Notch and VEGF pathways play distinct but complementary roles in tumor angiogenesis

Background Anti-angiogenesis is a validated strategy to treat cancer, with efficacy in controlling both primary tumor growth and metastasis. The role of the Notch family of proteins in tumor angiogenesis is still emerging, but recent data suggest that Notch signaling may function in the physiologic response to loss of VEGF signaling, and thus participate in tumor adaptation to VEGF inhibitors. Methods We asked whether combining Notch and VEGF blockade would enhance suppression of tumor angiogenesis and growth, using the NGP neuroblastoma model. NGP tumors were engineered to express a Notch1 decoy construct, which restricts Notch signaling, and then treated with either the anti-VEGF antibody bevacizumab or vehicle. Results Combining Notch and VEGF blockade led to blood vessel regression, increasing endothelial cell apoptosis and disrupting pericyte coverage of endothelial cells. Combined Notch and VEGF blockade did not affect tumor weight, but did additively reduce tumor viability. Conclusions Our results indicate that Notch and VEGF pathways play distinct but complementary roles in tumor angiogenesis, and show that concurrent blockade disrupts primary tumor vasculature and viability further than inhibition of either pathway alone.


Background
The anti-vascular endothelial growth factor (VEGF) antibody bevacizumab (BV) is clinically validated to treat human cancers in adults, and is currently being evaluated as a therapy for pediatric tumors, including neuroblastoma [1]. However, it appears that essentially all patients, even those who respond to BV initially, will ultimately progress despite ongoing treatment. Thus, a similar limitation is anticipated in children treated with this agent. The mechanisms of tumor progression during BV treatment are not well understood.
We have previously shown that VEGF inhibition disrupts the blood vessels of the neuroblastoma NGP xenograft model, but resistance occurs [2][3][4]. VEGF blockade can increase cooption of existing vascular structures by encroaching tumors [2], a mechanism which may partly compensate for defective VEGF signaling. An additional mechanism for tumor progression during VEGF blockade could be the activation of an alternative angiogenic pathway. In support of this concept, we have shown that the Notch pathway is up-regulated in neuroblastoma when the VEGF receptor 2 (VEGFR2) is blocked with the anti-VEGFR2 antibody DC101 [3].
To examine the role of Notch in tumor angiogenesis, we have used the Notch1 decoy (N1D), a soluble construct derived from the extracellular domain of the Notch1 receptor that blocks Notch activation by sequestering Notch ligands [5]. Notch blockade using the N1D disrupts tumor vasculature and decreases tumor viability [5]. Based on our previous studies, we thus hypothesized that blocking the Notch and VEGF pathways simultaneously would disrupt the vasculature and inhibit tumor growth more effectively than either agent alone. To test this hypothesis, we evaluated the effects of expressing N1D in combination with bevacizumab (BV) in the NGP neuroblastoma xenograft model. Here, we report that combining N1D with BV had a profound effect on the tumor viability and the vasculature of NGP xenografts. N1D and BV additively increased tumor endothelial cell apoptosis, leading to blood vessel regression, and cell death.

NGP xenograft model
1 × 10 6 cells were injected intrarenally into 4-6 week old NCR female nude mice (Taconic, Germantown, NY) as previously described [3]. All animal experiments were approved by the Columbia University Institutional Animal Care and Use Committee. At sacrifice, organs were harvested and weighed as previously described [3].

Bevacizumab and placebo administration
NGP-LacZ and NGP-Notch1 decoy (N1D) tumorbearing mice were randomized one week after implantation. At this time, 28 mice in each group were treated with 100 μg mouse albumin (Sigma Aldrich) (placebo), and another 28 with 250 μg bevacizumab (10 mg/kg BV, Genentech) by intraperitoneal injection biweekly until time of sacrifice. Both the N1D and BV are fused to a human Fc fragment. Immunohistochemistry (Additional file 1: Figure S1A) and SDS-PAGE (Additional file 1: Figure S1B) to human Fc was performed on tumors, and confirmed expression of N1D in the tumors at time of sacrifice. Since BV is also fused to human Fc, it could also be detected in the NGP-LacZ + BV and NGP-N1D + BV tumors by SDS-PAGE (Additional file 1: Figure S1B, lower band).

Hypoxyprobe and quantification
Thirty minutes prior to sacrifice, mice received an intraperitoneal injection of 0.1 ml of pimonidazole (Hypoxyprobe®, Chemicon), and paraffin sections were used to stain for pimonidazole with MAB-1 (Chemicon), following the manufacturer's instructions. Four tumors per treatment group were quantified, with an average of 10 images taken from each tumor section at 20×. The images were then analyzed using Adobe Photoshop using an arbitrary threshold applied to all images.

Lectin perfusion and quantification
An intracardiac injection of 100 μl of Fluorescein labeled Lycopersicon esculentum (Vector Labs) was delivered 2 minutes prior to 1% PFA and PBS perfusion, followed by tumor collection. Tumors were fixed in 4% PFA and 50 μm thick sections were analyzed using confocal microscopy.

TUNEL assay quantification
TUNEL staining and quantification was performed as described [5]. An average of 20 images from each tumor section, and 5 tumors per group were analyzed.

Cooption quantification
5 μm sections of paraffin-embedded livers or kidneys were stained with hematoxylin and eosin (H&E). Tumor sections were scanned using an Epson 4870 Scanner. Glomeruli surrounded by at least one layer of tumor cells were counted as coopted vasculature. One section per tumor, and 28-30 tumors per group were analyzed.

VEGF ELISA
3 × 10 6 NGP-LacZ or NGP-N1D cells were plated on a 10 cm dish, and cultured in 6 ml of serum-free EBM-2 (Invitrogen) for 48 hrs. Supernatant from three independent collections was measured, using a human VEGF ELISA kit (R&D) following the manufacturer's instructions.

HUVEC cell death and BrdU incorporation assays
8,000 HUVEC per well (48 well plate) were seeded on a fibrinogen gel (Sigma), in basal EBM-2 media (Lonza). Two hrs later, media was replaced by conditioned media (CM) from NGP-LacZ or NGP-N1D cells (3×10 6 cells per 10 cm 2 plate, cultured for 24 hrs in EBM-2 basal media. We added 50 ng/ml BV or 20 ng/ml rhVEGF165 (R&D) to the CM. For BrdU assays, 2 μl BrdU (Millipore) was added per ml of CM. Cell death assays (Roche) or BrdU incorporation assays (Millipore) were measured 23 hrs later, following the manufacturer's instructions.

Immunoprecipitation and Western Blots
Tumors were lysed in a ratio of 100 ml of lysis buffer to 10 mg of tumor. 100 μl of tumor lysate was immunoprecipitated with A/G agarose beads (Santa Cruz Biotechnology) and blotted with anti-human Fc (Pierce).

Reverse transcription and quantitative PCR
RNA was extracted from cells following the manufacturer's instructions (Qiagen). Tissue RNA extraction was performed with Ambion ToTALLY RNA kit (Ambion) following the manufacturer's specifications. RNA from tissue or cells was reverse transcribed following the manufacturer's instructions (Invitrogen, DNAseI and Superscript). Quantitative PCR was performed with the Taqman system (Applied Biosystems) following the manufacturer's specifications, using primers designed not to amplify genomic DNA. GeNorm was used to select the normalization values out of 6 housekeeping genes.

Microarrays and probes preparation
HG-U133A 2.0 (human genome) Gene Chips (Affymetrix, Santa Clara, CA) were used to investigate gene expression in xenograft tumors. The cRNA probes were synthesized as recommended by Affymetrix, purified using RNeasy and fragmented according to the Affymetrix protocol, and 15 μg of biotinylated cRNA were hybridized to the microarrays. The samples were scanned with Affymetrix Gene Chip Scanner 3000. The data set, corresponding to 24 tumors (6 arrays per group), has been deposited in NCBI's Gene Expression Omnibus (GEO) and is available through accession number GSE (pending). Data were RMA normalized using the Bioconductor open source software.

Combined blockade of notch and VEGF increases necrosis and apoptosis in primary tumors
We have previously shown that inhibiting Notch signaling with the soluble Notch1 decoy (N1D) decreased viability (but not tumor weight), disrupted vasculature, disorganized the interaction of endothelial cells with pericytes in neuroblastoma NGP xenografts [5]. Given the critical role of VEGF in tumor angiogenesis, we hypothesized that combining Notch inhibition with VEGF blockade would synergistically inhibit tumor growth and angiogenesis. NGP-N1D cells or control NGP-LacZ cells were implanted into the left kidney of athymic mice. The mice were then randomized to treatment with either placebo or BV, and sacrificed at 5.5 weeks. Consistent with our previous data [5], there was no difference in tumor weight between control NGP-LacZ and NGP-N1D ( Figure 1A). VEGF blockade with BV, however, reduced NGP mean tumor weight by 61% (p < 0.001) compared to NGP-LacZ. Combined treatment with N1D and BV (NGP-N1D + BV), reduced mean tumor weight by a similar extent, 60%, as compared to NGP-LacZ, (p < 0.001).

Figure 1
Although there was no difference in tumor weight as compared with BV treatment alone, the NGP-N1D + BV tumors were noticeably more necrotic ( Figure 1B), which was confirmed by quantification of necrosis in H&E sections ( Figure 1C). Trichrome staining confirmed that these differences were due to necrosis and not fibrosis (data not shown). When assessed by TUNEL, the NGP-N1D + BV tumors exhibited significantly more apoptosis then NGP-LacZ controls (14.4 fold), BV alone (3.6 fold), or N1D alone (2.5 fold) ( Figure 1D). Blockade of Notch or VEGF did not have a direct effect on tumor cell proliferation, as there was no difference in phosphohistone H3 positive cells in areas of viable tumor ( Figure 1E). These data indicate that combined Notch and VEGF blockade additively decreases tumor viability.

Combined blockade of notch and VEGF markedly decreases perfusion
We next evaluated the effect of Notch and VEGF blockade on lectin perfusion (Figure 2A,B). We found that with combined blockade (NGP-N1D + BV), there was a paucity of vessels, with a significantly reduced (P < 0.05) lectin-positive area (16% of control NGP-LacZ, Figure 2B). Blockade with N1D or BV alone did not reduce overall perfusion as measured by lectin-positive area ( Figure 2B). However, in the case of BV, bound lectin outlined predominantly larger vessels (Figure 2A, arrow), or coopted renal glomeruli (Figure 2A, arrowhead), which contribute disproportionally to the total lectin-positive area. There was a marked reduction in vessel branching when VEGF was blocked ( Figure 2C), either with BV alone (12%) or with N1D + BV (7%), indicating that VEGF blockade inhibits the small sprouts that initiate neoangiogenesis.
The decrease in lectin-positive area for NGP-N1D + BV relative to NGP-BV tumors suggested that there could be a decrease in the number of coopted glomeruli. We quantified coopted vessels by counting the number of glomeruli surrounded by tumor cells in H&E sections ( Figure 2D, arrows). The number of coopted glomeruli in NGP-LacZ + BV tumors was increased 2.3 fold compared to NGP-LacZ tumors ( Figure 2E). Notch blockade significantly attenuated this BV-induced increase in cooption, with NGP-N1D + BV tumors having significantly fewer coopted glomeruli as compared to the NGP-LacZ + BV group (115 vs 178, respectively, Figure 2E). These findings suggest that Notch blockade attenuates BV-induced cooption in NGP xenografts. Furthermore, immunostaining for cleaved Notch1, reflecting processing of the Notch receptor after ligand binding, showed increased activated Notch1 in coopted glomeruli as a result of BV treatment in NGP tumors (Additional file 2: Figure S2). These data implicate Notch signaling in BV-induced

Blockade of VEGF results in tumor hypoxia
The lectin perfusion results demonstrate that combined Notch and VEGF blockade reduces both neoangiogenesis and VEGF blockade-induced vascular cooption. To determine if this resulted in increased hypoxia, we quantified pimonidazole (Hypoxyprobe®) by IHC. There was significantly more hypoxia in NGP-LacZ + BV (P < 0.05) and NGP-N1D + BV (P < 0.001) as compared to NGP-LacZ. (Figure 3A,B). NGP-N1D was not significantly different from NGP-LacZ tumors.

Figure 3
We also quantified the expression of two hypoxia responsive genes, VEGF and PlGF , in the tumors by real-time PCR. VEGF expression was significantly increased 3.1 fold (P < 0.01) in NGP-LacZ + BV and by 2.7 fold (P < 0.01) in NGP-N1D + BV tumors compared to NGP-LacZ controls and to NGP-N1D ( Figure 3C). PlGF expression behaved in a similar pattern: relative to NGP-LacZ controls, NGP-N1D tumors had no change, while NGP-LacZ + BV tumors had 4.7 fold more (p < 0.001), and NGP-N1D + BV tumors had 3.79 fold more PlGF than controls (p < 0.001).
We further analyzed the tumors by gene expression profiling using HG-U133A 2.0 (human genome) Gene Chips (Affymetrix). VEGF expression was increased 2.36 fold in NGP-LacZ + BV and 2.56 fold in NGP-N1D + BV tumors (see Additional file 3 for significantly increased genes). Gene Set Enrichment Analysis (GSEA) [6] demonstrated that VEGF blockade resulted in enrichment of hypoxia related gene sets in comparison to control NGP-LacZ tumors (Table 1, Figure 3D). For NGP-N1D + BV there were 7 hypoxia related gene sets that were enriched (3 positively, 4 negatively); for NGP-LacZ + BV there were 3 hypoxia related gene sets that were enriched (3 positively); but none for NGP-N1D. Comparing NGP-LacZ + BV to NGP-N1D + BV did not show any significantly enriched gene sets, suggesting that Notch blockade did not further increase the hypoxia caused by VEGF blockade.

Combined notch and VEGF blockade disrupts tumor vasculature
Combined blockade of VEGF and Notch reduced tumor viability, but did not further increase tumor hypoxia. To further understand the effect of combined blockade, we quantified tumor vasculature, using IHC for the basement membrane marker collagen IV, which forms a perivascular envelope that can persist after vascular regression. We detected a~50% increase in collagen IV in NGP-LacZ + BV tumors as compared to both control NGP-LacZ and NGP-N1D + BV ( Figure 4B), which suggested increased collagen IV contributed by coopted vasculature (Figure 4A, arrowheads). Notably, however, we noted persistent collagen IV in necrotic areas for NGP-N1D and for NGP-N1D + BV tumors ( Figure 4A, areas denoted by "N" and delimited by dashed line). These results suggest that Notch blockade disrupts vasculature, with vascular regression marked by the collagen IV sleeves at sites of previously functioning vessels [7].

Figure 5
Previous studies have suggested that both Notch and VEGF can act as survival signals for ECs. We therefore hypothesized that combined blockade would result in increased EC apoptosis. Quantification of apoptotic ECs was done using TUNEL ( Figure 5C, red), PECAM-1 ( Figure 5C, green), and nuclear DAPI immunostaining ( Figure 5C, blue). A cell was considered an apoptotic EC when its nucleus, stained by DAPI, overlapped with TUNEL stain (red), and was surrounded by PECAM-1 (green) ( Figure 5C, arrowheads). The number of apoptotic ECs in each section was then normalized to the PECAM positive area in each tumor. This quantification showed a significant increase in apoptotic EC in NGP-N1D + BV tumors ( Figure 5D, P < 0.01). This increase in apoptotic ECs suggested that Notch and VEGF are both survival signals in tumor ECs, which act independently.
In order to test the effects of VEGF and Notch on EC survival in vitro , we used human umbilical vein endothelial cells (HUVEC). Conditioned media (CM) from NGP-N1D cells was collected as a source of N1D, with CM from NGP-LacZ as a control. CM from NGP-LacZ and NGP-N1D contained similar amounts of VEGF (1.26 ± 0.79 vs. 1.28 ± 1.4 ng/ml, respectively, P = n.s.). We therefore added 50 ng/ml of BV to block the VEGF that was present in the CM. We incubated the HUVECs in CM + BV or CM + VEGF (20 ng/ml of rhVEGF165) for 23 hrs and quantified apoptosis using a cell death ELISA. HUVECs incubated with N1D CM + BV had 1.5 to 2 fold higher apoptosis levels than HUVECs incubated with control CM + BV ( Figure 5E). HUVEC apoptosis was significantly reduced when VEGF was added to the CM containing N1D ( Figure 5E).
Inhibition of Notch signaling with γ-secretase inhibitors or anti-Dll4 antibodies have found that there is increased EC proliferation in vitro [8], while Notch1 activation leads to endothelial cell cycle arrest [9]. Using conditioned medium from NGP-LacZ or NGP-N1D cells, with either 50 ng/ml BV or 20 ng/ml rhVEGF165, we found that N1D caused a small increase (9%) in HUVEC proliferation in vitro , and which was not affected by addition of VEGF (10%, Figure 5F). Together, these data suggest that both endothelial proliferation and apoptosis increase simultaneously in NGP-N1D, as compared to NGP-LacZ tumors. ECs in NGP-N1D tumors treated with BV, proliferate more than controls, but also display more EC apoptosis than either N1D or BV treatment alone.

Notch1 decoy and BV alter pericyte coverage of blood vessels
Immunostaining for the mature pericyte marker αSMA demonstrated discontinuity in this stromal cell envelope in NGP-N1D tumors as compared to NGP-LacZ, along with a more flattened morphology ( Figure 6A, arrows), in agreement with our previous data [5]. BV treatment led to a thinner layer of pericytes, although immunostaining revealed a continuous pattern ( Figure 6A, bottom left). Combined inhibition with N1D + BV, similar to N1D, led to discontinuity together with altered pericyte morphology ( Figure 6A, right panels, white arrows). Quantification revealed an increase in total αSMA(+) area in NGP-N1D compared to the other groups ( Figure 6B). NGP-N1D + BV had decreased αSMA(+) area compared to NGP-LacZ and NGP-N1D, indicating that combined blockade both diminishes the amount of pericytes coverage as well as disrupting the coverage ( Figure 6B). Similar findings were seen using NG2, a marker of more immature pericytes (data not shown). HUVECs incubated with N1D and BV had increased apoptosis compared to HUVECs with BV only. HUVECs incubated with N1D and VEGF had decreased apoptosis compared to N1D with BV, but were not different from HUVEC incubated with BV only. Consistent with the known role of VEGF promoting endothelial cell survival, HUVEC incubated with VEGF had lower apoptotic levels than HUVECs incubated with BV, and lower apoptotic levels than HUVECs incubated with N1D and VEGF. Error bars represent SD. *P < 0.05. (F) N1D with BV resulted in increased BrdU incorporation by HUVECs compared to BV only. Addition of VEGF did not affect N1D-induced proliferation. Error bars represent SD. *P < 0.05.

Notch1 decoy effects on vasculature are direct
Immunofluorescence for cleaved Notch1 demonstrated Notch activation in vasculature of NGP-LacZ and NGP-LacZ + BV, but a paucity of activation in NGP-N1D and NGP-N1D + BV tumor vessels (Additional file 3). These data suggest that Notch blockade predominantly affects vascular cells. However, quantitative real time PCR analysis revealed that NGP tumor cells also express Notch1. In order to test the specificity of the effects of Notch blockade, we silenced Notch1 (shN1) in NGP. In vitro, NGP-shN1 had a 70% decrease in Hes1 and 50% decrease in Hey1 mRNA compared to empty vector (NGP-EV). In vitro , NGP-EV and NGP-shN1 cells proliferated at comparable rates, assessed by MTT assay (P = n.s.). We implanted the NGP-EV and NGP-shN1 cells into athymic mice and analyzed the tumors 6 weeks after implantation. We confirmed a 78 ± 18% (P < 0.05) reduction of Notch1 mRNA, as well as reduction in expression of the Notch downstream target gene Hes1 by 90% (P = 0.04), in NGP-shNotch1 as compared to NGP-EV. We found no changes in tumor weight ( Figure 7A), or apoptosis ( Figure 7B) as a result of silencing Notch1. Immunostaining revealed that shN1 did not affect Notch1 activation in the tumor blood vessels ( Figure 7C). The vasculature as assessed by percentage of PECAM-1 and αSMA positive area IHC, was similar in NGP-shNotch1 tumors compared to control NGP-EV tumors ( Figure 7D-G). Lastly, the vasculature in NGP-shN1 tumors did not appear to be disrupted ( Figure 7E,F), as was seen with Notch blockade ( Figure 6A). Together, these data support the notion that the effects of N1D are due to Notch blockade in the tumor vasculature, rather than to an effect on the tumor cells. Discussion A growing body of literature demonstrates crosstalk between Notch and VEGF pathways in angiogenesis in vitro [10][11][12][13], as well as in the mouse retina model [5,8,14,15]. The interaction between Notch and VEGF in tumor angiogenesis, however, is still in the process of definition. Our data suggests that Notch and VEGF have independent but coordinated functions in tumor angiogenesis, leading to additive effects when both pathways are simultaneously inhibited. VEGF blockade limits initial vessel sprouting and branching, and then later promotes vascular cooption [2] and the recruitment of pericytes that can stabilize the vasculature in the absence of VEGF [16,17]. Our results demonstrate that Notch blockade inhibits the later processes of cooption and EC/pericyte interaction.
Our data indicate that while VEGF blockade induces vascular cooption, this does not rescue tumors from hypoxia. Furthermore, while Notch blockade decreases cooption, this does not increase the VEGF-blockade induced hypoxia. This suggests that vascular cooption is relatively inefficient as a mechanism of providing oxygen to tumors; other potential contributions of co-opted vessels to tumor homeostasis remain unclear.
Our current results demonstrate that while Notch blockade can disrupt EC/pericyte interaction, it is only in the absence of VEGF that there is marked loss of vasculature, perfusion, and subsequent induction of tumor necrosis. In the context of loss of VEGF, Notch blockade causes EC apoptosis both in vitro and in vivo. This suggests that VEGF can rescue ECs when Notch is blocked.
Other groups have shown that antibodies against Dll4 and Notch1 increase non-functional EC vasculature in subcutaneous tumor models [8,14,15]. Endothelial Jagged1, however, has been shown to decrease sprouting in the mouse retinal model [18]. Therefore, the ability of the N1D to block both signaling by both Dll4 and Jagged1 [5] may explain the observed decrease in EC in NGP-N1D tumors. Our studies also suggest that the effect of Notch blockade on tumor angiogenesis is contextdependent. Thus, this effect may vary with different tumor models, and perhaps with tumor types.

Conclusions
Our data lead us to propose that the Notch and VEGF pathways play distinct but complementary roles in tumor angiogenesis, and that Notch is required for VEGF-mediated vascular remodeling. Here we show that concurrent blockade disrupts primary tumor vasculature and viability further than inhibition of either pathway alone, and that manipulation of each pathway is reflected in distinct vascular defects. Further investigation of the interaction between VEGF and Notch signaling in vasculature may allow refinement of a combined approach to targeting tumor angiogenesis. The percentage of PECAM-1 coverage did not differ between NGP-EV and NGP-shN1 tumors. Error bars represent SD. P = n.s. (F) The pericyte marker αSMA does not show erratic, discontinuous immunostaining in NGP-shN1 tumors. Bar = 100 μm. (G) There was no difference in percentage of αSMA coverage between NGP-EV and NGP-shN1 tumors. Error bars represent SD. P = n.s. Additional file 1: Figure S1: N1D is expressed in NGP-N1D and NGP-N1D + BV tumors. A) Immunostaining for human Fc indicates the presence of N1D in NGP-N1D tumors (right panel, red), but not in NGP-LacZ tumors (left panel). Bar = 200 μm B) SDS-Page shows the presence of N1D (upper band) in NGP-N1D + BV, but not in NGPLacZ + BV tumors. Presence of BV, which also contains Fc, is observed in both NGP-LacZ + BV and NGP-N1D + BV tumors (lower band). (PDF 617 KB) Click here to view.