HIF-2α promotes the formation of vasculogenic mimicry in pancreatic cancer by regulating the binding of Twist1 to the VE-cadherin promoter

Vasculogenic mimicry (VM) is a blood supply modality that occurs independently of endothelial cell angiogenesis. Hypoxia and the epithelial-mesenchymal transition (EMT) induce VM formation by remodeling the extracellular matrix. Our previous study demonstrated that hypoxia-inducible factor-2 alpha (HIF-2α) promotes the progress of EMT in pancreatic cancer; however, whether HIF-2α promotes VM formation in pancreatic cancer remains unknown. In this study, we investigated HIF-2α expression and VM by immunohistochemistry in 70 pancreatic cancer patients as well as the role of Twist1and Twist2 in HIF-2α-induced VM in vitro and in vivo. We found that the overexpression of HIF-2α and VM were correlated with poor tumor differentiation, late clinical stage and lymph node metastasis, and a poor prognosis in pancreatic cancer. Moreover, the upregulation of HIF-2α in SW1990 cells induced VM formation, whereas the opposite results were found after silencing HIF-2α in AsPC-1 cells. A mechanistic study indicated that HIF-2α might regulate the binding of twist1 to vascular endothelial cadherin (VE-cadherin) to promote VM formation in pancreatic cancer cells, and that the P1 (-421bp) and P4 (-2110bp) regions of the Twist1 binding sequences are positive regulatory elements for VE-cadherin. In addition, we confirmed that the overexpression of HIF-2α increased Twist1 expression and promoted tumor growth and VM formation in pancreatic cancer xenografts in nude mice. These findings indicated that HIF-2α might play a critical role in VM and that HIF-2α and the pathway of HIF-2α inducing VM formation are potential therapeutic targets for pancreatic cancer.


INTRODUCTION
Pancreatic cancer, one of the most lethal cancers among all malignances, is characterized by aggressive local invasion and metastatic spread, with a median overall survival of less than 1year and a 5-year survival of approximately 5% [1]. There are no obvious symptoms in the early stage of pancreatic cancer, and more than half of all pancreatic cancer patients are diagnosed after metastases have occurred, when the chance for surgical resection has been lost [2]. Pancreatic cancer is a complex, multifactorial disease involving mutations in cancer-associated genes, and has a poor prognosis because traditional radiation or chemotherapy strategies yield disappointing results [1,3]. Thus, it is necessary to understand the molecular pathogenesis of pancreatic cancer, and identify new molecular targets. Tumor growth and metastasis depend on tumor angiogenesis, a type of neovascularization that occurs in embryonic development and is conventionally regarded as the only way to provide tumor blood supply [4][5][6]. Anti-angiogenesis drugs that are specific to endothelial cells are a traditional therapy for tumors; however, anti-angiogenesis drugs yield little clinical effect for some malignant tumors. Maniotiset al. [7] described an angiogenesis-independent pathway called "Vasculogenic Mimicry (VM)," which is a newly-defined pattern of tumor blood supply in which highly aggressive tumor cells mimic endothelial cells and form extracellular matrix (ECM)-rich structures to supply red cells and plasma without the appearance of endothelial cells. VM channels are surrounded by tumor cells and extracellular matrix, which stain periodic acid-Schiff (PAS) positive, and endothelial cells or makers are absent [8]. The name of VM was created to describe these channels of tumor cells: vasculogenic refers to the channels which contain red blood cells are not come from preexisting vessls; and mimicry indicates that the channels mimic blood vessels but not true blood vessels. Since its discovery, VM has been detected in many other malignant tumors, including hepatocellular carcinoma [9,10], gallbladder cancer [11], lung cancer [12], prostate cancer [13],and gastric cancer [14]. In fact, VM formation is a multistep and complex process, including activation and proliferation of tumor cells, differentiation of tumor cells into endothelial-like cells and appearance of new vessels. VM has also been linked to increased tumor metastasis rates, disappointing treatment outcomes, and a poor prognosis [15,16]. Some studies have showed that blood vessels are a threestage phenomenon containing VM patterns, mosaic blood vessels, and endothelium-dependent vessels [17]. Following tumor growth, VM occurs first, whereas endothelium-dependent vessels ultimately provide the tumor with sufficient blood and promote cancer progression [18]. In various carcinomas, several key proteins are necessary for the formation of VM, among which VE-cadherin is the most critical for VM channels [19,20].
Along with tumor growth, hypoxia is a characteristic of most solid tumors. Hypoxia-inducing factors (HIF) are involved in VM, angiogenesis, cellular differentiation, tumor progression, chemoresistance, apoptosis, and glucose metabolism, all of which are correlated with a poor prognosis in cancer patients [21]. Hypoxia is involved in angiogenesis and VM formation via many signaling pathways, including the direct modulation of VEGF-A, VEGFR1, EPHA2, TWIST, COX-2, and NODAL expression [22]. Hypoxia inhibits the degradation of HIF, providing a chance for the nuclear localization of HIF-1α and HIF-2α. Previously, HIF-2α has been shown to play an important role in many aspects of digestive cancers, including proliferation, angiogenesis, metastasis, and resistance to chemotherapy [21]. Our previous study indicated that HIF-2α is a vital factor during hypoxia and participates in the epithelial-mesenchymal transition (EMT) in pancreatic cancer [23]; however, there is limited evidence of the role and mechanisms of HIF-2α during VM formation in pancreatic cancer.
In this study, we first investigated the expression and clinical significance of HIF-2α and VM in pancreatic cancer. Next, we studied the role of HIF-2α in the formation of VM in pancreatic cancer cells in vitro and in vivo. Last, we further investigated the possible mechanism between HIF-2α and VM during the progression of pancreatic cancer.

Overexpression of HIF-2αin pancreatic cancer predicts a poor prognosis
To detect the effects of HIF-2αon the development of pancreatic cancer, 70 pancreatic cancer tissues and matching adjacent non-tumor tissues were analyzed to investigate the expression of HIF-2α by IHC staining. Based on the results obtained, the localization of HIF-2α was primarily in the cytoplasm and nucleus ( Figure 1A). As well, HIF-2α was expressed in 67.1% (47/70) of the pancreatic cancer tissues but in only 11.4% (8/70) of the adjacent non-tumor pancreatic tissues. Further, the difference in the positive rate of HIF-2α expression between the pancreatic cancer tissues and the non-tumor tissues was significant (χ2=45.549, P<0.05).
As shown in Table 1, we also investigated the correlation of HIF-2α expression with clinicopathological features in the pancreatic cancer patients and found that HIF-2α expression was significantly correlated with tumor differentiation (χ2=6.921, P=0.026), clinical stage (χ2=6.460, P=0.017), and lymph node metastasis (χ2=5.250, P=0.040). However, the expression of HIF-2α had no obvious association with gender, age, tumor location, or tumor size (P>0.05). These results indicated that HIF-2α might be involved in poor differentiation and advanced clinical stages of pancreatic cancer. To assess the prognostic significance of HIF-2α expression on survival of pancreatic cancer patients further, a Kaplan-Meier analysis was conducted. The results indicated that patients with HIF-2α-negative expression had a significantly longer survival time than those with HIF-2α-positive expression (log-rank test, P<0.05; Figure 2B). In addition, the results of univariate analysis revealed that clinical stage (P=0.001), lymph node metastasis (P=0.001), and HIF-2α expression (P=0.001) were significantly associated with patient survival (Table 2). Finally, the results of the multivariate analysis revealed that clinical stage (P=0.002), lymph node metastasis (P=0.001), and HIF-2α expression (P=0.001) were independent and significant prognostic factors in pancreatic cancer patients (Table 3). In conclusion, the overexpression of www.impactjournals.com/oncotarget  HIF-2α was correlated with poor prognosis and was an independent prognostic marker of pancreatic cancer.

Correlation between HIF-2α expression and VM in pancreatic cancer
To determine whether VM structures could be observed in tissues obtained from 70 pancreatic cancer patients, we utilized anti-CD34 and PAS staining to identify the endothelium and VM channels, respectively. CD34-positive staining identified vessels formed by endothelial cells (yellow arrow), whereas CD34-negative, PAS-positive vascular-like structures containing red blood cells formed by cancer cells or ECM were regarded as VM (red arrow; Figure 1B). We found that 51.4% (36/70) of the tissues contained VM channels within the pancreatic cancer tissues; however, none of the tissues in the nontumor tissue group exhibited a VM pattern. Further, we found that the expression of VM channels was correlated with tumor differentiation (χ2=8.588, P=0.014), clinical stage (χ2=11.712, P=0.001), and lymph node metastasis (χ2=6.863, P=0.016), whereas the expression of VM had no obvious association with gender, age, tumor location, or tumor size (P>0.05). As well, the patients in the VM-positive group had a poorer prognosis than those in the VM-negative group (log-rank test, P<0.05; Figure 2B). The univariate analysis results indicated that patients with VM expression had a high risk of reduced survival time (P=0.001, Table 2). In addition, the results of the multivariate analysis using Cox's proportional hazards model revealed that VM expression was also an independent and important prognostic factor in all patients (P=0.012, Table 3). These results indicated that VM is associated with the prognosis of pancreatic cancer patients.
To investigate the relationship betweenHIF-2α expression and VM in 70 pancreatic cancer tissues, we examined the expression of HIF-2α and VE-cadherin in both the VM-positive and VM-negative groups. As the most critical protein for VM, VE-cadherin expression was significantly higher in the VM-positive group (25/36) than in the VM-negative group (13/34; χ2=6.044, P<0.05). Further, HIF-2α expression was detected in 80.6% (29/36) of the VM-positive group but in only 52.9% (18/34) of the VM-negative group. As well, the difference in the positive rate of HIF-2α expression between the VM-positive and VM-negative groups was significant (χ2=6.863, P<0.05; Figure 2A). In addition, there was a significant positive correlation between the expression of HIF-2α and VM (r=0.294, P<0.05) and between the expression of VEcadherin and VM (r=0.313, P<0.05) in the pancreatic cancer tissues. We performed stratifed analysis of survival time between HIF-2α + VM + group and HIF-2α + VMgroup. The results indicated that patients with HIF-2α + VM + group   Figure 1). The result indicated that HIF-2α positive group with VM expression had a high risk of reduced survival time.

HIF-2α promotes the migration, invasion, and formation of VM in pancreatic cancer in vitro
To validate the expression of HIF-2α in pancreatic cancer cells, we performed western blotting using the SW1990, AsPC-1, CaPan-2, Patu8988, BxPc-3, and CFPANC-1 pancreatic cancer cell lines. Based on the results, we determined that AsPC-1 cells had the highest HIF-2α expression, in contrast with the SW1990 cells, which expressed the lowest level ( Figure 3A). To investigate the effect of HIF-2α on VM, we selected AsPC-1 cells to conduct the HIF-2α silencing experiments by siRNA, whereas SW1990 cells were transfected with HIF-2α cDNA to upregulate HIF-2α expression. As shown in Figure 3B, the expression of HIF-2α was obviously decreased in AsPC-1 cells after si-HIF-2αtreatment (P<0.05), whereas the expression of HIF-2α was significantly upregulated in SW1990 cells following transfection with OE-HIF-2α (P<0.05).
Because VM is closely related to invasion and migration in tumor cells, we measured invasion ability and cell migration using the Transwell® system and a wound healing assay after performing HIF-2α knockdown or HIF-2α ectopic transfection, respectively. We found that the si-HIF-2α pancreatic cells with knocked-down HIF-2α exhibited decreased tumor cell invasion compared with the si-Scramble group (P<0.01). However, the upregulation of HIF-2α expression in the OE-HIF-2α group promoted cell invasion compared with the vector group (P<0.01; Figure 4A). A similar trend was observed in the wound healing assay, as the si-HIF-2α cells displayed relatively slower migration towards the wound space compared with the si-Scramble cells; however, upregulation of HIF-2α expression could enhance the migration speed toward the wound space (P<0.05; Figure 4B).
To confirm the role of HIF-2α in VM formation, 3D tumor cell culture was used to observe VM formation in vitro. Notably, the number of complete tubular channels formed by AsPC-1 cells was significantly decreased in the si-HIF-2α cells compared with si-Scramble cells in 3D cultures (P<0.05). In contrast, the overexpression of HIF-2α in SW1990 cells promoted the formation of more typical ECM-rich vessel-like networks by tumor cells in the OE-HIF-2α group compared with the vector group (P<0.05; Figure 4C). These data indicated that HIF-2α promoted the formation of VM in pancreatic cancer cells in vitro.
Next, we examined the levels of the VM related marker VE-cadherin, as well as Twist1, Twist2, MMP2, and MMP9, which are important to cell plasticity and VM formation, in cells exhibiting altered HIF-2α expression via si-RNA or overexpression. As shown in Figure 4D, the expression of VE-cadherin, Twist1, Twist2, MMP2, and MMP9 were significantly decreased in si-HIF-2α AsPC-1 cells compared with si-Scramble cells (P<0.05). Similarly, after upregulating HIF-2α by transfecting HIF-2α cDNA, the expression of VM-associated proteins including VE- cadherin, Twist1, Twist2, MMP2, and MMP9 were all increased in the OE-HIF-2α group compared with the negative vector group in the SW1990 cells(P<0.05). These results confirmed our hypothesis that HIF-2α enhanced VM-like channel formation, which might be involved in regulating VM-related proteins.

HIF-2α promotes VM formation through Twist1 binding to VE-cadherin in pancreatic cancer
Twist is reportedly a main transcription factor associated with the EMT progress [23]. First, we used Patch software to identify potential transcription factor binding sites in the promoter region of the VE-cadherin gene. Among these transcription factors, we found that Twist was important. Four potential Twist protein binding sites, separately designated P1 (-421bp), P2 (-714bp), P3 (-2000bp), and P4 (-2110bp) were identified in the Patch transcription factor binding site database (Table 4). Next, we performed a ChIP assay using AsPC-1 cells overexpressing Twist1 and Twist2 to investigate whether Twist1 and Twist2 could bind to the promoter region of VE-cadherin. Based on the results presented in Figure 5A, we found that the Twist1 antibody only specifically immunoprecipitated a Twist1-DNA complex in the P1 and P4 regions of the VE-cadherin promoter, and that Twist1 had no positive binding capacity to the other two VE-cadherin binding sites. However, there was no significant evidence to illustrate that Twist2 had the capacity to bind to the four potential transcription regions of VE-cadherin via the ChIP assay. The results suggested that Twist1 directly bound to the VE-cadherin promoter through the P1 and P4 regions.
To understand the effect of the P1 and P4 elements on VE-cadherin transcription, we constructed recombinants containing wild type and mutated promoters of VE-cadherin and conducted luciferase reporter gene assays to evaluate their effects. The normal pVE2000 luciferase promoter construct displayed increased transcription activity; however, the deletion of P1 or P4 in the promoter caused an obvious decrease in transcription activity ( Figure 5B). Our results suggest that the P1 and P4 Twist1 binding sequence regions are positive regulatory elements for VE-cadherin transcription.
To confirm that Twist1 is involved in HIF-2αinduced VM channel formation, we first silenced Twist1 expression by RNAi and then upregulated Twist1 expression by pcDNA3.1-Twist1 (P<0.05, respectively; Figure 6A). Figure 6B, 6C and 6D indicated that Twist1, a main VM-related proteins, promotes the migration, invasion and VM formation in pancreatic cancer cells; While downregulating Twist1 in AsPC-1 cells had the opposite effect (P<0.05, respectively). Next, we examined whether the restoration of Twist1 could rescue VM formation after HIF-2α silencing. The co-transfection of pcDNA3.1-Twist1 and si-HIF-2α in AsPC-1 cells efficiently restored Twist1 expression, and partially reversed the suppression of VM formation induced by HIF-2α silencing (P<0.05, respectively; Figure 6D). These results indicated that Twist1 could contribute to VM formation, and that Twist1 could also, at least partly, revert the ability of HIF-2α to induce VM formation in pancreatic cancer cells. These findings indicated that HIF-2α might regulate Twist1 binding to VE-cadherin to promote VM formation in pancreatic cancer cells.

HIF-2α promotes the expression of Twist1 and VE-cadherin and contributes to VM formation in pancreatic cancer in vivo
We reported that HIF-2α can promote VM formation in vitro. To investigate whether HIF-2α contributes to VM in vivo, we established xenografts in nude mice and then transplanted AsPC-1 cells with stably silenced HIF-2α and SW1990 cells that stably overexpressed HIF-2α, as well as the respective empty vectors, into the pancreas of each mouse. After two weeks, we observed the tumor volume and investigated the expression of VM, Twist1, and VE-cadherin in the murine pancreatic tumor tissues. As shown in Figure 7A, the tumors were significantly smaller in the si-HIF-2α group than in the si-Scramble group (P<0.05), whereas the tumors in the OE-HIF-2α group were larger compared with the vector group (P<0.05). CD34/PAS staining revealed that VM was obviously reduced in the si-HIF-2α xenografts compared with the si-Scramble group (P<0.05). In contrast, VM was markedly increased in the OE-HIF-2α xenografts compared with the vector group (P<0.05). Twist1 and VE-cadherin, which are vital VM proteins, were more highly expressed in the OE-HIF-2α xenografts compared with the vector group (P<0.05); however, the expression of both Twist1 and VE-cadherin was lower in the si-HIF-2α group compared with the matched group (P<0.05; Figure 7B).

DISCUSSION
An increasing number of studies have indicated that tumor angiogenesis not only includes endotheliumdependent vessels, but is also dependent on VM formed by cancer cells and ECM [24]. In fact, VM channels, mosaic blood vessels, and endothelial vessels coexist in malignant tumors and can exchange the stage of each other when the tumor microenvironment changes [25]. Following the progression of pancreatic cancer, tumors require additional  perfusion pathways to obtain sufficient oxygen and nutrients from the blood. As a new and unusual source of blood supply to solid tumors, VM is clinically significant in pancreatic cancer and is associated with poor prognosis [20,26]. In our study, we observed VM channel patterns in pancreatic cancer tissues, and demonstrated that VM channels were correlated with poor tumor differentiation, advanced clinical stage, and lymph node metastasis. In addition, VM-positive expression was associated with shorter overall survival than VM-negative expression in pancreatic cancer patients. Growing solid tumors reach hypoxia because the blood supply in their normal development state is lacking, which induces the expression of HIF-1α and HIF-2α and contributes to the growth of the tumor in this condition [27]. In the present study, we discovered that the expression of HIF-2α was high in human pancreatic cancer tissues, and that the increased expression of HIF-2α caused clinicopathological disadvantages in pancreatic cancer patients. Hypoxia can stimulate the formation of tumor vessel structures including VM and angiogenesis, which further provides a convenient environment for tumor invasion and metastasis [28]. Our study indicated that HIF-2α could promote invasion and cell migration in pancreatic cancer cell lines in vitro. We also found that HIF-2α enhanced the ability of pancreatic cancer cells to form VM structures both in vitro and in vivo. Based on our results, it appears that HIF-2α not only plays a vital role in tumor cell invasion and migration but also closely associated with tumor cell plasticity in VM patterns.
To elucidate the role of HIF-2α in promoting VM formation, additional experiments were conducted. Tumor cells involved in VM formation exhibit a high degree of plasticity and an obvious multipotent phenotype that is similar to embryonic stem cells [29]. In malignant tumors, unlike normal embryonic progenitors, plastic tumor cells lack vital regulatory checkpoints, which stimulates the aberrant activation of embryonic signaling pathways and leads to unregulated growth and aggressive behavior in these cells [30]. As a member of the cadherin family of transmembrane proteins, VE-cadherin is an adhesive protein expressed specifically in endothelial cells that accelerates homotypic cell-to-cell interactions and is involved in vasculogenic activities [17,31]. Highly aggressive tumor cells assume the endothelial cell phenotype to mimic the structure of the vessels. Further, VE-cadherin is an important vascular-associated gene that participates in VM formation. VE-cadherin is highly expressed in aggressive melanoma cells, and the downregulation of VE-cadherin expression inhibits VM formation [32]. Additionally, studies have reported that MMP2 and MMP9 also participate in VM formation by regulating the PI3K/Akt signal pathway [33]. In this study, the expression of HIF-2α and VE-cadherin was higher in the VM-positive group than in the VM-negative group. We likewise discovered that the expression of VE-cadherin, MMP2, and MMP9 was also upregulated following the upregulation of HIF-2α and that the downregulation of HIF-2α resulted in decreased VE-cadherin, MMP2, and MMP9 expression. Thus, it seems reasonable to conclude that HIF-2α is involved in VM formation through the regulation of several important VM-associated proteins, especially VE-cadherin.
The transcription factor Twist, a master regulator of embryonic morphogenesis, is closely related to many kinds of cancers and promotes metastasis of tumor cells by contributing to VM formation. As a main VM-related proteins, twist is a member of the basic helix-loop-helix (bHLH) transcription factor family and structural similarity with HIF at the bHLH [20]. The function of HIF and Twist may have some similarity. So, we focused on Twist protein including Twist1 and Twist2. Twist1 and Twist2 belong to the bHLH family of transcription factors, and they share high sequence homology and structural similarity [34,35]. In our study, the ChIP assay results showed that Twist1, but not Twist2, could promote VE-cadherin transcription in the -421 bp and -2110 bp promoter region sites of the VE-cadherin gene. As well, the results of the luciferase reporter gene assays indicated that the -421 bp and -2110 bp regions of the Twist1 binding sequences are positive regulatory elements for VE-cadherin transcription and can increase the activity of VE-cadherin. However, the role of Twist2 in VM remains unclear. As important transcription factors in VM formation, the expression of Twist1 was decreased after HIF-2α was downregulated in AsPC-1 cells, whereas the opposite results were observed after upregulating HIF-2α in SW1990 cells in vitro and in vivo. Furthermore, restoring the expression of Twist1 rescued the ability of HIF-2α to form VM structures. These results indicated that Twist1 could promote VM formation, and that HIF-2α promoted VM formation through the binding of Twist1 to the VE-cadherin promoter.
VM, a novel and angiogenesis-independent pathway, was considered as a challenge to traditional anti-angiogenic treatments and poor prognosis in tumor patients. In the past, potential therapies aimed at endothelial cells have gained much attention [36]. However, the effects of anti-angiogenesis drugs have been unsatisfactory in these malignant tumors that have other blood supply mechanisms such as VM. Furthermore, when endothelial vessels are blocked or the blood vessel density is reduced due to anti-angiogenic agents, the tumor might reach hypoxia. Subsequently, the oxygen and nutrient deficiencies contribute to VM formation and indirectly promote tumor progression. Therefore, strategies that combine traditional endothelium-dependent drugs with VM-targeted therapies have garnered much attention and interest. Recently, many studies have focused on various mechanisms and genes involved in VM formation, such as VE-cadherin and MMPs. The development of an anti-VM therapy should consider several important aspects including suppressing the plasticity of tumor cells, cutting off the molecular signaling pathways of VM formation, and the remodeling of the tumor microenvironment [37,38]. However, VM and anti-VM are still in the initial stages and additional studies are required to elucidate the specific mechanisms of VM. Nonetheless, whether blockage of HIF-2α could be an efficient way to inhibit VM formation deserves further exploration.
In conclusion, our study demonstrated that HIF-2α and VM were overexpressed in pancreatic cancer tissues and were associated with poor pathological characteristics. HIF-2α contributes to VM formation by regulating the expression of VE-cadherin through the binding of the transcription factor Twist1 to the promoter of VE-cadherin in pancreatic cancer both in vitro and in vivo. Thus, VM therapies should focus on HIF-2α or other VM-associated genes to find more effective ways to treat pancreatic cancer.

Clinical samples
Tumor tissues were obtained from 70 patients at the First Affiliated Hospital of Soochow University from January 2011 to December 2013, including tumor samples and matching adjacent non-tumor tissues. Detailed clinicopathological data were recorded, including the age and gender of each patient, and the tumor size, tumor differentiation, and lymph node metastasis. The tumor clinical stages were classified according to the Union for International Cancer Control (UICC) staging system. None of the patients received chemotherapy, radiotherapy, or immunotherapy prior to surgery. All samples were obtained following patient consent and approval by the Ethics Committee of Soochow University.

Lentiviral vector production and cell infection
Lentiviral vector system was purchased from GeneChem (Shanghai, China). This vector system include three plasmids: GV115 or GV358, pHelper 1.0 and pHelper 2.0. According to the manufacturer's instruction, lentiviral constructs GV115 containing the siRNA sequence targeting human HIF-2α (5'-GCAAATGTACCCAATGATA-3') was used to establish cell lines constitutively repressing HIF-2α (named si-HIF-2α). A nonspecific scrambled siRNA sequence (sequence: 5'-GTTCTCCGAACGTGTCACGT-3') was used as a negative control (si-Scramble). Full-length HIF-2α was synthesized and subcloned into a GV358 vector. Lentiviral vector encoding the human HIF-2α gene was designated OE-HIF-2α. The empty vectors served as a negative control (Vector). Pancreatic cancer cell lines were grown to 30-40% confluence before lentiviral infection. The viral supernatant was added at a multiplicity of infection (MOI) of 5 per well with 5 ug/ml Polybrene. Cells were cultured in normal growth medium for 72 h after infection and expanded for downstream experiments.

Three-dimensional (3D) culture
A 24-well culture plate was coated with 250 µl/well of growth factor-reduced Matrigel™ (BD Biosciences), which was allowed to polymerize for 1 hour at 37°C. A cell suspension (5 × 10 5 cells/well) was seeded on top of the gel, and three wells were used for each group of AsPC-1 cells and SW1990 cells. After adding DMEM (HyClone, Shanghai, China) containing 10% FBS (HyClone, Shanghai, China), the cells in the 3D culture system were observed at 12 and 24 hours for the number and the completeness of the tubules. Images were captured using an inverted light microscope at 200× magnification (Leica Microsystems, Mannheim, Germany) after incubation for 24 hours.

Wound healing assay
Cells from each group were plated in 6-well plates. When the cells reached 80% confluence, a small wound area was created in the confluent monolayer by making a lengthwise swipe using a 200-µl pipette tip. The cells were washed twice with PBS and incubated at 37°C. The speed of wound closure was measured by calculating the ratio of the distance of the wound closure at 24 and 48 hours compared to 0 hours, respectively. The wound width was measured at 100× magnification using a microscope (Leica Microsystems, Mannheim, Germany). Each experiment was performed in triplicate.

Cell invasion assay
In vitro cell invasion assays were performed using Transwell® cell culture chambers with 8 μm pores (Corning, NY, USA). The inserts in the membrane filter were coated with 40ul configured Matrigel™ on the upper surface. The cells were resuspended in serum-free DMEM at a concentration of 5 × 10 5 cells/ml and placed in the upper chamber. The lower chamber was filled with DMEM with 10% FBS. After incubation at 37°C for 48 hours, the cells on the upper surface of the filter were removed with a cotton swab. The invading cells at the bottom of the Matrigel™ were fixed in methanol and stained with 0.1% crystal violet. The number of invading cells in five random fields per well was calculated using a microscope at 200× magnification. Each assay was performed in triplicate.

Luciferase reporter assay
The VE-cadherin promoter regions (approximately 2 kb surrounding the transcription start site) were generated by PCR amplification of genomic DNA and cloned into the pGL3-basic promoter vector (Promega). The sequence of mutated P1 is 5'-CACGTG-3' and the sequence of mutated P4 is 5'-CAGAAAAATC-3'. Luciferase reporter assays were performed by transfecting the mutated VEcadherin promoter reporter plasmid (separate deletions of P1 or P4 in the promoter), together with the basic pRL vector, into AsPC-1 cells (Shanghai, China) in triplicate using Lipofectamine® 2000 (Invitrogen). The cell lysates were analyzed for luciferase activity using the Dual-Luciferase® Assay Kit according to the manufacturer's instructions. Each experiment was performed in triplicate.
The staining intensity was scored as follows: 0, no staining; 1, weak staining; and 2, moderate to strong staining. The percentage of positively stained cells was scored as follows: 0, <10%; 1, 10%-50%; and 2, >50%. The final score was calculated as the sum of the intensity and quantity scores. A score >2 indicated positive expression. CD34 staining was performed to identify endothelial cells, and any structure containing CD34positive immunoreactivity was defined as a blood vessel. PAS staining was used to identify matrix-associated vascular channels in the pancreatic cancer tissues. Vessels lined by endothelial cells, regardless of the presence of a basement membrane, were counted as endotheliumdependent vessels. In contrast, VM structures were defined as CD34-negative, PAS-positive structures.

Xenografts in nude mice
Twenty 6-week-old nude female mice were randomized into four groups (si-Scramble, si-HIF-2α, vector, and OE-HIF-2α, respectively). All protocols used in the animal experiments were approved by the institutional animal ethics committee. To evaluate tumor formation, 3 × 10 6 cells/0.2 mL were suspended and injected into the pancreas of each nude mouse. Tumor size was determined by caliper measurements of the length and width. Tumor volume was calculated using the formula tumor volume = (length × width) 2 /2. Two weeks after injection, the mice were sacrificed and the tumors were harvested, fixed with formalin, and embedded in paraffin.

Statistical analysis
All data in the study were evaluated using SPSS version 18.0 software. All data were presented as mean ± SD. The continuous variables were compared using one-way analysis of variance (ANOVA) and categorical variables were compared using a Chi-square test. The correlation analysis was performed using a Spearman's Rank Correlation test. Survival was assessed according to the Kaplan-Meier method and compared using the logrank test. The multivariate analysis of prognostic markers was performed using the Cox proportional hazards regression model. Differences were considered significant at values of P<0.05.

Abbreviations
HIF-2α: is short for hypoxia-inducible factor-2α; VM: is short for vasculogenic mimicry; EMT: is short for Epithelial-mesenchymal transition; IHC: is short for immunohistochemistry; ChIP is short for Chromatin immunoprecipitation; ECM: is short forextracellular matrix.

Author contributions
JY and JZ conceived and designed the experiments. JY, DMZ and XGZ carried out experiments. NY and YZ participated in statistical analysis. DMZ and DCL took part in interpretation of data. JY and JZ wrote this manuscript. All authors read and approved the final manuscript.