Novel paracrine modulation of Notch-DLL4 signaling by fibulin-3 promotes angiogenesis in high-grade gliomas

Abstract

High-grade gliomas are characterized by exuberant vascularization, diffuse invasion and significant chemoresistance, resulting in a recurrent phenotype that makes them impossible to eradicate in the long-term. Targeting pro-tumoral signals in the glioma microenvironment could have significant impact against tumor cells and the supporting niche that facilitates their growth. Fibulin-3 is a protein secreted by glioma cells, but absent in normal brain, that promotes tumor invasion and survival. We show here that fibulin-3 is a paracrine activator of Notch signaling in endothelial cells and promotes glioma angiogenesis. Fibulin-3 overexpression increased tumor VEGF levels, microvascular density, and vessel permeability, while fibulin-3 knockdown reduced vessel density in xenograft models of glioma. Fibulin-3 localization in human glioblastomas showed dense fiber-like condensations around tumor blood vessels, which were absent in normal brain, suggesting a remarkable association of this protein with tumor endothelium. At the cellular level, fibulin-3 enhanced endothelial cell motility and association to glioma cells, reduced endothelial cell sprouting, and increased formation of endothelial tubules, in a VEGF-independent and Notch-dependent manner. Fibulin-3 increased ADAM10/17 activity in endothelial cells by inhibiting the metalloprotease inhibitor TIMP3; this resulted in increased Notch cleavage and increased expression of DLL4 independently of VEGF signaling. Inhibition of ADAM10/17 or knockdown of DLL4 reduced the pro-angiogenic effects of fibulin-3 in culture. Taken together, these results reveal a novel, pro-angiogenic role of fibulin-3 in gliomas, highlighting the relevance of this protein as an important molecular target in the tumor microenvironment.

High-grade gliomas are the most common primary tumors in the Central Nervous System (CNS) and one of the most aggressive and difficult to treat forms of cancer (1). Glioblastoma (GBM), the most common form of adult glioma, has a particularly dismal prognosis with a median survival of approximately 15 months (2). Multiple factors contribute to this poor outcome, including tumor isolation within the CNS, cellular heterogeneity within the tumor, and the highly invasive nature of GBM cells (1).

GBMs are, in addition, among the most vascularized type of tumors (3). Aberrant GBM blood vessels exhibit glomeruli of proliferating endothelial cells, reduced astrocyte and pericyte coverage, and thrombotic obstructions. The resulting capillaries are “leaky” and have abnormal flow, resulting in tumor hypoxia despite the increased microvascular density (4, 5). The local environment around these abnormal blood vessels forms the anatomic and functional niche for glioma stem cells (GSCs), which proliferate and disperse just beyond the reach of conventional therapeutics.

Anti-angiogenic treatments have been an attractive strategy to disrupt the supporting microenvironment of GBM (4, 6). The anti-VEGF antibody bevacizumab is widely used for adjuvant therapy of recurrent GBM, enhancing quality of life and extending progression-free survival. However, results from recent clinical trials suggest lack of significant benefits for bevacizumab in primary GBM (7), highlighting the need for better anti-angiogenic approaches. More importantly, mounting evidence suggests that anti-angiogenic treatments may trigger dispersion of residual tumor, making the treatment of recurrent GBM even more challenging (8). Novel approaches to hinder simultaneously tumor vascularization and invasion, which are supported by multiple common molecular mechanisms, would be a welcome strategic addition for combination therapy of GBM.

GBM invasion results from concerted mechanisms of tumor cell adhesion, motility, and remodeling of the extracellular matrix (ECM). Our laboratory was the first to describe the ECM protein fibulin-3 (gene EFEMP1) in GBMs and to demonstrate the key protumoral role of this extracellular protein in gliomas (9, 10). Fibulin-3 is a matrix protein detected in the skin and elastic tissues but absent from normal brain, including the brain vasculature (11). Expression of fibulin-3 is associated with tumor progression towards a more malignant, metastatic phenotype: this protein is unchanged or downregulated in several primary solid tumors but upregulated in the late, metastatic stages of carcinomas (12, 13). In gliomas, fibulin-3 expression correlates with tumor grade and is particularly increased in primary and recurrent GBMs. Fibulin-3 promotes tumor invasion and supports survival of GBM cells challenged by apoptotic stimuli. Mechanistically, this protein promotes activation of Notch signaling (9), being one the first paracrine activators of this pathway described in cancer.

The Notch pathway is a highly conserved signaling mechanism that involves activation of Notch receptors by ligands from the Delta-Like and Jagged families located on adjacent cells. This pathway is highly active in gliomas (14), where it has been correlated with tumor cell proliferation, survival to apoptosis, self-renewal of tumor stem cells, and invasion (15). In addition, activation of Notch in endothelial cells by its ligand DLL4 regulates the formation of tumor blood vessels (16) and is critical to stabilize a functional vascular network (17). Notch and DLL4 are upregulated in subsets of GBMs presenting an angiogenic phenotype (18), and Notch-DLL4 signaling has been shown to mediate glioma resistance to anti-VEGF therapy (19). Whether fibulin-3 could regulate Notch-DLL4 signaling and be involved in glioma vascularization is unknown.

We demonstrate here that fibulin-3 promotes vascularization of high-grade gliomas in vivo and may play a role in the association of glioma cells to blood vessels for tumor growth and dispersion. Moreover, we show that fibulin-3 activates endothelial Notch-DLL4 signaling independently of VEGF and through a novel TIMP3/ADAM-dependent mechanism, resulting in increased pro-angiogenic behavior of endothelial cells.

Materials and methods

Cells and tissue specimens

The rat GBM cell line CNS1 and human GSCs GBM8 and GBM34 were cultured as previously described (9, 10). GSCs were validated for self-renewal, tumorigenicity in low numbers, and multi-lineage differentiation, and were always used in early passages. Primary cultures of human brain microvascular endothelial cells (HBMECs) were prepared from normal brain tissue obtained during expedited autopsy and cultured in RPMI-1640 with 10% fetal bovine serum, 2 mM L-glutamine, 1 mM sodium pyruvate, 20 mM HEPES, 50 U/ml penicillin and 50 μg/ml streptomycin. The cells used in this study were originally described as HBMEC-3 (20) and validated for expression of factor VIII-reactive antigen and uptake of acetylated low density lipoprotein. We further validated HBMECs for expression of endothelial markers and negligible expression of fibulin-3 (Figure S1), and authenticated them by short-tandem repeat DNA profiling (Idexx Bioresearch, Columbia MO). Human umbilical vein endothelial cells (HUVECs) were obtained from American Type Culture Collection (Manassas VA) and cultured in the same medium as HBMECs. Human GBM tissues were provided by the Cooperative Human Tissue Network, funded by the National Cancer Institute. Normal brain tissues were provided by the Brain and Tissue Bank for Developmental Disorders, funded by the National Institute of Child Health and Human Development.

DNA constructs, lentivectors and siRNAs

Full-length fibulin-3 cDNA (gene EFEMP1, Genbank # BC098561.1) was cloned into pcDNA3.1(+) (Life Technologies, Carlsbad CA) and then subcloned into the lentiviral expression vector pCDH-EF1-copGFP (System Biosciences, Mountain View CA) for constitutive expression in glioma cells (10). Full-length TIMP3 cDNA (Genbank # BC014277.2) was cloned into pcDNA3.1-V5/6xHis for transient expression. Fibulin-3 shRNAs have been previously described and validated; these shRNAs were cloned into the lentiviral vector pLVET-tTR-KRAB-eGFP for doxycycline-dependent co-expression of shRNAs and eGFP, as described (9). A Notch-reporter construct carrying firefly luciferase under control of 4xCBF1 binding elements (pGL2Pro-CBF1-Luc) and the truncated, constitutively active mouse Notch-1 intracellular domain (pSG5-FLAG-NICD) have also been described (9). SiRNA oligonucleotides against DLL4 were purchased from Qiagen (Valencia, CA) and validated at the mRNA and protein levels (Figure S1).

Animal procedures

All animal experiments were approved by the Institutional Animal Care and Use Committees at The Ohio State University and Harvard Medical School/Brigham and Women's Hospital. A total of 7.5×10⁴ GFP-expressing CNS1 cells (2.5×10⁴ cells/μl) were implanted intracranially in the striatum of Lewis rats as previously described (10) and allowed to form tumors for 15 days before histological processing. To assess vessel leakiness animals were injected in the tail-vein with 200 μl of 1 mg/ml tetramethylrhodamine-labeled dextran (average molecular weight 155 kDa, Sigma-Aldrich, St. Louis MO), euthanized after 5 minutes, and the brain immediately fixed and processed for cryosectioning. Human GSCs were first transduced with the lentiviral vector carrying inducible fibulin-3 shRNA/eGFP and selected by fluorescence as described (9). A total of 5×10⁴ cells (2.5×10⁴ cells/μl) were implanted in the striatum of athymic mice and the shRNA induced with 1 mg/ml doxycycline in the drinking water, starting 3 days after tumor implantation (9). Animals were euthanized and tumors harvested for cryosectioning and histological analysis 21 days after implantation. Blood vessels were stained with endothelial markers (CD31 in mouse and RECA1 in rat) and analyzed within the tumor boundaries (identified by GFP fluorescence). Vessels were identified as elongated structures of arbitrary length >25 μm, using the image analysis software ImageJ, and quantified in at least 3 separate sections per tumor. In some cases tumors and contralateral brain tissue were recovered after 14 days and dissociated with collagenase/hyaluronidase as described (21). Brain microvascular endothelial cells were isolated from these homogenates by flow cytometry, using fluorescently-labeled anti-CD31 antibody.

Cell culture experiments

For cell-cell adhesion experiments, CNS1 cells were seeded at 5×10⁴ cells/well in 48 well plates and allowed to form monolayers overnight. HBMECs (1×10⁴ cells/well) labeled with the red-fluorescent cell tracker CM-DiI (Life Technologies) were added to the monolayers the following day and washed off with phosphate buffered saline (PBS) after 2 h. The number of red-fluorescent adhered cells per well was counted using fluorescence microscopy. For cell migration experiments, HBMECs (5×10³ cells/well) were seeded in Transwell™ inserts (8 μm pore) pre-coated on their underside with fibronectin (10 μg/ml) to induce cell motility. Cells were then allowed to migrate towards CNS1 cells or purified fibulin-3 (300 ng/ml, Origene Technologies, Rockville MD) for 8 h. Cells were subsequently fixed with methanol, stained and counted. For invasion experiments, the Transwell inserts were first covered with 50 μl Matrigel™ (7.5 mg/ml) and cells were allowed to migrate through this barrier for 24 h. For co-culture experiments, fluorescent HBMECs (5×10³ cells/well) were first seeded in 24-well plates containing an aligned-nanofiber substrate (Nanofiber Solutions, Columbus OH) that restricted cell movement and facilitated cell tracking (22). Glioma cells (5×10⁴ cells/well) were seeded on Transwell inserts (0.4 μm pore) hanging on top of the nanofibers. Both cell types were co-cultured in HBMEC culture medium overnight. HBMECs were imaged by time-lapse fluorescence microscopy using an Olympus IX81 microscope adapted with an environmental chamber. Migration of individual HBMECs was quantified using the software ImagePro 6.1 (Media Cybernetics, Rockville MD). For tubulogenesis assays, HBMECs were seeded at 1×10⁴ cells/well in 96-well plates pre-coated with 50 μl Matrigel and allowed to aggregate overnight. The following day, cells were stained with Calcein-AM (Life Technologies) and imaged by fluorescence microscopy. Tubules were identified and analyzed using the software ImageJ loaded with the Skeletonize and AnalyzeSkeleton plugins. For endothelial cell sprouting assays we followed a previously described method (23). Briefly, HBMECs (5×10⁴ cells/well) were cultured in ultra-low attachment 96-well plates (Corning Life Sciences, Tewksbury, MA) to form floating spheroids of 250-300 μm diameter. Spheroids were manually seeded in 96-well plates on a Matrigel substrate (50 μl/well) and incubated for 5 to 6 days to measure total cell sprouting from each spheroid. In all experiments, transfection with cDNAs or siRNAs was performed 48 h before cell culture assays.

Biochemical assays

Cells were recovered from culture, lysed and processed for Western blotting using standard protocols (antibodies are listed in Supplementary Table I). For semi-quantitative RT-PCR, cells were processed using Trizol reagent (Life Technologies) and total RNA was purified by ethanol precipitation (primers are listed in Supplementary Table II). For Notch-reporter assays, HBMECs were transfected with the Notch-reporter construct and Renilla luciferase as loading control (9). Reporter-transfected cells were treated with purified fibulin-3 for 16 h and processed to quantify luciferase activity. To measure alpha-secretase (ADAM10/17) activity, HBMECs were lysed in 50 mM Tricine buffer (pH 7.5) containing 100 mM NaCl, 10 mM CaCl₂, 1 mM ZnCl₂, and 0.1% Triton X-100. Lysates (50 μg total protein) were incubated with a fluorogenic ADAM10/17 substrate peptide (TACE substrate III, 10 μM; R&D systems, Minneapolis, MN) and development of fluorescence was followed with a microplate reader as recommended by the peptide manufacturer. Cultures were treated for 1 to 16 h with purified fibulin-3 (300 ng/ml), purified TIMP3 (1 μg/ml, Sigma-Aldrich), gamma-secretase inhibitor DAPT (25 μM, Tocris Bioscience, Bristol UK) and alpha-secretase inhibitor TAPI-2 (1 μM, Cayman Chemical, Ann Arbor MI)

Immunohistochemistry

Frozen human tissues were mounted in cryo-protectant and sectioned at 20 μm. Sections were allowed to air-dry and then fixed for 20 minutes at room temperature using buffered acidic alcohol (8.5 mM sodium acetate buffer, pH 5, in 90% ethanol). Fixed sections were washed with PBS, blocked with PBS containing 5% normal goat serum and 0.1% Triton X-100, and incubated with primary antibodies (16 h) and secondary antibodies (2 h) following standard procedures. Stained sections were imaged using a confocal microscope Zeiss LSM710.

Statistics

All in vitro experiments were repeated at least in triplicate with three independent replicates per group. Results are shown as means ± S.E.M. Animal studies were performed with five animals per experimental condition. Multivariate results were analyzed by one- or two-factor ANOVA with Tukey's multiple comparison test. A value of p<0.05 was taken to indicate statistically significant differences.

Results

Fibulin-3 expression correlates with tumor vascularization in rodent models of high-grade glioma

We have previously shown that fibulin-3 correlates with overall size and invasion of intracranial gliomas: Fibulin-3-overexpressing gliomas are significantly larger and more invasive than controls (10), while tumors with downregulated fibulin-3 are smaller and less disperse (9). Interestingly, gross inspection of these tumors often showed increased hemorrhage in fibulin-3-overexpressing tumors and reduced bleeding in fibulin-3-deficient tumors, independently of their size, which prompted the hypothesis that fibulin-3 could also regulate tumor vascularization.

To test this hypothesis, we first analyzed microvascular density of CNS1-derived tumors implanted in their syngeneic hosts (Lewis rats). Fibulin-3-overexpressing tumors exhibited significantly higher microvascular density and increased average vessel length (Figure 1A-B). Identical results were obtained with the same tumors implanted as xenografts in nude mice (data not shown). Furthermore, intravenous injection of fluorescent dextran showed remarkable accumulation of extravascular dextran in the parenchyma of fibulin-3-overexpressing tumors (Figure 1C-E), indicating an increase of vessel permeability typical of aberrant tumor neovasculature.

Figure 1. Fibulin-3 correlates with increased tumor vascularization and vessel permeability.

CNS1-derived gliomas were processed for immunohistochemistry to identify blood vessels within the tumor parenchyma. In separate experiments, animals were injected with rhodamine-labeled dextran to quantify vessel leakiness. A) Representative image of control and fibulin-3-overexpressing (fibulin-3) tumor sections stained with RECA1 (bar: 100 μm). B) Quantification of microvascular density, stratified by vessel length (* p<0.05; ** p<0.01 by repeated measures ANOVA). C) Representative images from two independent control and fibulin-3-overexpressing tumors, showing total tumor distribution (dashed line) and dextran accumulation (arrows) in the tumor parenchyma. D) Quantification of integrated dextran fluorescence (IF) along the anteroposterior axis of the tumors; each line represents a separate animal (RU, relative units). All sections were imaged under the same illumination conditions by automated imaging software. E) Average signal of fluorescent dextran in control and fibulin-3-overexpressing tumors (p=0.0175 by Student's t-test).

Conversely, GSCs transduced for conditional knockdown of fibulin-3 resulted in tumor xenografts that were visually less hemorrhagic (Figure 2A) and had significantly reduced vascular density in the tumor parenchyma and borders (Figure 2B-C). Taken together, results from fibulin-3 overexpression and knockdown strongly suggested that fibulin-3 secreted by glioma cells is a positive regulator of tumor vascularization.

Figure 2. Knockdown of fibulin-3 correlates with reduced tumor vascularization.

Human glioma stem cells (GBM8 and GBM34) were transduced for doxycycline-dependent induction of fibulin-3 shRNA and implanted intracranially. Fibulin-3 knockdown was initiated after surgery and tumors were collected after 21 days. A) Representative GBM34 tumors showing differences in gross hemorrhage between controls (control shRNA) and fibulin-3-knockdown tumors (fibulin-3 shRNA). Arrows indicate the approximate position of the fibulin-3-deficient tumors. B) Representative images of control and fibulin-3-knockdown tumors showing the tumor parenchyma (green) and vessels stained with anti-CD31 (red). B) Quantification of blood vessel density and stratification of density by vessel length (* p<0.05; ** p<0.01; *** p<0.001 by repeated measures ANOVA).

Fibulin-3 forms a fibrillar wrapping around glioma blood vessels

Although fibulin-3 can be detected throughout the tumor parenchyma (9), the results above prompted us to investigate if this protein could have preferential localization near tumor vessels. Indeed, confocal imaging of human GBM sections with well-preserved vasculature revealed that fibulin-3 was intensely localized around tumor blood vessels (Figure 3A-B) but was absent in normal brain vessels (Figure 3C), even in brain tissue adjacent to tumor. Higher magnification images revealed well-defined fibulin fibrils that tightly wrapped around small tumor capillaries (Figure 3D) and larger vessels (Figure 3E-F and movie S1). These striking fibrils were detected with different antibodies that recognize non-overlapping domains of fibulin-3 and do not cross-react with other perivascular fibrillar proteins (e.g., elastin or fibrillin), indicating that they were not staining artifacts.

Figure 3. Fibulin-3 associates with blood vessels in human GBMs.

Fresh-frozen sections of GBM and normal human brain cortex were briefly fixed and processed for immunohistochemistry. A-C) Sections of primary GBM (A), recurrent GBM (B) and age-matched normal brain (C) were probed with a rabbit anti-fibulin-3 antibody (Ab3911) and tomato lectin to detect vessels (left), or with a mouse monoclonal anti-fibulin-3 antibody (mAb3-5) and anti-CD31 (right). Notice the intense staining of blood vessels in both cases (bars: 200 μm). Cell nuclei were counterstained with DAPI (blue). Punctate, non-specific green fluorescence was observed in normal brain tissue in absence of primary antibodies, likely due to lipofuscin accumulation. D-F) High-magnification images revealed a striking wrapping of fibulin-3 fibrils around single capillaries (D) and larger blood vessels (E) (bars: 50 μm). F) Orthogonal projections of the image in (E) show the fibrils associated with endothelial cells..

We were intrigued about the source of these fibrils because fibulin-3 is absent in normal brain endothelium. Analysis of fibulin-3 mRNA in HBMECs, which do not express a detectable amount of fibulin-3 (Figure S1), showed that their fibulin-3 expression was unchanged during co-culture with glioma cells (Figure S2). Moreover, microvascular endothelial cells recovered from GBM-bearing mice revealed that mouse fibulin-3 mRNA expression was also extremely low and undistinguishable between endothelial cells from the tumor, the contralateral brain side, or naïve mouse brain tissue (Figure S2). Together, these results strongly suggested that fibulin-3 in human GBM is likely secreted by GBM cells and accumulates around tumor blood vessels, where it could cause the pro-angiogenic effects described above.

Fibulin-3 promotes endothelial cell migration and tubulogenesis

We next investigated if fibulin-3 secreted by glioma cells could induce pro-angiogenic behavior in brain microvascular endothelial cells.

HBMECs co-cultured with CNS1 cells showed increased rates of migration and invasion towards fibulin-3-overexpressing glioma cells compared to controls (Figure 4A-B). Time-lapse videomicroscopy specifically demonstrated increased velocity of individual HBMECs co-cultured with fibulin-3-overexpressing CNS1 cells (Figure 4C). Moreover, HBMEC had increased adhesion to fibulin-3-overexpressing glioma cells (Figure 4D) and, reciprocally, these glioma cells showed increased adhesion to pre-formed HBMEC tubules (Figures 4E and S3). Finally, conditioned medium from fibulin-3-overexpressing glioma cells increased number and length of HBMEC tubules in vitro (Figure 4F) but had negligible effects on endothelial cell proliferation (data not shown). Taken together, these results suggested that paracrine stimulation from fibulin-3-secreting glioma cells increased the overall ability of HBMEC to migrate, form tubular structures, and associate with glioma cells, all of which likely contributed to the increased vascularization observed in vivo.

Figure 4. Fibulin-3 promotes angiogenic behavior in endothelial cells.

A-B) HBMECs cultured in Transwell inserts showed increased migration (A) and invasion through matrigel (B) towards fibulin-3 expressing CNS1 glioma cells. C) HBMEC motility on nanofiber scaffolds was quantified using time-lapse fluorescence microscopy; average cell velocities were 10.97 ± 0.19 μm/h for HBMEC exposed to control glioma cells (N=121) and 16.52 ± 0.29 μm/h for HBMEC exposed to fibulin-3-expressing cells (N=109; p<0.01 by paired t-test). D) Fluorescently-labeled HBMECs added to monolayers of CNS1 cells for 2h adhered more to fibulin-3-expressing glioma cells than control cells. E) Similarly, fibulin-3-expressing glioma cells adhered more than control cells to pre-formed HBMEC tubules. F) HBMECs cultured on Matrigel formed longer tubules in presence of conditioned medium from CNS1 overexpressing fibulin-3 (CNS1-fibulin3); bars: 300 μm. G) VEGF-A mRNA expression increased in CNS1 cells transfected with fibulin-3 cDNA or treated with purified fibulin-3 (300 ng/ml) overnight. H) VEGF-A mRNA was also higher in tumors derived from fibulin-3-overexpressing CNS1 compared to control CNS1 (N=3/group); VEGF levels in contralateral brain tissue were taken as baseline. (* p<0.05; **p<0.01 by repeated measures two- way ANOVA). I-J) HBMECs cultured in Transwell inserts showed increased chemotaxis towards purified fibulin-3 (I) while HBMEC cultured on Matrigel showed increased tubule length when treated with fibulin-3 overnight (J). For experiments A-G: * p<0.05; ** p<0.01; *** p<0.001 by Student's t-test for each assay. K) A tubulogenesis experiment was performed in presence of purified fibulin-3 and increasing concentrations of the pan-VEGFR inhibitor axitinib. Fibulin-3 promoted tubule elongation even at concentrations of axitinib that abolished control tubulogenesis (> 5 nM). Asterisks indicate significant differences for each treatment against maximum inhibition at 10 nM axitinib (** p<0.01; ***p<0.001 by 2-way ANOVA and Tukey's multiple comparison test).

One possible explanation of these results would be increased VEGF expression in glioma cells transfected with fibulin-3 cDNA. Indeed, VEGF mRNA levels were higher in fibulin-3-overexpressing CNS1 cells (Figure 4G) and their corresponding intracranial gliomas (Figure 4G) compared to controls. Therefore, we re-examined the angiogenic behavior of HBMECs using highly purified fibulin-3. Remarkably, soluble fibulin-3 was sufficient to increase HBMEC migration (Figure 4I) and tubule formation (Figure 4J) in absence of VEGF from glioma cells. Moreover, fibulin-3-enhanced tubulogenesis was maintained in presence of concentrations of the VEGFR inhibitor axitinib that abolished basal tubule formation (Figure 4K). Together, these results suggested that, while fibulin-3 increases VEGF levels in gliomas, its pro-angiogenic effects are, at least in part, independent of VEGF.

Fibulin-3 activates DLL4-Notch signaling to promote angiogenic behavior of endothelial cells

Since Notch signaling is a critical mechanism that regulates tumor vascularization and fibulin-3 is a paracrine activator of Notch in glioma (9), we next explored if this pathway would be the mediator of fibulin-3 in our model.

Fibulin-3 caused remarkable activation of Notch signaling in HBMECs as shown by increased production of the Notch intracellular domain (NICD, Figure 5A), activation of a NICD-dependent reporter construct (Figure 5B) and increased expression of Notch-regulated genes (Figure 5C). These effects of fibulin-3 were inhibited by the gamma-secretase inhibitor DAPT, which blocks Notch cleavage and production of NICD. DAPT also abolished fibulin-3-stimulated migration of HBMECs (Figure 5D) and tubulogenesis (Figure 5E). Since fibulin-3 is not expressed in HBMECs, fibulin-3 knockdown experiments were instead performed in HUVECs, resulting in significant decrease of Hes5 and MASH mRNAs (Figure S4). Together, these results indicated that activation of endothelial Notch signaling is the likely mechanism underlying the pro-angiogenic effects of fibulin-3.

Figure 5. Fibulin-3 activates endothelial Notch to promote tubulogenesis.

A) Western blotting of HBMECs incubated overnight in presence of fibulin-3 (300 ng/ml) showed increased Notch-1 cleavage and expression of Notch-regulated genes (Hes1, Hes5); these effects were inhibited by the gamma-secretase inhibitor DAPT (25 μM) (NICD: Notch-1 intracellular domain). B) HBMECs were transiently transfected with a Notch-reporter plasmid and treated with fibulin-3 and DAPT as in (A). Results show increased Notch activation by fibulin-3. A truncated, constitutively active fragment of Notch (FLAG-NICD) was co-transfected with the reporter plasmid as positive control (*** p< 0.001 by 1-way ANOVA and Bonferroni's post-hoc test). The inset shows expression of endogenous NICD (~110 kDa) and FLAG-NICD (~68 kDa) in the cells. C) Semiquantitative RT-PCR of fibulin-3-treated HBMECs showed significant increase of Notch-regulated genes (** p<0.01; *** p<0.001 by Student's t-test for each gene). D) Increased chemotaxis of HBMECs towards fibulin-3 was abolished by DAPT. E) Similarly, DAPT also prevented the enhancing effect of purified fibulin-3 on endothelial tubule elongation. Results in D-E: ** p<0.01; *** p<0.001, by two-way ANOVA.

To further understand this mechanism we focused on the major endothelial Notch ligand, DLL4. Treatment with fibulin-3 increased the expression of DLL4 in HBMECs and glioma cells (Figure 6A-B). Fibulin-3 also increased the mRNA level of the transcription factor Foxn4 that is known to upregulate DLL4 expression (24), but not the transcription factor Klf4 that is a negative regulator of DLL4 (25)(Figure 6A). Furthermore, knockdown of DLL4 reduced HBMEC tubulogenesis and abolished the pro-tubulogenic effect of fibulin-3 (Figure 6C). These results suggested that the effects of fibulin-3 not only required activation of Notch receptors but also the specific presence of the Notch ligand DLL4.

Figure 6. The pro-angiogenic effects of fibulin-3 require endothelial DLL4.

A-B) Treatment of HBMEC with fibulin-3 (300 ng/ml) increased mRNA (A) and protein (B) expression of DLL4. Fibulin-3 also increased mRNA expression of the DLL4-regulatory factor Foxn4 but not the transcription factor Klf4 (* p<0.05; *** p<0.001 by two-way ANOVA). C) Transient transfection of DLL4 siRNAs in HBMEC reduced tubulogenesis and prevented fibulin-3 pro-tubulogenic effect (*p<0.05; **p<0.01 by two-way ANOVA). Bars: 300 μm. D) Transient DLL4 knockdown also stimulated HBMEC sprouting, which was significantly reduced, but not completely abolished, by fibulin-3 (** p< 0.01; *** p<0.001 by two-way ANOVA). Bars: 120 μm.

Loss of DLL4 also caused significant effects when we evaluated the effect of fibulin-3 on endothelial cell sprouting, although not as marked as in the tubulogenesis assays. DLL4 knockdown caused a drastic increase in HBMEC sprouting (Figure 6D), as expected and described elsewhere (23). This time fibulin-3 was able to antagonize in part the effect caused by loss of DLL4 while having no effect on sprouting on its own, suggesting that fibulin-3 was still able to activate Notch signaling or reduce sprouting by a different mechanism. We interpreted these results as indication that fibulin-3 mechanism involves primarily Notch activation in HBMECs, while its effects on vessel lengthening require DLL4-Notch signaling in formed tubules.

Activation of Notch-DLL4 signaling by fibulin-3 is ADAM10/17-dependent

We finally examined how fibulin-3 was activating Notch signaling in endothelial cells. Treatment of HBMECs with purified fibulin-3 for up to 1h did not have effect on NICD production, DLL4 expression, or activation of VEGFR2 and Akt (not shown). Longer incubations (>= 6h) with purified fibulin-3 did not activate VEGFR2 or Akt but nevertheless resulted in increased production of NICD and increased expression of DLL4 and Hes1 (Figure 7A). These effects were inhibited with DAPT but not with axitinib, suggesting that they were dependent on Notch signaling but not VEGF signaling. Notch signaling depends not only on Notch ligands but also on the enzymes that process the Notch receptors; since fibulin-3 increases metalloprotease activity in GBMs (10), we hypothesized that this protein could also regulate secretase activity to increase Notch activation.

Figure 7. Fibulin-3 activates ADAM10/17 to trigger DLL4-Notch signaling.

A) HBMECs treated with fibulin-3 (300 ng/ml, 6h) did not show activation of VEGFR2 or Akt but exhibited increased production of NICD, upregulation of DLL4 and Hes1, and activation of Erk1/2. These effects were blocked by preincubation with DAPT (25 μM) but not by axitinib (5 nM). B) Incubation of HBMECs with the ADAM10/17 inhibitor, TAPI-2, prevented the enhancing effects of fibulin-3 on expression of DLL4 and Foxn4. C) TAPI-2 also blocked fibulin-3-dependent increase of Notch cleavage, Hes1 expression, and Erk1/2 activation. D) Incubation of HBMEC lysates with purified fibulin-3 dramatically increased ADAM10/17 activity on a fluorogenic substrate (RFU: relative fluorescent units); this effect was abolished by purified TIMP3 (1 μg/ml). E) Purified TIMP3 prevented the enhancing effect of fibulin-3 on NICD production and expression of DLL4 and Hes1. F) Similarly, TIMP3 abolished the pro-tubulogenic effect of fibulin-3. Results in (B) and (F): ** p< 0.01; *** p<0.001, by one-way ANOVA.

We specifically focused on the alpha-secretase activity that must precede gamma-secretase cleavage during activation of Notch receptors. Alpha-secretase activity is mediated by the membrane-bound enzymes ADAM10 and ADAM17, which can be specifically inhibited with TAPI-2. Strikingly, preincubation of HBMEC with TAPI-2 prevented all enhancing effects of fibulin-3 on Notch-DLL4 signaling (Figure 7B-C): In presence of TAPI-2 fibulin-3 did not upregulate Foxn4 and DLL4 mRNAs, and did not increase NICD levels or Hes1 expression. These results confirmed that the effects of fibulin-3 were mediated by activation of Notch receptors and could be blocked by inhibitors of the different Notch-processing enzymes.

One of the few confirmed ligands of fibulin-3 is the extracellular metalloprotease inhibitor TIMP3 (26), which can also bind ADAM17 to regulate angiogenesis (27). Using a co-immunoprecipitation assay we first confirmed that purified fibulin-3 could bind secreted TIMP3 in our model (Figure S5). Next, using a fluorogenic assay to measure ADAM10/17 activity in HBMEC lysates, we observed that exogenous fibulin-3 caused a dramatic increase in this enzymatic activity, which was completely abolished by TIMP3 (Figure 7D). Moreover, purified TIMP3 was sufficient to prevent the enhancing effect of fibulin-3 on DLL4 expression and Notch activation (Figure 7E). Finally, TIMP3 also prevented the pro-tubulogenic effect of fibulin-3 in HBMECs (Figure 7F). Together, these results strongly suggested that fibulin-3 likely activates Notch-DLL4 signaling by binding TIMP3 and inhibiting its ability to inactivate ADAM10/17, leading to increased cleavage of Notch and expression of DLL4 that result in the observed pro-angiogenic effects.

Discussion

The contribution of fibulin family members to cancer development, and in particular to tumor angiogenesis, is complex and depends on the tumor type and model analyzed (28). Both tumor-promoting and tumor-suppressive effects have been described for fibulin-1 (29, 30), fibulin-2 (31, 32), fibulin-3 (10, 12, 33) and fibulin-5 (30, 34). This multiplicity of effects has been attributed to the presence of multiple isoforms within the family, with different functions and differential regulation by extracellular signals (35).

Fibulin-3 in particular has been described as an enhancer of tumor progression and metastasis in pancreatic, cervical and ovarian carcinomas (12, 36-38), and a potential tumor-suppressor in colorectal, lung and hepatic cancers (39, 40). We provided the first comprehensive description of fibulin-3 in human gliomas (10), its paracrine Notch-activating mechanism, and its ability to promote tumor growth and invasion (9, 10).

Our present results demonstrate that the effects of fibulin-3 on the glioma microenvironment are even more extensive: fibulin-3 secreted by glioma cells can activate endothelial Notch-DLL4 signaling in a paracrine manner and promote angiogenic behavior. This is remarkable because fibulin-3 was originally reported as an anti-angiogenic protein (41) due to the in vitro use of a short recombinant form lacking the critical Notch-activating, N-terminal domain (9). This short isoform, which was hypothesized but never observed in vivo (42), could also explain the lack of pro-tumor effects of fibulin-3 in another glioma study (43), highlighting how critical the Notch pathway is to mediate fibulin-3 functions in these tumors. Several studies have since demonstrated that fibulin-3 upregulation is in fact associated with angiogenesis in carcinomas (36-38). Moreover, fibulin-3 accumulation has long been known to correlate with vascular proliferation in age-related macular degeneration in the retina (44), suggesting that this protein may have a pro-angiogenic role beyond the context of cancer.

Our study conclusively identifies fibulin-3 as a pro-angiogenic signal secreted by glioma cells, encouraging future work to further elucidate its molecular mechanisms. We have shown here that fibulin-3 activates ADAM10/17, likely by inhibiting TIMP3, and that this mechanism then promotes Notch activation and DLL4 expression, both of which are needed to mediate the pro-angiogenic effects of fibulin-3. Blockade of DLL4 upregulation by TAPI-2 (Figure 7B) and DAPT (Figure 7A) suggest that increased expression of this Notch ligand (and its regulatory factor Foxn4) would depend at least in part on Notch activation and could form a feed-forward loop as previously proposed (45). Whether fibulin-3 activity is, in addition, sufficient to promote ligand-independent activation of Notch receptors as proposed in certain models (46) is still uncertain but is nevertheless an attractive hypothesis in the context of aberrant receptor activation in glioma. While the molecular mechanisms of fibulin-3 are not yet fully defined, our results show robust modulation of Notch signaling and overlapping effects (i.e. VEGF upregulation, Notch activation and DLL4 upregulation) that make fibulin-3 a strong positive modulator of tumor angiogenesis. Fibulin-3 could therefore be a major target to help disrupt Notch-DLL4 signaling, with significant impact in strategies to overcome anti-VEGF resistance (47). Moreover, since fibulin-3 has an established pro-invasive role in gliomas (9, 10), targeting this protein could disrupt not only angiogenic compensation mechanisms but also dispersion mechanisms in gliomas treated with anti-VEGF. As an additional benefit, targeting the pro-angiogenic role of fibulin-3 could help devise strategies for eye vascular diseases where this protein is clearly involved.

Finally, we consider important to remark on the distinctive fibulin-3 fibrils wrapping the blood vessels of human GBMs. Expression of fibulin-3 in adult organs is diffuse and restricted to the basal lamina of connective tissues (11). A pattern of vascular fibrils has been observed in embryonic lung tissue (48) but never in the CNS, underscoring the unique association of fibulin-3 with the wall of glioma blood vessels. Fibulins −4 and −5, which share considerable homology with fibulin-3, form perivascular fibrils by association with elastin and fibrillin-1, but fibulin-3 does not associate with these proteins (48, 49). This suggests that unique docking molecules, possibly restricted to embryonic and tumor-associated endothelial cells, may account for this unusual pattern of fibulin-3 that could help associate glioma cells to the tumor vasculature. Interestingly, purified fibulin-3 can self-polymerize in vitro (Viapiano Lab, unpublished observation), which could contribute in part to the formation of the fibrils.

In sum, the localization of fibulin-3 around glioma blood vessels, together with its promotion of angiogenesis, association of glioma and endothelial cells, and glioma cell dispersion, suggest that this matrix protein may play multiple key roles in tumor progression, enhancing both formation of new vessels and co-option of existing vessels by dispersing tumor cells. We propose fibulin-3 as a relevant target restricted to the tumor parenchyma that can be exploited to disrupt Notch-dependent pro-angiogenic and pro-invasive mechanisms in glioblastoma.