Journal list menu

Volume 598, Issue 1 p. 84-106
Free Access

Structural view on autophagosome formation

Nobuo N. Noda

Corresponding Author

Nobuo N. Noda

Institute for Genetic Medicine, Hokkaido University, Sapporo, Japan

Institute of Microbial Chemistry (BIKAKEN), Tokyo, Japan


N. N. Noda, Institute for Genetic Medicine, Hokkaido University, Kita 15, Nishi 7, Kita-ku, Sapporo 060-0815, Japan

Tel: +81 11 706 5069.

E-mail: [email protected]

Search for more papers by this author
First published: 27 September 2023
Citations: 1
Edited by Michael Thumm


Autophagy is a conserved intracellular degradation system in eukaryotes, involving the sequestration of degradation targets into autophagosomes, which are subsequently delivered to lysosomes (or vacuoles in yeasts and plants) for degradation. In budding yeast, starvation-induced autophagosome formation relies on approximately 20 core Atg proteins, grouped into six functional categories: the Atg1/ULK complex, the phosphatidylinositol-3 kinase complex, the Atg9 transmembrane protein, the Atg2–Atg18/WIPI complex, the Atg8 lipidation system, and the Atg12–Atg5 conjugation system. Additionally, selective autophagy requires cargo receptors and other factors, including a fission factor, for specific sequestration. This review covers the 30-year history of structural studies on core Atg proteins and factors involved in selective autophagy, examining X-ray crystallography, NMR, and cryo-EM techniques. The molecular mechanisms of autophagy are explored based on protein structures, and future directions in the structural biology of autophagy are discussed, considering the advancements in the era of AlphaFold.


AIM, Atg8-family interacting motif

BARA, beta-alpha repeated autophagy-specific

BATS, Barkor/Atg14(L) autophagosome targeting sequence

EM, electron microscopy

ER, endoplasmic reticulum

FIR, FIP200 interacting region

FR, flexible region


HR, handle region

IDR, intrinsically disordered region

IM, isolation membrane

LIR, LC3-interacting region

MIMs, MIT-interacting motifs

MIT, microtubule interacting and transport

PAS, pre-autophagosomal structure

PE, phosphatidylethanolamine

PI, phosphatidylinositol

PI3P, phosphatidylinositol-3 phosphate

PROPPINs, β-propellers that bind polyphosphoinositides

SARs, selective autophagy receptors

The key event in macroautophagy is de novo autophagosome formation (Fig. 1A), which plays a crucial role in determining degradation targets by sequestering them within autophagosome lumens for subsequent lysosomal degradation [[1, 2]]. In budding yeast, this process involves approximately 20 Atg proteins (known as core Atg proteins), organized into six evolutionarily conserved functional groups: the autophagy initiation complex (Atg1, Atg13, Atg17, Atg29, Atg31) [[3-6]], the autophagy-specific phosphatidylinositol (PI)-3 kinase complex (Atg6/Vps30, Atg14, Vps34, Vps15, Atg38) [[7-9]], the transmembrane protein Atg9 [[10]], the Atg2–Atg18 complex [[11-13]], the Atg8 lipidation system (Atg8, Atg7, Atg3, Atg4) [[14]], and the Atg12–Atg5 conjugation system (Atg12, Atg7, Atg10, Atg5, Atg16; Fig. 1B) [[15, 16]]. In 1993, the discovery and reporting of yeast variants related to these proteins occurred when little was known about their structure and function [[17]]. Since then, structural biologists have extensively investigated these proteins and their homologs, leading to the unveiling of the structures and biological functions of nearly all core Atg proteins by 2020 [[18]]. Concurrently, studies on the proteins involved in cargo selection during selective autophagy have also been conducted. This review provides insights into the history of structural studies on core Atg proteins and selective autophagy, highlights recent developments, and discusses the molecular mechanisms of fundamental events in autophagy as well as future research directions in the era of AlphaFold [[19]].

Details are in the caption following the image
Core Atg proteins responsible for autophagosome formation (A) Membrane dynamics during autophagosome formation. (B) Six functional groups organized by core Atg proteins essential for autophagosome formation. Numbers represent Atg protein identifiers. Disk size is proportional to the molecular weight.

Structural view on the Atg8 and Atg12 conjugation systems

Atg8 is a ubiquitin-like protein that undergoes initial processing by the cysteine protease Atg4, followed by activation and transfer to phosphatidylethanolamine (PE) through E1–E2–E3 enzymatic reactions [[14]]. The resulting Atg8–PE complex tightly associates with isolation membranes (IMs)/phagophores, playing pivotal roles in autophagy [[20, 21]]. Similarly, Atg12 is another autophagy-related ubiquitin-like protein that forms a conjugate with Atg5 through E1 and E2 enzymes [[15, 22]]. This Atg12–Atg5 conjugate then associates with Atg16, functioning as the E3 enzyme for Atg8 lipidation [[16, 23, 24]]. The study of autophagy structure commenced with the investigation of the Atg8 and Atg12 conjugation systems.

Atg8-family proteins

In 2000, the crystal structure of Bos taurus GABARAPL2/GATE-16, a mammalian homolog of yeast Atg8 and later confirmed as a functional counterpart, was reported in its pro-form (Fig. 2A, left) [[25]]. The GABARAPL2 structure consists of a ubiquitin fold and two N-terminal α-helices. Subsequently, during the 2000s, structures of other mammalian orthologs such as GABARAP [[26-30]] and LC3B [[31, 32]], as well as yeast Atg8 [[33, 34]], were reported, all exhibiting a similar architecture that confirmed the conservation of the N-terminal two α-helices in Atg8-family proteins (Fig. 2A). To date, over a hundred structures of Atg8 from various species have been determined through X-ray crystallography or NMR, including species such as Saccharomyces cerevisiae and mammals (including six orthologs such as LC3A [[35]], LC3C [[36]], and GABARAPL1 [[37]]), as well as Schizosaccharomyces pombe [[38]], Arabidopsis thaliana [[39]], Solanum tuberosum [[40]], Caenorhabditis elegans [[41]], Trypanosoma brucei [[42]], Plasmodium falciparum [[43]], and Bombyx mori [[44]], all of which demonstrated the conserved architecture. Many of these structures were determined as complexes with short peptides from selective autophagy receptors (SARs), establishing the essential role of Atg8-family proteins in cargo recognition during selective autophagy (discussed later).

Details are in the caption following the image
Structural view on the Atg8 and Atg12 conjugation systems (A) Structure of Atg8-family proteins. From left to right: crystal structures of GABARAPL2 (PDB 1EO6), GABARAP (PDB 1GNU), LC3B (PDB 1UGM), Atg8 (PDB 2ZPN), and a close-up view of the closed conformation of GABARAP and GABARAPL2 (PDB 1GNU, 1EO6). (B) Structural model of Atg8–PE (phosphatidylethanolamine) on a nanodisc. (C) Crystal structure of human ATG4B alone (left, PDB 2CY7) and complexed with LC3B (right, PDB 2Z0E). (D) Crystal structure of plant ATG12b (PDB 1WZ3). (E) Crystal structure of Atg7 (left, PDB 3VH2), Atg3 (middle, PDB 2DYT), and Atg10 (right, PDB 2LPU). (F) Proposed model of Atg8/Atg12 transfer from the catalytic Cys of Atg7 to that of each E2 (Atg3/Atg10) via a trans mechanism. “C” indicates catalytic cysteines. (G) Crystal structure of Atg5 complexed with the N-terminal region of Atg16 (PDB 2DYO). (H) Crystal structure of Atg16 (PDB 3A7P). (I) Crystal structure of the yeast Atg12–Atg5 conjugate complexed with the N-terminal region of Atg16 (left, PDB 3W1S) and the human ATG12–ATG5 conjugate complexed with the N-terminal region of ATG16L1 and the Atg3 interacting motif of ATG3 (right, PDB 4NAW).

Six mammalian Atg8 orthologs can be classified into two groups: the LC3-family (LC3A-C) and the GABARAP-family (GABARAP, GABARAPL1/L2), with subtle structural differences observed between them [[41]]. In the LC3-family, the N-terminal tail is detached from the ubiquitin fold (open conformation), while in the GABARAP-family, it is associated with the ubiquitin fold using the N-terminal methionine and the aromatic residue at position 3 (closed conformation; Fig. 2A). Although it is necessary to consider the artificial effect of crystal packing on the terminal conformation, many crystal structures of the two families in the different crystal packing fit this rule. Interestingly, yeast Atg8 adopts an open conformation, whereas C. elegans' two Atg8 orthologs, LGG-1 and LGG-2, exhibit closed and open conformations, respectively. The significance of this small structural variation in terms of functional roles remains uncertain. Notably, a structural study revealed that the helix structure of GABARAP α1 is unfolded and bound to neighboring GABARAP in the crystal, which was proposed to be related to non-autophagic functions of GABARAP: promoting tubulin polymerization and GABA(A) receptor clustering [[30]]. It is interesting whether this conformational change is also related to the autophagic functions of Atg8-family proteins. In the case of S. cerevisiae Atg8, α1 and α2 were found to be highly mobile and challenging to discern their helical structure. However, introducing a mutation of Lys26 to Pro in the linker between α2 and the ubiquitin fold stabilized it [[34]]. These findings suggest that the N-terminal two α-helices may adopt different conformations. These two α-helices have been linked to membrane tethering, hemi-fusion, or full-fusion activities of Atg8-family proteins [[41, 45-47]]. Nonetheless, the exact relationship between these in vitro activities and autophagosome formation remains elusive.

Most structural studies on Atg8-family proteins have been conducted using a non-lipidated soluble form. However, in autophagy, Atg8 functions as a conjugate with PE [[14, 21]]. In 2010, to study the membrane-bound structure, GABARAP was chemically linked to a lipid and incorporated into lipid bilayer nanodiscs, and NMR analysis revealed that GABARAP retained rotation freedom and exhibited a similar structure to unlipidated GABARAP [[48]]. Subsequently, in 2021, an NMR study on Atg8 conjugated with PE in nanodiscs through E1-E2-E3 enzymatic reactions showed that Atg8–PE was tightly associated with the membrane in a preferred orientation, reducing its motion (Fig. 2B) [[49]]. Moreover, Phe77 and 79 of Atg8-PE were proposed to be involved in membrane association, leading to an increase in the area difference between the outer and inner layers of membranes. This membrane perturbation activity of Atg8–PE has been demonstrated to be crucial for efficient autophagy in both yeast and mammals [[49]]. Mammalian ATG8 proteins were also suggested to exhibit membrane perturbation activity through interactions with the N-terminal region for membrane binding [[50]]. However, this membrane perturbation activity alone cannot induce in-bud formation in membranes, which is an essential process during autophagosome formation. In 2015, a membrane scaffold model was proposed, suggesting that Atg8–PE interacts with the Atg12–Atg5–Atg16 complex through Atg8–Atg12 interaction, forming a coat on membranes and inducing in-bud formation, which is necessary for autophagosome formation [[51]].

Enzymes mediating Atg8 lipidation

Atg4 is a cysteine protease responsible for processing Atg8 before the conjugation reaction and for deconjugating Atg8–PE for recycling [[20]]. In 2005, the crystal structure of human ATG4B was determined, revealing a Papain-like architecture with unique insertions (Fig. 2C, left) [[52, 53]]. Later in 2009, the structure of the ATG4B–LC3B complex was reported, demonstrating that the catalytic site of ATG4B is autoinhibited by the N-terminal tail and the regulatory loop. Upon LC3B binding, both inhibitory elements are removed, resulting in an open catalytic site (Fig. 2C, right) [[54]]. Atg7 serves as the E1 enzyme for Atg8 and Atg12, another ubiquitin-like protein essential for autophagy [[55]]. Atg12 consists of a ubiquitin fold with the intrinsically disordered region (IDR) attached at the N terminus (Fig. 2D) [[56]]. In 2011 and 2012, several research groups independently reported the crystal structures of Atg7, either in its free form or in complex with Atg8, Atg3 (E2 for Atg8), or Atg10 (E2 for Atg12) [[57-61]]. Atg7 is composed of the C-terminal adenylation domain, which is conserved among all E1s, and the N-terminal domain unique to Atg7 (Fig. 2E, left). The structures of Atg3 and Atg10 were also reported [[62, 63]], showing that Atg3 consists of the E2 fold with two unique insertions: the flexible region (FR) and handle region (HR) (Fig. 2E, middle), while Atg10 is simply comprised of the E2 fold. Atg7 binds to Atg3 and Atg10 using the N-terminal domain, and it interacts with Atg8 and Atg12 using the adenylation domain (Fig. 2E, right) [[57-59]]. Atg7 functions as a homodimer through the dimerization of the adenylation domain [[64]], providing two identical binding sites for ATP, Atg8, Atg12, Atg3, and Atg10, as well as two catalytic cysteines. Intriguingly, Atg8 and Atg12 thioester-bonded to the catalytic cysteine of one Atg7 protomer are transferred to the catalytic cysteine of Atg3 and Atg10, respectively, which are bound to another Atg7 protomer within the dimer (Fig. 2F) [[57, 58, 60]]. Hence, homodimerization is essential for Atg7 to initiate the enzymatic cascade.

After the formation of Atg8 ~ Atg3 and Atg12 ~ Atg10 thioester intermediates, Atg3 and Atg10 transfer Atg8 and Atg12 to PE and Atg5, respectively. Atg10 can directly recognize Atg5 using a β-hairpin structure within the E2 fold and mediate the conjugation reaction between Atg12 and Atg5 without the need for E3 enzymes [[63]]. In contrast, Atg3 requires E3 for mediating the conjugation reaction between Atg8 and PE in cells, although Atg3 can facilitate Atg8–PE formation without E3 in vitro when using liposomes with non-physiologically high PE content [[24, 65]]. The Atg12–Atg5 conjugate formed by Atg7 and Atg10-mediated reactions forms a complex with Atg16 (ATG16L1 in mammals) [[16, 23]]. The resultant Atg12–Atg5–Atg16 complex functions as the E3 enzyme for Atg8 lipidation [[24, 66]]. In 2007, the crystal structure of Atg5 complexed with the N-terminal region of Atg16 was reported, revealing the unique structure of Atg5, consisting of two ubiquitin-like domains and a helix-rich domain (Fig. 2G) [[67]]. In 2010, the full-length structure of Atg16 was reported, unveiling its parallel dimeric coiled-coil architecture (Fig. 2H) [[68]]. In 2013, the structure of the yeast and human Atg12–Atg5 conjugate was reported, demonstrating that Atg12 tightly interacts with Atg5 to form a continuous conserved surface that is crucial for the E3 function in Atg8 lipidation (Fig. 2I) [[69, 70]]. The Atg12–Atg5–Atg16 complex targets autophagic membranes via interaction with Atg21, a PI-3 phosphate (PI3P) binding protein [[71]] (WIPI2 in the case of mammals [[72]]; structural studies are described below), and recruits Atg8-loaded Atg3 to the membrane via the interaction between Atg12 and the Atg12-interacting motif in the FR of Atg3 (Fig. 2I) [[73, 74]]. This process promotes the transfer of Atg8 from Atg3 to PE. It was proposed that the binding of the Atg12–Atg5–Atg16 complex to Atg3 reorganizes the catalytic site of Atg3 from an inactive to an active conformation, further promoting the transfer of Atg8 from Atg3 to PE [[75, 76]].

Structural view on the autophagy initiation complex

The Atg1 complex

Structural studies on the autophagy-initiating Atg1/ULK complex were initially conducted on the budding yeast Atg1 complex, which consists of Atg1, Atg13, Atg17, Atg29, and Atg31. In 2012, the crystal structure of the Atg17–Atg29–Atg31 complex was determined, revealing the crescent-shaped structure of Atg17 with Atg29 and Atg31 at the concave surface (Fig. 3A) [[77]]. Atg17 forms a homodimer, resulting in an unprecedented S-shaped architecture. Although the concave surface was speculated to recognize lipid vesicles with approximately 20 nm diameter, in vitro binding assays did not detect membrane-binding activity of Atg17 [[77]]. Atg31 consists of the β-sandwich domain that binds to Atg29 and the C-terminal helix that binds to Atg17, thereby linking Atg29 to Atg17 (Fig. 3A). Atg29 consists of the N-terminal β-strand that is incorporated into the β-sandwich domain of Atg31, as well as tandem three α-helices and the C-terminal IDR [[77]]. In 2013, crystallographic analysis was applied to the N-terminal domain of Atg13, revealing its similarity in structure to the HOP1, REV7, and MAD2 (HORMA) domains (Fig. 3B) [[78]]. The non-HORMA region of Atg13, consisting of approximately 470 residues, was predicted to be intrinsically disordered, which was later confirmed using high-speed atomic force microscopy [[79, 80]]. In 2014, the minimum binding regions for Atg1 and Atg17 were determined biochemically in Atg13 IDR, and their structures were determined as a complex with the C-terminal domain of Atg1 or the Atg17–Atg29–Atg31 complex by crystallography (Fig. 3C) [[81]]. The C-terminal domain of Atg1 consists of two microtubule interacting and transport (MIT) domains that interact with each other to constitute one globular fold. Atg13 IDR contains two MIT-interacting motifs (MIMs) that bind to the MIT domains of Atg1 in a similar manner to canonical MIT–MIM interactions observed in other proteins [[81, 82]]. Atg13 IDR has at least two binding regions for Atg17, one named Atg17-binding region (17BR) and another named Atg17-linking region (17LR) [[79, 81]]. Crystallographic studies have shown that Atg1317BR binds to the N-terminal part of Atg17, while Atg1317LR binds to the part of Atg17 where two protomers of the protein meet (Fig. 3D). When a mixture of Atg13 and Atg17–Atg29–Atg31 was studied, it behaved as a higher-order assemblage, which was disassembled into the dimeric Atg13–Atg17–Atg29–Atg31 complex by mutations at either 17BR or 17LR. These data suggested that the Atg1 complex forms a higher-order assemblage, which is mediated by Atg13-mediated bridging of Atg17 molecules using the two binding regions [[79]]. The crystal structure has also revealed the interaction of Atg13 with the vacuolar membrane protein Vac8, which is responsible for tethering the pre-autophagosomal structure (PAS) to the vacuole (Fig. 3E) [[83-86]]. Subsequently, it was shown that the higher-order assemblage of the Atg1 complex triggers liquid–liquid phase separation to form liquid droplets, which were proposed to constitute the PAS responsible for autophagosome formation in yeast (Fig. 3F) [[86, 87]]. These studies have successfully determined the structure of all parts of the structured region of the Atg1 complex except for the kinase domain of Atg1. The kinase domain structure was determined for human ULK1, which showed a canonical kinase fold consisting of two lobes (Fig. 3G) [[88]].

Details are in the caption following the image
Structural insights into the autophagy initiation machinery (A) Crystal structure of the Atg17–Atg29–Atg31 complex (PDB 4HPQ). (B) Crystal structure of the HORMA domain of Atg13 (PDB 4J2G). (C) Crystal structure of the tandem MIT domain of Atg1 complexed with Atg13 MIM (PDB 4P1N). (D) Crystal structure of the Atg13 BR and LR bound to the Atg17–Atg29–Atg31 complex (PDB 5JHF). (E) Crystal structure of the C-terminal region of Atg13 bound to Vac8 (PDB 6KBM). (F) Model of the PAS as a phase-separated droplet. (G) Crystal structure of the kinase domain of ULK1 (PDB 4WNO). (H) Crystal structures of the fission yeast Atg101–Atg13 HORMA complex (left, PDB 4YK8), the human ATG101–ATG13 HORMA complex (middle, PDB 5C50), and the ATG9A fragment (colored green) bound to the human ATG101–ATG13 HORMA complex (right, PDB 8DO8). (I) Crystal structure of the C-terminal region of FIP200 (PDB 6GMA). (J) Structural model of the FIP200 N-terminal domain complexed with the tandem MIT domain of ULK1 and ATG13 MIM in the 2 : 1 : 1 stoichiometry (PDB 8SOI).

The ULK complex

The mammalian counterpart of the yeast Atg1 complex is the ULK complex, which consists of ULK1 or ULK2, ATG13, FIP200/RB1CC1, and ATG101 [[89-93]]. ULK1/2 and ATG13 are orthologs of yeast Atg1 and Atg13, respectively. Despite FIP200 being a much larger protein (1594 amino acids) than Atg17 (417 amino acids), these two proteins are considered functionally related because the N-terminal region of FIP200 shows weak sequence homology with Atg17, and both of them directly bind to Atg13/ATG13 [[94]]. ATG101 is not conserved in budding yeast, whereas mammals lack Atg29 and Atg31 [[95]]. The fission yeast Atg1 complex is a hybrid of the budding yeast Atg1 complex and the mammalian ULK complex, consisting of Atg1, Atg13, Atg17, and Atg101. Since ATG101 is absent from the budding yeast Atg1 complex, structural studies initially focused on this protein. In 2015, the crystal structure of the Atg101-Atg13HORMA complex from fission yeast was reported [[96]], followed by structural studies on human ATG101 alone or in complex with ATG13HORMA [[97, 98]]. These structural studies revealed that Atg101/ATG101 has a HORMA structure similar to Atg13/ATG13 (Fig. 3H). A well-known HORMA domain protein, Mad2, has two conformations named O-Mad2 and C-Mad2 [[99]]. A structural comparison shows that Atg101 resembles O-Mad2, whereas Atg13 resembles C-Mad2. Budding yeast Atg13 has a unique insertion named Cap, which stabilizes the C-Mad2 conformation (Fig. 3B), whereas human ATG13 and fission yeast Atg13 lack Cap, and the C-Mad2 conformation is stabilized by ATG101/Atg101 [[78, 96]]. ATG101 conserves a loop with exposed Trp and Phe residues, termed the WF finger (Fig. 3H, left), which is responsible for recruiting downstream autophagy factors such as WIPI1 and DFCP1 [[96]]. Additionally, human ATG101 has a C-terminal extension that interacts with and recruits the PI-3 kinase complex [[100]]. In 2023, the crystal structure of the carboxyl-terminal tail of ATG9A bound to the ATG101-ATG13HORMA heterodimer was reported, revealing that the HORMA dimer-interacting region of ATG9A binds to the interface between ATG101 and ATG13HORMA, which is crucial for recruiting ATG9A vesicles (Fig. 3H, right) [[101]]. Thus, the ATG101-ATG13 complex plays a central role in recruiting various downstream factors for autophagy progression. Moreover, in 2023, it was demonstrated by in vitro analyses that ATG101-ATG13HORMA undergoes slow spontaneous metamorphic conversion between distinct folds upon binding to ATG9, and the ATG9-ULK complex recruits PI3-kinase and ATG2-WIPI complexes to initiate autophagy [[102]]. On the other hand, in yeast cells, Atg9 was shown to recruit Atg2 alone and Atg18 later [[103]]. The temporal sequence of ATG recruitment in mammalian cells requires further investigation.

In 2019, structural studies on FIP200 focused on the C-terminal region and utilized crystallography (Fig. 3I). The C-terminal region consists of a coiled coil followed by a new globular domain named Claw, forming a parallel homodimer through the coiled-coil region [[104]]. The Claw domain contains a positively charged pocket, which is proposed to interact with the phosphorylated FIP200 interacting region (FIR) of p62 (Fig. 3I, the Claw-FIR interaction depicted below). In 2020, negative stain electron microscopy (EM) and cryo-EM were used to analyze the N-terminal region of FIP200, revealing a crescent shape similar to the Atg17-Atg29-Atg31 complex but closer to the scaffold-like and ubiquitin-like domains of TBK1 responsible for selective autophagy initiation. The N-terminal region forms a flexible dimer, with its conformation becoming fixed in a C-shape upon ATG13 binding [[105]]. Additionally, negative stain EM and biochemical data indicated that ULK1, ATG13, and FIP200 form a 1 : 1 : 2 complex, with the detailed structure revealed later using cryo-EM assisted with AlphaFold2 (Fig. 3J) [[106]].

This stoichiometry differs from that observed in the Atg1 complex, making it challenging to form a higher-order assemblage as seen in the Atg1 complex. The overall shape of FIP200 was visualized through negative stain EM, revealing a C-shaped N-terminal region followed by a coiled coil that extends up to 50 nm and further leads to the Claw [[107]]. In mammalian cells, FIP200 was found to create a liquid droplet in response to calcium transients in the endoplasmic reticulum (ER) [[108]]. This suggests that the mechanism of forming a liquid droplet is likely conserved between yeast and mammalian autophagy initiation complexes, although possibly through different processes. In 2023, cryo-EM analyses revealed that the ULK complex and PI3-kinase complex interact with each other to form a super complex. In this super complex, PI3KC3-C1 induces a rearrangement of ULK1 by dislocating an ATG13 loop from an inhibitory site on the FIP200 dimer, changing the stoichiometry of ULK1, ATG13, and FIP200 in the complex from 1 : 1 : 2 to 2 : 2 : 2. This stoichiometric change was proposed to promote self-phosphorylation of ULK1 to activate its kinase activity [[106]]. Further investigation is needed to fully understand the biological significance of the formation of this super complex.

Structural view on the autophagy-specific PI-3 kinase complex

Yeast PI-3 kinase complex I and mammalian Class III PI-3 kinase complex I are crucial for autophagosome formation, comprising Vps34, Vps15, Atg6/Beclin 1, Atg14, and Atg38/NRBF2. Among these components, Atg14 and Atg38/NRBF2 specifically function in autophagy, while Atg6/Beclin 1 plays a vital role not only in autophagy but also in the endosomal pathway [[7-9, 109, 110]]. In 2007, initial structural studies focused on the BH3 domain of Beclin 1, which bound to Bcl-XL and was related to apoptosis rather than autophagy [[111, 112]]. Subsequently, in 2012, the coiled-coil domain of Beclin 1 [[113]] and the beta-alpha repeated, autophagy-specific (BARA) domain of Atg6/Beclin 1 were investigated (Fig. 4A) [[114, 115]]. While these domain-based structural analyses were valuable in understanding individual domain functions, such as the identification of the aromatic finger in Beclin 1 for membrane association (Fig. 4A, right top) and the PAS-targeting role of Atg6 BARA [[114, 115]], they were not sufficient for comprehending the functions of the large PI-3 kinase complex in autophagy. To explore the whole PI-3 kinase complex, structural studies were conducted using EM. In 2014, negative stain EM provided insights into the V-shaped architecture of the human PI-3 kinase complex, composed of VPS34, VPS15, Beclin 1, and ATG14, at a resolution of 27 Å (Fig. 4B). The complex consists of two arms, with one arm formed by the Atg6/Beclin 1-Atg14 heterodimer and the Vps15 WD40 domain, and the other arm comprised of the helical and kinase domains of Vps34 and Vps15 [[116]]. In 2015, the crystal structure of the yeast PI-3 kinase complex II was determined at 4.4 Å resolution, providing a detailed architecture of the kinase complex with Vps38 instead of Atg14 (Fig. 4C) [[117]]. Subsequent cryo-EM studies from 2017 to 2023 further improved the resolution, starting at 8.5 Å and reaching 3.96 Å, revealing the overall structure of the human PI-3 kinase complex (Fig. 4D) [[106, 118, 119]]. These studies unveiled the intricate architecture of this large complex (Fig. 4E). One edge of the V-shape is formed by Atg6 BARA that possesses the aromatic finger and the membrane-binding amphipathic helix known as the Barkor/Atg14(L) autophagosome targeting sequence (BATS) of Atg14 [[120]]. These regions are believed to facilitate targeting the complex to autophagic membranes. The other edge of the V-shape contains the kinase domain of Vps34, which is positioned near the membrane and catalyzes the phosphorylation of PI to produce PI3P. The kinase domain of Vps34 exhibits high dynamics and changes its positioning relative to other regions of the PI-3 kinase complex, likely playing a crucial role in mediating the catalytic reaction [[116]]. Atg38 and NRBF2 were identified to interact with the tip of the V-shape, which is opposite to the membrane-binding regions. In yeast, Atg38 stabilizes the integrity of the complex, while in mammals, NRBF2 promotes complex dimerization [[8, 119, 121]]. In these functions, the MIT and coiled-coil domains of Atg38/NRBF2, whose structure was separately determined by X-ray crystallography (Fig. 4F), mediate protein–protein interaction and homodimerization, respectively. Through these interactions, Atg38 and NRBF2 contribute to the stability and functionality of the PI-3 kinase complex during autophagosome formation.

Details are in the caption following the image
Unraveling the structural complexity of the autophagy-specific PI-3 kinase complex (A) Structure of Atg6/Beclin 1. Left, crystal structure of the homodimeric coiled-coil domain of Beclin 1 (PDB 3Q8T). Right top, crystal structure of the BARA domain of Beclin 1 (PDB 4DDP). Right bottom, crystal structure of the BARA domain of Atg6 (PDB 3VP7). (B) Structure of Atg38/NRBF2. Top, crystal structure of the MIT domain of NRBF2 (PDB 4ZEY). Bottom, crystal structure of the homodimeric coiled-coil domain of Atg38 (PDB 5KC1). (C) Negative stain EM density map of the human autophagy-specific PI-3 kinase complex (EMD-2846). (D) Crystal structure of the budding yeast PI-3 kinase complex that contains Vps38 instead of Atg14 (PDB 5DFZ). (E) Cryo-EM density map of the human autophagy-specific PI-3 kinase complex (EMD-40669). (F) Cryo-EM structure of the human autophagy-specific PI-3 kinase complex (PDB 8SOR) and the model of membrane binding and PI3P production by this kinase complex.

Structural view on the membrane-expanding proteins

In budding yeast, Atg2 forms a stable complex with Atg18 and localizes at the contact site between the ER and the IM along with the transmembrane protein Atg9 [[103, 122, 123]]. These proteins in both yeast and mammals mediate phospholipid transfer from the ER to the IM for IM expansion [[124-130]].


PROPPINs (β-propellers that bind polyphosphoinositides) are a family of proteins that specifically bind to PI3P and PI3,5-bisphosphate [[131]]. Budding yeast possesses three PROPPINs, namely Atg18, Atg21, and Hsv2, among which Atg18 and Atg21 form stable complexes with Atg2 and Atg16, respectively, playing crucial roles in autophagy [[71, 132]]. Mammals have four WIPI proteins (WIPI1-4) that are the mammalian homologs of PROPPINs both in structure and potential function in the autophagy pathway. WIPI3 and WIPI4 interact with ATG2, while WIPI2 interacts with ATG16L1 [[72, 133-135]]. In 2012, the crystal structure of Hsv2 was independently reported by three groups, revealing the structural basis of PROPPINs. It consists of a seven-bladed β-propeller fold with two PI3P binding pockets located at blades 5 and 6 (Fig. 5A, left) [[136-138]]. Structure-based mutational analyses identified blade 2 of Atg18 as the binding site for Atg2 [[136, 139]]. Additionally, one loop of blade 6 of Atg18 reinforces the membrane interaction, and its phosphorylation negatively regulates this process [[140, 141]]. Subsequent crystallographic analyses of loop-deleted forms confirmed the conservation of this architecture in Atg18 and WIPI3 in the late 2010s [[142, 143]]. During the 2020s, significant progress was made in elucidating the structural basis of interactions involving Atg18/WIPI family proteins. Crystal structures of the fragments of Atg16, ATG16L1, and ATG2A bound to Atg21, WIPI2, and WIPI3, respectively, were reported successively (Fig. 5A) [[133, 144-146]]. In these interactions, the coiled coil of Atg16 binds to the hydrophobic pocket between blades 2 and 3 of Atg21 through hydrophobic interactions, further stabilized by salt bridges [[144]]. Similarly, ATG16L1 has two binding regions for WIPI2, both interacting with the hydrophobic pocket between blades 2 and 3 of WIPI2 in an α-helical fold, indicating evolutionary conservation of this interaction [[145, 146]]. The WIPI-interacting-region motif from ATG2A binds to blades 1–3, forming an intermolecular β-sheet with blade 2, and also binds to the hydrophobic pockets between blades 1 and 2 and between blades 2 and 3 [[133]]. Notably, the binding regions on the PROPPINs fold for ATG2 and ATG16L1 largely overlap with each other (Fig. 5A), suggesting that WIPIs cannot form a ternary complex with these two proteins. However, ATG16L1 is capable of binding to RAB33B using a region that does not overlap with the WIPI2 binding region, allowing it to interact with both proteins. This binding versatility is proposed to assist in targeting ATG16L1 to the IM or recruiting RAB33B-containing vesicles as a lipid source for autophagosome formation [[147-149]].

Details are in the caption following the image
Structural view on the membrane-expanding proteins (A) Structure of PROPPINs. From left to right: the crystal structure of Hsv2 (PDB 3VU4), the Atg21–Atg16 complex (PDB 6RGO), WIPI2 complexed with two regions in ATG16L1 (PDB 7XFR, 7MU2), and the WIPI3–ATG2AWIR complex (PDB 6KLR). (B) Structure of Atg2-family proteins. Left: negative stain EM density map of the ATG2A–WIPI4 complex (EMD–8899). Middle: crystal structure of fission yeast Atg2 complexed with PE (PDB 6A9J). Right: AlphaFold2 structure of full-length budding yeast Atg2 (AF-P53855). The cavity within Atg2 is shown with a surface model colored based on the electron static potentials (blue, positive; red, negative). (C) Model of lipid supply from the ER to the IM for membrane expansion, which is mediated by the collaboration of Atg2 and Atg9. (D) Structure of Atg9-family proteins. From left to right: cryo-EM structure of human ATG9A determined using detergents (PDB 6W9Z), amphipols (PDB 7JLO), and nanodiscs (PDB 7JLP), and fission yeast Atg9 determined using detergents (PDB 7D0I). TM indicates a transmembrane region.


Structural studies on Atg2-family proteins began with negative staining in EM. In 2018, two research groups visualized the overall architecture of mammalian ATG2A and ATG2B complexed with WIPI4, revealing a rod-like shape of ATG2 that spans up to 20 nm, with WIPI4 binding to one edge of the rod (Fig. 5B, left) [[134, 135]]. Moreover, the EM analysis of ATG2A with liposomes demonstrated that ATG2A interacts with liposomes using both edges, acting as a bridge between them [[135]]. Similar studies on yeast Atg2 indicated the presence of two membrane-binding regions, one at the N-terminus, responsible for ER-binding, and another near the C-terminal side, responsible for IM-binding [[150]]. In 2019, the crystal structure of the N-terminal region of Atg2 was determined in both free and PE-bound forms, revealing a hydrophobic cavity at the core of the fold that accommodates the acyl chains of PE without recognizing its head moiety (Fig. 5B, middle) [[124]]. Crucially, in vitro analyses demonstrated that Atg2 exhibits lipid transfer activity of phospholipids between liposomes, with little specificity toward the head group [[124]]. This lipid transfer activity was also observed for mammalian ATG2A and ATG2B, suggesting that Atg2-family proteins evolutionarily conserve this lipid transfer capability [[125, 126, 151, 152]]. Low-resolution cryo-EM analysis of ATG2A revealed the presence of a ~ 16 nm pore along the long axis of its rod-shaped structure [[125, 126]]. This observation, combined with the localization of Atg2/ATG2 at the ER-IM contact site, led to an intriguing model proposing that Atg2-family proteins act as bridges between the ER and the IM, facilitating the transfer of phospholipids from the ER to the IM and promoting IM expansion (Fig. 5C) [[124-126]]. In 2021, the AlphaFold2 model of Atg2-family proteins further supported the idea of the existence of a hydrophobic pore along the long axis of these proteins (Fig. 5B, right) [[153]]. In 2023, the cryo-EM structure of the ATG2A-WIPI4 complex was determined at a higher resolution of 3.2 Å, confirming the presence of the hydrophobic pore along the rod-shaped architecture, measuring approximately 18–20 nm in length and 1.6–2.4 nm in width [[154]]. The proposed Atg2-mediated IM expansion model explains the observation that autophagosomal membranes in both yeast and mammals contain minimal membrane proteins [[155, 156]]. However, this model faces a critical challenge: while Atg2 can transfer phospholipids to the cytosolic leaflet of the IM, it cannot transfer phospholipids to the luminal leaflet. As a result, there must be additional mechanisms responsible for distributing phospholipids to both leaflets of the IM to support its expansion process.


Atg9 stands out as the only transmembrane protein among the core Atg proteins [[1, 2]]. It is found in vesicles called Atg9 vesicles or tubulovesicular Atg9 reservoirs, which are believed to target the PAS during periods of starvation and act as the initial seed membranes for IM generation [[157, 158]]. However, for a long time, the structure and molecular function of Atg9 remained elusive. In 2020, the advent of cryo-EM allowed for significant progress in understanding the structure of Atg9-family proteins. The structure of plant ATG9 was reported at a resolution of 7.8 Å, which revealed a homotrimeric architecture; however, detailed structural information, such as the number of transmembrane helices, remained unclear due to the low resolution [[159]]. Soon after, the structures of human ATG9A and fission yeast Atg9 were determined at higher resolutions of 2.9–3.4 Å in detergent micelles, amphipols, and nanodiscs (Fig. 5D) [[127, 130, 160]]. These studies demonstrated that the homotrimeric architecture is evolutionarily conserved across Atg9-family proteins. Moreover, the atomic-resolution structures provided valuable insights into the structural details of Atg9-family proteins, revealing that they consist of four transmembrane helices and two laterally membrane-buried helices. The proteins share two transmembrane helices with neighboring protomers within the trimer and possess two distinct pores: one at the center of the trimer, vertically penetrating the membrane, and the other at the center of the protomer, laterally open to the membrane [[127, 130, 160]]. Recent structural studies and molecular dynamics simulations have proposed that the membrane curvature affects the orientation of protomers within the Atg9 trimer, thereby influencing the size of the pore at the trimer center. Notably, in vitro analyses have demonstrated that both yeast Atg9 and human ATG9A possess the ability to scramble fluorescently labeled lipids between the leaflets of liposomes [[127, 130]]. Additionally, these proteins have been shown to scramble PI3P, a phospholipid abundant in the IM [[127]]. Mutations at the pores have been found to inhibit both scramblase activity in vitro and autophagy in cells, indicating that scramblase activity is mediated by these pores and is essential for autophagosome formation [[127, 130]]. These findings have updated the lipid transfer model from the ER to the IM, suggesting that Atg2-family proteins transfer phospholipids from the cytosolic leaflet of the ER to that of the IM, which is then transferred to the luminal leaflet of the IM by Atg9-family proteins, facilitating IM expansion (Fig. 5C) [[127, 129, 130]]. In 2022, cryo-EM analysis of the ATG2A–ATG9A complex at 32 Å resolution revealed that the C-terminal region of the ATG2A monomer directly binds to the ATG9A trimer in a 1 : 3 stoichiometry, suggesting that ATG2A may directly transfer phospholipids to ATG9A for scrambling [[161]]. Subsequently, in 2023, another group performed cryo-EM analysis of the ATG2A–ATG9A complex and identified two distinct complexes in which the N-terminal and the C-terminal regions of ATG2A bind to the ATG9A trimer, respectively [[154]]. Since the pore of ATG2A was found to be located near the lateral pore of ATG9A in the latter complex, the authors proposed that phospholipids could be directly transferred between these two pores. However, further investigations are required to fully understand the collaboration between Atg2 and Atg9 in phospholipid transfer.

Structural view on selective autophagy

During selective autophagy, specific cargos such as mitochondria, ER, protein aggregates/droplets, and invasive bacteria are recognized by the IM and sequestered into the autophagosome [[162]]. Key players in this process are SARs, which interact with both specific cargos and core Atg machinery, linking the cargos to the IM for selective sequestration [[163]].

SAR–Atg8 interaction

Atg8-family proteins play a central role in cargo recognition during selective autophagy. In 2001, the first SAR, Atg19, was identified in yeast, which is responsible for recognizing vacuolar enzymes during the cytoplasm-to-vacuole targeting (Cvt) pathway, a prototype of selective autophagy [[164, 165]]. Atg19 was found to directly interact with Atg8 through its 10 C-terminal residues [[166]]. In mammalian cells, p62/SQSTM1 was identified as the first SAR responsible for selective autophagy of polyubiquitinated protein aggregates in 2005 [[167]]. Subsequent studies in 2007 showed that p62 directly binds to LC3 using the LC3-binding region (LIR) identified at residues 321–342 of p62, with mutational analyses revealing the importance of three consecutive acidic residues and its C-terminal Trp in the LIR for LC3 binding [[168]]. In 2008, the crystal structures of yeast Atg8 complexed with the C-terminal fragment of Atg19 [[33]] and LC3B complexed with the LIR from p62 [[33, 169]] were reported, showing that the Trp–X–X–Leu sequence in both Atg19 and p62 binds to Atg8/LC3B in a similar manner [[33, 168]]. This binding involves forming an intermolecular β-sheet with Atg8 β2 and inserting the side chains of Trp and Leu into the W-site located between α2 and β2 and the L-site located between β2 and α3, respectively (Fig. 6A, left). These structural studies established the concept that this type of sequence is the conserved binding motif for Atg8-family proteins and thus was named the Atg8-family interacting motif (AIM) (Fig. 6A, right) [[170]]. As the LIR in p62 was the first identified fragment for interacting with mammalian Atg8 proteins, the LIR motif, equivalent to AIM, has become widely used among mammalian autophagy researchers [[171]]. Subsequently, numerous structures of Atg8-family proteins complexed with various AIM/LIR sequences were experimentally determined, contributing to the refinement of the AIM/LIR consensus sequence (current consensus is Trp/Tyr/Phe-X-X-Leu/Ile/Val) and the identification of several accessory interactions that enhance binding affinity. For more comprehensive information about AIM/LIR, please refer to review articles dedicated to this motif [[172]].

Details are in the caption following the image
Structural view on selective autophagy (A) Structure of the Atg8–Atg19AIM complex (left, PDB 2ZPN) and summary of the core AIM–Atg8 interaction (right). Atg8 is shown with a surface model, with colors indicating electron static potentials (blue for positive; red for negative). (B) Structure of the FIP200 Claw–CCPG1FIR2 complex (left, PDB 7D0E) and summary of core LIR–Claw interaction (right). The FIP200 Claw is shown with a surface model, with colors indicating electron static potentials (blue for positive; red for negative). LHP and SHG stand for the large hydrophobic pocket and the small hydrophobic groove, respectively. (C) AlphaFold2 structure of p62 (left, AF-Q13501) and Atg32 (right, AF-P40458), with colors representing pLDDT values (red for values < 50, blue for values > 90). Red-colored regions may correspond to intrinsically disordered regions (IDRs). In boxes, experimentally determined structures are shown (PB1, PDB 2KKC; ZZ, PDB 5YP7; UBA, PDB 1Q02; Pseudo-receiver, PDB 5WLP). (D) Crystal structure of the Atg44 octamer (left) and protomer (right) (PDB 7YDO). PE bound to Atg44 is shown with a stick model. Protomer structure is depicted with a surface model, with colors indicating electron static potentials (blue for positive; red for negative).

SAR–Atg11/FIP200 interaction

The interaction between SARs and Atg8 is crucial for the tight association between specific cargo and the IM during selective autophagy. However, this interaction alone is not sufficient for most types of selective autophagy. It is essential for specific cargos to be localized in close proximity to the autophagosome formation sites. In budding yeast, Atg19 was found to directly interact with Atg11, a functional homolog of FIP200 responsible for organizing the “selective PAS,” the autophagosome formation site for selective autophagy [[166]]. Similarly, other SARs in yeast, including Atg32, Atg36, and Atg40, were later shown to directly interact with Atg11 [[173-176]]. In mammals, SARs such as p62, Optineurin, and CCPG1 were shown to interact with FIP200. The binding region for both Atg11 and FIP200 is short, typically containing phosphorylation sites [[104, 177-181]]. In 2021, a structural study was conducted on the Claw domain of FIP200 complexed with the FIR of CCPG1 and Optineurin [[181]], as well as with NAP1 [[182]]. This study provided insights into the interaction mode between FIP200 Claw and FIR: FIR adopts an extended conformation and forms an intermolecular anti-parallel β-sheet with Claw. Additionally, phosphoryl groups and/or acidic side chains of FIR interact with the basic surface of Claw, and two hydrophobic residues of FIR bind to two hydrophobic pockets of Claw. Based on this structural information, the core sequence of the FIR motif was defined as [D/E/phospho-S/phospho-T][I/L/V/W/F/Y]XX[LIV] (Fig. 6B, right) [[181]]. Interestingly, this core sequence is similar to the core sequence of AIM/LIR, indicating that many mammalian SARs may utilize the same region as both AIM/LIR and FIR. However, it remains unclear whether Atg11 also interacts with SARs in a similar manner to the FIP200-FIR interactions. Further investigation is needed to elucidate this aspect of the SAR-Atg11 interactions.

Structural studies on SARs

Selective autophagy receptors typically consist of multiple domains connected by IDRs, making their architecture highly conformationally dynamic and challenging for full-length structural studies. Therefore, researchers have focused on studying individual domains of SARs to understand their structural mechanisms. Examples include the PB1, UBA, and ZZ domains of p62 (Fig. 6C, left box) [[183-185]]; the PB1, UBA, and FW domains of NBR1 [[186-188]]; the SKICH and UBZ domains of TAX1BP1 [[189, 190]]; the UBAN and UBZ domains of Optineurin [[191-193]]; the SKICH and UBZ domains of NDP52/CALCOCO2 [[36, 194]]; the ABD and coiled-coil domains of Atg19 [[195, 196]]; the ABD domain of Atg34 [[195]]; and the pseudo-receiver domain of Atg32 (Fig. 6C, right box) [[197]]. These structural studies have provided valuable insights into the target recognition and self-assembly mechanisms of various SARs during selective autophagy. However, to further understand the mode of action of SARs in selective autophagy, it is necessary to conduct structural studies on the full-length or more natural states of these proteins. Some progress has been made in this direction, such as the use of correlative light and EM to visualize the filamentous assembly state of p62 in cells, which is crucial for the autophagic degradation of p62 cargo [[198]].


The size of mitochondria is larger than that of autophagosomes. Therefore, mitochondria must undergo fragmentation, reducing their size to that of autophagosomes, before being engulfed during mitophagy. However, the known mitochondrial fission factors (Dnm1 in yeast and DNM1L/Drp1 in mammals) have been shown to be dispensable for mitophagy [[199]]. In 2023, a novel mitochondrial fission factor essential for mitophagy was identified in fission and budding yeasts and named Atg44, Mdi1, or mitofissin [[200, 201]]. The crystal structure of Atg44 revealed that its protomer consists of three α-helices, exhibiting an amphipathic overall structure with one side being hydrophobic and the opposite side being hydrophilic (Fig. 6D, right) [[200]]. Atg44 forms a stable octamer in both solution and crystals, utilizing the hydrophobic face and accommodating eight phospholipids within the octamer (Fig. 6D, left). High-speed atomic force microscopy demonstrated that Atg44 binds to membranes with high curvature in an oligomerization state smaller than an octamer, and simulations suggested that Atg44 dimers and tetramers bind to positively curved membranes using the hydrophobic face [[200]]. Although the involvement of Atg44 in membrane fission might result from insertion into the membrane and/or protein crowding on the membrane, further studies are required to unveil the molecular mechanisms of Atg44-mediated mitochondrial fission.

Conclusions and perspectives

Over the last 30 years, a considerable amount of structural information on the core Atg factors has accumulated, and the molecular functions of each factor have been elucidated one after another. X-ray crystallography and NMR have greatly contributed to structure determination, and recent developments in cryo-EM have made structural analysis of larger complexes possible. Furthermore, since the release of AlphaFold2 in 2021 [[19]], it has become possible to predict the static structures of not only full-length autophagy factors but also their complexes with high accuracy although experimental validation work is essential. Going forward, we expect to transition from the era of structure determination to the era of structure utilization. The elucidation of molecular functions, not only in core Atg factors but also in factors involved in selective autophagy, is expected to accelerate using structural information.


This work was supported in part by JSPS KAKENHI Grant Number JP19H05707, by CREST, Japan Science and Technology Agency Grant number JPMJCR20E3, and by grants from the Takeda Science Foundation and from the Chugai Foundation for Innovative Drug Discovery Science.