Structure-based Analyses Reveal Distinct Binding Sites for Atg2 and Phosphoinositides in Atg18*

Background: Atg18 plays a critical role in autophagy as a complex with Atg2 and phosphatidylinositol 3-phosphate. Results: The structure of the Atg18 paralog was determined, and important residues in Atg18 were identified. Conclusion: Atg18 recognizes Atg2 and membranes using distinct regions. Significance: This study will be a basis for elucidating the function of Atg18 in autophagy. Autophagy is an intracellular degradation system by which cytoplasmic materials are enclosed by an autophagosome and delivered to a lysosome/vacuole. Atg18 plays a critical role in autophagosome formation as a complex with Atg2 and phosphatidylinositol 3-phosphate (PtdIns(3)P). However, little is known about the structure of Atg18 and its recognition mode of Atg2 or PtdIns(3)P. Here, we report the crystal structure of Kluyveromyces marxianus Hsv2, an Atg18 paralog, at 2.6 Å resolution. The structure reveals a seven-bladed β-propeller without circular permutation. Mutational analyses of Atg18 based on the K. marxianus Hsv2 structure suggested that Atg18 has two phosphoinositide-binding sites at blades 5 and 6, whereas the Atg2-binding region is located at blade 2. Point mutations in the loops of blade 2 specifically abrogated autophagy without affecting another Atg18 function, the regulation of vacuolar morphology at the vacuolar membrane. This architecture enables Atg18 to form a complex with Atg2 and PtdIns(3)P in parallel, thereby functioning in the formation of autophagosomes at autophagic membranes.

Macroautophagy (hereafter referred to as autophagy) is an intracellular degradation system conserved among eukaryotes from yeast to mammals. During autophagy, a double-membrane structure called an autophagosome sequesters a portion of the cytoplasm and fuses with a vacuole (or lysosome in the case of mammalian autophagy) to deliver its inner contents to the lumen of the organelle (1). Autophagy is important in a wide range of physiological processes, such as adaptation to starvation, quality control of intracellular proteins and organelles, embryonic development, elimination of intracellular microbes, and prevention of neurodegeneration and tumor formation (2)(3)(4).
Currently, Ͼ30 genes involved in autophagy have been isolated in yeast and termed autophagy-related (ATG) genes. Among these genes, ATG1-10, ATG12-14, ATG16 -18, ATG29, and ATG31 are essential for autophagosome formation during starvation-induced autophagy, and the 18 Atg proteins they encode are classified into six functional groups (1,5): (i) starvation-responsive Atg1 kinase complex, (ii) class III phosphatidylinositol (PtdIns) 2 3-kinase complex I, (iii) proteins involved in the ubiquitin-like conjugation of Atg12 with Atg5, (iv) proteins involved in the ubiquitin-like conjugation of Atg8 with phosphatidylethanolamine, (v) multimembrane-spanning protein Atg9, and (vi) the Atg2-Atg18 complex. These Atg proteins localize, at least in part, to the pre-autophagosomal structure (PAS), which is proximal to the vacuole and plays a central role in autophagosome formation (6). The characterization of each of these proteins is ongoing, and the interrelationships among these functional groups have also been studied systematically. However, except for the proteins involved in ubiquitin-like conjugation (7), structural studies on the Atg proteins involved in autophagosome formation have been limited.
Atg18 plays a critical role in autophagy (8,9) by forming a protein complex with Atg2 (10). Besides Atg2, Atg18 also binds to PtdIns(3)P and PtdIns(3,5)P 2 and is therefore considered to be one of the effectors of these molecules (11,12). Atg18 and Atg2 localize to the PAS in an interdependent manner, for which the ability of Atg18 to bind PtdIns(3)P is required (10,13). Furthermore, the production of PtdIns(3)P at the PAS by PtdIns 3-kinase complex I is required for autophagosome for-* This work was supported in part by Japan Society for the Promotion of mation (14,15). These observations suggest that the interaction of Atg18 with Atg2 and PtdIns(3)P at the PAS is essential for the formation of autophagosomes. In addition to autophagy, Atg18 also has a role in regulating the vacuolar morphology of yeast, for which Atg18 localizes to the vacuolar membrane through its interaction with PtdIns(3,5)P 2 , but not with PtdIns(3)P (11,16). Previous studies predicted the structure of Atg18 and its homologs as a seven-bladed ␤-propeller fold and identified a putative phosphoinositide-binding motif (FRRG) within the predicted ␤-propeller structure (11,17,18). However, little is known about the molecular mechanisms underlying how Atg18 recognizes Atg2 and PtdIns(3)P via its ␤-propeller structure and how the Atg2-Atg18 complex functions in the formation of autophagosomes.
In yeast, two Atg18 paralogs have been identified: Atg21 and Hsv2 (11,12,19). Although Atg21 and Hsv2 are not essential for autophagy, they contain an FRRG motif and bind to PtdIns(3)P and PtdIns(3,5)P 2 (12,17,20). Moreover, they were also predicted to have a seven-bladed ␤-propeller similar to Atg18. To reveal the architecture of Atg18, we herein report the crystal structure of Hsv2 from a thermotolerant yeast, Kluyveromyces marxianus (KmHsv2), at a resolution of 2.6 Å. The structure reveals a seven-bladed ␤-propeller fold. Mutational analyses of Atg18 based on the structure of KmHsv2 showed that Atg18 possesses two binding sites for phosphoinositides at blades 5 and 6, whereas the loop regions in blade 2 are specifically required for recognizing Atg2 and thus for autophagy. These results suggest that Atg18 tethers Atg2 to the PAS and autophagic membranes through its simultaneous interaction with Atg2 and PtdIns(3)P, thus playing a critical role in the formation of autophagosomes.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-KmHsv2 was amplified by PCR and cloned into the pGEX-6P-1 vector (GE Healthcare) to produce GST fusion proteins. The construct was sequenced to confirm its identity and expressed in Escherichia coli BL21(DE3) cells that were cultured in 2ϫ YT medium containing 10 g/liter yeast extract, 16 g/liter Tryptone, and 5 g/liter sodium chloride. After cell lysis by sonication, GST-fused proteins were purified by affinity chromatography using a glutathione-Sepharose 4B column (GE Healthcare). The GST tag was then cleaved with PreScission protease (GE Healthcare) and removed by affinity chromatography using a glutathione-Sepharose 4B column. This process left a Gly-Pro-Leu-Gly-Ser sequence at the N terminus of Hsv2. Further purification was performed using a Superdex 200 gel filtration column (GE Healthcare) and elution with 20 mM Tris-HCl (pH 8.0) and 150 mM NaCl.
X-ray Crystallography-Crystallization of Hsv2 was performed using the sitting drop vapor diffusion method at 20°C. Drops of 10 mg/ml Hsv2 in 20 mM Tris-HCl (pH 8.0), 150 mM NaCl, and 2 mM dithiothreitol were mixed with equal amounts of reservoir solution (1.2 M (NH 4 ) 2 SO 4 and 100 mM acetate buffer at pH 5.5) and equilibrated against 100 l of the same reservoir solution by vapor diffusion. Crystals, typically with dimensions of 0.30 ϫ 0.25 ϫ 0.25 mm, were obtained within one week. Diffraction data of the native and selenomethioninelabeled crystals were collected on an ADSC Quantum 210 charge-coupled device detector using beamline AR-NW12A at KEK (Ibaraki, Japan). The diffraction data were indexed, integrated, and scaled using the HKL2000 program suite (21). The initial phasing was performed by the multiwavelength anomalous dis- rystal Structure of KmHsv2 persion method using the peak, edge, and remote data from the selenomethionine-labeled crystals. After the 28 selenium sites were identified using the SHELXD program, the initial phase and density modification were calculated using the SHELXE program (22). Automated model building was performed using the Buccaneer program (23) in the CCP4 program suite (24). Further model building was performed manually using the Coot program (25), and crystallographic refinement was performed using the crystallography and nuclear magnetic resonance system software (26). Data collection, phasing, and refinement statistics are summarized in Table 1.
Yeast Strains and Media-We utilized standard methods for yeast manipulation (27). The Saccharomyces cerevisiae strains used in this study are listed in Table 2. Yeast cultures were incubated in rich YPD medium (1% Bacto-yeast extract, 2% Bacto-peptone, and 2% D-glucose) or SDCA medium (0.17% yeast nitrogen base (without amino acids and ammonium sulfate), 0.5% ammonium sulfate, 0.5% casamino acid, and 2% D-glucose) containing appropriate amino acids. Gene disruption or epitope tagging was carried out as reported previously (28,29). To induce autophagy, the cells were grown to mid-log phase in YPD or SDCA medium and then incubated in 0.17%  Plasmid Construction for Yeast Experiments-The pRS316based plasmid for Atg18 with a 3ϫHA-enhanced GFP (EGFP) tag (hereafter referred to as Atg18-HG) was generated as reported previously (30). Point mutations were introduced by PCR-based site-directed mutagenesis using the pRS316-based plasmid for Atg18-HG as a template. The successful introduction of the point mutations was confirmed by sequencing.
Microscopy Observations-The intracellular localization of monomeric red fluorescent protein (mRFP)-or EGFP-tagged proteins was visualized using an inverted fluorescence microscope (IX-71, Olympus) equipped with an EM-CCD digital camera (ImagEM, Hamamatsu Photonics K.K.). Images were acquired using AquaCosmos 2.6 software (Hamamatsu Photonics K.K.) and processed using Photoshop CS4 software (Adobe Systems). To observe the PAS, yeast cells were treated with rapamycin (final concentration, 0.2 g/ml) for 1 h to induce autophagy.

RESULTS
Overall Structure of KmHsv2-We first tried to crystallize S. cerevisiae Atg18 but failed to obtain good diffracting crystals, so we adopted a strategy to obtain crystals from various Atg18 homologs/paralogs, including S. cerevisiae Atg21, Pichia pastoris Atg18, Arabidopsis thaliana Atg18b, and Homo sapiens Atg18 homologs (WIPI-1-4). However, these trials also failed. Recently, we succeeded in determining the solution structure of Atg10 by switching the target from S. cerevisiae Atg10 to K. marxianus Atg10 (32). K. marxianus is a thermotolerant yeast, so the homologs in this yeast could be expected to have higher stability than those in other eukaryotes. We thus attempted to crystallize K. marxianus Atg18 paralogs and succeeded in obtaining good diffracting crystals of KmHsv2 (referred to simply as Hsv2 hereafter), and we determined the crystal structure of Hsv2 by the multiwavelength anomalous dispersion method using a selenomethionine-substituted crystal (supplemental Fig. S1). The structure was refined against 2.6 Å data to an R-factor of 0.224 and a free R-factor of 0.252 (Table 1).
The asymmetric unit of the crystal contains two Hsv2 molecules (Hsv2A and Hsv2B). The Hsv2 model obtained lacks 13 N-terminal residues and some loop regions (residues 164 -181 in Hsv2A and residues 163-181 and 268 -284 in Hsv2B) due to undefined electron density. Hsv2 possesses a seven-bladed ␤-propeller fold, in which each blade consists of a four-␤stranded antiparallel ␤-sheet, resembling each other (Fig. 1A). Circular permutation, which is frequently observed in ␤-propeller proteins, is not observed in this fold. Residues 263-289 of Hsv2A form a large extended loop connecting the C and D ␤-strands (CD loop) in blade 6, which protrudes from the ␤-propeller fold as far as ϳ25 Å. The ordered conformation of this loop is stabilized by crystal packing. In contrast, the electron density of the equivalent residues of Hsv2B is disordered, suggesting that these residues have a flexible conformation in solution.
Two Binding Sites for Phosphoinositides-The phosphoinositide-binding FRRG motif (residues 229 -232) of Hsv2 is located at the D ␤-strand in blade 5 and the loop connecting blades 5 and 6. Intriguingly, the side chains of the two arginine residues of the motif, Arg-230 and Arg-231, point in opposite directions (Fig. 1B). In the proximity of Arg-230, there is a basic pocket composed of Ser-209, Thr-213, and Arg-216 from blade 5 and His-189, Thr-190, and Asn-191 from the loop connecting blades 4 and 5. A sulfate ion is bound to this basic pocket, suggesting that it has a role in accommodating phosphoinositides; thus, this pocket was named site 1. The structure of site 1 is very similar between Hsv2A and Hsv2B (Fig. 1B). Arg-231 is involved in the construction of another basic pocket composed of Ser-254, Lys-256, Thr-258, and His-260 from blade 6. The structure of this second pocket is somewhat distinct between Hsv2A and Hsv2B. In Hsv2A, the loop connecting blades 5 and 6 has a conformation closer to blade 6, and the side chain of Asp-234 on the loop is bound deeply in the pocket, whereas in Hsv2B, the pocket is occupied by the side chain of Glu-336 of the crystallographically adjacent Hsv2 molecule. In both cases, the basic pocket is occupied by a negatively charged carboxyl group, suggesting that it has a similar role to site 1 in accommodating phosphoinositides; thus, the second pocket was named site 2. The residues constituting sites 1 and 2 are highly conserved among the Atg18 homologs/paralogs, suggesting the possibility that Atg18 and its relatives possess two binding pockets for phosphoinositides. During the preparation of this manuscript, Baskaran et al. (33) and Krick et al. (34) reported the crystal structure of Kluyveromyces lactis Hsv2. The overall structure of K. lactis Hsv2 is similar to that of KmHsv2, and it possesses two basic pockets similar to sites 1 and 2 in KmHsv2. They showed that both pockets are important for recognizing phosphoinositides according to in vitro mutational analyses. We also observed that a single mutation at either Arg-230 (site 1) or Arg-231 (site 2) resulted in a partial defect, and simultaneous mutation at both residues resulted in a more severe defect in autophagy (supplemental Fig. S2), suggesting that both sites are important for recognizing PtdIns(3)P. Besides two binding pockets, Baskaran et al. (33) also showed that a long loop in blade 6 of K. lactis Hsv2, which is equivalent to the long loop (residues 263-289) of KmHsv2, is important for its association with membranes.
Effect of Mutating Conserved Residues of Atg18 on Vacuolar Morphology-Structurally annotated multiple sequence alignment of KmHsv2 with Atg18 orthologs (S. cerevisiae, K. marxianus, P. pastoris, and H. sapiens WIPI-1) showed that the residues in blades 2, 3, 5, and 6 are highly conserved among the Atg18 orthologs ( Fig. 2A). Fig. 2B shows the location of the conserved residues in the Atg18 orthologs on the structure of KmHsv2. Conserved exposed residues are especially clustered at blades 2 and 3. To evaluate the functional significance of these conserved residues in the regulation of vacuolar morphology, we introduced point mutations at these sites, especially those conserved among Atg18 orthologs but not in Hsv2, and prepared the following six mutants: F54A/S55A and S57A/ L58A (both at the AB loop in blade 2), I49K/L96K (at strand A in blades 2 and 3), P72A/R73A (at the BC loop in blade 2), M121A/R122A/L123A (at the CD loop in blade 3), and T126R/ N132R (at strand D in blade 3). As mentioned above, many of the residues in sites 1 and 2 are conserved among Atg18 homologs/paralogs and are responsible for the recognition of phosphoinositides. We selected His-244 from site 1 and His-315 from site 2 and prepared an Atg18 mutant with an alanine substitution at both histidine residues (H244A/H315A). These seven mutants and wild-type Atg18 were expressed as fusion proteins with a 3ϫHA-EGFP tag (Atg18-HG) in atg18⌬ cells using the pRS316 centromeric plasmid and visualized by fluorescence microscopy (Fig. 3, middle panels). At the same time, the vacuoles were visualized using FM4-64 staining (Fig. 3, right  panels). Although wild-type Atg18-HG localized to the vacuolar membrane and properly regulated the vacuolar morphology, Atg18(H244A/H315A)-HG was not recruited to the vacuolar membrane, and cells expressing Atg18(H244A/H315A)-HG showed abnormally enlarged vacuoles. We confirmed that the H244A/H315A mutations abrogated the binding affinity of Atg18 for PtdIns(3,5)P 2 by an in vitro pulldown assay using PIP beads (Echelon Biosciences) (data not shown). These results are consistent with previous reports showing that the binding activity of Atg18 for PtdIns(3,5)P 2 is required for the localization of Atg18 to the vacuolar membrane and for the maintenance of vacuolar morphology (11,16). Conversely, all of the six mutants at blades 2 and 3 showed normal localization to the vacuolar membrane, and cells expressing these mutants showed normal vacuolar morphology. These data indicate that the mutations at blades 2 and 3 do not affect the affinity of Atg18 for phosphoinositides or abrogate its role in the regulation of vacuolar morphology.
Next, we monitored aminopeptidase I (Ape1) maturation. The premature form of Ape1 (prApe1) is transported to the vacuole via the cytoplasm-to-vacuole targeting pathway under nutrient-rich conditions and by autophagy in response to starvation or rapamycin treatment (35). In the vacuole, prApe1 is processed into a mature form (mApe1), which can be monitored by Western blotting for Ape1. As shown in Fig. 4B, cells expressing wild-type and mutant Atg18-HG constructs, except for Atg18(P72A/R73A)-HG, showed a strong mApe1 band and a weak prApe1 band in response to rapamycin treatment. Because monitoring Ape1 maturation is a much more sensitive technique to detect autophagic activity than the Pho8⌬60 assay, the normal Ape1 maturation observed in the Atg18(F54A/S55A)-HG, Atg18(S57A/L58A)-HG, and Atg18(M121A/R122A/L123A) -HG cells may be due to the high sensitivity of this method. Nevertheless, Atg18(P72A/R73A)-HG cells showed a significantly weaker mApe1 band and a stronger prApe1 band, indicating that these cells retained minimal autophagy activity. These data, together with the results shown in Fig. 3, suggest that the AB loop (containing Phe-54 and Ser-55) and BC loop (containing Pro-72 and Arg-73) in blade 2 in Atg18 are important for autophagy, but they are dispensable for the regulation of vacuolar morphology and phosphoinositide recognition.
The BC Loop in Blade 2 Is Essential for PAS Targeting of Atg18-The AB and BC loops in blade 2 are not responsible for recognizing phosphoinositides; nevertheless, they were shown to be important for autophagy. To function in autophagy, Atg18 has to localize to the PAS; therefore, we speculated that the loops in blade 2 are important for PAS localization of Atg18. To confirm this speculation, we monitored the PAS localization of Atg18-HG constructs upon rapamycin treatment. As a PAS marker, mRFP-Ape1 was coexpressed with Atg18-HG constructs. As shown in Fig. 5, wild-type Atg18-HG co-localized with mRFP-Ape1 at a dot proximal to the vacuole, suggesting its PAS localization. Atg18(F54A/S55A)-HG and Atg18 (H244A/H315A)-HG also localized to the PAS, which may reflect their remaining autophagic activity. In contrast, Atg18(P72A/R73A)-HG did not localize to the PAS. These data suggest that the BC loop in blade 2 is essential for the PAS targeting of Atg18.
Atg18 Recognizes Atg2 Using Blade 2-Atg18 and Atg2 localize to the PAS interdependently. This observation suggests that the formation of the complex between Atg18 and Atg2 is essential for their PAS targeting. Because the BC loop in blade 2 is essential for PAS localization of Atg18, it is possible that the loop is involved in the interaction with Atg2. To study this possibility, we examined the Atg18-Atg2 interaction using coimmunoprecipitation. Wild-type Atg18-HG and mutant Atg18-HG (F54A/S55A, P72A/R73A, and H244A/H315A) were coexpressed with Atg2 fused to a tandem affinity purification (TAP) tag in atg18⌬ cells, and Atg18-HG constructs were pulled down using GFP-Trap magnetic beads. As shown in Fig. 6A, wild-type Atg18-HG, Atg18(F54A/S55A)-HG, and Atg18(H244A/H315A)-HG, but not Atg18(P72A/R73A)-HG, interacted with Atg2. Compared with wild-type Atg18-HG, the interaction of Atg18(F54A/S55A)-HG with Atg2 was mildly but significantly attenuated, whereas that of Atg18(H244A/ H315A)-HG was slightly enhanced. Similar results were obtained when Atg2 fused to a TAP tag was pulled down using IgG-conjugated magnetic beads (Fig. 6B). These results show that the AB and BC loops in blade 2 are important for the interaction of Atg18 with Atg2.

DISCUSSION
In this study, we determined the crystal structure of KmHsv2 and revealed the seven-bladed ␤-propeller architecture conserved among the Atg18 family of proteins. Furthermore, using FIGURE 6. Analysis of the interaction between Atg18 and Atg2. Co-immunoprecipitation (IP) experiments were performed as described under "Experimental Procedures." A, Atg18-HG constructs were pulled down using GFP-Trap magnetic beads. The protein bands for Atg2 and Pgk1 were detected using rabbit IgG and anti-Pgk1 antibody, respectively. The protein bands for Atg18 were detected using anti-GFP or anti-HA antibody. B, Atg2 fused to a TAP tag was pulled down using IgG-conjugated magnetic beads. The protein bands for Atg2, Atg18, and Pgk1 were detected using anti-Atg2, anti-HA, and anti-Pgk1 antibody, respectively.
in vivo mutational analyses of Atg18, we showed that the loop regions in blade 2 play a critical role in autophagy through their interaction with Atg2, whereas the two basic pockets in blades 5 and 6 are responsible for phosphoinositide binding and play an essential role in the regulation of vacuolar morphology. Immunoprecipitation experiments showed that the AB and BC loops in blade 2, which are located on opposite surfaces of the ringlike structure of Atg18, are important for its interaction with Atg2. These observations suggest that Atg2 recognizes both surfaces of the ring simultaneously, which might be achieved by gripping the ring from the side of blade 2 (Fig. 7). It appears to be easy for Atg2 to perform such interactions because of its very large size (ϳ180 kDa). Very recently, Baskaran et al. (33) and Krick et al. (34) showed that Hsv2 could interact edge-on to the membrane such that the two binding pockets for phosphoinositides in blades 5 and 6 might contact the membrane surface, whereas the CD loop in blade 6 might penetrate the membrane (Fig. 7). According to this model, the membrane will not interfere with the interaction between Atg18 and Atg2 because blade 2 of Atg18 will be located distally from the membrane, thus enabling the simultaneous interaction of Atg18 with both Atg2 and the membrane. This model of Atg18-Atg2 interaction appears to be conserved in their human counterparts because the residues responsible for the interaction with Atg2 are conserved between S. cerevisiae Atg18 and H. sapiens WIPI-1, an Atg18 ortholog. This hypothesis is supported by the report that H. sapiens Atg2A directly interacts with S. cerevisiae Atg18 (36).
In addition to the autophagic function of Atg18, it also has a role in the maintenance of vacuolar morphology. The interaction between Atg18 and Vac14, one of regulators of PtdIns(3,5)P 2 , appears to be required for the proper regulation of PtdIns(3,5)P 2 and vacuolar morphology (37). In this study, Atg18(P72A/R73A)-HG cells had normal sized vacuoles (Fig.  3), indicating that Atg18(P72A/R73A)-HG could be bound to Vac14. This observation suggests that the Vac14-binding site of Atg18 is distinct from its Atg2-binding site. Because the function of Atg18 in the regulation of vacuolar morphology is not conserved among the homologs in higher eukaryotes, such as mammals, the Vac14-binding site may be located in the nonconserved region of Atg18.
We recently showed that Atg18 is most likely to be important for the localization of Atg2 to the PAS (30). Conversely, the interaction of Atg18 with Atg2 was demonstrated to be essential for the PAS targeting of Atg18 (Fig. 5). This kind of interdependence between two Atg proteins for PAS targeting is also observed for Atg6-Atg14 and Atg5-Atg16 complexes (38,39). However, the mechanisms underlying this interdependent targeting of Atg proteins to the PAS have yet to be elucidated. After targeting to the PAS, the Atg2-Atg18 complex plays an absolutely critical role in autophagosome formation together with other Atg groups. However, the specific roles of this complex also have yet to be established. A structural study of the Atg2-Atg18 complex is required to uncover these critical issues and will create a path to understanding the molecular mechanisms of autophagosome formation.