Delta-like 4 Notch Ligand Regulates Tumor Angiogenesis, Improves Tumor Vascular Function, and Promotes Tumor Growth In vivo

Introduction

The vascular endothelial growth factor (VEGF) plays a major role in tumor angiogenesis, and inhibition of the VEGF pathway inhibits tumor growth in some preclinical tumor models in mice ( 1). In clinical trials, bevacizumab, when combined with standard chemotherapy, significantly affected tumor progression and improved overall survival of colorectal and lung cancer patients and progression-free survival of breast cancer patients ( 2). However, disruption of the VEGF pathway did not significantly affect tumor growth and tumor angiogenesis in some preclinical tumor models, and in almost all clinical trials of the antiangiogenic monotherapy, there is evidence of resistance ( 2, 3). This suggests that other factors or pathways are also important in tumor growth and tumor angiogenesis, and that other antiangiogenic therapies are required for tumors that are resistant to VEGF inhibitors.

The Notch signaling is an evolutionarily conserved intercellular signaling pathway affecting many differentiation processes and cell fate determination during embryonic and postnatal development ( 4). Five transmembrane Notch ligands [Jagged1, Jagged2, Delta-like 1 (DLL1), DLL3, and DLL4] and four Notch receptors (Notch1–4) have been described in mammalian cells. Ligand receptor binding mediates Notch signaling by stimulating cleavage of the Notch intracellular domain (NICD) via intramembrane proteolysis by γ-secretase. The NICD then translocates to the nucleus, where it typically interacts with recombination signal binding protein Jκ (RBP-Jκ)/transcription factor C promoter-binding factor 1 (CBF1) to activate the transcription of target genes such as members of the Hes and Hey families of basic helix-loop-helix transcription factors and EphrinB2 ( 5, 6).

There is accumulating evidence that Notch signaling has an essential role in vascular development and angiogenesis. Numerous components of the Notch pathway are expressed in the vasculature at various stages of development ( 7, 8). Mutations of the Jagged1 and Notch3 genes in humans lead to the autosomal dominant disorders Alagille syndrome and cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy, respectively, both of which exhibit severe vascular defects ( 9– 11). Mice deficient for a variety of components of the Notch pathway display embryonic lethality with severe vascular abnormalities ( 8). Remarkably, haploinsufficiency of DLL4 resulted in embryonic lethality due to severe vascular defects in mice ( 12– 14), a phenotype previously only reported for VEGF haploinsufficiency in angiogenic pathways ( 15).

Although these studies have established a well-defined role for Notch signaling in vascular development, less is known about the function of Notch signaling in tumor angiogenesis ( 16, 17). Recently, it was reported that Jagged1 is able to stimulate angiogenesis and xenograft tumor growth ( 18). We and others have previously shown that DLL4 expression is restricted to arterial endothelial cells (ECs) in embryonic vascular development and to tumor vessels of human breast and bladder cancer and clear cell renal cell carcinoma ( 19– 22). Knockdown of basal DLL4 levels in ECs inhibited multiple EC functions ( 21), whereas overexpression of DLL4 in ECs reduced cell proliferation, migration, and tube-like formation ( 23), suggesting that an optimal level of DLL4 expression is essential for EC functions. In addition, DLL4 expression is induced not only by VEGF and basic fibroblast growth factor ( 21, 24), but also by hypoxia ( 19, 21), a characteristic feature of the tumor microenvironment known to induce angiogenesis.

During the preparation of this study, it was reported that DLL4, when transduced into C6 rat glioma tumor cells, decreased angiogenesis of C6 xenograft tumors by reduced vascular sprouting and branching ( 25), whereas blockade of DLL4 signaling with soluble DLL4-Fc or anti-DLL4 antibody inhibited tumor growth in several tumor models by promoting nonproductive angiogenesis ( 25– 27). Systemic treatment with soluble DLL4-Fc or anti-DLL4, unlike with dibenzazepine, an inhibitor of γ-secretase, had no visible toxic effects on the mice ( 25, 26). Vascular proliferation in Sarcoma 180 tumors implanted in adult DLL4^+/− mice was increased, and maturation was lacking due to a deficiency in pericyte coverage ( 27).

In this study, we have investigated the function of DLL4 in tumor angiogenesis and tumor progression in vivo through overexpression of DLL4 in five tumor cell lines grown in mice. We found that DLL4 expressed in these tumor cells activated Notch signaling in mouse endothelial cells, improved vascular structure and function, and importantly, in two human tumor cell lines, promoted tumor growth in vivo. VEGF inhibition using bevacizumab reduced tumor growth induced by DLL4 expression. Inhibition of DLL4 signaling by D4ECD-Fc inhibited the growth of tumors by disrupting vascular structure and function despite increased tumor vessel number. We have also shown for the first time that DLL4 is up-regulated in tumor cells and tumor ECs of human glioblastoma, indicating that our model is clinically significant.

Materials and Methods

Cell culture and reagents. U87GM (U87, human glioblastoma), PC3 (human prostate adenocarcinoma), MDA-231 (human breast carcinoma), HT1080 (human fibrosarcoma), B16F10 (B16, mouse melanoma) and Chinese hamster ovary (CHO) cell lines were obtained from Cancer Research UK Cell Services and maintained in RPMI 1640 or DMEM supplemented with 10% fetal bovine serum (FBS), 2 μmol/L l-glutamine, 50 IU/mL penicillin, and 50 μg/mL streptomycin sulfate. Human umbilical vascular endothelial cells (HUVEC) were isolated and cultured as previously described ( 21). Recombinant human DLL4 extracellular domain (rDLL4) was purchased from R&D Systems. The full-length human DLL4 retroviral construct and retroviral packaging and infection were done as previously described ( 23).

Establishment of cell lines secreting D4ECD-Fc fusion proteins. The cDNA coding for human DLL4 extracellular domain (D4ECD, amino acid residues 1–486) was cloned in frame with a mouse immunoglobulin G₁ into the pFUSE-mIgG₁-Fc1 expression vector (Invivogen Insight). The resulting D4ECD-Fc construct was transfected into CHO or U87 cells using LipofectAMINE 2000 (Invitrogen) selected with 200 μg/mL Zeocin. The parental pFUSE-mIgG₁-Fc1 vector was transfected into the corresponding cells as a control.

Xenograft and allograft mouse models. Six- to eight-week-old female BALB/c severe combined immunodeficiency (SCID) mice (Harlan Sprague-Dawley, Inc.) were implanted s.c. with 100 μL of cell suspension with an equal volume of Matrigel (BD Bioscience). U87 (10⁷), MDA-231 (10⁷), HT1080 (5 × 10⁶), or PC3 (5 × 10⁶) cells were injected into each mouse. B16 cells (2 × 10⁴) were implanted s.c. into the flank of SCID mice without Matrigel. For the experiments of cell co-implantation, CHO cells (2 × 10⁶) plus U87 or PC3 cells (5 × 10⁶) or U87-D4ECD-Fc (5 × 10⁶) plus U87-DLL4 (5 × 10⁶) with an equal volume of Matrigel were coinjected s.c. into the flank of SCID mice. Each group consisted of five mice. Tumor growth was monitored twice to thrice every week by measuring the length, width, and height of each tumor using calipers. Tumor volumes were calculated from the formula L × W × H/0.52. When a tumor reached the maximum size (1.44 cm² surface area) permitted by the Home Office license, the mouse was sacrificed, and the tumor was excised. Where appropriate bevacizumab was given i.p. (10 mg/kg every 3 days) for a total of five injections (U87) or seven injections (PC3), starting from day 0 when the tumor cells were implanted.

Immunohistochemistry. Tumor-bearing mice were injected i.v. with 2 mg pimonidazole (Hypoxylprobe-1) in 0.9% saline, 100 μg biotinylated tomato lectin (Vector Laboratories), or 0.6 mg Hoechst 33342 in PBS 30 min, 10 min, or 1 min before sacrifice of the mouse, respectively. Paraffin-embedded tissue blocks from formalin-fixed tumor samples were sectioned, dewaxed, and rehydrated using standard techniques. For immunofluorescent staining, 7-μm cryo-sections were fixed in acetone for 10 min and dried in air for 1 h. Following blocking of nonspecific binding with 5% filtered goat serum in PBS for 1 h, sections were incubated with primary antibody for 1 h at room temperature. After a PBS wash, sections were incubated with secondary antibody Alexa-Fluor 594 goat anti-rabbit, Alexa-Fluor 555 goat anti-rat (0.5 μg/mL, Molecular Probe, Invitrogen), rabbit anti-rat FITC (1:20, DAKO) or Texas Red–conjugated streptavidin (1:100, Molecular Probe, Invitrogen) for 30 to 60 min in the dark. Following a PBS wash, slides were mounted in DAKO Fluorescent Mounting Medium. Vascular density and vessel size of tumors were quantified after staining ECs with rat anti-mouse CD31/PECAM (2 μg/mL, Clone ER-MP12, Acris GmBH). Fluorescent microscopy images were acquired at a resolution of 1,300 × 1,030 pixels. Vessel density corresponds to the percentage of each field (0.59 mm²) occupied by a CD31-positive signal (as determined by the percentage of black pixels per field after transforming the RGB pictures into binary files). Vessel number is determined by the number of continuous CD31-positive structures per field. Vessel size quantification is relative to the average number of pixels per continuous CD31-positive structure. Tumor vessel perfusion was assessed by Hoechst 33342 fluorescence or biotinylated tomato lectin. Tumor hypoxia was determined as the percentage section area that stained positively for using the Hypoxyprobe-1 kit (Chemicon) to detect pimonidazole-protein adducts in tumor samples following the manufacturer's instructions. Pericyte coverage around the vessel was assessed by staining with the pericyte marker NG2 (1:200, Chemicon), Cy3-conjugated α-SMA (1:2,000, Sigma) or endosialin (kindly provided by Dr. C.M. Isacke, Institute of Cancer Research, London). Tumor proliferation was quantified following staining for Ki67 either in 5-μm formalin-fixed, paraffin-embedded sections using the MIB-1 antibody (1:100, DAKO) and hematoxylin counterstain or in cryo-sections and double stained for CD31. Necrosis was quantified histologically on hematoxylin-stained sections. Necrotic areas were identified as areas displaying cells with basophilic pyknotic nuclei undergoing karyorrhexis and/or karyolysis. Tumor apoptosis was measured by quantification of positive tumor cells in sections stained with an anti-human active caspase-3 antibody (1:2000, R&D Systems) and hematoxylin counterstain. For each tumor section, the total number of positively stained cells was counted from five randomly chosen fields.

Statistics. All data are presented as mean ± SE (X ± SE). Statistical analysis including t test, one-way ANOVA, and parametric generalized linear model with random effects were done using GraphPad Prism 4.0b or SPSS 15.0 software. Statistical significance is indicated in the figures by *, where P < 0.05, and **, where P < 0.01.

Results

Generation of tumor cells over expressing DLL4. We transduced a range of human (U87, PC3, HT1080, and MDA-231) and mouse (B16) tumor cell lines with an LZRSpBMN-linker-IRES-EGFP retroviral vector. This led to increased DLL4 expression in all tumor cell lines, although the up-regulation of DLL4 was much higher in U87, HT1080, and B16 than in PC3 and MDA-231 (Supplementary Fig. S1A). In contrast, infections with empty vector control (EV) retroviruses did not increase DLL4 expression in tumor cells. The exogenous enhanced green fluorescent protein (EGFP) encoded by the retroviral vector itself was expressed at a similar level within the DLL4 and EV control cells (Supplementary Fig. S1A).

To assess whether DLL4 activated Notch signaling in these cell lines, we measured the accumulation of the activated cleaved N1ICD and the induction of the Notch target genes, Hes1, Hey1, and Hey2, by qPCR and Western blotting. In DLL4-overexpressing cells, N1ICD was increased in HT1080, MDA-231, and B16 cells and, to a lesser extent, in U87 cells. Hes1 protein was also increased in U87, MDA-231 and B16 cells overexpressing DLL4 compared with the corresponding EV-control cells (Supplementary Fig. S1B). Hes1 and Hey1 mRNA was significantly increased in U87 cells (DLL4 cells versus EV cells, P < 0.01 for both genes; Supplementary Fig. S1C). Similarly, expression of mouse Hes1 (mHes1) and mHey2 was also increased in B16 cells (DLL4 cells versus EV cells, P < 0.05 for both genes). mHey1 was not detected in B16 cells by qPCR. However, there was no significant change in expression of any tested genes in PC3, HT1080, and MDA-231 DLL4-overexpressing cells compared with the corresponding EV control cells (P > 0.05), implying variable expression of Notch pathway components and Notch target genes in different cell types.

DLL4 differentially affects proliferation of tumor cells in vitro and tumor growth in vivo. We tested the effect of DLL4 overexpression on the proliferation of transduced tumor cells in vitro. As shown in Fig. 1A , DLL4 did not affect the growth of PC3, MDA-231, and B16 cells assessed by counting cell number. However, DLL4 slightly but significantly inhibited the growth of HT1080 and retarded the growth of U87. Similar results were obtained with MTS cell viability assays (data not shown).

Next, we s.c. implanted each cell line into SCID mice. In U87 and PC3 xenografts, DLL4 increased the rate of tumor growth, but DLL4 did not alter the rate of tumor growth in HT1080, MDA-231, and B16 xenografts ( Fig. 1B). As shown in Fig. 1B, U87 tumors overexpressing DLL4 grew more rapidly, and the tumor volume was significantly larger than EV control tumors after 10 days (P < 0.01). Kaplan-Meier survival curves showed that all mice in the DLL4-overexpressing group had to be sacrificed within 3 weeks due to excessive tumor size with a median survival of 18 days, whereas mice in the EV control group were sacrificed within 5 weeks, with a median survival of 31 days (P < 0.01; hazard ratio = 0.2). A similar but smaller effect was observed for PC3 tumors overexpressing DLL4.

DLL4 activates Notch signaling in host stromal/endothelial cells. We investigated whether DLL4 expressed in tumor cells was able to activate Notch signaling in host ECs and thereby affect EC function. When immobilized on a plate, recombinant human DLL4 (rDLL4) activated Notch signaling ( 23), induced expression of several arterial genes in ECs, including Hey2, Efnb2, and DLL4 itself ( Fig. 2A ), and inhibited EC proliferation ( Fig. 2B). Thus, DLL4 has a positive-feedback mechanism to enhance Notch signaling in ECs and affects endothelial functions.

We then assessed whether DLL4 expressed in tumor cells activates Notch signaling in host stromal/endothelial cells by measuring mRNA levels of Notch target genes in xenograft tumors by qPCR using mouse-specific primers ( Fig. 2C). Because mDLL4 is specifically expressed in arterial ECs and up-regulated in the tumor vessels ( 19), the ratio of mDLL4 to the EC marker mCD31 could reflect the level of mDLL4 per EC in tumors ( 21). For U87 xenograft tumors, mDLL4 expressed in tumor ECs and mHes1 expressed in host stromal/endothelial cells were significantly increased in tumors overexpressing DLL4 (both P < 0.05). For PC3 xenograft tumors, however, there was no significant change in expression of any tested genes (data not shown). In contrast, for B16 allograft tumors, all tested genes were significantly increased in DLL4-overexpressing tumors (P < 0.05). Because mHey1 and mEfnb2 were not detectable in B16 cells in vitro (see Supplementary Fig. S1C), their up-regulation in the tumor samples presumably reflected gene expression in mouse stromal/endothelial cells.

DLL4 modulates structure and function of tumor vasculature. We assessed the effects of DLL4 expressed in tumor cells on the morphology and function of the tumor vasculature. Immunostaining for mCD31 showed abundant and irregular vascular networks in the EV control tumors of U87, with numerous tortuous, disorganized vessels ( Fig. 3A ). In contrast, DLL4-overexpressing tumors displayed decreased vessel density and vessel number, yet exhibited larger vessels with large lumina ( Fig. 3A). Importantly, similar findings in terms of vessel size and number were observed for all the DLL4-overexpressing tumors tested (Supplementary Fig. S2); although the basal vessel level in PC3 tumors was much lower (∼8-fold) than that in U87 tumors, we still detect a smaller number of larger vessels in tumors expressing DLL4.

To assess vessel maturity, tumor sections were co-stained for mCD31 and for the NG2 proteoglycan, endosialin, and α-smooth muscle actin (α-SMA), respectively. NG2 is a used marker for pericytes of poorly differentiated growing vessels, for example, those usually observed in tumors. During later stages of vessel maturation, pericytes progressively lose their ability to express NG2, and NG2 is totally absent from fully differentiated and functional vessels ( 28, 29). As shown in Fig. 3B, overexpression of DLL4 in U87 cells correlated with a total loss of NG2 staining surrounding the observed vessels. In contrast, the vessels observed in the control tumors were strongly positive for NG2 expression. Endosialin staining produced a similar result to that of the NG2 marker (Supplementary Fig. S3). However, α-SMA–positive cells were hardly detected in either EV control and DLL4-overexpressing tumors (data not shown). Thus, DLL4 expressed in tumor cells resulted in a decrease of pericyte coverage around the vessels.

Hoechst 32224 and tomato lectin fluorescence indicated that the vasculature was better perfused in DLL4-overexpressing U87 tumors compared with EV control tumors ( Fig. 3B). Double staining for hypoxyprobe-1 (pimonidazole) and mCD31 showed much reduced regions of intratumoral hypoxia in DLL4-overexpressing U87 tumors, significantly less than those in EV control tumors (5% compared with 18%; P < 0.01; Fig. 3C).

We previously showed that DLL4 reduces the expression of VEGF receptor-2 (VEGFR2) in ECs in vitro ( 23). As shown in Fig. 3D, in vivo expression of mouse VEGFR2 protein in the large vessels was lower in DLL4-overexpressing U87 tumors compared with those of EV control tumors, suggesting that the large vessels were particularly modulated by the DLL4 signal.

Taken together, the above data suggest that DLL4 expressed in tumor cells induces larger blood vessels with well-defined lumina, resulting in better perfusion and decreased hypoxia despite decreased tumor vessel number and pericyte recruitment.

DLL4 affects proliferation, apoptosis, and necrosis in vivo. To further address the mechanism of the DLL4-induced increase in tumor growth in vivo, we investigated cellular proliferation, apoptosis, and necrosis of U87 tumors. The Ki67 labeling index was significantly lower in DLL4-overexpressing tumors than in EV control tumors (P < 0.01; Fig. 3C). Immunostaining for activated caspase-3, a marker of apoptosis, showed that the number of apoptotic cells was decreased in DLL4-overexpressing tumors compared with EV control tumors (P < 0.01; Fig. 3C). Histologic analysis of tumor tissues revealed that more than 25% of the section area of EV control tumors was necrotic, whereas necrosis was rarely detected in DLL4-overexpressing tumors (P < 0.01), and the necrosis was associated with poor vessel distribution ( Fig. 3C). Similar effects on proliferation and apoptosis were found for two other xenografts investigated, PC3 and HT1080.

We have carried out double staining of CD31 with either MIB-1/Ki67 (proliferation) or activated caspase-3 (apoptosis). This staining showed that in U87-EV tumors, Ki67-positive nuclei are around the blood vessels, whereas in tumors expressing DLL4, Ki67-positive nuclei are more evenly dispersed. Similar results for activated caspase-3 were found, but inversely, away from vessels in EV and more uniform in DLL4.

Effect of DLL4 on the response of tumors to VEGF inhibition. To explore whether DLL4 affects the response of tumors to VEGF inhibition in vivo, we first assessed VEGF expression in U87 and PC3 cells under normoxia and hypoxia in vitro by ELISA. As shown in Fig. 4A , DLL4 overexpression in tumor cells did not affect the production of VEGF protein compared with EV control cells for either U87 or PC3 (all P > 0.05). Interestingly, hypoxia significantly increased VEGF production of PC3 (P < 0.05) but not U87 cells (P > 0.05) compared with normoxia. However, U87 cells produced about 35-fold more VEGF than PC3 cells under normoxia and about 10-fold more under hypoxia.

We then investigated the therapeutic effects of bevacizumab, a humanized anti-human VEGF antibody, on tumor growth of U87 and PC3 overexpressing DLL4 compared with EV controls. As shown in Fig. 4B and C, i.p. administration of bevacizumab strongly inhibited the growth of U87 EV control tumors but did not significantly affect the growth of PC3 EV control tumors compared with the corresponding control groups treated with PBS. Xenografts of the DLL4-overexpressing tumors were treated with bevacizumab in the same experiments. For U87 tumors, although bevacizumab was given, tumor growth was inhibited in DLL4-overexpressing and EV control tumors. However, following the cessation of bevacizumab treatment, the expression of DLL4 was associated with regrowth that parallels that of the original control-treated tumors of comparable size ( Fig. 4B). For PC3 tumors, bevacizumab significantly reduced the tumor growth of DLL4 tumors compared with the PBS control, such that the growth curve of DLL4 tumors matched that of the EV controls without therapy ( Fig. 4C). There was a minimal effect on the EV control tumors, which were resistant compared with U87.

mCD31 staining revealed that treatment with bevacizumab resulted in a marked decrease of vessel density in both EV control and DLL4-overexpressing tumors of U87 by pruning small, disorganized immature vessels compared with tumors treated with the PBS control ( Fig. 4D). However, there were residual well-organized large vessels with regular shape and large lumina in DLL4-overexpressing tumors after bevacizumab treatment ( Fig. 4D). To a lesser extent, similar results were found for PC3 tumors (Supplementary Fig. S4).

D4ECD-Fc inhibits growth of bevacizumab-sensitive and bevacizumab-resistant tumors in vivo. Soluble forms of Notch ligands have previously been shown to act as dominant negatives ( 30– 32). To assess the effects of soluble DLL4 on tumor angiogenesis, we established a U87 cell line that expressed and secreted the soluble D4ECD-Fc ( Fig. 5A ). U87 D4ECD-Fc cells exhibited the same growth rate to U87-IgG control (data not shown). Co-implantation of U87 D4ECD-Fc with an equal number of U87 cells overexpressing DLL4 into the mice significantly reduced the tumor growth (P < 0.01) and improved mouse survival (P < 0.05) when compared with those of U87 IgG control ( Fig. 5B). Thus, soluble D4ECD-Fc inhibits tumor growth in vivo.

As shown by CD31 staining, expression of D4ECD-Fc totally reversed the phenotype of U87-DLL4 tumor to that observed in U87 EV tumors, which is a greater number of smaller, abnormal vessels within U87 tumors, resulting in increased vessel density (P < 0.05). Vessel lumina were rarely observed in D4ECD-Fc–expressing tumors ( Fig. 5C). Similar observations were obtained in a CHO/U87 and CHO/PC3 co-implantation experiments, where CHO cells secreting D4ECD-Fc inhibited tumor growth in vivo (Supplementary Fig. S5 and 6). NG2 staining showed that NG2-positive cells occurred in the Fc control tumors but not in those expressing and secreting D4ECD-Fc (Supplementary Fig. S7). Thus, both up-regulation and down-regulation of DLL4 function affected pericyte coverage.

DLL4 expression is up-regulated in tumor ECs of human glioblastoma. Because U87 tumors were most sensitive to the effect of DLL4 expression in vivo, we investigated DLL4 mRNA expression using in situ hybridization in 20 surgical glioblastoma specimens (Supplementary Table S1). As shown in Fig. 6 , expression of DLL4 in tumor ECs was detected in 75% of glioblastoma specimens (15/20 cases; see Supplementary Table S1), including DLL4 expressed in garland-like vascular proliferations (8/20 cases, 40%), glomeruloid vascular proliferations (11/20, 55%), and capillaries (14/20 cases, 70%; Fig. 6A). In contrast, expression of DLL4 was not detected in parenchymal or vascular cells of any of the 10 non-neoplastic control specimens ( Fig. 6B) that were collected from the temporal lobes of patients surgically treated for intractable temporal lobe epilepsy. Interestingly, expression of DLL4 was also observed in some tumor cells (4/20 cases, 20%) located either in perinecrotic areas or in areas distant from necrosis ( Fig. 6C).

Discussion

It was reported that DLL4 expressed in the rat C6 tumor cells affected tumor angiogenesis but not tumor growth in the xenografted mouse model ( 25). We found that DLL4 expressed in tumor cells did not stimulate growth in vitro, but increased tumor growth in two of the five cell lines in vivo, and these two made it possible for us to further investigate the mechanism why DLL4 enhances tumor progression. However, in all five types, there were consistent and reproducible effects on the vascular phenotype.

Molecular analysis showed that DLL4 induced known Notch targets in ECs in vitro (DLL4, Efnb2, and Hey2) and in mouse stromal/endothelial cells in vivo (mHes1 and mDLL4) in U87 tumors. There was no difference for PC3 xenografts, which may be related to the much sparser vasculature and lower DLL4 levels. By in situ hybridization, we have previously shown the induction of mDLL4 specifically in tumor vessels of xenografts. Therefore, DLL4 expressed in tumor cells activates Notch signaling in host stromal/endothelial cells.

DLL4 expressed in tumor cells dramatically inhibited tumor angiogenesis, as recently reported ( 25), but surprisingly improved the structure and function of tumor vasculature by inducing larger vessels with large lumina and increased vessel perfusion and tumor oxygenation, resulting in increased tumor growth in vivo. Thus, DLL4 functions as a negative regulator of tumor angiogenesis, but acts as a positive driver of tumor progression by improving vascular function within tumors. The mechanism underlying DLL4 stimulation of tumor growth was related to a marked reduction in apoptosis and intratumoral hypoxia. There was a reduction in the number of dividing cells, but this was outweighed by the reduction in apoptosis. Clearly, Notch activation did not stimulate growth by stimulating proliferation in concordance with the in vitro results, where there was either no effect on growth or inhibition. Decreased apoptosis in response to angiogenesis, rather than increased proliferation, has been described by Folkman et al. ( 33). In studies of tumor dormancy, the reciprocal effect occurred. Thus, activation of DLL4-Notch signaling in vessels could overcome tumor dormancy.

In early development, DLL4 plays an important role in the vascular arterialization. DLL4 is specifically expressed in arterial ECs of embryonic vascular development ( 19, 22) and of developing retinal arteries ( 7). In addition, DLL4 induced expression of various arterial genes in HUVECs and endothelial precursor cells in vitro ( 6, 23). Moreover, genetic studies in zebrafish indicated that arterial endothelial fate is determined early by gridlock/Hey2, and that VEGF acts upstream of Notch signaling during arterial endothelial differentiation ( 34, 35).

Thus, we investigated the effects of DLL4 up-regulation and also DLL4 antagonism on pericytes. Using two markers of immature pericytes, we found that they were present in control tumors and absent in tumors expressing high DLL4. However, inhibition of Notch function with secreted soluble D4ECD-Fc also reduced pericytes, as recently reported ( 27). This effect of either up- or down-regulation of Notch signaling having similar responses has been reported for other aspects of angiogenesis. It was previously shown that up- or down-regulation of DLL4 in ECs both reduced cell proliferation, migration, and tube-like formation ( 21, 23), and that EC-specific knock-in of the Notch4 active form or knock-out of either Notch1 or Notch1/Notch4 triggers similar phenotypes in vivo (vascular defects causing embryonic lethality; ref. 36). Thus, our results indicate an important interaction of DLL4 with pericytes, but also show that pericyte coverage is not essential for an effective tumor vasculature. Although we showed that DLL4 expression correlated with vascular maturation in human bladder cancers ( 20), there are no data on other cancer types yet, and final effects may depend on the optimal amount of DLL4.

There are major interactions between VEGF and DLL4 because VEGF induces DLL4 expression ( 21, 24) and DLL4 expression reduces the response to VEGF and down-regulates VEGFR2 expression in vitro of ECs ( 23). We found in vivo down-regulation of VEGFR2 in the characteristic, large vessels of DLL4-overexpressing tumors (see Fig. 3D), which could imply that tumors would be less dependent on VEGF. We chose two cancer cell lines with more than 10-fold difference in endogenous VEGF production to investigate these interactions and found that their in vivo sensitivity to bevacizumab was directly correlated with their VEGF production. Bevacizumab treatment significantly reduced the vasculature, but in the DLL4-overexpressing tumors, the characteristic, larger vessels were maintained, although at a lower number, suggesting that they were partly VEGF dependent. This residual effect could be related to endogenous mouse VEGF that would not be inhibited by bevacizumab.

Because the growth of DLL4-overexpressing tumors following bevacizumab treatment was similar to that of PBS-treated EV control tumors, the vasculature of these two tumors, although morphologically different, must have a similar ability to deliver oxygen and nutrients to the tumor. Thus, fewer but larger vessels can be as efficient as a much greater number of smaller vessels. Because the effects of DLL4 expression on the vasculature can be reduced by bevacizumab, this implies a synergy between VEGF signaling and DLL4 in vivo. Indeed, the endogenous mouse DLL4 transcripts were significantly reduced by bevacizumab in DLL4-overexpressing and EV control tumors (data not shown). Therefore, our results, together with others ( 25, 26), suggest that DLL4 in tumors is a target of VEGF. Even when DLL4 is expressed exogenously, there is a positive interaction with VEGF in EC function in vivo. Furthermore, bevacizumab inhibited the early growth of DLL4-expressing tumors, and this may reflect the lack of pericytes still allowing tumors to maintain sensitivity. The final effect on tumor vasculature and sensitivity to bevacizumab may depend on the ratio of these two factors.

Disruption of Notch signaling by D4ECD-Fc induced a diametrically opposed vascular phenotype to that in tumors that overexpressed DLL4, with increased vessel density but impaired vessel function within tumors, resulting in the inhibition of tumor growth ( 25, 27). D4ECD-Fc not only restrained the progression of tumors that are sensitive anti-VEGF therapy, but also inhibited the growth of tumors that are resistant to anti-VEGF therapy ( 25). This lack of cross-resistance has important implications for therapy.

The U87 glioblastoma cell line was most sensitive to the effects of DLL4 expression. Glioblastoma, the most common primary brain tumor in adults, is characterized by very prominent vascular proliferation ( 37, 38) and has a poor prognosis. DLL4 was expressed in ECs, both of capillaries and the atypical vascular proliferations of glioblastoma, although not in all cases. Interestingly, DLL4 was also expressed in tumor cells from some patients, but not in non-neoplastic brain tissue, indicating that our model reflects some types of clinical cancer directly. There may be a paracrine signaling pathway from tumor cells to ECs as reported previously for Jagged1 in another tumor type ( 18).

In summary, we have shown that DLL4 expressed in tumor cells was able to activate Notch signaling, enhance the functional tumor vasculature despite reduced tumor angiogenesis, and consequently accelerate tumor growth in some tumors by reducing intratumoral hypoxia, tumor apoptosis, and tumor necrosis. The growth of tumors expressing DLL4 still responded to bevacizumab treatment of early tumors, although residual large vessels remained. Disruption of DLL4 signaling by D4ECD-Fc increased tumor angiogenesis, but inhibited the growth of both bevacizumab-sensitive and bevacizumab-resistant tumors. We also showed up-regulation of DLL4 in human glioblastoma. Our findings provide a rational basis for the development of novel therapeutic strategies via blockade of DLL4/Notch signaling and suggest that combined approaches for interrupting both DLL4 and VEGF pathways may improve the antiangiogenic therapeutic efficacy.