Structural basis of phosphatidylinositol 3-kinase C2α function

Phosphatidylinositol 3-kinase type 2α (PI3KC2α) is an essential member of the structurally unresolved class II PI3K family with crucial functions in lipid signaling, endocytosis, angiogenesis, viral replication, platelet formation and a role in mitosis. The molecular basis of these activities of PI3KC2α is poorly understood. Here, we report high-resolution crystal structures as well as a 4.4-Å cryogenic-electron microscopic (cryo-EM) structure of PI3KC2α in active and inactive conformations. We unravel a coincident mechanism of lipid-induced activation of PI3KC2α at membranes that involves large-scale repositioning of its Ras-binding and lipid-binding distal Phox-homology and C-C2 domains, and can serve as a model for the entire class II PI3K family. Moreover, we describe a PI3KC2α-specific helical bundle domain that underlies its scaffolding function at the mitotic spindle. Our results advance our understanding of PI3K biology and pave the way for the development of specific inhibitors of class II PI3K function with wide applications in biomedicine.

P hosphoinositide 3-kinases (PI3Ks) are a family of lipidmodifying enzymes that phosphorylate the 3′-OH group of inositol phospholipids and play key roles in physiology ranging from cell growth and metabolism to organismal development. Dysfunction of PI3K signaling is implicated in human diseases including cancer, immunodeficiency, diabetes and neurological disorders [1][2][3] . Mammalian PI3Ks are grouped into three classes based on their structural organization. Class I PI3Ks are receptor-activated heterodimeric enzymes with pivotal roles in cell signaling (for example, cell growth) via synthesis of phosphatidylinositol (PI) 3,4,5-trisphosphate (PI (3,4,5)P 3 ) at the plasma membrane 2,4-6 . Isoform-specific pharmacological inhibitors of class I PI3K activity have undergone clinical development as anticancer therapeutics and for the treatment of human disorders caused by PI3K pathway hyperactivation.
Complexes of Vps34, the sole class III PI3K member, produce PI 3-phosphate (PI(3)P) in the endolysosomal system and during autophagy to regulate vesicle-mediated sorting en route to lysosomes 1 . Recent structural studies 7,8 have enabled the development of selective Vps34 inhibitors that have been instrumental for the analysis and manipulation of class III PI3K function in autophagy and in the regulation of nutrient signaling.
The class II PI3K isoforms PI3KC2α, PI3KC2β and PI3KC2γ are unique in directly synthesizing PI 3,4-bisphosphate (PI(3,4)P 2 ) from PI 4-phosphate (PI(4)P) at the plasma membrane and within the endolysosomal system 9,10 , in addition to synthesis of PI 3-phosphate (PI(3)P) [11][12][13][14] . The mechanistic basis for the ability of class II PI3Ks to recognize PI(4)P as a substrate to directly produce local pools of PI (3,4)P 2 at defined endocytic membrane nanodomains is unknown. PI3KC2α is essential in mice 13 . Loss of its catalytic activity is associated with cellular defects in endocytosis 15,16 , angiogenesis and endothelial cell function [17][18][19] , regulation of blood pressure 20 , viral replication 21,22 , platelet formation 23,24 and primary cilia signaling 13 . Abrogation of PI3KC2α activity in animal models and in humans leads to kidney cyst formation, skeletal abnormalities, neurological symptoms and cataract formation 25 . In addition to these catalytic roles, PI3KC2α is also required for genome stability by acting as a scaffold at the mitotic spindle during cell division 26 .
In contrast to class I 4,6 and class III PI3Ks (refs. 4,7,8,27 ) that are understood in structural detail, little is known about the structural and functional architecture, and mechanism of activation of class II PI3Ks including PI3KC2α. Unlike their class I and class III relatives that are targeted to their site of action via associated subunits, class II enzymes such as PI3KC2α lack stable association with other subunits 4,14 and, thus, must be activated via a distinct regulatory mechanism that so far has remained elusive. The lack of structural information on PI3KC2α and related class II PI3Ks has also greatly hampered the development of isoform-selective pharmacological inhibitors for clinical applications. Moreover, the molecular basis of the scaffolding function of PI3KC2α at the mitotic spindle via its association with the microtubule-binding protein TACC3 at kinetochore fibers to prevent aneuploidy 26 is unknown.
To address these important unresolved issues, we have determined high-resolution crystal structures as well as a 4.4-Å cryogenic-electron microscopic (cryo-EM) structure of PI3KC2α in different conformational states and in the presence of nonselective small molecule inhibitors. Our results identify a hitherto unknown coincident mechanism of lipid-induced activation of PI3KC2α that is distinct from that of all other PI3Ks and can serve as a model for the entire class II PI3K family. Moreover, our structural and Structural basis for PI3KC2α catalytic activity. To gain insights into the mechanism of catalysis and to enable the development of small molecule inhibitors of PI3KC2α, we determined crystal structures of the kinase core domain. Based on the PI3KC2α ΔN+ΔC−C2 construct, we designed a recombinant PI3KC2α core variant that lacks the N-terminal low complexity region and the distal PX and C-C2 domains. The HBD was substituted by a short seven residue-long linker ( Fig. 1a and Extended Data Fig. 2). The 2.5-Å crystal structure of PI3KC2α core determined in the absence or presence of Mg 2+ and ATP confirmed the architecture of the PI3KC2α catalytic core (Fig. 2a, Extended Data Fig. 3a and Table 1) and revealed important insights into the catalytic mechanism (Fig. 2b,c).
The PI3KC2α KD displays a typical PI3K KD fold comprising a smaller N-and a larger C-lobe linked by a hinge (Extended Data Fig. 3b). The ATP-binding P-loop is located between strands kβ1 and kβ2, and contacts the α-phosphate via a conserved serine (S1114). The catalytic loop located between kα6 and kβ7 contains a conserved 1250 DRH 1252 motif, in which R1251 stabilizes the N-terminal part of the activation loop containing the 1268 DFG 1270 motif. H1252 within the DRH motif could act as catalytic base in lipid kinases to deprotonate the 3′-OH of the inositol substrate, a reaction aided by D1250 and D1268 (ref. 31 ). An atypical feature of the crystallized PI3KC2α catalytic core is the presence of only a single Mg 2+ ion that is coordinated by the βand γ-phosphates of ATP and D1146 in kα3 and D1268 in the DFG motif. In this Mg 2+ -bound state, N1255, that is a residue bound to a second Mg 2+ ion in PI3Kγ 5 and other PI3Ks 4 , interacts with D1250 and H1252 in the HRD motif, apparently to inhibit the catalytic function ( Fig. 2c and Extended Data Fig. 3b). Hence, this conformation of PI3KC2α likely reflects an early precatalytic state. Binding of the second Mg 2+ conceivably induces a conformational change of N1255 to release the HRD motif from intramolecular inhibition, thereby enabling catalysis.
Class II PI3Ks including PI3KC2α among other features are distinguished from class I and class III enzymes by their substrate selectivity, most notably their unique ability to use PI(4)P as a substrate to directly produce PI(3,4)P 2 at endocytic membranes 1,14 , for example to facilitate endocytosis 15,33 . We confirmed the preference of purified recombinant PI3KC2α to synthesize PI(3,4)P 2 from PI(4)P over PI(3)P synthesis from PI (Fig. 2e), in agreement with our earlier cell-based data 15,16 . Cellular production of PI(3,4)P 2 versus PI(3)P in addition to substrate availability is likely modulated by the specific membrane environment, explaining the distinct functions of PI3KC2α at the plasma membrane and at endosomes 1,14 . Lipid substrate binding in PI 3-kinases is encoded within the activation loop. The primary sequence of the activation loop is highly conserved also among class II PI3Ks, suggesting that they bind their substrates via similar mechanisms (Extended Data Fig. 2). While the N-terminal part of the activation loop harboring the catalytic DFG motif is clearly resolved, the C-terminal part of the PI3KC2α activation loop containing the phospholipid headgroup-binding basic residues remains disordered. We therefore used the class I PI3Kα complexed to PI(4,5)P 2 (ref. 34 ) as a reference to model PI(4) P headgroup binding to PI3KC2α. In this model, the 3′-OH group of inositol is oriented toward the γ-phosphate of ATP, whereas the 4-phosphate binds to a flexible positively charged surface created by K1283 and R1284 on the C-terminal activation loop (Fig. 2d). We tested this model experimentally by creating charge neutralization mutants of PI3KC2α. Consistently, we found that alanine substitution of K1283 and R1284 abrogated catalytic activity toward PI(4) P (Fig. 2f) and greatly reduced phosphorylation of PI (Extended Data Fig. 3c), whereas alanine substitution of K1283 only had very minor effects.
These data provide a firm structural basis for the unique ability of PI3KC2α and related class II PI3Ks to directly synthesize PI(3,4) P 2 from PI(4)P to control its biological function and activity.
Small molecule inhibition of PI3KC2α catalytic function. Given the multiple important roles of PI3KC2α and related class II PI3Ks in cell physiology and in disease 1,14 , we next sought to probe the structural basis for inhibition of its catalytic activity by small molecules. Although no specific inhibitors of PI3KC2α have been identified so far, the enzyme is known to be targeted by pan-PI 3-kinase inhibitors (that is, PIK-90, Extended Data Fig. 4b) 35 and by Torin-2, an ATP-competitive inhibitor of the PI3K-related kinase superfamily member mTOR 36 (Extended Data Fig. 4a). To explore the determinants of inhibitor potency and specificity, we determined the structures of PI3KC2α core bound to Torin-2 or PIK-90 and compared these to the structure of the apo-form of the enzyme (Fig. 3a).
Torin-2 binds to PI3KC2α in a mode similar to mTOR ( Fig. 3b and Extended Data Fig. 4c). The tricyclic benzonapththyridine ring of Torin-2 occupies the hydrophobic adenine pocket and a sulfur-pi interaction contributed by the conserved M1125 located in hydrophobic pocket II. Binding is further enabled by local small scale conformational changes of K1138 and D1268 to accommodate the amino-pyrimidine group of Torin-2 and by hydrophobic contacts of the benzotrifluoride group with M1136 and F1112 (Fig. 3b,c). The reduced half-maximum inhibitory concentration (IC 50  PI3KC2α ∆N + C-C2 PI3KC2α N-and C-terminal lobes of the kinase core domain. amino acids 1-377 of PI3KC2α are predicted to be disordered. HBD specific to class II PI3Ks identified in this study. recombinant PI3KC2α constructs used for X-ray protein crystallography and single-particle cryo-EM are indicated. b, Overall structure of PI3KC2α ΔN+ΔC−C2 . The compact core comprises an N-terminal helix (black), the rBD (red), N-C2 (yellow), HD (green), N-KD (cyan) and C-KD (blue). The HBD (orange) points away from the compact core region and forms the stalk of the inverted lollipop. In this open conformation, a short helix is observed in the activation loop (yellow) and the kα12 helix within the KD is well folded. The N terminus of the PX domain is located 15.6 Å away from the C terminus of KD. c, Comparison of PI3K architectures. Surface representations of PI3Kα (PDB 2rD0; that is, a class I PI3K), PI3KC2α ΔN+ΔC−C2 (as in b) and its C-C2 domain (PDB 6BTY) (that is, a class II PI3K), and Vps34 (PDB 5DFZ; that is, class III PI3K). The HD and N-KD of PI3Kα contact the N-terminal domain of PI3Kα; the p85α regulatory subunit associates with the aBD. PI3KC2α lacks regulatory subunits but contains a unique HBD as well as lipid-binding PX and C-C2 domains that regulate membrane binding and activity. In Vps34 the N-terminal C2 domain acts as a protein interaction hub for its associated subunits Vps15, Vps30 and Vps38. region 36 , a residue not conserved in PI3KC2α ( Fig. 3b and Extended Data Fig. 4c). PIK-90, which also displays profound off-target activity toward PI3KC2α (IC 50 PI3KC2α = 78 nM), occupies a similar position to Torin-2 in the ATP site. The imidazoquinazoline ring of PIK-90 binds to the adenine pocket via a single hydrogen bond and hydrophobic region II ( Fig. 3d and Extended Data Fig. 4d), while the pyridine ring in the terminal hinge of PIK-90 targets the innermost region of the affinity pocket (Fig. 3d,e). In contrast to Torin-2, association of PI3KC2α with PIK-90 involves comparably minor conformational movements of K1138 and D1268 (Fig. 3e). We conclude that Torin-2 and PIK-90 bind to PI3KC2α via the conserved ATP-binding site common to PI3K-related kinases and PI3Ks and in a manner similar to complex formation with their target enzymes (Extended Data Fig. 4c,d), providing a structural basis for their high affinity but moderate selectivity.
Based on these data and the size and chemical composition of the catalytic site in PI3KC2α we predict that PI3KC2α should in principle be amenable to selective targeting by high-affinity small molecule inhibitors.
Conformational control of PI3KC2α activity. Unlike class I and class III PI3Ks that are activated by membrane binding of their associated subunits 2,4,7,27,34 , class II PI3Ks such as PI3KC2α are autoregulated by their lipid-binding distal PX and C2 domains 30 . To structurally dissect the mechanism of PI3KC2α autoregulation, we purified near full-length PI3KC2α only lacking the intrinsically disordered N-terminal region but containing the distal PX and C2 domains (PI3KC2α ΔN ) (Extended Data Fig. 1a,b). We then determined the three-dimensional (3D) structure of PI3KC2α ΔN by single-particle cryo-EM ( Fig. 4 and Extended Data Fig. 5). Tilted data collection followed by two-dimensional (2D) classification resulted in a reconstruction with a nominal resolution of 4.4 Å from roughly 600,000 particles (Table 2) with substantial resolution anisotropy and the highest resolution in the core domain (Extended Data Fig. 5). The cryo-EM structure of PI3KC2α ΔN unequivocally showed the catalytic core of PI3KC2α, which was overlaid almost perfectly with the PI3KC2α crystal structure. Two additional EM densities were located underneath the core region: a donut-shaped density in 2D classes that adopted a barrel shape in the 3D map. This density could accommodate the previously solved crystal structure of the C-terminal C2 domain of PI3KC2α (ref. 37 ) (Fig. 4). A second, less well-defined density was located in the vicinity of the C-terminal lobe of the catalytic kinase core domain, which likely corresponds to the lipid-binding PX domain (Fig. 4).
As the PX domain exhibited significant flexibility within PI3KC2α, we complemented our results from single-particle cryo-EM by crosslinking-mass spectrometry (XL-MS). This analysis (Supplementary Table 2) imposed distance constraints (30 Å, measured by the Cα-Cα distance between two crosslinked residues) of the PX domain relative to the C-C2 domain and the C-terminal lobe of the kinase core domain (Extended Data Fig. 6a-c), and supported docking of the PX domain into the cryo-EM structure 38 . We then used the information derived from cryo-EM, XL-MS and the crystal structure of PI3KC2α core to generate a composite 3D model of PI3KC2α. Ιn this model, only the N-terminal half of kα12 (kα12(N)) is folded whereas the remainder of it is disordered. kα12(N) interacts with the loop between kα7 and kα8 of the KD forming closed contact I. The disordered C-terminal half of kα12 provides the flexibility necessary to place the C-C2 domain in the vicinity of the RBD, where it forms another set of interactions referred to as closed contact II (Fig. 5a). These contacts precisely map to putative autoinhibitory interfaces defined earlier by HDX-MS analysis 30 . In particular, we had reported that mutation of 1303 EKP 1305 to 1303 KKT 1305 (that is, 'KKT mutant'), in the kinase core domain in closed contact I, results in elevated lipid kinase activity. Moreover, HDX-MS analysis had identified a putative intramolecular interaction between the RBD and the distal PX-C2 domain module 30 . Our integrative structural analysis identifies this interaction as closed contact II, formed by K426, W458, D461 and D462 on the RBD. Based on these data, we hypothesized that the composite structure of PI3KC2α ΔN represents an inactive conformation that is stabilized by closed contacts I and II. We further experimentally tested this model. Affinity chromatography experiments showed that the immobilized glutathione S-transferase-(GST-)tagged C2, but not the PX domain, directly associates with PI3KC2α ΔN+ΔPX−C2 but less well with a mutant version of PI3KC2α, in which the RBD binding interface with the distal C2 domain had been perturbed by mutations in the closed contact I interface (K426A, W458A, D461A, D462A; that is, the 'RBD mutant') (Extended Data Fig. 6d). Disruption of either closed contact I in the KKT mutant ( 1303 EKP 1305 to 1303 KKT 1305 ) or of contact II in the RBD mutant (K426A, W458A, D461A, D462A) significantly increased the PI(3,4)P 2 -synthesizing activity of PI3KC2α, a phenotype further augmented in a double KKT/RBD mutant of PI3KC2α, in which both inhibitory interfaces were disrupted (Fig. 5c).
As PI3KC2α transitions from the inactive closed to the active open conformation, the C2 and PX domains likely are dislodged from their positions at the RBD and kinase core, respectively, while kα12 undergoes repositioning and folding into a complete helix (Fig. 5a,b). Consistently, we observed the PX domain to be displaced from the kinase core, and closed contact I involving the N terminus of kα12 to be disrupted in the crystal structure of PI3KC2α ΔN+ΔC−C2 (Fig. 1b and Extended Data Fig. 3a). Instead, in this open conformation of PI3KC2α, the kα12 helix interacts via hydrogen bonding of H1391 with the backbone of a short helical segment within the intermediate section of the activation loop (Fig. 5b). In the open conformation, the C-terminal part of the activation loop known to be crucial for lipid binding remains flexible to enable catalysis.
This mechanism is distinct from the function of kα12 in other PI3Ks. PI3Kγ uses kα12 to capture the catalytic loop in an inactive state (Extended Data Fig. 7a), whereas in VPS34, kα12 interacts with the N terminus of VPS15 to trap the activation loop 7 (Extended Data Fig. 7b). Finally, conformational opening requires the lipid-binding distal C2 domain to be displaced from its inhibitory contact with the RBD. In the closed conformation, association with the RBD renders the lipid-binding surface of the C2 domain difficult to access and misaligns it with the substrate lipid-binding site in the activation loop (Extended Data Fig. 7c). Hence, we predict the distal C2 domain to be flexibly positioned away from the kinase core in the open conformation of the enzyme ( Fig. 6e and below). A further prediction from this structure-based activation mechanism is that disruption of the interaction between kα12 and the activation loop in the open form should abrogate lipid kinase activity. In vitro kinase assays confirmed that the H1391A mutation in the center of the kα12-activation loop interface resulted in a complete loss of enzymatic activity (Fig. 5c). H1391 is thus required to stabilize the open catalytically active conformation of PI3KC2α.
Our combined data indicate a molecular model for the conformational control of PI3KC2α activity by large-scale rearrangements in the position of the lipid-binding PX and C2 domains that is accompanied by refolding and repositioning of the kα12 helix critical for catalysis.

Conformational control of PI3KC2α function in cells.
We tested this structure-based model for the conformational activation of PI3KC2α at membranes by analyzing the PI(3,4)P 2 -synthesizing activity of PI3KC2α during endocytic membrane dynamics. Depletion of PI3KC2α from Cos7 cells resulted in reduced levels of PI(3,4)P 2 at endocytic plasma membrane coated pits and a concomitant reduction in clathrin-mediated endocytosis of transferrin. These defects were rescued by re-expression of the small interfering RNA-resistant wild-type (WT) enzyme (Fig. 5d-f) or a mutant lacking the HBD (Fig. 6a), in agreement with its presumed scaffolding role during mitosis. In contrast, PI3KC2α mutant versions defective in PI(4)P substrate binding (K1283A,R1284A) or lacking critical hydrogen bonding via H1391 to stabilize the open conformation (H1391A) failed to restore PI(3,4)P 2 levels and defective endocytosis. Conversely, conformational activation of PI3KC2α by disrupting closed contacts I and II via the combined KKT and RBD mutations led to elevated cellular PI(3,4)P 2 synthesis and a gain in endocytic transferrin uptake (Fig. 5d-f). These results confirm that structural changes in the position of the lipid-binding PX and C2 domains and the kα12 helix underlie the conformational activation of PI3KC2α at membranes in vivo.
A further prediction from our combined structural and biochemical studies is that distinct structural elements mediate the catalytic roles of PI3KC2α in endocytic membrane dynamics and its noncatalytic function at the mitotic spindle 1,14 . The unique HBD of PI3KC2α, which points away from the KD (Figs. 1b and 4c), is dispensable for catalytic activity in vitro (Extended Data Fig. 1c) and for endocytosis in vivo (Fig. 6a). We therefore hypothesized, that the HBD might facilitate targeting of the enzyme to the mitotic spindle by associating with the microtubule-binding protein TACC3 (ref. 26 ). Consistently, we found that the HBD of PI3KC2α with its four antiparallel α-helices displays strong structural homology to cytoskeletal proteins, such as the focal-adhesion targeting domain of Crk-associated substrate (Cas) and to the F-actin binding domains of vinculin and α-catenin (Fig. 6b). To probe the possible function of the HBD in targeting of PI3KC2α to the mitotic spindle, we examined the subcellular localization of an N-terminally  truncated PI3KC2α lacking the clathrin binding region, a mutant version thereof, in which the HBD was deleted (PI3KC2α ΔN+ΔHBD ), or the isolated HBD alone (PI3KC2α HBD ). The HBD was sufficient for targeting to the mitotic spindle and for association with TACC3 (Fig. 6c,d), whereas deletion of the HBD abrogated the spindle localization of PI3KC2α ΔN (Fig. 6c) and complex formation with TACC3 (Fig. 6d). These data uncover the unique HBD as an important structural element that underlies the scaffolding function of PI3KC2α at the mitotic spindle 26 . We note that while the presence of the HBD is conserved among the members of the class II PI3K subfamily, its relatively low level of sequence conservation (Extended Data Fig. 2) suggests that they interact with different protein binding partners to execute putative noncatalytic functions.

Discussion
Our integrated structural analysis of PI3KC2α reveals different conformational states of the enzyme that suggest a molecular model for the local activation of PI3KC2α at endocytic membranes (Fig. 6e). In its cytosolic form, the enzyme is present in a closed inactive conformation that is stabilized by intramolecular contacts within the kinase core domain that occlude catalysis and an inhibitory interface between the RBD and the distal C2 domain, which may be further augmented by placement of the PX domain at the interface between the distal C2 and the C-terminal lobe of the kinase core domain, in agreement with our earlier biochemical data 30 . Clathrin-mediated recruitment 15,39 and activation of PI3KC2α at endocytic membranes involves a large-scale conformational change within the single subunit enzyme that releases PI3KC2α from autoinhibition to enable local PI 3-phosphate synthesis (Fig. 5). In this active conformation, the substrate PI(4)P is bound by basic residues (that is, K1283, R1284) within the PI3KC2α activation loop (Fig. 2d-f). Of note, these residues are absent from Vps34, a PI3K that is unable to use PI(4)P as a substrate, providing a molecular explanation for the distinct catalytic activities of PI3KC2α and related class II PI3Ks (refs. 1,2,14 ). The mechanism of activation of PI3KC2α is distinct from that of all other PI3Ks, in which membrane binding and catalytic activation are induced by conformational transitions in tightly associated accessory subunits (compare Fig. 1c). In class I PI3K, hydrophobic residues in the C-terminal tail of the KD not present in PI3KC2α as well as basic amino acids in the charged activation loop are only allowed to contact the membrane once the enzyme has been released from allosteric inhibition by its regulatory p85 subunit 40 .   The class III Vps34 complex binds to membranes via the tips of two arms that is three of its four subunits: one contact is formed by the catalytic Vps34 subunit and the Vps15 myristoylation site, the other one involves the Vps30/Beclin 1 BARA domain 7 . The unique mechanism of membrane binding and activation of PI3KC2α is not only interesting from a mechanistic viewpoint, but also bears important implications for our understanding of class II PI3K biology. Our structural data predict that the conformational activation and, thereby, the catalytic function of PI3KC2α is triggered by multiple coincident signals, most notably, the membrane association of its PX and C2 domains 30,37 . The exquisite lipid-binding specificity of these domains for PI(4,5)P 2 thereby limits PI3KC2α activity to nanoscale sites enriched in PI(4,5)P 2 , providing a structural explanation for the observed spatiotemporal restriction of PI3KC2α-mediated synthesis of PI(3,4)P 2 or PI(3)P at late-stage endocytic pits 15,33,41 and the base of primary cilia 13,42 . Our structural and biochemical data further predict that the conformational activation of PI3KC2α, and likely other class II PI3K family members, further requires or is facilitated by complex formation of the RBD with an endocytic Rab protein 12,13 . Rab association would aid displacement of the distal C2 domain from the RBD and, thereby, act as a third coincidence determinant, in addition to clathrin 15,39 and PI(4,5)P 2 (ref. 30 ). The structure-based mechanism for the activation of PI3KC2α at membranes described here therefore predicts that the multiple physiological functions of PI3KC2α, for example in endocytic receptor internalization and recycling 12,15,16,33 , VEGF-driven angiogenesis 19 and viral replication 21,22 , result from and are defined by the coincident interaction of PI3KC2α with PI(4,5)P 2 and different Rab proteins that steer its catalytic activity to distinct nanoscale sites. Identifying the respective Rab protein underlying these activities will be key to our understanding of the physiological functions of PI3KC2α in cell physiology and disease.
Additionally to providing insights into the mechanism of PI3KC2α activation and function, our structural analysis of PI3KC2α in complex with nonselective PI3K inhibitors will undoubtedly serve as a door-opener for rational development of isoform-selective PI3KC2α inhibitors and other class II PI3K family members that may provide new therapeutic avenues for the treatment of important human diseases such as thrombosis 28 , viral infection 21,22 , diabetes 29 or cancer 1,14 . Finally, we provide a structural basis for the scaffolding function of PI3KC2α at the mitotic spindle that involves the association of its unique HBD with the microtubule-associated kinetochore protein TACC3 (Fig. 6). These structural insights will be of relevance to develop new therapeutics to fight cancer and cancer metastasis 14,26 .

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41594-022-00730-w.

Methods
Oligonucleotides. Oligonucleotide sequences used in this study are listed in Supplementary Table 1. Cell lines. HeLa, human embryonic kidney 293T (HEK293T) and Cos7 cells were obtained from ATCC and cultured in DMEM with 4.5 g l −1 glucose (Lonza) containing 10% heat-inactivated FBS, 100 U ml −1 penicillin and 100 μg ml −1 streptomycin (Gibco). Cells were routinely tested for and devoid of mycoplasma contamination.
Negative stain screening of PI3KC2α ΔN in buffer with different salt concentrations. Purified PI3KC2α ΔN (roughly 2.4 mg ml −1 , in 20 mM Tris-HCl buffer, with 300 mM NaCl at pH 7.4) was diluted to a final concentration of 0.02 mg ml −1 into buffers containing 20 mM Tris-HCl, and various concentrations of NaCl (50 to 300 mM) at pH 7.4, before being negatively stained with 2% (w/v) uranyl formate. Negatively stained images were collected using a Tecnai Spirit BioTwin transmission electron microscope (Thermo Fisher Scientific) at 120 kV at a nominal magnification of ×49,000 (2.26 Å per pixel) on a Gatan Rio CCD camera (4,000 × 4,000). The sample diluted in a buffer containing 100 mM NaCl presented a good particle distribution without showing significant aggregates. A group of images was collected at a defocus around −1.5 to −3.5 μm. Contrast transfer function (CTF) estimation was performed with CTFFIND 4.1 (ref. 48 ). Particles were auto-picked using Gautomatch-0.53 (https://www2.mrc-lmb.cam.ac.uk/ research/locally-developed-software/zhang-software/), and 2D class averages were performed with RELION-2.0 and higher 49 .
Cryo-EM sample preparation of PI3KC2α ΔN . Purified PI3KC2α ΔN (0.8 mg ml −1 in 20 mM Tris-HCl, 100 mM NaCl at pH 7.4) were used for plunge-freezing. Double-application of 3 μl of diluted PI3KC2α ΔN were applied onto freshly plasma-cleaned (NanoClean, model 1070, Fischione Instruments) QUANTIFOIL Holey Au-carbon-R2/2 specimen grids, and vitrified by plunge-freezing into liquid ethane using a Mark IV Vitrobot device (Thermo Fisher Scientific). Cryo-EM specimen prepared with different blotting conditions (blot force, blot time) were screened. Final datasets were collected from the cryo-EM grids with thinner uniform ice thickness and good particle orientation distribution.
Cryo-EM single-particle data collection of PI3KC2α ΔN specimen. The cryo-EM single-particle datasets of PI3KC2α ΔN were collected without and with −30° stage-tilting on a Titan Krios cryo-transmission electron microscope (Thermo Fisher Scientific) operated at 300 kV and equipped with a K3 direct electron detector (Gatan, Inc.) device at a nominal magnification of ×105,000 yielding a pixel size at the specimen of 0.837 Å per pixel in counting mode (Table 2). Videos were collected with EPU, a data collecting automation software package (Thermo Fisher Scientific), with an imaging setting of total exposure equal to 60 electrons over 50 fractions in 3 s and a defocus range between −1.5 and −2.8 μm, at the Max Planck Institute of Biophysics, Frankfurt, Germany.
Cryo-EM image processing. During data collection, all datasets were preprocessed 'on-the-fly' using cryoSPARC live 50 running video motion correction, CTF estimation and automatic particle picking and stream 2D classification to estimate particle quality. As the streamed 2D classification showed particle features as expected, more data automated positions were defined. Once data collection was completed, all videos from both datasets were imported into cryoSPARC, processed with patch motion correction and CTF estimation, and auto-picked separately. The auto-picked particles were inspected and extracted with a box size of 260 pixels to perform particle cleaning using several rounds of 2D classification and 3D heterogeneous refinement (3D classification, if effective) with C1 symmetry, accordingly. The most homogenous particle sets after cleaning and separation from both datasets were merged, containing about 1 million particles, and taken to perform 3D homogeneous refinement and further 2D classifications as necessary. 3D variability was performed to show the heterogeneity of the PI3KC2α ΔN specimen due to its flexibility. Fourier shell correlation (FSC) estimation was done both in cryosparc and Relion. Postprocessing and local resolution was estimated in RELION-3.0, and global directional resolution estimation was performed using 3DFSC (ref. 51 ). A detailed image processing pipeline is shown in Extended Data Fig. 5. Cryo-EM data were deposited in the PDB and are available under accession number EMD-12191. Original EM micrographs were deposited in the Electron Microscopy Public Image Archive (EMPIAR) (code EMPIAR-10665).
Transferrin uptake and surface labeling. Cos7 cells transfected with siRNA and/or PI3KC2α (WT or mutant)-encoding plasmids or pretreated for 4 h with 0.1% dimethylsulfoxide or 20 μM PITCOIN1 were starved in serum-free DMEM media for 1 h. For transferrin uptake, cells were incubated with 25 mg ml −1 Alexa647 labeled transferrin (Molecular Probes, Invitrogen) for 10 min at 37 °C in a humidity chamber. Cells were washed twice with ice-cold PBS supplied 10 mM MgCl 2 and then acid washed twice at pH 5.3 (0.2 M sodium acetate, 200 mM sodium chloride) on ice for 2 min to remove surface-bound transferrin. Cells were then washed twice more with ice-cold PBS containing 10 mM MgCl 2 and fixed with 4% PFA for 45 min at room temperature. For surface labeling, cells were incubated with 25 mg ml −1 Alexa647 labeled transferrin at 4 °C for 45 min and then washed three times with ice-cold PBS (10 mM MgCl 2 ) on ice for 1 min. Cells were fixed with 4% PFA for 45 min at room temperature. Transferrin labeling was analyzed using the Nikon Eclipse Ti microscope and ImageJ software. Internalized transferrin per cell was quantified and normalized to the amount of surface-bound transferrin determined in the same experiment as a measure for the efficiency of internalization.
Statistical analysis. All data are presented as mean ± s.e.m. and were obtained from ≥3 independent experiments with total sample numbers provided in the figure legends. Statistical significance was evaluated by Prism software (GraphPad), using one simple, two-tailed t-test with theoretical mean of 100 or one-way analysis of variance (ANOVA) test with Tukey's multiple comparisons. Specific P values are indicated in the legends to figures. Significant differences were marked as *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Fig. 7 | Comparison of the inhibitory conformations of PI3Ks. (a) auto-inhibited conformation of PI3Kγ (pdb: IE8X). kα12 (red) bends inward towards the catalytic pocket. (b) auto-inhibited conformation of yeast Vps34/Vps15 in complex II (pdb: 5DFZ). The activation loop of VPS34 (yellow) is locked by kα12 (red) and the N-terminus of VPS15 (blue). (c) auto-inhibited conformation of PI3KC2α ΔN as ribbon and surface representation. The distal C2 domain binds to the rBD. The yellow and orange regions represent the catalytic pocket and the activation loop, where lipid substrates bind. The PI(4,5)P 2 -binding surface in the distal C2 domain is colored in blue. The locations of the lipid-binding pockets are distributed on different sides of the PI3KC2α molecule in the closed conformation.