JCI - CD93 promotes β1 integrin activation and fibronectin fibrillogenesis during tumor angiogenesis

CD93 and MMRN2 interact in human endothelial cells and colocalize in tumor vessels. To investigate the mechanisms through which CD93 regulates vascular function, interacting proteins were identified by coimmunoprecipitation from human endothelial cell protein extracts using a monoclonal anti-CD93 antibody followed by mass spectrometry. Analysis of 5 independent experiments identified the endothelial-specific secreted matrix-binding protein MMRN2 as a potential CD93-binding molecule. MMRN2 was identified in all 5 CD93 coimmunoprecipitated samples and never in samples coimmunoprecipitated with IgG. The CD93-MMRN2 interaction was confirmed by Western blot analysis of CD93 coimmunoprecipitated samples using antibodies directed against MMRN2 (Figure 1A). In situ proximity ligation assay (PLA) technology was used to verify CD93-MMRN2 interaction in human endothelial cell monolayers. Positive PLA signals indicating formation of CD93-MMRN2 complexes were homogeneously distributed in endothelial cells (Figure 1B). The specificity of the PLA assay was confirmed by quantification of CD93-MMRN2 PLA signals per cell, comparing with signals in negative controls (Figure 1C).

CD93 is highly expressed in the tumor vasculature of human high-grade gliomas (15) as well as in tumor vessels in the orthotopic murine GL261 glioma model (11). To determine whether the observed interaction between CD93 and MMRN2 is likely to occur in tumor vessels, we examined the expression pattern of MMRN2 in tumors. MMRN2 was expressed in CD31-positive tumor vessels of human glioblastoma (grade IV glioma), colocalizing with CD93 expression (Figure 1D). Analysis of y-z 3D stack of the grade IV glioma vessels revealed that MMRN2 and CD93 were expressed at the abluminal side of the CD31-positive glioblastoma vessels (arrowheads in Supplemental Figure 1, A and B, respectively; supplemental material available online with this article; https://doi.org/10.1172/JCI97459DS1), indicating that the interaction between MMRN2 and CD93 occurs abluminally in proximity to extracellular matrix (arrowheads in Supplemental Figure 1, C and D). Similarly, MMRN2 was highly expressed in murine GL261 glioma vasculature, colocalizing with CD93 (Figure 1E). A significantly lower basal expression of MMRN2 colocalizing with CD93 was observed in the brain vasculature adjacent to the GL261 tumor (Figure 1F and quantification in Figure 1G).

CD93 colocalizes with MMRN2 during retinal angiogenesis and regulates filopodia formation and vessel sprouting. To further understand the role of CD93 in sprouting angiogenesis, we analyzed the developing vasculature in mouse retina. CD93 was expressed in the sprouting front of the postnatal day 6 (P6) mouse retinas, including the filopodia protrusions, colocalizing with the endothelial marker isolectin B4 (Figure 2A). MMRN2 colocalized with CD93 in the retinal plexus and sprouting front, but colocalization was not detectable in filopodia extensions (high magnification, Figure 2, A and B). Instead, MMRN2-positive signals were found in the extracellular matrix surrounding the filopodia at the vascular front (arrowheads, Figure 2A) and within the vascular plexus (arrowheads, Figure 2B). This indicates that MMRN2 is not present within the filopodia, but interacts with CD93 in filopodia after being secreted. To investigate whether loss of CD93 affects retinal angiogenesis, we analyzed P6 retinas from CD93-deficient (CD93^–/–) mice and WT littermates. The radial expansion of the vascular plexus was similar in CD93^–/– and WT retinas (Figure 2C, quantified in Figure 2F). However, the mean length of the sprouts in the angiogenic front was significantly reduced in the CD93^–/– retinal vasculature in comparison with WT littermates (Figure 2D, quantified in Figure 2G). In addition, a significant reduction in filopodia protrusions was observed in CD93^–/– mice compared with WT mice (Figure 2E, quantified in Figure 2H). No differences were observed when the sprout length and the number of filopodia were compared between WT and CD93 heterozygous mice (CD93^+/–; Figure 2, G and H). The importance of CD93 in filopodia formation was further analyzed through siRNA-mediated knockdown of CD93 in human dermal blood endothelial cells (HDBECs). In line with a reduced number of filopodia in CD93^–/– P6 retinas, sparsely seeded CD93 siRNA-treated endothelial cells formed fewer filopodia than control cells (Supplemental Figure 2, A and B). These data indicate that CD93 regulates filopodia formation and the extension of endothelial sprouts during angiogenesis.

MMRN2 protects CD93 from proteolytic cleavage. To investigate whether the interaction between CD93 and MMRN2 affects the expression or stability of either protein, CD93 and MMRN2 mRNA and protein levels were examined in HDBECs after siRNA-mediated knockdown of CD93 or MMRN2. Transfection of endothelial cells with siRNA targeting CD93 reduced CD93 mRNA expression and protein level by approximately 95% compared with controls (Figure 3A). An equally high knockdown efficiency was obtained using siRNAs against MMRN2 (Figure 3B). siRNA-mediated knockdown of CD93 did not affect MMRN2 mRNA or protein levels (Figure 3, C and E). Similarly, CD93 mRNA levels were not affected by MMRN2 downregulation (Figure 3D). However, Western blot analysis revealed a significant reduction of CD93 protein levels in siMMRN2-treated cells compared with controls (Figure 3F). This suggests that MMRN2 affects CD93 protein stability but not CD93 gene expression.

To determine whether MMRN2 regulates CD93 protein stability in endothelial cells, protein synthesis was blocked with cycloheximide (CHX) treatment over a time course of 0, 30, 60, and 240 minutes in siMMRN2 and control cells, and CD93 protein levels were determined by Western blot analysis. A gradual reduction of CD93 over time was observed in the control conditions (Mock and siCtrl) in response to CHX treatment, indicating a time-dependent turnover of CD93 protein (Figure 3G and graph in Figure 3H). When MMRN2 expression was knocked down by siRNA treatment, the reduction of CD93 protein levels was accelerated (Figure 3, G and H). These results indicate that the interaction with MMRN2 stabilizes CD93 protein in endothelial cells.

To further characterize the mechanism by which MMRN2 regulates CD93 protein levels, we assessed the possible impact of MMRN2 in CD93 degradation, using proteasome and lysosome inhibitors. Inhibition of either proteasomal degradation by 20 μM lactacystin during 8 hours or lysosomal degradation using 10 mM NH₄Cl during 18 hours resulted in accumulation of CD93 protein in controls, indicating that CD93 is degraded through both proteasome- and lysosome-dependent mechanisms (Supplemental Figure 3, A and B). However, the siMMRN2-mediated decrease in CD93 protein level was not restored to control levels by treatment with either proteasome or lysosome inhibitors. These results indicate that the reduction of CD93 protein levels observed after MMRN2 knockdown is not due to enhanced intracellular protein degradation.

CD93 can be cleaved from the plasma membrane in response to inflammatory conditions, releasing a soluble CD93 fragment containing the EGF-like domain (9). We therefore assessed whether the interaction with MMRN2 affected shedding of CD93 from endothelial cells. Conditioned media from control cells and siMMRN2-treated cells were concentrated and analyzed by Western blot to detect soluble CD93 (Figure 3I). Notably, the level of soluble CD93 was increased more than 60% in the conditioned medium derived from siMMRN2-treated cells as compared with the soluble CD93 levels in that derived from control cells (Figure 3J). These results indicate that MMRN2 protects CD93 from cleavage in endothelial cells, and that decreased expression of MMRN2 leads to an increased release of soluble CD93.

MMRN2 stabilizes CD93 in filopodia and in the front edge of migrating endothelial cells. We have previously demonstrated that CD93 is required for endothelial cell adhesion and migration (11). To investigate whether the interaction with MMRN2 decreases cleavage of CD93 in endothelial filopodia, the subcellular localization of CD93 was assessed by confocal microscopy in sparsely cultured HDBECs transfected with siRNA targeting MMRN2 or control siRNA. In line with our observation of CD93 expression in filopodia in endothelial sprouts of the developing retinal vasculature (Figure 2), CD93 was localized in filopodia structures in control cells (Figure 4A). Downregulation of MMRN2 resulted in a striking decrease of CD93 in filopodia. In siMMRN2-transfected cells, CD93 was mainly accumulated in the perinuclear region, and only low amounts were detected in filopodia (Figure 4A). The only partial reduction of CD93 in filopodia did not result in a decreased number of filopodia in siMMRN2-transfected HDBECs, which is consistent with similar numbers of filopodia in the retinas of CD93^+/– and WT mice (Supplemental Figure 4, A and B, and Figure 2, G and H).

To further analyze the localization of CD93 during endothelial cell migration, an in vitro wound healing assay was used, and the localization of CD93 in the leading edge of migrating cells was analyzed by confocal microscopy. A strong CD93 staining was consistently detected in the migrating front and filopodia of control endothelial cells (Figure 4B). In contrast, only low levels of CD93 were observed at the migration front in MMRN2-silenced cells. The CD93-positive signal in siMMRN2-treated endothelial cells was mainly located in the perinuclear region (Figure 4B).

MMRN2 is an extracellular matrix protein secreted by endothelial cells. Therefore, we analyzed whether MMRN2 produced and secreted by control cells can rescue CD93 localization to endothelial filopodia and the migrating front in siMMRN2-transfected cells. The rescue experiment was performed using cell culture inserts. Control cells and siMMRN2 cells were initially cultured in separate wells until confluence. When the culture insert was removed, control and siMMRN2-treated endothelial cells were allowed to migrate for 6 hours, sharing the same culture media (Supplemental Figure 4C). CD93 was strongly localized in the migrating front and filopodia when control endothelial cells were cocultured on each side of the insert. However, a significant reduction of CD93-positive signal was detected in the migrating front when siMMRN2-treated cells were present on both sides of the insert (Figure 4C). Coculture with control cells restored CD93 levels in the migrating front of MMRN2-silenced cells (Figure 4D). CD93-positive signals in the migrating front (area delimited by dotted line in Figure 4, C and D) were quantified, showing a complete rescue of CD93 levels in MMRN2 siRNA-treated endothelial cells when cocultured with control endothelial cells (Figure 4E). Thus, MMRN2 secreted by control cells can stabilize CD93 in the migrating front and filopodia of MMRN2-deficient endothelial cells. This indicates that CD93 in filopodia interacts with MMRN2 in the extracellular matrix. This result was further confirmed by costaining of CD93 and MMRN2 during migration. MMRN2 colocalized with CD93 in the migrating front in the control condition (arrowheads in the high-magnification picture of siCtrl, Figure 4F). Knockdown of MMRN2 resulted in a loss of CD93 in the migrating front (siMMRN2_3 panel, Figure 4F). CD93 localization to filopodia is rescued by treatment of siMMRN2-treated endothelial cells with conditioned medium (CM) derived from control cells. In association, we observed that MMRN2, derived from conditioned medium, bound to endothelial filopodia on siMMRN2-treated cells (arrowheads in siMMRN2_3+siCtrl CM, Figure 4G). As a functional readout, we quantified endothelial migration using the scratch wound assay. Downregulation of MMRN2 (siMMRN2_3 and siMMRN2_6, black dashed lines, Figure 4H) significantly impaired migration at 6 hours after wounding in comparison with control cells (Mock and siCtrl, solid black lines, Figure 4H). This inhibitory effect was reverted by culturing of siMMRN2 cells in conditioned media from control cells containing secreted MMRN2 (siMMRN2+Mock or siCtrl CM, red dashed lines, Figure 4H).

CD93 is predominantly expressed in tip cells, while MMRN2 is distributed throughout endothelial sprouts in embryoid bodies. We used the embryoid body (EB) model of angiogenesis to examine the expression patterns of CD93 and MMRN2 during blood vessel formation. EBs are aggregates of embryonic stem (ES) cells that can be induced to differentiate into various cell types and, when cultured in the presence of VEGF, recapitulate sprouting angiogenesis (16). EBs were grown on a 2-dimensional (2D) glass slide surface or in a 3D collagen I matrix and stimulated with VEGFA to induce sprouting angiogenesis. Immunofluorescent staining of EBs cultured on 2D glass slide surfaces showed that CD93 was exclusively localized in the sprouting front of the vascular plexus with high levels detected in endothelial filopodia (Figure 5A). In contrast, MMRN2 expression was observed throughout the vascular network and colocalized with CD93 in the sprouting front (high magnification in Figure 5A). MMRN2 was specifically expressed in endothelial cells within the retina, consistent with previous publications (17, 18). A similar localization of CD93 and MMRN2 was observed in sprouting vessels formed by EBs cultured in a 3D collagen I matrix (Figure 5B). CD93 was mainly observed in the tip cells of endothelial sprouts, whereas MMRN2 was expressed both in the tip and in the stalk cells, colocalizing with CD93 in the tip cell filopodia (Figure 5B).

CD93 and MMRN2 are required for endothelial sprouting in EBs. To determine the necessity of CD93 and MMRN2 for sprouting angiogenesis, ES cells were transduced with lentiviruses expressing shRNAs designed to knock down CD93 (shCD93) or MMRN2 (shMMRN2). Empty vector–transduced (shCtrl) and untransduced cells (Ctrl) were used as controls. Transfection of ES cells with shRNA targeting CD93 or MMRN2 reduced the mRNA expression of CD93 and MMRN2 by approximately 90% compared with controls in EBs (Supplemental Figure 5, A and B). GFP detection indicating shRNA-transduced cells showed an efficient transduction in 2D and 3D EBs (Supplemental Figure 5, C and D). Knockdown of either CD93 or MMRN2 in ES cells was associated with deficient formation of endothelial sprouts in VEGF-treated EBs (Figure 5, C–F). Indeed, the area of CD31-positive sprouts was significantly decreased in 2D EBs when either CD93 or MMRN2 was knocked down (Figure 5, C and D). Similarly, shCD93 and shMMRN2 EBs cultured in 3D collagen matrix showed a significant inhibition in the number of endothelial sprouts as compared with controls (Figure 5, E and F).

Consistent with a role of secreted MMRN2 in interacting with and stabilizing CD93 during endothelial migration and sprouting, deficient sprouting in 2D shMMRN2 EBs (Figure 5G) could be rescued by coculturing with control EBs (Figure 5, G and H, quantified in Figure 5I). In contrast, endothelial sprouting in CD93-deficient EBs was not rescued by coculturing with control EBs, consistent with a key role of cell surface–bound CD93 in this context (Figure 5, G and H, quantified in Figure 5I). These results support an important role of CD93 and MMRN2 during sprouting angiogenesis.

CD93 or MMRN2 downregulation disrupts fibronectin fibrillogenesis in endothelial cells. Costaining of CD93 and MMRN2 in confluent endothelial cells showed colocalization of these proteins along fibrillar structures distributed on the endothelial monolayer (Figure 6A). CD93 did not colocalize with filamentous F-actin or other cytoskeletal proteins, including tubulin or vimentin, in confluent endothelial cells (Supplemental Figure 6, A–C). Instead, a deeper characterization of the subcellular distribution of CD93 and MMRN2 revealed that both proteins colocalize with fibronectin fibers (siCtrl in Figure 6, B and C). The colocalization between CD93 and fibronectin was dramatically inhibited in MMRN2-silenced endothelial cells (siMMRN2_3 in Figure 6B). In contrast, downregulation of CD93 did not affect the colocalization between MMRN2 and fibronectin fibers (siCD93_1 in Figure 6C). These results indicate that MMRN2 is required for CD93 interaction with fibronectin in extracellular matrix.

Costaining with vesicular markers revealed that CD93 colocalized with caveolin-1–coated vesicles in endothelial cells, while MMRN2 was found partially associated with vWF-positive vesicles (Supplemental Figure 6, D and E). Our data suggest that the interaction between CD93 and MMRN2 occurs in the cell membrane–extracellular matrix interface.

Fibronectin is a major component of the extracellular matrix and is essential for sprouting angiogenesis and cell migration (19). Therefore, we investigated whether the interaction of CD93 with MMRN2 affects the organization of fibronectin fibrils. Control endothelial cells grown on gelatin formed a dense fibronectin matrix and a well-organized fibrillar network (Mock and siCtrl in Figure 6D). However, downregulation of either CD93 or MMRN2 in endothelial cells induced a disruption in the fibronectin fibrillar network as compared with controls (siCD93 and siMMRN2 in Figure 6D). In contrast, fibronectin gene expression in siCD93 and siMMRN2 cells was similar to that in controls (Figure 6E). These results indicate that CD93 and MMRN2 are important for the organization of fibronectin into fibrillary structures, but that they do not regulate fibronectin expression in endothelial cells. Consistent with these results, decreased fibronectin fibrillogenesis was observed in brain endothelial cells isolated from CD93^–/– mice compared with those isolated from WT mice (Figure 6F). Disrupted fibronectin fibrillogenesis was restored by lentiviral transfection with a WT CD93 construct in endothelial cells silenced for endogenous CD93 (Figure 6G). A similar rescue effect on fibronectin fibrillogenesis was observed when siMMRN2 cells were cultured in conditioned medium derived from control cells (Figure 6H). These data point out a novel role of MMRN2 and CD93 interaction in regulating organization of fibronectin in extracellular matrix.

Knockdown of CD93 or MMRN2 impairs β₁ integrin activation and phosphorylation of focal adhesion kinase in endothelial cells. Integrins are transmembrane proteins with a well-described function in the regulation of cell-matrix adhesion, extracellular matrix organization, and cell migration (20). In particular, α₅β₁ integrin plays a key role in the regulation of fibronectin fibrillogenesis and extracellular matrix deposition (21). We therefore investigated whether β₁ integrin activity was altered in CD93- or MMRN2-downregulated cells. β₁ Integrin activation was assessed by immunofluorescent staining using an antibody that specifically recognizes the active conformation of β₁ integrin (clone 12G10). A striking reduction of active β₁ integrin was observed in endothelial cells when either CD93 or MMRN2 was silenced (Figure 7, A and B). This reduction in β₁ integrin activation was confirmed by Western blot analysis, indicating that CD93 and MMRN2 affect integrin activation rather than expression (Supplemental Figure 7A). The important role of CD93 in regulating β₁ integrin activation was confirmed in brain endothelial cells isolated from CD93^–/– and WT mice, showing a striking reduction in β₁ integrin activation in the absence of CD93 (Supplemental Figure 7B).

β₁ Integrin can interact with several integrin α subunits in endothelial cells, and the main receptor for fibronectin is the α₅β₁ integrin. CD93 colocalized with α₅β₁ integrin in confluent HDBECs and in filopodia of sparsely seeded endothelial cells (arrowheads in Figure 7C and Supplemental Figure 7C). Costaining with the 12G10 antibody revealed colocalization between CD93 and activated β₁ integrin in endothelial cells. α_vβ₃ Integrin can also bind to fibronectin in endothelial cells, but we did not observe colocalization between CD93 and α_vβ₃ integrin (Figure 7C). The CD93-α₅β₁ interaction was further confirmed by PLA showing statistically significant interaction between CD93 and α₅β₁ as well as CD93 and active β₁ integrin (Figure 7D).

Focal adhesion kinase (FAK) is activated downstream of integrin activation and plays a key role in focal adhesion turnover as well as cytoskeleton organization and cell migration (22). In line with a requirement of CD93 and MMRN2 for full β₁ integrin activation, we observed reduced FAK activation, monitored as Y397-phosphorylated FAK (p-FAK), in endothelial cells treated with siRNA targeting either CD93 or MMRN2. We detected high levels of p-FAK in the focal adhesion sites at the leading edge of migrating control endothelial cells, while a striking reduction of p-FAK levels were found in the migrating front of endothelial cells silenced for either CD93 or MMRN2 (Figure 7E, quantified in Figure 7F).

These data indicate that CD93 interacts with α₅β₁ integrin and modulates its activation, leading to the phosphorylation of the downstream molecule FAK, and inducing fibronectin fibrillogenesis required for cell migration. The extracellular matrix protein MMRN2 stabilizes CD93 and thereby enhances the activation of α₅β₁ integrin. The proposed mechanism through which the CD93-MMRN2 complex interacts with integrins and fibronectin, thus affecting endothelial cell dynamics, is depicted in Figure 7G.

β₁ Integrin is critical for vessel growth and maturation and is indispensable for efficient endothelial sprout formation (23). Given the significant inhibition in endothelial sprouts and endothelial filopodia observed in CD93^–/– retinas (Figure 2, D and E), integrin activation and fibronectin fibrillogenesis was assessed during developmental angiogenesis in P6 WT and CD93^–/– mouse retinas. Consistent with a key role of CD93 in β₁ integrin activation, a dramatic inhibition of the active form of β₁ integrin (recognized by the 9EG7 antibody) was observed in the sprouting front and filopodia of CD93^–/– retinas compared with WT mice (Figure 8A). In addition, fibronectin staining showed disruption of fibronectin fibers in the sprouting front and filopodia of CD93^–/– retinas (Figure 8B).

CD93 and MMRN2 expression correlates with increased fibronectin deposition in human glioma vessels. Fibronectin fibrillogenesis is a known feature of tumor angiogenesis. To determine whether CD93 is coexpressed with MMRN2 and fibronectin fibrils in human glioma vessels, we analyzed the fraction of blood vessels positive for MMRN2 (Figure 9A) and fibronectin (Figure 9B) in glioma tissue microarrays, containing a total of 235 samples of human control brain and WHO grade II–IV glioma tissues, and correlated the scores with our previous scoring of CD93 expression (11). Since MMRN2 and fibronectin were heterogeneously expressed in tumor vessels, a semiquantitative scoring method was used based on the proportion of positive vessels in each core. Similar to our previous observation that CD93 expression is increased in the vasculature of human high-grade glioma (HGG) (11), MMRN2 and fibronectin were significantly elevated in WHO grade III and IV glioma compared with low-grade glioma (LGG; WHO grade II) or control brain samples (Figure 9, C and D). When correlating the scoring of MMRN2 staining with the scoring of CD93 staining in the same tissue microarray cores, we found that high levels of CD93 corresponded to significantly higher levels of MMRN2 in samples from LGG (Figure 9E) and HGG (Figure 9H). Similar results were obtained when we correlated fibronectin scores with CD93 scores in the vessels of LGG (Figure 9F) and HGG (Figure 9I). HGG samples with high levels of vascular MMRN2 showed significantly higher levels of fibronectin associated with the vessels (Figure 9J). No significant association was found between fibronectin and MMRN2 in LGG (Figure 9G). These results indicate that CD93, fibronectin, and MMRN2 are coexpressed in tumor vessels in glioma, and that the levels of these proteins are increased in higher malignancy grades, suggesting a functional role of the CD93-MMRN2-fibronectin complex during tumor angiogenesis.

CD93 is essential for β₁ integrin activation and fibronectin fibrillogenesis in GL261 tumor vessels. To determine whether CD93 is required for efficient β₁ integrin activation and fibronectin fibrillogenesis during tumor angiogenesis, we analyzed tumor vessels in intracranial GL261 gliomas in WT and CD93^–/– mice. By immunofluorescence staining using an antibody that specifically recognizes the active form of β₁ integrin (clone 9EG7), we found that β₁ integrin was highly activated in tumor vessels (Wild-type in Figure 10A). In contrast, a dramatic reduction of β₁ integrin activation was observed in GL261 tumor vessels in CD93^–/– mice (CD93^–/– in Figure 10A). In normal vessels adjacent to the tumor site, β₁ integrin activation was low or absent in both WT and CD93^–/– mice (Supplemental Figure 8A).

Immunostaining revealed an abundant deposition of fibronectin in the GL261 tumor tissue in WT mice, predominantly associated with the tumor vessels (Figure 10B). Notably, a substantial reduction of fibronectin deposition was found in the CD93^–/– GL261 tumor, and disrupted fibronectin fibers were observed associated with the tumor vasculature (Figure 10B). Quantification of the fibronectin signal revealed a significant reduction of fibronectin deposition in GL261 tumors from CD93^–/– mice compared with tumors from WT mice (Figure 10, C and D). No obvious differences in fibronectin deposition were found in the normal vasculature of CD93^–/– mice compared with WT mice, and fibronectin levels in normal vessels were considerably lower than those in tumor vessels (Supplemental Figure 8B). Fibronectin gene expression analysis showed a significant upregulation of fibronectin in tumor tissue as compared with nontumor brain tissue, and no differences were found when WT and CD93^–/– mice were compared (Supplemental Figure 8C). Fibronectin expression was considerably higher in brain endothelial cells than in GL261 cells, suggesting a main contribution of the vasculature in fibronectin production in this tumor model (Supplemental Figure 8D). Consistent with our previous report, the mean GL261 tumor area was significantly reduced in CD93^–/– mice as compared with WT mice (Figure 10E and ref. 11). These results indicate a nonredundant role of CD93 in promoting β₁ integrin activation and fibronectin fibrillogenesis during glioma angiogenesis. A 3D volume reconstruction of fibronectin deposition and CD31-positive vessels in GL261 tumors revealed that fibronectin fibrils are consistently associated with blood vessels (arrowheads Figure 10F). Thus, our data show that fibronectin matrix is mainly deposited and organized in fibrillary structures by endothelial cells in GL261 tumors, and indicate that CD93 expression is critical for correct organization of fibronectin during tumor angiogenesis.