INTRODUCTION
Macroautophagy (henceforth autophagy) is a conserved cellular degradative process that maintains cellular homeostasis under a variety of stress conditions including starvation, mitochondrial damage, and microbial invasion. One of the major functions of autophagy is to selectively target and degrade various unneeded or dangerous cellular cargoes. Dysfunction of selective autophagy and the consequent accumulation of problematic cargoes contribute to a multitude of human disease (
1). Parkinson’s disease (PD) is one of the clearest examples of an autophagy defect linked to human disease. In
PRKN and
PINK1 patients with PD, the failure to clear damaged mitochondria via mitophagy is thought to contribute to disease due to the progressive loss of dopaminergic neurons in the substantia nigra pars compacta (
2,
3). Selective autophagy proceeds from the recognition of cargoes, formation of autophagy initiation sites, de novo synthesis of a double lipid bilayer termed the phagophore (or isolation membrane), maturation of the phagophore into a closed autophagosome that sequesters cargo, and, lastly, autophagosome-lysosome fusion leading to cargo degradation (
4–
6). A set of core autophagy initiation proteins bridge cargo recognition to isolation membrane biogenesis and elongation into the cup-shaped phagophore (
4,
5). In mammalian autophagy, the Unc-51–like autophagy activating kinase (ULK1) complex, PI3KC3 complex I (PI3KC3-C1), ATG12-5-16L1 autophagy-related protein 8 (ATG8) conjugation machinery, ATG2A/B, and ATG9A are all fundamental to autophagosome formation (
7). How these complexes interact to trigger initiation has been challenging to dissect due to the presence of overlapping and partially redundant activities and the large size and dynamic character of the protein complexes.
The ULK1 complex is composed of four proteins: the ULK1 kinase, ATG13, ATG101, and focal adhesion kinase family interacting protein of 200 kDa (FIP200) (
8–
11). The ULK1 and ULK2 kinases regulate much of the autophagy pathway through phosphorylation events (
12). The nonkinase subunits of the ULK1 complex have numerous scaffolding and bridging roles, some of which act upstream of ULK kinase activity. FIP200 is recruited by autophagic cargo receptors to mark sites of autophagy initiation and scaffolds the assembly of the ULK1 complex (
13–
16). Cargo receptors including p62 and Nuclear domain 10 protein 52 (NDP52) engage with the C-terminal coiled coil and Claw domains of FIP200 (
13–
15). These interactions place the ULK1 complex in an upstream position in many forms of selective autophagy, such that it is thought to be responsible for bridging the rest of the autophagy initiation machinery (
17).
ATG13 translocation to initiation sites is a key early step in both starvation-induced autophagy and mitophagy (
18,
19). The ATG13 and ATG101 subunits of the ULK1 complex form a heteromeric dimer through homologous Hop1/Rev7/Mad2 (HORMA) domains (
Fig. 1A) (
20,
21). HORMA domains contain a “safety belt,” which conformationally regulates binding to protein proteins (
22). ATG101 contains a protruding Trp-Phe (WF) finger motif, which is important for autophagy initiation (
20,
21,
23), but whose precise function is unknown. In addition to a HORMA domain, ATG13 contains a C-terminal intrinsically disordered region (IDR) of ~300 residues. This IDR is responsible for interaction with FIP200 and the ULK1 kinase (
24–
26).
ATG9A is the only known ubiquitously expressed transmembrane protein of the autophagy initiation cascade in mammals (
27). ATG9A forms Golgi-derived vesicles, which are recruited to autophagy initiation sites through a complex trafficking process (
28–
32). Global knockout (KO) of
ATG9A in mice results in impaired autophagosome biogenesis and accumulation of p62 aggregates (
33), and conditional KO in the brain leads to progressive neurodegeneration (
34). Structurally, ATG9A is composed of two distinct domains, a transmembrane domain (TMD; 1–525) and a C-terminal IDR (526–839) (
Fig. 1A). The TMD of ATG9A forms an interlocked trimer and functions as a lipid scramblase (
35–
37). ATG9A distributes incoming endoplasmic reticulum (ER)–synthesized lipids, transported via ATG2 proteins, across the growing phagophore from the outer to the inner leaflet (
38,
39). The IDR region of ATG9A contains sites of regulatory signaling including phosphorylation via TANK-binding kinase 1 (TBK1) and ULK1, ubiquitination, and direct protein-protein interactions (
40–
42). It has been shown that yeast Atg9 is capable of acting as a seed for phagophore initiation (
43), and it seems reasonable to expect that mammalian ATG9A might do the same. To carry out any of their lipid transfer, regulatory, assembly, or putative seeding functions, ATG9A vesicles must first be recruited to sites of autophagy initiation, the focus of the present study.
ATG9A, ATG13, and ATG101 form a multifunctional hub at an early stage of starvation-induced autophagy and mitophagy initiation (
18,
19,
44). A Biotinylation identification (BioID) mass spectroscopy approach revealed that ATG13:ATG101 can recruit ATG9A during p62-dependent autophagy, independent of FIP200 and ULK1 (
45). Deletion of ATG13 or ATG101 led to a mislocalization of ATG9A, leading, in turn, to an accumulation of p62 aggregates identical to the ATG9A KO phenotype (
33,
45). This study highlighted the importance of the ATG9A, ATG13, and ATG101 nexus and focused our attention on this subnode of the autophagy interaction network. Clearance of damaged mitochondria requires both the ULK1 complex and ATG9A to promote efficient mitophagy (
46). During PINK1/Parkin-dependent mitophagy, the ubiquitination of outer mitochondrial membrane proteins promotes recruitment of Optineurin (OPTN), NDP52, and p62 (
47–
49). However, TBK1-phoshorylated OPTN has been reported to directly recruit ATG9A during mitophagy, thereby bridging cargo to ATG9A vesicles in a manner that could potentially bypass the need for ATG13 to recruit ATG9A (
50,
51). This contrasts with the strict dependence of NDP52-mediated selective autophagy on the ULK1 complex for initiation, which led us to focus on the role of the ATG9A:ATG13:ATG101 complex in NDP52-dependent PINK1/Parkin mitophagy. Here, we have identified the binding interface between ATG9A, ATG13, and ATG101. We term the region of the extreme C terminus of ATG9A that binds to the ATG13:ATG101 HORMA dimer the “HORMA dimer–interacting region” (HDIR). We determined the structure of the human ATG9A HDIR bound to the ATG13:ATG101 HORMA dimer and confirmed that the interface functions in NDP52-dependent mitophagy.
DISCUSSION
Understanding the interactions responsible for autophagy initiation is central to efforts to therapeutically modulate bulk and selective autophagy pathways and to understand their functioning at a fundamental level. Here, we determined the structure of the ATG9A HDIR bound to the ATG13:ATG101 HORMA dimer (
Fig. 2C). As a central part of the autophagy initiating the ULK1 complex, the HORMA dimer is considered a key element, enabling the ULK1 complex to recruit ATG9A vesicles for autophagy initiation (
18,
19,
44). The structure helps bring an understanding of ATG9A binding in the context of other structural information on the HORMA dimer and HORMA domains in general. The HDIR site overlaps a hydrophobic binding site for the small-molecule benzamidine that was identified in the apo structure of the human HORMA dimer (
20), explaining the normal function of this site. Other HORMA domain proteins function as protein interaction hubs regulated by conformational changes in the safety belt region (
22). This posed the question whether interactions of the ATG13:ATG101 dimer could be regulated by conformational switching as seen for Mad2 and other HORMA domains (
22). The interaction between the ATG9A HDIR and ATG13:ATG101 is distal to and thus apparently independent of the safety belt region.
Our biochemical experiments confirmed that the ATG13 residues that bind the ATG9A HDIR in the structure are required for full function. We noticed that none of these mutations completely abolishes mitophagy (
Fig. 4, A and B). On the other hand, the deletion of the entire ATG13 HORMA domain has a nearly complete loss of function, as seen by Kannangara
et al. (
45). In our Halo assay, mutation of ATG13 W50D, which disrupted the HDIR interaction in vitro (
Fig. 3D), had no in cellulo defect (
Fig. 4, A and B). This discrepancy may arise from technical differences in the assays or it may point toward an unknown function of W50D. Either way, further work is required to fully elucidate the function of this site on ATG13. In addition, the HORMA dimer has additional functions distinct from the ATG9A tail binding pocket, which may play a role. We confirmed that the essential WF motif of ATG101 (
21) is not involved in ATG9A binding. An intact ATG13 HORMA dimer is required for ATG101 to associate with the rest of the ULK1 complex via the ATG13 IDR (
20). The unaccounted-for residual function of the HORMA dimer seems likely related to the role of the WF finger, which will be important to establish.
Human ATG9A contains a long C-terminal IDR, whereas budding yeast Atg9 contains an extensive N-terminal IDR. Sequence alignment of the ATG9A tail shows a strong conservation across most of the Opisthokonta (
Fig. 2A) except budding yeast, which lacks the HDIR. In addition to ATG9A, mammals express the homologous ATG9B in a few tissues; however, ATG9B does not contain the HDIR motif. The physiology of human ATG9B is relatively little explored; thus, the importance of this difference is currently hard to interpret. The mammalian
S. cerevisiae, as well as related thermophilic yeasts such as
Kluyveromyces lactis and
Lachancea thermotolerans that have been used as model systems in autophagy, has distinct initiation machinery when compared to other Opisthokonta (
7). The ULK1 (human) kinase complexes contain ATG13 along with its constituent binding partner ATG101. However, budding yeast entirely lacks an ATG101 ortholog, and its functions are apparently replaced by the single HORMA domain in yeast Atg13 (
56). Thus, although budding yeast Atg9 binds to the HORMA domain of Atg13 (
54), the mammalian sequence motif and structural interactions do not seem to exist in yeast.
We combined biochemical analysis and predictive modeling to generate a fusion ATG9A-ATG101 protein for experimental structure determination. Protein structure prediction has rapidly advanced via deep learning algorithms such as AF2 (
53,
57). Predictive modeling can determine the binding pockets of polypeptide chains with an RMSD of <2 Å (
58). However, validation of structural models remains a critical step in guiding our understanding of biological mechanisms. Minimizing the amount of primary sequence during modeling can greatly improve the accuracy of predicted interfaces. Our truncation data limited the primary sequence for the ATG9A component to be modeled on the ATG13:ATG101 dimer, making AF2 prediction of the complex feasible. Determining which of the two modeled binding sites was correct relied on experimental information. The potential for predictive models combined with in vitro biochemical analysis to correctly determine protein-protein interfaces will continue to grow as more methods are explored and structural studies are guided by predictive models. In our study, we found that experimental interaction mapping provided essential constraints before AF2 prediction of the complex, and downstream model validation by mutational and structural analysis was important to establish confidence in the model.
Ubiquitinated autophagy cargoes recruit receptors (OPTN, p62, NBR1, NDP52, and TAX1BP1) and some bind directly to the ULK1 complex via FIP200 (
13–
15). This includes NDP52 recognition of Parkin ubiquitination marks on mitochondria damaged by uncoupling agents (
13). These interactions drive clustering and allosteric activation (
16) of the ULK1 complex to promote the autophagy initiation cascade. While the micromolar affinity of the ATG9A HDIR:ATG13:ATG101 interaction is moderate, it is easy to see how the clustering of multiple ULK1 complexes on an extended ubiquitinated platform such as a damaged mitochondrion would contribute avidity to recruitment of ATG9A trimers. The 1:1 stoichiometry established here implies that a single ATG9A trimer can bind to three ULK1 complexes presented in a cluster on a cargo substrate (
Fig. 4C). The interaction between the C-terminal IDRs of the ATG13 molecules and the N-terminal crescent domains of the FIP200 molecules would position the massive C-terminal coiled coil domains of FIP200 distal to the ATG9A vesicles. On the basis of the dimensions of FIP200 (
24), this would allow for fly casting for ATG9A vesicles as far as roughly 35 nm from sites of clustered ubiquitin.
While our model presents a satisfying and plausible model for ATG9A vesicle recruitment downstream of the ULK1 complex in autophagy initiation, this is almost certainly not the complete mechanism or the only mechanism. Further mechanistic research is required to elucidate the role of the ATG9A HDIR motif on both selective and bulk autophagy pathways outside of mitophagy. Our attempts to perform KO and rescue experiments of ATG9A failed due to known mislocalization of ATG9A upon overexpression. It remains to be determined whether the ATG9A HDIR is the sole interactor at the ATG13-ATG101 interface. Kannangara
et al. (
45) found that ULK1 could still interact with ATG9A in the absence of ATG13 as measured by BioID, although this was insufficient to support normal levels of p62 autophagy. Thus, at least one additional point of contact between ATG9A and the ULK1 complex is likely to exist, at least in some forms of autophagy. A more significant departure from this mechanism is suggested by the finding that OPTN can recruit ATG9A directly, bypassing the need for the ATG9A-binding role of the ULK1 complex (
50). The physiological rationale for these various pathways remains largely unknown, but the distinct role of different cargo receptors in neurodegeneration suggests that it will be important to better understand. Elucidating the molecular details of distinct initiation mechanisms in various “flavors” of mitophagy should provide windows into disease mechanism and ultimately therapy.
Acknowledgments
We thank members of the Hurley Lab, D. Fracchiolla, and others in Aligning Science Across Parkinson’s (ASAP) Team mito911 for advice and discussions. We thank C. Smith and L. Dunn at SSRL beamline BL12-2 for assistance with data collection. We thank the Mizushima Laboratory for sharing their ATG13 KO/penta KO cell line. Use of the Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences under contract no. DE-AC02-76SF00515. The SSRL Structural Molecular Biology Program is supported by the DOE Office of Biological and Environmental Research and by the National Institutes of Health (NIH), National Institute of General Medical Sciences (NIGMS) (P30GM133894). The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of NIGMS or NIH.
Funding: This work was supported by the Michael J. Fox Foundation for Parkinson’s Research (MJFF) and ASAP initiative. MJFF administers the grant ASAP-000350 (to J.H.H. and M.L.) on behalf of ASAP and itself. This work was also supported by NIGMS, NIH, R01 GM111730 (J.H.H.); National Health and Medical Research Council (NHMRC) GNT1106471 (M.L.); and Australian Research Council (ARC) Discovery Project DP200100347 (M.L.).
Author contributions: Conceptualization: A.L.Y. and J.H.H. Methodology: X.R., T.N.N., W.K.L., and C.Z.B. Investigation: X.R., T.N.N., W.K.L., and C.Z.B. Visualization: X.R., T.N.N., W.K.L., and A.L.Y. Supervision: M.L., A.L.Y., and J.H.H. Writing—original draft: A.L.Y. and J.H.H. Writing—review and editing: All authors.
Competing interests: J.H.H. is a cofounder of Casma Therapeutics and receives research funding from Casma Therapeutics, Genentech, and Hoffmann-La Roche. M.L. is a member of the Scientific Advisory Board of Automera. The other authors declare that they have no competing interests.
Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Coordinates and structure factors have been deposited in the Protein Data Bank under accession code PDB 8DO8. Protocols have been deposited in protocols.io. Plasmids developed for this study were deposited at
Addgene.org. Raw data files for gel scans have been uploaded to Zenodo (DOI:
10.5281/zenodo.7632198 ).