Research ArticleCell Biology

Phosphoproteomic Analysis Implicates the mTORC2-FoxO1 Axis in VEGF Signaling and Feedback Activation of Receptor Tyrosine Kinases

See allHide authors and affiliations

Science Signaling  16 Apr 2013:
Vol. 6, Issue 271, pp. ra25
DOI: 10.1126/scisignal.2003572

Abstract

The vascular endothelial growth factor (VEGF) signaling pathway plays a pivotal role in normal development and also represents a major therapeutic target for tumors and intraocular neovascular disorders. The VEGF receptor tyrosine kinases promote angiogenesis by phosphorylating downstream proteins in endothelial cells. We applied a large-scale proteomic approach to define the VEGF-regulated phosphoproteome and its temporal dynamics in human umbilical vein endothelial cells and then used siRNA (small interfering RNA) screens to investigate the function of a subset of these phosphorylated proteins in VEGF responses. The PI3K (phosphatidylinositol 3-kinase)–mTORC2 (mammalian target of rapamycin complex 2) axis emerged as central in activating VEGF-regulated phosphorylation and increasing endothelial cell viability by suppressing the activity of the transcription factor FoxO1 (forkhead box protein O1), an effect that limited cellular apoptosis and feedback activation of receptor tyrosine kinases. This FoxO1-mediated feedback loop not only reduced the effectiveness of mTOR inhibitors at decreasing protein phosphorylation and cell survival but also rendered cells more susceptible to PI3K inhibition. Collectively, our study provides a global and dynamic view of VEGF-regulated phosphorylation events and implicates the mTORC2-FoxO1 axis in VEGF receptor signaling and reprogramming of receptor tyrosine kinases in human endothelial cells.

Introduction

Vascular endothelial growth factor–A (VEGF-A; VEGF hereafter) is a major regulator of vasculogenesis and angiogenesis and is also a key mediator of pathological angiogenesis (1, 2). VEGF can exist as one of multiple splice isoforms that differ in their affinity for heparin. VEGF165 is the most abundant and physiologically relevant isoform (3). VEGF binds to two highly homologous endothelial cell–specific receptor tyrosine kinases: VEGF receptor 1 (VEGFR1) and VEGFR2 (4, 5). In addition, multiple VEGF isoforms can bind neuropilin-1, resulting in enhanced VEGFR2 signaling (6, 7). The biological effects of VEGF in developmental and pathological angiogenesis are largely mediated by VEGFR2, and the role of VEGFR1 is less clear (4, 8, 9). VEGF triggers activation of VEGFR2 and subsequent phosphorylation of various intracellular effectors, which leads to endothelial cell proliferation and migration during blood vessel growth (9, 10). Investigation of individual signaling molecules in specific pathways has expanded our knowledge of the VEGF-regulated signaling cascades. However, so far, most efforts have focused on one signal node at a time in a static form, making it difficult to obtain a global kinetic profile of the VEGF receptor signaling network.

On the other hand, current views of signal transduction are becoming increasingly multidimensional, such that cells are now seen as complex networks of interacting and intersecting pathways (1113). Downstream of VEGFR2, three core signaling pathways are the PI3K (phosphatidylinositol 3-kinase)–Akt pathway, which promotes cell survival and vascular permeability, the Raf–MEK (mitogen-activated or extracellular signal–regulated protein kinase kinase)–MAPK (mitogen-activated protein kinase) pathway, which activates cell proliferation, and the Src–FAK (focal adhesion kinase) pathway, which increases cell motility (9, 14). The importance of each pathway has been exemplified by the vascular phenotypes of genetic models. Accordingly, several studies have suggested that certain small-molecule inhibitors targeting these pathways also exhibit direct antiangiogenic effects in preclinical models (1517). With more than 100 kinase inhibitors currently in clinical development, a better understanding of the VEGF-regulated kinome and crucial substrates will be key in determining the action of novel drugs on vasculature, identifying potential pharmacodynamic biomarkers and preventing unfavorable side effects.

To systematically characterize the endothelial cell circuitry activated by VEGF, we applied a pathway-based differential phosphoproteomic approach (18, 19) to identify and quantify VEGF-regulated protein phosphorylation sites in a human endothelial cell model system. We used sixplex tandem mass tag (TMT) technology (20) to generate a high-resolution time profile of the phosphorylation events. We found that signaling molecules in the PI3K–mTORC2 (mammalian target of rapamycin complex 2) axis were enriched among the phosphorylated proteins, and we identified FoxO1 (forkhead box protein O1) as a critical downstream effector that mediated both the cell survival signal and the endothelial receptor tyrosine kinase (RTK) reprogramming. These data provide insights into early VEGF receptor signaling events and should allow rational evaluation of the therapeutic efficacy of kinase inhibitors on pathological angiogenesis.

Results

A focused phosphoproteomic strategy based on phospho-motif antibodies

We applied an integrated phosphoproteomic technology to comprehensively analyze the VEGF-regulated phosphorylation events (fig. S1A). We used human umbilical vein endothelial cells (HUVECs) because they are a particularly well-characterized and extensively investigated model system for both endothelial cell biology and VEGF signaling (2123). HUVECs are relatively easy to isolate and grow, making the large-scale proteomic experiments feasible. However, HUVECs express both arterial and venous markers, reflecting their origin from the umbilical vein, which carries arterial-like oxygen-enriched blood. Therefore, it may be argued that HUVECs are not the ideal cell type to investigate angiogenesis, a process that primarily involves microvascular endothelial cells. Furthermore, being a purely cell culture system, they do not fully model physiological processes such as angiogenesis, arteriogenesis, or vascular permeability. These theoretical limitations notwithstanding, the hypotheses generated by our study should provide the basis for further investigation in additional systems.

First, VEGF-treated HUVECs were profiled using 16 phospho-motif antibodies to provide a kinome-wide view of cellular phosphorylation (Fig. 1A and fig. S1B). To enrich phosphopeptides by immunoaffinity, we conducted five experiments with five selected antibodies (Fig. 1A), which either detected many changed phosphoproteins in the Western blot analysis (for example, the Akt motif Ab) or represented kinases that are activated by VEGF (for example, the MAPK motif Ab) (24). TMT labeling was then used to permit six-way multiplexed relative quantification of phosphopeptide abundances on a linear ion trap/Orbitrap mass spectrometer (LTQ Orbitrap).

Fig. 1 A focused phosphoproteomic strategy based on phospho-motif antibodies.

(A) HUVECs were treated with VEGF or sorafenib and immunoblotted with the indicated phospho-motif antibodies. Red arrows indicate phosphoproteins stimulated by VEGF. Data represent two independent experiments. (B) Mass spectra of a phosphopeptide from VEGFR2. (C) Immunoblotting verification of phosphorylation sites identified by proteomic experiments with different phospho-motif antibodies. Data represent two independent experiments. (D) Western blot analysis of total protein abundance. Data represent two independent experiments.

To obtain a high-resolution time profile of the phosphoproteome, we stimulated serum-deprived HUVECs with VEGF for various time points. Sorafenib, a Food and Drug Administration (FDA)–approved antiangiogenic inhibitor that targets VEGF receptors, was included as a control. Representative identification and quantification of a previously characterized phosphorylation site on VEGFR2 (Tyr951) are shown in Fig. 1B. Using commercially available phospho-specific antibodies, we verified results generated from quantitative mass spectrometry (MS) by Western blot analysis (Fig. 1C). During the course of the experiment, we did not observe apparent changes in total abundance of various proteins (Fig. 1D).

VEGF-regulated phosphoproteome dynamics

In total, we identified 5682 unique phosphopeptides on 1324 distinct proteins in HUVECs [table S1; false discovery rate (FDR) <1%]. Scatter plots of raw data demonstrated a linear correlation between any two groups, suggesting that the TMT-based quantification did not introduce systematic bias (fig. S2A). Despite serum and growth factor starvation, sorafenib treatment resulted in a further overall decrease in protein phosphorylation compared to control HUVECs (fig. S2B). We observed changes in phosphopeptides across different time points of VEGF stimulation (fig. S2C). These analyses confirmed that our quantitative phosphoproteomic approach enabled us to carry out unbiased, large-scale discovery of protein phosphorylation events in primary human endothelial cells.

To analyze the dynamic changes as a result of VEGF treatment, we chose a conservative cutoff value of a twofold difference (increase or decrease relative to control) at any time point (2527). There were a total of 1530 phosphopeptides in this category, which constituted the VEGF-regulated phosphoproteome (table S2). To determine the patterns in time profiles of the phosphopeptides regulated by VEGF, we applied fuzzy c-means clustering on our time series data set (27, 28). By setting the cluster size to 6 and the fuzzification parameter to 1.75, we found three clusters of peptides that showed increased phosphorylation (clusters 1 to 3) and three clusters of peptides (clusters 4 to 6) that showed decreased phosphorylation (Fig. 2A and table S2). We observed that clusters 4 to 6 are mostly enriched in cell adhesion, cell-cell junction, and cytoskeleton proteins (table S2). Although the decreased phosphorylation of a large number of proteins is somewhat unexpected, it is consistent with previous observations in tumor cells stimulated with epidermal growth factor (EGF) or heregulin (27, 29), and the decreased phosphorylation may be caused by decreased protein abundance, by dephosphorylation by phosphatases, or by translocation to another subcellular compartment.

Fig. 2 Temporal dynamics and functional assays of the VEGF-regulated phosphoproteins.

(A) Clustering of VEGF-regulated phosphopeptides. Temporal profiles were assigned to six clusters using fuzzy c-means clustering. (B) Gene list for siRNA screens (four individual siRNAs targeting each gene). The amalgamated data from three independent experiments are shown. Red, gene knockdown resulted in increased cell viability (V), motility (M), or both (B); green, gene knockdown resulted in decreased phenotype; yellow, gene knockdown did not affect endothelial cells compared with nontarget controls. White indicates excluded siRNAs after the quality control screen. (C) HUVEC growth upon FoxO1 knockdown was analyzed by the xCELLigence system. Data represent three independent experiments. (D) VEGF-induced wound closure was monitored by the IncuCyte imaging system. Data represent three independent experiments. Scale bar, 200 μm. NTC, nontargeting control. Quantification and statistical analysis are shown in fig. S3A.

Because VEGF triggers activation of VEGFRs, which are tyrosine kinases, the downstream substrate proteins with increased phosphorylation (clusters 1 to 3 in Fig. 2A) were of particular interest. Temporal profiles indicated that the phosphorylation of these proteins was increased by VEGF in a specific sequence (Fig. 2A). The early responders to VEGF stimulation (cluster 1) included cytoskeleton proteins (GO:0005856) and GTPase (guanosine triphosphatase) regulators (GO:0030695), which could affect the dynamics of endothelial cell adhesion and motility (table S3). These membrane-proximal signaling events were followed by phosphorylation of nuclear proteins (GO:0005634) such as transcription factors (cluster 2), which presumably regulate gene expression (GO:0010468). At the later stage (cluster 3), many phosphoproteins were involved in protein binding (GO:0005515) or were serine/threonine kinases (GO:0004674), implying the establishment of an integrated signaling network (table S3). Pathway analysis revealed that VEGF-induced phosphoproteins were enriched in focal adhesions (KEGG hsa04510), the VEGF signaling pathway (KEGG hsa04370), and regulation of actin cytoskeleton (KEGG hsa04810). Overall, these data provided a detailed view of the VEGF-regulated phosphoproteome dynamics.

Functional assays of the VEGF-regulated phosphoproteins

We hypothesized that some phosphorylated proteins may participate in VEGF-mediated signaling pathways and endothelial cell activities. To investigate the biological functions of VEGF-regulated phosphoproteins, we performed small interfering RNA (siRNA) screens and characterized the roles of ~90 proteins in VEGF-induced cell viability and motility (table S4). To rule out potential off-target effects, we used sets of four individual siRNAs and selected gene candidates with at least two siRNAs exhibiting similar phenotype. We performed the xCELLigence experiments to assess cell viability in real-time and image-based wound closure assays to monitor cell migration (Fig. 2B). We identified 24 proteins that controlled VEGF-mediated endothelial cell viability (V), 15 proteins that specifically regulated cell motility (M), and 9 proteins (B) that affected both (Fig. 2B). Although further studies are required to investigate whether phosphorylation changes the function of these proteins, we conclude that some substrates identified in our profiling study may be important effectors downstream of the VEGF receptors.

Intriguingly, a considerable percentage of candidate proteins were inhibitors of VEGF signaling, including several putative Akt substrates and multiple MEKK kinases (Fig. 2B). One highly scored example was FoxO1, and knockdown of FoxO1 significantly promoted HUVEC growth induced by VEGF (Fig. 2C). Similar results were also recapitulated in cell migration assays (Fig. 2D and fig. S3A). Further analysis indicated that FoxO1 was phosphorylated in response to VEGF (fig. S3B).

VEGF signaling mechanisms that modulate protein phosphorylation

We combined in silico analysis and experimental perturbation to uncover the molecular mechanisms through which VEGF regulates protein phosphorylation. Initially, we performed kinase enrichment analysis (30) on the proteins showing increased phosphorylation (table S5). Substrates of the AGC kinase group were significantly overrepresented (P = 1.4 × 10−12, Fisher’s exact test). Among the AGC kinases, Akt in the PI3K pathway was predicted to be responsible for the phosphorylation of many proteins (P = 1.89 × 10−9, Fisher’s exact test). In addition to Akt, we also noticed SGK (serum- and glucocorticoid-inducible kinase) and PKC (protein kinase C) kinases in the top list, which are substrates of mTORC2 (31). These results suggest that the PI3K-mTORC2-Akt pathway was activated upon VEGF stimulation and phosphorylated various substrate proteins identified in our proteomic study. Indeed, we detected multiple phosphorylated proteins in this pathway by Western blots, some of which were not captured in the phosphoproteomic experiments (fig. S3C). The potentially critical role of mTORC2 in VEGF signaling is underappreciated and is particularly interesting.

As a complement to the computational approach, we used inhibitors of different pathways to test their effects on VEGF-induced protein phosphorylation and cell activities (fig. S4). PI3K inhibitors suppressed VEGF-induced phosphorylation events (fig. S5A) and endothelial cell viability (Fig. 3A). The mTOR kinase inhibitors (GNE-236 and AZD8055), but not rapalogs (everolimus), were comparable to the PI3K inhibitors in reducing protein phosphorylation and cell growth (Fig. 3, A and B, and fig. S5B). Because the mTOR kinase inhibitors target both mTORC1 and mTORC2, we specifically knocked down Raptor (a component specific to mTORC1), Rictor (a component specific to mTORC2), or mTOR and found that mTORC2 was more important for VEGF-induced HUVEC viability (fig. S5C). Furthermore, mTOR inhibitors significantly blocked vessel assembly (Fig. 3C) and bead sprouting (Fig. 3D and fig. S5D), both of which require an intact VEGF receptor pathway. Finally, to elaborate on the specific role of the PI3K-mTORC2-Akt axis in VEGF signaling, we developed a gene signature as the molecular fingerprint of the VEGF stimulation in HUVECs (fig. S5E and table S6). The PI3K, mTOR, or Akt inhibitors affected the expression of multiple VEGF signature genes, whereas the effect of the MEK inhibitor selumetinib was limited (Fig. 3E and table S7). Therefore, the PI3K-mTORC2-Akt axis is a major pathway that modulates the endothelial responses to VEGF at the transcriptional, posttranslational, and functional levels.

Fig. 3 VEGF signaling mechanisms to modulate protein phosphorylation.

(A) HUVEC viability measured by CellTiter-Glo luminescent assays (n = 8 biological replicates). (B) HUVECs were pretreated with mTOR inhibitors, stimulated with VEGF, and probed with Akt motif antibody. Data represent two independent experiments. DMSO, dimethyl sulfoxide. (C) HUVEC assembly analyzed by CellPlayer kinetic angiogenesis assay. Left: quantification of tube length. Data points are means of eight biological replicates ± SD. Right: representative images of tube networks. Scale bar, 500 μm. (D) Phalloidin staining of HUVEC sprouts. Scale bar, 100 μm. Quantification is shown in fig. S5D. (E) Perturbation profiles of different inhibitors (two biological replicates). The inhibitors (columns) and their effects on VEGF signature genes (rows) are shown. Red: expression was increased by VEGF stimulation; blue: expression was decreased by VEGF stimulation; white: no substantial changes upon VEGF stimulation. (F) Phosphorylation of FoxO1. Data represent three independent experiments. (G) Immunofluorescence staining of FoxO1. Scale bar, 20 μm. Quantification is shown in fig. S5G. (H) Inhibition of HUVEC growth by GDC-0941 or GNE-236 was significantly decreased by FoxO1 knockdown (n = 8 biological replicates; P < 0.05, Kruskal-Wallis followed by Mann-Whitney tests). (I) HUVEC apoptosis induced by GDC-0941 or GNE-236 was impaired by FoxO1 knockdown. Data represent three independent experiments.

We sought to determine the crucial effectors downstream of PI3K-mTORC2-Akt. FoxO1 is phosphorylated by Akt and was a likely candidate because knockdown of FoxO1 inhibited VEGF-dependent cell growth (Fig. 2C). Consistent with our finding that VEGF increases phosphorylation of FoxO1 (fig. S3B), PI3K, mTOR, or Akt inhibitors almost completely abrogated its phosphorylation at Thr24 (Fig. 3F and fig. S5F), which resulted in translocation of FoxO1 from cytoplasm into nucleus, where it could serve as an active transcription factor (Fig. 3G and fig. S5G). Upon FoxO1 knockdown, the inhibitory effects of PI3K or mTOR inhibitors were significantly impaired (Fig. 3H), which could be, at least partially, attributed to the failure to activate the apoptotic machinery (Fig. 3I).

Reprogramming of RTKs in endothelial cells in response to mTOR inhibition

The PI3K pathway is important for VEGFR2 signaling (32). Our findings identify a function for mTORC2 downstream of PI3K, which prompted us to further investigate whether targeting PI3K or mTOR has an equivalent effect on endothelial cells. Although both PI3K and mTOR inhibitors immediately reduced protein phosphorylation in HUVECs, the mTOR kinase inhibitors induced only transient effects (Fig. 4A), unlike PI3K inhibitors (fig. S6A). The phosphorylation of Ser473 in Akt and Ser235/236 in S6 remained blocked 48 hours after addition of GNE-236 or AZD8055, indicating that the kinase activity of mTOR was fully inhibited (Fig. 4B and fig. S6B). However, phosphorylation of Thr308 in Akt, Akt substrates [PRAS40, FoxO1, and GSK-3 (glycogen synthase kinase–3)], and SGK substrate (NDRG1) rebounded as early as 4 hours after adding the mTOR kinase inhibitors (Fig. 4B and fig. S6B), implying that PI3K was activated in response to mTOR inhibition.

Fig. 4 Endothelial RTK reprogramming in response to mTOR inhibition.

(A) Protein phosphorylation after mTOR inhibition was probed with the Akt motif antibody. Data represent three independent experiments. (B) HUVECs were treated with GNE-236 or AZD8055. Cell lysates were immunoblotted with indicated antibodies. Data represent three independent experiments. Quantification is shown in fig. S6C. (C) HUVECs were treated with GNE-236 or AZD8055 and analyzed by phospho-RTK antibody array. Data represent two independent experiments. (D) HUVECs were treated with GNE-236 or AZD8055. Phosphorylation and total protein abundance of receptors were analyzed by immunoblotting. Data represent three independent experiments. Quantification is shown in fig. S7A. (E) HUVECs were treated with GNE-236 or AZD8055. The expression of genes encoding different RTKs was analyzed by microarray. Expression values were normalized by row (n = 3 biological replicates). (F) HUVECs were treated with AZD8055 after RNA interference (RNAi) knockdown of FoxO1. Cell lysates were analyzed by phospho-RTK antibody array. Data represent two independent experiments. (G) HUVECs transfected with FoxO1 siRNA or nontarget control sequence were immunoblotted to assess phosphorylation and total abundance of receptor tyrosine kinases. Data represent three independent experiments. (H) Cell viability was assayed in HUVECs treated with GNE-236, GDC-0941, or both compounds (n = 8 biological replicates).

In tumor cells, mTOR inhibitors relieve inhibition of upstream receptor tyrosine kinases, leading to subsequent increased activation of PI3K (33). However, this phenomenon has not been fully investigated in endothelial cells. We tested a panel of 42 kinase receptors in HUVECs exposed to mTOR kinase inhibitors and found that phosphorylation of multiple RTKs was induced, including EGFR, VEGFR2, IGF-1R (insulin-like growth factor 1 receptor), and insulin receptor (Fig. 4C). These results were confirmed by immunoblotting using phospho-specific antibodies (Fig. 4D and fig. S7A). Increased RTK phosphorylation was in part due to increased abundance of these receptors (Fig. 4D and fig. S7A). Using microarray analysis, we examined changes in the expression of genes encoding all RTKs in response to the mTOR kinase inhibitors (table S8) and identified 16 different receptor-encoding genes with enhanced transcription (Fig. 4E). Quantitative polymerase chain reaction (PCR) validated these data and revealed that rapalog or MEK inhibitor had limited or no effects (fig. S7B).

The FoxO family of transcription factors is involved in the induction of RTKs by Akt inhibitors in tumor cells (34). Because the mTOR inhibitors activated FoxO1 in HUVECs, we investigated its role in endothelial RTK reprogramming. Knockdown of FoxO1 in HUVECs diminished the induction of RTK phosphorylation by AZD8055 (Fig. 4F). Similar results were obtained using GNE-236, and receptor protein abundance was reduced when FoxO1 was depleted (Fig. 4G and fig. S8A). Note that increased phosphorylation of VEGFR2 was coupled with decreased VEGFR2 abundance (Fig. 4D). We speculate that there are alternative mechanisms to activate endothelial RTKs, such as increased activity of protein kinases, decreased activity of protein phosphatases, or stabilization of receptors on cell membrane, which remain to be elucidated. Nevertheless, FoxO1 was also required for the activation of VEGFR2 by mTOR kinase inhibitors (Fig. 4G and fig. S8A). Together, these data demonstrated that FoxO1 is regulated by VEGF through the PI3K-mTOR2-Akt pathway, and in turn, FoxO1 activated both cell apoptosis and RTK reprogramming in endothelial cells.

The PI3K inhibitor (GDC-0941) or PI3K-mTOR dual inhibitor (GDC-0980) similarly led to the RTK reprogramming in endothelial cells (fig. S8B). However, in contrast to the mTOR inhibitors, GDC-0941 induced sustained inhibition of the PI3K pathway, illustrated by ablated phosphorylation of Thr308 and Ser473 in Akt, Thr24 in FoxO1, and Thr246 in PRAS40 (fig. S8C). These findings predicted that at pharmacologically equivalent exposures, PI3K inhibitors may be superior to mTOR kinase inhibitors in reducing tumor angiogenesis. Furthermore, suboptimal concentrations of PI3K and mTOR inhibitors might have synergistic effects and fully inhibit endothelial cell growth. We tested this hypothesis in HUVECs and found that combination therapy was superior to single agents starting at low concentrations (Fig. 4H). When 0.25 μM GNE-236 and GDC-0941 were combined, growth inhibition was >80%, which required >2 μM single treatment to achieve.

Discussion

Here, we have systematically profiled the VEGF-regulated phosphoproteome in primary human endothelial cells. We provided detailed and time-dependent information on numerous phosphorylation events controlled by the VEGF signaling pathway. Furthermore, by perturbing pathways using pharmacological inhibitors, we described the molecular mechanisms through which VEGF activates protein phosphorylation. Our findings and approach illustrate the power of phosphoproteomic tools for understanding endothelial cell biology and open up new avenues for future investigations.

VEGF is a key regulator of normal vascular development (35, 36). Also, VEGF is implicated in pathological angiogenesis (1). To date, 10 VEGF pathway inhibitors have been approved by the FDA as therapies for cancer or intraocular neovascular disorders (37, 38). Our study provides a more thorough understanding of the VEGF signaling pathways, which may help identify biomarkers and improve clinical index of antiangiogenic treatments. Furthermore, two mTOR inhibitors (rapalogs) have been approved for cancer therapy (39), and various PI3K, mTOR, and Akt kinase inhibitors are presently undergoing clinical trials (4042). The potential interference of these inhibitors with the VEGFR2 pathway and their impact on normal or pathological vasculature have not been rigorously evaluated. Therefore, it is particularly important and timely to understand the phosphorylation and dephosphorylation of intracellular proteins in endothelial cells, which are the earliest events after VEGF stimulation, as well as the reciprocal interactions and feedback loops among these different signaling pathways.

Our study demonstrates that the phospho-motif antibody-based MS technology can robustly probe the VEGF-regulated phosphoproteome in detail. In addition, it can also be used to study perturbations such as those due to drug treatment. Application of this approach enabled us to identify and quantify 5682 phosphopeptides in HUVECs, which formed a large phosphorylation data set of primary human endothelial cells. Furthermore, the high-resolution time profiles of phosphoproteins revealed intrinsic patterns in phosphoproteome dynamics and brought attention to underexplored molecules in the VEGF receptor pathway. Our initial functional characterization using siRNA screens suggested that some proteins could participate in VEGF-mediated growth or migration of endothelial cells. Further research is warranted to determine the molecular mechanisms and investigate the impact of phosphorylation on their functions. Our in vitro findings provide the basis for more complex in vivo studies, which are required for understanding VEGF signaling in physiological contexts.

In theory, our phosphoproteomic experiments measured the net effects of both VEGFR1 and VEGFR2 signaling on protein phosphorylation in HUVECs. There is compelling evidence showing that VEGFR2 is the major mediator of VEGF-induced mitogenesis, migration, permeability, and angiogenesis in endothelial cells (43, 44). VEGFR1 has a more limited signaling potential in endothelial cells and may function in some circumstances as a “decoy” receptor, which inhibits the activity of VEGFR2 by preventing VEGF binding to VEGFR2 (8). Indeed, a targeted mutation in VEGFR1 resulting in lack of the intracellular tyrosine kinase domain does not lead to overt defects in vascular development in mice (45). VEGFR1 activation may generate mitogenic or chemotactic signals in certain tumor cells expressing VEGFR1, although not in endothelial cells, under the same experimental conditions (46). Therefore, we reasoned that VEGFR2 activation is largely responsible for the VEGF-regulated phosphorylation events identified in the current study. In future studies, it will be interesting to compare the phosphoproteome in response to VEGF, VEGFR1-specific ligands [PlGF (placenta growth factor) and VEGFB], or receptor-specific VEGF mutants (47), not only in endothelial cells isolated from different vascular beds (48) but also in cell types other than endothelial cells, including bone marrow cells (49) and tumor cell lines expressing VEGFR1 (46). The phosphorylation spectrum induced by VEGFR1 remains to be elucidated.

Our phosphoproteomic analyses and follow-up experiments pinpoint the critical role of PI3K-mTORC2-FoxO1 axis in VEGF signaling and endothelial cell activity. The physiological importance of PI3K, mTORC2, and FoxO1 in the vasculature is underscored by previous work showing that p110α-deficient (50), mLST8-deficient (51), or FoxO1-deficient (52, 53) embryos exhibit defective vascular development. mTORC2 has been reported to be a crucial regulatory component of the Akt-FoxO, but not the Akt-mTORC1, signaling cascade (51, 54). These previous findings are consistent with our observations that mTORC2 was more important than mTORC1 for VEGF to promote HUVEC viability. Among the phosphoproteins regulated by VEGF, FoxO1 was validated as a central focal point for the PI3K-mTORC2 pathway in vascular biology. Knockdown of FoxO1 significantly facilitated VEGF-induced endothelial cell growth and impaired the suppressive effects of PI3K or mTOR inhibitors. The major effectors of FoxO1 in endothelial cells remain to be elucidated. Our microarray analysis revealed that Sprouty2 and Bim, two FoxO transcriptional targets (55, 56), are increased in abundance in HUVECs in response to PI3K or mTOR inhibitors (table S8). It is tempting to speculate that Sprouty2 and Bim are two candidates downstream of FoxO1 that inhibit endothelial cell proliferation and survival, respectively.

Although our phosphorylation data set provides comprehensive information on VEGF-regulated phosphoproteome, it is unlikely that all critical pathways were captured, considering limitations in proteomic technologies as well as in the endothelial model system. It is noteworthy that genetic studies have revealed the essential role of VEGFR2 Tyr1173 (human Tyr1175) in endothelial cell development, because vegfr21173F homozygous mice, similar to vegfr2 null mice, died in utero between embryonic day 8.5 and 9.5 without organized blood vessels or yolk sac blood islands (57). Phosphorylation of Tyr1175 in VEGFR2 is crucial for VEGF-dependent activation of the phospholipase C–γ (PLC-γ)–PKC–MAPK pathway in cultured endothelial cells (58). Although MAPK was phosphorylated upon VEGF stimulation (Fig. 2A, cluster 1), our phosphoproteomic approach did not identify many MAPK substrates phosphorylated in response to VEGF (table S1). One reason might be that some of this information was lost during sample preparation and data collection because of intrinsic limitations of the current proteomic platform. Alternatively, it is possible that under our experimental conditions, HUVECs do not rely to a large extent on the MAPK pathway. Indeed, the MEK inhibitor selumetinib had limited effects on VEGF-induced transcription (Fig. 3E). It is also noteworthy that the vascular phenotype in VEGFR21173F homozygous mice may not be entirely caused by lack of MAPK activation because phospho-MAPK staining was not appreciably reduced relative to wild-type embryos (57). Indeed, it has been reported that VEGFR2 Tyr1175 is required not only for activating the PLC-γ–PKC–MAPK pathway but also for additional signaling pathways, including VEGF-induced PI3K activation (59). The importance of the MAPK pathway and its effectors in VEGF receptor signaling during developmental and pathological angiogenesis clearly remains to be further elucidated.

Extensive negative feedback loops have emerged as ubiquitous features of oncogenic signaling networks (60). Drug-induced relief of negative feedbacks can be detrimental to the efficacy of targeted therapies in cancer patients. Here, we show that similar mechanisms might also exist in tumor-associated stromal cells such as endothelial cells. Specifically, mTOR kinase inhibitors induced the expression of multiple RTK-encoding genes and signaling by promoting the dephosphorylation of FoxO1, which resulted in increased activation of PI3K in endothelial cells. In contrast, although PI3K inhibitors also caused RTK reprogramming in HUVECs, they maintained constant inhibition of PI3K and downstream components, indicating that targeting PI3K would be a more favorable approach to control pathological angiogenesis. Furthermore, our data suggest that mTOR kinase inhibitors may render endothelial cells more prone to PI3K inhibition, which in principle could lower drug exposure without compromising clinical efficacy. Indeed, we observed synergistic effects of PI3K and mTOR inhibitors to restrain HUVEC growth. Considering that treatment with a single agent inhibiting PI3K or mTOR may not be well tolerated at high doses because of their importance in normal tissues, combination or serial treatment at suboptimal concentrations could provide a therapeutic window to manage aberrant vasculature in tumors or other diseases.

In conclusion, our study documents the multitude, variety, and kinetics of endothelial proteins that undergo changes in phosphorylation status upon VEGF stimulation. This data set provides the basis for future investigations in the area of vascular biology and cell signaling. The complexity and multiplicity, in addition to vigorous reciprocal negative feedback, involving VEGF signaling networks are only beginning to be delineated. A more thorough understanding of the expansive molecular programs in endothelium will unmask the potential for future pharmacologic drug discovery and for identifying novel combination therapies.

Materials and Methods

Reagents

GDC-0980 (61), GDC-0941 (62), GNE-236 (63), and GDC-0068 (64) were synthesized at Genentech. AZD8055, sorafenib, axitinib, vandetanib, BEZ235, MK2206, selumetinib, and everolimus were purchased from Selleckchem or LC Laboratories. All inhibitors were reconstituted in DMSO (Sigma-Aldrich) at a stock concentration of 10 mM. Inhibitor concentrations used for experiments were selected based on fig. S4. Cell viability assays (CellTiter-Glo) were performed in 96-well black-walled plates according to the manufacturer’s protocol (Promega). For gene knockdown, siRNA sequences (Dharmacon) were transfected with Lipofectamine RNAiMAX Reagent (Invitrogen). Recombinant human VEGF165 was purchased from Sigma-Aldrich.

KinomeView profiling

Primary HUVECs were purchased from Lonza and cultured in EGM-2 medium (Lonza). HUVECs were treated as indicated and then lysed in radioimmunoprecipitation assay (RIPA) buffer [50 mM tris (pH 7.4), 150 mM NaCl, 1% NP-40, 0.1% SDS, 2 μM EDTA] containing proteinase inhibitors (Roche) and phosphatase inhibitors (Sigma). The cell lysates (20 μg of protein) were subjected to SDS–polyacrylamide gel electrophoresis (SDS-PAGE) and Western blot. Phospho-motif antibodies in KinomeView Profiling Kit (Cell Signaling Technology) were used to detect tyrosine, serine, and threonine phosphorylation mediated by diverse kinase families throughout the kinome.

Immunoaffinity phosphopeptide enrichment

Motif-specific phosphopeptides were enriched from basal, VEGF-stimulated (10, 20, 30, and 60 min), or inhibitor-treated HUVECs as previously described (65). Briefly, immunoaffinity precipitations (IAPs) were performed on 10 mg of tryptic digests (1:100 enzyme/substrates) with anti-pY (Cell Signaling Technology, #9411) and anti-PXS*P, S*PX(K/R) (MAPK substrate motif, #2325), or on 10 mg of Glu-C digests (1:25 enzyme/substrates) with anti-RXX(S*/T*) (Akt substrate motif, #9614), anti-(K/R)(K/R)X(S*/T*) (PKA substrate motif, #9624), and anti-(K/R)XS*X(K/R) (PKC substrate motif, #6967).

Sixplex TMT labeling

IAP eluents were dried and reconstituted in 10 μl of 20 mM Hepes at pH 8.0. Immunoprecipitated Glu-C peptides were further digested with trypsin (Promega) for 4 hours to facilitate sequence identification. Phosphopeptides enriched from the six conditions were labeled with 0.1 U of TMT sixplex (Thermo Scientific), respectively, and combined afterward following the manufacturer’s instructions. Samples were further enriched for phosphopeptides with a TiO2 Tip (GL Sciences USA) as described previously (66).

Liquid chromatography–tandem MS, peptide identification, quantitation, and data analysis

TMT-labeled phosphopeptides were analyzed with a nanoACQUITY UPLC System (Waters) directly coupled to an LTQ Orbitrap Velos mass spectrometer (Thermo Scientific). Peptides were reconstituted in 0.1% formic acid (FA) with 2% acetonitrile (ACN), loaded onto a Symmetry C18 column (1.7 mm BEH-130, 0.1 × 100 mm; Waters), and separated with a 90-min gradient from 2 to 25% solvent B (0.1% FA, 98% ACN) at 1 μl/min flow rate. Peptides were eluted directly into the mass spectrometer with a spray voltage of 1.2 kV. Full MS data were acquired in FT for 400 to 1600 mass/charge ratio (m/z) with a 30,000 resolution. The eight most abundant ions found in the full MS were selected for MS/MS through a 2-dalton isolation window. Parent ions were fragmented with higher-energy collision dissociation (HCD) at 50% normalized collision energy and acquired in FT with a 7500 resolution. Automatic gain controls were set to 1,000,000 and 50,000 for the full MS and MS/MS, respectively.

Acquired MS/MS spectra were searched with the Mascot (Matrix Sciences). Search criteria included a full MS tolerance of 50 parts per million (ppm), MS/MS tolerance of 0.02 dalton with oxidation (+15.9949 daltons) of methionine and phosphorylation (+79.9663 daltons) of serine, threonine, and tyrosine as variable modifications, and TMT sixplex (+229.1629 daltons) on lysine and peptide N termini and carbamidomethylation (+57.0215 daltons) of cysteine as static modifications. Data were searched against the human and contaminant subset of the UniProt database that consists of the reverse protein sequences and then filtered with linear discriminator analysis (67) at the peptide level at 5% FDR and further filtered at the protein level at 2% FDR. Phosphorylation sites were localized with the Ascore algorithm (68), and only the phosphopeptides matching to motif antibody specificity were selected for further analysis. TMT reporter ions were quantified by peak height with an in-house developed software package named Mojave with a 0.01-dalton matching window of the theoretical reporter masses and adjusted for isotope purity distribution (69).

Western blot

HUVECs were lysed in RIPA buffer [50 mM tris (pH 7.4), 150 mM NaCl, 1% NP-40, 0.1% SDS, 2 μM EDTA] containing proteinase inhibitors (Roche) and phosphatase inhibitors (Sigma). The cell lysates (20 μg of protein) were subjected to SDS-PAGE and Western blot analysis. Antibodies directed toward the following proteins were used: pTyr951 VEGFR2, pTyr1059 VEGFR2, pTyr1175 VEGFR2, VEGFR2, pTyr1068 EGFR, EGFR, pTyr1135/1136 IGF-1R, IGF-1R, pTyr702 AXL, AXL, pTyr410 p130 Cas, pSer588 AS160, AS160, pSer252 catenin δ, catenin δ, pSer1177 eNOS, eNOS, pSer507 MYPT1, MYPT1, pThr246 PRAS40, PRAS40, pSer422 eIF4B, eIF4B, pSer330 NDRG1, NDRG1, pSer167/170 MARCKS, MARCKS, pThr308 Akt, Akt, pThr24 FoxO1, FoxO1, pSer21/9 GSK-3α/β, GSK-3α/β, pThr389 S6K1, pSer133 CREB (adenosine 3′,5′-monophosphate response element–binding protein), Raptor, Rictor, mTOR, PARP [poly(adenosine 5′-diphosphate–ribose) polymerase], caspase-3, actin (Cell Signaling Technology); pSer417 Girdin (Abcam); and Girdin, p130 Cas (Santa Cruz Biotechnology). Human phospho-RTK antibody array was performed following the manufacturer’s instructions (R&D Systems).

siRNA screening

A customized siRNA library (Dharmacon) containing 360 oligos (four siRNAs targeting each gene) was synthesized and resuspended in siRNA buffer (Dharmacon) at a concentration of 1 μM. Transfection solution containing 15 nM siRNA and 0.135 μl per well of Lipofectamine RNAiMAX Reagent was prepared with the Mini Oasis liquid handler. Five thousand HUVECs were grown overnight in 100 μl of EGM-2 media (Lonza) and transfected in 96-well plates. A primary screen was performed to control the siRNA quality, and for each gene, it was necessary for at least two siRNAs to present similar HUVEC growth curves, which were automatically recorded on the xCELLigence System (Roche) in real time. To screen for siRNAs that regulate VEGF-induced cell growth, we switched HUVECs to VEGF (50 ng/ml) in EBM-2 24 hours after transfection, and cell viability was analyzed on the xCELLigence System for an additional 72 hours. To screen for siRNAs that regulate VEGF-induced cell migration, the Essen 96-well WoundMaker was used to generate consistent wounds in the confluent monolayer of HUVECs 48 hours after transfection. HUVECs were switched to VEGF (50 ng/ml) in EBM-2, and cell migration was monitored by the IncuCyte imaging system for an additional 48 hours.

Microarray analysis and quantitative PCR

HUVEC RNA was prepared with RNeasy Plus Mini Kit (Qiagen) according to the manufacturer’s protocol. Total RNA was subjected to microarray analysis with Affymetrix Human Genome U133 Plus 2.0. Three biological replicates per treatment group were included for statistical analyses. Affymetrix microarray probe-level data were normalized by robust multiarray average procedure (70). Differential gene expression was analyzed with linear models for microarray data (Limma) (71). TaqMan Gene Expression Assays (Applied Biosystems) were performed to verify the microarray results. The relative expression of each gene was normalized to human β-actin. At least three biological replicates were included for each condition. To generate the VEGF gene signature, we stimulated HUVECs with VEGF (50 ng/ml) for 0, 2, 4, 6, and 8 hours. We selected genes that were significantly (P < 0.05) changed more than twofold in response to VEGF at any time point. Quantitative PCR was performed to verify differential gene expression, and HUVECs with VEGFR2 knockdown served as negative controls. Seventy-two genes comprised the VEGF signature.

NanoString nCounter analysis

HUVECs were stimulated with VEGF (50 ng/ml) for 6 hours in the presence or absence of small-molecule kinase inhibitors. Total RNA samples were prepared and provided to NanoString for analysis. Samples were assayed on NanoDrop 8000 (Thermo Scientific) to determine the concentrations. Each test sample was analyzed with 100 ng of RNA material. Code sets were synthesized targeting 76 genes (VEGF signature genes and housekeeper genes) and 14 controls (90-plex code set). The hybridizations were carried at 65°C for about 16 to 20 hours. The raw transcript counts were first normalized to the sum of six positive control RNA spikes to account for slight differences in assay efficiency (hybridization, purification, binding, etc.). To further correct for input mRNA amounts, we then normalized the data to the geometric mean of the three housekeeper genes [ACTB, B2M, and GAPDH (glyceraldehyde-3-phosphate dehydrogenase)]. Changes in gene expression in response to VEGF stimulation were calculated for each kinase inhibitor and plotted in a heatmap.

CellPlayer kinetic angiogenesis assay

CellPlayer 96-well kinetic angiogenesis assay (Essen BioScience) was performed according to the manufacturer’s protocol. Briefly, lentivirally infected green fluorescent protein (GFP)–HUVECs were cocultured with normal human dermal fibroblasts in a 96-well microplate. The plate was placed in IncuCyte, and images were automatically acquired in both phase and fluorescence every 6 hours for 10 days. At day 4, small-molecule inhibitors were added on the endothelial tube networks and kept throughout the experiment. We used the Angiogenesis Analysis Module to quantify tube length and branch points. Eight biological replicates were included for each condition.

HUVEC sprouting assay

HUVEC sprouting assay was performed as described previously (72). HUVECs were mixed with Cytodex microcarrier beads (Sigma-Aldrich) in a ratio of 106 cells per 2500 beads and incubated for 4 hours at 37°C. Coated beads were cultured in a six-well dish overnight. The following day, ~200 coated beads were dissolved in a solution of fibrinogen (2 mg/ml) (Sigma-Aldrich) in EGM-2 and added to thrombin (0.625 U/ml) (Sigma-Aldrich) in one well of a 24-well plate. Skin fibroblasts (D551) were plated on top of the clot and incubated with D551 condition medium and EGM-2 (3:1). HUVEC sprouts were visualized by immunostaining with Alexa Fluor 488 phalloidin (1:100) and Hoechst 33258 (1:1000) (Invitrogen) overnight at 4°C.

Statistical analysis

To cluster VEGF-regulated phosphopeptides, we log2-transformed and normalized the raw data. The processed data set was then clustered with fuzzy c-means algorithm implemented in GProX (73). The membership values were assigned to each phosphopeptide and allowed color-coding graph items according to their goodness of fit to the cluster consensus profile. Gene ontology and pathway analyses were performed with DAVID Bioinformatics Resources (74, 75), with GOTERM ALL categories and KEGG pathways. Terms were defined as significantly enriched if they contained at least five counts and had a P value of <0.001 and an FDR of <1%. For microarray analyses, FDR was calculated by the Benjamini-Hochberg procedure. In all experiments, comparisons between two groups were based on two-sided Student’s t test, and one-way analysis of variance (ANOVA) was used to test for differences among more groups. P values of <0.05 were considered statistically significant.

Supplementary Materials

www.sciencesignaling.org/cgi/content/full/6/271/ra25/DC1

Fig. S1. KinomeView profiling of endothelial cells.

Fig. S2. Quality control of phosphoproteomic experiments.

Fig. S3. Regulation of PI3K-mTORC2-Akt-FoxO1 axis by VEGF.

Fig. S4. Different classes of small-molecule inhibitors used in the study.

Fig. S5. VEGF signaling mechanisms to regulate phosphorylation.

Fig. S6. mTOR kinase inhibitors in endothelial RTK reprogramming.

Fig. S7. Feedback activation of endothelial RTKs.

Fig. S8. PI3K inhibitors in endothelial RTK reprogramming.

Table S1. The complete data set for identified phosphopeptides.

Table S2. VEGF-regulated phosphopeptides and membership values in fuzzy c-means clustering.

Table S3. Gene ontology and pathway analysis of the VEGF-regulated proteins.

Table S4. Gene list and siRNA sequences for RNAi screen.

Table S5. Kinase enrichment analysis of the VEGF-regulated phosphoproteins.

Table S6. Microarray analysis of HUVECs stimulated with VEGF.

Table S7. NanoString nCounter analysis of HUVECs stimulated with VEGF.

Table S8. Microarray analysis of HUVECs treated with mTOR kinase inhibitors.

References and Notes

Acknowledgments: We thank the Genentech Microchemistry and Proteomic Laboratory, the Microarray Laboratory, and the Department of Protein Chemistry. We are grateful to M. Yan and J. Le Couter for their critical reading of our manuscript. We acknowledge R. Sheng, D. Bustos, D. Kirkpatrick, K. Lin, J. P. Stephan, and Q. Song for their help. Author contributions: G.Z. and N.F. designed the study and wrote the manuscript. K.Y. and J.R.L. performed the phosphoproteomic studies and data analyses. C.H., K.T., and R.S. performed microarray experiments. G.Z., A.C., J.Y., B.H., and C.B. performed the other experiments. Z.J. provided statistical analyses. E.B. and D.S. developed the kinase inhibitors. All authors helped to finalize the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Phosphoproteomics data have been deposited online (https://proteomecommons.org/dataset.jsp?i=YBg2flmrIhFppz2kRuiirrSTulIewK1urEHDUK9A%2FF% 2BZh1PlwKbzqM%2FdtNsDdcWXJk7PZVVsEbVstyLh6JhkJ4gR9p4AAAAAAAAEnA%3D%3D).
View Abstract

Stay Connected to Science Signaling

Navigate This Article