Development of selective inhibitors of phosphatidylinositol 3-kinase C2α

Phosphatidylinositol 3-kinase type 2α (PI3KC2α) and related class II PI3K isoforms are of increasing biomedical interest because of their crucial roles in endocytic membrane dynamics, cell division and signaling, angiogenesis, and platelet morphology and function. Herein we report the development and characterization of PhosphatidylInositol Three-kinase Class twO INhibitors (PITCOINs), potent and highly selective small-molecule inhibitors of PI3KC2α catalytic activity. PITCOIN compounds exhibit strong selectivity toward PI3KC2α due to their unique mode of interaction with the ATP-binding site of the enzyme. We demonstrate that acute inhibition of PI3KC2α-mediated synthesis of phosphatidylinositol 3-phosphates by PITCOINs impairs endocytic membrane dynamics and membrane remodeling during platelet-dependent thrombus formation. PITCOINs are potent and selective cell-permeable inhibitors of PI3KC2α function with potential biomedical applications ranging from thrombosis to diabetes and cancer.

P hosphoinositide 3-kinases (PI3Ks) are a family of ubiquitously distributed lipid kinases that phosphorylate inositol lipids at the 3′-OH group to regulate cell and organismal physiology. Mammals including humans express three distinct classes of PI3Ks, consistent with the large repertoire of processes that are controlled by phosphatidylinositol (PI) 3-phosphates [1][2][3] . Heterodimeric class I PI3Ks signal downstream of plasma membrane-bound receptors and small GTPases (for example, Ras) via synthesis of PI 3,4,5-trisphosphate (PI(3,4,5)P 3 ) 2,4-6 to regulate metabolism, proliferation and migration. Aberrant class I PI3K signaling has a prominent role in cancer and immunity and, hence, isoform-selective as well as pan-class I PI3K inhibitors have been the focus of extensive drug development efforts and hold promise as anticancer therapeutics among other biomedical applications 1,7 . Vps34, the sole class III PI3K member, is present in two related protein complexes that synthesize PI 3-monophosphate (PI(3)P) within the endolysosomal system and during autophagy as well as in cytokinesis 1 . The recent development of selective Vps34 inhibitors 8,9 has greatly facilitated the analysis of class III PI3K function in biology and may be of potential clinical use in cancer and other disorders related to autophagy or rare genetic diseases caused by imbalance of PI 3-phosphate levels (for example, Charcot-Marie-Tooth disease type 4B) 10,11 .
Class II PI3Ks comprise PI3KC2α, PI3KC2β and PI3KC2γ isoforms and are unique in their ability to directly synthesize PI 3,4-bisphosphate (PI(3,4)P 2 ) from PI 4-monophosphate (PI(4)P) in vivo. In addition, they have been suggested to contribute to cellular PI(3)P production on endosomes and in the autophagy/lysosomal pathway 7,[12][13][14] . PI3KC2α and PI3KC2β are ubiquitously expressed, whereas PI3KC2γ is mostly expressed in exocrine glands and liver. PI3KC2α is encoded by an essential gene in mice 14 and has been implicated in a variety of cellular and organismal functions. On a cellular level, PI3KC2α controls clathrin-mediated endocytic vesicle formation 15,16 and internalization of vascular endothelial growth factor (VEGF) 17 and transforming growth factor β receptors 18,19 , endocytic recycling 13 , cytokinesis 20 , primary cilia function 14 and insulin signaling 21 . Accumulating evidence further implicates PI3KC2α in cancer 1,22,23 and cataract formation 20 , regulation of blood pressure 24 , viral replication 25,26 , angiogenesis 17 and platelet-dependent thrombus formation as well as membrane shear-dependent platelet adhesion [27][28][29] . In spite of its importance for cell and organismal physiology and the potential biomedical use of PI3KC2α inhibitors, for example, as antithrombotic, antidiabetic or antiangiogenic drugs, no isoform-selective inhibitors of PI3KC2α catalytic activity are available.
Here we report on the development and characterization of PhosphatidylInositol Three-kinase Class twO INhibitors (PITCOINs) 1-3 (compounds 1-3), potent and highly selective cell-permeable small-molecule inhibitors of PI3KC2α activity and function. High-resolution co-crystal structures of PI3KC2α in complex with PITCOIN inhibitors reveal a unique mode of interaction with the adenosine triphosphate (ATP)-binding site that provides a rationale for the exclusive selectivity of these inhibitors for PI3KC2α with no off-target activity toward other lipid and protein kinases. PITCOINs impair PI3KC2α-mediated synthesis of PI (3,4

)P 2 and
Structural basis for PI3KC2α inhibition by PITCOINs. To unravel the structural determinants of inhibitor potency and specificity for PI3KC2α, we capitalized on the recent determination of high-resolution structures of PI3KC2α in its active and inactive conformations by cryogenic electron microscopy (cryo-EM) and protein X-ray crystallography 30 . PI3KC2α contains a compact PI3K catalytic core (PI3KC2α core ) that comprises a Ras-binding domain and an N-terminal C2 domain as well as helical and kinase domains and displays a typical PI3K kinase domain fold 30 . The ATP-binding site of the unliganded apo form of PI3KC2α core is located in a cavity between the N-and C-lobes of the kinase domain (Fig. 2a) and encompasses the (1) adenine-binding pocket, (2) affinity pocket and (3) specificity pocket. The adenine-binding pocket consists of a hinge region (L1186, V1187 and P1188) that connects the N-and C-lobes of the kinase domain, a hydrophobic region (F1255, M1257 and I1267) that forms the mouth of the pocket and a conserved aromatic residue (F1172). The small affinity pocket harbors conserved hydrogen-bond donor or acceptor residues (for example, K1138 and D1268). The specificity pocket comprises P-loop residues (F1112 and S1113) and the PI3KC2α-unique residue M1136 (Fig. 2a, inset). A notable feature of the ATP-binding site of PI3KC2α is that L1186 within the kinase hinge is tightly packed against the P-loop residue F1112, which might rigidify its conformation.
To explore the structural basis of the inhibitory mechanism of PITCOIN1-PITCOIN3, we determined the structures of PI3KC2α core in complex with inhibitors at maximal resolutions ranging from 2.5 to 2.9 Å (Fig. 2b, Table 3). PITCOINs adopt nearly identical positions in the ATP-binding site. Their pteridinone scaffold occupies the adenine-binding pocket, while the two vicinal (R 1 and R 2 ) arms that extend outward from the pteridinone moiety display specific interactions with the N-lobe of the kinase domain and cause the inhibitors to adopt a propeller-shaped conformation. The affinity pocket remains empty. The pteridinone scaffold of PITCOIN1-PITCOIN3 forms a single hydrogen bond with the backbone of V1187 located in the kinase hinge region and embeds into hydrophobic surfaces provided by M1136 and L1186 at the top, M1257 and I1267 at the bottom, and F1172 on the inside of the adenine-binding pocket. The terminal phenyl group that forms the R 2 arm binds to F1112 on the P-loop via hydrophobic interactions (Fig. 2b,c and Extended Data Fig. 3a-d). The importance of the hydrophobic R 2 interactions is underscored by the fact that substitution of the isopropyl group in the initial hit compound 4 by a phenyl moiety led to considerably increased inhibitory potency (that is, a >5-fold-improved IC 50 ; Fig. 1c). The R 1 arm displays an L-shaped conformation, in which one edge contacts S1190 within the C-lobe, while the thiazole group is placed into a hydrophobic pocket formed by L1186 within the kinase hinge and the N-lobe residues K1120 and N1134 ( Fig. 2c and Extended Data Fig. 3b,d), that is, a site that has not been identified previously as an inhibitor target site in other PI3Ks.
PITCOIN2 and PITCOIN3 are distinguished from PITCOIN1 by the presence of bulky hydroxyphenyl (that is, PITCOIN2) or N-phenylmethanesulfonamide (that is, PITCOIN3) substituents on the 4′ position of the thiazole group. The space demand of these polar and bulky groups leads to additional hydrophobic interaction between P1188 within the kinase hinge and the phenyl group on the R 1 arm of PITCOIN2 and PITCOIN3. As a result of these alterations, the mild off-target activity of PITCOIN1 against PI3KC2γ is lost in PITCOIN2 and PITCOIN3. Interestingly,     the terminal hydroxyphenyl and N-phenylmethanesulfonamide groups cause the respective phenyl moiety to adopt different orientations upon complex formation of PITCOIN2 and PITCOIN3 with PI3KC2α. The hydroxyphenyl group of PITCOIN2 faces inward and is stabilized through interaction with the solvent, whereas the N-phenylmethanesulfonamide group of PITCOIN3 forms hydrophobic interactions with E1131 (Fig. 2b,c and Extended Data Fig. 3c,d).
Interestingly, many of the target residues of the protein backbone contacted by the R 1 and R 2 arms of PITCOIN1-PITCOIN3 are not conserved in other class II PI3K isoforms, for example, N1134 (PI3KC2β: R1079, PI3KC2γ: S916), L1186 (PI3KC2β: M1131, PI3KC2γ: M968) and S1113 (PI3KC2β: N1058, PI3KC2γ: T895) (Extended Data Fig. 4a). Together with the different lengths and space requirements of the R 1 arms of PITCOIN1 versus PITCOIN2 or PITCOIN3, these features provide a structural explanation for the high degree of selectivity of PITCOINs for PI3KC2α over other class II PI3K isoforms and further members of the PI3K family. Consistent with this, we observe that placement of PITCOIN1 (Extended Data Fig. 4b,c) or PITCOIN3 (Extended Data Fig. 4d,e) into the ATP-binding pockets of class I PI3Kγ or Vps34 results in a steric clash of the R 1 arm and the thiazole phenyl group of PITCOIN3 with residues in the ATP-binding site.
Collectively, these results from structural-biochemical analysis of the protein-ligand interaction provide a molecular explanation for the potency and specificity of PITCOINs for PI3KC2α.

PITCOINs impair PI(3,4)P 2 synthesis and endocytosis.
To assess the use of PITCOINs for targeting PI3KC2α activity in mammalian cells, we determined whether these compounds might be cytotoxic. Sustained exposure for 20 h of HeLa cells to different concentrations of PITCOIN1 or PITCOIN3 up to 100 µM in the presence or absence of serum did not result in detectable cytotoxicity as measured by trypan blue staining (Fig. 3a). PITCOINs also were nontoxic if analyzed by the lactate dehydrogenase (LDH) assay in a variety of different cell types, for example, Cos7, HEK293 and HepG2 cells (Extended Data Fig. 5a-c).
Previous studies have shown that PI3KC2α synthesizes a local plasma membrane pool of PI(3,4)P 2 that drives endocytic membrane remodeling by the PI(3,4)P 2 -binding PX-BAR domain protein sorting nexin 9 (SNX9) 15,33,34 . Consistent with this, pharmacological inhibition of PI3KC2α activity in the presence of PITCOIN1 or PITCOIN3 led to a progressive dose-dependent block of plasma membrane tubule formation induced by constitutively active enhanced green fluorescent protein (eGFP)-SNX9 with half-maximal effective concentration (EC 50  4.6 µM, respectively ( Fig. 3b and Extended Data Fig. 5d), that is, phenocopying genetic depletion of PI3KC2α in cells 33 . To probe the cellular effects of PITCOINs on PI3KC2α-mediated synthesis of PI(3,4)P 2 directly, we monitored the local levels of PI(3,4)P 2 at endocytic clathrin-coated pits (CCPs) in Cos7 cells expressing eGFP-clathrin light chain. Acute perturbation of PI3KC2α activity by pretreatment of cells with PITCOIN1 or PITCOIN3 depleted PI(3,4)P 2 from endocytic CCPs (Fig. 3c,d) and resulted in impaired clathrin-mediated endocytosis of transferrin, akin to genetic loss of PI3KC2α (Fig. 3e). Clathrin-independent fluid-phase uptake of fluorescent large dextrans 35 proceeded unperturbed in the presence of PITCOINs (Extended Data Fig. 5e,f). Depletion of PI(3,4)P 2 from CCPs and inhibition of transferrin endocytosis were accompanied by attenuated dynamics of CCPs as evidenced by a significant increase in CCP lifetime (Fig. 3f,g), that is, a hallmark of PI3KC2α deficiency 15,16,36 . In contrast, synthesis of plasma membrane PI(4,5) 2.0 P 2 or PI(4)P was unaffected by PITCOIN1 or PITCOIN3 (Extended Data Fig. 6a-d). Application of PITCOINs also had no effect on the Golgi pool of PI(4)P, which is mainly synthesized by PI 4 kinase IIIβ 37 (Extended Data Fig. 6e,f). These results demonstrate that PITCOIN1 and PITCOIN3 potently and specifically inhibit the activity of PI3KC2α but no other PI kinases and, thereby, impair endocytic plasma membrane remodeling in living mammalian cells.

PI3KC2α contributes to endosomal PI(3)P synthesis.
Apart from the synthesis of PI(3,4)P 2 at plasma membrane endocytic sites and at the cytokinetic bridge 20 , PI3KC2α has been implicated in contributing to the formation of endosomal pools of PI(3)P in a cell-typeand/or context-specific manner 7,12,14,17,21,27,38 . A major alternative route of PI(3)P formation on endosomes involves Vps34 (refs. 4,37 ), an enzyme that can be selectively inhibited by VPS34-IN1 (ref. 9 ). To assess the contribution of PI3KC2α and Vps34 to the overall synthesis of PI(3)P on endosomes, we monitored the cellular levels of PI(3) P in Cos7 cells treated with PITCOINs or VPS34-IN1 using the recombinant eGFP-2×FYVE domain of Hrs as a sensor. Inhibition of Vps34 in the presence of VPS34-IN1 greatly reduced endosomal PI(3)P levels, consistent with prior data 8,9,39 . Interestingly, pharmacological blockade of PI3KC2α by either PITCOIN1 or PITCOIN3 also led to a significant, albeit less pronounced, reduction in endosomal PI(3)P (Fig. 4a,b). Endosomal membrane recruitment of the PI(3)P-binding effector early endosomal antigen 1 (EEA1) was concomitantly reduced (Fig. 4c,d). This was surprising as previous work by us had failed to detect significant changes in PI(3)P levels in Cos7 cells with sustained genetic depletion of PI3KC2α 16 . To rule out possible off-target effects of PITCOINs on other pathways of PI(3)P synthesis, we tested the effect of PITCOINs on endosomal PI(3)P in PI3KC2α-depleted Cos7 cells (Fig. 4e,f). Lack of PI3KC2α had no effect on PI(3)P, corroborating our earlier data 16 . Notably, application of PITCOIN1 or PITCOIN3 failed to elicit changes in PI(3)P levels in PI3KC2α-depleted Cos7 cells, confirming their exquisite PI3KC2α target specificity. Furthermore, treatment of starved Cos7 cells with PITCOIN1 or PITCOIN3 did not affect the formation of LC3-II-containing autophagosomes, a process that largely depends on Vps34-mediated formation of an autophagy-specific pool of PI(3)P in most cell types (Extended Data Fig. 7a,b).
These results show that PI3KC2α contributes to endosomal PI(3)P synthesis, either directly by phosphorylating PI or indirectly via production of PI(3,4)P 2 , which can subsequently be hydrolyzed to PI(3) P by endosomal INPP4A/INPP4B 4-phosphatases 3,16 . They further suggest that sustained genetic loss of PI3KC2α in Cos7 fibroblasts leads to compensatory changes in PI(3)P metabolism, for example, via regulation of lipid phosphatases that remain to be identified.

PITCOIN3 impairs platelet membrane and thrombus formation.
A major cell type reported to depend on PI(3)P synthesis mediated by PI3KC2α is platelets 27-29 , megakaryocyte-derived blood cells of major biomedical relevance due to their physiological role in thrombus formation and, thus, as a target for antithrombotic drugs. Previous studies in kinase-inactive PI3KC2α knock-in mice have revealed a critical role for the enzyme in controlling platelet membrane morphology and function via synthesis of an agonist-insensitive pool of PI(3)P 27,29 . Moreover, application of PIK-90-derived PI3K inhibitors with activity against PI3KC2α and other PI3Ks including PI3KC2β and class I PI3Kα has provided proof of principle that acute perturbation of PI3KC2α function may impair thrombosis without suppression of canonical platelet activation mechanisms that affect bleeding 28 . The interpretation of these data is, however, compromised by the off-target activities of the broad-spectrum PI3K inhibitors used.
Based on these prior works, we hypothesized that acute specific inhibition of PI3KC2α activity by PITCOIN3 should recapitulate key phenotypes elicited by genetic PI3KC2α kinase inactivation and counteract thrombosis. Consistent with this hypothesis, we observed a dramatic decrease in basal PI(3)P levels in platelets treated with PITCOIN3 ( Fig. 5a,b), illustrating the important role of PI3KC2α in PI(3)P production in platelets. Ultrastructural analysis of PITCOIN3-treated platelets by transmission electron microscopy (EM) revealed major defects in membrane morphology, including an aberrant invaginated shape of the plasma membrane (Fig. 5c,d). The open canalicular system (OCS) that constitutes a reservoir of plasma membrane to enable platelet shape changes during activation was also altered and expanded (Fig. 5e). Moreover, acute inhibition of PI3KC2α activity by PITCOIN3 caused dramatic defects in filopodia extension in collagen-related peptide (CRP)-stimulated platelets visualized by scanning EM (Extended Data Fig. 7c,d). Platelet processes known to be independent of PI3KC2α 27 such as CRP-or thrombin-stimulated P-selectin exposure (Extended Data Fig. 7e), ATP release (Extended Data Fig. 7f), αIIbβ3-integrin activation (Extended Data Fig. 7g) or platelet aggregation (Extended Data Fig. 7h) were not affected significantly by PITCOIN3. Hence, acute inhibition of PI3KC2α by PITCOIN3 selectively causes major defects in platelet membrane morphology, thereby recapitulating key features of genetically induced PI3KC2α activity loss in mice 27,29 .
We therefore tested whether PITCOIN3 displays antithrombotic activity in mouse blood. To this end, we analyzed platelet-dependent thrombus formation ex vivo on a collagen matrix at an arterial shear rate of 500 s −1 during 2 min of perfusion. We found that acute inhibition of PI3KC2α activity in the presence of PITCOIN3 caused a pronounced and highly significant reduction in the prothrombotic capacity of platelets compared to DMSO-treated controls ( Fig. 5f and Extended Data Fig. 7i). Moreover, the surface covered by platelets was significantly decreased (Fig. 5g).
These data corroborate the key physiological role of PI3KC2α in controlling platelet membrane morphology and function and identify PITCOIN3 as a lead compound for further development as a potential antithrombotic drug.

Discussion
We have combined HTS with subsequent medicinal chemistry approaches to identify potent and selective small-molecule inhibitors of PI3KC2α activity and provide proof of principle for their use in dissecting the multiple roles of this enzyme in cell physiology and to combat human disease.
Multiple independent lines of evidence support the notion that PITCOIN1-PITCOIN3 potently and specifically inhibit the activity of PI3KC2α but no other PI kinases in vitro and in living cells. PITCOINs were found to be inactive against other PI3Ks including PI3Kα and Vps34 in vitro, and further selectivity profiling revealed no off-target activity of PITCOIN1-PITCOIN3 toward any of more than 117 lipid and protein kinases studied. Our MS-based Kinobead profiling showed selective competition of PIK3C2α among the 137 quantified kinases in HEK293T cells. Structural studies by protein X-ray crystallography demonstrate that PITCOIN molecules display a unique mode of interaction with the ATP-binding site of PI3KC2α. Complex formation is found to involve key interactions between the R 1 and R 2 arms of PITCOIN molecules and target residues that are not conserved in other class II PI3K isoforms (Fig. 2), providing a molecular explanation for the potency and unprecedented high selectivity of these molecules for PI3KC2α. We note that this binding mode is distinct from the off-target association of pan-PIK inhibitors such as PIK-90 with PI3KC2α, which capitalizes exclusively on binding interfaces that are conserved between PI3K family members 30 . Notably, we demonstrate that the cellular application of PITCOIN1 or PITCOIN3 recapitulates key phenotypes elicited by genetic loss of PI3KC2α 7,14,15,17,21,29 . Specifically, we show that acute pharmacological inhibition of PI3KC2α activity causes the local depletion of PI(3,4)P 2 from CCPs, resulting in impaired endocytic plasma membrane remodeling, and reduces endosomal PI(3)P levels (Figs. 3  and 4). These phenotypes are abrogated in the absence of the target enzyme. Finally, experiments in platelets show that acute pharmacological targeting of PI3KC2α recapitulates key features of genetically induced PI3KC2α activity loss in mice 27,29 , that is, defects in PITCOIn3, 50 platelets); *P < 0.05 (P = 0.0183) versus DMSO, two-tailed unpaired Student's t-test. f, Thrombus volume was measured using Imaris. Mean ± s.e.m.; n = 4 mice per condition. *P < 0.05 (P = 0.0133), ***P < 0.001 (P = 0.001) and ****P < 0.0001 versus DMSO, two-way anOVa followed by Sidak's multiple comparisons. g, Surface covered by platelets measured using ImageJ. Mean ± s.e.m.; n = 4 mice per condition. ****P < 0.0001 versus vehicle, two-way anOVa followed by Sidak's multiple comparisons test. platelet membrane morphology caused by depletion of PI(3)P and a prominent reduction in prothrombotic capacity (Fig. 5).
PITCOINs thus are first-in-class potent and selective inhibitors of PI3KC2α catalytic function and represent unique pharmacological tools to further decipher the multiple roles of PI3KC2α in biology. The observation that acute inhibition of PI3KC2α causes antithrombotic effects that occur concomitantly with a reduction in PI(3)P levels and changes in platelet membrane morphology (that is, the OCS) opens the possibility that PITCOIN3 and related compounds may serve as a new class of shear stress-dependent 12,27 antithrombotic agents that, unlike current therapies, may not increase bleeding risk. Future studies will be directed to explore this exciting perspective and to further improve PITCOINs, for example, with respect to cell membrane permeability. In addition, PITCOINs and their derivatives could also provide new therapeutic avenues for the treatment of other important human diseases related to PI3KC2α function, including viral infection 25,26 , diabetes 21 or cancer 1,7 that pertain to the established physiological roles of PI3KC2α in viral replication 25,26 , VEGF signaling and endocytosis 15,17 and the abscission reaction of cytokinesis 23 , respectively.
Finally, the structure-based development of selective PI3KC2α inhibitors reported here will serve as a door opener for the pharmacological targeting of other class II PI3Ks (that is, PI3KC2β and PI3KC2γ) with key roles in mTORC1 signaling 40 , cell migration 38,41 , inherited muscle disease, stroke 42 and insulin signaling or metabolism 43 .

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/ s41589-022-01118-z.

Methods
Animals. All mice were C57BL/6J background males purchased from Janvier Labs. Mice were housed in conventional cages under specific pathogen-free conditions in an animal room with constant temperature (20-22 °C) and humidity (50-60%) with a 12 h light/12 h dark cycle (lights on at 7:00 AM) and free access to food and water. All procedures were performed in accordance with institutional guidelines for animal research and were approved by the French Ministry of Research in agreement with European Union guidelines.
Oligonucleotides. Oligonucleotide sequences used in this study are listed in Supplementary Table 4.
Cloning and mutagenesis. Constructs for baculovirus-mediated expression in insect cells, cDNA encoding human PI3KC2α ΔN (amino acids 376-1686) amplified by PCR and cloned into pFL10His via KasI/XbaI restriction sites and mouse PI3KC2α core (amino acids 377-1400; engineered internal loop with amino acids 533-544 replaced by GSGS; HBD (amino acids 550-665) replaced by the sequence SGAGSGA) were engineered by multiple steps of site-directed mutagenesis using PCR.
Protein expression and purification. His 10 -tagged PI3KC2α ΔN and PI3KC2α core were expressed in Sf21 insect cells, using SF900-II serum-free medium (ThermoFisher). Sf21 cells (800 ml) grown to a density of 1.5-2 × 10 6 cells per ml were infected with 8 ml amplified baculovirus encoding the desired construct. Cells were collected around 72 h after virus infections. The cell viability was around 80-90%. Cell pellets were stored frozen at −20 °C until purification. For protein purification, cell pellets from each 200-ml culture were resuspended in 35 ml lysis buffer (50 mM Tris pH 7.2, 300 mM NaCl, 10 mM imidazole, 1 mM DTT, 0.5% Triton X-100, 1 tablet per 50 ml of protease inhibitor cocktail), sonicated for 1 min (1-s pulse on, 5-s pulse off) and centrifuged for 20 min at 87,200g. Each 50 ml of supernatant was incubated with 0.5 ml nickel NTA beads (Sigma) on a rotating wheel for 1 h at 4 °C. Beads were collected in a plastic open column (Bio-Rad) and washed with 20 ml lysis buffer and then with 30 ml wash buffer (50 mM Tris pH 7.5, 300 mM NaCl, 20 mM imidazole, 1 mM DTT). Protein was eluted with 8 ml elution buffer (20 mM Tris pH 7.5, 300 mM NaCl, 300 mM imidazole, 5 mM DTT). The His 10 tag was released by Tobacco Etch Virus nuclear-inclusion-a endopeptidase (TEV) (10 mg protein per 0.25 mg TEV) cleavage overnight, while dialyzing against size exclusion chromatography (SEC) buffer (20 mM Tris pH 7.5, 300 mM NaCl, 5 mM DTT) at 4 °C. Proteins were purified on a Superdex 200 gel filtration column at 4 °C with SEC buffer (20 mM Tris pH 7.5, 300 mM NaCl, 5 mM DTT). Proteins were concentrated to about 1.2 mg ml −1 for human PI3KC2α ΔN and 5 mg ml −1 for mouse PI3KC2α core . Purified human PI3KC2α ΔN was flash frozen in liquid nitrogen and stored at −80 °C, and mouse PI3KC2α core was immediately used for crystallization. VPS34 complex II was purified as described previously 44 .
Compound library screening and IC 50 determination. Primary screens were performed using the ADP-Glo assay in a 384-well format. In total, 37,224 small molecules from the FMP library were screened for their ability to inhibit the lipid kinase activity of purified PI3KC2α. The concentration of each compound used in the primary screens was 10 μM. Kinase incubations with ATP but without PI were used as negative control samples. For IC 50 determination, purified human PI3KC2α ΔN was prediluted to 0.5 mg ml −1 in SEC buffer used for purification and subsequently diluted to 30 μg ml −1 in kinase buffer (5 mM HEPES/KOH pH 7.2, 25 mM KCl, 2.5 mM magnesium acetate, 150 mM potassium glutamate, 10 μM CaCl 2 , 0.2% CHAPS). Native liver PI was dissolved to a concentration of 1,000 µM with kinase buffer by water bath sonification and then suppliment with 50 μM ATP for IC 50 measurement. A 2× compound dilution series (with a total of 11 concentrations) starting from 10 μM was prepared in kinase buffer. Reactions were started by mixing 5 μl of compound with 3 μl of purified PI3KC2α ΔN for 5 min, followed by initiation of kinase reactions by adding 2 μl of substrates. The kinase reactions were performed for 20 min at room temperature. The reactions were stopped by adding 10 μl of ADP-Glo reagent (Promega). After incubation for 40 min, 20 μl of kinase detection reagent was added. After a further incubation for 20 min, luminescence was read with a TECAN plate reader. IC 50 was calculated according to the manufacturer's protocol for the ADP-Glo assay.
Chemical synthesis of PITCOINs. Chemical synthesis of PTCOINs was performed as indicated in the Supplementary Note.

Kinobead profiling.
Kinobead pulldown assays were performed as previously described using 5 mg ml −1 HEK293 cell lysates in IGEPAL CA-630-containing buffer 32 . Briefly, for profiling of each PITCOIN, 12 wells of a 96-well plate were filled with lysate (2.5 mg of total proteins per well) and incubated for 45 min at 4 °C in an end-over-end shaker with 0 nM (DMSO control), 30 nM, 100 nM, 300 nM, 1 μM, 3 μM, 10 μM, 30 μM, 100 μM, 300 μM and 1 mM of the PITCOINs dissolved in DMSO. Subsequently, the treated lysates were incubated with Kinobeads-ε for 30 min at 4 °C in a 96-well filter plate on an end-over-end shaker. The beads were then washed before the bound proteins were denatured and alkylated with chloroacetamide. Addition of trypsin (300 ng per well) started overnight on-bead digestion. Acidified peptide eluates were then subjected to C18 StageTip desalting 45 for LC-MS/MS analysis on an Orbitrap Fusion Lumos Tribrid (ThermoFisher Scientific) mass spectrometer coupled to an online Dionex Ultimate3000 equipped with a micro flow Vanquish pump UHPLC (ThermoFisher Scientific). MaxQuant (v.1.5.3.30.)/Andromeda was used to quantify proteins using the Swissprot reference database containing all canonical protein sequences with standard settings. Dose-response curves, Kdapps and the kinome tree were derived from the proteinGroup.txt file using a set of R scripts 46 . The MS proteomics data have been deposited to the ProteomeXchange Consortium (http:// proteomecentral.proteomexchange.org) via the PRIDE 47 partner repository dataset identifier PXD032284.
Crystallization of PI3KC2α core . Diffracting crystals of mouse PI3KC2α core were obtained as described previously 30 . In brief, concentrated protein samples (2.5 mg ml −1 ) were complexed with 0.5 mM of PITCOIN1, PITCOIN2 or PITCOIN3 overnight on ice. Samples were filtered with 0.2-μm spin filters to remove precipitates. Crystals were grown in mother liquid containing 0.1 M Tris pH 7.5, 100-200 mM MgSO 4 and 7-10% PEG 3,350. Crystals were washed and cryo-protected with mother liquid supplement with 10% ethylene glycol. Crystals were mounted in a nylon loop and flash cooled in liquid nitrogen. Data collection, model building and refinement. Diffraction data were collected at station BL14.1 of BESSY/Helmholtz Center Berlin (HZB). Images were processed with XDSAPP 48 . The CC 1/2 cutoff of diffraction data was 0.996 (0.272), 0.998 (0.172) and 0.998 (0.251) with respect to the crystal of PI3KC2α core in complex with PITCOIN1, PITCOIN2 and PITCOIN3, respectively. Values in parentheses are for the highest-resolution shell. The PI3KC2α ΔN+ΔC−C2 structure was determined by molecular replacement with the PHENIX suite 49 using the previously solved structure of PI3KC2α (PDB: 7BI4) as a search model. The structure was manually built using COOT and iteratively refined using Refmac 50 and the PHENIX suite. The different orientations of the thiazole group in PITCOIN1, PITCOIN2 and PITCOIN3 were validated via a model fitting into the electron density map and further refinement of the model. Data collection and structure refinement statistics are summarized in Supplementary Table 2. The Ramachandran statistics in the order of favored, allowed and outliers for each structure are as follows: PI3KC2α core in complex with PITCOIN1 (97.15, 2.85, 0%), in complex with PITCOIN2 (96.80, 3.08, 0.12%) and in complex with PITCOIN3 (98.27, 1.73, 0%). The rotamer outlier percentage for each structure was as follows: PI3KC2α core in complex with PITCOIN1 (4.81%), in complex with PITCOIN2 (0%) and in complex with PITCOIN3 (0%).
LDH cell toxicity assay. Toxicity was assayed by measurement of LDH released from cells due to cell death. Cos7 cells were seeded in a 96-well plate. The cells were treated with a concentration series of PITCOIN1 and PITCOIN3 for 20 h. The supernatant was collected and added to the LDH assay reagent (Sigma-Aldrich).
The reactions were performed for 20 min. Absorbance was measured at 490 nm. Values were normalized to the drug/medium background value and converted into the percentage of toxicity with 100% lysed cell treated with Triton X-100 as a control.
Analysis of CCP dynamics. A stable Cos7 cell line expressing eGFP-clathrin light chain was used for live cell imaging of CCP dynamics via TIRF microscopy. Time-lapse series of Cos7 cells pretreated with 20 µM compound for 6 h were recorded at a frame rate of 0.5 Hz for 3 min. Cells treated with 0.1% DMSO were used as a control. To quantitatively analyze CCP dynamics we used the associated cmeAnalysis MATLAB package 51 . Here the main features of the cmeAnalysis MATLAB package were combined to customized adaptations and newly written analysis functions in ref. 52 . The analysis includes first detecting CCPs independently in individual frames and then tracking and closing tight gaps in between them using the functionality from the cmeAnalysis package. To ensure that trajectories of long-lived, low-motility structures with fluctuating intensities were not broken up into small segments, which would affect the result, the tracking parameters were adapted to the experimental data. As the dynamics of CCPs should be studied in this experiment, the CCP selection criteria were adjusted in a way that allowed analysis of true internalizations only. For this purpose, the large pool of short-lived (and often highly motile) structures of unclear origin were excluded. The same intention was kept for the subset of extremely long-lived CCPs, meaning that only structures with a duration above 20 s and below 160 s were included in lifetime analysis.
Transferrin uptake and surface labeling. Cos7 cells were grown on Matrigel-coated glass coverslips and treated with 0.1% DMSO or 20 μM PITCOIN1 and PITCOIN3 for 6 h followed by starvation in serum-free DMEM for 1 h. For transferrin uptake, cells were incubated with 25 μg 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 supplemented with 10 mM MgCl 2 and then washed with acid 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 another two times with ice-cold PBS containing 10 mM MgCl 2 and fixed with 4% paraformaldehyde (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 a 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.
Fluid-phase endocytosis of fluorescent dextran. HeLa cells were grown on Matrigel-coated glass coverslips to near confluency before treatment for 6 h with 0.1% DMSO or 20 μM PITCOIN1 or PITCOIN3. Cells were incubated with 0.5 mg ml −1 Alexa488-labeled dextran (10k) (Thermo, 22910) in the continued presence of DMSO or PITCOINs for 60 min at 37 °C in a humidity chamber. Coverslips were washed three times with PBS for 2 min each before fixation for 45 min in 4% PFA. Cell nuclei were stained with Hoechst33342 (Thermo). Internalized dextran was quantitatively analyzed using a Nikon Eclipse Ti microscope equipped with ImageJ software.

PI immunocytochemistry and imaging. PI(3)P detection in Cos7 cells.
Cos7 cells were grown on Matrigel-coated glass coverslips and treated with 0.1% DMSO or 20 μM PITCOIN1 or PITCOIN3 for 6 h. Cells were washed with PBS containing 10 mM MgCl 2 once and fixed in 2% PFA with 2% sucrose in PBS for 15 min at room temperature. Cells were washed three times with PBS containing 50 mM NH 4 Cl and permeabilized with PBS containing 20 μM digitonin (5% stock; Invitrogen) in buffer A (20 mM PIPES, pH 6.8, 137 mM NaCl, 2.7 mM KCl) for 5 min. Following three washes with buffer A, cells were stained for PI(3)P by incubation with the purified recombinant eGFP-2×FYVE domain of Hrs (0.5 μg ml −1 ) for 1 h at room temperature. Samples were stained with antibodies against GFP followed by labeling with Alexa488-conjugated secondary antibodies. Cells were imaged by laser scanning confocal microscopy, and PI(3)P levels were quantified using ImageJ software.

PI(3)P detection in platelets.
Washed resting mouse platelets were adhered on poly(l-lysine)-coated slices for 30 min at 37 °C, fixed in 2% formaldehyde for 30 min and permeabilized with PIPES buffer (20 mM PIPES pH 6.8, 137 mM NaCl, 2.7 mM KCl) containing 20 µM digitonin for 5 min at room temperature. Following washes in PIPES buffer, samples were saturated with 10% goat serum in PIPES buffer for 1 h at room temperature and incubated with the anti-PI3P antibody (Echelon, 1:100) in 10% goat serum in PIPES buffer overnight at 4 °C followed by Alexa Fluor 488 anti-mouse antibody (Life Technologies, 1:500) in 10% goat serum in PIPES buffer for 1 h at room temperature. After several washes and slice mounting, samples were examined using the Zeiss LSM780 confocal microscope with a ×63, 1.4-NA Plan-Apochromat lens. PI(3)P levels were quantified using ImageJ software. PI(3,4)P 2 detection. Cos7 cells stably expressing eGFP-clathrin light chain were grown on Matrigel-coated glass coverslips and treated with 0.1% DMSO or 20 µM PITCOIN1 or PITCOIN3 for 6 h. Cells were washed with PBS containing 10 mM MgCl 2 once and fixed in 2% PFA with 2% sucrose in PBS for 15 min at room temperature. Cells were washed three times with PBS containing 50 mM NH 4 Cl and permeabilized with PBS containing 0.5% saponin and 1% BSA for 30 min. Samples were incubated with primary (2 h) and secondary (1 h) antibodies diluted in PBS containing 1% BSA and 10% normal goat serum. Cells were analyzed by TIRF microscopy (Nikon TI Eclipse, 488-and 561-nm lasers, ×60 1.49-NA objective equipped with an sCMOS Andor mNeo camera). PI(3,4)P 2 levels at CCPs were quantified using ImageJ software with eGFP-clathrin as a mask.
PI(4)P and PI(4,5)P 2 detection. Cos7 cells were grown in matrix gel-coated glass coverslips and treated with 0.1% DMSO or 20 µM PITCOIN1 or PITCOIN3 for 6 h. Cos7 cells were washed with PBS containing 10 mM MgCl 2 once. For detection of plasma membrane PI(4)P or PI(4,5)P 2 , cells were fixed in 2% PFA, 1% glutaraldehyde and 2% sucrose in PBS for 20 min at room temperature and permeabilized with PBS containing 0.5% saponin and 1% BSA in PBS for 30 min. For detection of the intracellular Golgi/TGN pool of PI(4)P, cells were fixed in 2% PFA and 2% sucrose in PBS for 20 min and permeabilized with 20 µM digitonin in buffer A for 5 min.
Analysis of starvation-induced autophagosome formation. HEK293T cells were seeded at a density of 0.4 × 10 6 cells per well into six-well plates coated with poly(l-lysine). Cells were grown overnight and pretreated with DMSO (0.1%, 6 h), VPS34-IN1 (5 µM, 2 h) or PITCOIN1 or PITCOIN3 (20 µM, 5 h) in DMEM containing 10% FBS. Cells were washed twice with EBSS, followed by co-treatment with bafilomycin A1 (100 nM) in EBSS for 1 h to block acidification. Cells were lysed in 200 μl TBS supplemented with 1% NP-40, 0.1% SDS and a protease inhibitor cocktail (Roche). Total protein concentrations were measured using the Bradford assay. Cell lysates (50 μg total protein) were analyzed by 15% SDS-PAGE and immunoblotting using antibodies against LC3-I, LC3-II and β-actin. IRDye 800CW-conjugated secondary antibodies were used for detection with a LI-COR imager system. EM analysis of platelet ultrastructure. Transmission EM. Washed mouse resting platelets were fixed in 2.5% glutaraldehyde in 0.05 M sodium cacodylate and 30 mM glucose at 4 °C for 24 h. Samples were embedded in 2% agarose, cut in 1-mm pieces, washed in 0.2 M sodium cacodylate buffer, postfixed with 1% osmium tetroxide in 0.2 M sodium cacodylate buffer for 1 h at room temperature and dehydrated in a series of graded ethanol solutions (30%, 50% and 70% 10 min each at room temperature) before being embedded in Embed 812 resin (Electron Microscopy Sciences) using a Leica EM AMW automated microwave tissue processor for EM. Samples were then sliced into 70-nm thick sections (Ultracut Reichert Jung) and mounted on 100-mesh copper grids before staining with 3% uranyl acetate in 50% ethanol and Reynold's lead citrate. Examinations were carried out on a Hitachi HT7700 transmission electron microscope at an accelerating voltage of 80 kV. The OCS area and total platelet area were quantified using ImageJ software.
Scanning EM. Washed platelets were stimulated for 3 min with CRP and were fixed in 2.5% glutaraldehyde in 0.05 M sodium cacodylate and 30 mM glucose at 4 °C for 24 h. After resuspending in distilled water, platelets were allowed to adhere on poly(lysine)-coated coverglasses, rinsed and dehydrated in a graded ethanol series. Once dehydrated in 100% ethanol, samples were submitted to critical point drying with a Leica EM CPD 300. Finally, samples were coated with 6 nm platinium on a Leica EM Med 020 before being examined on an FEI Quanta 250 FEG scanning electron microscope at an accelerating voltage of 5 kV.
Platelet flow assay on collagen. Microcapillaries were coated with Horm collagen suspension (100 μg ml −1 ) for 1 h at room temperature and saturated with a solution of 1% BSA in PBS for 1 h at room temperature. Whole blood was drawn from the inferior vena cava of mice anesthetized with a mixture of Imalgene (25 mg kg −1 ) and Rompun (10 mg kg −1 ) at 1 IU ml −1 and labeled with DIOC 6 (2 μM) for 10 min at 37 °C. Using a syringe pump (PHD-2000, Harvard Apparatus), DIOC 6 -labeled whole blood was perfused through a collagen-coated microcapillary for 2 min at an arterial shear rate of 500 s −1 . Thrombus formation was visualized with a ×40 oil-immersion objective for both fluorescence and transmitted light microscopy using the Zeiss Axio Observer microscope and recorded in real time (one frame every 20 s on 50 sections of 2 µm) with an ORCA camera (Carl Zeiss). Surface covered by platelets and three-dimensional thrombus volumes were calculated by thresholding using, respectively, ImageJ and Imaris.
Platelet activation assays. For platelet aggregation and ATP release analysis, 300 µl of 2 × 10 8 cells per ml washed platelets were stimulated in HEPES-Tyrode's buffer (pH 7.38) with CRP (1 and 3 µg ml −1 ) or thrombin (0.1 and 0.3 IU ml −1 ) for 5 min at 37 °C under continuous stirring at 1,000 r.p.m. Aggregation measurements were performed using a Born lumi-aggregometer (Chrono-Log) alongside ATP release quantification by addition of 10 μl of Chronolume 1 min before stimulation. For P-selectin membrane exposure and αIIbβ3-integrin activation, 1 × 10 6 platelets were stimulated with CRP (1 and 3 µg ml −1 ) or thrombin (0.1 and 0.3 IU ml −1 ) for 5 min at 37 °C under nonstirring conditions in the presence of 1 mM CaCl 2 and PE-conjugated JON/A antibody. Platelets were further labeled with FITC-conjugated anti-mouse P-selectin for 30 min in the dark at room temperature before being analyzed with a BD LSRFortessa cytometer and BD FACSDiva software (BD Biosciences).
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 with Prism software (GraphPad), using a one-sample t-test with a theoretical mean of 1, two-tailed Student's t-test, one-way ANOVA with Tukey's multiple-comparisons test or two-way ANOVA followed by a Sidak's multiple-comparisons test. Specific P values are indicated in the legends to figures. Significant differences are marked as follows: *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. (e, f) PITCOIn1 and PITCOIn3 do not affect fluid-phase endocytosis of dextrans. (e) Representative confocal microscopy images of HeLa cells treated with DMSO, PITCOIn1, or PITCOIn3 (20 μM for 6 h) and incubated with alex488-conjugated dextran for 60 mins. Scale bar, 10 μm. (f) Quantification of representative data shown in (e). Data are from n = 3 independent experiments and depicted as mean ± SEM. n.S., non-significant, one-way anOVa with Tukey's multiple comparison test. Fig. 6 | PiTCOiN1 and PiTCOiN3 do not affect cellular Pi(4,5)P 2 and Pi(4)P levels. (a) Representative confocal microscopy images of fixed Cos7 cells treated with DMSO, PITCOIn1, or PITCOIn3 (20 μM for 6 h) and stained for PI(4,5)P 2 using specific antibodies. (b) Quantification of representative data shown in (a). Data are from n = 3 independent experiments and depicted as mean ± SEM. n.S., non-significant, one-way anOVa with Tukey's multiple comparison test. (c) Representative confocal microscopy images of fixed Cos7 cells treated with DMSO, PITCOIn1, or PITCOIn3 (20 μM for 6 h) and stained for PI(4)P on the cell surface using specific antibodies. (d) Quantification of representative data shown in (c). Data are from n = 3 independent experiments and depicted as mean ± SEM. n.S., non-significant, one-way anOVa with Tukey's multiple comparison test. (e) Representative confocal microscopy images of fixed Cos7 cells treated with DMSO, PITCOIn1, or PITCOIn3 (20 μM for 6 h) and stained for internal pools of PI(4)P using specific antibodies. Cells treated with 5 µM PI4KIIIβ-In-10 for 2 h were used as a positive control. (f) Quantification of representative data shown in (e). Data are from n = 3 independent experiments and depicted as mean ± SEM. n.S., non-significant, one-way anOVa with Tukey's multiple comparison test.