Imatinib inhibits VEGF-independent angiogenesis by targeting neuropilin 1–dependent ABL1 activation in endothelial cells

Because NRP1 is known to interact with FN-binding integrins (Fukasawa et al., 2007; Robinson et al., 2009; Valdembri et al., 2009), we first examined the requirement of NRP1 for EC adhesion, spreading, and actin remodeling. For these experiments, we used human dermal microvascular ECs (HDMECs) because dermal vasculature naturally undergoes extensive angiogenesis during wound healing. We additionally used human umbilical vein ECs (HUVECs), as they have been widely used to study VEGF-induced signaling mechanisms (e.g., Soker et al., 2002; Pan et al., 2007a; Chen et al., 2010). We transfected primary cells with a previously validated siRNA that targets NRP1 or a control nonsense siRNA. Having confirmed the efficacy of this approach (Fig. 1 A), we tested HDMEC and HUVEC adhesion to tissue culture dishes coated with 10 µg/ml FN, a concentration which effectively promotes cell adhesion and migration (Clark et al., 1986; Sottile et al., 1998; Tvorogov et al., 2005). In contrast to earlier studies, which reported that NRP1 loss inhibits HUVEC adhesion at FN concentrations <5 µg/ml (Murga et al., 2005; Valdembri et al., 2009), we found that adhesion was not compromised by NRP1 deficiency in either HDMECs or HUVECs at 10 µg/ml FN (Fig. 1, B and C). We had therefore identified conditions suitable to study FN-induced cell spreading, actin remodeling, and cell migration of ECs in the absence of prior defects in cell attachment.

Phalloidin staining of filamentous (F) actin showed that NRP1-deficient HDMECs (Fig. 1, D and E) adopted an abnormal round morphology with abundant cortical actin when plated on FN (Fig. 1 F). In contrast, control cells appeared elongated and contained stress fibers typical of adherent cells (Fig. 1 F). VEGFR2 protein levels (Fig. 1 D) and mRNA expression (Fig. 1 E) were slightly but significantly decreased in NRP1-deficient cells, but cells transfected with siRNA targeting VEGFR2 spread well on FN and assembled many stress fibers (Fig. 1 F), indicating that cytoskeletal defects in NRP1-deficient cells were not caused by low VEGFR2 expression. High magnification images revealed phalloidin-positive, filopodia-like microspikes in HDMECs 60 and 120 min after plating on FN, which were significantly reduced after NRP1 but not VEGFR2 down-regulation (Fig. 1 F, right columns). Quantitation confirmed that NRP1 but not VEGFR2 down-regulation significantly impaired cell spreading (Fig. 1 G) and microspike extension (Fig. 1 H) in cells plated on FN. Strikingly, addition of VEGF165, which is known to bind both VEGFR2 and NRP1, did not rescue the cytoskeletal defects of FN-stimulated HDMECs lacking NRP1 (Fig. 1 I). NRP1 therefore functions independently of VEGF and VEGFR2 to regulate ECM-induced cell spreading and actin remodeling in ECs.

Impaired cell spreading and actin remodeling in NRP1-deficient ECs predicts defective cell motility. By tracking the behavior of individual HDMECs after plating on FN, we indeed found that control cells were significantly more motile than cells lacking NRP1 (Videos 1 and 2; Fig. 1, J and K). Consistent with reduced motility, transwell assay measuring ECM-induced haptotaxis demonstrated reduced migration of NRP1-deficient compared with control cells toward FN, but VEGFR2 knockdown did not affect migration (Fig. 1 L). Similar results were obtained with HUVECs (Fig. 1 L). To examine if NRP1 deficiency impaired EC migration in another species, we used mouse lung ECs (MLECs) from Nrp1 conditional null (Nrp1^fl/fl) mice in the transwell assay after infecting them with adenovirus expressing GFP as a control or CRE recombinase to down-regulate NRP1 (Fig. 1 M). As observed for human ECs, migration onto FN substrates was significantly reduced in MLECs lacking NRP1 compared with controls (Fig. 1 N). NRP1 deficiency therefore impairs ECM-induced EC motility and migration independently of VEGFR2 in a pathway that is conserved between mice and humans.

Because FN stimulation activated NRP1-dependent actin remodeling and cell migration independently of VEGF165 and VEGFR2, we sought to identify candidate effectors with a phosphokinase antibody array (e.g., Zhuang et al., 2012). For this experiment, HDMECs were transfected with si-NRP1 or control siRNA (Fig. 2 A), and then serum-starved and stimulated with VEGF165 (Fig. 2 B) or plated on FN (Fig. 2 C). We observed reduced VEGF165-induced activation of the MAPK p38 in ECs lacking NRP1, as expected (Kawamura et al., 2008), and additionally impaired FN-induced P38 activation in ECs lacking NRP1 (Fig. 2, B and C). The screen also showed that NRP1 down-regulation in VEGF165-stimulated HDMECs reduced AKT activation (Fig. 2 B), as previously reported for HUVECs (Pan et al., 2007a,b; Koch et al., 2011). Unexpectedly, however, NRP1 down-regulation did not impair AKT phosphorylation in FN-stimulated ECs (Fig. 2 C).

Consistent with findings in HUVECs and zebrafish (Lanahan et al., 2010, 2013), NRP1 loss decreased both VEGF165- and FN-induced activation of PLCγ1 (Fig. 2, B and C), known to promote the phosphorylation of the focal adhesion protein PXN downstream of FN-mediated integrin activation to increase cell spreading and migration (Choi et al., 2007; Zaidel-Bar et al., 2007). Consistent with reduced PLCγ1 activation and a requirement for NRP1 in integrin ligand-mediated EC spreading and motility, PXN phosphorylation on tyrosine residue (Y) 118 was also reduced in FN-stimulated HDMECs lacking NRP1 (Fig. 2 C). In contrast, VEGF165-induced PXN phosphorylation was unaffected by NRP1 loss (Fig. 2 B). Loss of NRP1 did not affect the phosphorylation of FAK on residue Y397 downstream of either VEGF or FN (Fig. 2 B). This was unexpected, as phosphorylation of this residue activates FAK kinase, which is then able to phosphorylate PXN (Chen et al., 1996; Parsons et al., 2000). Collectively, the phosphoantibody screen indicated essential and distinct roles for NRP1 in VEGF165- versus ECM-stimulated signal transduction and suggested that NRP1 promotes PXN phosphorylation downstream of ECM activation independently of VEGFR2 or FAK.

Immunoblotting validated the results from the phosphoantibody screen and allowed us to more directly compare NRP1 and VEGFR2 dependence of specific signal transduction pathways (Fig. 2, D and E). For these experiments, HDMECs were serum-starved and plated on FN or grown on tissue culture plastic, serum-starved, and then stimulated with VEGF165. As expected, VEGF165-induced VEGFR2 phosphorylation was inhibited by NRP1 or VEGFR2 down-regulation (Fig. 2 D). In contrast, FN stimulation did not induce comparable VEGFR2 stimulation (Fig. 2 E). Also as expected, NRP1 or VEGFR2 knockdown decreased VEGF165-induced AKT phosphorylation; however, FN-induced AKT phosphorylation was unaffected by NRP1 or VEGFR2 deficiency (Fig. 2, D and E). The AKT pathway therefore transduces growth factor but not ECM signals in a VEGFR2- and NRP1-dependent fashion. Agreeing with experiments in other types of ECs (Pan et al., 2007a; Lanahan et al., 2010; Ren et al., 2010), VEGF165-induced ERK1/2 phosphorylation was impaired after NRP1 or VEGFR2 knockdown in HDMECs (Fig. 2, D and F). Surprisingly, however, ERK1/2 phosphorylation was significantly increased in response to FN (Fig. 2, E and F). Finally, we used immunoblotting to examine how NRP1 and VEGFR2 knockdown affected PXN tyrosine phosphorylation. Although NRP1 loss reduced PXN phosphorylation in FN-stimulated HDMECs, VEGFR2 knockdown did not (Fig. 2 E). In contrast, PXN was constitutively phosphorylated in adherent HDMECs, with phosphorylation increasing slightly after VEGF165 stimulation in a VEGFR2-dependent manner (Fig. 2 D; Birukova et al., 2009), but NRP1 was not required for VEGF165-induced PXN phosphorylation (Fig. 2 D). Together with the phosphoantibody screen, these findings demonstrate that ECM- and VEGF165-induced signal transduction pathways are differentially affected by NRP1 loss, and that NRP1-dependent PXN phosphorylation occurs independently of VEGFR2.

Because pPXN is recruited to focal adhesions to promote their turnover during cell migration (Zaidel-Bar et al., 2007; Pasapera et al., 2010) and NRP1-deficient ECs have reduced pPXN (Figs. 1 and 2), we examined NRP1-dependent pPXN localization in FN-stimulated HDMECs by immunostaining. In control cells, pPXN increased over time and was present in focal adhesions at the end of F-actin stress fibers, correlating with an elongated cell shape (Fig. 3 A). FN-stimulated HDMECs, therefore, displayed the hallmarks of polarized cells. Although VEGFR2 down-regulation did not impair these responses, HDMECs lacking NRP1 displayed their characteristic rounded morphology with abundant cortical actin; correlating with the lack of stress fibers, pPXN levels were significantly decreased, with remaining pPXN being localized mainly to the cell periphery (Fig. 3, A and B). Immunoblotting established that siRNA targeting had been effective and further confirmed reduced pPXN in NRP1-deficient cells (Fig. 3 B); in contrast, pPXN was slightly but significantly increased in VEGFR2-deficient cells (Fig. 3, B and C). These findings demonstrate that NRP1 promotes PXN phosphorylation in a VEGFR2-independent fashion.

Because both the phosphoantibody screen and immunoblotting suggested a functional relationship between NRP1 and pPXN, we examined if both proteins interact. In agreement, NRP1 immunoprecipitation, followed by pPXN immunoblotting, demonstrated an FN-dependent interaction in HDMECs (Fig. 3 D). Moreover, immunofluorescence showed partial colocalization in peripheral cell areas resembling focal adhesions (Fig. 3, E and F). The finding that NRP1 and pPXN associate in FN-stimulated ECs agrees with a study showing that overexpressed NRP1 colocalizes with α5β1 integrins at endothelial adhesion sites (Valdembri et al., 2009).

To examine the relationship between NRP1 and pPXN in a gain-of-function experiment, we transfected a construct encoding murine NRP1 into HDMECs and validated NRP1 overexpression with an antibody specific for murine NRP1 (Fig. 3 F). This experiment demonstrated increased PXN phosphorylation after FN stimulation in transfected compared with neighboring untransfected cells (Fig. 3 F, top). Importantly, overexpressing a VEGF165 binding-deficient NRP1 mutant (mNRP1^D320A; Herzog et al., 2011) similarly increased PXN phosphorylation (Fig. 3 F, bottom), confirming that FN/NRP1-dependent pPXN induction does not depend on VEGF binding to NRP1. Collectively, these results establish that NRP1 regulates PXN phosphorylation independently of VEGF.

Because NRP1 lacks catalytic activity, it requires a partner kinase to promote FN-induced PXN phosphorylation, but this kinase is not VEGFR2 (Fig. 1) or FAK (Fig. 2, B and C). A good candidate is the cell adhesion–associated kinase ABL1, which interacts with PXN in FN-stimulated fibroblasts (Lewis and Schwartz, 1998) as well as integrins β1 and β2 (Cui et al., 2009; Baruzzi et al., 2010). Furthermore, the Y118 residue that is phosphorylated in an NRP1-dependent fashion resides in an ABL1 phosphorylation consensus site (Cujec et al., 2002), and ABL1 is an effector of NRP1 and integrins in tumor matrix remodeling (Yaqoob et al., 2012). To investigate ABL1 function in FN-stimulated ECs, we used two independent but complementary methods: siRNA-mediated knockdown of ABL1 (Fig. 4) and pharmacological inhibition of ABL1 kinase activity (Fig. 5).

ABL1 knockdown was achieved by transfecting HDMECs with control siRNA or siRNA targeting ABL1 (si-ABL1) and confirmed by qPCR (Fig. 4 A). We found that targeting ABL1 caused a phenotype similar to NRP1 knockdown. Thus, immunostaining revealed significantly reduced pPXN levels, with residual pPXN accumulating in the cell periphery (Fig. 4, B and C). Moreover, there was a conspicuous absence of pPXN-positive focal adhesion contacts and stress fibers, but abundant cortical actin, correlating with impaired cell spreading and a round rather than elongated cell shape, as observed after NRP1 knockdown (Fig. 4 B). Immunoblotting for the phosphorylated form of the ABL1 target CRKL (Lewis et al., 1996) confirmed ABL1 down-regulation as well as significantly reduced FN-stimulated pPXN in ABL1 knockdown compared with control cells (Fig. 4, D and E). This analysis also showed increased ERK1/2 phosphorylation after ABL1 knockdown (Fig. 4 D), similar to NRP1 knockdown (Fig. 2, E and F). Unexpectedly, NRP1 protein expression also decreased in a time-dependent manner after ABL1 knockdown (Fig. 4 D), but this decrease was not caused by reduced NRP1 mRNA levels (si-ABL1 relative to si-control: 1.18 ± 0.13 at 0 min; 1.17 ± 0.11 at 30 min; 1.07 ± 0.07 at 60 min; mean ± SD; n = 3; P > 0.05). ABL1 therefore positively regulates NRP1 protein stability in HDMECs in response to FN stimulation. Hence, ABL1 may affect NRP1-dependent pPXN up-regulation in a dual fashion, first by directly promoting PXN phosphorylation, and second by stabilizing NRP1 protein and therefore increasing the capacity of HDMECs to respond to FN and up-regulate PXN phosphorylation via an NRP1-dependent pathway.

Because PXN phosphorylation promotes the turnover of focal adhesions that serve as traction points during cell migration (Zaidel-Bar et al., 2007), we predicted that ABL1 loss would impair EC migration on FN. In agreement, a transwell assay showed a 40% reduction in haptotactic migration after ABL1 knockdown (Fig. 4 F), which was similar to the reduction observed after NRP1 knockdown (Fig. 1, K and L). NRP1 and ABL1 therefore similarly regulate PXN phosphorylation, actin remodeling, and cell migration in FN-stimulated ECs.

The similar cellular changes caused by ABL1 and NRP1 deficiency raised the possibility that both proteins interact to regulate PXN phosphorylation. To investigate this idea, we cotransfected HEK293 cells with expression constructs for NRP1 and ABL1. ABL1 immunoprecipitation followed by NRP1 immunoblotting, and vice versa, showed that both proteins formed a complex (Fig. 4, G and H). Immunoprecipitating ABL1 from HDMECs followed by NRP1 immunoblotting confirmed the existence of an endogenous complex and further showed that this complex had recruited pPXN 60 min after FN stimulation (Fig. 4 I). To investigate whether this complex also forms in murine ECs, we inactivated two conditional Nrp1-null alleles in newborn mice with a tamoxifen-inducible endothelial Cre transgene (Fantin et al., 2013) and then isolated ECs (Fig. 4 J). As observed for HDMECs, ABL1 immunoprecipitated NRP1 in control MLECs adhering to FN, and complex formation was increased after detaching the cells and plating them for 30 min on FN (Fig. 4 K). Moreover, pPXN was present in the ABL1–NRP1 complex of MLECs from both conditions (Fig. 4 K), as seen in HDMECs (Fig. 4 I). In contrast, NRP1 was not immunoprecipitated effectively in MLECs from tamoxifen-treated mice, and complex formation of ABL1 with pPXN was reduced (Fig. 4 K). These data demonstrate that ABL1 and NRP1 form a complex and that ABL1 recruits pPXN in an NRP1-dependent fashion.

To examine if ABL1 kinase activity was required for FN-induced PXN phosphorylation, we treated HDMECs with Imatinib (Glivec), a small molecule inhibitor that effectively targets ABL1 but not VEGFR2 (Buchdunger et al., 2002; Anselmi et al., 2012) and has been approved for therapy in cancers with up-regulated ABL1 kinase activity (Druker et al., 1996). As observed in si-ABL1–transfected cells, Imatinib-treated HDMECs formed few stress fibers, but abundant cortical actin, and they adopted a round shape with reduced cell spreading; moreover, they had low pPXN phosphorylation, with residual pPXN in the cell periphery rather than in areas where stress fibers normally terminate in focal adhesions (Fig. 5 A). The quantitation of pixel intensities in immunostains (Fig. 5 B) and immunoblots (Fig. 5 C) confirmed significantly reduced PXN phosphorylation in Imatinib-treated compared with control cells. NRP1 immunoprecipitation of Imatinib-treated HDMECs followed by ABL1 immunoblotting showed that both proteins formed a complex before FN stimulation, independently of ABL1 kinase activity (Fig. 5 D). We conclude that ABL1 stimulates PXN phosphorylation in an NRP1-dependent fashion when ECs migrate on FN.

To examine if NRP1 and ABL1 also promote PXN phosphorylation in an ECM-dependent angiogenesis model in vivo, we studied the perinatal mouse retina; in this organ, endothelial sprouts headed by filopodia-studded tip cells migrate toward astrocyte-localized VEGF in the retinal periphery, with filopodia being guided by astrocyte-derived FN (Fig. 6 A; Ruhrberg et al., 2002; Gerhardt et al., 2003; Stenzel et al., 2011). Immunostaining confirmed FN deposition around IB4-stained vessels and ahead of the vascular front in a fine meshwork characteristic of astrocytes (Fig. 6 B). To investigate how NRP1 deficiency affected pPXN levels during retinal angiogenesis, we could not use Nrp1^−/− mice, as they die before birth (Kawasaki et al., 1999). Instead, we compared Nrp1^+/+ and Nrp1^+/− littermates because the latter are viable but nevertheless have mild angiogenesis defects (Fantin et al., 2013). Immunoblotting confirmed significantly decreased NRP1 levels in P6 Nrp1^+/− mice (Fig. 6 C). Single optical slices, acquired by confocal microscopy after immunolabeling, revealed pPXN staining in ECs at the IB4-positive vascular front, including tip cells and their filopodia, and also some pPXN ahead of the vascular front (Fig. 6 D). As observed for si-NRP1–targeted HDMECs, pPXN staining appeared reduced in Nrp1^+/− littermates, both in vascular and avascular areas (Fig. 6 D). To confirm that this defect was cell autonomous in vascular endothelium, we also used Cre-LoxP recombination approach to delete NRP1 (Fig. 1). Thus, conditional Nrp1-null mice lacking or expressing a tamoxifen-inducible, endothelial-specific Cre transgene were treated with tamoxifen from postnatal day 2 (P2) to P5 and stained for IB4 and pPXN. As seen in Nrp1^+/− mice, tamoxifen-treatment reduced pPXN in Cre-expressing but not Cre-negative (control) littermates (Fig. 6 D).

To quantify pPXN specifically in ECs, we isolated IB4-positive areas by applying a virtual mask over IB4-negative areas (Fig. 6 D, right). Quantification of pPXN relative to IB4 pixel intensity in masked 3D projections of z-stacks showed significantly reduced endothelial pPXN staining at the vascular front of heterozygous Nrp1-null mutants (Fig. 6 E) and endothelial-specific Nrp1-null mutants (Fig. 6 F) compared with controls.

To examine if ABL1 was also required for PXN phosphorylation in retinal ECs, we treated neonates with Imatinib or vehicle (Fig. 6, G and H). Single confocal slices showed that pPXN staining was reduced in P6 retinas after 2 days of treatment (Fig. 6 G), as observed for NRP1 mutants. To quantify pPXN specifically in ECs, we again isolated IB4-positive areas with a virtual mask (Fig. 6 G, right) and found that endothelial pPXN pixel intensity was significantly reduced after Imatinib treatment (Fig. 6 H). ABL1 therefore promotes PXN phosphorylation in angiogenic ECs in vivo, similar to NRP1 and as observed for FN-stimulated ECs in vitro.

Because ABL1 is involved in NRP1-mediated actin remodeling, cell migration, and PXN phosphorylation, we investigated next if it is also essential for blood vessel growth in vivo. IB4 labeling showed reduced vascular extension and network density in retinal flat mounts of P6 Imatinib-treated compared with vehicle-treated mice (Fig. 7 A). Quantitation confirmed a small but significant reduction in vascular extension across the retina (Fig. 7 B), as previously reported for loss of astrocyte FN (Stenzel et al., 2011). In agreement, FN immunostaining showed reduced astrocyte FN deposition ahead of the vascular front (Fig. 7, C and D), likely due to Imatinib targeting of PDGFR signaling in astrocytes (Buchdunger et al., 2002). In contrast, we observed abundant FN deposition around retinal vessels in both control and Imatinib-treated mice, with no difference in FN pixel intensity in IB4-positive areas (Fig. 7, C and D). Vascular FN assembly was therefore not compromised by ABL1 targeting.

High magnification images of the retinal vascular front in control and Imatinib-treated mice showed that vessels had sprouted and formed numerous tip cell filopodia (Fig. 7 E), consistent with the concept that VEGF gradients rather than matrix pathways drive these processes (Ruhrberg et al., 2002; Gerhardt et al., 2003). However, sprouts in Imatinib-treated mice appeared longer and wider, with fewer lateral protrusions connecting to neighboring sprouts (Fig. 7, E and F, arrows), and filopodia often appeared thin, wavy, and misoriented (Fig. 7 E, arrowheads). This phenotype agrees with a role for the NRP1–ABL1 complex in endothelial actin remodeling (Fig. 1). Moreover, the filopodia defects resemble those previously seen at the retinal vascular front of mice with a genetic lack of astrocyte FN (Stenzel et al., 2011). Yet loss of astrocyte FN in itself does not impair vessel sprouting and branching (Stenzel et al., 2011), suggesting that the more severe vascular defects of Imatinib-treated retinas are due to an EC-autonomous role for ABL1.

To further define the vascular defects caused by ABL1 inhibition, we compared vessel sprouting and branching in mice treated with Imatinib for different times (Fig. 7, F and G). Quantitative analysis revealed significantly fewer tip cells, defined by the presence of filopodial bursts at the vascular front (Fig. 7 G). Consistent with reduced tip cell density and impaired lateral sprout extension, we also observed a significant and Imatinib dose-dependent reduction in branch points behind the vascular front, indicative of reduced sprout fusion (Fig. 7 G). These findings therefore agree with the recently shown requirement of actin-based filopodia dynamics for sprout fusion and therefore vascular network formation (Phng et al., 2013).

Because our data suggest that ABL1 is involved in NRP1-mediated angiogenic processes, we also compared the retinal vascular phenotype of Imatinib-treated mice to that of endothelial-specific Nrp1 mutants. For this experiment, we treated Cre-positive and Cre-negative littermates with tamoxifen in a regimen similar to that used for Imatinib. We observed a significant reduction in both tip cell density and vascular branch points in mutants, similar to Imatinib-treated mice (Fig. 7 G).

Based on the in vitro and in vivo data we have described here, we conclude that NRP1 facilitates angiogenesis in response to integrin ligands via ABL1-mediated PXN phosphorylation and actin remodeling. Because this pathway operates independently of VEGF and VEGFR2, NRP1 plays a dual role in angiogenesis and vascular morphogenesis by independently promoting ECM-stimulated and growth factor–induced signals in ECs (Fig. 7 H). In agreement, vascular extension is more severely inhibited in EC-specific Nrp1 knockout compared with Imatinib-treated retinas (reduction in vascular extension compared with littermate controls: 32 ± 4% for endothelial Nrp1 knockouts vs. 14 ± 1% for Imatinib-treated mice).

To assess the relevance of the FN-induced NRP1–ABL1 pathway for pathological vessel growth, we used a mouse model of oxygen-induced retinopathy (OIR; Connor et al., 2009). In this model, the sequential exposure of mouse pups to hyperoxia followed by normoxia leads to the formation of retinal neovascular lesions that resemble those observed in PDR patients and in babies with ROP, which arises after moving them out of incubators with high oxygen tension (Connor et al., 2009). Specifically, the exposure of neonatal mice to hyperoxia from P7 to P12 induces vasoobliteration of central retinal capillaries, which causes central retinal hypoxia on return to room air. The ensuing up-regulation of VEGF and other proangiogenic factors then activates angiogenesis, but this process fails to effectively revascularize the retina and instead leads to abundant tuft-like vascular malformations that protrude into the vitreous (Smith et al., 1994). Immunostaining of flat-mounted retinas on P17, 5 days after return to room air, showed that these neovascular tufts were FN positive (Fig. 8 A), suggesting that OIR is a suitable model to examine the role of NRP1 and ABL1 in ECM-mediated angiogenesis pathways. In agreement, Imatinib treatment of mouse pups from their return to normoxia until P17 significantly reduced revascularization of avascular areas (AV; Fig. 8, B and C) and effectively inhibited the formation of neovascular tufts (Fig. 8, B–D).

In agreement with ABL1 acting downstream of NRP1, a reduction in both revascularization of avascular areas and neovascular tuft formation was also seen in conditional endothelial Nrp1-null mutants treated with a tamoxifen regimen analogous to that used for Imatinib treatment (Fig. 8, E and F). Even though the vascular OIR phenotype of Nrp1 mutants was similar to that of Imatinib-treated mice, the reduction in neovascular growth was greater, consistent with a dual role for NRP1 in ECM-stimulated ABL1 and VEGF-induced VEGFR2 pathways in postnatal angiogenesis (reduction in neovascular tuft formation compared with littermate controls: 60 ± 13% for endothelial Nrp1 knockouts vs. 28 ± 10% for Imatinib-treated mice; Fig. 7 H).

Finally, we assessed the efficacy of Imatinib in targeting angiogenesis in the ear wound healing model (Fig. 8 G), in which the healing of ear punch injuries depends on neoangiogenesis (Cho et al., 2006) and the wound margin is rich in FN (Fig. 8 H). Daily Imatinib treatment of adult mice from day 3 until day 20 after ear punch inhibited wound closure, with a significantly larger wound diameter at both 14 and 21 d compared with vehicle-treated controls (Fig. 8, H and J). Although FN deposition at the wound was not obviously affected by Imatinib treatment, blood vessel density in the wound margin was reduced (Fig. 8, H–I′). Imatinib therefore targets pathological vessel growth in both the eye and skin.

In vertebrates, organ formation and homeostasis require the delivery of oxygen and nutrients through blood vessel networks, which form in response to signals provided by the vascular endothelial growth factor VEGF (Ruhrberg, 2003). VEGF-driven vascular morphogenesis entails the processes of vasculogenesis, angiogenesis, and arteriogenesis, which all rely on signal transduction by the VEGF receptor tyrosine kinase VEGFR2 (Koch et al., 2011). However, it is poorly understood how VEGF signaling is integrated with other signaling pathways to ensure that these complementary processes take place in appropriate contexts (Carmeliet and Jain, 2011). We show here that the VEGF receptor NRP1 has an unexpected role in a non-VEGF–driven angiogenesis pathway that is activated by ECM signals and involves the tyrosine kinase ABL1. The discovery of this pathway has important implications for our understanding of both physiological and pathological vessel growth.

The coactivation of VEGF-induced and ECM-stimulated signaling pathways benefits ordered blood vessel growth in complex tissues. Thus, it was previously shown that endothelial tip cells extend filopodia to sense VEGF gradients and to align themselves along matrix templates for directional migration (Ruhrberg et al., 2002; Gerhardt et al., 2003; Uemura et al., 2006; Stenzel et al., 2011). Moreover, signaling responses induced by ECM-bound VEGF165 involve integrin activation and differ mechanistically from those induced by soluble VEGF165 (Chen et al., 2010). However, the functional significance of NRP1’s ability to associate with integrins independently of VEGFR2 (Soker et al., 2002; Fukasawa et al., 2007; Robinson et al., 2009; Valdembri et al., 2009) was not previously understood. We demonstrate here that NRP1 promotes integrin ligand-induced EC motility in a VEGFR2-independent mechanism that involves ABL1-dependent PXN phosphorylation, focal adhesion formation, and actin remodeling. We further show that this pathway is essential for normal vessel sprouting and vascular plexus formation in vivo. NRP1 therefore has a dual role in ECs by contributing to both VEGF-dependent VEGFR2 activation and ECM-stimulated but VEGF/VEGFR2-independent signaling via ABL1.

The novel proangiogenic NRP1 function we have identified differs fundamentally from prior models, which proposed that endothelial NRP1 acts exclusively as a VEGF coreceptor to enhance VEGFR2 signaling. Collectively with recent findings that NRP1 associates with VEGFR2 to stimulate VEGF-induced arteriogenesis (Lanahan et al., 2013), endothelial NRP1 appears to activate specific tyrosine kinase pathways in a context-dependent fashion to regulate vascular morphogenesis. The existence of an ABL1-dependent but VEGFR2-independent role for NRP1 in angiogenesis may also explain why blood vessel growth is affected more severely by loss of NRP1 than loss of NRP1-binding VEGF isoforms (Ruhrberg et al., 2002; Gerhardt et al., 2004; Raimondi and Ruhrberg, 2013), why the loss of VEGF binding to NRP1 causes milder embryonic vascular defects than loss of NRP1 (Fantin et al., 2014), and why anti-NRP1 and anti-VEGF treatments synergize to block angiogenesis-dependent tumor growth (Pan et al., 2007a). Importantly, our finding that NRP1 mediates ABL1 signaling to regulate actin cytoskeleton and cell migration in response to extracellular signals has far reaching implications beyond understanding angiogenesis, as NRP1 is important for the behavior of diverse cell types, including immune cells, neural crest cells, neurons, and tumor cells.

VEGFR2-independent functions of NRP1 in ECs in vitro have previously been ascribed to NRP1 cytoplasmic domain signaling. For example, the extracellular part of the EGF receptor, when linked to the NRP1 transmembrane and cytoplasmic domains, enables EGF-stimulated EC migration (Wang et al., 2006). The cytoplasmic NRP1 domain also promotes FN fibrillogenesis in arterial ECs by regulating endosomal trafficking of activated α5β1 integrin (Valdembri et al., 2009) and is involved in ABL1-mediated FN fibrillogenesis in myofibroblasts (Yaqoob et al., 2012). However, NRP1 cytoplasmic domain–mediated signal transduction pathways or integrin endocytosis are unlikely to play major roles in angiogenesis because genetic mouse studies showed that this NRP1 domain is dispensable for physiological and pathological angiogenesis (Fantin et al., 2011; Lanahan et al., 2013). Instead, NRP1 contributes to angiogenesis solely via its membrane-tethered extracellular domain (Fantin et al., 2011). Our finding that NRP1 promotes matrix signaling to enhance angiogenesis therefore agrees with prior observations that the extracellular rather than intracellular NRP1 interacts with α5β1 integrin (Fukasawa et al., 2007; Valdembri et al., 2009). Because the NRP1 extracellular domain complexes with integrins, but the NRP1 cytoplasmic tail is not required for angiogenesis, NRP1 most likely associates with ABL1 indirectly, possibly via integrins, as they are known to recruit ABL1 (Cui et al., 2009; Baruzzi et al., 2010).

The NRP1 dependency of FN-stimulated EC motility and migration was reflected in the defective activation of several previously identified integrin targets in NRP1 knockdown cells, including PLCγ1, ABL1, and PXN. However, targeting the FN-stimulated NRP1–ABL1 pathway impaired retinal angiogenesis more severely than deleting astrocytic FN or single integrins (Stenzel et al., 2011). The greater dependency of angiogenesis on NRP1 than on individual ECM components or different integrins may be explained by the partial redundancy of proangiogenic matrix components for at least some signaling responses (Stenzel et al., 2011) and the presence of multiple endothelial FN receptors, including α5β1 and αvβ3 (Dejana et al., 1990). Moreover, NRP1 can itself act as an adhesion receptor (Shimizu et al., 2000).

Our study identified endothelial NRP1 as a bimodal regulator of vascular morphogenesis that regulates ABL1-mediated signal transduction and actin remodeling during angiogenesis, even though it promotes VEGFR2 trafficking and signaling during arteriogenesis. In addition to regulating physiological vessel growth, the NRP1–ABL1 pathway promotes vascular pathology that can be inhibited by treatment with Imatinib, a small molecule inhibitor of ABL1, or through the genetic ablation of NRP1 in ECs; thus, both approaches significantly and similarly reduced vessel growth in a mouse model of human retinopathy. Targeting NRP1-mediated ABL1 signaling might therefore provide a novel therapeutic opportunity to enhance the efficacy of current anti-angiogenic therapies that focus on manipulating VEGF signaling through VEGFR2, for example to inhibit tumor angiogenesis or neovascular eye disease. In particular, the FDA-approved drug Imatinib might provide a complementary treatment to improve the responsiveness of patients to anti-angiogenic therapies, for example in metastatic colon cancer, PDR, or AMD. Additionally, Imatinib treatment may be beneficial for patients with cancers that are fuelled by vasculature that is resistant to anti-VEGF therapy. As shown in this preclinical study, Imatinib might be delivered systemically to treat vascular eye pathology to circumvent complications of intravitreal administration of anti-VEGF therapies, such as increased intraocular pressure or endophthalmitis. In conclusion, the unexpected finding that NRP1 promotes angiogenesis in a VEGFR2-independent fashion via ABL1 supports a model in which NRP1-mediated ECM signaling proceeds independently of, but synergistically with, VEGFR2-mediated growth factor signaling to drive angiogenic tissue invasion in development and disease (Fig. 7 H).

HDMECs (Promocell) were cultured in MV2 media with supplements (Promocell). HUVECs (Promocell) were cultured in EBM2 media (Lonza) containing 10% FBS (Life Technologies), 300 µg/ml EC growth supplement (ECGS; Sigma-Aldrich), and 1 µg/ml hydrocortisone (Sigma-Aldrich). Cells were transfected with lipofectamine RNAIMAX using SMARTpool siRNA targeting NRP1, VEGFR2, or ABL1 (Dharmacon), or Silencer negative control siRNA (Applied Biosystems). For overexpression studies, HDMECs were transfected with Lipofectamine according to the manufacturer’s protocol. In some experiments, cells were incubated with 10 µM Imatinib (Cambridge Bioscience), a concentration known to effectively target ABL1 kinase (Chislock et al., 2013). In some experiments, cells were stimulated with 50 ng/ml VEGF165 for the indicated times. Primary MLECs were isolated from mice between 1 and 2 mo of age by magnetic-activated cell sorting (MACS) with PECAM antibodies (BD). MLECs were cultured on 10 µg/ml FN in DMEM-GlutaMAX supplemented with 20% FBS, nonessential amino acids (Life Technologies), and ECGS. MLECs from Nrp1^fl/fl conditional null mice were infected with adenovirus-expressing CRE recombinase or with control virus–expressing GFP, incubated for 24 h in culture media, washed in PBS, and cultured in fresh media for 48 h before migration and immunoblotting experiments.

HUVECs or HDMECs transfected with targeting or control siRNA were detached with trypsin, resuspended in MV2 medium containing 0.5% BSA, and incubated at 37°C for 1 h on a rotating wheel. 10,000 cells/well were seeded on 96-well tissue culture plates (Nunc; Thermo Fisher Scientific) coated overnight with 10 µg/ml FN in PBS or, as an internal control, PBS containing 1% BSA, for the indicated times in triplicate, washed in PBS, fixed in 4% formaldehyde in PBS, and stained with 0.1% crystal violet for 30 min. After washing three times with PBS for 15 min, dye was extracted with 10% SDS in PBS for 15 min on an orbital shaker to determine absorbance at 570 nm.

HDMECs were transfected with siRNA and MLECs treated with adenovirus before overnight serum starvation and then plated onto FN-coated transwell inserts (8.0 µm pores, 10 mm diameter; Nunc) that had been preincubated with MV2 containing 0.5% BSA (HDMECs) or DMEM containing 0.5% BSA (MLECs). After 4 h, cells on the upper face of the insert were removed with a cotton bud, while transmigrated cells on the underside were fixed with 4% formaldehyde in PBS for 10 min and stained with 0.1% crystal violet solution for 10 min. Images were acquired with a phase-contrast light microscope using a c-plan 10×/0.22 objective (Leica). Transmigrated cells in duplicate inserts were counted in a minimum of five images per insert in three independent experiments.

HDMECs were transfected with siRNA, serum-starved overnight 56 h after transfection, and then detached and seeded in duplicates into 6-well plates that had been coated overnight with 10 µg/ml FN. Cells were then imaged in an environmental chamber maintaining a 5% CO₂ atmosphere at 37°C with a 10× phase contrast objective at 2-min intervals for 200 min using an inverted phase contrast microscope (Axiovert-200M; Carl Zeiss). The motility of 25 cells per each condition from two experiments was analyzed using the ImageJ MTtrackJ extension module (National Institutes of Health).

HDMECs were transfected with siRNA. 56 h after transfection, cells were serum-starved overnight and then stimulated with 50 ng/ml VEGF165 for 10 min or detached using 0.5× trypsin/EDTA (PAA), resuspended in MV2 serum-free media, and plated for 30 min on tissue culture plates coated with 10 µg/ml FN. ECs were lysed as described in Immunoblotting and immunoprecipitation and phosphokinase antibody array was performed according to the manufacturer’s instructions (R&D Systems).

mRNA was extracted using the RNeasy system (QIAGEN). cDNA was prepared using Superscript II reverse transcription (Invitrogen) and amplified with (SYBR Green; Applied Biosystems) and the following oligonucleotide primers: NRP1, 5′-GAAAAATCGAATGCTGAT-3′ and 5′-AATCCGGGGGACTTTATCAC-3′; ABL1, 5′-GAGGGCGTGTGGAAGAAATA-3′ and 5′-GGTAGCAATTTCCCAAAGCA-3′; VEGFR2, 5′-AGATGGTGTAACCCGGAGTG-3′ and 5′-ACATGTCAGCGTTTGAGTGG-3′; and GUS, 5′-AAACGATTGCAGGGTTTCAC-3′ and 5′-CTCTCGTCGGTGACTGTTCA-3′. Gene expression was analyzed with the quantitative 7500 Real-Time PCR System (Applied Biosystems). Data analysis was performed using the SDS software (version 2.3; Applied Biosystems).

Cells were lysed with lysis buffer (50 mM Tris-HCl, pH 7.4, 5 mM EDTA, 0.1% NP-40, and 250 mM NaCl) in the presence of protease inhibitor cocktail 2 and phosphatase inhibitor cocktail (Sigma-Aldrich). Lungs from P7 mice were lysed in GL35 (Cytoskeleton, Inc). 25 µg of protein was resuspended in Laemmli sample buffer, denatured for 5 min, separated by SDS-PAGE, and transferred to nitrocellulose membrane (Whatman).

Protein extracts were analyzed by SDS-PAGE and transferred to nitrocellulose membrane (Whatman) for immunoblotting with the following primary antibodies: rabbit anti-GAPDH (Abcam), goat anti-NRP1 C-19 (Santa Cruz Biotechnology, Inc.) or AF566 (R&D Systems); mouse anti-ABL1 (BD); rabbit anti-pCRKL (Y207), rabbit anti-pKDR (Y1175) or KDR, rabbit anti-pPXN (Y118), rabbit anti-pAKT (S473) or AKT, or rabbit anti-pERK1/2 or ERK1/2 (Cell Signaling Technology).

Serum-starved ECs were detached with 0.5× trypsin/EDTA and lysed or seeded into FN-coated plastic dishes for the indicated times before lysis. In some experiments, ECs were treated with 10 µM Imatinib (Glivec; Cambridge Bioscience) for 30 min before and during FN stimulation. ECs were lysed in 50 mM Tris, pH 8.0, 50 mM KCl, 1% (vol/vol) NP-40 in the presence of protease inhibitor cocktail 2 and phosphatase inhibitor cocktail. 1 mg of protein was incubated with 3 µg goat anti-NRP1 (Fantin et al., 2010), rabbit anti-ABL1 (Roig et al., 2000), or control goat or rabbit IgG (Santa Cruz Biotechnology) and then with 30 µl magnetic protein G Dynabeads (Life Technologies) at 4°C. Beads were collected with a Dynabead magnet, washed three times with lysis buffer on a rotating wheel at 4°C for 5 min, and resuspended in 50 µl Laemmli sample buffer for SDS-PAGE and immunoblotting.

HDMECs were transfected with siRNA. 56 h after transfection, cells were serum-starved overnight, detached using 0.5× trypsin/EDTA, resuspended in MV2, and seeded on FN-coated coverslips for various time points. ECs were fixed in 4% formaldehyde in PBS for 30 min, permeabilized in PBS containing 0.25% Triton X-100 for 2 min, and then stained with rabbit anti-pPXN or mouse anti-NRP1 (R&D Systems), followed by Alexa Fluor 594–conjugated goat anti–rabbit or anti–mouse antibodies (Life Technologies) and FITC-conjugated phalloidin (Sigma-Aldrich). Images were acquired with a Plan Apochromat 63× 1.4 NA oil objective on a LSM700 confocal microscope (Carl Zeiss). P6 mouse retinas or 20-µm cryosections through ear wounds were immunolabeled as described (Fantin et al., 2011) using rabbit anti-pPXN or rabbit anti-FN (Abcam) together with biotinylated isolectin B4 (IB4; Sigma-Aldrich) or rat anti-PECAM (BD), followed by Alexa Fluor 594– or 488–conjugated goat anti–rabbit or anti–rat antibodies and Alexa Fluor 647–conjugated streptavidin (Life Technologies). Confocal z-stacks were acquired with a laser scanning confocal microscope (LSM710; Carl Zeiss). FN mean pixel intensity in nonsaturated maximal intensity projections was determined with ImageJ for an area of 0.06 mm² ahead of the vascular front and plotted as fold change compared with control. Endothelial pPXN or FN staining were isolated by virtually masking out nonendothelial staining after IB4 surface rendering with IMARIS (Bitplan AG) and expressed as relative to IB4 mean pixel intensity.

All animal procedures were performed in accordance with institutional and UK Home Office guidelines. We used C57BL/6 wild-type mice (Charles River), littermate wild-type and heterozygous Nrp1-null mutants (Kitsukawa et al., 1997), and mice carrying two floxed conditional null Nrp1 alleles (Nrp1^fl/fl) combined with Pdgfb-iCre-ERT2-Egfp with codon-improved Cre on a C57BL/6 background (Fantin et al., 2013). For tamoxifen induction of CRE-mediated recombination, 0.2 mg of tamoxifen (Sigma-Aldrich) dissolved in peanut oil to 2 mg/ml was administered via intraperitoneal injection into pups at the indicated times. In some experiments, mice were injected intraperitoneally with 100 mg/kg/d Imatinib in 0.9% saline solution for the indicated times. We induced OIR as previously described (Connor et al., 2009). In brief, nursing dams and their pups were maintained at 75 ± 3% O₂ in an oxygen supply chamber from P7 to P12; dams were rested in room air for 1 h each day. Dams and pups were returned to room air on P12 and the pups culled on P17 to isolate retinas for immunolabeling; the size of the avascular and neovascular areas was determined with Photoshop (CS4; Adobe) in images acquired with a stereo fluorescent microscope (SZX16; Olympus) equipped with a digital camera (Hamamatsu Photonics). Ear wound healing was assessed in adult mice with a 2-mm hole in the ear center, which was made with a metal ear punch to identify mice during standard mouse husbandry.

Statistical analyses were performed with Prism (GraphPad Software) or Excel (Microsoft). Data obtained from cell lines were analyzed with a two-tailed, paired Student’s t test, whereas data from retina samples were analyzed using a two-tailed, unpaired Student’s t test or ANOVA.

Video 1 shows images of HDMECs transfected with control siRNAs that were acquired every 2 min for 200 min. Video 2 shows images of HDMECs transfected with NRP1 targeting siRNAs that were acquired every 2 min for 200 min.

We are grateful to the staff of the Biological Resources Unit at the UCL Institute of Ophthalmology for help with mouse husbandry and thank the Imaging Facility at the UCL Institute of Ophthalmology for maintenance of the confocal microscopes. We also thank Michael Simons and John Greenwood for critical reading of the manuscript.

This study was supported by a research fellowship from the British Heart Foundation to C. Raimondi (FS/13/35/30148), a research grant from the British Heart Foundation to C. Ruhrberg (PG/10/86/28622), and a New Investigator Award from the Wellcome Trust to C. Ruhrberg (095623/Z/11/Z).

The authors declare no competing financial interests.

Author contributions: C. Raimondi, A. Fantin, and C. Ruhrberg designed the research, performed the research, analyzed the data, and wrote the manuscript. A. Lampropoulou performed research and analyzed the data. L. Denti performed research. A. Chikh designed and performed research.