Hypoxia‐inducible factor‐3α promotes angiogenic activity of pulmonary endothelial cells by repressing the expression of the VE‐cadherin gene

Abstract

The variants of the hypoxia‐inducible factor‐3α gene HIF‐3α and NEPAS are known to repress the transcriptional activities driven by HIF‐1α and HIF‐2α. Although NEPAS has been shown to play an important role in vascular remodeling during lung development, little is known about the roles of HIF‐3α in adult lung function. Here, we examined pulmonary endothelial cells (ECs) isolated from wild‐type (WT) and HIF‐3α functional knockout (KO) mice. The expression levels of angiogenic factors (Flk1, Ang2 and Tie2) were significantly greater in the HIF‐3α KO ECs than those in the WT ECs irrespective of oxygen tension. However, the HIF‐3α KO ECs showed impaired proliferative and angiogenic activities. The impaired EC function was likely due to the excess vascular endothelial (VE)‐cadherin, an inhibitor of Flk1/PI3 kinase/Akt signaling, as treatment of the cells to a neutralizing antibody partly restored the phenotype of the HIF‐3α KO ECs. Importantly, we found that the mRNA levels of HIF‐2α and Ets‐1 were significantly increased by HIF‐3α ablation. Given that both factors are known to activate the VE‐cadherin gene, the transcriptional repression of these factors by HIF‐3α might be important for silencing the irrelevant expression of the VE‐cadherin gene. Collectively, these data show novel and unique roles of HIF‐3α for angiogenic gene regulation in pulmonary ECs.

Introduction

Hypoxia‐inducible factors (HIFs) are central transcriptional factors, which respond to low oxygen conditions (hypoxia) and regulate the expression of target genes related to cell proliferation, metabolism, migration and survival to overcome the disruption of oxygen homeostasis (reviewed in Semenza 2009; Majmundar et al. 2010). HIFs consist of two subunits: oxygen‐dependent HIF‐α and constitutively expressed HIF‐β (also called Arnt [aryl hydrocarbon receptor nuclear translocator]). Each subunit contains a basic helix–loop–helix–PAS (bHLH‐PAS) domain, which is necessary for heterodimerization and DNA binding (Wang et al. 1995; Jiang et al. 1996). Under well‐oxygenated conditions (normoxia), the HIF‐α protein is hydroxylated by prolyl hydroxylase domain proteins (PHDs), ubiquitinated by the von Hipple‐Lindau (VHL) protein and then degraded by the 26S proteasome (Huang et al. 1998; Epstein et al. 2001). As O₂ is required as a substrate for hydroxylation catalyzed by PHDs, the series of reactions is inhibited by substrate reduction during hypoxic stress. Thus, the HIF‐α protein is stabilized and dimerizes with Arnt, leading to the transcriptional activation of target genes via the hypoxic responsive element (HRE; Manalo et al. 2005; Wenger et al. 2005).

Three HIF‐α genes (HIF‐1α, HIF‐2α and HIF‐3α) have been identified so far (Wang et al. 1995; Tian et al. 1997; Gu et al. 1998). In addition, three splice variants are known to exist for the Hif3α gene, HIF‐3α, neonatal and embryonic PAS (NEPAS; Yamashita et al. 2008) and inhibitory PAS (IPAS; Makino et al. 2001). The expression of these splice variants is dependent on the oxygen conditions and developmental stage in mice. HIF‐1α and HIF‐2α possess both a C‐terminal transactivation domain (CTAD) and N‐terminal transactivation domain (NTAD), leading to potent transcriptional activity. In contrast, IPAS is known to be a negative regulator of HIF‐1α and HIF‐2α, as this factor lacks both the CTAD and NTAD and interacts with HIF‐1α or HIF‐2α, but not with Arnt (Makino et al. 2001, 2007). The transcriptional activities of HIF‐3α and NEPAS are lower than those of HIF‐1α and HIF‐2α as they lack the CTAD (Gu et al. 1998; Yamashita et al. 2008). As a consequence, HIF‐3α can suppress HIF‐driven transcriptional activities by competing for interactions between HIF‐1α or HIF‐2α and Arnt (the competitive model; Hara et al. 2001). The inhibitory and suppressive effects of these HIF‐3α variants against HIF‐1α or HIF‐2α have been shown by several in vitro studies; however, little is known about a role of HIF‐3α in cellular processes related to physiological and pathological phenomena.

Among of HIF‐3α variants, IPAS is predominantly expressed in the brain and eyes in mice. Interestingly, the expression of IPAS in the cornea is inversely correlated with that of vascular endothelial growth factor‐A (VEGF‐A), suggesting that IPAS is a negative regulator of angiogenesis, and may be involved in the maintenance of avascular tissue (Makino et al. 2001). Moreover, the hypoxia‐dependent expression of IPAS has been observed in heart and lung tissues, implying that IPAS might participate in a negative feedback loop with HIFs (Makino et al. 2007, 2002). In a previous report, we have generated HIF‐3α functional knockout (KO) mice in which GFP and a neomycin resistance gene cassette were inserted into the second exon of the Hif3α gene (Yamashita et al. 2008). We found that NEPAS was predominantly expressed at the neonatal and embryonic stages in the lungs, and HIF‐3α functional KO mice showed enlargement of the right ventricular and impaired lung remodeling during development (Yamashita et al. 2008). These data show that NEPAS plays an important role in pulmonary vascular remodeling. Furthermore, an increased level of endothelin‐1 (ET‐1) was observed in ECs isolated from the neonatal lungs of HIF‐3α KO mice, suggesting that the adequate suppression of ET‐1 by NEPAS is required for normal lung development. Another group reported that pulmonary epithelial cell‐specific HIF‐3α transgenic mice showed abnormal alveolarization (Huang et al. 2013). These observations led us to speculate that HIF‐3α also plays a critical role in adult lung function.

To clarify the role of HIF‐3α in the adult stage, we investigated pulmonary ECs isolated from HIF‐3α functional KO and WT adult mice. In the WT pulmonary ECs, the expression of known hypoxia‐inducible angiogenic genes, such as Flk1, Ang2 and Tie2, was very low and was not induced by hypoxia. We herein provide evidence that HIF‐3α tightly represses the expression of these genes, independent of the oxygen tension. Interestingly, despite the fact that the expression of angiogenic genes was increased, the proliferative and angiogenic activity was impaired, rather than enhanced, in the HIF‐3α KO ECs. We found that the increased expression of VE‐cadherin, which is known to inhibit the intracellular Akt activation required for VEGF/Flk1 signaling, was responsible for the phenotype. Importantly, our data show that HIF‐3α represses the expression of HIF‐2α and Ets‐1, and oxygen‐dependent and oxygen‐independent transcription factors, respectively, at the transcriptional level. Collectively, these results show a unique and novel role for HIF‐3α in regulating the angiogenic properties of pulmonary ECs.

Results

Identification and isolation of HIF‐3α‐expressing endothelial cells in the adult mouse lung

HIF‐3α functional KO mice were generated by inserting GFP into the Hif3α locus, so that the expression profile of this gene could be monitored by evaluating the expression of GFP. In the neonatal lungs, the GFP expression recapitulates that of NEPAS, which is predominantly expressed at the neonatal and embryonic stages. We previously reported that the NEPAS–GFP signals overlapped with the endothelial marker, CD31, and the type II alveolar marker, SP‐D, expression (Yamashita et al. 2008). To identify HIF‐3α‐expressing cells in adult lungs, serial sections of lungs prepared from HIF‐3α KO mice were immunohistochemically stained using antibodies against CD31 or alpha smooth muscle actin (Fig. 1A). Consistent with our previous observations at the neonatal stage, the GFP signals overlapped with the CD31‐positive cells in the adult lungs as well (Fig. 1A, right). In addition, the expression of alpha smooth muscle actin was co‐localized with the GFP signals (Fig. 1A, left). Taken together, these results indicate that HIF‐3α is expressed in both ECs and smooth muscle cells of the adult lung.

Figure 1

Identification and isolation of HIF‐3α‐expressing endothelial cells in the adult mouse lung. (A) Serial sections of HIF‐3α^+/− lungs were immunostained using anti‐GFP, anti‐CD31 and anti‐alpha smooth muscle actin (αSMA) antibodies. Isotype IgG was used as a negative control. Co‐localization of the GFP signal and CD31 or SMαA signals was observed. The bar indicates 100 μm. (B) FACS sorting of pulmonary ECs derived from WT (left top) and HIF‐3α KO (left bottom) mice using anti‐CD45 and CD31 antibodies. GFP‐positive cells were selected in the CD31⁺ or CD45⁻ fractions of HIF‐3α KO lungs. The morphologies of these cells are shown in the right upper panel (Scale bar = 100 μm). (C) The results of a RT‐PCR analysis of cell lineage markers. Both cell types expressed the endothelial marker, but not the smooth muscle marker. (D) The mRNA expression levels of HIF‐3α and IPAS under normoxic (20% O₂) or hypoxic (1% O₂) conditions in the WT ECs were determined by a RT‐PCR analysis. Note that while HIF‐3α was expressed constitutively, IPAS was not expressed in the pulmonary ECs derived from WT mice. The density of bands was measured using the image j free software program and was calculated as the mean ± SD (right). (E) The results of an immunoblot analysis of the WT ECs showing HIF‐3α and the internal control, Lamin B. HIF‐3α was expressed under normoxic and hypoxic conditions in the WT ECs.

We previously reported that pulmonary vascular remodeling failure was occasionally observed in HIF‐3α KO mice, thus suggesting the possibility that HIF‐3α plays a specific role in the vascular remodeling of pulmonary ECs (Yamashita et al. 2008). To test this possibility, we first prepared immortalized pulmonary cells by cross‐breeding WT or HIF‐3α KO mice with transgenic mice bearing the temperature‐sensitive SV40‐T‐antigen (tsSV40‐T; Yanai et al. 1991), and pulmonary cells from the compound mice were analyzed by FACS. To isolate EC fraction, the CD45‐negative and CD31‐positive cells were collected from the WT lung. From the HIF‐3α KO lung, CD31‐positive, CD45‐negative and GFP‐positive cells were collected as the EC fraction. The HIF‐3α KO and WT ECs were similar in their morphology, and both cells expressed endothelial‐specific genes, such as CD31 and VE‐cadherin (Fig. 1B,C). The CD31 expression level was significantly increased in the HIF‐3α KO ECs compared with that in the WT ECs (Fig. S1 in Supporting Information). Smooth muscle cell markers such as calponin1 and alpha smooth muscle actin were undetectable in these cells by semi‐quantitative RT‐PCR (Fig. 1C). Importantly, the expression level of HIF‐3α in the WT ECs was similar under normoxic (20% O₂) and hypoxic (1% O₂) conditions (Fig. 1D,E), thus suggesting that HIF‐3α is constitutively expressed in pulmonary ECs.

HIF‐3α dysfunction results in impaired proliferative activity and in vitro angiogenic potential in pulmonary ECs

Although the inhibitory effects of HIF‐3α on HIF‐driven transcriptional activities have been shown in several reports (Hara et al. 2001; Yamashita et al. 2008), the physiological and pathological roles of this factor remain unclear. In ECs, both HIF‐1α and HIF‐2α are expressed and regulate their appropriate target genes to activate cell survival, proliferation and angiogenic potential (Yamakawa et al. 2003; Yu et al. 2004; Manalo et al. 2005; Krotova et al. 2010). Therefore, we considered the possibility that the HIF‐3α dysfunction led to enhanced functions of HIF‐1α and HIF‐2α, and as a consequence, the HIF‐3α KO pulmonary ECs possessed greater proliferative and angiogenic potential than the WT ECs. To test this possibility, the cell proliferation assay was carried out by counting cells daily under normoxic (20% O₂) and hypoxic (5% O₂) conditions. Contrary to our expectations, the HIF‐3α KO ECs grew more slowly than the WT ECs under both hypoxic and normoxic conditions (Fig. 2A). The doubling time of the HIF‐3α KO ECs was significantly longer than that of the WT ECs (normoxic conditions: 40.4 ± 1.8 and 26.3 ± 0.5 h for HIF‐3α KO and WT ECs, respectively, P < 0.001; hypoxic conditions: 43.4 ± 1.7 and 29.6 ± 2.3 h for HIF‐3α KO and WT ECs, respectively, P < 0.01).

Figure 2

HIF‐3α dysfunction results in impaired proliferative activity and in vitro angiogenic potential in pulmonary ECs. (A) Pulmonary ECs were cultured under 20% O₂ (Normoxia) or 5% O₂ (Hypoxia) conditions and were harvested every day to count the cell number. The average cell number was determined (mean ± SD) for triplicate samples. The white circles indicate the WT ECs, and the black triangles indicate the HIF‐3α KO ECs. (B) The WT and HIF‐3α KO ECs were seeded on Matrigel and incubated under normoxic (N) or hypoxic (H) conditions. After 9 h, the tubes formed on Matrigel were counted in each field, and the mean ± SD value was obtained from triplicate wells (right; *P < 0.05, ***P < 0.001, n = 3). The white bars and black bars in the right graph indicate normoxic and hypoxic conditions, respectively. Bar = 500 μm. (C) The migration distance of each type of cell was determined by an in vitro wound scratch assay. Confluent monolayers were scratch‐wounded and incubated for up to 24 h. Photographs of the wounded monolayers taken at 0 and 24 h are shown (left). The migration distances are indicated by arrows and were measured. The white bars and black bars in the right graph indicate normoxic and hypoxic conditions, respectively. The white bars and black bars in the right graph indicate normoxic and hypoxic conditions, respectively. The values represent the means ± SD of four measurements (right; **P < 0.01, ***P < 0.001).

The in vitro Matrigel tube formation assay, which is frequently used to assess the angiogenic functions of ECs, showed a decreased number of tubes formed by HIF‐3α KO ECs compared with WT ECs (Fig. 2B). Again, these abnormalities were observed irrespective of the oxygen concentration.

We next examined the cell migration activity, which is an important process in angiogenesis, by the in vitro scratch assay (Fig. 2C). Confluent monolayers were wounded with a rubber cell scraper, and the migration distance was assessed (Fig. 2C). At 24 h, the migration distance of the HIF‐3α KO ECs was significantly shorter than that of the WT ECs under both oxygen conditions. In summary, contrary to our expectations, HIF‐3α disruption led to inhibition, rather than activation, of the known functions of HIF‐1α and HIF‐2α in ECs. In addition, it is of note that these effects were observed under both normoxic and hypoxic conditions.

Defective EC functions are restored by expressing HIF‐3α cDNA in HIF‐3α KO ECs

Among the three Hif3a gene products, HIF‐3α and IPAS are expressed in adult mice, whereas the expression of NEPAS is restricted to the neonatal stage (Yamashita et al. 2008). The RT‐PCR analysis using specific primers for HIF‐3α and IPAS showed that HIF‐3α is the major product of the Hif3a gene in pulmonary ECs under both normoxic and hypoxic conditions (Fig. 1D). In addition, we examined whether the abnormalities observed in the HIF‐3α KO ECs were restored by expressing full‐length HIF‐3α cDNA (i.e. ‘HIF‐3α rescued cells’). The HIF‐3α protein expression in the HIF‐3α rescued cells was confirmed by an immunoblot analysis (Fig. 3A). As expected, the cell proliferation assays showed that the growth rate of the HIF‐3α rescued KO ECs (KO + H3) was greater than that of the mock‐treated ECs (KO + mock) and was comparable to that of the WT ECs (Fig. 3B). The doubling time of the HIF‐3α rescued cells was significantly shortened compared with that of KO + mock ECs (28.6 ± 0.6 and 24.8 ± 0.4 h for KO + mock and KO + H3 ECs, respectively, P < 0.001). Furthermore, tube formation assays showed that the numbers of tubes formed by KO + H3 ECs were similar to that of the WT ECs under both oxygen conditions (Fig. 3C). The cell migration distance was also recovered by the HIF‐3α restoration (Fig. 3D). Taken together, these results indicate that the attenuated cell proliferation and angiogenic ability were caused by the loss of HIF‐3α in these cells.

Figure 3

Defective EC functions are restored by expressing HIF‐3α cDNA in HIF‐3α KO ECs. A HIF‐3α full‐length cDNA cassette was transfected into HIF‐3α KO ECs using a retrovirus vector (indicated as ‘KO + H3’ or ‘H3’ in subsequent figures). (A) The expression of HIF‐3α in the HIF‐3α‐transfected ECs was confirmed by an immunoblot analysis. (B) The results of the analysis of cell proliferation using WT ECs, HIF‐3α KO ECs, KO ECs transfected with empty vector (KO + mock) and KO + H3 under normoxic conditions. The proliferation rate of the KO + H3 cells was faster than that of the KO + mock cells and comparable to that of the WT ECs. (C) The tube formation assay was carried out with the cells described above (bar = 500 μm). The number of tubes on Matrigel was scored as the mean ± SD (right) (***P < 0.001). The white bars and black bars in the right graph indicate normoxic and hypoxic conditions, respectively. The gene transfer of HIF‐3α increased the number of tubes formed. (D) The effects of HIF‐3α gene transfection on cell migration were examined by the wound migration assay as previously above. The HIF‐3α rescued KO ECs migrated faster than the HIF‐3α KO ECs (right; ***P < 0.001).

Multiple genes related to vascular homeostasis are up‐regulated in HIF‐3α KO ECs

To identify the HIF target gene(s) responsible for the abnormalities in the HIF‐3α KO ECs, we examined the expression of the known HIF target genes by qRT‐PCR using four cell lines described above (WT, KO, KO + mock and KO + H3 cells, Fig. 4A). The expression of VEGF‐A was up‐regulated by hypoxia in all four cell lines, thus indicating that these cells were able to respond to low oxygen tension. We next examined the expression of other HIF target genes, such as VEGFR2 (Flk1), angiopoietin2 (Ang2) and its receptor, Tie2, and vascular endothelial‐cadherin (VE‐cadherin). Surprisingly, we found that the expression levels of these genes were very low and were not induced under hypoxic conditions in the WT cells. Furthermore, the expression of these factors was significantly up‐regulated in the HIF‐3α KO ECs, and the increase was diminished in the HIF‐3α rescued cells (Fig. 4A). These results suggest that the expression of these genes was repressed by HIF‐3α in the WT cells. VEGF‐A was the only exception, as its expression was induced under hypoxic conditions in both WT and HIF‐3α KO cells. It is possible that the expression of VEGF‐A and other HIF target genes examined in this study might be regulated in a distinct manner in pulmonary ECs. In agreement with the qRT‐PCR data, an immunoblot analysis confirmed that the protein levels of Flk1, Ang2, Tie2 and VE‐cadherin were significantly enhanced in the HIF‐3α KO ECs, regardless of oxygen tension, but this was not the case in the HIF‐3α rescued cells (Fig. 4B).

Figure 4

Multiple genes related to vascular homeostasis are up‐regulated in HIF‐3α KO ECs. (A) The mRNA expression levels of angiogenic factors, including VEGF‐A, Flk1, Ang2, Tie2 and VE‐cadherin, were measured by quantitative real‐time PCR. The white bars indicate normoxic data; the black bars indicate hypoxic data. All data are shown as the means ± SD (*P < 0.05, **P < 0.01, ***P < 0.001, n = 3). The white bars and black bars indicate normoxic and hypoxic conditions, respectively. (B) The protein levels of Flk1, Ang2, Tie2 and VE‐cadherin in the pulmonary ECs were confirmed by immunoblotting. Actin was used as internal control. Consistent with the results of quantitative real‐time PCR, all of the factors tested were significantly increased by HIF‐3α disruption and were reduced by forced expression of HIF‐3α. (C) The effects of enzyme inhibitors in pulmonary ECs were examined by the tube formation assay. The WT ECs were plated onto Matrigel and treated with inhibitors for 9 h. The inhibitors were added at the concentrations described below and are indicated by the following abbreviations: LY, PI3‐kinase inhibitor LY294002 (20 μm); Tie2 inhibitor, Tie2 kinase inhibitor (250 nm); SU, VEGFR2 kinase inhibitor SU5416 (20 μm); and PD, Erk inhibitor PD98059 (20 μm). A reduction of tube formation by inhibiting PI3 kinase and Flk1 phosphorylation was observed in the WT ECs, suggesting that the Flk1–PI3 kinase–Akt pathway was critical for pulmonary ECs to form capillary‐like tubes. (D) The effects of VEGF on the phospho‐Akt in the WT ECs and HIF‐3α KO ECs were examined by the immunoblot analysis. After the addition of VEGF‐A (50 ng/mL), the cells were harvested as indicated time points and analyzed. The HIF‐3α KO ECs did not induce the phosphorylation of Akt, whereas the phosphorylated Akt level was apparently increased in the WT ECs at 10 min after the treatment. (E) The levels of phosphorylated Akt and Erk1/2 were determined by immunoblotting. Note that despite the increased expression levels of angiogenic factors, the phospho‐Akt level was decreased in the HIF‐3α KO ECs. Representative photographs from three independent experiments are shown.

The VEGF/Flk1 and angiopoietin/Tie2 signaling pathways are critical for the maintenance of EC homeostasis (Scharpfenecker et al. 2005; Daly et al. 2006; Lee et al. 2007). Despite the fact that the expression of Flk1, Ang2 and Tie2 was up‐regulated, we did observe impaired cell proliferation, tube formation or migration activity in the HIF‐3α KO ECs (Fig. 2). One possible explanation for these inconsistent results is that the intracellular signaling from these angiogenic factors is attenuated in the HIF‐3α KO ECs. The activation of both Flk1 and Tie2 leads to the phosphorylation of Akt and MAP kinases, such as Erk1/2 (Olsson et al. 2006; Augustin et al. 2009), and promotes cell proliferation, survival and motility during angiogenic processes. We examined which signaling cascade was the major pathway required for pulmonary EC functions. To this end, WT ECs were examined by the tube formation assay using specific inhibitors of PI3 kinase (LY294002), Tie2 (Tie 2 inhibitor), VEGFR2 kinase (SU5416) and Erk (PD98059; Fig. 4C). Although no significant difference was observed in the presence of the Tie2 inhibitor or PD98059 in the tube numbers formed on Matrigel, the treatment with LY294002 or SU5416 significantly attenuated tube formation. These results suggest that the Flk1–PI3 kinase–Akt pathway was dominantly used for tube formation in pulmonary ECs. As shown in Fig. 4A, the mRNA expression level of VEGF‐A in the HIF‐3α KO ECs was lower than that of the WT ECs under both normoxic and hypoxic conditions. Thus, we considered that the impaired angiogenic properties of the HIF‐3α KO ECs might be caused by the decreased VEGF‐A expression. Importantly, however, the exogenous treatment of VEGF‐A increased the phospho‐Akt level in the WT ECs from 10 min after the treatment, whereas the phosphorylation of Akt was not observed in the HIF‐3α KO ECs even in the presence of exogenous VEGF‐A protein (Fig. 4D). These results indicate that the downstream signaling of VEGF‐A is impaired in the HIF‐3α KO ECs. An immunoblot analysis showed that the phosphorylated forms of both Akt and Erk1/2 were detectable in the WT cells (Fig. 4E). In contrast, the phospho‐Akt level was decreased even though the expression of Flk1 was increased, whereas the phospho‐Erk1/2 level was unaffected in the HIF‐3α KO ECs (Fig. 4E). The attenuated phospho‐Akt in the HIF‐3α KO ECs was observed under both oxygen tensions (Fig. S2 in Supporting Information). Collectively, these results indicate that the impaired angiogenic properties, at least tube formation of the HIF‐3α KO ECs, are likely caused by the impaired intracellular Flk1–PI3 kinase–Akt pathway rather than the decreased mRNA of VEGF‐A.

VE‐cadherin is responsible for the reduced proliferative and angiogenic ability of HIF‐3α KO ECs

We have shown that the Akt phosphorylation in response to the exogenous VEGF‐A stimulation was attenuated in the HIF‐3α KO ECs, and the increased expression of Flk1 did not promote Akt activation in the HIF‐3α KO ECs. That is, the angiogenic signaling from extracellular ligands and cell surface receptors was somehow ‘uncoupled’ to the activity of the intracellular molecule, Akt. To explore the mechanisms underlying these observations, we focused on examining the function of VE‐cadherin, whose expression is up‐regulated in the HIF‐3α KO ECs (Fig. 4A,B). VE‐cadherin is an endothelial‐specific adhesion molecule, which is located on cell–cell junctions between ECs (Matsuyoshi et al. 1997). VE‐cadherin‐mediated adhesion is essential for maintaining the vascular permeability and angiogenesis (reviewed in Vestweber 2008; Giannotta et al. 2013). However, VE‐cadherin possesses inhibitory effects on cell proliferation, migration and sprouting via modulation of the VEGF/Flk1 signaling pathways (Lampugnani et al. 2002, 2006; Abraham et al. 2009; Pirotte et al. 2011). Therefore, we suggested that the excess VE‐cadherin expression could break the signaling from VEGF/Flk1 and thus led to the inhibition of Akt phosphorylation in HIF‐3α KO ECs.

To prove this hypothesis, we investigated whether a VE‐cadherin neutralizing antibody (BV9) (Corada et al. 2001) could restore the cell proliferative and capillary‐forming ability of HIF‐3α KO ECs. As expected, the treatment with BV9 significantly increased the number of the HIF‐3α KO ECs on day 7 (Fig. 5A). Similarly, the tube formation and migration activity were both restored by blocking the VE‐cadherin function (Fig. 5B,C). To confirm whether the expression of VE‐cadherin was up‐regulated in HIF‐3α KO mice, the immunohistochemical staining was carried out. As shown in Fig. 5D, the expression of VE‐cadherin in the WT lung was restricted in a subset of ECs (arrows). However, the positive staining for VE‐cadherin was widely and more robustly observed in the HIF‐3α KO lung compared with that of the WT lung (Fig. 5D). These results indicate that the increased expression of VE‐cadherin is likely involved in the impaired proliferative and angiogenic activity of HIF‐3α KO ECs and in the structural abnormality of pulmonary capillaries in vivo.

Figure 5

VE‐cadherin is responsible for the reduced proliferation and angiogenic ability of HIF‐3α KO ECs. (A) The HIF‐3α KO ECs were plated at 2 × 10⁴ cells/dish and were treated with mouse IgG (50 ng/mL) or a VE‐cadherin neutralizing antibody, BV9 (50 ng/mL). The number of cells was determined on days 3, 5 and 7 by counting the cells with a hemocytometer. On day 7, the number of cells treated with the anti‐VE‐cadherin antibody (anti‐VE‐cad Ab) was significantly increased compared with the control cells. (B) The HIF‐3α KO ECs were plated onto Matrigel and incubated with an anti‐VE‐cad Ab (50 ng/mL). After nine hours, the number of tubes formed on Matrigel was counted (bar = 500 μm). The treatment with the anti‐VE‐cad antibody accelerated the formation of capillary‐like tubes by HIF‐3α KO ECs. (C) After the scratch wounding of monolayer cultures, the HIF‐3α KO ECs were incubated with mouse IgG or the anti‐VE‐cad Ab for the indicated times. The HIF‐3α KO ECs migrated faster after treatment with the anti‐VE‐cadherin antibody. The values represented the mean ± SD of five measurements from three independent wounds. In all experiments shown in figure, the white bars indicate the control (HIF‐3α KO ECs treated with mouse IgG) and the black bars indicate HIF‐3α KO ECs treated with BV9. The data are shown as the means ± SD (*P < 0.05, ***P < 0.001). (D) The immunohistochemical staining against VE‐cadherin in the lung tissues was carried out. Whereas a few and weak staining was observed in the WT lung (indicated with the arrows), the positive staining of VE‐cadherin was robustly detected in the pulmonary capillary‐like structure of the HIF‐3α KO lung. Scale bar = 200 μm.

HIF‐3α negatively regulates the expression of HIF‐2α, VE‐cadherin and Ets‐1 at the transcriptional level

Finally, we investigated the molecular mechanism underlying the regulation of angiogenic gene expression by HIF‐3α in pulmonary ECs. First, the expression of the HIF‐1α and HIF‐2α protein was determined by an immunoblot analysis using nuclear extracts. To our surprise, the HIF‐2α protein was barely detectable in the WT cells, but was clearly detectable in the HIF‐3α KO ECs under hypoxic conditions (Fig. 6A). In contrast, the HIF‐1α protein was induced under hypoxic conditions, and its expression levels were similar in all cell types (Fig. 6A). Furthermore, the HIF‐2α mRNA level was also very low in the WT ECs and was significantly increased by HIF‐3α ablation, as shown by qRT‐PCR (Fig. 6B). In addition, the increased HIF‐2α mRNA level was clearly restored in the HIF‐3α rescued cells (Fig. 6B). These results indicate that the expression of HIF‐2α, but not HIF‐1α, is negatively regulated at the transcriptional level by HIF‐3α.

Figure 6

HIF‐3α negatively regulates the expression of HIF‐2α, VE‐cadherin and Ets‐1 at the transcriptional level. (A) The protein levels of HIF‐1α and HIF‐2α were determined by an immunoblot analysis. The nuclear extracts prepared from each cell type stimulated under normoxia (20% O₂) or hypoxia (1% O₂ for 6 h) were used. The density of bands was measured by an imaging analyzer (image j) to evaluate the relative expression levels. The data are shown as the means ± SD (*P < 0.05, **P < 0.01, n = 3). The white and black bars indicate normoxic and hypoxic conditions, respectively. The hypoxic induction of HIF‐2α was accelerated by the dysfunction of HIF‐3α. (B) The quantitative real‐time PCR analysis was carried out to examine the mRNA expression levels of HIF‐1α and HIF‐2α. The HIF‐2α expression was markedly increased in the HIF‐3α KO ECs, whereas there were no significant differences in the HIF‐1α levels between WT ECs and HIF‐3α KO ECs. The data are shown as the means ± SD (***P < 0.001, n = 3). The white bars and black bars indicate normoxic and hypoxic conditions, respectively. (C) A chromatin immunoprecipitation (ChIP) assay was carried out for the VE‐cadherin promoter's putative HRE using an anti‐HIF‐1α antibody (left panel) and anti‐HIF‐2α antibody (right panel) in WT and HIF‐3α KO ECs. Input samples were used as an internal control; IgG was used as a negative control. Both HIF‐1α and HIF‐2α were associated with the putative HRE in VE‐cadherin in the HIF‐3α KO ECs. (D) The ChIP assay in WT ECs was carried out using an anti‐HIF‐3α antibody. In the WT ECs, HIF‐3α bound to the putative HRE in VE‐cadherin under both normoxic and hypoxic conditions. (E) The luciferase assay was carried out with the pGL4.10 −2486 to +25 VE‐cadherin promoter reporter gene. Expression vectors for HIF‐1α and HIF‐2α were transfected into 293T cells concomitant with/without the HIF‐3α expression vector. HIF‐1α and HIF‐2α activated the VE‐cadherin reporter gene. However, co‐expression of HIF‐3α slightly, but significantly, reduced the promoter activities of HIF‐1α or HIF‐2α. The relative luciferase unit values are presented as the means ± SD (***P < 0.001, n = 3). (F) The expression of Ets‐1 was determined by qPCR and an immunoblot analysis. The dysfunction of HIF‐3α led to an increase in Ets‐1 expression under normoxia, as well as the known angiogenic factors (*P < 0.05, **P < 0.01, n = 3). The white bars and black bars indicate normoxic and hypoxic conditions, respectively. A representative image of an immunoblot is shown in the right panel. HIF‐3α might be involved in the regulation of Ets‐1. (G), (H) The ChIP assay with anti‐HIF‐3α antibody in the WT ECs. HIF‐3α was associated with the HIF‐2α promoter irrespective of oxygen tension (G), but not the Ets‐1 promoter (H). The representative data from three independent experiments were shown. (I) The schematic model of the role of HIF‐3α in pulmonary ECs. Our data showed a potential role of HIF‐3α in repressing the VE‐cadherin expression that is required for normal angiogenic properties in pulmonary ECs.

We noted that mRNA levels of Flk1, Ang2, Tie2 and VE‐cadherin were correlated with those of HIF‐2α, suggesting that the expression of these factors was regulated positively and negatively by HIF‐2α and HIF‐3α, respectively (Figs 4A, 6B). Because VE‐cadherin has been shown to be one of the most important molecules responsible for the phenotype of HIF‐3α KO ECs, we next examined how the expression of the VE‐cadherin gene (Cdh5) is regulated by these HIF factors. A previous study reported that the VE‐cadherin gene possesses a putative HRE in the promoter region (Le Bras et al. 2007). Chromatin immunoprecipitation (ChIP) assays showed that the binding of both HIF‐1α and HIF‐2α was rarely detected in the promoter region in the WT ECs (Fig. 6C). Instead, HIF‐3α binding was clearly detectable in this region in the WT ECs (Fig. 6D). In the HIF‐3α KO ECs, the bindings of both HIF‐1α and HIF‐2α were detected in this region even under normoxic conditions (Fig. 6C). As the HIF‐1α and HIF‐2α proteins were undetectable by the immunoblot analyses in the HIF‐3α KO ECs under normoxic conditions (Fig. 6A), this might be an artifact due to the high sensitivity of this assay. Alternatively, it is possible that a small amount of HIF‐1α or HIF‐2α might preferentially bind to this region in the absence of HIF‐3α. In a reporter assay using 293T cells, both HIF‐1α and HIF‐2α activated the VE‐cadherin promoter luciferase activity, and this activation was partially, but significantly, repressed by the co‐expression of HIF‐3α (Fig. 6E). Collectively, these data suggest that HIF‐3α is a negative regulator of the VE‐cadherin gene under both normoxic and hypoxic conditions.

Despite the fact that the HIF‐1α and HIF‐2α protein levels were very low, we did observe significant induction of Flk1, Ang2, Tie2 and VE‐cadherin mRNA in the HIF‐3α KO ECs under normoxic conditions (Fig. 4A). These observations suggest that an oxygen‐independent transcription factor expressed in pulmonary ECs can activate the transcription of these genes under normoxic conditions, and this factor would be repressed by HIF‐3α. As a candidate, we considered Ets‐1, a member of the Ets family of transcriptional factors, which is known to play a critical role in regulating endothelial gene expression (Iwasaka et al. 1996; Tanaka et al. 1998). Interestingly, we found that the mRNA level of Ets‐1 was significantly induced in the HIF‐3α KO ECs under both normoxic and hypoxic conditions (Fig. 6F). Consistently, immunoblot analyses showed that the Ets‐1 protein levels were significantly induced in the HIF‐3α KO ECs, and this induction was diminished by the restoration of HIF‐3α cDNA. These data indicate that HIF‐3α regulates Ets‐1 expression at the transcriptional level. Given that Ets‐1 can bind regulatory regions of angiogenic genes directly, and its protein level is regulated independently of oxygen tension, it is possible that Ets‐1 by itself activates the transcription of angiogenic genes under normoxic conditions. We found that the Ets‐1 binding to the VE‐cadherin promoter was more prominently observed in the HIF‐3α KO ECs by the ChIP assays (Fig. S3 in Supporting Information), suggesting that Ets‐1 is involved in the remarkable induction of the angiogenic genes in the HIF‐3α KO ECs. To elucidate how HIF‐3α regulates the mRNA expression of HIF‐2α and Ets‐1, the association of HIF‐3α on the HIF‐2α and Ets‐1 promoter in the WT ECs was addressed by the ChIP assay with anti‐HIF‐3α antibody. As shown in Fig. 6G,H, the direct binding of HIF‐3α was observed in the putative HRE region within HIF‐2α promoter but not Ets‐1 promoter, suggesting that HIF‐3α suppresses the transcription of HIF‐2α and Ets‐1 mRNA in the direct and indirect manner, respectively.

Discussion

The present study aimed to define the physiological and pathological roles of HIF‐3α, a hypoxia‐inducible factor, in pulmonary ECs. The fundamental roles of pulmonary ECs are gas exchange, maintenance of a thrombosis‐free surface, interaction with circulating cells and modulation of the vascular tone (Rounds & Voelkel 2009). Additionally, the activation of pulmonary ECs is important for angiogenesis in the process of vascular remodeling under both physiological and pathological conditions (Taraseviciene‐Stewart et al. 2001; Thébaud 2010). Previous studies have shown that HIFs are closely linked to the hypoxic responses involved in lung development, as well as in diseases such as pulmonary hypertension, lung injury and cancer (reviewed in Shimoda & Semenza 2011). The activation of HIF‐1α and HIF‐2α enhances the angiogenic properties of pulmonary ECs, leading to cell growth, angiogenic sprouting and migration (Yamakawa et al. 2003; Manalo et al. 2005; Skuli et al. 2009; Krotova et al. 2010). The loss of HIF‐2α in mice with a mixed genetic background impaired fetal lung maturation due to insufficient production of surfactants by alveolar type 2 cells (Compernolle et al. 2002). Targeted disruption of the HIF‐3α/NEPAS gene in mice resulted in impaired vascular remodeling during the lung development (Yamashita et al. 2008). This was due to the loss of the suppressive effect on ET‐1 expression by HIF‐3α/NEPAS. As NEPAS is the major HIF‐3α/NEPAS gene product during the embryonic and neonatal stages, these findings suggested that NEPAS enhances the vascular remodeling in the lungs of developing embryos. HIF‐3α, a HIF‐3α/NEPAS gene product in adults, is also known to act as a repressor of HIF‐1α‐ or HIF‐2α‐driven transcriptional activities. On the basis of these findings, we surmised that a loss of HIF‐3α in pulmonary ECs would lead to the activation of HIF‐1α and HIF‐2α, and as a consequence, the angiogenic properties of ECs in response to hypoxia might be accelerated by the ablation of HIF‐3α. Contrary to our expectations, however, we found that HIF‐3α KO ECs were less angiogenic, and the effects of HIF‐3α ablation were observed under both ‘normoxic’ and hypoxic conditions. These findings show novel and unique cellular functions of HIF‐3α as a regulator of the angiogenic properties in pulmonary ECs.

We found that the expression of representative angiogenic factors, including Flk1, Ang2 and Tie2, was markedly increased by HIF‐3α disruption (Fig. 4A). Nevertheless, the proliferative and angiogenic abilities were reduced in the HIF‐3α KO ECs (Fig. 2). To define the molecular basis of this ‘uncoupling’ phenomenon, we examined the intracellular signaling involving angiogenesis in ECs. Surprisingly, the activation of Akt, a key molecule involved in Flk1‐mediated signaling, was severely attenuated, despite the fact that the expression of upstream factors, Flk1 and Tie2, was increased in the HIF‐3α KO ECs (Fig. 4A,B,E). In contrast, the activation of Erk1/2 was not affected by the HIF‐3α ablation. Because the treatment of VEGF‐A did not increase the phospho‐Akt level in the HIF‐3α KO ECs (Fig. 4C), we suggested that the inactivation of Flk1–PI3 kinase–Akt signaling might be responsible for the decreased angiogenic properties of the HIF‐3α KO ECs.

In fact, we noted that the up‐regulation of VE‐cadherin was accompanied by an increase in the known angiogenic factors (Fig. 4A,B), and we suggested that VE‐cadherin might be a candidate molecule to explain the contradictory observations. First, VE‐cadherin has been known to inhibit cell migration and angiogenic sprouting. Many groups have proposed molecular mechanisms underlying the inhibitory effects of VE‐cadherin (Lampugnani et al. 2003; Lampugnani et al. 2006; Abraham et al. 2009; Pirotte et al. 2011). VE‐cadherin associates with Flk1 and prevents the auto‐phosphorylation of Flk1 by mediating the expression of density‐enhanced phosphatase‐1 (DEP‐1), resulting in the inhibition of downstream signaling and the internalization of Flk1 in confluent resting ECs (Lampugnani et al. 2003; Lampugnani et al. 2006; Pirotte et al. 2011). Moreover, VE‐cadherin activates Rho kinase and myosin light‐chain 2 phosphorylation, leading to suppressive effects against migration via changes in actomyosin contractility and VEGF/Flk1‐mediated, Rac1‐dependent angiogenic sprouting (Abraham et al. 2009). Importantly, we showed that the inhibition of VE‐cadherin clustering by a neutralizing antibody, BV9, partially restored the cell proliferation, tube‐forming potential and migration activities (Fig. 5). These data suggest that the increased expression of VE‐cadherin possibly led to the inhibition of the Flk1 signaling and thereby repressed endothelial cell proliferation and tube formation in the HIF‐3α KO ECs. We also found that the expression of VE‐cadherin was appeared to increase in the HIF‐3α KO lung (Fig. 5D). Thus, the excess expression of VE‐cadherin might be involved in the impaired vascular remodeling under certain pathological conditions in vivo. Interestingly, it has been reported that the decreased expression of VE‐cadherin results in an increased vascular permeability in patients with respiratory distress syndrome (Grover et al. 2007; Herwig et al. 2013). Taken together with our findings, an adequate expression level of VE‐cadherin is necessary for normal angiogenic properties. Although other factors may also be involved, we found that the antiangiogenic effects induced by excess VE‐cadherin were dominant over the proangiogenic effects of Flk1, Ang2 and Tie2 in the HIF‐3α KO ECs.

Our data provide evidence that HIF‐3α plays a critical role in pulmonary EC functions, independent of the oxygen tension. Consistent with a previous report (Makino et al. 2002), we found that HIF‐3α expression was not induced by hypoxia (Fig. 1D,E). The reduced angiogenic abilities, as well as the increased expression of angiogenic factors other than VEGF‐A, were observed under both hypoxic and normoxic conditions in the HIF‐3α KO ECs. Importantly, we found that the expression levels of these factors were very low and not induced by hypoxia in WT pulmonary ECs (Fig. 4A). Collectively, these results suggest that HIF‐3α plays a fundamental role in repressing the expression of angiogenic factors, regardless of the oxygen tension, in pulmonary ECs.

The molecular mechanism underlying how HIF‐3α suppresses these factors is still open to debate. Under hypoxic conditions, it is likely that the increased expression levels of Flk1, Ang2, Tie2 and VE‐cadherin are mediated by the increased HIF‐2α protein in the HIF‐3α KO ECs (Fig. 6A). However, as the HIF‐2α protein was virtually undetectable under normoxic conditions, this cannot explain the up‐regulation of these factors under normoxic conditions in the absence of HIF‐3α. Importantly, our data showed that the Ets‐1 expression level was significantly up‐regulated and its binding to the VE‐cadherin promoter was enhanced in the HIF‐3α KO ECs (Fig. 6F, Fig. S3 in Supporting Information). These data suggest the possibility that Ets‐1 is a potential regulator of these factors under normoxic conditions. The Ets family of transcription factors plays a critical role in regulating endothelial gene expression (Iwasaka et al. 1996; Tanaka et al. 1998). Ets‐1, the well‐characterized factor among ETS family members, is highly expressed during early embryonic development and suppressed at very low levels in resting endothelium. Previous reports showed inconsistent results about the role of Ets‐1 in angiogenesis: The overexpression of Ets‐1 resulted in the induction of angiogenesis in vivo (Hashiya et al. 2004), whereas it reduces growth activity of ECs (Lelièvre et al. 2000). In this study, the tube formation was significantly decreased in the WT ECs transfected Ets‐1 cDNA (Fig. S4 in Supporting Information). Therefore, in pulmonary ECs, the excess Ets‐1 expression may lead to the impaired angiogenic properties.

Our data suggest that the lack of hypoxic induction of the HIF‐2α protein in the WT ECs was also due to the transcriptional repression of the HIF‐2α gene by HIF‐3α. The HIF‐2α gene possesses the putative HRE in the promoter region, and HIF‐2α can activate its own expression under hypoxia (Sato et al. 2001). As shown in Fig. 6G, in the WT ECs, HIF‐3α bound to this putative HRE region of HIF‐2α in the oxygen‐independent manner. It is likely that HIF‐3α inhibits the self‐activation of HIF‐2α by occupying the HRE or by interacting directly with HIF‐1α and HIF‐2α. In fact, a previous report showed that IPAS interacts directly with HIF‐1α, exerting its inhibitory effect (Makino et al. 2001). Another report showed that the HIF‐3α variants associated with HIF‐1α and HIF‐2α in the same way as Arnt, and the HIF‐1α/HIF‐3α complex failed to translocate into the nucleus in human cells (Heikkilä et al. 2011). Collectively, it can be illustrated that HIF‐3α suppresses the mRNA transcription of Ets‐1 and HIF‐2α, potential activators of VE‐cadherin gene. The VE‐cadherin expression is thereby maintained in an adequate level that is essential for normal EC functions (as shown in Fig. 6I). Given that HIF‐2α protein was not detected under normoxic conditions in immunoblot analyses (Fig. 6A), we speculate that Ets‐1 and HIF‐2α promote the VE‐cadherin expression under normoxic and hypoxic conditions, respectively. We need to note, however, that the binding of HIF‐2α and Ets‐1 to the VE‐cadherin promoter was observed under both normoxic and hypoxic conditions in the HIF‐3α KO ECs (Fig. 6C and Fig. S3 in Supporting Information, respectively). Thus, the precise molecular mechanisms in activating the VE‐cadherin promoter need to be elucidated in further experiments.

Notably, our data showed that the mRNA level of HIF‐1α is not affected by HIF‐3α (Fig. 6B). As a consequence, the hypoxia‐dependent regulation of gene expression is almost exclusively regulated by HIF‐1α in WT pulmonary ECs. These data suggest that HIF‐3α might also play a critical role to determine the preference between HIF‐1α and HIF‐2α in hypoxia‐dependent gene regulation.

It has been long considered that the activities of HIF transcription factors are regulated at the protein level. However, this may not be sufficient to repress hypoxia‐dependent gene expression under normoxic conditions, as oxygen‐independent transcription factors, such as Ets‐1, might be able to activate these genes in certain cell types. The present study indicates that HIF‐3α has a unique role in regulating the HIF activities at the transcriptional levels. HIF‐3α is a HIF factor whose expression is not induced by hypoxia in pulmonary ECs. Our data show that the absence of hypoxic induction per se enables this factor to repress angiogenic gene expression under normoxic conditions. In summary, we showed that HIF‐3α represses angiogenic genes independent of the oxygen tension to maintain the functions of pulmonary endothelial cells.

Experimental procedures

Reagents

The mouse monoclonal antibody against the extracellular domain of VE‐cadherin, BV9, was obtained from a commercial source (Abcam Inc., Cambridge, MA, USA). The following enzyme inhibitors were purchased from the respective companies: LY294002 (Cell Signaling, Boston, MA, USA), SU5416 (Tocris, Southampton, UK), PD98059 (InvivoGen, San Diego, CA, USA) and Tie2 kinase inhibitor (Calbiochem; Merck Millipore, Billerica, MA, USA). The recombinant VEGF‐A was obtained commercially (Peprotech, Rocky Hill, NJ, USA).

Immunohistochemical staining

The mice were maintained in accordance with the approval of the Animal Committee of the University of Tsukuba. The lung tissue sample from HIF‐3α^+/− mice was fixed in 4% paraformaldehyde (PFA)/phosphate‐buffered saline (PBS) overnight at 4 °C and was embedded in O.C.T. compound (Sakura Finetek, Tokyo, Japan). Serial cryostat sections were incubated with anti‐GFP (1 : 1000; MBL, Aichi, Japan), CD31 (1 : 1000; clone: MEC 13.3; BD Biosciences, San Diego, CA, USA) and α‐SMA (Dako, Carpinteria, CA, USA) antibodies. After being washed, the sections were incubated with a HRP‐conjugated secondary antibody (1 : 2000; Vector Laboratories, Burlingame, CA, USA). Positive signals were visualized with 3,3′‐diaminobenzidine and counterstained with hematoxylin solutions. To analyze the VE‐cadherin expression, the lung tissues from WT mouse and HIF‐3α KO mouse were harvested, fixed in 4% PFA/PBS and embedded, respectively. The sections were incubated with anti‐VE‐cadherin (1 : 200; Abcam) antibody, and color staining was detected using M.O.M.™ kit (Vector Laboratories) according to the manufacturer's instructions.

Isolation and culture of lung endothelial cells (ECs)

The lung tissues were harvested from 10‐week‐old WT and HIF‐3α^−/− mice that had been crossed with mice expressing the tsSV40‐T transgene (Yanai et al. 1991). The dissected lungs were minced and digested with 0.1% collagenase (Nitta Gelatin, Osaka, Japan)/PBS solution at 37 °C for 1 h. Then, a single cell suspension was stained with an allophycocyanin (APC)‐conjugated anti‐CD31 antibody (Pharmingen, San Diego, CA, USA) and phycoerythrin (PE)‐conjugated anti‐CD45 antibody (Pharmingen). The CD31‐positive and CD45‐negative fraction was collected by flow cytometry using a FACSVantage instrument (BD Biosciences) for WT lungs, and the CD31‐positive, CD45‐negative and GFP‐positive fraction was sorted for HIF‐3α^−/− lungs. The collected cells were cultured in HAVA medium (Ohneda et al. 1998) containing Dulbecco's modified Eagle's medium (DMEM)‐high glucose (Invitrogen, Carlsbad, CA, USA) with 10% FBS, 0.1 mm nonessential amino acids (Invitrogen), 2 mg/mL l‐glutamine (Invitrogen), 0.1 mm 2‐mercaptoethanol (Invitrogen) and 0.1% (v/v) penicillin–streptomycin (100 U/mL penicillin, 0.1 mg/mL streptomycin; Invitrogen) at 33 °C in 5% CO₂ in a humidified atmosphere. HEK293T cells were maintained in DMEM high glucose (Invitrogen) with 10% FBS and 0.1% (v/v) penicillin–streptomycin (100 U/mL penicillin, 0.1 mg/mL streptomycin; Invitrogen) for the luciferase reporter assay.

Establishment of HIF‐3α rescued cells

HIF‐3α KO ECs were infected with a MSCV retrovirus (Clontech, Palo Alto, CA, USA) encoding the full‐length HIF‐3α cDNA cassette and then were treated with puromycin for negative selection. The HIF‐3α expression of the HIF‐3α rescue cells was confirmed by an immunoblot analysis.

RT‐PCR and quantitative real‐time PCR analysis

Total RNA was extracted with Sepasol^®‐RNA I Super G (Nacalai tesque, Kyoto, Japan) and was reverse‐transcribed to cDNA using a ReverTra‐Plus‐^® kit (TOYOBO, Osaka, Japan). To detect cell‐specific markers, HIF‐3α and IPAS, the cDNAs were amplified using a GeneAmp PCR System 9700 (Applied Biosystems and Life Technologies, Carlsbad, CA, USA). The PCR products were electrophoretically separated on 2% agarose gels. The primers used for the PCRs were as follows: 18S rRNA sense 5′‐TATCAGATCAAAACCAACCCGGTGAGC‐3′ and antisense 5′‐CCAATTACAGGGCCTCGAAAGAGTCCT‐3′, Calponin1 sense 5′‐AGCAGGAGCTGAGAGAGTGG‐3′ and antisense 5′‐CAACTTCTCCGGCTCAAATC‐3′, SMAα sense 5′‐AGATTGTCCGTGACATCAAGG‐3′ and antisense 5′‐TTGTGTGCTAGAGGCAGAGC‐3′, CD31 sense 5′‐CACAGATAAGCCCACCAGAGACATGGA‐3′ and antisense 5′‐TCTCGCTGTTGGAGTTCAGAAGTGGAG‐3′, VE‐cadherin sense 5′‐GAACGAGGACAGCAACTTCACCCTCATA‐3′ and antisense 5′‐CTGACACATCATAGCTGGTGGTGTCCAT‐3′, HIF‐3α sense 5′‐ACCAAGACAGGTCGAACACCGAGCTGC‐3′ and antisense 5′‐CAGGTAGCAGGCGTCCAGTGGCTCTCC‐3′, IPAS sense 5′‐ATGGCGTTGGGGCTGCACGCGTG‐3′ and antisense (the same as for HIF‐3α).

The quantitative real‐time PCR analyses were carried out on a 7700 Sequence Detector (Applied Biosystems) with THUNDERBIRD qPCR Mix (TOYOBO). The obtained data were calculated with the ddCt method to determine the relative gene expression levels. The primer sequences for the gene sequencing were as follows: VEGF‐A sense 5′‐GGCTTTACTGCTGTACCTCCACCAT‐3′ and antisense 5′‐CTTCGCTGGTAGACATCCATGAACT‐3′, Flk1 sense 5′‐CTATCTCGCTGTCCCAGGAAATTCT‐3′ and antisense 5′‐CTATCTCGCTGTCCCAGGAAATTCT‐3′, angiopoietin‐2 sense 5′‐CGCTGGTGAAGAGTCCAACTACA‐3′ and antisense 5′‐GTCATTGTCCGAATCCTTTGTGCT‐3′, Tie2 sense 5′‐TGGGTGGCCACTACCTACTAGTGAA‐3′ and antisense 5′‐GGAGGTAAGACTCGGTTGACAGTGA‐3′, VE‐cadherin sense 5′‐CCGCCAGAATGCTAAGTATGTGC‐3′ and antisense 5′‐TCCACAATGAGGGCAGTAAGGAA‐3′, HIF‐1α sense 5′‐GTGCTGATTTGTGAACCCATTCC‐3′ and antisense 5′‐CGGCTCATAACCCATCAACTCAG‐3′, HIF‐2α sense 5′‐CCTGGACAGCAAGACTTTCCTGA‐3′ and antisense 5′‐GAACTCATAGGCAGAGCGTCCAA‐3′, Ets‐1 sense 5′‐CTGCATCCTATCAGCTCGGAAGA‐3′ and antisense 5′‐ACACCTCTTGCTTGATGGCAAAG‐3′ and β‐actin (as an internal control) sense 5′‐GTCGTACCACAGGCATTGTGATGGACT‐3′ and antisense 5′‐CACCAGACAGCACTGTGTTGGCATAGA‐3′.

Immunoblot analysis

Nuclear extracts or cytosolic was prepared from cells treated under normoxic or hypoxic (1% O₂, 6 h) conditions. Equal amounts of nuclear extracts were electrophoresed on a 7.5% sodium dodecyl sulfate (SDS)‐polyacrylamide gel and were transferred onto polyvinylidene difluoride membranes (Merck Millipore). After blocking with 5% skim milk/TBST buffer, the membranes were incubated with the following primary antibodies: rabbit anti‐HIF‐1α (Novus Biologicals, Littleton, CO, USA), rabbit anti‐HIF‐2α (Novus Biologicals), goat anti‐HIF‐3α, rabbit anti‐Flk1, rabbit anti‐angiopoietin2, rabbit anti‐Tie2, goat anti‐VE‐cadherin and rabbit anti‐Ets‐1 (all were purchased from Santa Cruz Biotechnology, Santa Cruz, CA, USA). To detect activated signaling cascades, rabbit anti‐Akt, rabbit anti‐phospho‐Akt (Ser473), rabbit anti‐Erk1/2 and rabbit anti‐phospho‐Erk1/2 (Thr202/204) were used (all were obtained from Cell Signaling). To confirm equal protein loading and transfer, goat anti‐Lamin B antibodies were used for nuclear extracts and goat anti‐actin antibodies (Santa Cruz Biotechnology) were used for cytosol extracts. After being washed, the membranes were incubated with HRP‐conjugated secondary antibodies, and positive signals were detected with chemiluminescence reagents (Merck Millipore). The intensity of positive signal was analyzed by a luminescence imager (Image Quant LAS4000; GE healthcare UK Ltd, Little Chalfont, Buckinghamshire, UK).

Growth curve analysis

The wild‐type and HIF‐3α KO ECs were seeded on 35‐mm tissue culture dishes (Sumitomo‐Bakelite, Tokyo, Japan) at a density of 2 × 10⁴ cells and were cultured under normoxic conditions (20% O₂) or hypoxic conditions (5% O₂). The cell culture medium was changed every 4 days. Dead cells were excluded by staining with trypan blue solution (Invitrogen), and the living cells were counted using a hemocytometer in triplicate, and the numbers of cells were scored at 24‐h intervals for 8 days.

In vitro tube formation assay

The in vitro tube formation assay was carried out as described previously (Nagano et al. 2007). Briefly, four‐well plates (Nunc; Thermo Fisher Scientific, Waltham, MA, USA) were coated with Matrigel (BD Biosciences) and incubated at 37 °C for 30 min. Then, WT ECs and HIF‐3α KO ECs were plated at a density of 5 × 10⁴ cells onto each Matrigel‐coated wells. After 9 h of incubation under normoxic or hypoxic (5% O₂) conditions, the numbers of tubes formed on the Matrigel were counted in five randomly selected fields in each well. The experiments were carried out independently in triplicate, and the average numbers of tubes from each sample were calculated.

In vitro wound migration assay

WT ECs and HIF‐3α KO ECs were seeded in four‐well plates at a density of 1.5 × 10⁵ cells. After 24 h, confluent monolayers were established, and scratch wounds were generated with a rubber cell scraper (1 mm width). Images were taken at the time points indicated in Figs 2C and 3D. The migration distances were calculated as the distances between the migration front of the wound and the edge at each time point. In Figs 2C and 3D, four and seven measurements from three and four independent wounds were evaluated, respectively.

Chromatin immunoprecipitation (ChIP) assay

The ChIP assay was carried out using the ChIP‐IT^® Express Enzymatic Kit (Active Motif, Carlsbad, CA, USA) according to the manufacturer's instructions. The extracted chromatin samples were enzymatically sheared and immunoprecipitated with the following primary antibodies: mouse anti‐HIF‐1α (Novus Biologicals), rabbit anti‐HIF‐2α (Novus Biologicals), goat anti‐HIF‐3α (Santa Cruz Biotechnology) and each of control IgGs (Santa Cruz Biotechnology). The precipitated genome fragments were subjected to a semi‐quantitative PCR analysis. The PCR primer set was designed to span the putative HRE sequence on VE‐cadherin (Le Bras et al. 2007), HIF‐2α (Sato et al. 2001) and Ets‐1 (Oikawa et al. 2001). The primer sequences were as follows: VE‐cadherin putative HRE sense 5′‐AACAGAACAGATTGTGGCAGAGAG‐3′ and antisense 5′‐ATGGAGTATTAGCGTCCTGAGACC‐3′, HIF‐2α putative HRE sense 5′‐GGGAAAGTGTCTAAGATCAAAGCAGGAG‐3′ and antisense 5′‐AGGTGCTCGTGGTAGACTGTTAGGTTTA‐3′ and Ets‐1 putative HRE sense 5′‐GATCTGGTTAGATTCCCAGTGCTGA‐3′ and antisense 5′‐GATGAGTCAATTGCTCCTTCCCTTC‐3′.

Luciferase reporter assay

The pGL4.10 VE‐cadherin −2486 to +25 promoter (Le Bras et al. 2007), pEF‐BOS‐HIF‐1α, pEF‐BOS‐HIF‐2α and/or HIF‐3α were transfected into HEK293T cells using Lipofectamine LTX and PLUS reagent (Life Technologies). After 24 h of transfection, the reporter assay was carried out using the dual‐luciferase reporter assay system (Promega, Madison, WI, USA). The pEF‐Renilla‐Luc plasmid was co‐transfected as an internal control to evaluate the transfection efficiency. The luciferase activities were measured using a Turner Designs Luminometer Model TD‐20/20 (Promega), and the activities of the reporter vector were normalized to each internal control value.

Statistical analysis

The data are shown as the means ± standard deviation (SD). Comparisons of two groups were conducted using an unpaired Student's t‐test. To compare multiple groups, an anova and the Tukey post hoc test were carried out to evaluate the statistical significance of differences between groups. A value of P < 0.05 was considered to be statistically significant.

Acknowledgements

This work was supported by grants‐in‐aid from the Japan Society for the Promotion of Science and the Japanese Ministry of Education, Culture, Sports, Science, and Technology.