Class II phosphatidylinositol 3-kinase-C2α is essential for Notch signaling by regulating the endocytosis of γ-secretase in endothelial cells

The class II α-isoform of phosphatidylinositol 3-kinase (PI3K-C2α) plays a crucial role in angiogenesis at least in part through participating in endocytosis and, thereby, endosomal signaling of several cell surface receptors including VEGF receptor-2 and TGFβ receptor in vascular endothelial cells (ECs). The Notch signaling cascade regulates many cellular processes including cell proliferation, cell fate specification and differentiation. In the present study, we explored a role of PI3K-C2α in Delta-like 4 (Dll4)-induced Notch signaling in ECs. We found that knockdown of PI3K-C2α inhibited Dll4-induced generation of the signaling molecule Notch intracellular domain 1 (NICD1) and the expression of Notch1 target genes including HEY1, HEY2 and NOTCH3 in ECs but not in vascular smooth muscle cells. PI3K-C2α knockdown did not inhibit Dll4-induced endocytosis of cell surface Notch1. In contrast, PI3K-C2α knockdown as well as clathrin heavy chain knockdown impaired endocytosis of Notch1-cleaving protease, γ-secretase complex, with the accumulation of Notch1 at the perinuclear endolysosomes. Pharmacological blockage of γ-secretase also induced the intracellular accumulation of Notch1. Taken together, we conclude that PI3K-C2α is required for the clathrin-mediated endocytosis of γ-secretase complex, which allows for the cleavage of endocytosed Notch1 by γ-secretase complex at the endolysosomes to generate NICD1 in ECs.

reduced the expression of PI3K-C2α protein by approximately 90% compared with control (ctrl)-siRNA (Fig. 1a), as reported previously [5][6][7] . Dll4 stimulation induced more than a sixfold increase in NICD1 level in ctrl-siRNAtransfected cells. Knockdown of PI3K-C2α reduced Dll4-induced increase in NICD1 by approximately 40% compared with ctrl-siRNA-transfected cells. A different PI3K-C2α-specific siRNA also decreased Dll4-induced NICD1 production ( Supplementary Fig. S1a). Compared with Dll4, another Notch ligand Jag1 induced NICD1 production slightly, which is also inhibited by PI3K-C2α knockdown ( Supplementary Fig. S1b). Quantitative PCR (qPCR) analysis showed that Dll4 increased the mRNA expression of the Notch target genes HEY1, HEY2, HES1 and FLT1 in control cells and that among them, PI3K-C2α knockdown inhibited Dll4-induced upregulation of HEY1 and HEY2 (Fig. 1b). Notch signaling regulates dynamic positive feedback loop of the expression of Notch itself 41 . HUVECs mainly expressed NOTCH1, NOCTH2 and NOCTH4 but rarely expressed NOCTH3 ( Fig. 1c and Supplementary Fig. S2a). Dll4 increased the mRNA expression of NOTCH1 and NOTCH3 in control cells, and PI3K-C2α knockdown inhibited Dll4-induced upregulation of NOTCH3 but not NOTCH1 (Fig. 1d). Because HUVECs expressed multiple Notch subtypes, we examined the involvement of Notch1 in Dll4-induced upregulation of the genes. Notch1 knockdown inhibited Dll4-induced NICD1 production (Fig. 2a) and upregulation of HEY1, HEY2 and NOTCH3 (Fig. 2b). In contrast, the forced expression of Flag-tagged NICD1 increased mRNA expression of HEY1, HEY2 and NOTCH3 compared with either non-transfected or GFP-transfected control cells (Fig. 2c and d). We also studied whether Dll4-induced Notch signaling was dependent on PI3K-C2α in another type of ECs, human microvascular ECs from lung (HMVECs-L). The expression profile of Notch paralogues in HMVECs-L was almost identical to that in HUVECs ( Fig. 3a and Supplementary Fig. S2b). Knockdown of PI3K-C2α in HMVECs-L inhibited Dll4-induced NICD1 production (Fig. 3b) and the expression of its target genes HEY1 and HEY2 (Fig. 3c), similar to HUVECs. These results suggest that PI3K-C2α is required for Dll4induced NICD1 production and its target gene expression via Notch1 in ECs.
We previously reported that endothelial-specific deletion of PI3K-C2α in mice resulted in embryonic lethality due to severe defects in angiogenesis whereas vascular smooth muscle-specific PI3K-C2α deletion did not impair normal growth or development 5 . Therefore, we studied the dependence of Notch signaling on PI3K-C2α in vascular SMCs. Human aortic smooth muscle cells (HAoSMCs) expressed NOCTH1, NOCTH2 and NOCTH3 but rarely expressed NOCTH4 differently from ECs ( Fig. 3d and Supplementary Fig. S2c). PI3K-C2α knockdown did not inhibit Dll4-induced NICD1 generation (Fig. 3e) or the expression of Notch target genes HEY1 and HES1 showing the effects of PI3K-C2α knockdown on Dll4-induced NICD1 production. PI3K-C2α-siRNA-or control (ctrl)-siRNA-transfected HUVECs were incubated on Dll4-or BSA-coated plates and 24 h later underwent Western blot analysis using anti-NICD1 and anti-PI3K-C2α antibodies. Right two panels show the quantified data of PI3K-C2α and NICD1 protein levels (n = 3). Statistical significance was assessed with two-tailed unpaired Student's t test for PI3K-C2α and two-way ANOVA followed by Bonfferroni's post-hoc test for NICD1. (b) The effects of PI3K-C2α knockdown on Dll4-induced Notch target gene expression. PI3K-C2α-siRNA-or ctrl-siRNA-transfected HUVECs were incubated on Dll4-or BSA-coated plates and 24 h later subjected to qPCR analysis (n = 5). Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test. (c) Relative mRNA expression levels of Notch paralogues compared with NOTCH1. The values on the top of bars represent the means ± SEM. HUVECs were cultured on BSA-coated plates for 24 h and underwent qPCR analysis (n = 3). (d) The effects of PI3K-C2α knockdown on mRNA expression of Notch paralogues. PI3K-C2α-siRNA-or ctrl-siRNA-transfected HUVECs were incubated on Dll4-or BSA-coated plates and 24 h later subjected to qPCR analysis (n = 3). In a-d, data are presented as means ± SEM from three to five independent experiments. Statistical significance was presented as *P < 0.05, **P < 0.01 or ***P < 0.001, respectively. ns: not significant. www.nature.com/scientificreports/ The effects of PI3K-C2α knockdown on Dll4-induced NICD1 production in HMVECs-L (b) and HAoSMCs (e). PI3K-C2α-siRNA-or ctrl-siRNA-transfected cells were incubated on Dll4-or BSAcoated plates for 24 h. Then cell lysate was subjected to Western blot analysis for NICD1, PI3K-C2α, and β-actin. The right panels show the quantified data of NICD1 protein level (n = 3). Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test.

Scientific Reports
(c, f) The effects of PI3K-C2α knockdown on Dll4-induced Notch target gene expression in HMVECs-L (c) and HAoSMCs (f). PI3K-C2α-siRNA-or ctrl-siRNA-transfected cells were incubated on Dll4-or BSA-coated plates, and 24 h later harvested for qPCR analysis (n = 3-4). Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test. Statistical significance was presented as *P < 0.05, **P < 0.01 or ***P < 0.001, respectively. ns: not significant. PI3K-C2α is not required for Dll4-induced Notch1 internalization but for cleavage of internalized Notch1. Previous studies [32][33][34][35][36][37] revealed that ligand binding of Notch1 results in the generation of NICD1 through the formation of NEXT by ADAM-mediated proteolysis of Notch1 and subsequent γ-secretase complex-mediated cleavage of NEXT at the plasma membrane and/or the endolysosomal membrane. The γ-secretase complex-mediated NICD1 generation in the endolysosomes requires endocytosis of both Notch1 and γ-secretase complex. Because we and others showed that PI3K-C2α is involved in clathrin-mediated endocytosis [5][6][7][8][9]42 , we studied how PI3K-C2α is involved in NICD1 production. Anti-Notch1 immunostaining using an antibody that recognizes the intracellular domain of Notch1 showed that the Notch1 immunoreactivity was detected mainly at the plasma membrane (green arrowheads in Fig. 4a and Supplementary Fig. S3) and the perinuclear regions in ctrl-siRNA-transfected, non-stimulated cells. Dll4 stimulation of ctrl-siRNA-transfected cells induced disappearance of Notch1 at the plasma membrane and robust accumulation of Notch1 immunoreactivity in the nuclei, which likely reflected nuclear translocation of NICD1. PI3K-C2α knockdown increased Notch1-positive perinuclear dots in non-stimulated cells. In PI3K-C2α-depleted cells, Dll4 induced disappearance of plasma membrane Notch1 as in ctrl-siRNA-transfected cells, but a marked reduction in the nuclear accumulation of Notch1 immunoreactivity with enhanced Notch1-positive perinuclear dots (red arrowheads in Fig. 4a) compared with ctrl-siRNA-transfected cells. Double immunofluorescent staining using anti-Notch1 and anti-organelle-specific markers showed that Notch1-positive perinuclear dots were also positive for the trans-Golgi network (TGN) marker TGN, the early endosome marker EEA1, the lysosomal markers LAMP1 and DQ-BSA in ctrl-siRNA-transfected cells (Fig. 4b). PI3K-C2α knockdown resulted in substantial enhancement of Notch1 immunoreactivity in the endolysosomes (white arrowheads in Fig. 4b).

Scientific Reports
In cell surface-biotinylated cells, Dll4 stimulation reduced biotinylated cell surface Notch1 (TM-IC) to the similar extents in control and PI3K-C2α-depleted cells (Fig. 5a), suggesting that Dll4-induced endocytosis of Notch1 was not dependent on PI3K-C2α. It was also noted that the NEXT was more abundant in the total cell lysate of PI3K-C2α-depleted cells compared with control cells. The biotinylated surface protein fractions from either control or PI3K-C2α-depleted cells did not contain a detectable level of NEXT, which may suggest that the produced NEXT was internalized in a manner independent of PI3K-C2α. The accumulation of Notch1 immunoreactivity at the endolysosomes was mimicked by pharmacological γ-secretase inhibition using DAPT (Fig. 5b). These findings, together with PI3K-C2α dependence of Dll4-induced NICD1 production (Fig. 1a), strongly suggest that γ-secretase-mediated cleavage of NEXT but not endocytosis of NEXT is dependent on PI3K-C2α. PI3K-C2α knockdown impedes γ-secretase-mediated cleavage of internalized NEXT, resulting in the accumulation of non-cleaved NEXT in the degradation pathway endolysosomes.
Because PI3K-C2α is essential for clathrin-mediated endocytosis, we studied the effect of clathrin heavy chain (CHC) knockdown on Dll4-induced Notch signaling in HUVECs. Knockdown of CHC inhibited Dll4-induced NICD1 production (Fig. 5c). Consistently, CHC knockdown decreased Dll4-induced expression of Notch1 target genes, HEY1 and NOTCH3 (Fig. 5d). In cell surface-biotinylation experiments, Dll4 induced similar reductions of biotinylated cell surface Notch1 in control and CHC-depleted cells (Fig. 5e), suggesting that Dll4-induced endocytosis of Notch1 is not dependent on clathrin. In contrast, NEXT was more abundant in the total cell lysate of CHC-depleted cells (Fig. 5e), which was likely due to impaired cleavage of NEXT. Taken together, these results suggest that both PI3K-C2α and clathrin are required for Dll4-induced cleavage of internalized Notch1 but not Notch1 internalization process itself. PI3K-C2α is required for clathrin-dependent endocytosis of γ-secretase complex which cleaves internalized Notch1. We tested the hypothesis that PI3K-C2α might be necessary for endocytosis of γ-secretase complex that cleaves internalized Notch1 at the endolysosomal compartment. We explored whether PI3K-C2α was required for endocytosis of nicastrin (Nct), a core component of γ-secretase complex 43 . Knock- Figure 4. PI3K-C2α knockdown causes the perinuclear accumulation of Notch1 at the endolysosomal compartment. (a) Immunofluorescent staining of HUVECs with anti-Notch1 antibody (black) and nuclear DAPI staining (blue). PI3K-C2α-siRNA-or ctrl-siRNA-transfected HUVECs were incubated on Dll4-or BSA-coated dishes for 24 h. Cells were fixed and stained with anti-Notch1 antibody. Left panels, representative confocal images of anti-Notch1-stained cells and the merged views with DAPI. Right panel, quantified data of relative abundance of perinuclear Notch1-positive vesicles (n = 3). The area of perinuclear Notch1-positive vesicles was quantified with Image-J software. Statistical significance was assessed with two-tailed unpaired Student's t test and presented as **P < 0.01. Notch1 immunoreactivity was detected at the plasma membrane (green arrowheads) on BSA-coated dishes and translocated to the nuclei (asterisks) upon Dll4 stimulation. Note the perinuclear Notch1-positive vesicles (red arrowheads) in PI3K-C2α-depleted cells on BSA-and Dll4-coated dishes. Scale bar, 20 μm. (b) Double immunofluorescent staining of Notch1 and the organelle-specific markers. PI3K-C2α-siRNA-or ctrl-siRNA-transfected HUVECs were incubated on Dll4-coated dishes for 24 h. Cells were fixed and stained with anti-Notch1 and either anti-p230 TGN (trans-Golgi network, top-left), anti-EEA1 (early endosome, top-right) or anti-LAMP1 (lysosome, bottom-left) antibodies. In a portion of dishes, cells were loaded with DQ Red BSA (bottom-right), which stains lysosomes, for 2 h before fixation. Nuclei were stained with DAPI. Representative confocal images illustrating the perinuclear Notch1-positive vesicles partially co-localized with EEA1-and LAMP1/DQ-BSA-positive endolysosomal compartments (white arrowheads). Asterisks denote the nuclei. Scale bar, 20 μm.  www.nature.com/scientificreports/ down of PI3K-C2α increased the cell surface glycosylated mature form of Nct to the similar extents in nonstimulated and Dll4-stimulated cells (Fig. 6a). Likewise, knockdown of CHC increased the amount of the cell surface Nct in non-stimulated and Dll4-stimulated cells (Fig. 6b). These findings suggested that Nct-containing γ-secretase complex is endocytosed in both non-stimulated and Dll4-stimulated conditions in PI3K-C2α-and clathrin-dependent manners. We studied the effects of CHC knockdown on nuclear and perinuclear accumulation of Notch1 immunoreactivity. Perinuclear accumulation of Notch1 immunoreactivity was observed only in CHC-depleted cells but not in non-depleted cells in the same microscopic field (Fig. 6c). We next examined the effects of PI3K-C2α knockdown on endocytosis of GFP-tagged Nct (Nct-GFP) and colocalization of Nct-GFP and Notch1 immunoreactivity and on the nuclear accumulation of Notch1 immunoreactivity ( Fig. 6d and e). We took advantage of exogenous Nct-GFP expression because commercially available anti-Nct antibodies were not properly validated for immunostaining of Nct. With high resolution confocal microscope, in non-stimulated, control siRNA-transfected cells on the BSA-coated dish, expressed Nct-GFP was observed in a punctate pattern diffusely in the cytoplasm and in a mesh-like pattern at the perinuclear region, and was hardly colocalized with Notch1 immunoreactivity, which was distributed in a punctate pattern in the cytoplasm and moderately in the nucleus (Fig. 6e). In contrast, in Dll4-stimulated, ctrl-siRNA-transfected cells, the colocalization of Nct-GFP and Notch1 immunoreactivity was observed at the perinuclear region, most of which were LAMP1-positive lysosomes (white arrowheads in Fig. 6e), at 2 h and 4 h after seeding. Dll4 stimulation also increased nuclear Notch1 immunoreactivity, which most likely represented NICD1. PI3K-C2α knockdown inhibited the perinuclear colocalization of Nct-GFP and Notch1 immunoreactivity and the nuclear accumulation of Notch1 immunoreactivity (SRRF views of Fig. 6e and f). In addition, PI3K-C2α knockdown resulted in the accumulation of Nct-GFP at the cell surface (yellow arrowheads at 4 h on the Dll4-coated dish in Fig. 6e) probably because of inhibited endocytosis of Nct-GFP.

PI3K-C2α mediates Dll4-induced expression of mesenchymal cell marker in HUVECs. Dll4
stimulation increased the expression of Notch3 in HUVECs, which was inhibited by PI3K-C2α depletion (Fig. 1d). Notch3 is predominantly expressed in SMCs but is rarely expressed in ECs in a non-stimulated condition 44 . It is reported that upregulation of Notch3 in ECs results in endothelial to mesenchymeal transition (EndMT) 45 . We investigated the expression of EndMT-related genes in HUVECs. Dll4 stimulation decreased the expression of the endothelial cell marker, CDH5 (VE-cadherin) and increased the expression of the mesenchymal cell marker COL1A in HUVECs. PI3K-C2α knockdown reversed these Dll4-induced changes, although the difference of only COL1A expression reached the statistical significance (Fig. 6g). It is reported that transcriptional repressor SNAI2 (also known as SLUG) mediates Notch-induced EndMT 46 . Dll4 stimulation increased SNAI2 expression in HUVECs, which was significantly impaired by PI3K-C2α depletion (Fig. 6g). These observations raised the possibility that PI3K-C2α is functionally involved in EndMT induced by Dll4-Notch signaling.

Discussion
In this study, we found that PI3K-C2α was required for Dll4-induced NICD1 production and the expression of Notch-target genes in HUVECs. Ligand-induced NICD1 generation in HUVECs involved endocytosis of ligandactivated Notch1 and γ-secretase complex. PI3K-C2α was necessary for the endocytosis of γ-secretase complex but not Notch1. The internalized γ-secretase complex mediated proteolytic cleavage of endocytosed Notch1 to generate NICD1. Thus, our study suggests that PI3K-C2α is required for intracellular activation of Notch1 by mediating endocytosis of γ-secretase complex in ECs (Fig. 7). PI3K-C2α knockdown decreased Dll4-induced production of NICD1 and the expression of Notch1 target genes in human vascular EC types, HUVECs and HMVECs-L (Figs. 1a, b, 3b and c). The present study, together (a) Representative Western blot images showing the effects of PI3K-C2α knockdown on the abundance of cell surface Notch1 and NEXT in HUVECs. PI3K-C2α-siRNA-or ctrl-siRNA-transfected cells were incubated on Dll4-or BSA-coated plates, and 24 h later total cell lysate and biotin-labeled surface proteins were prepared, followed by Western blot analysis for Notch1 and β-actin. Right panels show the quantified data of cell surface TM-IC level (n = 4) and the abundance of NEXT in total cell lysate as NEXT over NEXT + TM-IC (n = 5), respectively. Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test. (b) Representative confocal images of double immunofluorescent staining with anti-Notch1 and either anti-LAMP1 (left) or anti-EEA1 (right) antibodies. Cells received γ-secretase inhibitor DAPT (1 μM) before seeding and were incubated on Dll4-coated plates for 12 h before the fixation. Nuclei were stained with DAPI (asterisks). DAPT treatment almost completely abolished Dll4-induced nuclear translocation of Notch1. Note the endolysosomal accumulation of Notch1 (arrowheads) was observed in DAPT-treated cells. Scale bar, 20 μm. (c, d) The effects of CHC knockdown on Dll4-induced NICD1 production (c) and Notch target gene expression (d). CHC-siRNA-or ctrl-siRNA-transfected HUVECs were incubated on Dll4-or BSA-coated plates and 24 h later subjected to Western blot analysis (n = 5) and qPCR analysis (n = 3-4), respectively. Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test. (e) The effects of CHC knockdown on cell surface Notch1 level and NEXT abundance in HUVECs. CHC-siRNA-or ctrl-siRNA-transfected HUVECs were incubated on Dll4-or BSA-coated plates for 24 h. Total cell lysate and biotin-labelled surface proteins were prepared and analyzed as in (a). Right panels show the quantified data of cell surface TM-IC level and the abundance of NEXT in total cell lysate (n = 3). Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test. Statistical significance was presented as *P < 0.05, **P < 0.01 or ***P < 0.001, respectively. ns: not significant.  www.nature.com/scientificreports/ with our previous reports [5][6][7] , showed that PI3K-C2α is required for endocytosis-mediated signaling of several different classes of cell surface receptors. In our previous studies, we showed that PI3K-C2α is required for endosomal receptor signaling, which includes Rac activation by the G-protein-coupled receptor S1P 1 , Rho activation by the receptor tyrosine kinase VEGFR2 and Smad phosphorylation by the TGFβ1 receptor serine/threonine kinase ALK5, through being involved in endocytosis of these cell surface receptors. Differently from the cases of these cell surface receptors, PI3K-C2α in Notch signaling is required for endocytosis of the Notch-cleaving protease γ-secretase complex but not Notch receptor itself. By this action, PI3K-C2α is necessary for the generation of NICD1, an essential mediator of Notch signaling in ECs. In the present study, PI3K-C2α knockdown or pharmacological γ-secretase inhibition resulted in an increase of Notch1 immunoreactivity at the perinuclear endolysosomes, which implied the accumulation of internalized Notch1 that escaped proteolysis by γ-secretase complex (Figs. 4 and 5b). After ligand binding, Notch receptors are finally cleaved by γ-secretase complex at the cell surface 40 or at the intracellular vesicles to release NICD1 32,33 . γ-secretase complex is a member of the intramembrane cleaving proteases, which proteolyse type I transmembrane proteins at the hydrophobic compartment of the transmembrane domain. γ-secretase is hetero-tetrameric complex composed of presenilin (PS), Nct, anterior pharynx defective 1 (APH-1) and presenilin enhancer 2 (PEN-2). PS possesses a catalytic domain and has two paralogues, PS1 and PS2. There are also two APH-1 genes, APH1A and APH1B, in humans. Thus, at least four different γ-secretase complexes exist in humans 47,48 . All four components are required for γ-secretase activity and active γ-secretase complexes contain each component in a 1:1:1:1 stoichiometry [49][50][51][52] . Nct functions as a substrate receptor in the γ-secretase complex 53 . Because there is no isoform of Nct, it is present in all γ-secretase complexes. In the present study, knockdown of either PI3K-C2α or CHC inhibited internalization of Nct with or without Dll4 stimulation ( Fig. 6a, b, e and f), suggesting that PI3K-C2α is required for endocytosis of γ-secretase complex, which occurs in a manner independent of Dll4 stimulation. In contrast, knockdown of PI3K-C2α or CHC did not alter endocytosis of the cell surface Notch1 but rather increased the cellular protein level of ADAM-truncated Notch1, NEXT (Fig. 5a and e). Moreover, knockdown of PI3K-C2α or CHC induced the accumulation of Notch1 immunoreactivity at the endolysosomes, as well as the γ-secretase inhibitor DAPT (Figs. 4, 5b and 6c). Taken together, PI3K-C2α is required for clathrin-dependent endocytosis of γ-secretase complex but not Notch1, and the internalized γ-secretase complex mediates the final step of Notch1 cleavage to generate NICD1 likely at the endolysosomal compartments. When endocytosis of γ-secretase complex is impaired by knockdown of PI3K-C2α and CHC or γ-secretase activity is inhibited pharmacologically, γ-secretase-mediated cleavage of internalized Notch1, most likely NEXT, is hampered with the accumulation of Notch1 in the endolysosomes. In the present study, we could not determine the molecular entitiy of Notch1 immunoreactivity at the endolysosomes because cleavage-specific anti-Notch1 antibodies were not available. Further studies are required to exactly identify the molecular entity of Notch1 at the endolysosomes.
After translation, γ-secretase components undergo several maturation steps such as complex assembly, glycosylation of Nct and endoproteolysis of PS. Then, mature γ-secretase complexes are delivered to the cell surface or endosomes 47 . Indeed, growing evidence suggests that fully assembled γ-secretase complexes are localized at the plasma membrane and lysosomes [54][55][56] . Previous study reported that the mature form of γ-secretase is constitutively endocytosed from cell surface via clathrin-mediated pathway in a manner dependent on Clathrin Assembly Lymphoid Myeloid leukemia (CALM) 38 . PI3K-C2α is recruited to clathrin-coated pits through its N-terminal clathrin-binding domain to facilitate growth of clathrin-coated pits 9 . It is known that the proteolytic activity of γ-secretase is higher at low pH conditions 55 and impairment of lysosomal acidification disrupts the γ-secretase-mediated cleavage of Notch at endolysosomal compartments 57,58 . Therefore, PI3K-C2α-mediated delivery of γ-secretase complex to the endolysosomes provides a suitable condition of efficient Notch1 proteolysis by γ-secretase complex for generating NICD1. Because PI3K-C2α knockdown does not impair endocytosis of PI3K-C2α-siRNA-or ctrl-siRNA-transfected cells were incubated on Dll4-or BSA-coated plates and 24 h later harvested for qPCR analysis (n = 5). Statistical significance was assessed with two-way ANOVA followed by Bonfferroni's post-hoc test in (a, b, f, g). Statistical significance was presented as *P < 0.05, **P < 0.01 or ***P < 0.001, respectively. ns: not significant. www.nature.com/scientificreports/ Notch1, the suppression of γ-secretase endocytosis results in the retention of Notch1 immunoreactivity at the endolysosomes, as demonstrated in Fig. 4. Although PI3K-C2α depletion caused the accumulation of Notch1 at endolysosomal compartments, the condition of PI3K-C2α deficiency did not result in the activation of Notch signaling probably because γ-secretase was absent at that compartment. Some previous papers reported Notch1 is internalized through clathrin-dependent manner 32,36 . In contrast, other reports suggest that endocytosis in not required for Notch signaling 39,40 . It is also reported that there are several endocytic routes in Notch1 internalization 33 . These reports raise the possibility that Notch1 internalization pathway is cell-type-or context-dependent. It is also possible that several Notch1 internalization pathways coexist and orchestrate the signal intensity. PI3K-C2α knockdown inhibits γ-secretase-mediated Notch1 cleavage at the endolysosomes, but may not impair Notch1 cleavage at the cell surface. Therefore, PI3K-C2α knockdown-induced reductions of NICD1 production and target gene expression may be partial ( Fig. 1a and b). Further studies are required to fully understand the mechanism of PI3K-C2α-dependent NICD1 generation.
The involvement of the PI3K-C2α in Notch signaling is cell type-specific: Dll4-induced NICD1 production was not dependent on PI3K-C2α in vascular SMCs, differently from ECs (Fig. 3). Although the mechanisms underlying the cell type-specific differences are not fully understood, it might be due to the possible cell typespecific compensation of PI3K-C2α deficiency by another class II PI3K member PI3K-C2β. Alternatively, it might be due to the difference of the cell site, i.e. the endolysosomes or the plasma membrane, where γ-secretase complex proteolyses Notch1 to generate NICD1, in vascular SMCs and ECs. If the plasma membrane could be the major site of γ-secretase complex-mediated cleavage of Notch1 to generate NICD1, Notch1 cleavage would not require endocytosis of γ-secretase complex and, therefore, PI3K-C2α.
Notch ligands Dlls and Jagged are transmembrane molecules, as well as Notch. In the in vivo conditions, membrane trafficking may regulate not only Notch receptors in target cells but also Notch ligands in the neighboring cells. In the present study, we immobilized Dll4 onto plates or dishes as a means to stimulate Notch on seeded HUVECs. This allowed us to focus on membrane trafficking of Notch receptors in Dll4 signal-receiving cells. It would be intriguing to study a possible involvement of PI3K-C2α in the regulation of the endocytosis of Notch ligands. www.nature.com/scientificreports/ PI3K-C2α deficiency causes the defects in sprouting angiogenesis and vascular maturation resulting in the embryonic lethality 5 . We have previously reported that loss of PI3K-C2α disrupts several signaling pathways including VEGF, S1P and TGFβ, which are the possible causes of PI3K-C2α deletion phenotypes [5][6][7] . Notch1-null mouse embryo is also reported to show the vascular maturation defect 20 . This phenotype is similar to PI3K-C2αnull embryo, raising the possibility that PI3K-C2α-mediated Notch signaling is involved in the developmental vessel maturation. Finally, we investigated the possible involvement of PI3K-C2α in Dll4-mediated EndMT because Notch3 expression in ECs, which is caused by Dll4 stimulation (Fig. 1d), is reported to be involved in EndMT progression 45 . We revealed that PI3K-C2α was required for Dll4-induced expression of EndMT regulator SNAI2 as well as the downregulation of endothelial cell marker CDH5 and the upregulation of mesenchymal cell marker COL1A induced by Dll4 stimulation (Fig. 6g). It is reported that Dll4-Notch mediated EndMT is required for cardiac cushion development 46 . Further in vivo studies are needed to investigate the function of PI3K-C2α as a regulator of Notch signaling.
In summary, the present study indicates that PI3K-C2α is essential for clathrin-mediated endocytosis of γ-secretase complex and Dll4-induced cleavage of Notch1 at the endolysosomal compartment, which is followed by NICD1 production and target gene expression in ECs. Thus, PI3K-C2α is a novel mediator of Notch signaling pathway by functioning as an endocytosis regulator of γ-secretase complex. Further studies to elucidate the significance of PI3K-C2α in Notch signaling pathway provide new insights into previously unexplained diversification of organ development.
Western blotting. After the incubation on the Dll4-or BSA-coated plates, cells were washed with PBS and lysed with 2 X Laemmli's solubilizing buffer [100 mM Tris (pH 6.8), 2% SDS, 0.008% Bromophenol Blue, 2% 2-mercaptoethanol, 26.3% glycerol and 0.001% Pyronin Y]. Cell lysate was subjected to 8% SDS-PAGE and separated proteins were transferred onto PVDF membranes (Millipore). The membranes were blocked with PVDF Blocking reagent (Toyobo, Osaka, Japan) and then some membranes were devided into two or three pieces. Each part of the cut membrane pieces or uncut whole membrane was incubated with respective primary antibodies at 4℃ overnight. Primary antibodies used were as follows: www.nature.com/scientificreports/ β-actin as a loading control and expressed as multiples over the normalized values of untreated controls. See Supplementary Fig. S4, which shows the original membranes before the cropping and the cropped regions.

RNA isolation and quantitative PCR (qPCR) analysis.
Total RNA was isolated from cultured cells using TRIzol reagent (Invitrogen). 0.5-1.0 μg of total RNA was reverse-transcribed into cDNA using Quanti-Tect RT Kit (#205311, Qiagen). Quantitative real time PCR analyses were performed using FastStart Universal SYBR Green Master (#04913914001, Roche, Basel, Switzerland) and GeneAmp 7300 system (Applied Biosystems, Foster City, CA, USA) as previously described 59 . Threshold value was determined automatically within the exponential growth region and threshold cycle (Ct) value was defined as cycle number at which fluorescence passed the threshold. ΔCt value was determined by subtracting the Ct value of ACTB (β-actin gene) from that of the target gene. ΔΔCt value was calculated by subtracting the ΔCt value of ctrl-siRNA-transfected, non-stimulated control cells from the respective ΔCt value of each experimental condition. 2 -ΔΔCt was presented as a fold increase against control cells in the figures. Semi-quantification of NOTCH1-4 expression levels was performed by subtracting the ΔCt value of NOTCH1 from that of NOTCH1-4 to calculate the ΔΔCt value and (2 -ΔΔCt × 100) (%) was presented as the relative abundance of NOTCH1-4 mRNA expression against NOTCH1 in Figs. 1c, 3a and d and Supplementary Fig. S2. Determination of PCR amplification efficiency of NOTCH1-4 cDNAs was performed using serially diluted cDNA samples from Dll4-stimulated HUVECs. Regression lines and amplification efficiency were determined based on the Ct values of each diluted sample using Prism 6 software (v6.0 h, GraphPad Software, Inc., https ://www.graph pad.com/scien tific -softw are/prism /). Amplification efficiency (%) was determined by the formula of ((10 -1/slope -1) × 100) using the slope of a regression line. Because amplification efficiency was very similar between NOTCH1-4 ( Supplementary Fig. S2d), we did not incorporate it into the assessment of the relative mRNA abundance. Primers used in the experiments are listed in Supplementary  Table S1.
Immunofluorescent staining and lysosome labeling with DQ-BSA. To

Biotinylation of cell surface protein.
Cells on the Dll4-or BSA-coated dishes were washed with ice-cold PBS twice and then treated with 1.0 mg/mL of EZ-link sulfo-NHS-SS-biotin (#21331, Thermo Scientific) for 15 min on ice. Biotinylation reactions were terminated with 50 mM glycine in PBS. After washing with PBS, cell extracts were prepared in radioimmunoprecipitation assay (RIPA) buffer [10 mM Tris-HCl (pH7.4), 100 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1 mM NaF, 20 mM Na 4 P 2 O 4 , 2 mM Na 3 VO 4 , 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton-X 100, 10% glycerol] with the protease inhibitor cocktail cOmplete, Mini (#11836153001, Roche). Cell extracts were sonicated and centrifuged at 8,000 × g at 4℃ for 30 min. Aliquots were made from supernatant and used as "total cell lysate" in the immunoblotting analysis. The rest of the supernatant was mixed with PierceTM NutrAvidinTM Agarose (#29,200, Thermo Scientific) and rotated for 2-3 h at 4 ℃. The beads were then washed with RIPA buffer three times and biotinylated proteins were eluted with 2 × Laemmli's buffer at 65℃ for 10 min. Total cell lysate and biotinylated surface proteins were subjected to immunoblot analysis.

Statistical analysis.
Statistical analysis was performed with Prism 6 software. Data are presented as mean ± standard error of mean (SEM). Paired data were evaluated with two-tailed unpaired Student's t test and comparison of multiple groups was performed using two-way ANOVA followed by Bonfferroni's post-hoc test, unless stated otherwise. P < 0.05 was considered to be statistically significant. Statistical significance was presented as * (P < 0.05), ** (P < 0.01) or *** (P < 0.001) respectively.

Data availability
All data described in the manuscript are contained within the manuscript and available from the corresponding author on reasonable request.