A GABARAP−PtdIns3K-C1 positive feedback loop at the heart of the phagophore nucleation Authors: Antoine N Dessus, Yohei Ohashi, Maxime Bourguet, Tomos E Morgan, Anastasia Nunez, Maria Manifava, Nicholas T Ktistakis, Roger L Williams Published: 2026 ABSTRACT AbstractIn mammalian cells, autophagosomes can reach diameters of over 1000 nm within 30 min after triggering starvation, but how such substantial amounts of membranes can be synthesized remains elusive. The phagophore initiation needs the lipid kinase PIK3C3-Complex 1 (PtdIns3K-C1), which produces phosphatidylinositol-3-phosphate (PtdIns3P). PtdIns3P recruits WIPI2 that facilitates lipidation of mammalian ATG8 (mATG8) family proteins on phagophores. Here we show that recombinant membrane-coupled GABARAP binds to and potently activates PtdIns3K-C1. By a combination of cryo-electron microscopy, structural mass spectrometry, activity assays and mutagenesis, we show that GABARAP activates PtdIns3K-C1 through two binding sites. We propose that once GABARAP is indirectly recruited by PtdIns3P generated by basal activity of PtdIns3K-C1, a positive feedback loop is formed where PtdIns3K-C1 interacts with GABARAP and becomes activated to produce more PtdIns3P, thereby further stimulating GABARAP lipidation. This mechanism would be central for autophagosome biogenesis, where enlarged membranes need to be rapidly synthesized.Subject terms: Cryoelectron microscopy, Macroautophagy, Kinases Introduction Autophagy defines pathways where cytoplasmic material too large to be degraded by the proteasome, like protein aggregates and organelles, are targeted instead to the lysosomes for degradation. This includes microautophagy, chaperone-mediated autophagy and macroautophagy (hereafter referred to as autophagy)1–3, and defects in autophagy have been linked to pathologies including cancer, neurodegenerative diseases, and inflammatory and immune disorders4–7. Biogenesis of autophagosomes involves nucleation, expansion, closure of membranes (phagophores or isolation membranes) that sequester cargos, followed by their fusion with lysosomes8. Autophagy is initiated at the endoplasmic reticulum (ER) where stress signals are integrated by the unc-51 like kinase complex ULK1. This complex is co-recruited to membranes with the class III PI3K complex 1 (hereafter referred to as PtdIns3K-C1), which is composed of the PIK3C3 lipid kinase, the PIK3R4 pseudokinase, ATG14 (BAKOR) and BECN1 proteins9–11. PIK3C3 and PIK3R4 constitute the kinase arm of the complex whereas BECN1 and ATG14 form its adapter arm, giving a characteristic V-shape to PtdIns3K-C112–14. PtdIns3K-C1 phosphorylates phosphatidylinositol (PtdIns) to yield phosphatidylinositol-3-phosphate (PtdIns3P), at a low basal activity that can be increased by the small GTPase RAB1A, which recruits PtdIns3K-C1 to membranes and activates it15. PtdIns3P is essential for autophagosome generation because it recruits its effector WIPI proteins16. WIPI2d is known to interact with PtdIns3K-C1, and this interaction is believed to produce positive feedback17. Although it was shown that WIPI2d on PtdIns3P-containing membranes enhances the recruitment of PtdIns3K-C1 to membranes17, it remains unclear whether this results in an increase in the PtdIns3K-C1 activity. WIPI2d then recruits the ATG5-ATG12–ATG16L1 complex to autophagosomal membranes18. This complex acts as an E3 ligase involved in a ubiquitin-like enzymatic cascade along with the E1 ATG7 and the E2 ATG3. The substrates for this enzymatic cascade are Atg8 (yeast) or the mammalian-ATG8 proteins (mATG8s), which include LC3A, LC3B, LC3C, GABARAP, GABARAPL1 and GABARAPL219,20. The Atg8/mATG8s become covalently coupled to the phagophore’s lipids by the E3 complex, allowing the recruitment of various receptors, which in turn recruit the autophagic cargo to be degraded21. The mATG8s thus bind to a plethora of proteins, which insert their LC3-interacting regions (LIRs) into the LIR-docking sites (LDSs) of the mATG8s. The LIR is a four amino acid sequence conserved among mATG8s binders, consisting of the motif Θ-X-X-Γ (were Θ is an aromatic residue, X any residue and Γ a small aliphatic residue) interacting with hydrophobic pockets in the LDS22. Recently, such LIR-LDS interactions were found in peptides from three subunits of PtdIns3K-C1: PIK3C3, BECN1 and ATG14, with a particularly strong affinity for the mATG8 GABARAP23. However, no binding of mATG8s to the full PtdIns3K-C1 assembly was attempted, and the functional implications on PtdIns3P production remained unknown. Remarkably, during autophagy in mammalian cells, within a brief time24,25, one cell generates around 100 autophagosomes26, each 500–1500 nm in size27. To achieve this, a large amount of lipids is necessary, and the membranes need to be rapidly expanded and closed to engulf cargos and deliver them to lysosomes. Atg8/mATG8 proteins have roles in regulating autophagosome size28, elongation of phagophore membranes29,30, and autophagosome closure31. To execute these autophagic events efficiently, it is conceivable that many Atg8/mATG8 proteins need to be lipidated on the membranes in a brief time. Here we show that the acute inhibition of mATG8 lipidation does not cause the accumulation of the PtdIns3P effector WIPI2 relative to FIP200 in the cell, suggesting a reduction of PtdIns3P synthesis by PtdIns3K-C1. Using a combination of in vitro activity assays, cryo-electron microscopy (cryo-EM), structural mass spectrometry (MS), and mutagenesis we confirm that lipidated mATG8s activate PtdIns3K-C1, and we identify two binding sites for the mATG8 GABARAP on PtdIns3K-C1. These two sites, namely the N site and the C site, both contribute to the activation of PtdIns3K-C1 by lipidated GABARAP on membranes. Our results suggest a positive feedback loop happens in cells between GABARAP and PtdIns3K-C1, which could be critical for rapid PtdIns3P production and associated mATG8 lipidation on the growing phagophore. Results Translocation of PtdIns3P effectors is sensitive to lipidation of mATG8s We first examined the effect of the lack of mATG8 lipidation on its upstream PtdIns3P synthesis in the cell. A previous study showed that even in rich media, PtdIns3P-binding proteins WIPI1 and DFCP1 accumulated in MEF cells lacking the Atg5 gene that encodes an E3 subunit of the mATG8 lipidation machinery32. These results could have arisen from indirect effects over a long period of time in the Atg5 knockout cells. Because autophagy occurs in a brief time, we wanted to observe the acute effect of inhibition of mATG8 lipidation activity. To achieve this, we used an inhibitor of ATG7 which completely eliminates the lipidation of all mATG8 family members33, preventing their translocation to punctate structures upon autophagy induction (Fig. 1a, shown for LC3). Using this inhibitor, we determined whether translocation of early autophagy proteins to early autophagy structures was affected by the absence of lipidated mATG8, using as relevant markers FIP200 (a member of the ULK1 complex, which translocates first to autophagy puncta) and WIPI2 (an immediate effector of PtdIns3P, which requires activation of PtdIns3K-C1 for its translocation to autophagy puncta). WIPI2 puncta number and intensity were thus used here as estimates of total PtdIns3P levels. The number of FIP200 and WIPI2 puncta upon induction of autophagy with PP242 in the presence or absence of ATG7 activity was similar (Fig. 1b, c), suggesting the formation of the pre-autophagosomal structures was not affected by the loss of mATG8 lipidation. Interestingly, the intensity of FIP200 puncta was increased upon ATG7 inhibition (Fig. 1b small inserts and d), indicating that an accumulation of FIP200 was induced by stopping the downstream autophagy flux. However, the intensity of WIPI2 puncta was not increased upon ATG7 inhibition. This result contrasts to the WIPI accumulation seen in cells when the Atg8 lipidation is permanently deleted32, or when the autophagy flux is stopped downstream of mATG8s18,34. This suggests that the initial translocation of the early autophagy machinery to autophagy structures, which takes place upstream of mATG8 engagement, can occur in the absence of lipidated mATG8. However, the continuous translocation of the PtdIns3P effectors seemed to be affected by a transient loss of lipidated mATG8s, as WIPI2 does not further accumulate. Considering these results and a previously characterized interaction between peptides derived from PtdIns3K-C1 subunits and mATG8s23, we hypothesized that the lipidated mATG8s may directly affect the PtdIns3P production by PtdIns3K-C1 in a feedback loop. We therefore carried out the in vitro characterization of the interaction between these proteins. PtdIns3P production is reduced by inhibiting its downstream mATG8 lipidation machinery in cells. a HEK293 cells were treated for 60 min with 1 μM of the mTOR inhibitor PP242 in the presence or absence of an ATG7 inhibitor (inh, used as a pretreatment for 30 min) at 5 μM final concentration. The cells were then stained for immunofluorescence with antibodies to LC3. Representative micrographs from 10 repeats of the experiment. b HEK293 cells were treated as in panel a, but were double-stained with antibodies to endogenous FIP200 and endogenous WIPI2 as shown. Image acquisition and any intensity corrections were done to the same extent for both channels. c, d Quantitation of images shown in panel b. Five fields of each treatment were imaged, and the number (c) and intensity (d) of puncta per cell were estimated using ImageJ. Averages for n fields ± SD (c) or cells (d) were compared in a two-way ANOVA with Tukey’s correction for multitesting. GABARAP and other lipidated mATG8s bind to and activate PtdIns3K-C1 Three LIRs have been previously characterized on the PIK3C3, ATG14 and BECN1 subunits of PtdIns3K-C1, with a selectivity for the mATG8s GABARAP and GABARAPL123. However, this study used peptides from LIRs of PIK3C3, ATG14, and BECN1 to characterize their interaction with mATG8s, instead of the full PtdIns3K-C1 assembly. To determine whether the proposed binding between PtdIns3K-C1 subunits and mATG8s does happen within the full complex, we immobilized recombinant mCherry−PtdIns3K-C1 on beads and incubated it with mCerulean-fusions of each of the six human recombinant mATG8s (Fig. 2a). With a confocal microscope we could see clear binding of all mATG8s to PtdIns3K-C1 and confirmed stronger binding of PtdIns3K-C1 to the mATG8s GABARAP and GABARAPL1 (Fig. 2b). Titration gave an apparent Kd of 1.2 µM for the interaction between PtdIns3K-C1 and GABARAP (Fig. 2c, Supplementary Fig. 1a). This binding was not caused by non-specific interactions with mCerulean or mCherry (Supplementary Fig. 1b). Since mATG8s are covalently coupled to lipids on the growing phagophore in cells, we reconstituted this in vitro using recombinant ATG7 (E1), ATG3 (E2), the ATG12-ATG5–ATG16L1 (E3) complex17, and all six human mCerulean-mATG8 proteins on giant unilamellar vesicles (GUVs) (Fig. 2d). This successfully immobilized all mCerulean-mATG8s, but not mCerulean, on GUVs only in the presence of ATP and E1-E2 − E3 proteins (Fig. 2e left, mCerulean channel, Supplementary Fig. 1c). These vesicles were incubated with PtdIns3K-C1 to measure the efficiency of the PtdIns3P production in an endpoint assay, using AF647-p40phox PX as a product fluorescent reporter. Interestingly, all lipidated mATG8s except LC3B significantly increased the production of PtdIns3P (Fig. 2e right). GABARAP and GABARAPL1 were the most potent activators of PtdIns3K-C1, which correlated with them having the strongest affinity for PtdIns3K-C1 compared to the other mATG8s (Fig. 2b)23. Because GABARAP showed the strongest binding and the highest PtdIns3K-C1 activation, we examined its activation mechanism in a time course experiment. To simplify the lipidation system, we immobilized GABARAP on GUVs by incubating a cysteine-introduced GABARAP mutant (L117C) with DOPE-maleimide−containing GUVs (Fig. 2f). We obtained a clear activation of PtdIns3K-C1 by maleimide-lipidated GABARAP, to the same extent as the enzymatic lipidation in endpoint assays (Supplementary Fig. 1d), whereas soluble GABARAP (not membrane-bound) was not able to activate PtdIns3K-C1 (Supplementary Fig. 1e). The time course experiment showed that membrane-anchored GABARAP dramatically increased the initial rate up to 14-fold (Fig. 2g). Taken together, these results suggest that the binding of PtdIns3K-C1 to the lipidated mATG8s results in PtdIns3K-C1 activation on membranes. For the rest of this study, we focused on GABARAP as a model of an mATG8 able to activate PtdIns3K-C1, considering its high affinity and activation potency for the complex among all six mATG8s. mATG8s bind to and activate PtdIns3K-C1. a Design of the bead-based binding assay. RFP-trap agarose beads were coated with mCherry-tagged PtdIns3K-C1, incubated with different mCerulean (mCer)-tagged mATG8 proteins and imaged by confocal microscopy. b Quantification of the mCerulean (mCer)-tagged mATG8 proteins binding to PtdIns3K-C1. The plot shows the mean of n beads ± SD and is representative of three independent repeats of the experiment. Means were compared to the mCerulean control with a one-way ANOVA with Tukey’s correction for multitesting. c Quantification of the interaction between PtdIns3K-C1 and different concentrations of mCerulean-GABARAP. Top: The plot shows the mean of 29 beads ± SD and is representative of three independent repeats of the experiment (Supplementary Fig. 1a). Bottom: Representative confocal micrographs of the experiment in the top panel. Scale bars: 50 µm. d In vitro reconstitution of the mATG8 enzymatic lipidation. Vesicles were incubated with ATG7 (E1), ATG3 (E2), ATG12-ATG5 − ATG16L1 (E3), different mATG8s and ATP. e Activation of PtdIns3K-C1 by lipidated mCerulean-mATG8s on GUVs. E1-E2−E3 represent the lipidation enzymes listed above. Left: Representative confocal micrographs of GUVs, showing Lissamine-Rhodamine (lipids), mCerulean (mATG8) and AF647 (PtdIns3P) channels. Scale bars: 5 µm. Right: Average of the AF647-PX fluorescence intensity in arbitrary units for n GUVs ± SD. Means were compared to the PtdIns3K-C1 only control with a two-way ANOVA with Tukey’s correction for multitesting. f Chemical lipidation of GABARAP using a maleimide reaction. Cys = cysteine. g GUV-based time course PtdIns3K-C1 activity assay with or without maleimide-lipidated mCerulean-GABARAP, for one representative experiment of 21 repeats. Left: PtdIns3P production over time expressed in the mean of AF647-PX fluorescence ± SD for n GUVs (in arbitrary units). Middle: Representative micrographs of GUVs at the end of the reaction in the AF647 channel. Scale bars: 5 µm. Right: Initial rates of the reaction curves in the left panel, expressed in the mean of AF647-PX fluorescence intensity/min for n GUVs ± SD. The means were significantly different in a two-tailed unpaired t-test. GABARAP binds PtdIns3K-C1 through an atypical bipartite interaction To gain insight into which sites are responsible for the GABARAP interaction in the full complex assembly, we solved a 3.6 Å cryo-EM structure of PtdIns3K-C1 with the MIT domain of NRBF2 bound to soluble GABARAP (Fig. 3a). A beads-based interaction assay confirmed that NRBF2 MIT does not affect the binding of GABARAP to PtdIns3K-C1 (Supplementary Fig. 1b). We observed an additional density at the N-termini of the ATG14 and BECN1 subunits that clearly fitted one GABARAP molecule (Fig. 3a, b). Two interaction sub-sites with GABARAP could be distinguished: one on ATG14 (NA), and one on BECN1 (NB), both NA and NB sites forming the GABARAP-binding N site (Fig. 3c). The NB site showed density for the LIR of BECN1 97-FTLI-100 contacting the LDS formed by hydrophobic pockets (HP) HP1 and HP2 in GABARAP, and we built a model for the interaction guided by the previously described crystal structure of GABARAP bound to the BECN1 LIR (Fig. 3d)23. Whereas the NB site forms a canonical LIR-LDS interaction with GABARAP, the contacts we saw between GABARAP and the NA site have no precedent. In the presence of GABARAP, the ATG14 N-terminus showed an additional density compared to the previous PtdIns3K-C1 structure35, suggesting its interaction with GABARAP stabilized this flexible region. In the NA site, we saw ATG14 residues V38 and A39 making clear hydrophobic contacts with the GABARAP ɑ2-β1 loop residues K24 and P26, respectively. Although LIR peptide/LDS interactions have been well characterized36, this GABARAP ɑ2-β1 loop-mediated interaction with an mATG8 has not, to the best of our knowledge, been described. Two CXXC zinc fingers could be modeled in the vicinity of the NA site (Fig. 3d), one involving BECN1 C137 and C140 and ATG14 C43 and C46, and another with BECN1 C18 and C21 and ATG14 C55 and C58. Interestingly, these BECN1 cysteines were reported in PtdIns3K-C2 to form an intramolecular zinc finger37 but are seen here in PtdIns3K-C1 in two intermolecular zinc fingers. Our results also show that FIP200 is not required for the zinc fingers in the N-termini of BECN1 and ATG14 to be ordered, contrary to a previous claim38. For the other two previously reported GABARAP binding sites in PtdIns3K-C1 subunits23, the ATG14 LIR 435-WENL-438 was not observed in our density as it belongs to a long flexible region from residues 402 to the C-terminus of ATG14 that cannot be resolved in our structure (see below); and the PIK3C3 LIR 250-FELV-253 was seen inserted into a pocket of PIK3R4 residues including R46, N79, Q88, K89 and E334 that likely hold the LIR and thus prevent any binding to an mATG8 (Fig. 3e). Taken together, our cryo-EM structure revealed that GABARAP binds PtdIns3K-C1 at the N-termini of ATG14 and BECN1. The BECN1 LIR and residues V38 and A39 of the ATG14 N-terminus are respectively engaging the LDS of GABARAP and its ɑ2-β1 loop in a newly discovered bipartite interaction. PtdIns3K-C1 binds GABARAP via an atypical bipartite interaction. a Cryo-EM composite density map of PtdIns3K-C1 − NBRF2 MIT domain bound to GABARAP (main conformation, EMD-56357). A composite map from 5 local refinements was built at 3.58 Å resolution (resolution of the consensus map). b Atomic model of PtdIns3K-C1 − NRBF2 MIT domain bound to GABARAP (main conformation, PDB 9TW2). c Close-up view of the bipartite interaction between PtdIns3K-C1 N site and GABARAP. GABARAP is represented as a surface, and its interactions with ATG14 37-40 and BECN1 97-100 are defined and labeled as NATG14 (NA) and NBECN1 (NB) sites, respectively, both constituting the N site. The NB site consists of the LIR of BECN1, interacting with the LDS of GABARAP. d Close-up view of the residues involved in GABARAP binding in NA and NB sites, and neighboring zinc fingers. The interaction between PtdIns3K-C1 and GABARAP involves ATG14 residues 37 to 40 (NA), BECN1 residues 97 to 100 (NB), and GABARAP residues 24 and 26 on the ɑ2-β1 loop. Helix ɑ2 and β-strand β1 are indicated on the GABARAP model. Two CXXC zinc fingers are seen, one Zn2+ ion being coordinated by BECN1 C18 and C21 and ATG14 C55 and C58, and another Zn2+ by BECN1 C137 and C140 and ATG14 C43 and C46. e The PIK3C3 LIR is not accessible to GABARAP due to extensive contacts with PIK3R4. Close-up view of the PIK3C3 250-253 LIR in the cryo-EM structure, showing PIK3C3:PIK3R4 contacts between residues pairs E251:R46, L252:Q88, L252:K89 and K254:Q88, respectively. Overall, the PtdIns3K-C1 structure adopted a similar conformation to the one previously described for PtdIns3K-C1 bound to RAB1A35. However, a focused 3D classification on the adapter arm of PtdIns3K-C1 identified a minor class of particles (98,630 out of 403,177) that showed a translation of up to 7 Å of the end of the adapter arm made of the BARA domain of BECN1 and the C-terminal domain of ATG14 (Fig. 4a). We described this translation of the adapter arm as the ‘alternative’ conformation, in opposition to the ‘main’ conformation of PtdIns3K-C1 adapter arm. This alternative conformation was not observed in the published RAB1A-bound PtdIns3K-C1 structure35 and could be induced by GABARAP to facilitate membrane binding, allowing the activation of the complex. We could not see any class of particles resembling the ‘active’ conformation as described in Cook et al35. (PDB 9MHH), and this was also the case for another cryo-EM dataset we collected with PtdIns3K-C1 in the presence of the NRBF2 MIT domain and ADP-MgF3 as an ATP transition-state mimic (Fig. 4b)39. However, in this nucleotide-containing dataset, we clearly saw a distinct ADP density in the ATP-binding pocket, along with a Mg2+ ion coordinating the phosphates of the ADP and PIK3C3 residues N748 and D761. PIK3C3 residues from the ATP-loop (P-loop, S614), hinge (F684), the catalytic loop (D743, R744, H745) and the activation loop (D761) are all visible in this structure, as previously reported for other PI3K-family members40,41. To compare models of the kinase domains from datasets in the presence and absence of nucleotide, we aligned them on the PIK3C3 C-lobe (686–887), and this gave a root mean square deviation (RMSD) of 0.77 Å and an angle of 0.68° between the N-lobes (553–682). Remarkably, the conformation of the PIK3C3 kinase is identical in both our datasets with or without a nucleotide, similarly to what was previously seen in DmVps34 with or without inhibitors42. Structural features of the activated PtdIns3K-C1. a PtdIns3K-C1 adapter arm conformation change induced by GABARAP. Two classes were obtained in a 3D focused classification on the adapter arm of PtdIns3K-C1: the ‘main’ conformation class 1 (white) and the ‘alternative’ conformation class 2 (blue). Black and red markers (respectively) were positioned on the structures to estimate the distances of the shift. b Left and middle: Cryo-EM density (EMD-56458) and model (PDB 9TZ3) for PtdIns3K-C1 − NRBF2 MIT domain bound to ADP-MgF3, with a close-up view on the nucleotide binding pocket. PIK3C3 ATP-loop (P-loop), hinge, catalytic loop and activation loop are visible around the ADP nucleotide. Mg2+ is coordinated by PIK3C3 residues N748 and D761. Right: PIK3C3 kinase domain organization with the ATP-loop, hinge, catalytic loop and activation loop. Identification of another GABARAP binding site with structural mass spectrometry Because the ATG14 LIR23 is part of a flexible region that we did not observe by cryo-EM, we carried out a hydrogen/deuterium exchange coupled to mass spectrometry (HDX-MS) as an orthogonal approach to cryo-EM to investigate the interaction between PtdIns3K-C1 and GABARAP in solution. Upon comparing the exchange profiles between PtdIns3K-C1 in the presence of GABARAP and a control without GABARAP, two main regions in PtdIns3K-C1 showed significant HDX changes upon GABARAP binding (Fig. 5, Supplementary Fig. 2a). The first one was localized in the C-termini of BECN1 and ATG14 (hereafter called the C site), where strong protections were observed for residues spanning the ATG14 LIR (residues 429−443) and the BECN1 BARA domain (residues 278−285). The second region involved the GABARAP-binding N site of PtdIns3K-C1 that we characterized by cryo-EM, with the NA site showing protection (ATG14 residues 39−48), while the NB LIR site was not covered in our HDX-MS dataset. Additionally, we observed protection in the N-termini of BECN1 (residues 19−31 and 125−139) and ATG14 (residues 51−56), close to the N site. Interestingly, these regions and the ATG14 39−48 region are involved in the intermolecular zinc fingers identified in cryo-EM, suggesting the potential involvement of these zinc fingers in the stabilization of the interaction with GABARAP. Identification of two GABARAP-binding sites on PtdIns3K-C1 by a combination of structural approaches. HDX changes (relative fractional uptake difference [PtdIns3K-C1_GABARAP − PtdIns3K-C1]) and identified inter-crosslinked sites between GABARAP and PtdIns3K-C1 are shown on the PtdIns3K-C1 − GABARAP structure (PDB 9TW2). Strong ( ≤ −10%) and light ( ≤ −5%) HDX protections are shown in dark and light blue, respectively, and inter-crosslinked sites are represented by dotted lines. Two GABARAP-binding sites were identified in PtdIns3K-C1, the N site with ATG14 and BECN1 N-termini, and the C site spanning the ATG14 LIR motif and BECN1 BARA domain, both sites showing protected regions and inter-crosslinked sites. The two Zn2+ ions of the N-site zinc fingers are represented as spheres. HDX-MS difference plots and crosslinking-MS maps are shown in Supplementary Fig. 2. Next, we used crosslinking coupled to mass spectrometry (XL-MS) to identify spatial proximities between PtdIns3K-C1 and GABARAP, by using succinimidyl-ester diazirine (SDA). Firstly, both HDX-MS and XL-MS suggest that PtdIns3K-C1 has a binding site involving the ATG14 LIR, the C-site. Crosslinks were identified in the C site involving GABARAP residue K46 with ATG14 LIR residue E436 and PIK3R4 residues Y1290 and Y1291 (Fig. 5, Supplementary Fig. 2b). Interestingly, ATG14 residue E436 is part of the protected region identified in HDX-MS upon GABARAP binding. Secondly, crosslinks also were identified in the N site, involving residues D45 and K46 of GABARAP and ATG14 residues 36 to 40 of the NA site. This site showed protection in our HDX-MS data and directly involved in GABARAP binding in cryo-EM. To ensure that both the N site and the C site can bind GABARAP when it is lipidated, we conducted another crosslinking experiment with GABARAP lipidated on LUVs using maleimide coupling (Fig. 2f, Supplementary Fig. 3). Crosslinks were identified in both the C site and the N site of PtdIns3K-C1, confirming these two regions can bind GABARAP when it is covalently coupled to membranes. Altogether, our data suggest that both the N site (made of NA and NB sites) and the C site (made of the ATG14 LIR) can bind GABARAP, either as a soluble protein or in its lipidated form. Mutagenesis of the C site and the N site to disrupt GABARAP binding To examine the respective roles of the N and C sites on the binding to PtdIns3K-C1, we mutated residues involved in the interaction with GABARAP. The NB and C sites are canonical LIRs, for which the mATG8-binding disrupting mutations are well characterized (the Θ-X-X-Γ LIR motif mutated to A-X-X-A)23,43. For the C site, we thus mutated 435-WENL-438 into 435-AENA-438 (hereafter called Cmut), or we deleted the whole BATS domain carrying this region (∆C) for the bead-binding assays (see below). For the NB site, we mutated 97-FTLI-100 into 97-ATLA-100 (NBmut). To disrupt the hydrophobic contacts of GABARAP with the NA site, we deleted the whole ATG14 YVAV motif (residues 37 to 40, ∆NA). We first conducted bead-binding assays using ATG14-BECN1 heterodimers carrying these mutants (Fig. 6a), since all the binding sites we identified are only in these two subunits out of four in PtdIns3K-C1. For each heterodimer, a Kd between mCerulean-GABARAP and mCherry-ATG14−BECN1 was estimated (Fig. 6b, Supplementary Fig. 4a). We only tested the N site mutants in combination with the ∆C mutant, as the C site has the strongest affinity for GABARAP and thus masked the effect on the other sites23. While it appeared that mutating any site (C, NA or NB) affected the binding to GABARAP, the strongest effects were observed with the ∆C NBmut and ∆C ∆NA NBmut combinations, both increasing the Kd for GABARAP from 1.8 µM for the wild type ATG14-BECN1 heterodimers to about 28 µM, which likely corresponds to a non-specific interaction with GABARAP. Taken together, these results confirm that the mutagenesis strategy we applied to the different GABARAP-binding sites was successful and can eliminate binding of PtdIns3K-C1 to GABARAP. Both C and N sites cooperate for PtdIns3K-C1 activation by GABARAP. a Design of the bead-based binding assay. RFP-trap agarose beads were coated with mCherry-ATG14 − BECN1 heterodimers, incubated with different concentrations of mCerulean-tagged GABARAP and imaged by confocal microscopy. b Affinity for mCerulean-GABARAP of different C site and N site mutants. All titrations are shown in Supplementary Fig. 4a. The plot shows the average ± SD of three Kd values determined in technical repeats of the experiment. Means were compared to the wild-type control with a one-way ANOVA with Tukey’s correction for multiple testing. c GUV-based activity assay of full PtdIns3K-C1 assembly carrying C site and N site mutants in the presence or absence of maleimide-lipidated mCerulean (mCer)-GABARAP. Top: PtdIns3P production initial rates expressed in AF647-PX fluorescence/min, average ± SD for n GUVs. Averages were compared in a two-way ANOVA with Tukey’s correction for multitesting. Left, middle and right panels correspond to separate experiments, representative of three independent repeats each (Supplementary Fig. 4b). Bottom: Representative micrographs of GUVs at the end of the reaction in the AF647 channel. Scale bars: 5 µm. The C and N sites cooperate for PtdIns3K-C1 activation by GABARAP The mutant combinations in Fig. 6b were incorporated into the full PtdIns3K-C1 assembly and activation assays were done in the presence of lipidated GABARAP on GUVs during time courses (Fig. 6c, Supplementary Fig. 4b). Mutating the ATG14 LIR in the C site (Cmut) halved the initial rate of GABARAP activation, without abolishing it completely. Mutating the BECN1 LIR (NBmut) with or without a deletion of the ATG14 YVAV motif (∆NA) had no effect on the activation of PtdIns3K-C1 by GABARAP, however a combination of both C and NB mutations completely abolished the ability of GABARAP to activate PtdIns3K-C1. Thus, the NB site mutation had an observable effect on activity only when combined with a mutation of the C site, which is in line with the latter having the greatest affinity for GABARAP (see previous section). In the absence of GABARAP, all PtdIns3K-C1 constructs had the same basal PtdIns3P production, confirming none of these mutations affected the integrity of the complex. Taken together, these results functionally validate the interaction sites between PtdIns3K-C1 and GABARAP we identified and indicate that both the C site and the N site in PtdIns3K-C1 are important for GABARAP to fully activate the complex, with the C site being the main contributor to the activation and the resulting positive feedback loop. Discussion In this study, we showed that among the three LIR motifs in PtdIns3K-C1 that were originally proposed to bind GABARAP23, only the ones in BECN1 (NB site) and ATG14 (C site) are actually accessible to GABARAP in the full PtdIns3K-C1 assembly, whereas the PIK3C3 LIR is occluded due to numerous interactions with the PIK3R4 protein as seen in our cryo-EM structure (Fig. 3e). However, it is still possible that this region could bind an mATG8 if a conformational change would occur to allow the PIK3C3 LIR to dissociate from PIK3R4, induced, for instance, by the phosphorylation of PIK3C3 S249 by ULK123, or by a change in the nucleotide-binding region of PIK3R4 located near the LIR. The affinities of GABARAP for PtdIns3K-C1 or ATG14:BECN1 heterodimer that we measured (Kds of 1.2 µM and 1.8 µM, respectively) were very similar to the ones obtained with peptides containing LIR motifs derived from ATG14 or BECN1 (Kd of 2.8 and 3.6 µM, respectively)23, albeit slightly higher thanks to the cooperativity of the C, NA and NB sites for the binding to GABARAP. An alternative ‘active’ conformation of PtdIns3K-C1 was recently proposed based on a low-resolution class from a cryo-EM study35. In this conformation, the PIK3C3 helical domain/kinase unit (FATKIN) rotates by about 140° from the conformation that we observed in our PtdIns3K-C1 structures. Even in this alternative conformation, the interactions with PIK3R4 that occlude the PIK3C3 LIR from binding an mATG8 would persist. In addition to the C and NB LIR-mediated binding sites, our MS and cryo-EM results identified a GABARAP binding site NA in the N-terminus of ATG14, forming hydrophobic contacts with the ɑ2-β1 loop of GABARAP (Supplementary Fig. 5). Such a bipartite interaction of a protein with an mATG8 is atypical and unique to PtdIns3K-C1. Interestingly, the N-terminal region of the mATG8s is an evolutionarily conserved feature in the mATG8 family that distinguishes these proteins from other ubiquitin-like proteins44 and is a region with multiple roles, ranging from p62 interaction45 to membrane tethering46,47. Our cryo-EM structure of PtdIns3K-C1 with GABARAP provides structural details about the interaction between PtdIns3K-C1 and GABARAP in their soluble form; however, we could not determine the structure of the complex on reconstituted membranes with lipidated GABARAP. Future work will be needed to fully understand how PtdIns3K-C1 is recruited and activated by GABARAP on membranes and to determine whether PtdIns3K-C1 interaction with GABARAP affects GABARAP orientation on membranes48. The transition state of a kinase occurs with both ATP and phosphoacceptor bound to the enzyme. There is no structure of PtdIns3K-C1 with either ATP or the PtdIns substrate, despite recent attempts to capture the complex in an activated conformation35,49, and a structure of a similar complex, PtdIns3K-C2 bound to membranes15. Our cryo-EM structures of PtdIns3K-C1 bound to GABARAP or to the nucleotide ADP-MgF3 both have conformations that resemble the ‘inactive’ state described previously for PtdIns3K-C135, however, they provide structural insights into two essential components of PtdIns3K-C1 activation: the binding of a nucleotide and the recruitment of PtdIns3K-C1 to membranes by GABARAP50. We could see for the first time at the side chain level how the PIK3C3 kinase active site engages with a nucleotide. In addition, our work paves the way for future efforts aimed at resolving the structure of PtdIns3K-C1 bound to a nucleotide and the lipid substrate on membranes, helped by an activating protein like GABARAP. We found that the mATG8s GABARAP and GABARAPL1 are more potent activators of PtdIns3K-C1 than members of the LC3 family, in particular LC3A and LC3B, in line with the previously reported binding selectivity of PtdIns3K-C1–derived LIR motifs23. In this previous study, the authors gave a structural explanation for the higher affinity of the C site (ATG14 LIR) and NB site (BECN1 LIR) for GABARAP and LC3C compared to LC3A and LC3B, involving hydrophobic interactions between residues in positions +4 and +6 of the LIR and mATG8 residues of the hydrophobic pocket HP2. Interestingly, the NA site we discovered also shows residues that could explain the selectivity of PtdIns3K-C1 for the GABARAP family of mATG8s. The GABARAP-NA interaction we saw in cryo-EM consists of hydrophobic contacts between residues 24-KYPD-27 of GABARAP and residues 37-YVAV-40 of ATG14. A multi-sequence alignment of all human mATG8s (Supplementary Fig. 6) shows that the KYPD sequence of GABARAP is conserved among all members of the GABARAP family but is not shared with the LC3 family since LC3A/LC3B have a 26-QHPS/T-29 sequence that is likely to prevent such hydrophobic contacts with the PtdIns3K-C1 NA site. Interestingly, LC3C shows on this site a 32-KFPN-35 sequence closely resembling that of GABARAP, which might explain why, among the LC3 family, LC3C showed a particularly high affinity for and activation of PtdIns3K-C1 (Fig. 2b, e). In addition, the Y25 of GABARAP is known to be involved in a cation-π interaction with GABARAP R28, an interaction that is conserved in the GABARAP family but lost in the LC3 family members LC3A and LC3B, where this pair of residues is replaced by H27 and K30, respectively44. LC3C at these positions too (F33 and K36) is more like the GABARAP family. As this cation-π interaction is in the vicinity of the GABARAP residues interacting with NA in our cryo-EM structure, we suggest this conserved cation-π interaction could be critical to limit the flexibility of GABARAPs and LC3C in this site, stabilizing the interaction with the PtdIns3K-C1 NA site. We propose that the PtdIns3K-C1 NA site could contribute to the selectivity of PtdIns3K-C1 for GABARAPs and LC3C, explaining why these mATG8s have a greater affinity for PtdIns3K-C1 and result in a higher PtdIns3K-C1 activation. Our results are in line with the preference for the GABARAP family of mATG8s in core autophagy components21 and are reminiscent of the activation of ULK1 by GABARAP51. Our finding that mATG8s, in particular GABARAP, can efficiently activate PtdIns3K-C1 changes what was previously accepted as the order of events in phagophore nucleation and elongation during autophagy. Because the lack of PtdIns3P fails to recruit Atg8/mATG8 proteins both in yeast and mammals during starvation32,52, this PtdIns3K-C1−driven PtdIns3P production > recruitment of WIPIs > Atg8/mATG8 lipidation axis is evolutionarily conserved and is an essential pathway for autophagy. Based on the genetic and molecular hierarchy of recruitment, this process was thought to be unidirectional so far. Our results suggest that once lipidated on the growing phagophore, GABARAP can in turn recruit PtdIns3K-C1 and activate it, creating a positive feedback loop causing more PtdIns3P to be generated and thus more GABARAP to be recruited and lipidated (Fig. 7). This positive feedback loop could be essential to sustain an efficient PtdIns3P production on the elongating phagophore, especially during the engulfment of large cargoes53. The GABARAP−PtdIns3K-C1 loop is in line with recent studies suggesting other positive feedback loops between proteins involved in the initial steps of autophagy, such as PtdIns3K-C1 and WIPI217, ULK1 and PtdIns3K-C154, ULK1 and GABARAPs55, ULK1 and ATG16L156 and ULK1 and WIPI257. This would suggest that during autophagy, many positive feedback loops involving PtdIns3K-C1 and other autophagy components happen concomitantly to drastically increase the PtdIns3P production, allowing a rapid growth of the phagophore. This work revealed the structural details of the GABARAP−PtdIns3K-C1 positive feedback loop and helps the understanding of autophagy at a fundamental and molecular level. The GABARAP−PtdIns3K-C1 positive feedback loop. During autophagy, PtdIns3K-C1 initially produces PtdIns3P at the ER (left), which can then recruit WIPI2. WIPI2 recruits the ATG5-ATG12–ATG16L1 E3 complex to allow the lipidation of GABARAP on the growing phagophore (middle). GABARAP then recruits PtdIns3K-C1 to the phagophore and activates it, allowing for more PtdIns3P to be produced in a positive feedback loop (right). Methods Protein expression and purification Purification of PtdIns3K-C1 (untagged and mCherry-tagged) PtdIns3K-C1 constructs were expressed and purified as published before15. Briefly, 2 L of Expi293F suspension cells (ThermoFisher A14527) were transfected with different combinations of plasmids containing PtdIns3K-C1 subunits (see Table 1). The cell pellet was lysed in 100 mL of buffer containing 50 mM HEPES pH 8.0, 150 mM NaCl, 0.5 mM tris(2-carboxyethyl)phosphine (TCEP), 1% (v/v) Triton-X-100 (Sigma X100-500), 12% (v/v) glycerol, 2 mM MgCl2, 2 protease inhibitor tablets (Roche 05056489001) and 1 mM phenylmethylsulfonyl fluoride (PMSF), then centrifuged at 140,000 g for 30 min in a Ti45 rotor (Beckman Colter). The supernatant was passed through a 5 µm syringe filter (Sartorius), then mixed with 2 mL of IgG beads (Cytiva Sepharose 6 Fast flow 17096902) and incubated for 3.5 h at 4 °C with 8 rpm rotation. The lysate was then transferred to a gravity flow column and washed with 100 mL of wash buffer containing 50 mM HEPES pH 8.0, 150 mM NaCl, 0.5 mM TCEP, 0.1% Triton-X-100, 12% (v/v) glycerol, 5 mM ATP, 50 mM MgCl2 and 4 µg/mL of RNAse A (Sigma 83834), then with 100 mL of TEV buffer containing 50 mM HEPES pH 8.0, 150 mM NaCl and 0.5 mM TCEP. The protein was cleaved overnight with TEV protease in the TEV buffer at 4 °C. The next day, the protein was eluted with a buffer containing 20 mM HEPES, pH 8.0, 150 mM NaCl, 0.1% CHAPS (Calbiochem 220201) and 1 mM TCEP, then concentrated to about 500 µL using a 100 kDa cut-off Millipore large concentrator (Millipore, UFC901024) for gel filtration on a Superdex 200 10/30 column (Cytiva 17517501). Proteins were concentrated, aliquoted and frozen in liquid nitrogen and stored at −80 °C. Table 1PlasmidsIDConstructBackboneMarkerPurposeReferenceAddgene #pOP7966xHis-HsNRBF2 MIT (1–89)pOPTHampCryo-EMOhashi, et al.58235701pYO1025HsPIK3C3 untagged and HsPIK3R4-3xTEV-ZZpCAGampBeads binding assay, GUV assays, cryo-EM, HDX-MS, XL-MSOhashi, et al.58235705pYO1100HsATG14, untaggedpCAGampBeads binding assay, GUV assaysThis study235707pYO1006HsBECN1, untaggedpCAGampGUV assaysOhashi, et al.59235724pYO1645GST-PX domain with Cys at 2 and C85SpOPTGampGUV assaysBuckles, et al.79235719pYO1101HsBECN1 and HsATG14, both untaggedpCAGampGUV assays, cryo-EM, HDX-MS, XL-MSOhashi, et al.59235708Addgene 1690796xHis-TEV-HsATG3pETDuet1ampGUV assays, enzymatic lipidationFracchiolla, et al.17169079Addgene 16907610xHis-TEV-HsATG5-HsATG7-HsATG10-HsATG12-HsATG16L1-TEV-StrepIIpGBdestkan + genGUV assays, enzymatic lipidationFracchiolla, et al.17169076pYO1562mCherry-HsBECN1pEGFP-C1kanBeads binding assayThis study252672pYO16016xHis-TEV-mCerulean-HsLC3A (1–120)pOPTHampBeads binding assays, GUV assaysThis study252673pYO16096xHis-TEV-mCerulean-HsLC3B (1–120)pOPTHampBeads binding assay, GUV assaysThis study252674pYO16106xHis-TEV-mCerulean-HsLC3C (1–126)pOPTHampBeads binding assay, GUV assaysThis study252675pYO16116xHis-TEV-mCerulean-HsGABARAP (1–116)pOPTHampBeads binding assay, GUV assaysThis study252676pYO16126xHis-TEV-mCerulean-HsGABARAPL1 (1–116)pOPTHampBeads binding assay, GUV assaysThis study252677pYO16136xHis-TEV-mCerulean-HsGABARAPL2 (1–116)pOPTHampBeads binding assay, GUV assaysThis study252678pYO1594mCerulean-TEV-14xHispOPampBeads binding assaysThis study252679pYO1495mCherry-TEV-GSTpOPampBeads binding assaysThis study252680pYO1499ZZ-3xTEV-mATG7pCAGampGUV assays, enzymatic lipidationThis study252681pYO1928ZZ-TEV-HsBECN1pRS424ampBeads binding assaysThis study252682pYO1932ZZ-3xTEV-HsATG14-mCherrypRS426ampBeads binding assaysThis study252683pYO1933ZZ-3xTEV-HsATG14 (1–434)-mCherrypRS426ampBeads binding assaysThis study252684pYO1943ZZ-TEV-HsBECN1 (F97A I100A)pRS424ampBeads binding assaysThis study252685pYO1948ZZ-3xTEV-HsATG14 (1–434 ∆YVAV)-mCherrypRS426ampBeads binding assaysThis study252686pYO1797HsATG14 (W435A L438A), untaggedpCAGampGUV assaysThis study252687pYO1845HsBECN1 (F97A I100A)pCAGampGUV assaysThis study252688pYO1982HsBECN1 (F97A I100A) and HsATG14 (∆YVAV), both untaggedpCAGampGUV assaysThis study252689ADp1GST-TEV-HsGABARAP (1–116)pOPTGampEnzymatic lipidation, cryo-EM, HDX-MS, XL-MSThis study252690ADp196xHis-TEV-mCerulean-HsGABARAP (L117C)pOPTHampGUV assaysThis study252703ADp17GST-TEV-HsGABARAP (L117C)pOPTGampMaleimide lipidation, XL-MSThis study252702 Plasmids Purification of mCherry-tagged ATG14-BECN1 heterodimers All mCherry-tagged ATG14-BECN1 heterodimers in Fig. 6, Supplementary Fig. 4 were purified using a Saccharomyces cerevisiae protein expression and purification system modified from Rostislavleva et al.13 and Ohashi et al.58. The plasmids were transformed into a yeast strain YOY193 (Matα pep4::HIS3 prb::LEU2 bar1:HISG lys2::GAL1/10-GAL4 can1 ade2 ura3 leu2–3,112 trp1 TRX1-ADE2::ade2∆13). For each heterodimer, yeast colonies were grown at 30 °C until OD600 = 0.8–1.5 in 100 mL of medium lacking uracil and tryptophan (–URA –TRP) (0.67% yeast nitrogen base without amino acids, 0.5% casamino acids, 0.002% adenine sulfate, 0.002% tyrosine and 0.002% leucine) supplemented with 2% glucose. The whole culture was added to 700 mL of –URA –TRP medium supplemented with 2% galactose, 2% glycerol, 3% lactic acid (Merck L4263), and 10 µM ZnCl2, and grown at 30 °C for 26 h. Cells were harvested in 50 mL tubes, and lysed in lysis buffer (50 mM Tris at pH 9.1, 150 mM NaCl, 0.5 mM TCEP, 1% Triton-X, 2 mM MgCl2, 0.5 mM phenylmethylsulfonyl fluoride [PMSF], and a protease inhibitor tablet [Roche 05056489001]) with 3 g of 0.5 mm zirconia beads (YTZ® GRINDING MEDIA, Tosh) using a cell disruptor (FastPrep-24, MP Biomedicals) at intensity 6.5 for 40 s. Lysates were spun at 3000 g for 1 min. Supernatants were further spun at 140,000 g for 30 min in a Ti45 rotor (Beckman Colter). The supernatant was passed through a 5 µm syringe filter (Satorius), then incubated with 0.5 mL of IgG beads (GE Healthcare 17-0969-02) for 3.5 h at 4 °C with 8 rpm rotation. The beads/lysate mixture was transferred to a gravity flow column, and washed with 100 mL of wash buffer (50 mM HEPES, pH 8.0, 150 mM NaCl, 0.1% Triton X-100, 0.5 mM TCEP, 5 mM ATP [Sigma, A2383], 50 mM MgCl2, 5 µg/mL RNaseA [Sigma, 83834]) and 100 mL of TEV buffer (50 mM HEPES, pH 8.0, 150 mM NaCl, 0.5 mM TCEP). Five mL of TEV buffer and 10 µL of 4.7 mg/mL TEV protease were added and incubated at 4 °C overnight without rotation. The eluate was collected, and the proteins were eluted three more times with 5 mL buffer each. The elution fractions were combined and passed through a 0.45 µm syringe filter (Thermo Fisher 15191499), then concentrated using a 50 kDa concentrator (Millipore, UFC905024) before running on a Superdex 200 10/30 column equilibrated with 20 mM HEPES, pH 8.0, 150 mM NaCl, 0.5 mM TCEP. The main peak fractions were combined and concentrated using a 50 kDa concentrator (Millipore, UFC905024). The protein was aliquoted, frozen in liquid nitrogen, and stored at −80 °C. Purification of mCerulean-mATG8s, mCerulean, and ATG3 proteins All mCerulean-mATG8 protein constructs (see Table 1) were designed by truncating the mATG8 C-terminus to expose the last glycine, to allow the enzymatic lipidation. For maleimide reaction (ADp19), the last C-terminal residue of GABARAP L117 was replaced by a cysteine. For all constructs, the plasmids were expressed in E. coli C41 (DE3) bacteria, induced with 0.3 mM IPTG at 16 °C for 18−20 h, then the bacterial pellet was lysed using a sonicator (Sonics VCX-750-110) for 5 min on ice (2 s on, 3 s off, and 60% amplitude) in 50 mL buffer containing 50 mM HEPES pH 8.0, 150 mM NaCl, 12% (v/v) glycerol, 20 mM imidazole, 5 mM 2-mercaptoethanol, a protease inhibitor tablet (Roche 05056489001) and 1 mM PMSF. The lysate was centrifuged at 140,000 g for 30 min in a Ti45 rotor (Beckman Colter). The supernatant was passed through a 5 µm syringe filter (Sartorius), then incubated with Ni-NTA agarose beads (Qiagen 30210) for 1 h at 4 °C with 8 rpm rotation and was transferred to a gravity flow column. The beads were washed with buffer containing 50 mM Tris pH 8.0, 300 mM NaCl, 12% (v/v) glycerol, 20 mM imidazole and 5 mM 2-mercaptoethanol and eluted with a buffer containing 50 mM Tris pH 8.5, 20 mM NaCl, 300 mM imidazole, 5 mM 2-mercaptoethanol and 1 mM PMSF. The protein was further purified on a Q column (Cytiva 17115401) in buffer 50 mM Tris pH 9.0, 5 mM 2-mercaptoethanol and eluted with buffer 50 mM Tris pH 9.0, 1 M NaCl and 5 mM 2-mercaptoethanol. The peak fractions were combined and concentrated down to ~ 500 µL using a 10 k cut-off Millipore large concentrator (Millipore, UFC901024), then the protein was subjected to a gel filtration on a Superdex 75 16/60 column (Cytiva 17-1068-01) in a buffer containing 20 mM HEPES pH 8.0, 150 mM NaCl and 0.5 mM TCEP. Proteins were concentrated, aliquoted and frozen in liquid nitrogen and stored at −80 °C. For the ADp19 construct, the final buffer contained 5 mM TCEP instead of 0.5 mM. ATG3 protein was purified essentially in the same way as above, except the plasmid was Addgene 169079 and the Q elution buffer was 50 mM Tris, pH 9.0, 2 M NaCl and 5 mM 2-mercaptoethanol. Purification of mCherry The plasmid pYO1495 was transformed into C41(DE3)RIPL cells. The bacteria were induced in 900 mL of 2TY medium with 0.3 mM IPTG at 20 °C for 20 h. The bacterial pellet was lysed using a sonicator for 5 min on ice (2 s on, 3 s off, and 60% amplitude) in 20 mL of lysis buffer containing 50 mM Tris, pH 8.0, 300 mM NaCl, 1 mM DTT, a protease inhibitor tablet (Roche 05056489001) and 1 mM PMSF. The lysate was centrifuged at 140,000 g for 15 min in a Ti45 rotor (Beckman Colter). The supernatant was incubated with 1 mL of Glutathione Sepharose 4B beads (Cytiva, 17075605) at 8 rpm roation at 4 °C for 1 h. The lysate/beads mixture was transferred to a gravity flow column and washed with 250 mL of wash buffer (50 mM Tris pH 8.0, 300 mM NaCl, 1 mM DTT, and 12% glycerol), and 200 mL of TEV buffer (50 mM Tris pH 8.0, 300 mM NaCl, 1 mM DTT), then 10 mL of TEV buffer with 80 µL of 4.7 mg/mL His6-LIPOYL-TEVS219V was added to the beads and incubated at 4 °C overnight without rotation. The eluate was collected, and the proteins were eluted three more times with 5 mL TEV buffer each. The elution fractions were combined, and imidazole was added to a final concentration of 20 mM, then passed through a 5 mL HisTrap HP column (Cytiva 17524801). The flow-through fraction was concentrated using a 10 kDa cut-off concentrator (Millipore 901024). The concentrated protein was subjected to gel filtration on an S75 16/60 column equilibrated with GF buffer (20 mM Tris 8.0, 150 mM NaCl, 0.5 mM TCEP). The peak fractions were combined and concentrated down to 18 mg/mL (641 µM). The protein was aliquoted, frozen in liquid nitrogen and stored at −80 °C. Purification of GABARAP (1−116) and GABARAP L117C As for the other mATG8 expressed, the C-terminal glycine of GABARAP was exposed by an ∆L117 deletion (ADp1 in Table 1). For the maleimide reaction, a C-terminal cysteine was added (ADp17). For both constructs, the plasmids were expressed in bacteria (E. coli C41 (DE3)) induced in 900 mL of 2TY medium with 0.3 mM IPTG at 18 °C for 18 h. The bacterial pellet was lysed using a sonicator (Sonics VCX-750-110) on ice for 3 min (2 s on, 2 s off, and 60% amplitude) in a buffer containing 50 mM HEPES pH 7.4, 150 mM NaCl, 10 mM Dithiothreitol (DTT), 12% (v/v) glycerol, a protease inhibitor tablet (Roche 05056489001) and 1 mM PMSF. The lysate was centrifuged at 140,000 g for 15 min in a Ti45 rotor (Beckman Colter) and incubated with Glutathione Sepharose 4B beads (GE Healthcare 17-0756-05) for 1 h at 4 °C with 8 rpm rotation, then the beads were washed in a gravity flow column with a wash buffer containing 50 mM HEPES pH 7.4, 300 mM NaCl, 12% (v/v) glycerol and 10 mM DTT. The protein was cleaved with TEV protease overnight to remove the N-terminal GST tag, in a buffer containing 50 mM HEPES 7.4, 300 mM NaCl and 5 mM 2-mercaptoethanol. The TEV protease was removed using Ni-NTA agarose beads (Qiagen 30210), the GABARAP proteins concentrated using a 10 kDa cut-off concentrator (Millipore 901024) and further purified with a Superdex 75 16/60 column (Cytiva 17-1068-01) equilibrated in a buffer containing 20 mM HEPES pH 7.4, 300 mM NaCl and 5 mM DTT. For GABARAP L117C (ADp17), no DTT was present in the gel filtration buffer. The protein was concentrated, aliquoted and frozen in liquid nitrogen and stored at −80 °C. Purification of mATG7 protein The plasmid pYO1499 at 1.1 mg/L was transfected into 1 L of Expi293f cells in Expi293 Expression Medium (ThermoFisher A1435102) with polyethylenimine (PEI) ‘MAX’ (Polysciences 24765, 1 mg/mL in PBS) at 3 mg/L culture, and grown at 37 °C, 8% CO2, and 125 rpm shaking for 2 days. Cell pellets were suspended in 30 mL lysis buffer (50 mM HEPES, pH 8.0, 150 mM NaCl, 1% Triton X-100 [Sigma, X100], 12% glycerol, 0.5 mM tris[2-carboxyethyl]phosphine [TCEP, Soltec Ventures, M115], 2 mM MgCl2, 1 × EDTA-free inhibitor tablet [Roche, 05056489001], and 1 mM PMSF), then spun at 140,000 g for 30 min in a Ti45 rotor (Beckman Colter). The supernatant was passed through a 5 µm syringe filter (Satorius), then incubated with 2 mL of IgG beads (GE Healthcare 17-0969-02) for 3.5 h at 4 °C with 8 rpm rotation. The beads/lysate mixture was transferred to a gravity flow column, and washed with 150 mL of wash buffer (50 mM HEPES, pH 8.0, 150 mM NaCl, 0.1% Triton X-100, 0.5 mM TCEP, 5 mM ATP [Sigma, A2383], 50 mM MgCl2, 5 µg/mL RNaseA [Sigma, 83834]) and 150 mL of TEV buffer (50 mM HEPES, pH 8.0, 150 mM NaCl, 0.5 mM TCEP). Ten mL of TEV buffer and 30 µL of 4.7 mg/mL TEV protease were added and incubated at 4 °C overnight without rotation. The eluate was collected, and the proteins were eluted three more times with 5 mL buffer each. The elution fractions were combined, passed through a 0.45 µm syringe filter (Thermo Fisher 15191499), then concentrated using a 50 kDa concentrator (Millipore, UFC905024) before running on a Superose6 10/30 column equilibrated with 20 mM HEPES, pH 8.0, 300 mM NaCl, 0.5 mM TCEP. The main peak fractions were pooled, and TCEP was added to a final concentration of 2 mM, then concentrated using a 50 kDa concentrator at 17.57 mg/mL (225 µM). The protein was aliquoted, frozen in liquid nitrogen and stored at −80 °C. Purification of ATG5-ATG12–ATG16L1 complex The Addgene 169076 plasmid was transformed into DH10EMBacY cells, and white colonies were selected on LB agar containing kanamycin, gentamycin, tetracycline, Bluo-gal, and IPTG for bacmid purification. The generated bacmid was transfected into Sf9 cells at a cell density of 1.2 × 106 using FuGENE HD (Promega No E2311) and grown in 6-well plates in Insect XPRESS w/ L-Gln (Lonza LZBE12-730Q) at 27 °C for 6 days for virus production. The virus was used to infect 1 L of Sf9 cells. Cells were grown for 2 days at 27 °C. The cell pellets were suspended in 75 mL of lysis buffer (25 mM HEPES, pH 7.4, 300 mM NaCl, 1 mM DTT, 2 mM MgCl2, and 12% glycerol, 1 × EDTA-free inhibitor tablet (Roche, 05056489001), and 1 mM PMSF), lysed by sonication with 60% power, 2 s on 8 s off for 2 min 30 s. then spun at 140,000 g for 30 min in a Ti45 rotor (Beckman Colter). The supernatant was passed through a 5 µm syringe filter (Satorius), then passed through a 5 mL strep column (StrepTrap HP, Cytiva 10298754). The column was washed with 50 mL lysis buffer and 100 mL wash buffer (25 mM HEPES pH 7.4, 300 mM NaCl, 1 mM DTT, 2 mM MgCl2, and 12% glycerol), then eluted with 50 mL of elution buffer (25 mM HEPES pH 7.4, 150 mM NaCl, 1 mM DTT, and 10 mM desthiobiotin). The elute was concentrated using a 100 kDa cut-off concentrator (Millipore, UFC901024) for gel filtration using a Superose6 10/30 column equilibrated with gel filtration buffer (25 mM HEPES 7.4, 300 mM NaCl, and 0.5 mM TCEP). The peak fractions were concentrated using a 100 kDa cut-off concentrator (Millipore, UFC901024) at 12.6 mg/mL (108 µM). The protein was aliquoted, frozen in liquid nitrogen and stored at −80 °C. Purification and labeling of p40phox PX domain Bacteria (E. Coli C41 [DE3]) were harvested after expressing the plasmid pYO1645 upon induction in 900 mL of 2TY medium with 0.3 mM IPTG at 18 °C for 19 h. The pellet was sonicated on ice for 5 min (2 s on, 3 s off, and 60% amplitude) in lysis buffer containing 20 mM HEPES, pH 8.0, 200 mM NaCl, 1 mM TCEP, 0.05 μL/mL universal nuclease (ThermoFisher 88702) and 0.5 mg/mL lysozyme (MP Biomedicals 195303). The lysate was centrifuged at 140,000 g for 15 min in a Ti45 rotor (Beckman Colter) and incubated with 2 mL of affinity Glutathione Sepharose resin (GE Healthcare 17-0756-05), washed with 100 mL of wash buffer containing 20 mM HEPES pH 8.0, 300 mM NaCl and 1 mM TCEP and 100 mL of TEV buffer containing 20 mM HEPES pH 8.0, 200 mM NaCl and 1 mM TCEP. The N-terminal GST tag was cleaved with TEV protease by overnight incubation. Fractions were eluted, concentrated using a 10 kDa Amicon Ultra15 concentrator (Millipore, UFC901024) and loaded on a Superdex 75 16/60 column (Cytiva 17-1068-01) for further purification by gel filtration, in a buffer containing 20 mM HEPES pH 8.0, 200 mM KCl and 1 mM TCEP. The purified PX domain was labeled using the Alexa Fluor 647 (AF647) C2 Maleimide kit (Invitrogen A20347), using a Heparin column (Cytiva 17040701) to separate the labeled protein from the free dye. The protein was concentrated, aliquoted and frozen in liquid nitrogen and stored at −80 °C. Purification of NRBF2 MIT domain The plasmid pOP79658 was expressed in E.coli C41(DE3) cells in 2xTY medium supplemented with 0.1 mg/mL ampicillin, and induced with 0.3 mM IPTG for 12 h at 20 °C. Bacteria pellets were centrifugated at 6700 g for 20 min and resuspended in 75 mL lysis buffer (25 mM TRIS, pH 7.5, 300 mM NaCl, 10 mM imidazole, 2 mM 2-mercaptoethanol, 0.5 mg/mL lysozyme, 0.05 µL/mL universal nuclease) and sonicated (Sonics VCX-750-110) for 6 min (10 s on, 10 s off, 60% amplitude). The lysate was centrifugated at 142,000 g for 30 min at 4 °C, and the supernatant was filtered through a 0.45 µm syringe filter. The sample was loaded onto a 5 mL HisTrap FF column (Cytiva 17-5255-01) equilibrated with wash buffer (20 mM TRIS pH 7.5, 150 mM NaCl, 10 mM imidazole, 1 mM TCEP), washed with 100 mL wash buffer, 50 mL Ni-A buffer (20 mM TRIS pH 7.5, 150 mM NaCl, 10 mM imidazole, 1 mM TCEP), and eluted with a gradient of Ni-B buffer (20 mM TRIS pH 7.5, 150 mM NaCl, 300 mM imidazole, 1 mM TCEP). Peak fractions were pooled, diluted two-fold with Q-A buffer (20 mM TRIS, pH 7.5, 0.5 mM TCEP), and applied to a 5 mL HiTrap Q HP column (Cytiva 17115401), washed with 50 mL Q-A buffer, and eluted with a gradient of Q-B buffer (20 mM TRIS, pH 7.5, 1 M NaCl, 0.5 mM TCEP). Peak fractions were pooled and concentrated using a 3 kDa concentrator (Millipore, UFC900324) and loaded onto an S75 16/60 gel filtration column (Cytiva 17-1068-01) equilibrated in 20 mM HEPES pH 7.0, 100 mM NaCl, 0.5 mM TCEP. Peak fractions were pooled and concentrated to 136 µM, then aliquoted and frozen in liquid nitrogen and stored at −80 °C. Confocal microscopy beads-based binding assays Essentially, 1 µL of anti-RFP agarose beads (Chromotek RTA-10) was used per well of a 384-well plate (Greiner 781856), and the bead volume was calculated depending on the number of wells. The beads were resuspended three times with 1 mL of buffer containing 25 mM HEPES pH 8.0, 100 mM NaCl, 0.5 mM TCEP pH 7.4 and 0.1 mg/mL BSA (Sigma A7030) and centrifugated at 600 g. Excess buffer was removed, and the beads were incubated with 1.7 µM of mCherry, mCherry−PtdIns3K-C1 or different mCherry-ATG14 − BECN1 constructs for 30 min on ice. Some beads were also added 5 µM NRBF2 MIT. Beads were washed 3 times as above, and 3 µL of beads were added to 27 µL of 10 µM mCerulean-GABARAP (1 − 116) in a 384-well glass-bottom plate (Greiner 781856). Reactions were incubated for 30 min on ice in the sealed plate covered from light. Beads were then imaged on an inverted confocal microscope (LSM 780, Zeiss) with a 10x objective (EC Plan-Neofluar 10x/0.3 M27, Zeiss). Beads were visualized using the DIC channel, the mCerulean channel was excited using a 458 nm Argon laser and collected with a band of 463−556 nm, and the mCherry channel was excited using a diode-pumped solid-state (DPSS) 561 nm laser and collected with a 578−696 nm band. Images were taken using ZEN software (2.1 SP3, Zeiss) and analysed with FIJI software (ImageJ2 2.14.0). The surface of beads was selected as an elliptical ROI (regions of interest), and the fluorescence intensity was determined for each channel using a previously described macro37,59. Ten background ROI without beads were selected, and their corresponding intensity average subtracted from the fluorescence of the beads for each channel. The mCerulean fluorescence intensity was normalized by the mCherry fluorescence for each bead to account for coating discrepancies on the beads as well as out-of-focus beads. Data was analysed in Prism10 (GraphPad), and binding affinities and corresponding R2 were determined with the non-linear regression fitting tool with the ‘one site-binding’ option. The negative control with mCherry immobilized on beads showed no binding to mCerulean-GABARAP, nor did mCherrry–PtdIns3K-C1 to mCerulean (Supplementary Fig. 1b). GUV preparation A thin PVA gel was created by depositing 88 µL of 5% w/v polyvinyl alcohol (PVA, Sigma 8.14894, MW approx. 145 kDa) on a 13 mm coverslip (VWR 631-0150), followed by a 30 s centrifugation at 2000 g on a microfuge (IKA mini G) using a custom-made coverslip adapter. The PVA gel was dried at 60 °C for 1 h. Lipid stocks were purchased from Avanti Polar Lipids, Inc. with the following catalog numbers: Liver PI (mixed chains, bovine, 840042 C), DOPS (840035 C), DOPE (850725 C), DOPC (850375 C), Lissamine-Rhodamine-DOPE (810150 C), DSPE-PEG(2000)-Biotin (880129 C) and DOPE-MCC (780201 C). The lipid stocks were mixed to prepare the lipid mixtures ‘GUV DO base’ at 1 mg/mL in chloroform, as shown in Table 2. The lipid mixture was carefully deposited on the PVA gel by two applications of 7.5 µL, and the remaining chloroform was evaporated by 1 h desiccation under vacuum. The GUVs were swollen by adding 220 µL of swelling solution (500 mM D-Glucose, Fisher 50-99-07) on the coverslips in a 24-well plate (Costar 3526, Corning) and incubated for 1 h at room temperature. GUVs were then pipetted into a 2 mL tube (previously coated with 5 mg/mL BSA [Sigma A7030] and washed with swelling solution). Table 2Lipid MixturesIDCompositionDescriptionPurposeADguv315% liver PI, 10% DOPS, 20% DOPE, 55% DOPC, 0.017% Lissamine-Rhodamine-DOPEGUV DO baseGUV assay (enzymatic lipidation)ADguv1015% liver PI, 10% DOPS, 15% DOPE, 55% DOPC, 5% DOPE-MCC, 0.03% DSPE-PEG-Biotin, 0.017% Lissamine-Rhodamine-DOPEGUV DO base + 5% PE-MCCGUV assay (maleimide lipidation)ADluv1515% liver PI, 10% DOPS, 15% DOPE, 55% DOPC, 5% DOPE-MCCLUV DO base + 5% PE-MCCLUV XL-MS Lipid Mixtures PtdIns3K-C1 activation assay on GUVs For the GUV enzymatic lipidation, ADguv3 GUVs were added to the wells of a 384-well glass-bottom plate (Greiner 781856), 12 µL GUVs for a 50 µL final volume in observation buffer containing 50 mM HEPES pH 7.4 and 271 mM NaCl. ATP-containing buffer was also added, for a final concentration of 25 mM HEPES, pH 7.4, 2 mM MnCl2, 1 mM MgCl2, 1 mM TCEP pH 7.4, 1 mM ATP. Proteins were diluted in buffer containing 25 mM HEPES pH 7.4, 150 mM NaCl, 1 mM TCEP pH 7.4 and 0.5 mg/mL BSA (Sigma A7030), and added to the GUV mixture for a final concentration of 20 nM PtdIns3K-C1, 3.5 µM AF647-PX, 200 nM ATG7, ATG3 and ATG12-ATG5–ATG16L1, and 2.5 mM mCerulean-mATG8. The reaction was incubated for 2 h at room temperature. For the endpoint GUV maleimide lipidation, a 384-well glass-bottom plate (Greiner 781856) was coated with 0.1 mg/mL avidin (Invitrogen A2667) in 1 mg/mL BSA (Sigma A7030)-containing PBS. Wells were washed twice with observation buffer containing 50 mM HEPES, pH 7 and 271 mM NaCl, the buffer was removed and 5 µL of 0.1 mg/mL BSA-biotin (Sigma A8549) was added to the wells. Next, 10 µM of mCerulean-GABARAP L117C (ADp19) or GABARAP L117C (ADp17) and 12 µL ADguv10 GUVs in buffer containing 25 mM HEPES pH 7.0, 150 mM NaCl and 0.5 mg/mL BSA (Sigma A7030) were added, for a total volume of 29 µL. After an overnight incubation at 4 °C, the maleimide reaction was quenched with 4 washes of 360 µL of a solution containing 31.8 mM HEPES pH 8.0, 173 mM NaCl, 181.8 mM D-Glucose (Fisher 50-99-07) and 5 mM 2-mercaptoethanol (Aldrich M6250). PtdIns3K-C1 and AF647-PX in the ATP-containing buffer were then added at the same concentrations as described above for a final volume of 50 µL and incubated for 2 h at room temperature for the endpoint assays. GUVs were imaged on an inverted confocal microscope (LSM 880, Zeiss), using a x63 oil immersion objective (Plan-Apochromat 63x/1.4 Oil DIC M27, Zeiss) and ZEN software (Zeiss). The mCerulean channel was excited with a 458 nm Argon laser and collected with a band of 463−556 nm, the Lissamine-Rhodamine channel was excited with a Diode-pumped solid-state (DPSS) 561 nm laser and collected with a 566−629 nm band, and the AF647 channel was excited with a HeNe 633 nm laser and collected with a 638−755 nm band. Micrographs were analysed in FIJI, where the surface of GUVs was selected as elliptical ROIs (regions of interest) and the fluorescence intensity was determined for each channel using the previously described macro37,59. Ten background ROI without GUVs were selected, and their corresponding intensity average subtracted from the fluorescence of the GUVs for each channel. Data was analysed in Prism 10 (GraphPad) for statistics. For the time course assays, GUVs were prepared following the GUV maleimide lipidation protocol above, but an 8-well glass bottom chamber (Ibidi 80827) was used instead, for a final volume of 200 µL. A chamber system insert plate (PeCon 000 470) was added to the universal mounting frame to immobilize the 8-well chamber. GUV positions were recorded on ZEN software before adding to the GUVs the PtdIns3K-C1 and AF647-PX proteins in the ATP-containing buffer. Imaging of the GUVs was done as detailed above, except that all the fluorescence channels were tracked for all positions every 2 min over a time course of 180 min total, using the Positions and Time Series options in ZEN software, and Definite Focus.2 (Zeiss) was used to keep the GUVs in focus every 8 positions and every time point. Images were analysed as above, but using the version of the macro designed for time courses37. Curves were manually aligned by their linear phases in Microsoft Excel, and the initial slopes were calculated with the LINEST function. Data was further analysed in Prism 10 (GraphPad) for statistics. Cryo-electron microscopy sample preparation and screening For the structure of PtdIns3K-C1 and GABARAP, PtdIns3K-C1 at a final concentration of 3.6 µM was incubated on ice for 30 min with 3.6 µM NRBF2 MIT domain and 40 µM GABARAP (1 − 116), in a buffer containing 25 mM HEPES, pH 8.0, 100 mM NaCl, 1 mM TCEP, pH 7.4, 4 mM CHAPSO (Sigma C3649 lot SLBG1674V) and 0.005% (v/v) Nonidet P-40 substitute (Santa Cruz Biotechnology, sc29102 lot G1015). A sample volume of 3.5 µL was applied on a UltrAuFoil R 1.2/1.3 Au300 grid previously glow-discharged (40 mA, 60 s, air, Edwards Sputter Coater S150B), and plunge-frozen in liquid ethane using a Vitrobot Mk 2 (Thermo Fisher) plunger at 14 °C, 100% humidity, with 20 s waiting time, 4 s blotting time, and blot force set to +8. Grids were screened on a Glacios (Thermo Fisher) microscope using a Falcon3 detector in linear mode. For the structure of PtdIns3K-C1 with ADP-MgF3, PtdIns3K-C1 at a final concentration of 3.6 µM was incubated on ice for 30 min with 3.6 µM NRBF2 MIT, in a buffer containing 25 mM HEPES, pH 8.0, 100 mM NaCl, 1 mM TCEP, pH 7.4, 4 mM CHAPSO, 0.005% (v/v) Nonidet P-40 substitute, 13 mM MgCl2, 22.5 mM NaF and 2.5 mM ADP. Grids were prepared and screened as detailed above. Cryo-electron microscopy data collection For the structure of PtdIns3K-C1 with GABARAP, two datasets with identical parameters were collected on an identical sample, one with 4776 movies and another with 13,450 movies. Micrographs were collected on a Titan Krios G1 TEM (Thermo Fisher) with a Gatan K3 detector (Gatan) in counting mode and energy filter set at 20 eV, using a 100 µm objective aperture. The magnification was 105,000x, and the pixel size was 0.725 Å/px. Micrographs were collected with EPU (Thermo Fisher) in AFIS mode, with one movie of 40 frames per hole, at a defocus range of –0.8 to 2 µm. The dose rate was 27 e−/px/s over an exposure length of 0.97 s, making 50 e−/Å2 total dose. See Table 3 for the statistics of the cryo-EM data collection and processing. Table 3Cryo-EM data collection, refinement and validation statisticsPDB 9TW2, EMD-56346 (consensus)PDB 9TW3, EMD-56345 (consensus)PDB 9TZ3, EMD-56454 (consensus)Data collectionMagnification105,000 x105,000 xVoltage (kV)300300Electron exposure (e−per Å2)5050Defocus range (μm)−0.8 to −2−0.8 to −2Pixel size (Å)0.7250.725Symmetry imposedC1C1Initial particle images (no.)973,900306,134Final particle images (no.)643,471577,838186,859Map resolution (Å) FSC threshold3.580.1433.600.1433.660.143Map resolution range (Å)FSC threshold2.97−47.90.51.53−52.20.53.27−49.40.5Model resolution (Å) FSC threshold3.90.53.90.54.40.5RefinementMap sharpening B factor (Å2)147.2152.0136.2Model composition (no.) Non-hydrogen atoms Protein residues Ligands234272897523432289752233027607B factors (Å2) Protein Ligand43.1235.9281.6169.2459.3662.82R.m.s. deviations Bond lengths (Å) Bond angles (°)0.0050.8440.0040.8010.0030.612Validation MolProbity score Clashscore Poor rotamers (%)1.041.190.391.001.110.390.910.870.49Ramachandran plot Favored (%) Allowed (%) Disallowed (%)96.693.31096.903.070.0397.252.750 Cryo-EM data collection, refinement and validation statistics Map resolution (Å) FSC threshold 3.58 0.143 3.60 0.143 3.66 0.143 Map resolution range (Å) FSC threshold 2.97−47.9 0.5 1.53−52.2 0.5 3.27−49.4 0.5 Model resolution (Å) FSC threshold 3.9 0.5 3.9 0.5 4.4 0.5 Model composition (no.) Non-hydrogen atoms Protein residues Ligands 23427 2897 5 23432 2897 5 22330 2760 7 B factors (Å2) Protein Ligand 43.12 35.92 81.61 69.24 59.36 62.82 R.m.s. deviations Bond lengths (Å) Bond angles (°) 0.005 0.844 0.004 0.801 0.003 0.612 Validation MolProbity score Clashscore Poor rotamers (%) 1.04 1.19 0.39 1.00 1.11 0.39 0.91 0.87 0.49 Ramachandran plot Favored (%) Allowed (%) Disallowed (%) 96.69 3.31 0 96.90 3.07 0.03 97.25 2.75 0 For the structure of PtdIns3K-C1 with ADP-MgF3, 8483 movies were collected on a Titan Krios G1 TEM (Thermo Fisher) with a Gatan K3 detector (Gatan) in counting mode and energy filter set at 20 eV, using a 100 µm objective aperture. The magnification was 105,000x, and the pixel size was 0.725 Å/px. Micrographs were collected in EPU (Thermo Fisher) in AFIS mode, with one movie of 50 frames per hole, at a defocus range of –0.8 to 2 µm. The dose rate was 16 e−/px/s over an exposure length of 1.66 s, making 50 e−/Å2 total dose. See Table 3 for the statistics of the cryo-EM data collection and processing. Cryo-electron microscopy data processing and model building For the structure of PtdIns3K-C1 with GABARAP, both datasets were separately imported into CryoSPARC60. The movies were corrected for beam-induced motion using patch motion correction61 and their CTF estimated with the CTFFIND4 wrapper62. Movies with CTF-estimated resolution below 10 Å were selected for further processing. Particle picking was done in crYOLO-1.7.563, using a custom-trained model from a previous study37. A total of 973,900 particles were combined in CryoSPARC and extracted at a box size of 452 px, then downsampled to 226 px. Particles were subjected to one round of 2D classification, and 932,517 particles were separated from obvious noise. Particles were aligned in a non-uniform refinement64, using a density of PtdIns3K-C1 and MIT as a 30 Å-low passed reference (EMD-54326, unpublished data). Four masks of different regions of the complex were created in UCSF ChimeraX65: 1) PIK3C3 and PIK3R4 kinases, 2) The core of PtdIns3K-C1 (PIK3R4 WD40, PIK3C3 C2 domain and helical domain), 3) The PIK3R4 helical solenoid, BECN1 and ATG14 N-termini (NTDs) and GABARAP and 4) The BECN1 and ATG14 C-termini (CTDs). The masks were generated in CryoSPARC and used for focused 3D classification of all particles, followed by a local refinement on the best 3D classes with unbinned particles at a box size of 452 px. The particles that went through a local refinement of the BECN1 and ATG14 NTDs went through an additional round of focused 3D classification and local refinement, using a smaller mask on the GABARAP protein, done in UCSF ChimeraX. For the focused 3D classification of the CTDs of BECN1 and ATG14 (the adapter arm of PtdIns3K-C1), 2 classes with different conformation (‘main’ and ‘alternative’) were separately refined in local refinements. All particles were used in CryoSPARC non-uniform refinements to generate two consensus maps, using one class of the adapter arm conformation or the other. Using the same particles as for the consensus maps, two composite maps were generated with TEMPy-ReFF66 using the five local refinements. See Table 3 for the statistics of the cryo-EM data collection and processing, and Supplementary Fig. 7 for a detailed diagram of the processing workflow. An initial model of all proteins was generated using AlphaFold367. This model was fitted to each composite map by rigid-body fit in UCSF ChimeraX, followed by manual adjustment assisted by local fit using the ISOLDE plugin68. The models were then refined using TEMPy-ReFF, followed by real-space refinement in PHENIX69. The statistics presented in Table 3 were determined in comparison to the consensus maps using MolProbity70 and PHENIX, and the local resolutions and FSC resolutions were determined in CryoSPARC. For the structure of PtdIns3K-C1 with ADP-MgF3, the movies were imported in CryoSPARC, motion-corrected and CTF-estimated as above. The trained crYOLO model was used to pick 306,134 particles, which were imported into CryoSPARC and extracted at a box size of 452 px, then downsampled to 150 px. One round of 2D classification was used to get rid of obvious noise, then particles were aligned with a non-uniform refinement using the map of PtdIns3K-C1 and GABARAP (EMD-56346) obtained above as a 30 Å-low passed reference. Three masks of different regions of PtdIns3K-C1 were created in UCSF ChimeraX and CryoSPARC: 1) The core of PtdIns3K-C1, including ATG14 and BECN1 N-termini, 2) PIK3C3 and PIK3R4 kinases, and 3) ATG14 and BECN1 C-termini. The masks were used for three focused 3D classifications on all particles, followed by a local refinement on the best classes with unbinned particles at a box size of 452 px. For the PIK3C3 and PIK3R4 kinases, a non-uniform refinement was done instead of a local refinement and a smaller mask on the kinases was applied for a focused 3D classification, followed by a local refinement using the same mask. All local refinements were combined in a composite map in TEMPy-ReFF. The model of PtdIns3K-C1 (PDB 9TW2) was modified and refined as detailed above. See Table 3 for the statistics of the cryo-EM data collection and processing, and Supplementary Fig. 8 for a detailed diagram of the processing workflow. Hydrogen-deuterium exchange coupled to mass spectrometry HDX experiments were performed as follows: PtdIns3K-C1 was preincubated for 30 min on ice in the absence or presence of GABARAP, and exposed for different deuteration times (3 s on ice, 3, 30, 300 and 3000 s at room temperature) to a 20 mM HEPES pD8, 150 mM NaCl, 0.5 mM TCEP deuterated buffer (D2O, Thermo Scientific, 166300100) to yield solutions that were 2 μM PtdIns3K-C1 alone or in presence of 50 μM of GABARAP in 80% D2O. Triplicated experiments were performed for each time point. Exchange reactions were quenched in 2 M guanidine-hydrochloride, 100 mM glycine, formic acid (Fisher Chemical, 10596814), LC-MS grade water (Romil, H949), pH 2.3 (final pH was 2.5). Samples were then flash-frozen in liquid nitrogen after quenching and stored at −80 °C until analysis. Prior to LC-MS analysis, samples were quickly thawed and injected on an HDX Manager coupled to an Acquity UPLC M-Class system (Waters) set at 0.1 °C. Samples were then digested at 15 °C using a 20 mm × 2.0 mm home-made pepsin column (Pepsin from Thermo Scientific, 20343 and hardware from Upchurch Scientific, C130-B) and loaded on a UPLC pre-column (ACQUITY UPLC BEH C18 VanGuard pre-column, 2.1 mm I.D. × 5 mm, 1.7 μM particle diameter, Waters, 186003975) for 2 min at 100 μL/min. Digested peptides were then eluted over 20 min from the pre-column onto an ACQUITY UPLC BEH C18 column (1.0 mm I.D. x 100 mm, 1.7 μM particle diameter, Waters, 186002346) using a 5 − 43% gradient of Acetonitrile (Romil, H050), 0.1% formic acid. MS data were acquired using a Synapt G2Si-HDMS (Waters) with electrospray ionization in positive ion mode, using the data-independent acquisition mode (MSE) over a m/z range of 50–2000, and a 100 fmole/μL Glu-FibrinoPeptide solution was used for lock-mass correction and calibration. The following parameters were used during the acquisition: capillary voltage, 3 kV; sampling cone voltage, 40 V; source temperature, 90 °C; desolvation gas, 150 °C and 650 L.h−1; scan time, 0.3 s; trap collision energy ramp, 20–45 eV. Peptide identification was performed using ProteinLynx Global Server 3.0.1 (PLGS, Waters) with a home-made protein sequence library containing PIK3C3, PIK3R4, ATG14, BECN1, GABARAP, and pepsin sequences, with peptide and fragment tolerances set automatically by PLGS. Peptides were then filtered using HDExaminer 3.4.0 (Trajan Scientific and Medical) with a minimum fragment per amino acid of 0.2, a minimum intensity of 103, a peptide length between 5 and 25 residues, a minimum PLGS score of 6.62, and an MH+ tolerance of 5 ppm. An initial automated spectral processing step was conducted by HDExaminer, followed by a manual inspection of individual peptides for sufficient quality, where only one charge state per peptide was kept. Deuterium uptakes were not corrected for back-exchange and are reported as relative. HDX-MS results were statistically validated using Archaeopteryx (https://github.com/Glenn-Masson/Archaeopteryx), where significant differences required 0.3 Da, 5%, and p-value < 0.05 (n = 3) using an unpaired two-tailed t-test. HDX-MS results were illustrated on the PDB 9TW2 model using PyMOL 2.5.4 (Schrödinger). Deuteration tables are available in Supplementary Data 1. The mass spectrometry proteomics data (including raw data, tables of every peptide included within the dataset and all relative uptake plots) have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository71 with the dataset identifier PXD074963. Large unilamellar vesicles (LUVs) preparation Lipid stocks were purchased from Avanti Polar Lipids, Inc. with the following catalog numbers: Liver PI (mixed chains, bovine, 840042 C), DOPS (840035 C), DOPE (850725 C), DOPC (850375 C) and DOPE-MCC (780201 C). Lipid stocks were mixed to prepare the lipid mixture ADluv15 at 5 mg/mL in chloroform, as shown in Table 2. The solvent was then evaporated with a stream of nitrogen gas and desiccated for 1 h under vacuum. The dried lipid film was resuspended in a buffer with 25 mM HEPES, pH 7.0, 150 mM NaCl for a final concentration of 5 mg/mL. The liposomes were vortexed for 2 min (Vortex Genie 2, Scientific Industries), followed by a sonication for 1 min in a water bath sonicator (Ultrawave U50). Liposomes were frozen and thawed 10 times in a cycle between liquid nitrogen and a 43 °C water bath, followed by an extrusion for 30 cycles using a 100 nm diameter extruder (Whatman Anotop 10 syringe filter 0.1 μm pore size, 68091112). Liposomes were aliquoted, flash-frozen in liquid nitrogen and stored at –80 °C. In vitro reconstitution of the mATG8 lipidation on LUVs For the maleimide lipidation in the XL-MS experiment, 0.5 mg/mL of ADluv15 was incubated with 10 µM of GABARAP (L117C) in a buffer containing 25 mM HEPES, pH 7.0 and 150 mM NaCl at 4 °C overnight. Next day, the reaction was quenched with 5 mM 2-mercaptoethanol (Aldrich M6250) and the LUVs aggregates removed by pelleting with a centrifugation at 1000 g for 1 min at 4 °C. The LUVs in the supernatant were concentrated by a 20 min centrifugation at 21,000 g at 4 °C, and resuspended in 1/10 of the initial reaction volume. Crosslinking mass spectrometry For the sample with soluble GABARAP, 4 µM of PtdIns3K-C1 were incubated with 50 µM of GABARAP (1−116) for 30 min on ice in a buffer containing 20 mM HEPES pH8, 150 mM NaCl and 0.5 mM TCEP. For the sample containing LUVs, the vesicles were coated with GABARAP (L117C) by a maleimide reaction as detailed above. PtdIns3K-C1 was then added to the concentrated LUVs at a concentration of 3.6 µM, for a 50 µL final volume, and incubated on ice for 30 min. Protein crosslinking reactions were carried out by adding 2 mM of sulfo-SDA to the protein mixture, with 5 min incubation on ice, followed by 10 s 365 nm UV radiation from a home-built UV LED setup. Crosslinking reactions were quenched with the addition of ammonium bicarbonate to a final concentration of 50 mM. The quenched solution was reduced with 5 mM DTT and alkylated with 20 mM iodoacetamide. SP3 protocol72,73 was used to clean-up and buffer exchange the reduced and alkylated protein, shortly, proteins were washed with ethanol using magnetic beads for protein capture and binding. The proteins were resuspended in 100 mM NH4HCO3 and were digested with trypsin (Promega) at an enzyme-to-substrate ratio of 1:20, and protease max 0.1% (Promega). Digestion was carried out overnight at 37 °C. Clean-up of peptide digests was carried out with HyperSep SpinTip P-20 (ThermoScientific) C18 columns, using 60% Acetonitrile as the elution solvent. Peptides were then evaporated to dryness via Speed Vac Plus (Savant). Dried peptides were resuspended in 30% acetonitrile and were fractionated via size exclusion chromatography using a Superdex 30 Increase 3.2/300 column (GE Healthcare) at a flow rate of 20 µL/min using 30% (v/v) ACN and 0.1% (v/v) TFA as a mobile phase. Fractions were taken every 5 min, and the 2nd to 7th fractions containing crosslinked peptides were collected. Dried peptides were suspended in 3% (v/v) Acetonitrile and 0.1% (v/v) formic acid and analysed by nano-scale capillary LC-MS/MS using an Ultimate U3000 HPLC (ThermoScientific) to deliver a flow of 300 nL/min. Peptides were trapped on a C18 Acclaim PepMap100 5 μm, 0.3 μm x 5 mm cartridge (ThermoScientific) before separation on Aurora Ultimate C18, 1.7 μm, 75 μm x 25 cm (Ionopticks). Peptides were eluted on optimized gradients of 90 min and interfaced via an EasySpray ionization source to a tribrid quadrupole Orbitrap mass spectrometer (Orbitrap Eclipse, ThermoScientific) equipped with FAIMS. MS data were acquired in data-dependent mode with a Top-25 method, high-resolution scans full mass scans were carried out (R = 120,000, m/z 400–1550), followed by higher energy collision dissociation with stepped collision energy range 21, 30, 34% normalized collision energy. The tandem mass spectra were recorded (R = 60,000, isolation window m/z 1, dynamic exclusion 50 s). Mass spectrometry measurements were cycled for 3 s durations between FAIMS CV –45 and –60 V. Cross-linking data analysis was done by converting Xcalibur raw files to MGF files using ProteoWizard74, and cross-links were analysed by XiSearch75. Search conditions used 3 maximum missed cleavages with a minimum peptide length of 5. Variable modifications used were carbmidomethylation of cysteine (57.02146 Da) and Methionine oxidation (15.99491 Da). The false discovery rate was set to 5%. Data was imported into XiView76 for manual filtering and exported to the PDB 9TW2 in PyMOL 2.5.4 (Schrödinger). Crosslinks tables with scores are available in Supplementary Data 2. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD079187. WIPI2 and FIP200 colocalization experiments HEK293 cells (ATCC CRL-1573) were plated on glass coverslips to be 60–80% the next day for the experiment. Treatments with PP242 and the ATG7 inhibitor were done by withdrawing half the volume of the medium in each well, adding the compound at 2x final concentration and then adding the solution back to the well for the indicated times. After fixation in 3.7% formaldehyde, the coverslips were permeabilised in 0.1% NP40 and stained with antibodies to endogenous LC3 (Caltag Medsystems MBL-PM036), FIP200 (ProteinTech 10069-1-AP) and WIPI2 (Bio Rad Laboratories MCA5780GA) as indicated. Images were acquired on a ZEISS Axio Imager D2 and analysed using ImageJ. Multiple sequence alignment mATG8 protein sequences were obtained from UniProt (https://www.uniprot.org/)77 and aligned using Clustal Omega (https://www.ebi.ac.uk/jdispatcher/msa/clustalo)78. Both the multiple sequence alignment and the phylogenetic tree of Supplementary Fig. 6 were retrieved from Clustal Omega. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Reporting Summary Transparent Peer Review file Source data Source Data