. In contrast to Δypt7, proteinase K-resistant proAPI was not detected in Δapg8 cells (Fig. 10 F). This provides a biochemical evidence that Δapg8 cells do not accumulate autophagosome. Harding et al. 1995 reported that proteinase K is also completely accessible to proAPI at growing phase in cvt5/apg8 mutant, indicating that Cvt vesicles are not formed in apg8 mutants.

During this analysis, we found that proAPI was partially matured in the starved Δapg8 strain. This partial maturation was not detected in growing Δapg8 cells (data not shown). It is not a general feature of apg mutants. This suggests that there is some starvation-induced machinery delivering proAPI to the vacuole in apg8 null mutants. As shown in Fig. 10C–E, several kinds of autophagosome-like structures were observed in the starved Δapg8 cells. Although the frequency of these structures is very low, it is likely that these structures are responsible for the delivery of proAPI to the vacuole under starvation. ProAPI might be selectively sequestered to the vacuole via these membrane structures. However, because the Δapg8 mutant cannot carry out bulk protein degradation under starvation (Fig. 1), the structures must be insufficient for the bulk sequestration of cytoplasmic components to the vacuole. Maturation of proAPI should not reflect the nonselective bulk protein degradation exactly.

Lang et al. 1998 stated the accumulation of double membrane structure in aut7/apg8 mutant. Since the EM techniques that they used are quite different from ours, and the preservation of membrane structure is not good, it is hard to directly compare our morphological data with theirs. It is possible that their reported structure is one of autophagosome-like structures reported here. Our morphological and biochemical data demonstrate that Δapg8 strain never accumulates autophagosomes and strongly indicate that Apg8p plays a crucial role in autophagosome formation.

Here, we report characterization of Apg8p, one of the APG gene products essential for autophagy. APG8 turned out to be identical to AUT7, recently reported (Lang et al. 1998). They proposed that Apg8p functions in the delivery of autophagosome to the vacuole along microtubule. However, we found that autophagy proceeds normally, even if the microtubule is depolymerized (Fig. 2). This result suggests that microtubule does not play an essential role in the autophagy in yeast. The proposal by Lang et al. 1998 was based upon the physical interactions of Apg4/Aut2p with tubulin and Apg8/Aut7p by in vitro and two-hybrid analyses. But, they did not demonstrate that Apg8p is bound to microtubule via Apg4/Aut2p. We have found that recombinant GST–Apg4p is bound to glutathione-Sepharose column nonspecifically (our unpublished result). It may be due to the acidic nature of Apg4p (predicted isoelectric point is 4.4). One possible interpretation of their result is that Apg4p is bound to tubulin nonspecifically. They also described qualitatively that the Δapg8 cells accumulate autophagosomes in the cytoplasm. Here, we demonstrated quantitatively that the apg8 null mutant does not accumulate autophagosomes in the cytoplasm. So, we concluded that Apg8p participates in the autophagosome formation. This conclusion is not affected whether Apg4/Aut2p is indeed bound to tubulin.

Intermediate structures of autophagosome formation have been poorly characterized in both yeast and mammalian cells because of their dynamic feature and the lack of specific marker of autophagosome. In yeast, only a cup-shape structure was detected as a possible intermediate structure of autophagosome (Baba et al. 1994). In this study, we found that most Apg8p is bound to membrane (Fig. 4) and localized in autophagosomes and autophagic bodies (Fig. 6,Fig. 7,Fig. 8). Apg8p must be a key molecule to identify intermediate structures of autophagosome formation. Actually, we found that Apg8p is localized on premature autophagosomes (Fig. 8C and Fig. D) and isolation membranes (Fig. 9, A–C). In addition, an Apg8p-enriched region was observed in the cytoplasm close to the vacuole (Fig. 9 D), and a similar image was obtained in the open region of the semicircular isolation membrane (Fig. 9 C, small arrow). These regions were always electron less-dense, indicating that they contain Apg8p-localized membrane- or lipid-containing structures, possible precursor structures of autophagosomal membrane. Based on our observations, we propose a model for the scheme of autophagosome formation as follows. The isolation membrane is formed by sequential assembly of the precursor structures. As the isolation membrane becomes spherical (Fig. 8C and Fig. D), its intramembrane space gradually becomes thin and homogenous, and finally a mature autophagosome is formed (Fig. 8A and Fig. B). Our assembly model is distinct from the one that autophagosome is formed by enclosing the cytosol with preexisting membrane cisterna, such as ER or Golgi body. Autophagosomal membrane shows a unique feature morphologically distinct from well-known organelles (Baba et al. 1994, Baba et al. 1995). It is observed as a thinner membrane and has a much lower density of intramembrane particles in freeze-fracture images than the membrane of other organelles. These features may be derived from the nature of the precursor structures. The dot structures observed under growing condition may be the same to them and to one of the sources of autophagosomal membrane.

Immuno-EM analysis showed that Apg8p was enriched on the cytoplasmic faces of premature autophagosomes and intermediate structures (Fig. 8C and Fig. D, and Fig. 9). These data indicate that Apg8p bound to membrane plays its role in the formation of autophagosome. In mature autophagosomes, Apg8p was observed less than in those intermediate structures (Fig. 8A and Fig. B). In some typical autophagosomes, it was detected more readily in the lumen than on the membrane (Fig. 8 B). In the autophagic bodies, most Apg8p was detected in the lumen (Fig. 7). Furthermore, it was scarcely detected on the vacuolar membrane to which the outer membrane of the autophagosome had fused (Fig. 7). These results suggest that Apg8p starts to dissociate from the membrane of autophagosome after finishing its role. This may be a reason why more Apg8p was detected in HSS fraction during starvation (Fig. 4 A). Apg8p dissociated from the inner membrane would be entrapped in the lumen of autophagosome and transported to the vacuole, and Apg8p detached from the outer membrane may be recycled for the next autophagosome formation. Since Apg8p is not an integral membrane protein, it would be able to attach and detach from membrane. We found that some portion of Apg8p is tightly bound to membrane (data not shown). However, it is still unclear how Apg8p interacts with membrane structures, particularly autophagosomal membrane. The interaction of Apg8p with membrane is now under investigation.

The apg8 null mutation severely impairs autophagosome formation, leading to the defect in bulk nonselective protein transport and degradation (Fig. 1 and Fig. 10). However, a small amount of mature API was detected during starvation. ProAPI may be transported to the vacuole via structures rarely observed in Δapg8 cells (Fig. 10, C–E). Among them, there are the structures indistinguishable from autophagosome (Fig. 10 C), suggesting that a small number of autophagosomes are built up in the absence of Apg8p. Thus, Apg8p may modulate the efficiency of autophagosome formation. In addition, we found the starvation-induced structures showing abnormal morphology in the Δapg8 cells (Fig. 10D and Fig. E). Alternatively, Apg8p might regulate the morphogenesis of autophagosome. In any case, Apg8p would play an important role in the assembly of the precursor structures into autophagosomal membrane.

The Cvt pathway proceeds with topologically the same membrane dynamics to the autophagy (Baba et al. 1997; Scott et al. 1997). Proteinase K is accessible to proAPI without detergent in the growing cvt5/apg8 mutant cells (Harding et al. 1995), indicating that Apg8p also is required for the formation of Cvt vesicle. Since Apg8p is localized on autophagosomes, it would be reasonable to speculate that it is localized on the Cvt vesicles also. In the growing cells, small punctate signals of 3 × HA–Apg8p may represent the Cvt vesicles (Fig. 5 A; arrows). Between the two pathways, the most apparent difference is the size of vesicles. The autophagosomes are 300–900-nm in diam, while the Cvt vesicles are 140–160-nm (Baba et al. 1994, Baba et al. 1997). Surface area of autophagosomal membrane is calculated at ∼16-fold of Cvt vesicle, on average. Therefore, the autophagosome formation would require more Apg8p than the Cvt vesicle formation. Moreover, a significant amount of Apg8p is transported to the vacuole upon autophagy during starvation. These may be the reasons why Apg8p increases during starvation. However, it is still an open question whether the increase of Apg8p is actually necessary for the autophagy. At least, it is clear that the increase of Apg8p is not sufficient for the induction of autophagy, because overexpression of Apg8p did not induce autophagy under growing condition (data not shown). In addition, we showed that Apg8p is transcriptionally upregulated by inactivation of Tor signaling cascade. We found STRE-like sequences in the promoter region of APG8. STREs are found in the genes induced in response to various stresses, such as CTT1, encoding cytosolic catalase T, and play roles as positive control element (Belazzi et al. 1991; Schuller et al. 1994; Moskvina et al. 1998). Marchler et al. 1993 reported the mutation in the STREs of CTT1 (CTT1-23), which causes loss of induction activity. The corresponding mutation was introduced in the STRE-like sequences of APG8. However, it led to the increase of Apg8p at growing phase instead of the loss of the induction during starvation (data not shown), suggesting that a negative regulator is bound to the cis-elements to repress the expression under nutrient rich condition.

Apg8p is the first identified molecule that is localized on the autophagy-related membrane structures. Now it becomes a useful marker for further analysis on the whole process of membrane dynamics in autophagy. Primary structures of Apg8p are highly conserved among homologues in other organisms (Lang et al. 1998). We anticipate that some Apg8p homologues may function in the process of formation of autophagosome in higher eukaryotes. Further study of Apg8p will provide a breakthrough for elucidating the molecular mechanism of autophagy.

We are grateful to Dr. Masako Osumi for the use of EM facilities, Dr. Klionsky for the gift of anti-API antibody, and Dr Mizushima for the gift of the ypt7 null mutant.

This work was supported in part by Grants-in-Aid for the Ministry of Education, Science, and Culture of Japan.