Introduction

Mammalian cells evolved three classes of phosphoinositide 3-kinases (PI3Ks). The class I and II enzymes evolved in metazoa and are important elements of signal transduction pathways (Burke, 2018; Koch et al., 2021; Kucukdisli et al., 2023; Lo et al., 2023; Lo et al., 2022; Rathinaswamy & Burke, 2020; Vadas et al., 2011; Vogt et al., 2023), but the class III PI3K, VPS34, is the primordial PI3K, which is present in all clades of eukaryotes. VPS34 catalyzes phosphate transfer from ATP to the D3-OH of phosphatidylinositol, producing phosphatidylinositol 3-phosphate (PI3P), which is important for cellular sorting pathways, including maturation of endosomes, phagocytosis, and autophagy. Also, recent studies suggest that VPS34/PI3P-dependent pathways can be hijacked by viruses (Dahmane et al., 2022; Gong et al., 2024; Wang et al., 2021; Williams et al., 2021; Wu et al., 2024). VPS34 is a component in two complexes, VPS34 complex I (VPS34-CI) and VPS34-complex II (VPS34-CII). The complex-specific subunits cause VPS34-CI to localize to autophagosomes and VPS34-CII to endosomes. In mammalian cells, VPS34 complexes have four core subunits. Both complexes share the VPS34, VPS15, and BECLIN1 subunits. In addition, VPS34-CI has ATG14L, while VPS34-CII has UVRAG. In yeast, the homologous complexes have common subunits known as Vps34, Vps15, and Vps30, while the unique subunits are Atg14 and Vps38 (Itakura et al., 2008; Kihara et al., 2001). VPS34-CI and VPS34-CII have been studied by X-ray crystallography and cryo-EM. These studies revealed that the overall structures of both complexes are similarly V- or Y-shaped, with one arm (the kinase arm) composed of most of the VPS34 and VPS15 subunits and the other arm (adaptor arm) consisting of BECLIN1 with ATG14L in VPS34-CI and BECLIN1 with UVRAG in VPS34-CII (Baskaran et al., 2014; Cook et al., 2025; Ma et al., 2017; Rostislavleva et al., 2015; Tremel et al., 2021; Young et al., 2016; Young et al., 2019).

We recently found that organelle-specific recruitment of VPS34 complexes is achieved by RAB GTPases: ER/autophagosome-localized RAB1A interacts with and activates VPS34-CI, while early endosome-localized RAB5A interacts with and activates VPS34-CII (Tremel et al., 2021). The ‘Switch’ regions in the RAB proteins undergo substantial conformational changes when switching from GDP- to GTP-bound states (Pylypenko et al., 2018). RAB effectors interact with these regions and selectively recognize RABs in a nucleotide-dependent manner (Grosshans et al., 2006). A triad of invariant hydrophobic residues F, W, and Y located in Switch 1, Interswitch, and Switch 2 regions, respectively, can adopt different conformations to specifically bind effectors (Merithew et al., 2001).

The tomographic reconstruction of VPS34-CII on membranes provided a framework for understanding the mechanisms of VPS34-CII activation by RAB5A (Tremel et al., 2021), and subsequent single-molecule kinetic studies were consistent with RAB5A activating VPS34-CII by both recruitment and allosteric activation (Buckles et al., 2020b). To improve the resolution of VPS34-CII, relative to the tomographic reconstruction, we determined high-resolution single-particle cryo-EM structures of VPS34-CII bound to RAB5A. Remarkably, we find that VPS34-CII-RAB5A complex has two independent binding sites: one on VPS34 and one on the helical solenoid of VPS15. However, only the VPS34 site was detected in the tomographic reconstruction.

This VPS34 site (Cook et al., 2025; Tremel et al., 2021) has three walls (a tripartite binding site) made of the VPS15 small globular domain (SGD, residues 723-805), the VPS15 WD40 (residues 968-1358), and the helical hairpin insertion in the VPS34 C2 domain (VPS34 C2HH, residues 170-210). We previously attempted to eliminate RAB binding to VPS34 complexes by structure-guided mutation of the 199-REIE-202 stretch in VPS34 C2 HH helix 2 to AAAA (REIE>AAAA). VPS34-CI REIE>AAAA could neither bind nor be activated by RAB1A, while, surprisingly, the same mutation in VPS34-CII increased affinity for and activation by RAB5A (Tremel et al., 2021). The starkly different effects of this mutation on the two complexes remained unexplained by the cryo-ET study of VPS34-CII, but can be explained by the new high-resolution single-particle cryo-EM structures.

Results

High resolution structure of a mutant VPS34-CII bound to RAB5A

Using a confocal microscopy-based bead-binding assay, we showed that mCherry-tagged VPS34-CII bound specifically to RAB5A (Figs. 1A and B, Figs. S1A to S1C). Interaction of VPS34-CII with RAB1A was too weak to measure (Fig. S1D). Consistent with our previous activity assays, the VPS34-CII REIE>AAAA mutation in the VPS34 C2 domain bound RAB5A about three times more tightly than the WT complex (Figs. 1B and S1E) (Tremel et al., 2021). Because the VPS34-CII REIE>AAAA mutant bound more tightly to RAB5A than WT VPS34-CII (Fig. 1B), we used it for our single-particle cryo-EM sample of the VPS34-CII-RAB5 assembly, then built and refined a structure at a 3.2 Å resolution (Figs. S2A, B). The structure of the mutant bound to RAB5A is remarkably similar to VPS34-CII-BATS-RAB5A on membranes that we determined previously at low resolution by cryo-ET (Tremel et al., 2021). There were only small changes in kinase and adaptor arms of the complex (Fig. S2C). The VPS15 helical solenoid was the best resolved region of the complex, while N- and C-terminal extremities of the BECLIN1 and UVRAG subunits were less well resolved. RAB5A bound to a tripartite site composed of VPS15 SGD, VPS15 WD40 and VPS34 C2HH (Figs. S2D and E), which is consistent with our previous results (Tremel et al., 2021). We used this mutant complex as an initial model for building the structure of wild type VPS34-CII-RAB5A (see below). One surprising feature of the mutant reconstruction was that there was weaker density that resembled a second molecule of RAB5A bound to the helical solenoid of VPS15 (Fig. S2A). A model was built into this RAB5 density at partial occupancy, but this second site was even more apparent in the structure of a mutant in which the VPS34 RAB5 site was mutated (see next section).

Figure 1.

Structures of RAB5A-binding mutants suggest that VPS34-CII has an unexpected second RAB5A binding site on the VPS15 subunit

To examine the structural contribution of the VPS34 C2 HH to RAB5A binding, we introduced a reverse charge mutation in VPS34, 199-REIE-202 to 199-ERIR-202 (REIE>ERIR), to disrupt interactions between RAB5A Switch 1 and VPS34 C2 HH helix 2. Interestingly, the GUV kinase assay showed that although the reverse charge mutation did not eliminate VPS34-CII activation by RAB5A, the extent of activation was reduced two-fold compared to wild type VPS34-CII (Fig. 1C). Compared to wild type VPS34-CII, the reverse charge REIE>ERIR mutant VPS34-CII had a RAB5A binding affinity that was drastically reduced, a K_D of 27± 8.3 µM (Fig. 1D, Fig. S1F). To investigate whether the partial reduction in VPS34-CII REIE>ERIR activation by RAB5A was due to incomplete disruption of the VPS34-RAB5A interaction, we determined the structure of a VPS34-CII reverse charge mutant in the presence of RAB5A (Fig. 1E). As expected, we did not observe any density in the tripartite RAB binding site. However, consistent with the weak density on the VPS15 subunit in our structure of VPS34-CII with the VPS34 REIE>AAAA mutation, we saw strong density for RAB5A bound to the VPS15 subunit of VPS34-CII REIE>ERIR. We fit and refined an atomic model for RAB5A-GTP into the new density (Fig. 1F) on the VPS15 helical solenoid domain, with an interaction centered on the helix spanning VPS15 residues 570-585. RAB5A mediates this interaction mostly through its hydrophobic triad residues F57, W74, and Y89 (Fig. 1G). The Interswitch residue W74 contacts VPS15 residues S579 and H580, as well as the Switch 1 residue F57. In turn, F57 also interacts with VPS15 H580, and its nearby residue T583. The Switch 2 residue Y89 is not in a direct contact with any VPS15 residues and instead binds to other RAB5A residues. This contrasts with the RAB5A binding to VPS34 C2 HH, where Switch 1 F57 has a stabilizing role for the switch, but does not make direct interaction with the effector-binding site. It is evident that RAB5A has generous plasticity in its hydrophobic triad region that allows binding to different molecular surfaces.

To verify that RAB5A binding the VPS15 site was not an artefact of cryo-EM sample preparation, we used hydrogen – deuterium exchange mass spectrometry (HDX-MS) and examined RAB5A binding to the wild-type VPS34-CII in solution (Fig. S3A). We observed strong protection upon RAB5A binding not only in the VPS34 C2 HH helix 2 (Fig. S3B), but also in the VPS15 helical solenoid domain, including helix 570-585 (Fig. S3C). Protection in the VPS34 RAB5A binding site is consistent with our previous HDX-MS results (Tremel et al., 2021), and protection on the VPS15 helical solenoid residues 570-585 indicates that RAB5A binds the VPS15 site both in solution and in the cryo-EM sample.

Wild type VPS34-CII can bind two copies of RAB5A simultaneously

We next determined the structure of wild-type VPS34-CII-RAB5A to further confirm the presence of the new RAB5A binding site on VPS15. After several rounds of heterogeneous refinement and 3D classification, we found that 30% of particles were apo VPS34-CII, 23% of particles had RAB5A bound to VPS34 only (VPS34-CII-RAB5A_VPS34), 32% of particles had RAB5A bound to VPS15-only (VPS34-CII-RAB5A_VPS15), and 15% of particles had RAB5A bound to both VPS34 and VPS15 (VPS34-CII-(RAB5A)₂) (Fig. 1H). We subjected these particles to numerous rounds of focused 3D classifications to ensure that the density of VPS34-CII-(RAB5A)₂ originated from particles having two bound RAB5As and was not an artefact of combining VPS34-CII-RAB5A_VPS15 particles with VPS34-CII-RAB5A_VPS34 particles. Apart from the binding of two RAB5As, the structure of VPS34-CII-(RAB5A)₂ did not show significant conformational differences from the other RAB5-bound VPS34-CII structures, nor from the apo form. This indicates that VPS34-CII can interact with two RAB5A GTPases simultaneously, via two independent binding sites. The structure of the VPS34 REIE>AAAA mutant that bound RAB5 tighter than the WT VPS34 also has a second RAB5 binding site on VPS15, however, it is likely that the population of complexes in the cryo-EM sample contained both particles with and without RAB5A bound to VPS15, since the average density in the VPS15 RAB5 binding site was only about 40% of the average density in the VPS34 RAB5 binding site. Models were built and refined for all RAB5-associated VPS34-CII assemblies.

To test whether the RAB5A binding site on VPS15 has a role on VPS34-CII activation, we mutated the VPS15 579-SHMIT-583 stretch into 579-DDMIE-583, with the expectation that the charged, bulkier aspartic and glutamic acid residues would disrupt the RAB5A hydrophobic triad interaction with VPS15. As expected, the mCherry-tagged VPS34-CII VPS15 mutant could be activated by RAB5A but had about two-fold less activity than the wild-type VPS34-CII (Fig. 2A). The VPS34-CII VPS15 mutant had a K_D of 6.0 ± 0.1 µM for RAB5A (Fig. 2B), which is comparable to wild type VPS34-CII, where K_D = 7.4 ± 1.1 µM. To completely abolish RAB5A binding and activation of VPS34-CII, we combined the VPS34 REIE>ERIR and VPS15 SHMIT>DDMIE mutations. We verified that all the single and double mutants were structurally intact by nanoDSF (Fig. S4). The VPS15 + VPS34 double mutant VPS34-CII could not be activated by RAB5A (Fig. 2C), and we observed no binding of RAB5A to the double mutant complex II – coated beads (Fig. 2D). When the RAB5 binding sites in VPS34 and VPS15 are aligned with various species, the VPS34 RAB5-binding sequence REIE is not conserved in yeast S. cerevisiae and S. pombe, where it is rather reverse-charged (Fig. 2E). In contrast, the VPS15 RAB5-binding sequences are highly conserved through evolution. Indeed, the yeast Vps15 RAB5 non-binding mutant (SHLITY>DDLIEY) in yeast VPS34-CII was sufficient to abolish the binding to the yeast RAB5 orthologue, Vps21 in vitro (Fig S5). Also, the yeast vps15 SHLITY>DDLIEY mutant did not show strong colocalization with Vps21 (Fig. 2F), and the same mutant is defective in endocytic sorting measured by CPY processing (Fig. 2G). Taken together, these results show that RAB5A binds to two sites on human VPS34-CII, both sites are required for full activation, and that two RAB5A GTPases can bind VPS34-CII at the same time. It is likely that RAB5 binding in eukaryotic VPS34-CII originated at the VPS15 site, with the second binding site on VPS34 having been acquired during evolution.

Figure 2.

RAB5A binding to VPS34 site does not induce conformational changes in VPS34-CII

Cryo-EM analysis of a sample of wild-type VPS34-CII-RAB5A contained both the apo and RAB5A-bound forms of VPS34-CII, with about 30% of particles contributing to a reconstruction for apo VPS34-CII density at 3.9 Å resolution (Fig. S6A), and 23% of particles contributing to the structure of VPS34-CII-RAB5A at 4.0 Å resolution (Fig. 3A, right; Fig. S6B). We fit and refined atomic models into both densities and compared apo and RAB5A-bound forms of VPS34-CII. Similarly to what was reported for RAB1A binding to VPS34-CI (Cook et al., 2025), RAB5A binding to VPS34-CII did not induce large conformational differences (Fig. S6C). Furthermore, the WT VPS34-CII-RAB5A structure did not differ greatly from VPS34-CII REIE>AAAA-RAB5A, nor a reinterpreted VPS34-CII-BATS-RAB5A model based on our cryo-ET data previously reported (Tremel et al., 2021) (Figs. S6D and E).

Figure 3.

RAB specificity for VPS34 complexes arises from RAB Switch regions and from the VPS34 C2 HH conformation

We compared VPS34-CI bound to RAB1A (VPS34-CI-RAB1A) (Cook et al., 2025) with the VPS34-CII-RAB5A structures (Fig. 3). Both RABs bind VPS34 complexes in the same tripartite site with similar orientations (Fig. 3A). The most prominent element of the RAB binding site is the VPS34 C2 HH. In VPS34-CI, both helices of the C2 HH engage RAB1A (Fig. 3B, left), while in VPS34-CII, only one VPS34 C2 HH helix engages RAB5A (Fig. 3B, right). Also, in VPS34-CI, the VPS15 WD40 domain and RAB1A are separated by 27 Å (Fig. 2B, left), while in VPS34-CII the distance between equivalent points in RAB5A and WD40 is only 9 Å (Fig. 3B, right). These differences in the tripartite binding pocket are not due to RAB binding, since WT apo VPS34-CII has the same overall conformation as the RAB5-bound conformation (Fig. S6C). Instead, the difference in the binding site appears to arise from the influence of the UVRAG subunit in VPS34-CII instead of the ATG14L subunit of the VPS34-CI. The more open RAB pocket adjacent to the VPS15 WD40 in VPS34-CI suggests that the WD40 might make a larger contribution to RAB binding in complex VPS34-CII than in complex VPS34-CI, indicating that difference in the geometry of the tripartite site in the two complexes is the most important contribution to the RAB specificity.

In the VPS34 RAB-binding site, there are considerable differences between VPS34-CI and VPS34-CII in the interaction between the VPS34 C2 HH and the RAB hydrophobic triad, which is made of invariant hydrophobic residues F, W, and Y located in the Switch 1, Interswitch, and Switch 2 regions (Fig. 3C). While the same elements are involved in both RAB-VPS34 complexes, they are arranged differently. In RAB1A (Fig. 3C, left), Switch 1 residue F48 forms a cation-π stacking interaction with residue R195 in VPS34 C2 HH helix 2. Residue Y80 in Switch 2 interacts with R183 in VPS34 C2 HH helix 1. The Interswitch residue W65 does not engage VPS34 C2 HH, but instead contacts RAB1A residue F48 and others (Fig. S7A). In contrast, for the VPS34-CII-RAB5A structure (Fig. 3C right), Interswitch residue W74 forms a cation-π stacking interaction with the VPS34 R195. In this complex, VPS34 R195 is also engaged by Y89 in the RAB5A Switch 2. RAB5A Switch 1 residue F57 does not directly participate in VPS34 binding and instead forms an intramolecular π-π stacking interaction with RAB5A W74.

In both RAB1A and RAB5A, the Switch 1 region is close to VPS34 C2 HH helix 2 (Fig. 3D). In VPS34-CI-NRBF2-RAB1A, Switch 1 residue D47 forms a salt bridge with VPS34 R199, which, in turn, interacts with VPS34 E202 (Fig. 3D, left). The importance of the VPS34 R199 - RAB1A D47 interaction is supported by our previous finding that the VPS34 REIE>AAAA mutation abolishes VPS34-CI recruitment and activation by RAB1A (Tremel et al., 2021). In VPS34-CII-RAB5A, Switch 1 of RAB5A is also positioned close to helix 2 of the VPS34 C2 HH (Fig. 3D, right). However, instead of potential side chain hydrogen bonds or salt links between these elements, RAB5A has a non-polar A56 residue in the position equivalent to RAB1A D47 (Fig. S8), and it is bordered by aliphatic residues I53, G54, and A55. This aliphatic RAB5A Switch 1 region interacts more favorably with the non-polar VPS34 AAAA mutant than with the polar REIE motif in WT VPS34 C2 HH helix 2 (Fig. S2E), resulting in increased activation of VPS34-CII REIE>AAAA by RAB5A.

The Switch 2 region in both RABs interacts with different VPS34 C2 HH helices (Fig. 3E). When bound to VPS34-CI, the flexible RAB1A Switch 2 loops around the positively charged R183 in VPS34 C2 HH helix 1, forming polar contacts with it via residue S78 and the backbone carbonyl oxygens of A68 and F73 (Fig. 3E, left), with distances ranging from 2.8 Å to 3.4 Å, suggesting potential hydrogen bonding (Fig. S7B). This conformation of Switch 2 contrasts with a more helical conformation in other RAB1A-effector complexes (Cheng et al., 2012; Dong et al., 2012; Mishra et al., 2013). In a complex with VPS34-CII, RAB5A Switch 2 adopts a compact helical conformation stabilized by an intramolecular interaction between residues P87 and Y90. The M88 residue in this helical Switch 2 makes a non-polar interaction with F198 in VPS34 C2 HH helix 2 (Fig. 3E, right). Interestingly, RAB1A S78 and RAB5A P87 share the same position in the multiple sequence alignment (Fig. S8), suggesting that the function of this position is to stabilize a RAB-specific conformation of the Switch 2 region.

Overall, to bind VPS34-CII, RAB5A uses two interfaces (Switch 1 and the hydrophobic triad) and relies on the tripartite site in VPS34-CII. In contrast, RAB1A binding to VPS34-CI involves three interfaces (Switch 1, Switch 2, and the hydrophobic triad) that make polar and nonpolar interactions with both VPS34 C2 HH helices, but does not appear to contact other elements of the VPS34-CI tripartite site.

VPS34-CII has a zinc finger at the N-terminus of BECLIN1

Our higher-resolution structures of human VPS34-CII showed additional features at the N-terminal end of BECLIN1 that were previously not apparent. This BECLIN1 region has two highly conserved CXXC motifs (18-CXXC-21 and 137-CXXC-140) flanked by an intrinsically disordered region (IDR) (Mukhopadhyay et al., 2024), which has hindered detailed studies of the region (Chen et al., 2023; Rostislavleva et al., 2015; Young et al., 2019). Our cryo-EM densities allowed us to dock and refine atomic models of the BECLIN1 N-terminus predicted by AlphaFold3 (Abramson et al., 2024). In VPS34-CII there is a Zn²⁺ coordinated by the two BECLIN1 CXXC motifs to form a zinc finger (Fig. S9). In addition, the N-terminal C2 domain of UVRAG could be clearly built. The structures shows that the C2 domain is sandwiched between the N-terminal zinc finger of BECLIN1 and the long helical hairpin in the middle of the VPS15 solenoid (residues 465-545) (Fig. 4). There was a density at the edge of strand β2 of the UVRAG C2 domain that appeared to augment the second β-sheet of the C2 domain. The Alphafold model also had an augmented β-sheet and we have attributed this augmenting strand to BECLIN1 residues 78-83. This mostly ordered and relatively compact N-terminal region of BECLIN1 and UVRAG contrasts with the analogous region in VPS34-CI that was almost completely disordered (Cook et al., 2025). The N-terminal region of BECLIN1 contains the BH3 motif of BECLIN1 (residues 105-130) to which Bcl-2 family proteins can bind to inhibit autophagy. It is not immediately clear how Bcl-2 binding would inhibit the VPS34 complexes, but the BH3 forms one of the two contacts between the UVRAG C2 domain and other subunits of the VPS34-CII (purple helix in Fig. 4).

Figure 4.

Discussion

Our single-particle cryo-EM structures of assemblies of human VPS34-CII-RAB5A showed unanticipated aspects of VPS34 regulation. The VPS34-CII mutant VPS34 REIE>AAAA mutant bound RAB5A with a greater affinity than WT VPS34-CII and enabled the highest resolution VPS34-CII structure (3.2 Å). We have also determined apo structures of wildtype VPS34-CII at a resolution of 3.9 Å. These structures allowed us to delineate the molecular basis for RAB specificity of VPS34 complexes and determine that the differences in the overall shape of the VPS34 RAB binding pocket is not a consequence of RAB binding but a consequence of the complex-specific subunit (UVRAG versus ATG14L). We previously found that VPS34-CII was not activated by RAB1 but that VPS34-CI is weakly activated by RAB5 in vitro (Tremel et al., 2021). This differential selectivity may arise from the tripartite RAB5 binding site in VPS34-CII being narrower than the equivalent site in VPS34-CI. This may restrict the geometry of RAB binding in a way that would only allow RAB5A, but not RAB1A, to fit into the tripartite site of VPS34-CII.

One subunit difference between the two apo complexes causes conformational changes in the VPS34 tripartite RAB-binding site. Most importantly, in VPS34-CI, the C2 HH is positioned with both helices available for RAB1A, while in VPS34-CII only one helix from the C2 HH is accessible for RAB5A binding. The diminished interaction with the C2 HH in the VPS34-CII-RAB5A assembly appears to be compensated by a narrowing of the tripartite RAB5A pocket with additional, non-switch interactions between RAB5A and VPS15. Remarkably, the complex-specific ATG14L and UVRAG subunits do not come in direct contact with RABs. Given that the apo-VPS34-CII closely resembles the RAB5A-VPS34-CII, it appears that ATG14L tunes binding to RAB1A by twisting the common subunits VPS34, VPS15 and BECLIN1 into a RAB1-specific conformation.

RAB1A and RAB5A bind VPS34-CI and -CII via the Switch and Interswitch regions. In both RAB-VPS34 complex assemblies, the RAB hydrophobic triad is positioned at the linker between the VPS34 C2 HH helices and engages it in a similar manner. In contrast, the rest of the RAB Switch regions make VPS34 complex-specific interactions. RAB1A Switch 1 residue D47 particularly stands out, as it forms a salt bridge with VPS34, while in RAB5A Switch 1 is mostly hydrophobic and forms entirely different contacts with VPS34. The hydrophobic triad likely provides most of the binding energy, and the Switch regions confer specificity, as has been observed with other RAB-effector complexes (Lucas et al., 2014). The distinctive, hydrophobic RAB5A Switch 1 region interaction with the VPS34 C2 HH also explained why the VPS34-CII RAB5A binding site mutation REIE>AAAA, which was meant to inhibit RAB5A binding, increased the binding affinity and activation of VPS34-CII (Tremel et al., 2021).

Our VPS34 complex structures showed an N-terminal VPS15 myristoylation and a GDP bound to the VPS15 kinase domain (Fig. S10). The myristoyl-binding pocket is in the in the vicinity of the interface with VPS34 (Fig. S10). It was proposed that VPS34 is activated when the myristoyl moiety is ejected from the VPS15 N-lobe and inserts into a lipid membrane, however, the same study also showed that a non-myristoylatable mutant (G2A) is fully functional (Cook et al., 2025), which is consistent with a previous yeast Vps15 study (Herman et al., 1991). Therefore, the precise role of the VPS15 myristoyration in activation remains to be elucidated. The myristoyl group inserting into membranes would increase association with membranes, but it also has been shown that RAB5 makes a very prominent contribution the association with membranes (Buckles et al., 2020b).

What Cook et al. (Cook et al., 2025) designated as the ‘inactive’ conformation (PDB ID 9MHG) resembles the conformation that we report here for our cryo-EM studies of VPS34-CII. We did not observe in any dataset the minor ‘active’ conformation previously reported (Cook et al., 2025), which involved a 140° rotation of the VPS34 Helical-Catalytic (HELCAT) unit (PDB ID 9MHH). It may be that the VPS34-CII is poised for activation but not yet activated under the conditions in our cryo-EM study. It is possible that the transition state on a membrane is sufficiently short-lived that we have not been able to capture it, even though we have RAB5 bound in the single-particle structures in the absence of membranes and RAB5 bound in the presence of membranes from our previous cryo-ET study (Tremel et al., 2021).

Surprisingly, we discovered a new RAB5A binding site on the VPS15 subunit of VPS34-CII. Like the previously known VPS34 site and other RAB-effector interactions (Eathiraj et al., 2005; Hou et al., 2011; Lucas et al., 2014), this newly discovered VPS15 site has micromolar-range affinity for VPS34-CII. In addition, the VPS15 RAB5 binding site is α-helical, which also is typical in RAB effectors (Kawasaki et al., 2005; Rai et al., 2019). The VPS15 site consists of a single α-helix while the VPS34 site has two helices.

Altogether the avidity of VPS34-CII for RAB5 is dictated by interactions in three disparate regions: helix 2 of the VPS34 helical insertion in the C2 domain, the VPS15 WD40 in the tripartite binding site, and the VPS15 solenoid interaction with separate RAB5. In contrast, VPS34-CI interacts with RAB1 using only a single region, the helical insertion in the VPS34 C2 domain. VPS34-CI appears to compensate for lack of the other two regions by interacting with both helices of the helical insertion in the VPS34 C2 domain. This is enabled by the unique bend of the ATG14L-containing adaptor arm that shifts the helical insertion to bring both helices into contact with bound RAB1.

In contrast to the VPS34 site, RAB5A binding to VPS15 does not involve extensive specific interactions through the Switch regions beyond the hydrophobic triad. This might account for the lower affinity of the VPS15 binding site compared to the VPS34 site. RAB1A does not bind or activate VPS34-CII, suggesting either that the conformation of the hydrophobic triad and the Switch regions in RAB1A are not complementary to VPS15 helix 570-585, or that the VPS15 RAB-binding site is blocked by other elements in VPS34-C1. RAB5 binding assays are consistent with a higher affinity VPS34 site and a lower affinity VPS15 site binding independently. However, functional assays indicate that each RAB5A binding site contributes to activation equally. The lack of synergy in the binding assays, despite clear synergy in activity assays is consistent with the complex binding to two RAB5A molecules on the same membrane. Such synergy could not be observed in the binding assays because of the way they were conducted: VPS34-CII was bound to beads and soluble RAB5A was titrated. To observe the synergy that we observe on membranes would require the RAB5 be fixed to beads, while titrating VPS34-CII, and this arrangement would be impossible because of the limited solubility of the VPS34-CII. How the two RAB5A binding sites in VPS34-CII are coordinated in mammals remains to be seen.

There are only a few other cases where multivalent RAB effectors contain binding sites with different affinities (Rai et al., 2019). For example, MICAL1, a cytoskeleton regulatory protein, contains two RAB8 binding sites, one with high and one with low affinity for RAB8. The higher affinity site is used for initial MICAL1 recruitment to membranes, while the lower affinity site binds later (Rai et al., 2024). The two RAB5A binding sites on VPS34-CII might be a part of a similar mechanism that may control VPS34 activity on early endosomes. VPS34-CII would be fully activated only when present on highly concentrated RAB5A membrane domains and in absence of other RAB5A effectors that might bind RAB5A more tightly than VPS34-CII VPS15 site. On the other hand, VPS34-CII may cooperate with RAB5A effectors EEA1, Rabaptin-5, and Rabenosyn-5 to maintain early endosomal RAB5A domains that are required for endosome tethering and fusion (Zerial & McBride, 2001). In addition, VPS34-CII could also be involved in vesicle tethering of two RAB5 proteins from different vesicles bound a single VPS34-CII. More sensitive experiments such as single molecule analysis will be required to investigate these possibilities. Finally, two RAB5A molecules might be required to correctly orient the VPS34 complex on the membrane, as both RAB binding sites are on the same side of the complex. While we did not observe two RAB5A proteins bound to VPS34-CII-BATS in the cryo-ET structure (Tremel et al., 2021), the lower occupancy of RAB5A on vesicles used for the cryo-ET structure on membranes might have meant that the VPS15 RAB5A binding site was not occupied.

Interestingly, multiple sequence alignments suggest that yeast Vps34-CII has a single RAB5 binding site, and this would be on the Vps15 subunit. As with human RAB5, the yeast RAB5 orthologue Vps21 is required in endocytic trafficking (Singer-Kruger et al., 1994), and the RAB1 homolog Ypt1 is involved in autophagy initiation (Yao et al., 2024). Furthermore, Rab-pulldown experiments identified direct or indirect interactions with Vps15 (Graef et al., 2013; Nagano et al., 2024). We showed that disrupting the Vps21 binding site in yeast Vps15 abolished the binding of Vps21 to yeast complex II in vitro, and inhibits endocytic sorting. It is possible that the RAB binding site on VPS15 evolved first, and the second RAB binding site on VPS34 evolved in higher organisms, as the complexity of membrane compartments increased the demand for tighter spatial and temporal regulation of PI3P production. It will be interesting to investigate how these RAB5A binding sites contribute to VPS34-CII function in human cells.

Methods

Lipid mixtures

Lipids

Plasmids

Protein purification

Human VPS34 Complex II

To prepare VPS34-CII, Expi293 suspension cells (ThermoFisher A14527) were grown in Expi293 Expression Medium (ThermoFisher A1435102) at 37°C, 8% CO₂, shaking at 125rpm. At density of 2x10⁶/L, cells were transfected with 1.1 mg DNA/L cells using polyethyleneimine ‘MAX’ (Polysciences 24765, 1mg/mL in PBS) at 3mg/L cells as the transfection reagent. Cells were grown further in the same conditions as above for 68 hours, then harvested by centrifugation at 3000 g for 15 min, flash frozen in liquid nitrogen, and stored at -80 °C until use.

To purify the protein, cells were resuspended in 50 mL/2 L cells of lysis buffer (50 mM HEPES pH 8.0, 300 mM NaCl, 1% Triton X-100 (Fisher BioReagents BP151), 12% glycerol, 2 mM MgCl₂, 1 mM phenylmethylsulfonyl fluoride (PMSF, Thermo Fisher Scientific 36978), 0.5 mM tris(2-carboxyethyl)phosphine (TCEP), and 1x EDTA-free protease inhibitor cocktail tablet (Roche 11873580001)) and incubated with slow stirring. Lysate was cleared by centrifugation at 142,000 g for 30 min at 4 °C in a Type 45 Ti rotor (Beckman Coulter) and filtered through a 5 µm syringe filter (Sartorius 17594). Cleared lysate was incubated for 3.5 hours at 4 °C and 8 rpm rotation with 3 mL IgG beads (Cytiva 17-0969) equilibrated with lysis buffer. Beads were then transferred to a gravity column, washed with 50 mL ice-cold wash buffer (50 mM HEPES pH 8.0, 300 mM NaCl, 0.1% Triton X-100, 12% glycerol, 50 mM MgCl₂, 0.5 mM TCEP, 4 µg/mL bovine pancreatic RNase (Sigma 83834)), and incubated with 50 mL wash buffer supplemented with 5 mM ATP for 20 minutes. Beads were further washed with another 50 mL wash buffer and 100 mL TEV buffer (50 mM HEPES pH 8.0, 300 mM NaCl, 0.5 mM TCEP), then resuspended in 10 mL TEV buffer and incubated with 40 µL of 4.4 mg/mL TEV protease (house made) overnight without rotation to remove the C-terminal ZZ-tag on VPS15. Flow-through was collected and remaining protein eluted three times with 5 mL TEV buffer followed by 7 mL TEV buffer supplemented with 0.1% CHAPS (Calbiochem 220201). Eluted fractions were combined, passed through a 0.22 µm syringe filter (Elkay E25-PS22-50S), concentrated to 0.5 mL using a 100k MWCO concentrator (Millipore UFC810024), and injected onto an S200 10/300 GL gel filtration column equilibrated with 20 mM HEPES pH 8.0, 300 mM NaCl, 0.5 mM TCEP. Eluted fractions were inspected by SDS-PAGE, selected fractions were pooled, concentrated, frozen in liquid nitrogen in 2 µL aliquots, and stored at -80°C until use. Mutant VPS34 complexes and complexes with an N-terminal mCherry tag on BECLIN1 were purified in the same way as wild type complexes.

p40^phox PX domain purification and labelling

To prepare the p40^phox PX domain, plasmid pYO1645 was transformed into E.coli OverExpress™ C41(DE3) chemically competent cells (Lucigen 60442). Transformed bacteria were cultured in 6 L 2xTY medium supplemented with 0.1 mg/mL ampicillin (Formedium AMP50) at 37 °C until OD₆₀₀=0.8, then expression was induced with 0.5 mM IPTG, and bacteria were incubated for further 19 hours at 18 °C. Cells were pelleted by centrifugation at 6700 g for 25 minutes and pellets were resuspended in 100 mL lysis buffer (50 mM HEPES pH 8.0, 150 mM NaCl, 12% glycerol, 10 mM DTT, 1 mM PMSF, 1x EDTA-free protease inhibitor cocktail tablet). Cells were sonicated for 5 min on ice (2 s on, 3 s off, 60% amplitude), and lysate cleared by centrifugation at 142,000 g for 15 min at 4 °C in a Type 45 Ti rotor. Supernatant was filtered through a 5 µm syringe filter and incubated with 4 mL GS4B-GST beads (Cytiva 17-0756) equilibrated in lysis buffer for 1 hour at 4 °C at 8 rpm rotation. Beads were then transferred to a gravity flow column, washed with 50 mL lysis buffer, 200 mL wash buffer (50 mM HEPES pH 8.0, 300 mM NaCl, 12% glycerol, 10 mM DTT), and 50 mL TEV buffer (50 mM HEPES pH 8.0, 300 mM NaCl, 2 mM TCEP). Beads were resuspended in 10 mL TEV buffer, then 80µL of TEV protease at 4.4 mg/mL was added, and sample were incubated overnight without rotation to cleave the N-terminal GST-tag. Flow-through and four 5 mL elutions with TEV buffer were pooled and supplemented with 20 mM imidazole, then filtered through a 0.22 µm syringe filter and incubated with 0.5 mL Ni-NTA agarose beads (Qiagen 1018244) equilibrated with TEV buffer for 45 minutes at 4 °C at 8 rpm rotation to remove the His₆-tagged TEV protease. Ni-NTA beads were transferred to a gravity flow column and the collected flow-through was again filtered through a 0.22 µm syringe filter and supplemented with 5 mM TCEP. Protein was concentrated to 1 mL using a 10k MWCO concentrator (Millipore UFC901024) and injected onto an S75 16/60 gel filtration column equilibrated in 20 mM HEPES pH 8.0, 150 mM NaCl, 1 mM TCEP. The peak fractions were pooled and concentrated to 31.5 mg/mL.

To label the p40^phox PX domain, 1 mg of AlexaFluor 647 C2 maleimide (Invitrogen A20347) was dissolved in 104 µL 100% DMSO (Fisher Bioreagents BP231) for a final concentration of 7.7 mM. Protein and dye were mixed in labelling buffer (20 mM HEPES pH 7.0, 200 mM KCl, 1 mM TCEP), with final concentration of p40^phox PX domain at 112 µM and AF647 at 224 µM and incubated for 2 hours at room temperature with slow rotation. Reaction was quenched with 1 mM DTT and loaded onto a 5 mL HiTrap Heparin HP column (Cytiva 17-0406-01) equilibrated with 20 mM HEPES pH 7.4. The column was washed with 65 mL 20 mM HEPES pH 7.4 buffer and eluted with a gradient of Hep-B buffer (20 mM HEPES pH 7.4, 1M KCl). Peak fractions were pooled and concentrated to 168 µM with 90% labelling efficiency.

Human RAB1A purification and labelling

For maleimide labelling of human RAB1A, the pJB78 plasmid (GST-TEV-HsRAB1A Q70L C26S C126S,1-204) (Tremel et al., 2021) was used. To purify the protein, pJB78 was transformed into E.coli C41(DE3)RIPL cells, 2L of transformed bacteria were cultured in 2xTY medium supplemented 0.1 mg/mL ampicillin at 37 °C until OD₆₀₀=0.6, then induced with 0.2 mM IPTG and grown for further 18 hours at 18 °C. Cells were pelleted by centrifugation at 6700 g for 25 min, and pellets resuspended in 50 mL lysis buffer (25 mM HEPES pH 8.0, 200 mM NaCl, 1 mM TCEP, 0.5 mg/mL lysozyme (Sigma L6876), 0.05 µL/mL universal nuclease (Pierce 88702)). Cells were sonicated for 6 minutes (10 s on, 10 s off, 60% amplitude), and lysate cleared by centrifugation at 104,000 g for 45 minutes at 4 °C in a Type 45 Ti rotor (Beckmann Coulter). After filtering through a 0.45 µm syringe filter, lysate was incubated with 4 mL GS4B-GST beads equilibrated with lysis buffer for 1 hour at 4 °C at 8 rpm rotation. Beads were transferred to a gravity flow column and washed with 100 mL lysis buffer, then 100 mL wash buffer (20 mM HEPES pH 8.0, 300 mM NaCl, 5% glycerol, 1 mM TCEP). Beads were resuspended in 10mL TEV buffer (25 mM HEPES pH 8.0, 150 mM NaCl, 1 mM TCEP), 60 µL of TEV protease at 4.4 mg/mL was added, and sample was incubated overnight at 4 °C without rotation. Flow-through and five 5 mL TEV buffer elutions were collected and diluted with Hep-A buffer (25 mM HEPES pH 8.0, 1 mM TCEP) to a final NaCl concentration of 50-100 mM, then applied to a 5mL HiTrap Heparin HP column. The flow-through fraction with protein was concentrated and incubated with 16x molar excess of GTP (Jena Bioscience NU-1012) or GDP (Sigma G7127) and 27x molar excess EDTA for 90 minutes on ice, followed by addition of 55x molar excess MgCl₂ and further incubation for 30 minutes on ice. Protein was applied onto an S75 16/60 gel filtration column equilibrated with 25 mM HEPES pH 7.0, 150 mM NaCl. Main peak was pooled and concentrated to 816 µM.

To label RAB1A for affinity binding assays, 1 mg of AlexaFluor 488 C5 maleimide (Invitrogen A10254) was dissolved in 138.8 µL 100% DMSO (Fisher Bioreagents BP231) for a final concentration of 10 mM. Protein and dye were mixed in labelling buffer (20 mM HEPES pH 7.0, 200 mM KCl, 1 mM TCEP), with final concentration of RAB1A at 112 µM and AF488 at 224 µM and incubated for 2 hours at room temperature with slow rotation. Reaction was quenched with 1 mM DTT and applied to an S75 16/60 gel filtration column equilibrated with 20 mM HEPES pH 7.0, 300 mM NaCl. Peak fractions were pooled and concentrated to 88 µM with 14% labelling efficiency.

Human RAB5A purification

For maleimide labelling of human RAB5A, the pOP823 plasmid (HsRAB5A Q79L C19S,C63S, 1-212)(Tremel et al., 2021) was used. To purify the protein, pOP823 was transformed into E.coli OverExpress™ C41(DE3) cells, 4L of transformed bacteria were cultured in 2xTY medium supplemented with 0.1 mg/mL ampicillin at 37 °C until OD₆₀₀=0.6, then induced with 0.5 mM IPTG and grown for further 20 hours at 16 °C. Cells were pelleted by centrifugation at 6700 g for 25 min, and pellets were resuspended in 50 mL lysis buffer (25 mM HEPES pH 8.0, 200 mM NaCl, 10 mM imidazole, 1 mM TCEP, 0.5 mg/mL lysozyme, 0.05 µL/mL universal nuclease). Cells were sonicated for 7 minutes (10 s on, 10 s off, 60% amplitude), and lysate was cleared by centrifugation at 104,000 g for 45 minutes at 4 °C in a Type 45 Ti rotor (Beckmann Coulter). Supernatant was applied to a 5 mL HisTrap FF column (Cytiva 17-5255-01) equilibrated with Ni-A buffer (20 mM HEPES pH 8.0, 100 mM NaCl, 5% glycerol, 10 mM imidazole, 2 mM beta-mercaptoethanol (βME)). Protein was washed with 150 mL Ni-A buffer and eluted with a gradient of Ni-B buffer (20 mM HEPES pH 8.0, 100 mM Nacl, 5% glycerol, 200 mM imidazole, 2 mM BME). Eluted sample was dialysed in presence of ULP1 SUMO protease in dialysis buffer (20 mM HEPES pH 7.0, 150 mM NaCl, 5% glycerol, 1 mM TCEP) overnight at 4 °C in 10k MWCO dialysis tubing (Thermo Fisher 68100). Dialysed protein was applied again on a 5 mL HisTrap FF column to remove the cleaved His₆-SUMO tag. Flow-through and seven 10 mL elutions with dialysis buffer were collected and concentrated using a 10k MWCO concentrator. RAB5A was loaded with GTP or GDP, purified, and concentrated in the same way as RAB1A.

To label RAB5A for affinity binding assays, 1 mg of AlexaFluor 488 C5 maleimide (Invitrogen A10254) was dissolved in 138.8 µL 100% DMSO (Fisher Bioreagents BP231) for a final concentration of 10 mM. Protein and dye were mixed in labelling buffer (20 mM HEPES pH 7.0, 200 mM KCl, 1 mM TCEP), with final concentration of RAB1A at 112 µM and AF488 at 224 µM and incubated for 2 hours at room temperature with slow rotation. Reaction was quenched with 1 mM DTT and applied to an S75 16/60 gel filtration column equilibrated with 20 mM HEPES pH 7.0, 300 mM NaCl. Peak fractions were pooled and concentrated to 88 µM with 14% labelling efficiency. Final RAB5A-AF488 product was concentrated to 146 µM with 43% labelling efficiency.

Purification of mCherry-tagged yeast VPS34-CII

Plasmids carrying genes encoding the subunits of the WT ScVPS34-CII (pYO359 + pYO1922) or ScVPS34-CII carrying a RAB5 binding mutant (pYO1920 + pYO1922) were transformed into the YOY193 yeast strain [ADE2::TRX1 Bcy123 (pep4::HIS3 prb1::LEU2 bar1::HISG lys2::Gal1/10-Gal4 can1 ade2 trp1 his3 ura3 leu2-3,112 lys2); PMID: 26450213]. Cells were grown in YM4 medium (0.67% yeast nitrogen base without amino acids; 0.5% casamino acids and 2% glucose, supplemented with 0.002% adenine sulfate; and 0.002% tyrosine). Gene expression was induced by 2% galactose in the presence of 2% glycerol, 3% lactic acid, and 10 mM ZnCl₂ without glucose at 30 °C for 26 hours. Cells were lysed in lysis buffer [50 mM Tris at pH 8.8, 150 mM NaCl, 0.5 mM TCEP, 1% Triton-X, 0.5 mM phenylmethylsulfonyl fluoride (PMSF), and a protease inhibitor tablet (Roche 05056489001)] with zirconium beads using a cell disruptor (FastPrep-24, MP Biomedicals) at the intensity of 6.5 for 40 s. Lysates were spun at 3000 g for 1 min. Supernatants were transferred to 2 ml tubes, and spun at 21,000 g for 5 min at 4 °C on a bench top centrifuge. Supernatants were combined and passed through a 5 mm syringe filter. Immunoglobulin G (IgG) beads (GE Healthcare 17096902) were added to the lysate, incubated for 3 hours at 4°C at 8 rpm, then washed once with wash buffer (50 mM Tris at pH 8.0, 150 mM NaCl, 0.5 mM TCEP, 0.1% Triton-X, 50 mM MgCl₂, 5 mM ATP and 5 µg/mL RNaseA [Sigma, 83834]) and TEV buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 0.5 mM TCEP). The protein A (ZZ) tags were cleaved by TEV protease in TEV buffer overnight. The elution fractions were concentrated and subjected to gel filtration on Superdex 200 10/30 equilibrated in 20 mM HEPES at pH 8.0, 300 mM NaCl, and 0.5 mM TCEP. The peak fractions were combined and concentrated using a 100 MWCO concentrator (Millipore UFC810024).

Purification of Vps21 Q66L-EGFP

The pYO1919 plasmid was transformed into OverExpress C41(DE3) chemically competent cells (Lucigen 60442). Transformed bacteria were cultured in 4 L 2xTY medium supplemented with 0.1 mg/mL ampicillin (Formedium AMP50) at 37 °C until OD600=0.8, then expression was induced with 0.5 mM IPTG, and bacteria were incubated for further 16 hours at 15 °C. Cells were pelleted by centrifugation at 6700 g for 15 minutes, frozen in liquid nitrogen, and stored at -80 °C until use. Pellets were resuspended in 50 mL lysis buffer (50 mM HEPES pH 8.0, 150 mM NaCl, 12% glycerol, 5 mM beta-mercaptoethanol (BME), 1 mM PMSF, 1x EDTA-free protease inhibitor cocktail tablet). Cells were sonicated for 5 min on ice (2 s on, 3 s off, 60% amplitude), and lysate was cleared by centrifugation at 142,000 g for 15 min at 4 °C in a Type 45 Ti rotor. Supernatant was filtered through a 5 µm syringe filter then passed through a 5 ml Ni-NTA FF column (GE Healthcare). The column was washed with 50 mL wash buffer (50 mM Tris pH 8.0, 300 mM NaCl, 5 mM BME, and 12% glycerol), then the protein was eluted with 20 mL elution buffer (50 mM HEPES pH 8.5, 20 mM NaCl, 300 mM imidazole, and 5 mM BME). The eluted protein was immediately loaded on a 5 mL Q column (GE Healthcare). The bound protein was eluted with a NaCl gradient between Q-A buffer (50 mM Tris pH 8.5 and 5 mM BME) and Q-B buffer (50 mM Tris pH 8.5, 1 M NaCl, and 5 mM BME). Elution fractions were combined and concentrated down to 219 mg/ml (4.3 mM). The protein was diluted to 1 mM, and incubated with 11 mM GTP and 18 mM EDTA at rt for 90 min, then 36 mM MgCl2 was added and further incubated for 30 min at rt. The GTP-loaded protein was subjected to gel filtration on a S75 16/60 column in GF buffer (20 mM HEPES 8.0, 150 mM NaCl, and 0.5 mM TCEP). After combining the peak fractions, 10 mM TCEP at a final concentration was added and the protein was concentrated down to 130 mg/ml (2.57 mM).

Multiple sequence alignment

Sequences for human RAB GTPases were retrieved from UniProtKB (https://www.uniprot.org/(UniProt, 2025)) and multiple sequence alignments generated using Clustal Omega (https://www.ebi.ac.uk/jdispatcher/msa/clustalo (Sievers et al., 2011)). Alignments were inspected and corrected using JalView(Waterhouse et al., 2009), and results were analysed and visualised with ESPript 3.0 (https://espript.ibcp.fr/ESPript/ESPript/ (Gouet et al., 1999)).

GUV assay

GUVs were prepared using a hydrogel-assisted swelling protocol. 88 µL of 5% polyvinyl alcohol (PVA, Sigma 8.14894, MW approx. 145 kDa) solution was deposited onto a 13 mm cleaned coverslip (VWR 631-0150) and a thin layer of PVA was created by centrifuging for 30 s in a fixed-speed microfuge (IKA mini G, 6000 rpm) using a custom-made adaptor for cover slips. PVA-coated coverslips were dried at 60 °C for 30 minutes, and 15 µL of GUV lipid mixture at 1 mg/mL lipid concentration was deposited on the PVA layer. Coverslips were desiccated for 60 minutes, and GUVs swollen for 1 hour at room temperature in 220 µL 500 mM sterilised sucrose solution. Freshly generated GUVs were transferred to a 1.5 mL tube passivated by incubating with 1 mL of 5 mg/mL BSA (Sigma A7030) for 1 hour, followed by two 500 mM sucrose rinses.

To immobilise the GUVs, chambers of an 8-well glass bottom plate (Ibidi 80807) were incubated with 150 µL of avidin solution (0.1 mg/mL egg-white avidin, Invitrogen A2667, dissolved in PBS, and 1 mg/mL BSA) for 15 minutes at room temperature, and washed three times with 150 µL observation buffer (50 mM HEPES pH 7.0, 241.4 mM NaCl). 64 µL observation buffer was mixed with 48 µL GUVs in each chamber and incubated overnight at 4 °C in presence of 10 µM (4-fold molar excess to the PE-MCC lipid) RAB protein. GUVs were carefully washed four times with 360 µL wash buffer (31.8 mM HEPES pH 8.0, 172.7 mM NaCl, 181.8 mM sucrose, 5 mM BME).

GUVs were observed with a 63x oil-immersion objective (Plan-Apochromat 63x/1.4 Oil DIC M27, Zeiss) on an inverted confocal microscope (LSM 880 Indimo, AxioObserver, Zeiss), using ZEN software (Zeiss). The observation chamber was immobilised on a universal mounting frame (PeCon 010-800 486) using a chamber system insert plate (PeCon 000 470). After confirming that GUVs were immobilised on the wells, 20 µL of 10x buffer was added (250 mM HEPES pH 8.0, 20 mM MnCl₂, 10 mM TCEP, 1 mM ATP pH 8.0, 10mM EGTA in assays with EGTA). In ZEN software, Time Series and Positions were selected, and Definite Focus.2 (Zeiss) was used to maintain focus for every time point and position. The Lissamine-Rhodamine channel for GUV visualisation was excited using a diode-pumped solid-state (DPSS) 561 nm laser and collected with a 570-623 nm band. The AlexaFluor 647 channel for PX domain detection was excited using a HeNe 633 nm laser and collected with a 638-755 nm band. Four areas were selected per well to analyse at least 10 GUVs. To start the reaction, 48 µL of a solution containing the VPS34 kinase complex, labelled PX domain, in protein dilution buffer (25 mM HEPES pH 8.0, 150 mM NaCl, 1 mM TCEP, 0.67 mg/mL BSA) was added to the chamber. No BSA was added in the protein dilution buffer for assays without BSA. The final concentration of PX domain was 3.7 µM. The final concentrations of VPS34 complexes are described in figures and their legends. Images of each position were taken every 120 s for 2-3 h. Images were analysed using the strategy and macro described in Ohashi et. al. 2020 (Ohashi et al., 2020).

Confocal microscopy bead-based assay for RAB protein binding to VPS34 complexes

10-30 µL RFP-trap agarose beads (Chromotek RTA-10) were washed with 100 µL binding buffer (20 mM HEPES pH 8.0, 100 mM NaCl, 0.5 mM TCEP) and incubated with 2 µM mCherry-tagged VPS34 complex for 30 min on ice. Excess buffer was removed after centrifugation at 500g for 30 s and beads were washed with 100 µL binding buffer to remove unbound VPS34 complex. 30 µL reactions were prepared in PCR tubes with final bead concentration of 2% and AF488-labelled RAB proteins were added at varying final concentrations. Reactions were incubated on ice for 30 min and transferred to a 384-well plate (Greiner 781856) for imaging.

Beads were imaged with a 10x objective (EC Plan-Neofluar 10x/0.3 M27, Zeiss) on an inverted confocal microscope (LSM 880 Indimo, AxioObserver, Zeiss), using ZEN software (Zeiss). The 384-well plate was immobilised on a universal mounting frame (PeCon 010-800 486). DIC channel was used to visualise beads. The mCherry channel was excited using a diode-pumped solid-state (DPSS) 561 nm laser and collected with a 578-696 nm band. The AlexaFluor 488 channel was excited using an Argon 488 nm laser and collected with a 493-543 nm band. Images were analysed using Fiji software. Beads were selected using the ellipse ROI (region of interest) selection tool and added to the ROI manager, and fluorescence intensity of mCherry and AF488 channels was measured using a modified version of macro described in Ohashi et. al. 2020 (Ohashi et al., 2020). Background intensity was measured from 10 ROIs without beads and averaged to subtract from measured bead fluorescence intensities. For each bead, intensity of the AF488 channel was normalised by mCherry channel intensity to account for unequal mCherry-tagged VPS34 complex loading on the RFP-trap beads. Data were further analysed with GraphPad Prism (GraphPad) and binding affinities estimated using the non-linear fitting tool with ‘One site – specific binding’ option.

Nano differential scanning fluorimetry

To determine the melting temperatures (T_m) of VPS34 complexes, nanoDSF experiments were carried out on the Prometheus NT.48 instrument (NanoTemper). 11 µL samples were prepared with final VPS34 complex concentration of 0.5 mg/mL in SEC buffer (20 mM HEPES pH 8.0, 0.5 mM TCEP, 150 mM NaCl for complex I or 300 mM NaCl for complex II) and incubated on ice for 30 min. Prometheus standard capillaries (NanoTemper PR-C002) were filled with 10 µL sample and placed on the sample holder. Prior to the experiment, fluorescence signal was optimised by adjusting the excitation power. A temperature gradient was then applied at 1°C/min rate, from 15°C to 90°C, and intrinsic protein fluorescence emissions recorded at 330 nm and 350 nm. Data were visualised and analysed in GraphPad Prism (GraphPad).

Yeast strains

The S. cerevisiae BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) strain was from Open Biosystems. The vps15Δ null yeast was created by a PCR-based gene targeting with the primers yoo1652 (5’-GTTGAATTGGAGAAATACAAAAGCTGTAAGGTTATCAAAAAGGAAGGCATACAGT ATAgtttagcttgctcgtccccgcc-3’) and yoo1653 (5’-GAATACAAATTTATACGCATTTAGAATAAAGAAGCTGaaactggatggcggcgttagtat-3’), and the Schizosaccharomyces pombe HIS5 gene as a template (Longtine et al., 1998). For fluorescence microscopy, a vps15Δ strain integrated with a plasmid carrying N-terminally tandem-tomato-tagged VPS21 Q66L (pYO1879) was created by linearising the plasmid with NheI enzyme (YOY415).

Yeast plasmid transformation

For each transformation, 5 - 10 absorbance units at 600 nm (A₆₀₀) of yeast cells either from plates or liquid culture were suspended in 100 μL One-step buffer (40% PEG3350, 0.2 M lithium acetate, 10 mM Tris pH 7.4, 100 mM DTT), and 3 μL of a plasmid was added to the mixture and incubated at room temperature for 10 min. The mixture was then incubated at 42 °C for 30 min, spun at 3000 g for 30 s to remove the buffer. The yeast pellet was resuspended in 50 mL H₂O and spread on a -URA plate.

CPY assay

Yeast cells were grown to log phase (A₆₀₀ = 0.5-0.9) at 30 °C in -URA medium. Five A₆₀₀ units of cells were collected and resuspended in 500 μL of -URA medium containing 0.5 mg/mL bovine serum albumin (BSA) and 50 mM KPO4 at pH 5.7 and incubated at 30 °C for 1 hour. Cell and medium fractions were separated by centrifugation at 3000g for 1 min. Medium fractions were precipitated with final 10% trichloroacetic acid (TCA, from 100% stock) on ice for 15 min and centrifuged at 4 °C at 20,000g for 10 min. Pellets were washed once with 90% cold acetone. Both fractions were resuspended in 100 μL of lysis buffer (50 mM Tris at pH 8.0, 1% SDS, 8 M urea, 10 mM DTT, and 5 mM EDTA), 30 μL of 4x sample buffer, and 0.5 mm zirconia beads (YTZ® GRINDING MEDIA, Tosh). Cells and TCA pellets were disrupted at 5000 revolutions per minute (rpm) for 30 s with PowerLyzer (Mo Bio) at 4°C. Samples were boiled at 98 °C for 2 min and loaded on SDS-PAGE gel.

CPY is synthesized as a premature form (p1), glycosylated in the Golgi (p2), and then delivered to the vacuole, where it is cleaved to become a mature form (m). The m and p2 forms of CPY were detected both in cellular and medium fractions using anti-CPY (1/1000; Invitrogen A6428). The C-terminally FLAG-tagged Vps15 proteins were detected by anti-Flag-HRP (HRP, horseradish peroxidase) (1/1000; Sigma A8592). Pgk1 was detected by anti-Pgk1 antibody (1/1000; Abcam ab113687). The bands were detected using ChemiDoc™ Touch Imaging System (Bio-Rad) with Chemiluminescence mode.

For each condition, intensities of p2 and m bands in cell and medium fractions were summed as a total intensity, then the percentages of m and p2 were calculated. In the case p2 was detected both in cell and medium fractions, the percentage of the summed intensity from these fractions was shown. The band intensities were measured using Image Lab 6.0 (Bio-Rad). The quantification was from 3 independent experiments.

Yeast fluorescence microscopy

The YOY415 strain was transformed with the plasmids carrying WT VPS15-EGFP (pYO634) or RAB5 binding mutant VPS15-EGFP (pYO1880). Cells were grown to mid-log phase (A600=0.5-0.9), then examined with an inverted microscope (Nikon Eclipse Ti2) equipped with a CMOS camera (HAMAMATSU ORCA-Flash4.0 C11440) and a niji LED Light Source (bluebox Optics). Images were analysed and levels were adjusted using Fiji.

Hydrogen-deuterium exchange coupled to mass spectrometry

A preincubation was performed by adding either dilution buffer (20 mM HEPES pH 7.0, 150 mM NaCl) or RAB5A stock (611 μM RAB5A-Q79L-C19S-C63S-GTP, 20 mM HEPES pH 7, 150 mM NaCl) to a concentrated Complex II stock (57 μM Complex II, 20 mM HEPES pH 8.0, 300 mM NaCl, 0.5 mM TCEP) to yield solutions that were 48 μM of Complex II with or without 90 μM of RAB5A. Samples were then incubated for 30 minutes on ice. Complex II alone samples were then exposed for different deuteration times (3 seconds on ice, 3, 30, 300 and 3000 seconds at room temperature) to a deuterated buffer: 20 mM HEPES pD8, 300 mM NaCl, with D₂O (Thermo Scientific, 166300100) to yield a final concentration of 83% D₂O. Samples of Complex II with RAB5A were exposed to a deuteration buffer consisting of 20 mM HEPES pD 8.0, 300 mM NaCl, with D₂O and 90 μM RAB5A-Q79L-C19S-C63S-GTP to yield 83% D₂O. The final concentrations were 5 μM Complex II alone or 5 μM of Complex II and 90 μM of RAB5A. Triplicate samples were aliquoted for each deuteration timepoint. Exchange reactions were quenched in 2 M Guanidine-Hydrochloride, 100 mM glycine, formic acid (Fisher Chemical, 10596814), LC-MS grade water (Romil, H949), pH 2.1 (final pH was 2.5). Samples were then flash-frozen in liquid nitrogen immediately and stored at -80 °C until analysis. Prior to LC-MS analysis, samples were quickly thawed and injected on a 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 x 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. × 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, using the data-independent acquisition mode (HDMS^E) over an m/z range of 50-2000 and a 100 fmol/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 VPS34, VPS15, UVRAG, BECLIN1, RAB5A-Q79L-C19S-C63S-GTP, and pepsin sequences, with peptide and fragment tolerances set automatically by PLGS. Deuterium uptakes for all identified peptides were then filtered and validated manually using HDExaminer 3.4.0 (Trajan Scientific and Medical) as follows: peptides were kept with a minimum fragment per amino acid of 0.2, a minimum intensity of 10³, a 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 an in-house program (Archaeopteryx), where the p-value threshold was set to 0.05 for the t-test. Only statistically significant peptides showing a difference greater than 0.25 Da and greater than 5% for at least two timepoints were kept. A table of peptides within the dataset and the quality assessment statistics are available in Supplementary Data 1. HDX-MS results were illustrated on a Complex II model using PyMOL 2.5.4 (www.pymol.org). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (Perez-Riverol et al., 2022) with the dataset identifier PXD061277.

Cryo-ET and subtomogram averaging

We re-processed the cryo-ET dataset of Complex II-BATS on RAB5-coated LUVs described in (Tremel et al., 2021). The tilt series raw images were imported in Relion 5.0 (beta4-commit-dfea29,(Burt et al., 2024)) --tomo mode, and motion corrected using Relion 5.0 implementation of MOTIONCOR2 (Zheng et al., 2017). CTF was estimated using the build-in CTFFIND-4.1 (Rohou & Grigorieff, 2015), and tilt series were aligned using the IMOD (Kremer et al., 1996) wrapper with fiducial based alignment. 10 Å/pixel tomograms were reconstructed for picking, and particles on membranes were manually picked on 3 representative tomograms. CrYOLO 1.9.9 (Wagner et al., 2019) was used for picking in tomograms after training on the manually picked particles using the trained custom model described above as pretrained weights, with a threshold of 0.05 and a box size of 220 Å. 146,464 particles were re-imported in Relion 5.0, scaled to the raw movies and extracted at bin 8 with a box size of 375 Å. The EMD-12237 density was used as a reference for 3D classification using a low-pass filter of 60 Å, the best 3D class was selected with 23,237 particles and subjected to another round of 3D classification using bin 4 particles. Two classes with 8,441 particles were selected and refined at bin 1, masking out the membrane and using Blush regularisation. One refinement round of CTF refinement followed by Bayesian polishing was done and reached a resolution of 9.88 Å. To look for the Rab5 binding site in VPS15, a 3D classification was done on the 23,237 bin4 particles using the SPA structure of Complex II with two RAB5 molecules as a 40 Å low passed template. To search for alternative conformations of Complex II on the membrane, a 3D classification was done on the same particles using the membrane bilayer as a reference.

Cryo-EM sample preparation and screening

Complex II VPS34 199-REIE-202 > 199-AAAA-202 + RAB5A and Complex II VPS34 199-REIE-202 > 199-ERIR-202 + RAB5A

VPS34-CII at the final concentration of 6 µM was incubated with 90 µM RAB5A for 20 min on ice in buffer containing 50 mM HEPES pH 8.0, 200 mM NaCl, 1 mM TCEP, 5 mM MgCl₂, and 1 mM AMP-PNP. The sample was crosslinked with BS3 at a final concentration of 30 µM for further 20 min on ice, then quenched with 100 mM TRIS pH 8.0. Prior to blotting, the sample was supplemented with 0.005% (w/w) Nonidet P-40 substitute and 4 mM CHAPSO. 3.5 µL sample was applied to a glow-discharged (40 mA, 90 s, Edwards Sputter Coater S150B) UltrAuFoil R 1.2/1.3 Au300 grid and plunge-frozen in liquid ethane using a Vitrobot Mk 3 (Thermo Fisher) plunger at 14 °C, 100% humidity, with 0 s waiting time, 4 s blotting time, and blot force set to +8. Grids were screened on a Glacios (Thermo Fisher) microscope with a Falcon 3 detector in linear mode.

WT Complex II + RAB5A

VPS34-CII at the final concentration of 4 µM was incubated with 90 µM RAB5A for 20 min on ice in buffer containing 50 mM HEPES pH 8.0, 200 mM NaCl, 1 mM TCEP, 5 mM MgCl₂, and 1 mM AMP-PNP. The sample was crosslinked with BS3 at a final concentration of 20 µM for further 20 min on ice, then quenched with 100 mM TRIS pH 8.0. Prior to blotting, the sample was supplemented with 0.005% (w/w) Nonidet P-40 substitute and 4 mM CHAPSO. 3.5 µL sample was applied to a glow-discharged (40 mA, 90 s, Edwards Sputter Coater S150B) UltrAuFoil R 1.2/1.3 Au300 grid and plunge-frozen in liquid ethane using a Vitrobot Mk 3 (Thermo Fisher) plunger at 14 °C, 100% humidity, with 0 s waiting time, 4 s blotting time, and blot force set to +8. Grids were screened on a Glacios (Thermo Fisher) microscope with a Falcon 3 detector in linear mode.

Cryo-EM data collection

Complex II VPS34 199-REIE-202 > 199-AAAA-202 + RAB5A

Images were collected on a Titan Krios G3 TEM (Thermo Fisher) with a Gatan K3 detector (Gatan) in counting mode and energy filter set to 20 eV. 10,739 movies were collected at 105,000x magnification and pixel size of 0.86 Å / px using EPU software (ThermoFisher) in AFIS mode, with 2 movies collected per foil hole. The defocus range used was -0.8 nm to -2.2 nm, and the dose rate was 15.1 e^- / px / s over an exposure length of 1.96 s, making the total dose 40 e^- / Å. Movies were fractionated into 40 frames.

Complex II VPS34 199-REIE-202 > 199-ERIR-202 + RAB5A

Images were collected on a Titan Krios G3 TEM (Thermo Fisher) with a Gatan K3 detector (Gatan) in counting mode and energy filter set to 20 eV. 13,301 movies were collected at 105,000x magnification and pixel size of 0.86 Å / px using EPU software (ThermoFisher) in AFIS mode, with 3 movies collected per foil hole. The defocus range used was -0.8 nm to -2.2 nm, and the dose rate was 26.4 e^- / px / s over an exposure length of 1.12 s, making the total dose 40 e^- / Å. Movies were fractionated into 40 frames.

WT Complex II + RAB5A

Images were collected on a Titan Krios G1 TEM (Thermo Fisher) with a Gatan K3 detector (Gatan) in counting mode and energy filter set to 20 eV. In total, 19,844 movies were collected on two grids, at 105,000x magnification and pixel size of 0.73 Å / px using EPU software (ThermoFisher) in AFIS mode, with 3 movies collected per foil hole. The defocus range used was -0.8 nm to -2.2 nm. For the first grid, the dose rate was 15 e^- / px / s over an exposure length of 1.42 s, making the total dose 40 e^- / Å. For the second grid, the dose rate was 21.5 e^- / px / s over an exposure length of 1 s, making the total dose 40 e^- / Å. Movies were fractionated into 40 frames.

Cryo-EM data processing

VPS34-CII VPS34 199-REIE-202 > 199-AAAA-202 + RAB5A (Fig. S10A)

Data was pre-processed using RELION 4.0. Movie frames were motion-corrected using MotionCorr2, and CTF of motion-corrected micrographs was estimated using CTFFIND-4.1. Micrographs with CTF-estimated maximum resolution of ≤5 Å were selected for further processing. Particle picking was done with CrYOLO-1.7.5 using a custom-trained model. Particles were extracted with two-fold downsampling and at a box size of 204 x 204 px and imported to CryoSPARC for 2D classification. Selected particles were converted from .cs to star and re-imported to RELION for re-extraction at full box size of 408 x 408 px. Full size particles were again imported into CryoSPARC and used for homogeneous refinement with the density of VPS34-CI low-passed to 40 Å as an initial reference, yielding a 3.43 Å map after FSC-mask auto-tightening. An initial model for complex II REIE > AAAA + RAB5A was built into this density by rigid-body fitting PDB depositions of VPS34 kinase domain (PDB: 3IHY) and RAB5A (PDB: 1R2Q), AlphaFold2 predictions of VPS15, VPS34 C2 domain, BECLIN1 CC2-BARA domains, UVRAG CC2-BARA2-CTD domains, and AlphaFold3 prediction of BECLIN1 CC1 domain and UVRAG-CC1 domain. Models were consolidated into a single entity for complex II REIE > AAAA + RAB5A using Coot 0.9.8.7. From this initial model, UCSF ChimeraX command ‘molmap’ was used to generate masks on the adaptor arm, kinase arm, base of the complex, and the RAB5A binding site on VPS34. The particles from the homogeneous refinement job in CryoSPARC were classified into eight 3D classes, and selected classes (see Fig. S10A for details) were used to refine regions of the complex mentioned above. The adaptor arm was refined to 3.2 Å resolution; the kinase arm yielded a map with resolution of 3.15 Å; the map of the base of the complex had a resolution of 3.2 Å; the RAB5A binding site on VPS34 was refined to 3.25 Å. Maps from the consensus and local refinements were processed using CryoTEN. Local refinement maps were merged into a composite map in UCSF ChimeraX. The initial model for complex II REIE > AAAA + RAB5A was refined using the composite map with ISOLDE, StarMap, TEMPy-ReFF, PHENIX, and Servalcat. Model was validated through MolProbity.

Two of the eight classes (classes 1 and 6) from the final 3D classification job were later found to contain RAB5A bound to VPS15 subunit. These densities were used as initial references for heterogeneous refinement in the processing of complex II REIE>ERIR + RAB5A and wild type complex II + RAB5A datasets. One class did not contain RAB5A bound to the complex, and this density was used as an ‘apo complex II’ reference.

Complex II VPS34 199-REIE-202 > 199-ERIR-202 + RAB5A (Fig. S10B)

Movie frames were motion-corrected in RELION 4.0 with MotionCorr2, and CTF of motion-corrected micrographs was estimated using CTFFIND-4.1. Micrographs with CTF-estimated maximum resolution of ≤5 Å were selected for further processing. Particle picking was done with CrYOLO-1.7.5 using a custom-trained model. Picked particles were extracted with two-fold downsampling and a box size of 220 x 220 px, then imported to CryoSPARC for heterogeneous refinement with five initial references: a noise decoy, apo complex II, complex II + RAB5A on VPS34, complex II + RAB5A on VPS15, and complex II with two RAB5A bound simultaneously. The particles sorted into the classes with noise decoy and complex II + (RAB5A)₂ volumes did not produce meaningful densities and were removed. To find particles that had RAB5A bound to VPS34, the other three classes were subject to 3D classification into two classes with a focus mask on the RAB5A binding site on VPS34, and ‘force hard classification’ parameter set to ‘true’. The class with a density resembling a RAB5A bound to VPS34 was poorly resolved and further processing of the particles associated with this density did not yield a meaningful map. Therefore, the class without a density for RAB5A on VPS34 was selected for further processing. These particles were again force classified into two 3D classes, with a focus mask on RAB5A bound to VPS15. Particles from the class with an extra density on the VPS15 subunit were re-imported to RELION and re-extracted at full size with box size of 440 x 440 px, then imported back to CryoSPARC for non-uniform refinement that yielded a map at 3.68 Å after the FSC-mask auto-tightening. An initial model was built into this density by fitting the model of complex II REIE>AAAA + RAB5A. The RAB5A chain was moved from the binding site on VPS34 subunit to the VPS15 subunit and fitted into the density. This model was used in UCSF ChimeraX to generate masks for local refinements via the ‘molmap’ command. The base of the complex was refined to 3.52 Å; the kinase arm to 3.55 Å, the adaptor arm to 4.09 Å, and the local refinement of the RAB5A binding site on the VPS15 subunit yielded a map at 3.52 Å. Maps from the consensus and local refinements were processed using CryoTEN. Densities from local refinements were combined into a composite map on UCSF ChimeraX, which was used to refine the initial model for complex II REIE > ERIR + RAB5A with ISOLDE, StarMap, TEMPy-ReFF, PHENIX, and Servalcat. The model was validated through MolProbity.

WTVPS34-CII + RAB5A (Fig. S10C)

Two datasets were collected for this sample. For both datasets, the motion-correction was done in RELION 4.0 with MotionCorr2, and the CTF of motion-corrected micrographs was estimated using CTFFIND-4.1. Micrographs with CTF-estimated maximum resolution of ≤5 Å were selected for further processing. Particle picking was done with CrYOLO-1.7.5 using a custom-trained model. Picked particles were extracted with two-fold downsampling and a box size of 220 x 220 px, then imported to CryoSPARC.

For dataset 1, two rounds of heterogeneous refinement were performed with five initial references: a noise decoy, apo complex II, complex II + RAB5A on VPS34, complex II + RAB5A on VPS15, and complex II with two RAB5A bound simultaneously. Least noisy classes were further classified into ten 3D classes. Particles from seven classes were re-imported into RELION and re-extracted at full size with box size of 440 x 440 px, then imported back to CryoSPARC for merging with dataset 2.

For dataset 2, picked particles were subject to 2D classification, and then used in homogeneous refinement together with complex I + NRBF2 MIT domain as the initial reference. Particles were then classified into two 3D classes with a focus mask on the RAB5A binding site on VPS34. The class with a more prominent density on the VPS34 RAB5 binding site was used in two rounds of heterogeneous refinement, with four initial references: complex II + RAB5A on VPS15, complex II + RAB5A on VPS34, complex II with two RAB5A bound simultaneously, and apo complex II. Particles from the best densities were re-imported to RELION, re-extracted at full size with the box size of 440 x 440 px, imported back to CryoSPARC and merged with particles from dataset 1.

Combined 233,866 particles were subject to two rounds of heterogeneous refinement using five initial references: a noise decoy, apo complex II, complex II + RAB5A on VPS34, complex II + RAB5A on VPS15, and complex II with two RAB5A bound simultaneously. 66,323 particles were classed as apo complex II, with no density for RAB5A visible on VPS34 or VPS15 at any viewing threshold level on UCSF ChimeraX. These particles were used in non-uniform refinement and yielded a 3.87 Å reconstruction of apo complex II. 49,688 particles were assigned to the class with RAB5A bound to the VPS34 subunit. To confirm that in this particle set A) there were no particles with RAB5A bound to VPS15 and B) all particles had RAB5A bound to the VPS34 site, several rounds of ‘hard’ 3D classifications were performed with two classes and focus masks on the RAB5A binding site either on VPS34 or VPS15 subunits. No extra density on VPS15 was present, and none of the particles had an empty VPS34 RAB5A binding site; therefore, all particles were used in a non-uniform refinement job that yielded a 3.99 Å map.

44,127 particles contained complex II with the extra RAB5A density on the VPS15 subunit. These particles were further inspected using the procedure described above to ensure that A) no particles had RAB5A bound to the VPS34 subunit, and B) all particles had RAB5A bound to VPS15 subunit. This particle set was merged with the 26,318 particles found during the processing of the class with complex II + (RAB5A)₂ described below. The total of 70,445 particles for the density of complex II + RAB5A on the VPS15 subunit were used in non-uniform refinement that yielded a reconstruction at 4.00 Å, 58,253 particles that were classed as complex II with two RAB5A GTPases were used in a ‘hard’ 3D classification into two classes with a focus mask on the VPS34 RAB5A binding site. The class that did not have a RAB5A density on the VPS34 subunit contained 26,318 particles that were merged with the other 44,127 particles for complex II + RAB5A on VPS15 density. The other class with 31,935 particles was subject to further inspection as described above to confirm that all particles in this set A) had RAB5A bound to VPS15 and B) had RAB5A bound to the VPS34 site. All particles were confirmed to have two RAB5A proteins bound to complex II and were used for a non-uniform refinement job that produced a map at 4.03 Å resolution.

Maps from the consensus refinements were processed using CryoTEN. Models for complex II apo, complex II + RAB5A on VPS34, complex II + RAB5A on VPS15, and complex II + RAB5A on VPS34 and VPS15 were built by fitting a model of complex II REIE>AAAA + RAB5A into the density and adjusting the RAB5A chain appropriately. Models were refined using ISOLDE, StarMap, TEMPy-ReFF, PHENIX, and Servalcat. Models were validated through MolProbity. Consistent with what was reported previously for VPS34-CI (Cook et al., 2025), our highest resolution cryo-EM structures contain a density continuous with the VPS15 N-terminus that is consistent with a myristate moiety. When this density was was not apparent, the reconstructions had lower resolution. We also observed density for a GDP nucleotide in the VPS15 kinase domain. As it has been reported that VPS15 binds a GTP nucleotide in its active site (Cook et al., 2025), it may be that the GTP had been hydrolysed, and we were seeing the bound product of the reaction (Fig. S10B).

PDB models

Cryo-EM maps and refined VPS34-CII models were deposited in EMDB and PDB under accession codes:

WT VPS34-CII: PDB 9RX8, EMD-54361 (consensus map); WT VPS34-CII-RAB5A_VPS34: PDB 9RX9, EMD-54362 (consensus map); WT VPS34-CII-RAB5A_VPS15: PDB 9RXA, EMD-54363 (consensus map); WT VPS34-CII-RAB5A_VPS34,VPS15: PDB 9RXB, EMD-54364 (consensus map); VPS34-CII REIE>AAAA-RAB5A_VPS34,VPS15: PDB 9RX5, EMD-54358 (composite, CRYOTEN), EMD-54317 (composite), EMD-54316 (consensus), EMD-54313 (focused on kinase arm), EMD-54312 (focused on adaptor arm), EMD-54314 (focused on the base of the complex), EMD-54315 (focused on the RAB4A/VPS34 interface); VPS34-CII REIE>ERIR-RAB5A_VPS15: PDB 9RX6, EMD-54359 (composite, CRYOTEN), EMD-54322 (Composite), EMD-54321 (Consensus), EMD-54318 (Focused on kinase arm), EMD-54319 (Focused on adaptor arm), EMD-54320 (Focused on VPS15/RAB5A interface); Reprocessed tomography map for VPS34-CII(ΔBATS)-RAB5A_VPS34 bound to membrane (Tremel et al., 2021): PDB 9S47, EMD-54560.

Data availability

Cryo-EM maps and refined VPS34-CII models were deposited in EMDB and PDB under accession codes: WT VPS34-CII: PDB 9RX8, EMD-54361; WT VPS34-CII-RAB5AVPS34: PDB 9RX9, EMD-54362; WT VPS34-CII-RAB5AVPS15: PDB 9RXA, EMD-54363; WT VPS34-CII-RAB5AVPS34,VPS15: PDB 9RXB, EMD-54364; VPS34-CII REIE>AAAA-RAB5AVPS34,VPS15: PDB 9RX5, EMD-54358, EMD-54317, EMD-54316, EMD-54313, EMD-54312, EMD-54314, EMD-54315; VPS34-CII REIE>ERIR-RAB5AVPS15: PDB 9RX6, EMD-54359, EMD-54322, EMD-54321, EMD-54318, EMD-54319, EMD-54320; Reprocessed tomography map for VPS34-CII(∆BATS)-RAB5AVPS34 bound to membrane (Tremel et al., 2021): EMD-54365

Acknowledgements

We thank Anna Yeates, Grigory Sharov, Bilal Ahsan, Giuseppe Cannone of the MRC-LMB EM facility, and Madhanagopal Anandapadamanaban for help with cryo-EM data collection, Jake Grimmett, Toby Darling and Ivan Clayson for assistance and advice with scientific computing, Stephen McLaughlin and Chris Batters for help with biophysical instrumentation and experiment design, Ksenia Rostislavleva for purifying NRBF2 and NRBF2 MIT proteins, Olga Perisic for helpful discussions, Jerome Boulanger for the modified microscope macros, the Light Microscopy Facility for help with confocal microscopy,the mechanical workshop for custom instruments, the BioMass Facility for help with MS instruments, and Garib Murshudov for assistance with model refinement. The work was supported by the Medical Research Council (MC_U105184308 to R.L.W) and Cancer Research UK (grant DRCPGM\100014 to R.L.W.).

Additional information

Funding

UKRI | MRC | Medical Research Council Laboratory of Molecular Biology (LMB) (MC_U105184308)

• Roger L Williams

Cancer Research UK (CRUK) (DRCPGM\100014)

• Roger L Williams