NADPH Oxidase Activity Selectively Modulates Vascular Endothelial Growth Factor Signaling Pathways*

Vascular endothelial growth factor (VEGF) and reactive oxygen species (ROS) play critical roles in vascular physiology and pathophysiology. We have demonstrated previously that NADPH oxidase-derived ROS are required for VEGF-mediated migration and proliferation of endothelial cells. The goal of this study was to determine the extent to which VEGF signaling is coupled to NADPH oxidase activity. Human umbilical vein endothelial cells and/or human coronary artery endothelial cells were transfected with short interfering RNA against the p47phox subunit of NADPH oxidase, treated in the absence or presence of VEGF, and assayed for signaling, gene expression, and function. We show that NADPH oxidase activity is required for VEGF activation of phosphoinositide 3-kinase-Akt-forkhead, and p38 MAPK, but not ERK1/2 or JNK. The permissive role of NADPH oxidase on phosphoinositide 3-kinase-Akt-forkhead signaling is mediated at post-VEGF receptor levels and involves the nonreceptor tyrosine kinase Src. DNA microarrays revealed the existence of two distinct classes of VEGF-responsive genes, one that is ROS-dependent and another that is independent of ROS levels. VEGF-induced, thrombomodulin-dependent activation of protein C was dependent on NADPH oxidase activity, whereas VEGF-induced decay-accelerating factor-mediated protection of endothelial cells against complement-mediated lysis was not. Taken together, these findings suggest that NADPH oxidase-derived ROS selectively modulate some but not all the effects of VEGF on endothelial cell phenotypes.

Reactive oxygen species (ROS) have long been implicated in the pathogenesis of cardiovascular diseases, including atherosclerosis, hypertension, and diabetes (reviewed in Ref. 20). In addition, there is a growing appreciation for the role of ROS in physiological signaling in many cell types, including endothelial cells. NADPH oxidase is the primary source of superoxide in endothelial cells (21)(22)(23)(24). NADPH oxidase was originally identified and characterized in phagocytes, where it contributes to host defense. The NADPH oxidase complex consists of two membrane-bound components, gp91 phox (also known as Nox2) and p22 phox , and several cytosolic regulatory subunits, including p40 phox , p47 phox , p67 phox , and the small GTPase Rac (Rac1 or Rac2). Upon enzyme activation, the cytoplasmic units translocate to the cell membrane where they are assembled with gp91 phox /p22 phox . The resulting multisubunit complex transfers electrons from NAD(P)H to molecular oxygen. Each of the components of the neutrophil NADPH complex has been identified in endothelial cells (25)(26)(27). In addition, endothelial cells express a homologue of gp91 phox /Nox2, termed Nox4 (28,29). The endothelial NADPH oxidase differs from its leukocyte counterpart in several important ways. First, it is pre-assembled and displays constitutive low level activity. Second, agonist-mediated stimulation of NADPH oxidase results in slower and less potent activation of the enzyme (22). Third, the enzyme com-plex is localized in the perinuclear region of endothelial cells and produces an intracellular influx of ROS (30,31). Many different agonists have been shown to induce endothelial NADPH oxidase activity, including hemodynamic forces (32)(33)(34), VEGF (21,22,24), angiopoietin-1 (Ang1) (21,35,36), tumor necrosis factor-␣ (37), thrombin (38), angiotensin II (37,39), endothelin-1 (40), transforming growth factor-␤ (41), oxidized low density lipoprotein (27), and high potassium (42). Activation of NADPH oxidase is mediated by post-translational modification and/or increased transcription of the regulatory subunits.
We previously reported that VEGF induces NADPH oxidase activity and that NADPH oxidase activity is required for VEGFmediated migration and proliferation of endothelial cells (21,22). These results were confirmed and expanded upon by other groups. For example, PI3K and Rac1 were shown to be required for the VEGF-dependent oxidative burst in porcine aortic endothelial cells expressing VEGFR2/KDR (23). Ushio-Fukai et al. (24) demonstrated an important role for gp91 phox in mediating the effect of VEGF on endothelial cell migration and proliferation in vitro and in promoting angiogenesis in a mouse sponge implant model. Similar findings were observed in a hindlimb ischemia model (43). Moreover, we showed that NADPH oxidase-derived ROS are required for VEGF stimulation of manganese superoxide dismutase, activation of NF-B, and inactivation of FKHR in endothelial cells (44).
The mechanisms by which NADPH oxidase-derived ROS influence VEGF are poorly understood. ROS may result in oxidation of cysteine residues on receptor and nonreceptor protein kinases and phosphatases. Indeed, previous studies have demonstrated a role for ROS in VEGFR2 autophosphorylation (23,24). Here, we show that NADPH oxidase activity is required for some, but not all, downstream effects of VEGF in endothelial cells. These data argue against a global sensitivity of VEGF signal transduction pathways to the redox state of the cell, and suggest that therapeutic modulation of ROS will selectively influence the effect of VEGF on endothelial cell phenotypes.
Assay for NADPH Oxidase Activity in HCAEC and HUVEC-HCAEC and HUVEC were washed with ice-cold PBS, collected by a cell scraper, and Dounce-homogenized in buffer containing 20 mM KH 2 PO 4 (pH 7.0), 1ϫ protease mixture inhibitor (Sigma), 1 mM EGTA, 10 g/ml aprotinin (Calbiochem), 0.5 g/ml leupeptin (Sigma), 0.7 g/ml pepstatin (Sigma), and 0.5 mM phenylmethylsulfonyl fluoride (Calbiochem). NADPH oxidase activity of the cell lysate was measured using a modified assay (45). Briefly, photon emission from the chromogenic substrate lucigenin as a function of acceptance of electron/O 2 Ϫ generated by the NADPH oxidase complex was measured every 15 s for 20 min in a Berthold luminometer. NADPH oxidase assay buffer containing 250 mM HEPES (pH 7.4), 120 mM NaCl, 5.9 mM KCl, 1.2 mM MgSO 4 (7H 2 O), 1.75 mM CaCl 2 (2H 2 O), 11 mM glucose, 0.5 mM EDTA, 100 M NADH and 5 M lucigenin was used. The data were converted to relative light units/ min/mg of protein, using a standard curve generated with xanthine/xanthine oxidase. Lucigenin activity (light units/min/mg of protein) of control cells (Scram-AS-transfected) was arbitrarily set at 100%. Total intracellular levels of ROS were determined by FACS analyses of the oxidative conversion of cellpermeable 2Ј,7Ј-dichlorofluorescein diacetate (DCFH-DA; Molecular Probes Inc., Eugene, OR) to fluorescent dichlorofluorescein as described previously (22).
Oligonucleotide Microarray Analysis of Gene Expression-The transcriptional profile of control or VEGF-treated HCAEC with siRNA against p47 phox was characterized by oligonucleotide microarray analysis using the human U133A Affymetrix GeneChip, according to previously described protocols for total RNA extraction and purification, cDNA synthesis, in vitro transcription reaction for production of biotin-labeled cRNA, hybridization of cRNA with U133A Affymetrix gene chips, scanning of image output files, analysis of gene expression data, and hierarchical and functional clustering algorithms (46). The scanned array images were analyzed by dChip (47). In the dChip analysis a smoothing spline normalization method was applied prior to obtaining model-based gene expression indices, also known as signal values. There were no outliers identified by dChip so all samples were carried out for subsequent analysis. When comparing two samples (groups) to identify the genes enriched in a given phenotype, we used the lower confidence bound (LCB) of the fold change between the experiment and the base line. If the 90% LCB of the fold change between the experiment and the base line was above 1.2, the corresponding gene was considered to be differentially expressed. An LCB of Ͼ1.2 typically corresponds with an "actual" fold change of at least 3 in gene expression. GOTree machine was used to identify gene ontology categories for the input gene set (48).
Quantitative Real Time PCR-Real time PCR was carried out as described previously (49). Briefly, total RNA was prepared using the RNeasy RNA extraction kit with DNase-I treatment following the manufacturer's protocol (Qiagen, Valencia, CA). To generate cDNA, total RNA (100 ng) from each of triplicate samples was converted into cDNA using random primers and SuperscriptIII reverse transcriptase (Invitrogen). All cDNA samples were aliquoted and stored at Ϫ80°C. Primers were designed using the Primer Express oligo design software (Applied BioSystems, Foster City, CA) and synthesized by Integrated DNA Technologies (Coralville, IA). All primer sets were subjected to rigorous data base searches to identify potential conflicting transcript matches to pseudogenes or homologous domains within related genes. Amplicons generated from the primer set were analyzed for melting point temperatures using the first derivative primer melting curve software supplied by Applied BioSystems. The SYBR Green I assay and the ABI Prism 7700 sequence detection system were used for detecting real time PCR products from the reverse-transcribed cDNA samples, as described previously (50). 18 S rRNA, which exhibits a constant expression level across all the HCAEC samples, was used as the normalizer. PCRs for each sample were performed in duplicate, and copy numbers were measured as described previously (50). The level of target gene expression was normalized against the 18 S rRNA expression in each sample, and the data were presented as mRNA copies per 10 6 18 S rRNA copies.
Western and Northern Blot Analyses-HCAEC were harvested for total protein, and Western blots were carried out as described previously (19). Anti-Akt, anti-FKHR, anti-p38, anti-VEGFR2, anti-PLC␥-1, and anti-ERK1/2 antibodies and phosphospecific antibodies against Ser-256 FKHR, Ser-473 Akt, ERK1/2, JNK, p38 MAPK, Src, FAK, PLC␥-1, Tyr-951 VEGFR2, and Tyr-1175 VEGFR2 were purchased from Cell Signaling (Beverly, MA). Anti-␤-actin antibodies were from Sigma. Western blots were performed in triplicates using cell extracts prepared from three independent experiments. RNA extraction and Northern blot assays were performed as described previously (22). Immunoprecipitation was carried out as described (19). The protein kinase assay was carried out using the HTScan Src kinase assay kit with a slight modification of the manufacturer's protocol (Cell Signaling). Briefly, HCAEC were lysed in buffer (60 mM HEPES (pH 7.5), 1% Triton X-100, 150 mM NaCl, 1 mM EGTA) plus protease and protein-tyrosine phosphatase inhibitors. Reaction mixtures, containing 12.5 l of cell lysates and 37.5 l of kinase buffer (60 mM HEPES (pH 7.5), 5 mM MgCl 2 , 5 mM MnCl 2 , 3 M Na 3 VO 4 , 1.25 mM dithiothreitol, 20 M ATP, 1.5 M biotinylated peptide substrate), were incubated in a 96-well plate at room temperature for 30 min. 50 l of STOP buffer (50 mM EDTA (pH 8)) were added per well to stop the reaction. 25-l aliquots from each reaction were added to the well of a DELFIA streptavidin-coated, 96-well plates (PerkinElmer Life Sciences) containing 75 l of distilled H 2 O, and the resulting mixture was incubated at room temperature for 60 min. Each well was then washed three times with PBS/T. 100 l of primary mouse monoclonal antibody to phosphotyrosine (P-Tyr-100; 1:1000 dilution) were added to each well. After 60 min of incubation at room temperature, each well was washed three times with PBS/T. 100 l of secondary antimouse IgG (1:500 dilution in PBS/T with 1% BSA) were added to each well and incubated for 30 min at room temperature. Wells were washed with PBS/T five times, and 100 l of DELFIA enhancement solution (PerkinElmer Life Sciences) were added for 5 min. Fluorescence emission was detected at a wavelength of 615 nm.
Immunolocalization Studies-HCAEC were grown in 4-well chamber slides (Lab-Tek, Christchurch, New Zealand) and treated with or without VEGF for the times indicated. Subcellular localization of FKHR and ERK1/2 was determined using anti-FKHR and anti-ERK1/2 antibodies, respectively, and a Cy3-conjugated secondary antibody as described previously (19). Quantitative analyses were carried out by counting 200 cells per time point.
Complement-mediated Cell Lysis Assay-HCAEC were plated in 24-well plates and incubated at 37°C for 24 h prior to siRNA transfection. 24 h following siRNA transfections, the cells were serum-starved overnight and then incubated with or without VEGF for 24 h. 7 mol/liter calcein acetoxymethyl ester (Molecular Probes, Leiden, Netherlands) were added to each well. 30 min later, cells were washed with serum starvation medium containing 1% BSA and then incubated for 30 min with 250 l of monoclonal anti-endoglin (CD105) antibody (IgG2a) (Covance, Berkley, CA) to opsonize the cells. HCAEC were washed with HBSS containing 1% BSA and incubated with 250 l of 5-20% baby rabbit complement (Serotec, Oxford, UK) at 37°C for 30 min. The supernatant from each well was transferred to a 96-well microtiter plate. The remaining HCAEC in the 24-well plate were washed with HBSS plus 1% BSA, and the calcein remaining in the cells was released by incubation with 250 l of HBSS containing 1% BSA and 0.1% Triton X-100. The lysate was then transferred to another 96-well plate, and the calcein released by complement and detergent was estimated using a fluorescence plate reader (model 680; Bio-Rad). Percent specific lysis in triplicate wells was calculated as ((complement-mediated release Ϫ spontaneous release)/maximal release Ϫ spontaneous release)) ϫ 100%, where maximal release ϭ complement-mediated release ϩ detergentmediated release.
Assay for Thrombomodulin-dependent Activation of Protein C-Functional assay for cellular thrombomodulin was carried out as described previously (51), with slight modifications. Briefly, HCAEC transfected with 75 nM siRNAs were seeded on 24-well tissue culture plates. After 24 h, HCAEC were serumstarved in EBM-2 containing 0.5% fetal bovine serum for 18 h and treated with 50 ng/ml VEGF for 14 h. Cells were washed with PBS and incubated in a reaction mixture containing 2.5 mM CaCl 2 , 0.15 M NaCl, 5 nM thrombin, and 5 nM protein C for 3 h. After withdrawing 100 l at each time point, hirudin was added to inhibit thrombin. Equal volume of the chromogenic substrate, 400 M S2366, was added to the supernatant, and absorbance was measured at 405 nM.
Statistical Analyses-All values are presented as mean Ϯ S.D. where appropriate. Statistical significance between two groups was determined by use of a paired t test, and values of p Ͻ 0.05 were considered significant.

NADPH Oxidase Activity Is Abrogated by Down-regulation of p47 phox Expression in Primary Human Endothelial Cells-To
inhibit NADPH oxidase in endothelial cells, HCAEC or HUVEC were transfected with antisense or siRNA against p47 phox . Antisense (AS-p47 phox ) and siRNA (si-p47 phox ) resulted in significant (Ͼ90%) reduction in p47 phox protein levels, compared with scrambled antisense (ScramAS) and siRNA (Scram-si), respectively (supplemental Fig. 1, A and B). Transfection of HCAEC and HUVEC with AS-p47 phox or si-p47 phox also resulted in significant inhibition of NADPH oxidase activity as measured by lucigenin assay (Fig. 1A, HCAEC). As a negative control, treatment of endothelial cells with the nitric-oxide synthase inhibitor, L-NAME, had no effect on NAPDH oxidase activity (Fig. 1A). To determine the effects of NADPH oxidase inhibition on intracellular ROS production in HCAEC, FACS analyses were performed using DCFH-DA. Transfection with si-p47 phox significantly reduced intracellular levels of ROS (to 58.6 Ϯ 6.5%) (Fig. 1B). Consistent with the findings of the lucigenin assays, ROS levels were unaffected by L-NAME. As a positive control, the superoxide donor, phenazine methosulfate, resulted in 180% increase in ROS content (Fig. 1B). Thus, AS-p47 phox and si-p47 phox are each effective in reducing NADPH oxidase activity in endothelial cells.
NADPH Oxidase Activity Is Required for Late but Not Early VEGF-mediated Tyrosine Phosphorylation of VEGFR2/KDR-VEGF treatment of HCAEC resulted in a time-dependent increase in tyrosine phosphorylation of VEGFR2/KDR starting as early as 1 min and peaking at 10 min (data not shown). siRNA-mediated knockdown of p47 phox resulted in partial inhibition of VEGF-mediated phosphorylation of VEGFR2/KDR at 10 min (to 69 Ϯ 6.6%) but had no effect at 1 and 5 min (Fig. 2). Similar results were observed with AS-p47 phox (supplemental Fig. 2A). These findings suggest that NADPH oxidase activity is not required for early phosphorylation of VEGFR2/KDR by VEGF but may play a role in maintaining the receptor in the phosphorylated state.
Phosphorylated ERK1/2 is localized in the nucleus of endothelial cells (52). In keeping with the phosphorylation studies of ERK1/2, si-p47 phox -mediated inhibition of NADPH oxidase had no effect on nuclear translocation of ERK1/2 in VEGFtreated cells (64% nuclear in VEGF-treated control cells versus 68% nuclear in VEGF-treated si-p47 phox -transfected cells) (supplemental Fig. 3, A and B). In accordance with the effect of NADPH oxidase inhibition on VEGF-induced FKHR phosphorylation, si-p47 phox blocked VEGF-mediated nuclear exclusion of FKHR in HCAEC (27% nuclear in VEGF-treated control cells versus 72% nuclear in VEGF-treated si-p47 phox -transfected cells) (supplemental Fig. 4, A and B).
Inhibition of NADPH Oxidase in Primary Human Endothelial Cells Blocks VEGF-mediated Phosphorylation of Src and FAK-The above results indicate that NADPH oxidase activity is required for VEGF-mediated phosphorylation of Akt prior to any observable effect on VEGFR2/KDR phosphorylation (i.e. at 1 and 5 min). Based on these findings, we hypothesized that NADPH oxidase-derived ROS influence VEGF signaling at a level distal to VEGFR2/KDR and proximal to Akt. The nonre- FIGURE 3. NADPH oxidase activity is required for VEGF-mediated phosphorylation of Akt, p38 MAPK, but not ERK1/2 or JNK in HCAEC. Western blot analyses using lysates of HCAEC treated with VEGF (50 ng/ml) for the times indicated. A, HCAEC were transfected with either control siRNA (Scram-si) or si-p47 phox , serum-starved, and then treated with or without VEGF for the times indicated. Western blots were carried out using anti-phospho-Akt (p473-Akt) antibody. The membranes were stripped and reprobed with anti-phospho-ERK1/2 antibody. Anti-Akt antibody (Akt) was used as loading control. B, same as in A except Western blot analyses were carried out using anti-phospho JNK (p-JNK) antibody. The membrane was then stripped and probed with anti-phospho-p38 antibody (p-p38) and anti-p38 antibody (p38). Bar graphs show quantitative analyses of three independent Western blot experiments (mean Ϯ S.D.). *, p Ͻ 0.05, relative to time-matched control.   (Fig. 5A). These effects were blocked by si-p47 phox (Fig. 5A). Phosphorylation of Src has been correlated with the activity of this nonreceptor kinase (57)(58)(59). Consistent with the phosphorylation data, VEGF-mediated induction of Src activity was significantly reduced in cells transfected with si-p47 phox (Fig. 5C).
PLC␥-1 and Shb, through their interaction with the phosphorylated Tyr-1175 residue on VEGFR2, have also been shown to play a role in the activation of the PI3K-Akt signaling pathway in endothelial cells (60). VEGF treatment of HCAEC resulted in time-dependent phosphorylation of PLC␥-1 (Fig. 5B). These effects were not affected by si-p47 phox (Fig. 5B), suggesting that, unlike Src and FAK, VEGFmediated phosphorylation of PLC␥-1 is not dependent on NADPH oxidase activity in HCAEC. Together, these results suggest that NADPH oxidase-derived ROS are critical for mediating the effects of VEGF on Src/FAK signaling intermediates.
To determine whether the ROSsensitive Src and/or FAK play a role in VEGF-mediated activation of Akt, siRNAs were employed to inhibit expression of one or the other kinase in HCAEC. In these experiments, siRNA against Src resulted in Ͼ70% reduction of Src protein, whereas FAK siRNA resulted in Ͼ90% inhibition of FAK (Fig. 6A). Knockdown of Src significantly attenuated VEGF-mediated phosphorylation of Akt (by 58 Ϯ 5.4%) but not ERK1/2 (Fig. 6A). Inhibition of FAK had no effect on VEGF-mediated phosphorylation of Akt or ERK1/2 (Fig. 6A). Similarly, siRNA knockdown of PLC␥-1 (Ͼ75%) failed to inhibit Akt phosphorylation (Fig. 6B). These findings suggest that NADPH oxidase may exert its permissive effect on FIGURE 5. VEGF-induced phosphorylation of Src and FAK, but not PLC␥-1, is sensitive to NADPH oxidase activity in HCAEC. A, Scram-si or si-p47 phox -transfected HCAEC were serum-starved overnight and then treated with VEGF (50 ng/ml) for the times indicated. Western blot analysis was carried out using anti-phospho-Tyr-416 Src antibody (p-Src). The membrane was stripped and reprobed using anti-phospho-Tyr-925/861 FAK (p-FAK) antibody, followed by phospho-specific anti-ERK1/2 antibody (positive control for VEGF response, negative control for NADPH oxidase inhibition), and total anti-ERK1/2 antibody (as control for loading). Lower panel shows quantitative analysis of three independent experiments (mean Ϯ S.D. of fold changes versus control). *, p Ͻ 0.05, relative to time-matched control. B, same as in A except Western blots were carried out using anti-phospho-PLC␥-1 antibody, followed by stripping and reprobing of the membrane using anti-PLC␥-1 antibody as loading control. C, VEGF-induced Src kinase activity was measured in lysates from Scram-si or si-p47 phox -transfected HCAEC as described under "Experimental Procedures." *, p Ͻ 0.05, VEGF-treated versus untreated in Scram-si transfected endothelial cells. VEGF-PI3K-Akt signaling, at least partially, at the level of the nonreceptor tyrosine kinase Src.
Dependence of Src Kinase on NADPH Oxidase Activity Occurs at a Post-receptor Level-Phosphorylation of the tyrosine residue, Tyr-951, in the kinase insert domain of VEGFR2 has been shown to activate Src through its interaction with the T cell-specific adapter (TSAd/VRAP) molecule (61,62). However in immunoprecipitation assays, VEGF-induced phosphorylation of Tyr-951 was unaffected by inhibition of NADPH oxidase (Fig. 7A). These results suggest that NADPH oxidasederived ROS modulate the activity of Src kinase (and thus, PI3K-Akt) independently of VEGFR2 Tyr-951 in endothelial cells.
Phosphorylation of the tyrosine residue, Tyr-1175, in the C-terminal region of VEGFR2 promotes activation of PLC␥ (63). Activated PLC␥, in turn, generates inositol trisphosphate and diacylglycerol by hydrolyzing phosphatidylinositol 4,5bisphosphate, a common and rate-limiting substrate for both PLC␥ and PI3K. In addition, Tyr-1175 provides a binding site for Shb, an adapter molecule that has been implicated in the activation of the PI3K-Akt signaling pathway in porcine aortic endothelial cells (63). Inhibition of NADPH oxidase by siRNA did not inhibit VEGF-mediated phosphorylation at Tyr-1175 (Fig. 7B), arguing against a role for this residue in mediating redox sensitivity of the PI3K-Akt signaling pathway.

Inhibition of NADPH Oxidase in Primary Human Endothelial Cells Has a Differential Effect on VEGF-mediated Gene
Transcription-We next wished to determine whether the existence of NADPH oxidase-dependent and -independent VEGF signaling pathways is associated with differential sensitivity of downstream target genes. HCAEC were transfected with scrambled siRNA (control) or si-p47 phox and then incubated in the absence or presence of 50 ng/ml VEGF for 4 h. Total RNA was extracted and processed for DNA microarray studies. The clusters were analyzed for a pool of VEGF-responsive genes that were inhibited by si-p47 phox . A total of 1486 genes were induced by VEGF in scrambled siRNA-treated HCAEC (LCB 1.5). Of these, 402 were blocked by si-p47 phox (at LCB 1.5; data not shown). The existence of distinct ROS-sensitive and -insensitive classes of VEGF-inducible genes was validated by real time-PCR of selected genes (Tables 1 and 2). As expected, VEGF treatment of HCAEC also resulted in reduced expression of certain genes (Table 3). In each case, VEGF-mediated gene repression was blocked by si-p47 phox (Table 3). Selected examples of these three gene classes were verified by Northern blot experiments and are shown in Fig. 8.

VEGF Induction of NADPH Oxidase-sensitive Genes Is Mediated by ROS-dependent Signaling Pathways-We asked
whether the sensitivity of VEGF-responsive genes to NADPH oxidase activity could be explained by the involvement of one or more of the ROS-dependent signaling pathways elucidated in this study. Indeed, VEGF induction of two NADPH oxidase-dependent genes, VCAM-1 and E-selectin, was significantly blocked by LY294002 (PI3K inhibitor) or SB203580 (p38 MAPK inhibitor) but not PD98059 (ERK1/2 inhibitor) (Fig. 9, A  and B). Similarly, VEGF-mediated inhibition of GADD45A was attenuated by si-p47 phox (see Table 3), LY294002, and SB203580 but not PD98059 (Fig. 9C). In contrast, expression of FIGURE 7. VEGF-induced phosphorylation at the tyrosine residues 951 and 1175 of VEGFR2 is not dependent on NADPH oxidase activity in HCAEC. Immunoprecipitation (IP) of HCAEC lysates by anti-VEGFR2/KDR antibody followed by immunoblots with anti-phosphotyrosine 951 (A) and 1175 (B) antibodies are shown. Scram-si-or si-p47 phox -transfected HCAEC were serum-starved overnight and then treated with VEGF (50 ng/ml) for the times indicated. Immunoprecipitation followed by Western blot analyses were carried out as described under "Experimental Procedures." The membrane was stripped and reprobed using anti-VEGFR2/KDR antibody (as control for loading).

Inhibition of Akt Phosphorylation in VEGF-treated NADPH Oxidase-depleted Endothelial Cells Is Sufficient for Modulating
Downstream Gene Expression-We have demonstrated previously that VEGF reduces mRNA expression of GADD45A in endothelial cells through a mechanism that involves phospho-Akt-mediated exclusion of forkhead from the nucleus (49). Consistent with the redox sensitivity of Akt, this effect was reversed by si-p47 phox , DN-Akt, or constitutively active triplemutant (TM) FKHR (Fig. 10). In contrast, expression of a CA-Akt alone or in the presence of si-p47 phox mimicked the effect of VEGF on GADD45A expression (Fig. 10). Data demonstrating a role for Akt-FKHR in mediating the expression of another redox-dependent gene, MnSOD, were obtained (data not shown). Taken together, these findings suggest that activation of the redox-sensitive signaling intermediates may modulate activity of downstream transcription factors, which in turn lead to alterations in target gene expression.

Inhibition of NADPH Oxidase in Primary Human Endothelial Cells Has a Differential Effect on VEGF-mediated Endothe-
lial Cell Function-VEGF has been shown previously to induce the expression of thrombomodulin and DAF in cultured endothelial cells (49,64,65). VEGF-mediated induction of thrombomodulin, but not DAF, was blocked by si-p47 phox (Tables 1 and  2 and Fig. 7). To determine whether differential redox sensitivity occurred at a functional level, we carried out in vitro assays for thrombomodulin and DAF activity. The thrombomodulin assay relies on its ability to activate exogenously supplied protein C, which in turn catalyzes a substrate S2366 (measured at 405 nM wavelength). Treatment of HCAEC with VEGF resulted

Scrambled
Signal transduction a p Ͻ 0.001 is relative to scrambled siRNA basal level. b p Ͻ 0.05 is relative to si-p47 phox -transfected uninduced basal level. c p Ͻ 0.05 is relative to scrambled siRNA basal level. d p Ͻ 0.05 is relative to scrambled siRNA-transfected, uninduced/basal level. e p Ͻ 0.05 is relative to scrambled siRNA-transfected, VEGF-treated level.

TABLE 3 NADPH oxidase-dependent genes that are repressed by VEGF in endothelial cells as determined by real time-PCR analyses (-fold reduction)
The basal levels of expression for each gene in scrambled siRNA-transfected, unstimulated, serum-starved HCAEC were arbitrarily set at 1 (-fold) per 10 6   in Ͼ2-fold induction of thrombomodulin activity, an effect that was abrogated by si-p47 phox (Fig. 11A), suggesting VEGF-mediated induction in the activity of thrombomodulin was NADPH oxidase-dependent. The DAF assay measures complement-mediated lysis of endothelial cells (49). Consistent with the gene expression data, VEGF-mediated protection against C3b complement was unaffected by si-p47 phox (Fig. 11B). These findings suggest that VEGF induction of thrombomodulin activity but not DAF is sensitive to NADPH oxidase in endothelial cells.

DISCUSSION
Vascular NADPH oxidase is believed to play an important role in both physiology and pathophysiology. NADPH oxidase activity varies between vascular cells and between different sites of the vasculature. For example, vascular smooth muscle cells express high levels of Nox1 and Nox4, whereas endothelial cells express high levels of gp91 phox /Nox2 and Nox4 (reviewed in Ref. 67). Cultured human microvascular endothelial cells display higher NADPH oxidase activity compared with HUVEC (68). Human saphenous veins were shown to express more Nox2 and p22 phox , whereas internal mammary arteries expressed relatively high Nox4 (69). Many extracellular factors have been reported to induce NADPH oxidase activity in cultured endothelial cells. The nature of input signals varies at different sites of the vascular tree and over time. Together, these observations underscore the remarkable complexity of NADPH oxidase signaling in the intact vasculature.
Adding to this complexity is our finding that NADPH oxidase activity selectively modulates downstream signaling induced by a single agonist, namely VEGF. Differential effects of NADPH oxidase inhibition on VEGF signaling were observed at the level of signal transduction, transcription factor activation, target gene expression, and cell function.
Previous studies have implicated a role for ROS in mediating reversible receptor autophosphorylation in response to such ligands as insulin, epidermal growth factor, and platelet-derived growth factor (70,71,72). In addition, angiotensin IImediated transactivation of the epidermal growth factor receptor and platelet-derived growth factor receptor in vascular smooth muscle cells was shown to be redox-sensitive (73,74). Not all protein-tyrosine kinase receptors are similarly sensitive to ROS. For example, the NADPH oxidase inhibitor diphenyleneiodonium (DPI) or apocynin had no effect on Ang1-mediated phosphorylation of Tie2 (35). In this study, we demonstrated that si-p47 phox -mediated knockdown of ROS had no effect on VEGFR2/KDR tyrosine phosphorylation at 1 and 5 min following VEGF treatment and only partially reduced phosphorylation at 10 min. Similar results were obtained with DPI (data not shown). Two previous studies have shown a role for ROS in autophosphorylation of VEGFR2/ KDR. In one case, pretreatment of porcine aortic endothelial cells with catalase or the antioxidant NDGA partially inhibited VEGF induction of VEGR2 phosphorylation between 10 and 15 min (23). These results are consistent with our 10-min data. In another study, preincubation of HUVEC with chemical inhibitors of ROS or transfection with antisense against gp91 phox resulted in partial attenuation of VEGFR2/KDR autophosphorylation at 5 min (24). The reason for the discrepancy between the results of the latter report and the current study is not clear. Nevertheless, in neither case did ROS inhibition cause complete loss of VEGFR2/KDR autophosphorylation. Our results demonstrating insensitivity of VEGFR2 to NADPH oxidase activity at early time points are in accordance with the findings of Berk and co-workers (75). Our findings also suggest that inhibition of VEGFR2/KDR phosphorylation at later time points (10 and 30 min) was insufficient to completely abrogate signaling (as evidenced by the existence of a subclass of NADPH oxidase-independent pathways). The observation that NADPH oxidase is not required for VEGF-mediated phosphorylation of two major phosphorylation sites (Tyr-951 and Y Tyr-1175) argues against a tyrosine residue-specific effect of ROS on VEGF autophosphorylation. Although it is formally possible that other sites within the receptor are differentially sensitive to the redox state, a more likely explanation for our findings is that early and transient activation of VEGFR2/KDR is adequate for propagating downstream signals. That VEGF activation of PI3K/Akt is redox-sensitive is consistent with previous findings. For example, in HUVEC, VEGFmediated phosphorylation of Akt was blocked by DPI, apocynin, and catalase (76). Studies of Ang1 have yielded conflicting results. Chen et al. (35) demonstrated that Ang1-dependent phosphorylation of Akt at Ser-473 in porcine and murine endo-thelial cells was dependent on NADPH oxidase-derived ROS. In another study, Ang1-mediated phosphorylation of Akt in HUVEC was unaffected by overexpression of superoxide dismutase, catalase, or dominant-negative Rac1 (Rac1N17) (36). We have found that hepatocyte growth factor-mediated activation of Akt in HCAEC occurs independently of NADPH oxidase activity (supplemental Fig. 5). These findings suggest that the redox sensitivity of PI3K/Akt in endothelial cells is ligandand receptor-dependent.
In time course experiments, siRNA-mediated inhibition of p47 phox attenuated VEGF activation of Akt prior to any observable effect on VEGFR2/KDR phosphorylation. These data suggest that NADPH oxidase-derived ROS exert their positive effects on Akt at site(s) distal to the VEGF receptor. We have shown si-p47 phox blocks VEGF-mediated phosphorylation and activity of Src. This finding is consistent with a previous investigation in which antioxidants attenuated VEGF-induced Src phosphorylation in HUVEC (77). Importantly, we also demonstrated that Src knockdown attenuates VEGF stimulation of Akt phosphorylation. Thus, NADPH oxidase influences Akt signaling via an effect on Src (Fig. 12). The precise mechanism underlying redox sensitivity of Src requires further study. FIGURE 10. Redox-sensitive signaling intermediate Akt modulates activity of the downstream transcription factor FKHR that in turn regulates gene expression. A, HCAEC were transduced with replication-deficient control (Adv), dominant-negative Akt (DN-Akt), or constitutively active myristoylated Gag-Akt (CA-Akt) expressing adenoviruses as described under "Experimental Procedures." Whole-cell lysates (20 g/lane) were subject to Western blot analysis, and the membrane was probed with anti-Akt antibody. Adenovirally expressed DN-Akt was of the same mobility as native Akt (indicated on the left side of the panel), whereas Gag-Akt showed a slower mobility (indicated on the right side of the panel). B, HCAEC were either transfected with siRNAs (Scram-si or si-p47 phox ) or transduced with adenoviruses (Adv, DN-Akt, or CA-Akt) as indicated. The cells were then serum-starved overnight and treated without (Ϫ) or with (ϩ) VEGF (50 ng/ml) for 15 min. Western blots were carried out using anti-phospho-Ser-256 FKHR (p-FKHR) antibody. The membrane was then stripped and reprobed using anti-␤-actin antibody (␤-actin) as loading control. C, Northern blot analyses demonstrate that VEGFmediated down-regulation of GADD45A was reversed by si-p47 phox , DN-Akt, or constitutively active triple mutant FKHR. On the contrary, expression of a constitutively active (CA) Akt mimicked VEGF effects and counteracted the effects of si-p47 phox on GADD45A expression. HCAEC were transfected with siRNA or transduced with adenoviruses as described in B. Cells were serumstarved overnight followed by VEGF treatment for 4 h. Total RNAs were extracted and subjected to Northern analysis as described under "Experimental Procedures." Lower panel shows ethidium bromide-stained 28 S as loading control. A, control adenoviruses; D, DN-Akt; C, CA-Akt; Si, si-p47 phox ; SC, si-p47 phox plus CA-Akt; SD, si-p47 phox plus DN-Akt; T, TM-FKHR. FIGURE 11. VEGF-mediated thrombomodulin-dependent activation of protein C is NADPH oxidase-dependent, but protection from complement-mediated endothelial cell lysis is not. A, activation of protein C was measured as a function of thrombomodulin activity as described under "Experimental Procedures." Scram-si or si-p47 phox -transfected HCAEC were incubated with VEGF for 14 h and assayed for thrombomodulin-dependent activation of protein C (aPC) using chromogenic substrate S-2366 at 405 nm wavelength. *, p Ͻ 0.01 (VEGF-treated versus untreated); †, p Ͻ 0.05 (si-p47 phox -transfected plus VEGF versus Scram-si-transfected plus VEGF). B, C3b complement-mediated lysis of HCAEC was assayed as described under "Experimental Procedures." Scram-si or si-p47 phox -transfected HCAEC were serum-starved overnight and treated with or without VEGF for 24 h prior to cell lysis assay using calcein acetoxymethyl ester. *, p Ͻ 0.05 (VEGF-treated versus untreated). The data are obtained from three independent experiments and analyzed as mean Ϯ S.D.
We have demonstrated a role for NADPH oxidase activity in mediating the effect of VEGF on p38 MAPK but not ERK1/2 or JNK. These data contrast with those of Wu et al. (78) who showed that VEGF-mediated phosphorylation of JNK was oxidasedependent. Moreover, previous studies in HUVEC have demonstrated a role for ROS in VEGF-mediated activation of ERK1/2 in HUVEC and porcine aortic endothelial cells (23,36). The reason for the discrepancies between these latter reports and the present study is not clear. However, our findings that both ASoligonucleotides and siRNA against p47 phox had no effect on ERK1/2 or JNK phosphorylation in HUVEC and HCAEC in multiple independent time course experiments strongly argue against a role for NADPH oxidase-derived ROS in VEGF activation of these signal intermediates.
It is interesting to note that similar discrepancies exist regarding the role for ROS in Ang1-mediated MAPK signaling. One group demonstrated that overexpression of superoxide dismutase or Rac1N17 in HUVEC resulted in potentiation of Ang1-mediated phosphorylation of ERK1/2 but not p38 MAPK (36). Chen et al. (35), employing porcine coronary artery endothelial cells, as well as wild type and p47 phox mouse heart microvascular endothelial cells, demonstrated a role for NADPH oxidase in Ang1-stimulated phosphorylation of ERK1/2. These discrepancies notwithstanding, the effect of ROS on the MAPK family appears to be dependent on the agonist and cell type. For example, whereas the current results argue against a role for NADPH oxidase in mediating VEGF activation of ERK1/2 in endothelial cells, previous studies support a role for the enzyme in angiotensin II-mediated activation of ERK1/2 in endothelial cells (79). Urotensin-II-mediated activation of ERK1/2, p38 MAPK, and JNK was inhibited by DPI and antisense to p22 phox in human pulmonary artery smooth muscle cells (80). Thrombin-mediated activation of p38 MAPK but not ERK1/2 in vascular smooth muscles cells was p22 phox -depend-ent (81). Endothelin-1-mediated activation of JNK, but not ERK1/2, in vascular smooth muscle cells was inhibited by DPI (82). In murine cardiac microvascular endothelial cells, hypoxia-reoxygenation-mediated phosphorylation of ERK1/2 and Akt was NADPH oxidase-dependent (83). Another study demonstrated a critical role for p47 phox in tumor necrosis factor-␣mediated activation of ERK1/2 and p38 MAPK in cardiac microvascular endothelial cells (66).
Previously, we have shown that NADPH oxidase-derived ROS play an important role in mediating the effect of VEGF on the activity of the transcription factors, NF-B and forkhead (44). The current study extends these observations by demonstrating an inhibitory action of si-p47 phox on VEGF-mediated nuclear exclusion of FKHR. This study has also established a link between the relative redox sensitivity of the VEGF signaling intermediates (e.g. PI3K-Akt, p38 MAPK/ERK1/2) and the differential effects of NADPH oxidase-derived ROS on the expression of several VEGF-dependent genes. The data also supported the notion that redox-dependent alterations of a signaling molecule (e.g. Akt) by VEGF may result in the modulation of the activity of a downstream transcription factor(s) (e.g. FKHR), which in turn leads to modification of gene expression (e.g. GADD45A). Additionally, based on the results of the DNA microarrays, we predict the existence of a class of transcription factors whose activity is not sensitive to NADPH oxidase activity. Identification of those ROS-insensitive factors requires further studies.
Prior studies from our own group (21,22), as well as others (23,24,43,76), have demonstrated an important role for NADPH oxidase in mediating the effect of VEGF on endothelial cell migration and proliferation and angiogenesis. Here, we have shown that NADPH oxidase is also necessary for VEGFmediated induction of thrombomodulin activity. However, inhibition of NADPH oxidase had no effect on VEGF-stimulated DAF function. These findings are in accordance with the differential effect of NADPH oxidase on downstream signaling pathways, transcription factors, and gene expression, and they provide compelling evidence for the existence of NADPH oxidase-dependent and -independent VEGF functions.
The finding that NADPH oxidase is necessary for some, but not all, functions of VEGF have important biological and therapeutic implications. The data suggest that basal and/or inducible ROS are not required for global VEGF signaling but rather serve as signal intermediates in selected downstream pathways. In this way, alteration of NADPH oxidase activity either in response to agonists or drugs will have highly selective effects on VEGF signal transduction and endothelial cell phenotypes. Such an effect may be leveraged for therapeutic gain. For example, it is plausible that the therapeutic inhibition of NADPH oxidase would inhibit VEGF stimulation of migration and proliferation, while retaining certain protective effects of VEGF signaling, e.g. protection against complement lysis.
Acknowledgment-We are grateful to Lewis C. Cantley for helpful discussions.