Regulation of CXCR4 by the Notch Ligand Delta-like 4 in Endothelial Cells Free

Angiogenesis is essential for the growth of many types of cancer. As the tumors grow, low-oxygen environments within the tumor activate a variety of angiogenic pathways.

The Notch signaling pathway plays critical roles in vascular development and in tumor-induced angiogenesis (1). The Notch ligand Delta-like 4 (Dll4) is expressed at sites of active angiogenesis (2). Mice with targeted deletions of Dll4 revealed abnormal artery development and disrupted branching morphogenesis. Interestingly, heterozygous deletions of the Dll4 gene resulted in failure to remodel the primary vascular plexus and caused embryonic lethality, providing evidence for the critical importance of Dll4 expression levels in vascular development (3). Dll4 signals through the receptors Notch1 and Notch4 (4). Knocking out Notch1 alone or Notch1/Notch4 together caused severe defects in embryonic vascular remodeling similar to those seen in Dll4 knockouts, whereas induced expression of activated Notch4 in the developing mouse vasculature also caused abnormal vessel structure and patterning resulting in embryonic death (see ref. 5 for a review).

We and others have shown that Notch activation diminishes responses to vascular endothelial growth factor (VEGF), a principal proangiogenic factor (6, 7). However, integration of many pathways is required for angiogenesis. One such pathway is linked to the chemokine receptor CXCR4, expressed in endothelial cells, and its ligand, stromal-derived growth factor 1 (SDF1 or CXCL12). SDF1 is a chemoattractant for endothelial cells, which induces formation of capillary sprouts (see ref. 8 for a review). Blocking SDF1 or CXCR4 disrupts endothelial cell network formation on Matrigel (9). Knocking out either CXCR4 or SDF1 causes impaired vascular development, including defective formation of large vessels supplying the gastrointestinal tract (see ref. 10 for a review). CXCR4 and its ligand are also regulated by hypoxia via hypoxia-inducible factor-1α (11).

We hypothesized that Notch signaling regulates multiple pathways involved in tumor angiogenesis. Here, we report that Notch stimulation induced by Dll4 down-regulates CXCR4 expression, inhibiting responses toward SDF1. This represents a new integrative role for Notch signaling in endothelial function and has implications for therapeutic angiogenesis.

Cells, cell culture, constructs, and reagents. Human umbilical vascular endothelial cells (HUVEC) were isolated as described (9) and used through passage 7. The LZRSpBMN-linker retroviral construct with full-length Dll4 was made and propagated, and retroviral infections were performed as described (7). Infected HUVECs were fluorescence-activated cell sorting sorted for enhanced green fluorescent protein (EGFP) expression to achieve >90% purity; populations used were 60% to 90% EGFP positive. Endogenous Dll4 is usually not detected by reverse transcription-PCR (RT-PCR) in HUVEC. Recombinant human Dll4 (rhDLL4; R&D Systems) was diluted in 0.2% type B bovine gelatin (Sigma Chemical Co.) and coated on plates at a final concentration of 1 μg/mL. The γ-secretase inhibitor L-685,458 (Sigma Chemical or Peptide Institute, Inc.), dissolved in DMSO (Sigma Chemical), was used at 2 μmol/L.

Migration assay. Endothelial migration assays were performed as described (7). HUVECs were detached with 5 mmol/L EDTA and placed in the upper chambers, whereas migration medium with or without 100 ng/mL SDF1 was placed in the lower chamber. After 16 to 20 h of incubation at 37°C, viable cells in the lower chamber were counted.

Attachment assay. Endothelial cell attachment assays were performed using 96-well plates coated with 0.1% bovine serum albumin (BSA; Sigma Chemical), 5 mmol/L laminin (Trevigen), or 2 to 4 μg/mL SDF1 (PeproTech) overnight at 4°C. After blocking with 1% BSA for 1 h at room temperature, 2.5 × 10⁴ to 5 × 10⁴ of EDTA-detached HUVEC per well were plated in “attachment medium” [M199 (Mediatech) with 25 μg/mL porcine heparin (Sigma Chemical) and 1% heat-inactivated FBS (Biofluids)]. After 1 h at 37°C, plates were fixed with 4% paraformaldehyde (Sigma Chemical) and stained with 0.05% crystal violet (Sigma Chemical). Images were taken using a Retiga 1300 digital camera (QImaging) under phase-contrast microscopy (Olympus 1 × 51 with a 10 × 0.25 PhL lens; Olympus Optical Co.), and images from IPLab for Windows software (Scanalytics, Inc.) were imported into Adobe Photoshop. Percentage attachment was measured using ImageJ software correcting for pixels from adherent vector-transfected HUVEC.

Matrigel cord formation assay. The assay was performed as described (7). HUVECs were plated (40 × 10³ to 75 × 10³) onto 24-well plates precoated with solidified Matrigel (Collaborative; BD PharMingen). Neutralizing antibodies to human VEGF [mouse monoclonal antibody (mAb) A4.6.1; a gift of Genentech, Inc.] or to CXCR4 (mouse mAb clone 44716.111; R&D Systems) were added to the cultures at 10 μg/mL. After 18 h of incubation, cells were photographed under phase-contrast microscopy and images were imported into Adobe Photoshop (Adobe Systems).

Flow cytometry. Cells were detached with EDTA and stained with phycoerythrin-labeled CD184 (BD PharMingen). Data were collected using a FACSCalibur flow cytometer (Becton Dickinson) and analyzed using CellQuest software (Becton Dickinson).

Western blotting. Lysis buffer consisting of 1% Triton X-100, 25 mmol/L Tris (pH 7.5), 150 mL NaCl, and 5 mmol/L EDTA was supplemented with protease inhibitor cocktail set III (Calbiochem). Cell lysates (30 μg protein) were solubilized in NuPAGE lithium dodecyl sulfate sample buffer (Novex), incubated at 37°C for 30 min, and run through 4% to 12% NuPAGE Bis-Tris gels (Invitrogen). After transfer, Immobilon-P membranes (Millipore) were incubated overnight with antibodies against the following proteins: CD184 (2B11, 1:250), RDC1 (12870, 1:250; Abcam), SDF1 (BAF 310, 1:1,000; R&D Systems), VEGF (1:500; Santa Cruz Biotechnology), and cleaved Notch1 (Val 1744, 1:1,000; Cell Signaling Technology). Relative protein expression levels were estimated by membrane rehybridization with anti-actin antibody and detection as described (7).

RNA preparation and quantitative RT-PCR. Total RNA extraction from HUVEC and quantitative RT-PCR were performed as described (7). The following primers were used: glyceraldehyde-3-phosphate dehydrogenase (GAPDH), 5′-GAGTCCACTGGCTGCTTCA (sense) and 5′-GGGGTGCTAAGCAGTTGGT (antisense); CXCR4, 5′-GGCAAACTGGTACTTTGGGA (sense) and 5′-GACGCCAACATAGACCACCT (antisense). Total RNA was extracted from 25 mg frozen tumors using the FastRNA Pro Green kit (Qbiogene) and TRI Reagent. After treatment with DNA-free kit (Ambion), 1 μg total RNA was reverse transcribed into cDNA with the High Capacity cDNA Archive kit (Applied Biosystems). Quantitative PCR was performed in triplicate using the Exiqon system (Roche). Each reaction consisted of 25 μL containing 10 to 25 ng of cDNA, 12.5 μL of 2× ABsolute QPCR Mix (ABgene), 1 μL each of 10 μmol/L forward and reverse primers (Invitrogen), and 0.25 μL probe (Roche). The comparative C_t method was used for quantification. Expression level of mouse CXCR4 was normalized to that of mouse β-actin.

Transient transfection and luciferase assay. The full-length human CXCR4 promoter (pGL2-2632/+86) was a gift of Dr. W. Krek (Friedrich Miescher Institute of Biomedical Research, Basel, Switzerland; ref. 11); the control PGL-2-Basic vector was from Promega. Transient transfections of CXCR4 promoter and control in HUVEC were performed using Amaxa nucleofector system (Amaxa Biosystems), with Cell Line Nucleofector Solution V (100 μL/reaction). Cells were cotransfected with 2.5 μg phRL-SV40 (Renilla luciferase reference plasmid; Promega) to control for transfection efficiency. After transfection, cells were incubated (six-well plates) 5 to 6 h at 37°C in complete culture medium. Firefly luciferase and Renilla luciferase activities from transfected cells were measured in triplicate microtiter wells using Dual-Glo Luciferase Assay System (Promega). Relative luciferase activity is defined as the ratio of firefly luciferase to Renilla luciferase activity.

In vivo studies. The human glioblastoma U87GM cell line was maintained in RPMI 1640 supplemented with 10% FCS, 2 μmol/L l-glutamine, 50 IU/mL penicillin, and 50 μg/mL streptomycin sulfate. Cells were transduced with retrovirus containing full-length human Dll4 or empty vector as previously described (12). Retroviral supernatants were prepared using the bicistronic pLZRS-IRES-GFP plasmid (gift from Maarten van Lohuizen, Cancer Institute Antoni van Leeuwenhoek Hospital, Amsterdam, the Netherlands) with the Phoenix amphotropic packaging cell line (gift from Garry P. Nolan, Stanford School of Medicine, Stanford, CA).

Six- to 8-week-old female BALB/c severe combined immunodeficient mice (Harlan Sprague Dawley, Inc.) were implanted s.c. with 10⁷ U87 cells (200 μL injection volume consisting of culture medium and an equal volume of Matrigel). Each group consisted of five mice. When tumors reached the maximum size (1.44 cm² surface area) permitted by the Home Office license, mice were sacrificed and tumors were excised.

CXCR4 immunochemistry. Rat mAb MAB21651 (25 μg/mL; R&D Systems) was used for 90 min on cryostat tumor tissue sections fixed in acetone for 10 min and then allowed to dry. The secondary anti-mouse antibody (cross-reactive with rat 1:20 dilution) and the 3,3′-diaminobenzidine developer were from DAKO (Envision kit, ChemMate-real).

Statistical analysis. Group differences were evaluated by Student's t test; P values of <0.05 were considered significant.

Dll4 expression is associated with impaired endothelial cell migration and attachment mediated by SDF1. Previously, we reported that HUVECs retrovirally transduced with full-length Dll4 are deficient in their ability to respond to VEGF and to form cord-like structures on Matrigel, a mixture of extracellular matrix proteins (7). Because the chemokine receptor CXCR4 contributes to cord formation (9), we compared the effects of CXCR4 neutralization to overexpression of Dll4. CXCR4 neutralization impaired HUVEC cord formation similar to Dll4 overexpression, whereas VEGF neutralization minimally affected this process (Fig. 1A). This observation raised the possibility that Dll4 overexpression in HUVEC might reduce CXCR4 function. We tested the migration of Dll4-transduced HUVEC in response to SDF1 (100 ng/mL), the unique CXCR4 ligand, which is a chemoattractant for endothelial cells (9). We found that Dll4 overexpression significantly (P = 0.007) inhibited HUVEC migration toward SDF1 in Transwell migration assays (Fig. 1B).

We next examined whether CXCR4 binding to SDF1 might also be impaired in Dll4-overexpressing HUVEC. After coating ELISA plates with recombinant SDF1α (2–4 μg/mL in PBS), BSA (1% in PBS), or laminin (5 μmol/L in PBS) we measured HUVEC adherence. In wells coated with 1% BSA in PBS, no HUVEC attachment occurred by empty vector or Dll4 HUVEC (data not shown), whereas on laminin-coated wells both empty vector–transduced and Dll4-transduced HUVEC adhered alike. On wells coated with SDF1, HUVEC that overexpressed Dll4 attached less well than control HUVEC (Fig. 1C). Quantitative analysis of attachment in eight independent experiments revealed that Dll4-transduced HUVEC adhered significantly less (P = 0.02) to SDF1-coated wells than empty vector–transduced HUVEC (Fig. 1D).

CXCR4 down-regulation in endothelial cells that overexpress Dll4. CXCR4 is a G protein–coupled receptor that mediates SDF1 responses (8). We examined whether the diminished attachment and migration to SDF1 in Dll4-overexpressing cells might be attributable to reduced cell expression of CXCR4. By flow cytometry (Fig. 2A), surface CXCR4 levels were reduced in Dll4-overexpressing HUVEC [mean fluorescence intensity (MFI) = 67.1] compared with control HUVEC (MFI = 129.8). This CXCR4 reduction in Dll4-transduced HUVEC populations was similarly observed in Dll4-expressing HUVEC (60–90% of cells) and the Dll4-negative HUVEC (10–40% of cells), suggesting an extrinsic action of Dll4. Western blot analysis showed markedly diminished levels of CXCR4 protein in lysates from HUVEC-overexpressing Dll4 compared with vector-transduced HUVEC (Fig. 2B,, left). Measurement of relative ratios of CXCR4/actin band intensities revealed that CXCR4 protein levels were ∼6-fold lower in Dll4-HUVEC compared with vector-HUVEC. Recently, the orphan receptor RDC1 was identified as a second receptor for SDF1 (13). We analyzed RDC1 expression in Dll4-overexpressing and control HUVEC. By Western blotting using specific antibodies, we detected similar RDC1 expression in empty vector–transduced and Dll4-transduced HUVEC (Fig. 2B).

We used quantitative RT-PCR to assess levels of CXCR4 mRNA in Dll4- and vector-transduced HUVEC. CXCR4 mRNA levels were significantly (P = 0.0001, eight determinations) reduced in Dll4-overexpressing HUVEC compared with vector-infected HUVEC (Fig. 2C). To test whether Dll4 transcriptionally down-regulates CXCR4 expression, we transiently transfected a luciferase reporter containing the human CXCR4 promoter (PGL-2-CXCR4) or a control promoter (PGL-2-Basic) in Dll4-overexpressing and control HUVEC. Levels of relative luciferase activity induced by the CXCR4 promoter-reporter were significantly reduced (P = 0.003, five determinations) in Dll4-transduced cells compared with control cells, whereas levels of luciferase activity induced by the control reporter were similar (Fig. 2D).

Notch signaling is required for Dll4-induced CXCR4 down-regulation in HUVEC. To evaluate whether down-regulation of CXCR4 expression in Dll4-overexpressing HUVEC is a result of Notch activation, we looked for evidence of Notch1 cleavage. By immunoblotting, we detected significantly higher levels of the COOH-terminal Notch1 fragment in Dll4-transduced HUVEC compared with control cells (Fig. 3A). To assess the potential contribution of reverse signaling by Dll4, we used rhDLL4, which lacks the intracellular and transmembrane domains. Such Dll4 fragment is not known to exist in nature, but rhDLL4 was shown to activate Notch signaling in vitro if attached to the wells (7). We cultured HUVEC for 48 h on plates precoated with BSA (1 μg/mL) or rhDLL4 (1 μg/mL). In four determinations, we found significantly (P = 0.015) decreased levels of CXCR4 mRNA in HUVEC activated with rhDLL4 compared with controls (Fig. 3B,, left). Quantitative RT-PCR showed that expression of CXCR4 mRNA diminishes by 50% after HUVECs are cultured for 16 h onto immobilized rhDLL4 and continues to decline through 72 h (Fig. 3B,, right). By immunoblotting, we found a corresponding reduction of CXCR4 protein levels in HUVECs cultured on rhDLL4-coated wells for 72 h, whereas RDC1 levels remained similar (Fig. 3C). We also found similar levels of both SDF1 and VEGF in HUVEC lysates collected after 72 h of cell culture on BSA- or rhDLL4-coated wells (Fig. 3C).

We showed previously that the γ-secretase inhibitor L-685,458 can block activation of the Notch pathway and reverse the effects of Dll4 signaling by recovering VEGF receptor 2 (VEGFR2) expression and VEGF-induced proliferation (7). To further establish that extracellular Dll4 down-regulates CXCR4 expression by signaling through the Notch pathway, we used the γ-secretase inhibitor L-685,458. Treatment with the inhibitor reduced the levels of Notch1 cleavage (Fig. 3D,, left) and significantly reconstituted CXCR4 mRNA levels in HUVEC cultured on rhDLL4 (P = 0.003; Fig. 3D , right).

Dll4 regulates CXCR4 expression in vivo. To assess whether Dll4 down-regulation of CXCR4 is also observed in vivo, we used a human glioblastoma xenograft tumor model. Immunodeficient mice were injected s.c. with the human glioblastoma U87 cells, which were retrovirally transduced with human Dll4 or empty vector only. The Dll4 tumors that developed at the injection site displayed fewer but larger vessels than the vector control tumors (14). We assessed vascular CXCR4 expression in the tumors. By immunohistochemistry, the small vessels in the control tumors generally expressed CXCR4 (Fig. 4A,, top). However, the larger vessels typically found in the Dll4 tumors did not express CXCR4 (Fig. 4A,, bottom). In additional experiments, we evaluated the relative levels of mouse CXCR4 mRNA in tumor extracts, which includes the vasculature and other host cells in the tumor microenvironment. By real-time PCR using mouse-specific probes, we found that the relative levels of CXCR4 mRNA were significantly reduced in Dll4 tumors compared with controls (Fig. 4B). These results provide evidence that the Notch ligand Dll4 down-regulates CXCR4 expression in the tumor vasculature.

Here, we show that Dll4 transcriptionally down-regulates CXCR4 expression in endothelial cells and, by this mechanism, reduces responsiveness to SDF1 in HUVEC. Similar down-regulation of CXCR4 was seen by constitutive expression of full-length Dll4 and by stimulation with an immobilized extracellular portion of Dll4. This provides evidence that the intracellular and transmembrane domains of Dll4 are not required for activity in the current system. Because the extracellular domain of Dll4 seems sufficient to down-regulate CXCR4 expression, Dll4 likely acts through Notch and not through Dll4 signaling. Consistent with an involvement of Notch signaling, Dll4-transduced HUVEC displayed much higher levels of Notch1 COOH-terminal fragment, which is generated by enzymatic cleavage on ligand binding and is essential to Notch signaling (1, 15). Furthermore, partial recovery of CXCR4 expression was seen in the presence of the γ-secretase inhibitor L-685,458, as previously shown for VEGFR2 (7). These results suggest that down-regulation of CXCR4 by Dll4 is due to activation of the Notch pathway and provide the first evidence of association between these pathways.

The previously identified orphan receptor RDC1 was recently shown to be an additional SDF1 receptor in T lymphocytes (13). We questioned whether Dll4 down-regulates RDC1 but found that neither full-length Dll4 expression nor recombinant rhDLL4 down-regulates RDC1. Therefore, we conclude that impaired responses to SDF1 by Dll4-transduced HUVEC are likely attributable to down-regulation of CXCR4 expression.

VEGF signaling increases SDF1 protein levels in HUVEC (9), and Notch activation down-regulates the principal VEGF receptor, VEGFR2 (6, 7). The possibility existed that Dll4 might indirectly reduce SDF1 expression in HUVEC through down-regulation of VEGFR2. However, our immunoblotting results show similar levels of SDF1 in HUVEC cultured with and without rhDLL4. SDF1 activation of CXCR4 was reported to increase levels of VEGF mRNA (16). We previously determined that Dll4 up-regulation in HUVEC does not affect VEGF mRNA or protein secretion (7), and we now show that VEGF protein levels are similar in BSA- and rhDLL4-stimulated HUVEC. We conclude that, under these culture conditions, Dll4 does not significantly affect SDF1 protein levels, and Dll4-induced down-regulation of CXCR4 is not associated with diminished expression of VEGF. This latter conclusion is consistent with the observation that CXCR4 inactivation in vivo decreases tumor growth without affecting levels of VEGF (17).

The effects of Dll4 on tumor angiogenesis are complex. Several approaches to inhibit Dll4/Notch interactions have resulted in excessive tumor neovascularization, which was nonfunctional and resulted in reduced tumor perfusion and tumor growth (18–20). Dll4 overexpression in tumor cells was reported to reduce the number of tumor vessels, which were larger and more effective, and tumor growth was generally not reduced (14, 18). The observation we have made that Dll4/Notch activation reduces CXCR4 expression in tumor vessels adds yet another element of complexity to Dll4/Notch vascular effects. Based on the critical role of CXCR4 on angiogenesis (8–10), it is possible that some of the reported vascular abnormalities derived from blocking or promoting Dll4/Notch signaling may be contributed by CXCR4 modulation. CXCR4 inhibitors are currently in clinical trial and it would be of interest to assess vascular effects in patients on such trials.

Although this is the first report linking the Notch signaling pathway to CXCR4 regulation, evidence for this relationship can be seen in Kaposi sarcoma lesions. Although hypoxia has been shown to enhance transcription of CXCR4 in vitro (11), and Kaposi sarcoma lesions are known for their hypoxic environment, CXCR4 was undetectable in the angiogenic environment within Kaposi sarcoma lesions, whereas vessels in the surrounding normal tissues stained positive (21). Dll4 expression is increased by hypoxia at sites of active angiogenesis (2, 22). Our report showing that Dll4 stimulation down-regulates CXCR4 expression may explain absence of detectable CXCR4 within Kaposi sarcoma lesions.

Dll4 may serve as a switch for CXCR4 regulation in cells of endothelial lineage. The endothelial cells, which express Dll4 (2, 18) in the tumor microenvironment, would be expected to promote Notch activation and CXCR4 down-regulation in adjacent endothelial cells. Because CXCR4 plays critical roles in morphogenic processes that accompany neovascularization (8–10), its down-regulation in tumor endothelial cells would be expected to contribute to vascular defects that are commonly observed in the tumor vasculature. Dll4 may also serve to regulate CXCR4 in endothelial progenitor cells (EPC), which contribute to tumor neovascularization (23). As the tumor environment becomes hypoxic, the resulting secretion of angiogenic cytokines, such as VEGF and SDF1, serves to recruit VEGFR2- and CXCR4-expressing EPCs from circulation (23, 24). Because both these receptors are down-regulated in endothelial cells by Dll4 stimulation (7), once EPCs are recruited and make direct contact with endothelial cells that express Dll4, activation of Notch signaling could serve to dampen expression of CXCR4 and VEGFR2 and help retain the cells for incorporation within the tumor vasculature. Delicately balanced levels of Notch would be required to maintain appropriate levels of VEGFR2 and CXCR4 expression in endothelial cells.

Our discovery of Dll4 as a regulator of CXCR4 identifies a previously unknown relationship between Notch signaling and CXCR4 expression and reveals an additional mechanism by which Dll4 plays a role in angiogenesis regulation. This may have implications for antiangiogenic therapy targeting CXCR4, where Dll4 in tumor vasculature may induce resistance, and it would therefore be of interest to combine such therapy with Notch inhibition. It may also have important implications for antiangiogenic therapy targeting the Dll4/Notch signaling pathway, which has recently emerged an exciting and promising approach to cancer treatment (see ref. 25 for a review).

Grant support: Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research, and “The 6th Framework Programme of the European Union (Angiotargeting)”. C.K. Williams is a NIH-University of Oxford Health Science Scholar.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Drs. V. Kapoor and W. Telford for cell sorting and flow cytometry support; Dr. Garry P. Nolan for providing the retroviral vector; and Drs. M. Narazaki, N. Labo, P. Gasperini, P. McCormick, L. Harrington, O. Salvucci, and H. Turley for technical support and advice.