Structural pathway for PI 3-kinase regulation by VPS15 in autophagy

Abstract

The class III phosphatidylinositol-3 kinase complexes I and II (PI3KC3-C1 and -C2) have vital roles in macroautophagy and endosomal maturation, respectively. We elucidated a structural pathway of enzyme activation through cryo-EM analysis of PI3KC3-C1. The inactive conformation of the VPS15 pseudokinase stabilizes the inactive conformation, sequestering its N-myristate in the N-lobe of the pseudokinase. Upon activation, the myristate is liberated such that the VPS34 lipid kinase catalyzes PI3P production on membranes. The VPS15 pseudokinase domain binds tightly to guanosine triphosphate (GTP), and stabilizes a web of interactions to autoinhibit the cytosolic complex and to promote the activation upon membrane binding. These findings show in atomistic detail how the VPS34 lipid kinase is activated in the context of a complete PI3K complex.

One-Sentence Summary:

The activation of the VPS34 lipid kinase complexes is mediated by the myristoylated, GTP binding pseudokinase VPS15.

Macroautophagy (hereafter autophagy) is the primary eukaryotic catabolic response to insufficient supplies of amino acids, and the main cellular mechanism for clearing toxic aggregates and dysfunctional organelles (1). Deficits in autophagy are implicated as drivers of many human diseases (2), such as Parkinson’s disease in which the genes Parkin and PINK1 initiate damage-induced mitophagy (3). Autophagy is initiated by the concerted action of several core complexes (4), including the phosphatidylinositol 3-kinase class III complex I (PI3KC3-C1). PI3KC3-C1 is a heterotetrameric complex that comprises one subunit each of the VPS34 lipid kinase, the regulatory scaffold VPS15 (vacuolar protein sorting 15, also known as PIK3R4, phosphoinositide-3-kinase regulatory subunit 4), BECN1 (beclin 1, autophagy related), and the autophagy specific subunit ATG14 (autophagy-related 14) (5–9). The related complex II (PI3KC3-C2) which functions in both endosome maturation and autophagy, shares the three subunits VPS34, VPS15, and BECN1, and has the same overall architecture. In PI3KC3-C2, UVRAG (UV radiation resistance associated) replaces ATG14 as the fourth subunit (5). PI3KC3-C1 produces phosphatidylinositol (PI) 3-phosphate (PI3P) on the phagophore, which is the precursor to the autophagosome. PI3P recruits WIPI (WD repeat domain, phosphoinositide-interacting) proteins, which are in turn critical for all subsequent steps in autophagosome expansion (10). In endosome maturation, PI3P generated by PI3KC3-C2 (11, 12) recruits the early endosome fusion (13) and ESCRT machinery (14) and other essential factors. In both the autophagic and endosomal pathways, PI3KC3 activation is a decisive step.

Because of their central role in autophagy and endosomal sorting, the structure and dynamics of PI3KC3-C1 and -C2 complexes have been intensively studied (15–21). Yet a near-atomic picture of the activating transition has not been described, likely because of the dynamic character of the complex. The VPS34 lipid kinase domain (VPS34^KD) is highly mobile relative to the rest of PI3KC3-C1 (15, 17), and its mobility is necessary for function (17). This has made it challenging to determine the structures of conformational substates at high resolution. We obtained a structure of human PI3KC3-C1 at a local resolution of up to 3.3 Å. Building on the insight that the small GTPase RAB1A binds and recruits PI3KC3-C1 (21), we reconstructed the complex in the presence of RAB1A at a local resolution of nearly 2.3 Å in the best-resolved regions, and to classify and refine inactive, transitional, and active conformations, defining the major conformational states along the PI3KC3 activation pathway. We describe a large conformational change in VPS34^KD relative to the VPS15 pseudo-kinase domain (VPS15^PKD). We discovered that an Arg residue at the gatekeeper position of VPS15^PKD, not detected in other members of the kinome, confers GTP binding. We mapped the position of the covalent N-myristoyl modification of VPS15. Combined with molecular dynamics (MD) simulations, we use the structural information to propose a mechanism for the activation cycle of VPS34 as regulated by N-myristoylation and binding of GTP to VPS15^PKD.

Cryo-EM structure of PI3KC3-C1 alone and in complex with RAB1A

The structure of the PI3KC3-C1 and the related PI3KC3-C2 complex, which functions in late stages in autophagy and in endosomal sorting (9) has been extensively studied (15–19, 21, 22) yet still no structures of the autophagy-specific PI3KC3-C1 have been resolved at a high enough resolution to place amino acid side chains. We determined the PI3KC3-C1 structure at a best local resolution of 3.26 Å, allowing accurate model building with side-chain assignment (fig. S1 and fig. S2). The refined structure is in a similar overall conformation to that of the PI3KC3-C1-NRBF2 complex (18) and the quality of the density facilitated building a molecular model (fig. S2). The VPS34^KD (kinase domain) was not visualized, consistent with observations of its flexibility (15).

To gain insight into the activation of the VPS34^KD, we confirmed the reported activation of PI3KC3-C1 by RAB1A (21)(fig. S3), and reconstituted and determined the structure of the complex of catalytically inactive RAB1A(Q70L) with PI3KC3-C1 by cryoelectron microscopy (cryo-EM) (Fig. 1A,B, fig. S4). Extensive 3D classification of a 19,000-movie dataset (fig. S4A) allowed us to determine the structures of the PI3KC3-C1-RAB1A complex in three 3D classes that appeared to represent distinct states in the activation pathway. In the most populated 3D class, determined at nominal resolution of 2.7 Å, VPS34^KD was not visualized. We refer to it as the “transitional” structure because it is neither evidently active nor inactive. The overall domain architecture of the complex is very similar to our apo-PI3KC3-C1 structure (C_α r.m.s.d. = 2.7 Å). RAB1A binding induced the VPS34 helical hairpin (VPS34^HH) to rotate inward and upward, moving by up to 5.5 Å relative to apo-PI3KC3-C1 to accommodate the RAB1A switch-II motif (fig. S5). This led to a contraction of the complex along its long axis by 7 Å, a rotation of the VPS15 WD40 and helical linker domain by 5 degrees, and a global bending of the complex resembling a longbow pulled under tension (fig. S5B).

Fig. 1. Structure of the RAB1A-PI3KC3-C1 complex.

(A) Cryo-EM reconstruction of the RAB1A-PI3KC3-C1 and the associated atomic model. (B) Domain organization of the complex. Colors correspond to A. (C) Close up of the RAB1A interface with VPS34 C2 domain. The RAB1A switch-II motif (gold) and the HH (light purple) are indicated. (D) 180 degree rotation of the RAB1A C2 domain interface. The RAB1A switch-I motif is indicated (orange). (E) Inset, close-up of the switch-II motif contacts with the C2-HH cleft. The switch-II helix adopts an extended conformation, forming multiple hydrophobic and polar contacts with HH helix 1 and the C2 β-sheet core. (F) Close up of the switch I interface with HH helix 2. The Arg199 and Glu202 form a hydrogen bond network with the switch-I and interswitch domains.

RAB1A binds to PI3KC3-C1 through the cleft between the VPS34 C2 domain (VPS34^C2) and VPS34^HH (Fig. 1C,D) and the GTP-dependent switch-II region of RAB1A (Fig. 1C). RAB1A switch-I and the interswitch β-strand contacts the first α-helix of VPS34^HH. The local resolution of the density reached 2.5 Å in this region (fig. S6A–B), and we visualized ordered water molecules near the GTPase active site (Fig. 1F, S6C). VPS34^HH Arg199 and Glu202, two residues from the REIE motif essential for the RAB1A-PI3KC3-C1 interaction²⁰, form a salt bridge and hydrogen bond cage with Asp47 of the RAB1A interswitch motif. VPS34 Glu202 forms a salt bridge with Arg199, which in turn hydrogen bonds to the backbone carbonyl of RAB1A at Asp47. Asp47 of the RAB1A interswitch motif coordinates a pair of water molecules together with VPS34^HH Glu202. Glu202 also hydrogen bonds with the amide of the switch-I backbone at RAB1A Gly45, whereas the water coordinated between VPS34 Glu202 and RAB1A Asp47 contacts switch-I Thr43. Thr43 participates in the octahedral coordination of the Mg²⁺ in the GTP pocket. The N-terminal and GTP-proximal portion of switch-II is in essentially the same conformation as seen in other RAB1A:effector complexes (23), while the C-terminal portion from Thr75 to Ser78 diverges to accommodate the first VPS34^HH helix. This interaction network explains why RAB1A^GTP, but not RAB1A^GDP, can recruit PI3KC3-C1.

VPS15 N-myristate sequesters the activation loop of VPS34^KD

Through refinement of a second 3D class we built a model of the VPS34^KD interface with the VPS15 pseudo-kinase domain (VPS15^PKD) with residue-level detail (Fig. 2A, B). This conformation resembled PI3KC3-C1 and -C2 structures observed at lower resolution in complex with NRBF2(18) and as bound to RAB5A and liposomes(21). The lower resolution of the previous structures precluded mechanistic interpretation in side-chain level detail. We observed that the VPS34^KD activation loop (A-loop) containing the critical DFG motif is sequestered by the VPS15^PKD N-terminal region and VPS15^PKD phosphate-binding loop (P-loop) (Fig. 2C). The inaccessibility of the DFG motif and A-loop of VPS34 in this conformation is incompatible with substrate recognition and kinase activity. Therefore, we denote this conformation as inactive. We observed a density feature in a pocket formed from bulky hydrophobic residues of VPS15^PKD P-loop Phe37 and Phe38 and A-loop Tyr185 and Phe186, Phe55, and VPS15^PKD αC Leu64, Tyr67, and Leu71 (Fig. 2D). This density is contiguous with the polypeptide density for the N-terminus of VPS15 (Fig. 2D). VPS15 is N-myristoylated (24), and the density could be entirely accounted for by an N-myristoyl modification of Gly2 (Fig. 2D). To confirm that this post-translational modification was present in our sample, we performed trypsin digestion coupled to liquid chromatography—mass spectrometry (LC-MS) and identified a peptide corresponding to the myristoylated N-terminal 29 residues of VPS15 (fig. S7A). The VPS34 A-loop Arg768 and VPS15 P-loop made hydrogen bonds to the main-chain of the N-myristoyl Gly2 itself, and adjacent residues, including Ile8 (Fig. 2E). These observations establish the structure of the likely inactive conformation of PI3KC3-C1, the ordered binding of the VPS15 N-myristate, and show how the ordered binding of the N-myristate in the VPS15^PKD may help lock VPS34^KD in an inactive conformation.

Fig. 2. Structure of the VPS34^KD-VPS15^PKD interface and ordered myristate in the inactive conformation.

(A) Focused view of the Cryo-EM reconstruction of the VPS34 kinase domain in the inactive state. The inset at upper left shows the context of this VPS34^KD-VPS15^PKD interface in the full structure. (B) Model of the kinase domain derived from the map. (C) Close-up of the interface between the VPS34 A-loop and the VPS15 N-terminal domain. (D) The N-Myristoyl-Glycine is sequestered in a hydrophobic pocket lined with bulky hydrophobic residues from the VPS15 activation loop, the VPS15 P-loop, and the C α-helix domain. Density is shown for the N-myr-Gly at a contour of σ = 7.0 (E) The interaction between the VPS34 A-loop and the VPS15 N-Terminal Domain. Hydrogen bonds are denoted with black dashed lines. (F) Hydrophobic contacts with VPS15 NTD lock VPS34 Κα10 in a position that blocks membrane interaction by VPS34^KD. (G) Rotated view of the VPS34 inactive interface. (H) Inset view of the sequestration of VPS34 Κα10 and Κα11 by VPS15^PKD. The VPS34 Κα12 is indicated in transparent ribbon form as its position was inferred from the orientation of the other secondary structural features of the kinase domain and not visualized in the cryo-EM density. (I) VPS34 Κα10 forms several hydrogen bonds (black dashes) with the carbonyl oxygens of VPS15 A-loop and P-loop, as well as between VPS34 Glu835 and Arg191.

The C-terminal helix of VPS34, Κα12, embeds into the membrane and is essential for catalysis (25). Κα12 is positioned in part by the immediately preceding helices Κα10 and Κα11 (Fig. 2G, H.) The VPS15^PKD A-loop interacted extensively with VPS34 Κα10 (Fig. 2I), reaching across the narrow gap between VPS34 and VPS15. VPS34 Kα10 Glu835 interacts with VPS15 A-loop Arg191 and the VPS15 P-loop main chain amides of Thr35 and Arg36 (Fig. 2F). VPS34 Κα10 residues Lys837 and Lys840 donated hydrogen bonds to the carbonyl oxygens of VPS15 A-loop Thr189, Arg191, and Arg192 (Fig. 2I). VPS15 N-terminal region Ile13 formed hydrophobic contacts with VPS34 Κα10 Leu833 and the preceding Pro829 (Fig. 2F). In essence, the VPS15 N-terminus and A- and P-loops appear to form an extensive web of contacts that sequesters Κα10, resulting in the misorientation of Κα12 and making it impossible for VPS34^KD to engage with its lipidic substrate. Thus, the sequestration of the VPS15 N-myristate by the N-lobe blocks both the catalytic and membrane binding elements of VPS34^KD.

Liberation of the VPS34^KD catalytic and membrane-engagement motifs in the active conformation

Further 3D classification of the same dataset captured a third 3D class, representing a distinct conformation of VPS34^KD. We refined this state to a nominal overall resolution of 3.3 Å, although density for VPS34^KD and VPS15^PKD was of lower quality, which we attributed to a strong preferred orientation of this state, and to increased mobility in the VPS34^KD (fig. S4). The local resolution for VPS34^KD of 5 Å sufficed to place VPS34^KD as a rigid body, but was insufficient to place sidechains in the density and directly visualize the residues that contribute to this new interface. Relative to its position in the inactive state, VPS34^KD in this third state was rotated by 140 degrees (Fig. 3A–C, Movie 1), breaking all of the VPS15^PKD contacts visualized in the inactive state. A cluster of Arg residues, including Arg36 of the P loop, and Arg191 and Arg193 of the VPS15 A-loop, formed new interactions with VPS34 Kα11 (Fig. 3D and E). To assess the role of the Kα11-VPS15^PKD interface in favorably positioning VPS34^KD for activity, we made two VPS15 mutant constructs, the double mutant R191D and R193D, and the single mutant R36D. Reconstituted complexes containing these mutations had decreased lipid kinase activity (Fig. 3E), consistent with Kα11-VPS15^PKD interactions stabilizing the active conformation. To test the role of the active-state stabilizing contact in VPS15, we tested the ability of wild-type and R36D-R191D-R193D VPS15 to rescue autophagy activity in VPS15 knockout HeLa cells generated with CRISPR-Cas9 editing (fig. S8, Table S2). Wild-type and mutant VPS15 were expressed in similar amounts (fig. S8D), yet the R36D-R191D-R193D mutant did not rescue starvation-induced autophagic flux (Fig. 3F, G). These data show how contacts between VPS34^KD and two regions of VPS15^PKD are functionally required to stabilize the active conformation of PI3KC3-C1.

Fig. 3. Structure of the VPS34 active state.

(A) The locally refined VPS34^KD in the active conformation. (B) The model of the kinase domain in the active state. Key secondary structural features are indicated with labels in the color corresponding to the secondary structural element. The active site of the kinase is indicated with a red star. (C,D). Overview and close-up view of novel VPS34^KD-VPS15^PKD contacts specific to the active conformation. (E) Lipid kinase activity of VPS15 mutations in the active site contacts as measured by luminescence assay. The results of three biological replicates were compared by one-way ANOVA to test for significance, and then a post-hoc T-test with the Bonferonni parameter applied was used for pairwise comparisons. (F-G) VPS15 KO cells stably expressing Halo-LC3B without rescue or rescued with wild type (WT) or R36/191/193D VPS15 were treated with 50 nM TMR-conjugated Halo ligand for 15 min. Following that, cells were washed with 1x PBS and incubated with EBSS for indicated time periods, harvested, and analysed by immunoblotting with indicated antibodies (F) and the percentage of the cleaved Halo band was quantified (G). Data in (G) are mean ± SD from three independent experiments. ****P<0.0001 (two-way ANOVA).

To address the possible implications of this conformation for membrane binding, we performed all-atom molecular dynamics (MD) simulations of PI3KC3-C1 on a phospholipid membrane with a phagophore-like membrane composition (Fig. 4A). The inactive conformation was docked onto the membrane via the BARA (beta-alpha repeated, autophagy-specific) domain, with the two geranylgeranyl Cys residues of RAB1A inserted into the bilayer. Gentle pulling was applied to simulate the separation of the VPS34^KD and VPS15^PKD domains. Once the domains were separated, the N-myristate was removed from the hydrophobic pocket in the N-lobe of the VPS15 pseudokinase and pulled towards the membrane, which docked VPS34^KD onto the membrane (Movie 2). This conformation of PI3KC3-C1-RAB1A is consistent with the active-state cryo-EM structure and remained stable over a subsequent 160 ns of unrestrained dynamics, stabilized by interactions of the VPS15 N-myristate and VPS34 Κα12 with the membrane. To assess the catalytic competence, we tested for substrate lipid recruitment into the active site of membrane-bound VPS34. We performed ten 1 μs atomistic simulations of the kinase domain engaged with the membrane in the pose observed in the simulation of the full complex. These simulations show spontaneous PI entry into the active site, resulting in transient, yet frequent interactions between the γ-phosphate of ATP and the 3’-OH group of PI lipids, at distances compatible with a phosphorylation reaction (Figure 4C,D, fig. S9, Movie 3).

Fig. 4. MD simulation of the PI3KC3-C1 membrane-bound geometry.

(A) MD snapshot of the active PI3KC3 complex I engaged with a membrane. (B) Detailed views in two orientations of A-loop and catalytic loop conformations from snapshot shown in A. (C) Distance between the gamma phosphate of ATP and nearest 3’-O atom of POPI. Distances <5 Å (dashed lines) are regarded as contacts. (D) Snapshot of a POPI molecule at the catalytic site. (E) Basic patch of VPS34 that contacts the membrane in the simulations. (F) Lipid kinase activity of VPS34 mutations in the basic patch as measured by luminescence assay. (G,H) VPS34 KO cells stably expressing Halo-LC3B without rescue or rescued with wild type (WT) or R561E/R566E/K567E/K568E VPS34 were treated with 50 nM TMR-conjugated Halo ligand for 15 min. Following that, cells were washed with 1x PBS and incubated with EBSS for indicated time periods, harvested, and analysed by immunoblotting with indicated antibodies (G) and the percentage of the cleaved Halo band was quantified (H). Data in (H) are mean ± SD from three independent experiments. **P<0.01, ****P<0.0001 (two-way ANOVA)

As compared to the cryo-ET structure of an inactive PI3KC3-C2 bound to liposomes²⁰ through the aromatic tip of the BECN1 BARA domain alone, membrane contact in the simulation was far more extensive (Fig. 4A,B). From the docked model of the complex, we identified several membrane-interacting residues, including a basic patch encompassing Arg561, Arg566, Lys567, and Lys568 (Fig. 4E). To address the implications of these membrane contacts, we inverted the charge of the basic patch with the mutations R561D-R566D-K567D-K568D. As a control, we mutated two aromatic residues in the Kα12 helix, Tyr884 and Trp885, to Asp that are important for the activity of VPS34 (25). The Y884D-W885D mutant was essentially inactive (Fig. 4F). The mutations to the basic residues of VPS34 substantially reduced the in vitro lipid kinase activity of VPS34 (Fig. 4E). To assay the effects of these basic residues on autophagy in cells, we genetically depleted VPS34 from HeLa (human epithelia, Henrietta Lacks) cells using CRISPR-Cas9 (fig. S8) and tested the functional consequences of the VPS34 R561D-R566D-K567D-K568D mutation. The basic patch mutant was expressed at normal levels. This basic patch charge inversion of VPS34 reduced autophagic flux to less than 20% of that in cells expressing the wild-type protein (Fig. 4G,H), consistent with the loss of activity in the in vitro enzyme assay.

VPS15 is a GTP-binding pseudokinase

Refinement of the VPS15^PKD -VPS34^KD of the RAB1A bound complex yielded a local resolution range of 2.3–2.6 Å for the VPS15^PKD nucleotide pocket (fig. S6B). The quality of the density was sufficient to place every side chain that contributed to nucleotide recognition, including associated ordered water molecules. The shape of the purine base density and the position of the Arg103 guanidino group were incompatible with placement of ATP in the density. However, GTP fit the density perfectly, and N5 and O6 of the guanine base were ideally positioned to accept hydrogen bonds from the Arg103 guanidino moiety (Fig. 5A, B). To confirm the identity of the nucleotide, we denatured PI3KC3-C1 purified from HEK 293 cells and eluted the bound molecule by ion-pair reversed-phase acetonitrile gradient HPLC. In confirmation of the EM density map and hydrogen bonding geometry prediction, the bound nucleotide eluted at the same position as the GTP standard (Fig. 5C). We further characterized the nucleotide that was released by electrospray ionization MS (fig. S7B). This directly confirmed that the dominant nucleotide present was GTP, with no detectable ATP or ADP. The MS analysis did reveal the presence of a lesser proportion of GDP, consistent with the possibility that VPS15^PKD can slowly hydrolyze GTP.

Fig. 5. VPS15 is a GTP binding pseudokinase.

(A) The overall architecture of the VPS15^PKD. Broadly conserved secondary structural motifs are indicated with labels in the color of each motif. GTP is indicated in green. (B) Cryo-EM density (σ = 7) of the bound nucleotide is shown. The hydrogen bonds with Arg103 donate hydrogen bonds to the N5 and O6 groups on the guanine ring. The catalytic Asp166 is rotated to form a bidentate hydrogen bond interaction with the gatekeeper Arg103 (C) HPLC elution profile results. Material eluted from a denatured WT (pink) and R103K unmyristoylated (light blue) PI3KC3-C1 sample in the absence of RAB1A is shown to be retained at the same time as the GTP standard. Other nucleotides are shown in the indicated colors. (D) Local sequence alignment of VPS15 P-loop against PKA. The P-loop motifs are shown in lighter colors. (E) The γ-phosphate is aligned 50 degrees away from in-line attack on the location of a hypothetical substrate, indicated in blue. The P-loop of M. musculus PKA structure (yellow) is shown for comparison (PDB: 1ATP). (F) Local sequence alignment with Mus musculus protein kinase A (PKA) showing the substitution of Arg at the gatekeeper position, indicated in blue.(G) The N-myristoyl glycine makes contact with the GTP pocket via the VPS15 A-loop Phe170 and Lys171. (H) Lipid kinase activity of the R103K, unmyristoylated PI3KC3-C1. (I-J) VPS15 KO cells stably expressing Halo-LC3B without rescue or rescued with wild type (WT) VPS15 or other VPS15 mutants were treated with 50 nM TMR-conjugated Halo ligand for 15 min. Following that, cells were washed with 1x PBS and incubated with EBSS for indicated time periods, harvested, and analysed by immunoblotting with indicated antibodies (I) and the percentage of the cleaved Halo band was quantified (J). Data in (J) are mean ± SD from four independent experiments. ns = not significant, *P<0.05, **P<0.01, ****P<0.0001 (two-way ANOVA). (K) A model of VPS15-mediated membrane engagement and VPS34 activation.

The gatekeeper position responsible for nucleotide specificity in kinases (26) is Arg103 in VPS15. In functional kinases, which use ATP, the gatekeeper is almost invariably hydrophobic (26). VPS15 homologs are unusual among kinases and pseudokinases in containing an Arg at the gatekeeper position (Fig. 5B,F; fig. S10). Our map revealed both the position of Arg103, and the overall shape and hydrogen bonding network of the nucleotide (Fig. 5A,B). The unique presence of an Arg at the gatekeeper position reorients the Mg²⁺-coordinating Asp166 of the VPS15 DFG motif towards Arg103 (Fig 5B). The rotamer switch in Asp166 repositioned the catalytic Mg²⁺ such that it coordinated all three GTP phosphates. The corresponding Mg²⁺ in active protein kinases coordinates only the β- and γ -phosphates (27). The VPS15 P-loop has two changes relative to active kinases that cripple its ability to function as a phosphate stabilizing motif. In conventional kinases, the glycine residues at position 3 and 6 in the P-loop (Fig. 5D) stabilize the β and γ phosphates in a phosphotransferase competent geometry (Fig. 5E). The replacement of Gly35 and Gly38 by Thr and Phe, respectively, renders the VPS15 P-loop incapable of positioning the γ phosphate for in line attack on a phosphoacceptor. The collective effect of the loss of the two Gly residues and the reorientation of Asp166 is to shift the orientation of the GTP β and γ phosphates by 50° from their position in active kinases (Fig. 5E), such that they are no longer poised for in line transfer to a phosphoacceptor. This structure rules out that VPS15 is a GTP-dependent protein kinase.

The GTP binding and myristoyl pockets of VPS15 are both highly conserved, and intimately linked via the VPS15 activation loop residues Phe170 and Lys171 (Fig. 5G). We probed the role of GTP recognition by conservatively mutating the gatekeeper Arg103 to Lys, and Lys171 to Asn. We purified the recombinant unmyristoylated VPS15-R103K complex (Fig. S11) and found that this complex contained no GTP as assessed by HPLC (Fig. 5C). This complex had reduced but non-zero enzyme activity (Fig. 5H). The non-zero activity is consistent with the presence of an intact VPS34^KD. We assessed the functional effects of the myristoylation-blocking mutation G2A and the GTP site mutations R103K and K171N separately and together by reconstituting them into HeLa cells lacking VPS15 and assaying nutrient depletion--induced autophagy flux. Although the single mutations G2A, R103K, and K171N had no apparent loss of function (Fig. 5I, J), the double mutant combinations G2A-R103K and G2A-K171N reduced flux to near-background levels. Collectively, these data confirm the functional importance of the myristoyl modification and GTP binding site for proper initiation of autophagy. These results show that dysregulation of VPS34 by defective VPS15 inhibits autophagy even in the presence of partial VPS34 activity.

Discussion

Our results clarify key aspects of PI3KC3 activation. We characterized an apparent activated conformation of the VPS34^KD compatible with lipid kinase activity. The geometry of membrane docking, which is prerequisite for activation, showed that the active conformation appears to make much more extensive contact with membranes than has been observed in existing PI3KC3 structures. The ordered VPS15 N-myristate docking into the N-lobe of the VPS15^PKD that we infer from our studies may help understanding of coupling between membrane docking and enzyme activation that depends on sequestration of the VPS15 N-myristate in the inactive conformation and its release to dock onto the membrane in the activated conformation. Finally, we found that VPS15^PKD is an unusual pseudokinase that binds GTP instead of ATP. As a cornerstone of the function of the VPS15^PKD, the GTP binding site is intertwined closely with sequestration of the N-myristate and allosteric regulation of the VPS34 catalytic site.

RAB1A (21) and its yeast ortholog Ypt1 (28) are implicated in autophagy initiation, whereas RAB5 is fundamental to early endosome maturation (13, 29). The structure reported here reveals the RAB1A:VPS34 interface in high-resolution detail, including ordered water molecules. Interface features are conserved between RAB1A and RAB5, which is consistent with the observation that RAB1A interacts directly only with the VPS34 subunit that is common to both PI3KC3-C1 and -C2 complexes. The differential specificity of RAB1A for PI3KC3-C1 and RAB5 for -C2 is therefore be attributed to context and conformation, rather than to direct subunit-specific interfaces. The extensive interactions of the GTP-dependent switch regions of RAB1A explain the dependence of binding on the RAB1A GTP-bound state. RAB1A binding leads to a long-range structural contraction of the complex. A single molecule study of PI3KC3-C2 activation concluded that RAB5 both recruited and allosterically activated PI3P production (20). The long-range contraction in the PI3KC3-C1 structure shifts the equilibrium between the various conformational states to favor the activated state. The revised model for membrane docking indicates that RAB1A is positioned very close to the membrane, consistent with roles in membrane recruitment and activation.

Our high-resolution data make it possible to re-interpret past lower resolution structures. It is now clear that NRBF2- (18) and RAB5A and liposome-bound (21) complexes correspond to the inactive conformation. Previous observations of the necessity of VPS34^KD mobility (15, 17) for activity are now explained by the observation that a 140° rotation of VPS34^KD mediates activation. This rotation breaks all contacts with VPS15^PKD, and then forms new ones to reach the active conformation. The RAB5A and liposome-bound inactive structure (21) may represent a state that has been recruited to a membrane by RAB binding and the BECN1 BARA aromatic finger (30), and is thereby primed for the activating transition. Engagement of the BECN1 β sheet-1 peptide (19) with the membrane would be a subsequent step in the pathway, following the activating transition, and concurrent with membrane docking. The MD simulations of membrane docking shed light on how VPS34 accesses its membrane-bound PI substrate. The catalytic DFG motif of the VPS34 A-loop and DRH motif of the catalytic loop appear to cooperate to stably coordinate the ATP γ-phosphate in an orientation that is primed for a phosphotransfer reaction. The reaction center is predicted to be roughly 5 Å above the glycerophosphates of the lipids, almost exactly the distance spanned by a single inositol headgroup. The PI substrate lipid then likely diffuses into the VPS34 substrate cleft as directly observed by MD simulations, where it can be phosphorylated by the optimally oriented ATP molecule.

VPS15 has been known, since its discovery, to be myristoylated (12, 24), yet the function of this myristoylation has not been experimentally addressed. We found that the VPS15 N-myristate is sequestered in the N-lobe of VPS15^PKD and stabilizes the inactive conformation of the PI3KC3 complex through a web of interactions with the VPS15 ^PKD P and A-loops and the VPS34 catalytic and A-loops. Loss of myristoylation, in combination with destabilization of the VPS15 GTP site, leads to a loss of function. We attribute this to the dual role of the myristate in stabilizing the inactive state through a network of VPS15^PKD-VPS34^KD interactions, and conversely, stabilizing the active state through direct insertion into the membrane, as summarized in Fig. 5K and Fig. S12.

Pseudokinases, which are catalytically inactive members of the protein kinase superfamily, represent about one tenth of the human kinome (31). A subset of pseudokinases bind nucleotides with various degrees of specificity (31), but none were known to bind specifically to GTP. PI3KC3-C1 purified from HEK 293 cells had GTP bound in our inactive cryo-EM structure without the resupply of GTP in any purification buffer, indicative of low off and hydrolysis rates, like those of many small GTPases. The ability of VPS15 to bind to GTP is conferred by the unusual and conserved presence of an Arg at the gatekeeper position that dictates nucleotide specificity. This Arg not only switches purine base specificity from adenine to guanine, it also triggers a rearrangement of the catalytic site by forcing a rotamer change on the Mg²⁺-binding Asp of the kinase DFG motif. This in turn repositions the γ-phosphate such that an in-line attack on an external phosphoacceptor becomes impossible. This coupling between the determinant for purine specificity and γ-phosphate positioning explains why no protein kinases have been described that use only GTP as the phosphodonor. GTP binding to VPS15^PKD is clearly central to the organization of the VPS15^PKD N-lobe (Fig. 5A,B) and, by virtue of the interactions between VPS15^PKD and VPS34^KD (Fig. 2B, C, F, I), to regulation of VPS34. Mutation of the highly conserved residues responsible for GTP binding, Arg103 or Lys171, in conjunction with an unmyristoylated N-terminus essentially eliminates autophagic flux. We interpret this to mean that GTP binding functions in concert with the N-myristate to dock PI3KC3-C1 onto membranes and orient and activate VPS34^KD for PI3P production. On the basis of the cryo-EM structures the VPS15^PKD N-lobe functions to maintain both the inactive and active states. The activating role is the one visibly impacted by mutation, leading to our expectation that VPS34 activation is the most critical and dominant role of VPS15^PKD.

Therapeutic enhancement of autophagy and lysosome biogenesis is a major yet unrealised goal, that may hold promise for currently untreatable neurodegenerative diseases and other aging-associated diseases (32). We resolved the active conformation of one of the rate-limiting signaling complexes in autophagy initiation (PI3KC3-C1) and a key player in both autophagy and lysosome biogenesis via endosome maturation (PI3KC3-C2). We identified two small molecule binding sites: those for GTP and myristate. The sequestration of the myristate in the N-lobe of VPS15^PKD is associated with the inactivated conformation. Therefore, antagonizing the sequestration of the myristate in the N-lobe of VPS15^PKD, if it could be done without loss of stability or the integrity of the VPS34 activating surface of VPS15, would be an attractive path for therapeutic autophagy enhancement through allosteric activation of PI3KC3.

Materials and Methods summary

Protein expression, purification, and characterization

PI3KC3-C1 subunits were co-expressed from pCAG vectors in HEK GNTi cells. PI3KC3-C1 was purified using Strep-Tactin affinity and size-exclusion chromatography. RAB1A proteins were expressed in either E. coli or HEK cells and purified using Ni-NTA affinity chromatography. RAB1A-GTP was prepared by using EDTA to strip bound nucleotides followed by loading with an excess of GTP and MgCl₂. VPS34 lipid kinase activity was assessed via the ATP Glo luminescence assay in the presence of small unilamellar vesicles. Activity was also assayed on giant unilamellar vesicles containing phosphatidylinositol (PI) using a FYVE domain reporter for PI3P production. PI3KC3-C1 was subjected to electrospray mass spectrometry. Details and references are provided in the supplementary material.

Cryo-EM

PI3KC3-C1:RAB1A was vitrified in the presence of n-Octyl-β-D-Glucopyranoside using a Vitrobot. Cryo-EM images were recorded on a 300 kV Titan Krios microscope equipped with an X-FEG and energy filter set to 20 eV. Data were processed in cryoSPARC v3 and density was interpreted using ChimeraX. Initial models were obtained from AlphaFold and iteratively improved manually and with ISOLDE. Details and references are provided in the supplementary material.

MD simulations

Simulations were performed with GROMACS using the CHARMM36m force field. The membrane-docked simulation started with the active conformation of PI3KC3-C1 as determined by cryo-EM, with disordered regions modeled by AlphaFold. The simulation of the activation transition began with the cryo-EM structure of the inactive conformation, which was placed on the membrane by superimposition onto a membrane-engaged active complex after relaxation. For simulation of PI entry into the active site, the VPS34 lipid kinase domain was placed on the membrane in the active orientation derived above. Details and references are provided in the supplementary material.

Cell-based assays

CRISPR/Cas9 was used to generate VPS34 and VPS15 KO HeLa cell lines. Wild-type and mutant VPS34 and VPS15 were re-introduced and used to generate stable cell lines by retroviral transduction. Protein expression was verified by immunoblotting. Starvation-induced autophagic flux was measured by a Halo-based assay. Details and references are provided in the supplementary material.

Supplementary Material

Movie S1

Movie 1.

The activation mechanism of the VPS34 kinase domain. The kinase domain rotates by 140 degrees to expose membrane-binding domains.

Movie S2

Movie 2.

Steered molecular dynamics trajectory of the activation mechanism of the VPS34 kinase domain. The complex was docked onto the membrane in the inactive state. A constant pulling force was applied to the center of mass of the VPS34^KD to separate the domain from VPS15. Upon softening of this interface, a gentle pulling force on the N-myristoyl motif towards the membrane resulted in a quick insertion of the myristoyl chain in the lipid bilayer (time ~30 ns) and subsequent docking of the complex onto the membrane (~50 ns). Water and ions are omitted for clarity, phosphate groups of membrane lipids are shown as grey spheres.

Movie S3

Movie 3.

Simulation of POPI entry into the catalytic site of VPS34. Water and ions are omitted for clarity, phosphate groups of membrane lipids are shown as grey spheres. A time-resolved plot of the distance between the 3’-O atom of POPI and the gamma phosphate of ATP in the course of this simulation is provided in Supplementary Figure 4C.

Data and materials availability:

The cryo-EM maps have been deposited in the Electron Microscopy Data Bank (EMDB) under accession codes: intermediate structure: PDB: 9MHF, EMD-48276 EMD- EMD-48230, EMD-48231, EMD-48232, EMD-48233; inactive structure: PDB: 9MHG, EMD-48278, EMD- EMD-48257, EMD-48258, EMD-48259, EMD-48260; active structure: PDB:9MHH, EMD-48272, EMD-48273. Protocols are available on protocols.io. Plasmids developed for this study are being deposited at Addgene.org. Original gel scans, mass spectrometry RAW data files, HPLC data, GUV images, GUV data, ADP glo data, molecular dynamics trajectories, and codes used to produce plots in this manuscript are deposited at Zenodo.