Structure of Atg5·Atg16, a Complex Essential for Autophagy*

Atg5 is covalently modified with a ubiquitin-like modifier, Atg12, and the Atg12-Atg5 conjugate further forms a complex with the multimeric protein Atg16. The Atg12-Atg5·Atg16 multimeric complex plays an essential role in autophagy, the bulk degradation system conserved in all eukaryotes. We have reported here the crystal structure of Atg5 complexed with the N-terminal region of Atg16 at 1.97Å resolution. Atg5 comprises two ubiquitin-like domains that flank a helix-rich domain. The N-terminal region of Atg16 has a helical structure and is bound to the groove formed by these three domains. In vitro analysis showed that Arg-35 and Phe-46 of Atg16 are crucial for the interaction. Atg16, with a mutation at these residues, failed to localize to the pre-autophagosomal structure and could not restore autophagy in Atg16-deficient yeast strains. Furthermore, these Atg16 mutants could not restore a severe reduction in the formation of the Atg8-phosphatidylethanolamine conjugate, another essential factor for autophagy, in Atg16-deficient strains under starvation conditions. These results taken together suggest that the direct interaction between Atg5 and Atg16 is crucial to the performance of their roles in autophagy.

Autophagy mediates the bulk degradation of cytoplasmic components in lysosomes/vacuoles (1,2) and plays a critical role in numerous biological processes such as neurodegeneration and pathogen infection, as well as in the survival response during neonatal starvation (3)(4)(5)(6)(7). In autophagy, a double membrane structure called an autophagosome sequesters a portion of cytoplasm and fuses with the lysosome/vacuole to deliver its contents into the organelle lumen.
Atg5 was identified together with other Atg proteins by genetic screening in the yeast Saccharomyces cerevisiae (8). Because Atg5 has little sequence homology with proteins with known functions, it is difficult to predict its structure and function from the sequence. Thus far, biochemical analyses have shown that Lys-149 of Atg5 is conjugated to Atg12, a ubiquitinlike (Ubl) 2 modifier dependent on ATP and two enzymes Atg7 (E1-like) and Atg10 (E2-like) (9,10). Compared with other ubiquitin-like modifications, the Atg12-modification is unique in that it is irreversible and constitutive. The majority of Atg5 and Atg12 exist as Atg12-Atg5 conjugates irrespective of whether autophagy is induced or not and behave as a single protein (11). In yeast, the Atg12-Atg5 conjugate localizes to the pre-autophagosomal structure (PAS), a putative center for autophagosome formation (12). All Atg proteins involved in Atg12-Atg5 conjugation are also conserved in mammals. Localization studies of the Atg12-Atg5 conjugate in embryonic stem cells using green fluorescent protein-fused Atg5 showed that the Atg12-Atg5 conjugate is translocated from the cytosol to the isolation membranes upon nutrient deprivation and, immediately upon completion of autophagosome formation, the Atg12-Atg5 conjugate dissociates from the membrane, suggesting that it plays a significant role in autophagosome formation (13).
In addition to the covalent interaction with Atg12, Atg5 interacts non-covalently with a multimeric protein, Atg16 (14). Atg16 was originally obtained from a two-hybrid screen using Atg12 as a bait and was later confirmed to interact with Atg5 (but not Atg12) via its N-terminal region. Because Atg16 selfassembles via its C-terminal coiled-coil motif (residues 58 -123), the Atg12-Atg5 conjugate forms a multimeric complex with Atg16 (11,14). Further, as the majority of Atg12-Atg5 conjugates form a complex with Atg16 constitutively, the conjugates should function as a complex with Atg16 during autophagosome formation. In fact, the Atg12-Atg5 conjugate cannot localize to the PAS in ⌬atg16 yeast strains (12,15). In mammals, Atg16L, a functional counterpart of yeast Atg16, was shown to localize to the isolation membranes together with the Atg12-Atg5 conjugate during autophagosome formation (16).
Although the significance of the Atg12-Atg5⅐Atg16 complex in autophagosome formation has been shown, its molecular function is still not clearly understood. One identified function of the Atg12-Atg5 conjugate is to promote the formation of the Atg8-phosphatidylethanolamine (PE) conjugate, the other conjugate essential for autophagosome formation, and the targeting of the Atg8-PE conjugate to the PAS (12). Atg8 is a ubiq-uitin-like modifier that is first processed by a cysteine protease, Atg4 (17). The processed Atg8 is conjugated to PE dependent on ATP and two enzymes, Atg7 (E1-like) and Atg3 (E2-like) (18). Although Atg3, Atg7, Atg8 (processed form), and ATP are sufficient for Atg8-PE conjugation in vitro (19), the Atg12-Atg5 conjugate is also required for Atg8-PE conjugation in vivo (12). The expression of Atg8 is dramatically enhanced upon nutrient depletion. Under such conditions, Atg16 is also required for the efficient formation of Atg8-PE (12). However, the molecular mechanism by which the Atg12-Atg5 conjugate and Atg16 promote Atg8-PE formation is not clear. Recently, we reported the crystal structure of plant Atg12 and revealed that Atg12 is a ubiquitin-fold protein (20). However, structural information on Atg5 and Atg16 has been thoroughly lacking, preventing us from elucidating the molecular functions of the Atg12-Atg5⅐Atg16 complex. In this report, we describe the first structure of Atg5 in complex with the N-terminal region of Atg16. Furthermore, based on structural information, Atg16 mutants that lost the binding affinity to Atg5 were constructed and used to clarify the significance of the direct interaction between Atg5 and Atg16 in autophagy.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Expression and purification of Atg5⅐Atg16-(1-46) and Atg5⅐Atg16-(1-57) as well as the construction of the expression vector of hexahistidinetagged Atg5 were described previously (21). The expression vectors of GST-fused Atg16 and its mutants were constructed as follows. The full-length ATG16 gene was amplified by PCR and inserted into pGEX-6P-1 (GE Healthcare). Mutations leading to the indicated amino acid substitutions were introduced by PCR-mediated site-directed mutagenesis. All of the constructs were sequenced to confirm their identities and were expressed in Escherichia coli strain BL21 (DE3) cells. After cell lysis, GST-fused Atg16 and its mutants were purified by sequential chromatography using a glutathione-Sepharose 4B column (GE Healthcare) and a Superdex200 gel filtration column (GE Healthcare). Hexahistidine-tagged Atg5 was purified by sequential chromatography using a nickel-nitrilotriacetic acid column (Qiagen) and a Superdex75 gel filtration column (GE Healthcare).
Diffraction Data Collection-Crystallization of Atg5⅐Atg1-(1-46) and Atg5⅐Atg16-(1-57) was performed, and four crystal forms (I-IV) were obtained as described previously (21). Crystal form I (Atg5⅐Atg16-(1-46) complex) was used for phase determination, and crystal form IV was used for structure determination of Atg5⅐Atg16-(1-57). Collection of native data for Atg5⅐Atg16-(1-46) and Atg5⅐Atg16-(1-57) for final refinement was described previously (21) (Tables 1  and 2). Native data for phase determination as well as the mercury and platinum derivative data for Atg5⅐Atg16-(1-46) were collected on Rigaku R-AXIS IV and VII imaging plate detectors using CuK ␣ radiation from an in-house Rigaku rotating anode x-ray generator. Multiwavelength anomalous diffraction data for the selenomethionine-substituted crystal of Atg5⅐Atg16-(1-46) at three wavelengths were collected on the ADSC Quantum 315 chargecoupled device detector at the BL41XU beamline, SPring-8. All diffraction data were indexed, integrated, and scaled with HKL2000 (22).  Structure Determination-The initial phasing was performed by the combination of multiple isomorphous replacement and multiwavelength anomalous diffraction methods against Atg5⅐Atg16-(1-46) crystals using the crystallography NMR software program suite (23). After multiple isomorphous replacement phases were calculated using the data between 50 and 3.5 Å resolution for mercury and platinum derivatives, 35 selenium sites for selenomethionine-substituted crystals were identified from the anomalous difference Fourier map calculated using the peak data and the multiple isomorphous replacement phases. Multiwavelength anomalous diffraction phases were then calculated using the data between 50 and 3.0 Å resolution for a selenomethionine-substituted crystal at three wavelengths and were combined with the multiple isomorphous replacement phases. After density modification, the obtained phases were used to calculate the electron density map for model building. Model building was performed manually using programs O (24) and COOT (25), and crystallographic refinement was performed using crystallography NMR software. The structure of Atg5⅐Atg16-(1-57) was determined by the molecular replacement method using the structure of Atg5⅐Atg16-(1-46) as a search model.
In Vitro Pulldown Assay-GST-fused Atg16 and its mutants were mixed with hexahistidine-tagged Atg5 in phosphate-buffered saline and incubated at 277 K for 30 min. Subsequently, a slurry of glutathione-Sepharose 4B was added followed by further incubation at 277 K for 30 min. After washing three times with phosphate-buffered saline, proteins were incubated with 10 mM glutathione in 50 mM Tris-HCl buffer at pH 8.0. The elutes were subjected to 15% SDS-PAGE and stained with Coomassie Brilliant Blue.
In Vivo Assay-The significance of the complex formation between Atg16 and Atg5 in vivo was examined as follows. Point mutations were introduced by PCR-based site-directed mutagenesis using pRS314-or pRS424-based plasmids containing the ATG16-HA gene (14) as templates. Successful introduction of the point mutations was confirmed by sequencing. These plasmids were introduced into ⌬atg16::LEU2 cells on a SEY6210 (MAT␣ leu2 ura3 his3 trp1 lys2 suc2) (26) or BJ2168 (MAT␣ leu2 trp1 ura3 pep4-3 prb1-1122 prc1-407; Yeast Genetic Stock Center) background. SEY6210 ⌬atg16::LEU2 cells expressing Atg16 mutants were cultured in synthetic defined (SD) medium ϩ casamino acids (CAs) ϩ AdeUra medium to logarithmic phase and then transferred to SD (ϪN) medium. After culture for 4 h in SD (ϪN) medium, cells were collected. The lysates were prepared by the alkaline-trichloroacetic acid method, resuspended in SDS-PAGE sample buffer, and separated by SDS-PAGE, with subsequent immunoblotting using anti-aminopeptidase I (anti-API) and anti-Atg8 antibodies. Signals were detected using an ECL system (GE Healthcare, Buckinghamshire, UK) with a bioimaging analyzer (LAS1000; Fujifilm, Tokyo, Japan). To separate lipidated and non-lipidated Atg8 by mobility, separating gel containing 6 M urea was used (17). Accumulation of autophagic bodies was examined using BJ2168 ⌬atg16::LEU2 cells expressing the Atg16 mutants. Cells in logarithmic phase of growth were transferred to S (ϪNC) medium. After culture for 4 h in S (ϪNC) medium, cells were observed by phase-contrast microscopy. Yellow fluores-cent protein-tagged Atg16 mutants were engineered by PCRbased site-directed mutagenesis using a plasmid containing the ATG16-YFP gene (12) as a template. Successful introduction of the point mutations was confirmed by sequencing. The plasmids were introduced into SEY6210 ⌬atg16::LEU2 cells. Cells with the plasmids were cultured in SD ϩ CA ϩ AdeUra medium to logarithmic phase and transferred to SD (ϪN) medium. After culture for 4 h in SD (ϪN) medium, the cells were observed under an inverted fluorescence microscope (IX-81; Olympus, Tokyo, Japan) equipped with a cooled chargecoupled device camera (CoolSNAP HQ; Nippon Roper, Tokyo, Japan). Images were acquired using MetaMorph software (Universal Imaging, Buckinghamshire, UK) and processed using Adobe PhotoShop software (Adobe Systems, Mountain View, CA).
Atg5 Comprises Three Domains- Fig. 1B shows the topology of Atg5. Atg5 comprises two ubiquitin-like domains that flank a helix-rich domain. We named the N-and C-terminal ubiquitinlike domains UblA (yellow) and UblB (red), respectively, and the helix-rich domain between UblA and UblB, HR (green) (Fig.  1B). The linker regions connecting HR with UblA and UblB were named L1 and L2, respectively. In addition to the three domains and two linkers, Atg5 has an additional ␣-helix (␣1) at its N terminus (orange) (Fig. 1B). A color-coded ribbon diagram of Atg5 domains is shown in Fig. 1C.
Both UblA and UblB comprise a five-stranded ␤-sheet and two ␣-helices, which is a conserved feature in all ubiquitin superfamily proteins. Comparison of UblA and UblB of Atg5 with the Protein Data Bank (PDB) using the DALI search engine revealed that both domains show structural similarity to ubiquitin (PDB code 1UBI) and ubiquitin-like proteins, including LC3 (PDB code 1UGM, a mammalian ortholog of yeast Atg8) and AtATG12b (PDB code 1WZ3, a plant ortholog of yeast Atg12) with a Z-score of 5.8 -8.2 and a root mean square deviation of 2.3-3.0 Å for 66 -80 residues. Fig. 2 shows the sequence alignment of Atg5 homologs in which Val, Leu, Ile, Pro, Met, Phe, Tyr, and Trp are all considered to be hydrophobic residues and colored yellow, and other conserved residues are colored cyan. The sequences of ubiquitin and LC3 are also aligned based on their three-dimensional structures. Although the sequence identities of UblA and UblB with ubiquitin and LC3 are low (11-16%), hydrophobic residues that constitute the ubiquitin core are conserved, indicating that the two ubiquitin-like domains are part of the conserved architecture in Atg5 family proteins.
HR comprises three long and one short ␣-helices that form a helix-bundle structure. Comparison of HR with the PDB data base using the DALI search engine showed that HR is structurally similar to the domain III of CysG, a multifunctional siroheme synthase (27) (PDB code 1L3I), with a Z-score of 4.8 and a root mean square deviation value of 2.9 Å for 53 residues. HR also shows weak structural similarity to CUE (coupling of ubiquitin conjugation to ER), a ubiquitin binding domain, with a Z-score of 2.2 and a root mean square deviation value of 3.4 Å for 40 residues. Eleven hydrophobic residues constituting the core of HR are conserved among Atg5 homologs (Fig. 2), indicating that HR is also part of the conserved architecture in Atg5 family proteins. Lys-149, the conjugation site for Atg12, is located on ␣4 of HR and exposes its side chain (Fig. 1).
Structural Basis of the Atg5 Architecture-Although each domain constituting Atg5 has a common fold, the overall architecture of Atg5 is unique, and no similar structures have been reported. Three domains, two linkers, and ␣1 form many interactions with each other and are packed into a globular structure. Among all pairs, the UblA-UblB pair forms the largest interaction surface (1,574 Å 2 ). ␣3 of UblA is located on the central ␤-sheet of UblB and is surrounded by three loops connecting ␤6 -␤7, ␤8 -␤9, and ␣9 -␤10 of UblB (Fig. 3A, left), thus forming a wide point of contact with UblB. This binding mode is partly similar to that between ubiquitin and the ubiquitin-interacting motif (28), although the direction of the bound helix is inverse (Fig. 3A, right).
Compared with the interaction surface between UblA and UblB, those between UblA and HR (657 Å 2 ) and between UblB and HR (824 Å 2 ) are rather small. However, UblB and HR form extensive hydrophobic interactions with each other (Fig. 3B  and Val-177 of HR. Phe-273 also interacts with Tyr-44 and Pro-86 of UblA, thus playing a crucial role in gathering the three domains. Interestingly, the loop connecting ␣9 and ␣10 of UblB, which plays a major role in these interactions, is both sequentially and structurally similar to the equivalent loop of LC3 (Figs. 2  and 3B, right). In LC3, the loop is involved in the interaction with two N-terminal ␣-helices, ␣1 and ␣2 (29). Val-151 and Ile-254 of HR and Pro-201 and Ile-203 of UblB are also conserved in LC3 and are involved in the interaction between the ubiquitin fold region and ␣2 of LC3. Thus, the ubiquitin folds in both Atg5 and LC3 utilize the common residues for the construction of the architecture of each protein.
Sequence alignment shows that, in addition to the three domains, ␣1 and L1 are also conserved among Atg5 homologs, whereas L2 is not conserved at all (Fig. 2). L1 has four hydrophobic residues, Leu-126, Pro-127, Ile-131, and Pro-132, which are conserved among Atg5 homologs (Fig. 2). Leu-126 and Pro-132 interact with each other, whereas Pro-127 and Ile-131 interact with the conserved tryptophans Trp-73 and Trp-83 of UblA, respectively (Fig. 3C). Through these interactions, L1 takes on a rigid conformation and may contribute to fixing the UblA-HR arrangement. In addition to the interactions described above, the interactions mediated by ␣1 seem to play critical roles in the formation of the Atg5 architecture. ␣1 has two conserved hydrophobic residues Leu-8 and Trp-9 (Fig. 2); Trp-9 interacts with conserved hydrophobic residues, Val-177, Phe-182, and Phe-185 of HR, whereas Leu-8 interacts with a conserved hydrophobic residue, Leu-270, of UblB (Fig. 3D). Ile-4 of ␣1, which is partially conserved, also interacts with Leu-270. Thus ␣1, which is located prior to the N terminus of UblA, plays an essential role in the assembling of the three domains. In addition to the intramolecular interactions, Ile-4 and Leu-8 of ␣1 also interact with the hydrophobic residues in Atg16 (Fig. 3D). A, ribbon diagrams of the interaction between UblA and UblB (left) and that between ubiquitin-interacting motif and ubiquitin (right). UblA and ubiquitin-interacting motifs are yellow, whereas UblB and ubiquitin are red. B, ribbon diagrams of the interaction between HR and UblB (left) and that between the N-terminal and ubiquitincore regions of LC3 (right). The HR and N-terminal regions of LC3 are green, UblB and the ubiquitin-core of LC3 are red, and UblA is yellow. The side chains of the residues involved in these interactions as well as Lys-149 are shown as stick models. Residues observed in both Atg5 and LC3 are labeled with blue letters. C, a ribbon diagram of the interaction between L1 and UblA. UblA, HR, and L1 are yellow, green, and gray, respectively. Four conserved hydrophobic residues of L1 and two conserved tryptophan of UblA are shown as stick models. D, a ribbon diagram of the interaction of ␣1 with its neighbors. ␣1, UblA, UblB, HR, and Atg16 are orange, yellow, red, green, and cyan, respectively. The side chains of the residues involved in the interaction between ␣1 and its neighbors are shown as stick models.
To identify the residues of Atg16 crucial for the interaction with Atg5, an in vitro pulldown assay was performed using hexahistidinetagged Atg5 and GST-fused Atg16 mutants. Eleven Atg16 residues (labeled in Fig. 4A) were mutated with alanine. As shown in Fig. 4C, Arg-35 and Phe-46, especially Arg-35, were crucial for the interaction between Atg5 and Atg16.  Fig. 1B, and Atg16 is shown as a cyan ribbon diagram. The side chains of Atg16 are shown as stick models in which oxygen, nitrogen, and sulfur atoms are shown as red, blue, and yellow, respectively. B, stereoview of the detailed interaction between Atg5 and Atg16. Atg5 is colored as in Fig. 1B, whereas Atg16 is cyan. Residues involved in the interaction between Atg5 and Atg16 are shown as stick models, in which oxygen, nitrogen, and sulfur atoms are shown as red, blue, and yellow, respectively. Residues forming hydrophilic interactions are connected with broken lines. C, in vitro pulldown assay between Atg5 and Atg16 mutants. GSTfused Atg16 and its mutants were incubated with hexahistidine-tagged Atg5. Proteins pulled down with glutathione-Sepharose 4B resin were subjected to SDS-PAGE and stained with Coomassie Brilliant Blue.

Significance of the Atg5-Atg16 Interaction for Both Atg8
Lipidation and Autophagy-To reveal the significance of the direct interaction between Atg5 and Atg16, we measured the in vivo activities of two Atg16 mutants, R35A and F46A, that lack binding affinity to Atg5 in vitro. As a control, the Atg16 D25A mutant, which retains binding affinity to Atg5 in vitro, was also assayed. The activity of Atg16 mutants was first assayed by monitoring the maturation of the proform of API (aminopeptidase I), which is transported into vacuoles via the cytoplasmto-vacuole targeting pathway or by autophagy under nutritionrich and starvation conditions, respectively. The protein is then processed into a mature form within vacuoles. API maturation, which was not observed in ⌬atg16 cells, was restored in cells expressing the Atg16 D25A mutant to a similar extent as those expressing wild-type Atg16 (Fig. 5A). In contrast, there was little restoration of API maturation in cells expressing the R35A or F46A mutants of Atg16 (Fig. 5A). Next, the accumulation of autophagic bodies in vacuoles was examined using an optical microscope. Typical accumulation of autophagic bodies was observed in wild-type cells and ⌬atg16 cells expressing wildtype or D25A Atg16 (Fig. 5B, a, c, and f), whereas no such accumulation was observed in either ⌬atg16 cells or those expressing the R35A or F46A mutants of Atg16 (Fig. 5B, b, d, and e). These results support the notion that direct interaction between Atg5 and Atg16 is crucial for autophagy.
The Atg12-Atg5⅐Atg16 complex localizes to the PAS and plays a crucial role in autophagosome formation (12), where Atg5 and Atg16 (but not Atg12) is necessary for the localization. Thus, we next studied whether Atg16 mutants can localize to the PAS using yellow fluorescent protein-tagged Atg16 in FIGURE 5. Interaction between Atg5 and Atg16 is essential for autophagy. A, API maturation in cells expressing Atg16 mutants were monitored. Cells (SEY6210 background) grown in SD ϩ CA ϩ AdeUra medium were collected during the logarithmic phase (starvation Ϫ) and at 4 h after transfer into SD (ϪN) medium (starvation ϩ). The prepared lysates were separated by SDS-PAGE and subjected to immunoblotting with anti-API antibody. B, the accumulation of autophagic bodies was examined on the BJ2168 background, which lacks vacuolar proteases. Cells in the logarithmic phase of growth were transferred to S (ϪNC) medium. After culture for 4 h in S (ϪNC) medium, cells were observed by phase-contrast microscopy. WT cells bearing an empty vector (a) and ⌬atg16 cells bearing an empty vector (b) or expressiong Atg16 (c), Atg16 R35A (d), Atg16 F46A (e), or Atg16 D25A (f) are shown. Autophagic bodies were indicated by arrowheads. Bar, 5 m. C, intracellular localization of Atg16 mutants was observed. Cells (SEY6210 background) expressing yellow fluorescent protein-tagged Atg16 mutants were grown in SD ϩ CA ϩ AdeUra medium to logarithmic phase and transferred to SD (ϪN) medium. After culture for 4 h in SD (ϪN) medium, cells were subjected to fluorescence microscopy. Arrows indicate Atg16 mutants accumulated at the PAS. Bar, 5 m. D, Atg8 lipidation in cells expressing Atg16 mutants was monitored. Cells (SEY6210 background) grown in SD ϩ CA ϩ AdeUra medium were collected during the logarithmic phase (starvation Ϫ) and at 4 h after transfer into SD (ϪN) medium (starvation ϩ). The prepared lysates were separated by SDS-PAGE and subjected to immnoblottting with anti-Atg8 antibody.
⌬atg16 yeast strains. Wild-type Atg16 localized to a punctate structure proximal to the vacuole, which corresponds to the PAS (Fig. 5C, top). The Atg16 D25A mutant also localized to a similar dot structure (Fig. 5C, bottom). In contrast, Atg16 R35A and F46A mutants did not localize to any punctate structures and spread in the cytosol (Fig. 5C, middle two panels). These results suggest that the direct interaction between Atg5 and Atg16 is essential for the localization of Atg16 to the PAS.
The expression of Atg8 is significantly enhanced under starvation conditions, and the Atg8-PE level dramatically increases compared with that of unconjugated Atg8. Atg8-PE formation was severely diminished in the absence of the Atg12-Atg5 conjugate (12). Moreover, in the absence of Atg16, Atg8-PE formation was reduced, and unconjugated Atg8 was accumulated under starvation conditions but not under nutrient-rich conditions (12). Therefore, we examined Atg8-PE formation in ⌬atg16 cells expressing Atg16 mutants. In cells expressing the Atg16 D25A mutant, Atg8-PE formation was observed to a similar extent to those expressing wild-type Atg16 even under starvation conditions (Fig. 5D). In contrast, unconjugated Atg8 but not Atg8-PE was accumulated in cells expressing the R35A or F46A mutants of Atg16 under starvation conditions to a similar extent as cells transformed with a control vector (Fig.  5D). These results suggest that the direct interaction between Atg5 and Atg16 is critical for efficient Atg8-PE formation under starvation conditions.

DISCUSSION
In the yeast S. cerevisiae, many Atg proteins have been observed to cluster into a punctuate structure proximal to the vacuole, called the PAS. From the PAS, isolation membranes, the precursors of autophagosomes, are thought to be formed. The Atg12-Atg5⅐Atg16 complex was found to be localized to the PAS, suggesting that the complex plays a critical role in autophagosome formation. Thus far, it has been reported that the localization of the Atg12-Atg5⅐Atg16 complex to the PAS depends on both Atg5 and Atg16, but not Atg12 (12). However, the significance of the direct complex formation between Atg5 and Atg16 remained unclear. We have shown herein that the direct interaction between Atg5 and Atg16 is crucial not only for the localization, but also for autophagy (Fig. 5). Furthermore, we have also shown that the binding site for Atg16 and the conjugation site for Atg12 are located on the opposite sides of Atg5 (Fig. 1). These results suggest that the Atg12-Atg5⅐Atg16 complex localizes to the PAS using the surface formed by both Atg5 and Atg16. The autophagy-specific phosphoinositide 3-kinase (PI3K) complex, which itself localizes to the PAS (30), is known to be essential for targeting the Atg12-Atg5 conjugate to the PAS (12). Because PI3K produces phosphoinositide 3-phosphate (PI3P), it can be speculated that PI3P is produced on the PAS by PI3K and then interacts with the Atg5⅐Atg16 complex, either directly or via some PI3P binding protein(s), by which the complex is recruited to the PAS. Because the overall architecture of Atg5⅐Atg16-(1-57) is quite novel and the structure contains no known binding motifs for PI3P, the interaction between the Atg5⅐Atg16 complex and PI3P, if it exists, is also thought to be novel.
Atg3, Atg7, Atg8 (processed form), and ATP are necessary and sufficient for Atg8-PE formation in vitro (19). However, in addition to these factors, the Atg12-Atg5 conjugate is also required for Atg8-PE formation in vivo (12), and the significance of Atg12 in Atg8-PE formation was shown by mutational analyses (31). The expression level of Atg8 is dramatically enhanced upon nutrient depletion. Under such conditions, Atg16 is also required for the efficient formation of Atg8-PE (12). We demonstrated here that Atg16 mutants that lack affinity to Atg5 do not promote Atg8-PE formation (Fig. 5D). Because most Atg5 exists as a conjugate with Atg12 in vivo, this result suggests that Atg16 promotes Atg8-PE formation as a complex with the Atg12-Atg5 conjugate. This activity of the Atg12-Atg5⅐Atg16 complex is similar to that of E3 enzymes in the ubiquitin system. In most cases, E3 binds to both E2 and a target protein, thus mediating the transfer of ubiquitin between them. The Atg5⅐Atg16-(1-57) complex shows no structural similarity to E3 enzymes, and the molecular mechanism by which the Atg12-Atg5⅐Atg16 complex promotes Atg8-PE formation is still unclear. Because complex formation between Atg16 and the Atg12-Atg5 conjugate is crucial for both the promotion of Atg8-PE formation and targeting of the Atg12-Atg5⅐Atg16 complex to the PAS, these two events appear to have a strong relationship. Atg12 was shown to interact with Atg3, the E2 enzyme for Atg8 lipidation, by yeast two-hybrid screening (32). Taken together, these results suggest that the Atg12-Atg5⅐Atg16 complex binds to both Atg3 and the PAS, thus promoting the transfer of Atg8 between Atg3 and PE at the PAS. This idea is consistent with the observation that the Atg12-Atg5⅐Atg16 complex is also crucial for the localization of Atg8-PE to the PAS. However, further studies are required to elucidate this mechanism, including the identity of the PAS itself.