NRPs (neuropilins) are co-receptors for class 3 semaphorins, polypeptides with key roles in axonal guidance, and for members of the VEGF (vascular endothelial growth factor) family of angiogenic cytokines. They lack a defined signalling role, but are thought to mediate functional responses as a result of complex formation with other receptors, such as plexins in the case of semaphorins and VEGF receptors (e.g. VEGFR2). Mutant mouse studies show that NRP1 is essential for neuronal and cardiovascular development, whereas NRP2 has a more restricted role in neuronal patterning and lymphangiogenesis, but recent findings indicate that NRPs may have additional biological roles in other physiological and disease-related settings. In particular, NRPs are highly expressed in diverse tumour cell lines and human neoplasms and have been implicated in tumour growth and vascularization in vivo. However, despite the wealth of information regarding the probable biological roles of these molecules, many aspects of the regulation of cellular function via NRPs remain uncertain, and little is known concerning the molecular mechanisms through which NRPs mediate the functions of their various ligands in different cell types.
- vascular endothelial growth factor (VEGF)
NRP1 (neuropilin-1) was originally identified as the antigen of a specific monoclonal antibody called A5, raised against neuronal cell-surface proteins presumed to be involved in neuronal recognition between the visual centres and the optic-nerve fibres of Xenopus laevis [1,2]. It was named NRP because A5 bound in the tadpole optic tectum to the superficial layer of the dense network of glial processes, synapses, axons and dendrites interspersed with neurons, called the neuropile . Subsequently, analysis of NRP1 transgenic and deficient mouse and zebrafish (Danio rerio) models established an essential role of NRP1 in the development of the embryonic nervous and cardiovascular systems [4–7]. NRP1 and the structurally related molecule, NRP2, are receptors both for class 3 semaphorins, a family of secreted polypeptides with key roles in axonal guidance, and for various members of the VEGF (vascular endothelial growth factor) family of angiogenic cytokines, but are thought to transduce functional responses only when co-expressed with other receptors: plexins in the case of semaphorins and VEGFRs (VEGF receptors) for VEGFs. NRPs are also highly expressed in diverse tumour cell lines and human neoplasms and have been implicated in tumour growth and vascularization in vivo [8–11]. More recently, NRP1 has been implicated as a novel mediator of the primary immune response [12–14].
These findings suggest that NRPs are multi-functional co-receptors essential for neuronal and cardiovascular development, but potentially with additional roles in diverse physiological and disease-related settings. However, despite the wealth of information regarding the likely biological functions of these molecules, many aspects of the regulation of cellular function via NRPs remain uncertain, and little is known concerning the molecular mechanisms through which NRPs mediate the functions of their various ligands in different cell types.
This review will describe NRP structure and biology with a focus on recent advances that provide novel insight into the role of NRPs in angiogenesis and neurogenesis. Also highlighted is the recent emergence of NRPs as novel therapeutic targets in cancer.
NRP1 and NRP2 are transmembrane glycoproteins of up to 923 and 926 amino acids respectively, sharing a similar domain structure and an overall amino acid homology of 44% . NRPs comprise large extracellular regions containing two CUB [complement binding factors C1s/C1r, Uegf, BMP1 (bone morphogenetic protein 1)] (a1/a2) domains, two Factor V/VIII homology (b1/b2) domains , a MAM (meprin, A5 antigen, receptor tyrosine phosphatase μ) (c) domain, a single transmembrane domain and small cytoplasmic domains (44 amino acids for NRP1 and 43 for NRP2) (Figure 1). The C-terminal three amino acids, SEA, present in NRP1 and the NRP2a isoform, form a consensus PDZ [post synaptic density protein (PSD95), Drosophila disc large tumor suppressor (DlgA), and zonula occuldens-1 protein] domain binding motif, which mediates association with a PDZ domain protein called NIP1 (neuropilin-interacting protein-1), synectin or GIPC (RGS-GAIP-interacting protein) .
The CUB (a1/a2) domains share homology with the complement binding factors C1s/C1r, sea urchin fibropellins (called Uegf) and BMP1 (bone morphogenetic protein 1). CUB domains comprising approx. 110 amino acid residues are found in functionally diverse, often developmentally regulated, proteins. Most CUB domains contain four conserved cysteine residues which probably form two disulfide bridges and are predicted to form a β-barrel structure similar to that of immunoglobulins. CUB domains mediate protein–protein associations in the complement system, including Ca2+-dependent C1r–C1s association and serum mannose-binding-protein binding. In NRPs, the CUB domains are essential for binding of semaphorins, as described in greater detail below.
The b1/b2 domains consist of two tandem regions, each of approx. 150 amino acids, sharing homology with the C-terminal (C1/C2) domains of blood coagulation Factors V and VIII. In Factors V and VIII, the C1/C2 domains are part of a larger functional domain that promotes binding of membrane phospholipids (phosphatidylserine) on the surface of platelets and endothelial cells. The NRP1 b1 domain is essential for VEGF-A165 binding, although the b2 domain is also required for optimal binding. In addition, the b1/b2 domains are required for semaphorin binding to NRPs (described in detail below). The crystal structure of the NRP1 b1 domain displays a strong similarity to the three-dimensional structure of the Factor V/VIII C2 domain, but with significant differences . Crystal structures of the NRP1 b1 domain bound either to the peptide ligand, tuftsin , or to antibodies directed against either the VEGF or Sema3A binding domains , have identified key residues and structural features required for ligand binding.
The c or MAM domain is a 170-amino-acid region found in the extracellular domains of functionally diverse proteins, including meprin (a cell-surface glycoprotein), A5 antigen and receptor tyrosine protein phosphatase μ (hence MAM). MAM domains are thought to mediate homophilic protein–protein associations important for homodimerization or oligomerization, and have also been implicated in regulating protein stability. The MAM domain is thought to play a role in NRP-1 oligomerization [20,21].
Immunoblotting for NRP1 in certain cell types revealed a high- molecular-mass band of >250 kDa in addition to the major species of ∼130 kDa corresponding to the full-length protein (Figure 2) . This high-molecular-mass protein is modified by addition of a GAG (glycosaminoglycan) moiety to Ser612 in the linker region between the b2 and MAM domains (Figure 1). Shintani et al.  reported modification of NRP1 by both heparan sulfate and CS (chondroitin sulfate) GAGs , though findings from our laboratory indicate that high-molecular-mass NRP1 contains exclusively CS-GAG. NRP1-CS-GAG appears to be preferentially expressed in VSMCs (vascular smooth muscle cells) and several tumour cell lines, and expressed at a low or undetectable level in endothelial cells (; Figure 2; C. Pellet-Many, P. Frankel and I. Zachary, unpublished work). In contrast, there is no evidence for a high-molecular-mass-GAG-modified form of NRP2. In addition to GAG modification, both NRP1 and NRP2 appear to be modified by asparagine (N)-linked glycosylation, as indicated by the effect of tunicamycin inhibition of N-glycosylation on the apparent molecular mass of the 130 kDa NRP1 band ( and Figure 2). CS-GAG modification is reported to enhance both VEGF binding to NRP and cell survival, and to down-regulate VEGFR2 expression levels in VSMCs . Further studies are needed to fully determine how glycosylation alters the ligand binding and other functional properties of NRPs.
NRPs have the ability to bind with high affinity two structurally unrelated classes of ligands with distinct biological functions, the class 3 semaphorins and selected members of the VEGF family. These ligand interactions are summarized in Figure 3.
Semaphorins constitute a large protein family including both transmembrane and secreted species expressed in diverse metazoans (worms, insects, crustaceans, fishes and vertebrates) and also in viruses, but not in plants or protozoans [24,25]. They are divided into seven different classes containing more than 20 proteins in vertebrates [26–28]. The class 3 semaphorins are secreted proteins, most of which require NRP1 or 2 as obligate co-receptors. The major ligand for NRP1 is Sema3A (also named collapsin-1), a factor that induces collapse of the growth cone in selected sensory and sympathetic neurons such as the DRG (dorsal root ganglia), which also repels DRG axons and is essential for neurogenesis in development . NRP1 also binds Sema3F with lower affinity and several other secreted class 3 semaphorins (Figure 3), although the functional roles of these interactions are not clear in each case. The best characterized ligands for NRP2 are Semas3C and 3F, and NRP2 also recognizes Semas3B and 3G, but not Sema3A. There is evidence that several NRP2 ligands (Semas3B, 3C and 3F) may act as antagonists of Sema3A . The role of NRPs in semaphorin signalling is discussed in more detail below.
The binding of Sema3A to the CUB (a1/a2) domains of NRP1, depicted in Figure 4, is mediated by the sema domain, a conserved extracellular domain of ∼500 amino acids forming a β-sheet propeller structure held together by four disulfide bonds , while a C-terminal region of Sema3A rich in basic residues, and uniquely present in class 3 semaphorins, interacts with the b1/b2 domains [30–32]. Deletion of a1, a2 or b1 domains abolishes Sema3A binding to NRP1, whereas the b2 domain is not required . Both a and b domains are also essential for the Sema3F interaction with NRP2 . The binding interactions between semaphorins and the b domain appear to differ from those involved in VEGF-A binding at least in the case of NRP2, since mutagenesis of the NRP2 b1 domain can modify VEGF-A165 affinity without affecting Sema3F binding . On the basis of crystallization studies of the sema domain of Sema3A, it has been proposed that it also interacts with plexins (Figure 4), thereby creating a heterotrimeric complex . Whether direct interaction of Sema3A with plexins occurs in cells or is biologically relevant, awaits confirmation.
The members of the class 3 semaphorins have different affinities for NRP1 and NRP2 and require distinct plexins to transduce their signal. Thus in sensory neurons NRP1 acts mainly through plexin A4, whereas signalling via NRP2 is mediated largely by plexin A3 .
VEGF or VEGF-A, also called VPF (vascular permeability factor), plays a major role in vasculogenesis (early embryonic formation of blood vessels from haematopoietic precursor cells) and angiogenesis (the sprouting of new vessels from pre-existing ones occurring during development and also in the adult) . VEGF polypeptides are anti-parallel homodimeric 40 kDa secreted glycoproteins covalently linked by two intermolecular disulfide bridges, related to the PDGF (platelet-derived growth factor) family of growth factors. In mammals, the VEGF family has five members: VEGF-A, -B, -C, -D and PlGF (placental growth factor). In addition, VEGF-E is encoded by the orf virus, and VEGF-F is found in snake venom [36–40].
Alternative splicing generates multiple VEGF isoforms which differ in their receptor affinity and specificity. VEGF-A occurs in at least six different isoforms: VEGF121, VEGF145, VEGF165, VEGF165b, VEGF189 and VEGF206 [39–41]. VEGF121 lacks exons 6 (coding for a sequence rich in basic amino acids implicated in heparin binding) and 7, and has therefore been thought unable to bind either NRPs or heparin . The most active and abundant isoform, VEGF165 contains exon 7, but lacks exon 6, and is able to bind to both NRP1 and 2, indicating a key role for the domain encoded by exon 7 in NRP binding . However, VEGF145 lacks exon 7 and binds only to NRP2 . The NRP1 b1 domain is essential for VEGF165 binding, and the b2 domain is additionally required for optimal binding, whereas the a domains are largely dispensible . Although VEGF165 was originally believed uniquely able to bind to NRPs with high affinity through its exon-7-encoded C-terminal domain, it is now clear that the residues encoded by exon 8 are crucial for NRP recognition. Peptides corresponding to exon 8 inhibit VEGF binding to NRP1 , and a naturally occurring peptide, tuftsin, with homology with exon 8 also binds NRP1 [18,46].
Antibodies specifically blocking VEGF165 binding to NRP1 also inhibited endothelial-cell migration induced by VEGF121 . VEGF121, lacking exon 7 but containing exon 8, binds NRP1 in vitro in SPR (surface plasmon resonance) assays, but was unable to promote complex formation between NRP1 and VEGFR2 . These findings are consistent with a model in which initial VEGF contacts with NRP1 are mediated via an interaction between the extreme C-terminal residues encoded by exon 8 and the b1 domain (Figure 4); binding and NRP1–VEGFR2 complex formation is either stabilized by further interactions between NRP1 and exon-7-encoded residues and/or the exon 7 domain functions as a bridge between NRP1 and VEGFR2. The larger isoforms, VEGF189 and VEGF206, bind heparin, are not readily diffusible and are thought to remain sequestered by the extracellular matrix, but are theoretically able to bind NRP1. A novel isoform, VEGF-A165b, has a distinct C-terminus, RSLTRKD, encoded by an alternative exon 8 and is unable to stimulate endothelial-cell proliferation, angiogenesis or other biological activities . Possibly the weaker biological activity of VEGF-A165b is due to its inability to bind NRP1 and thereby promote productive signalling complexes with VEGFR2 .
In addition to structural features of VEGF-A and NRP1 discussed above, other factors may also influence binding. The affinity of VEGF binding to NRP1, at least in vitro, is enhanced by heparin and by increasing the density of NRP1 . As mentioned above, CS-GAG modification of NRP1 also enhances VEGF binding .
VEGF-B and VEGF-E have been shown to bind to NRP1 [50,51], whereas the VEGFR3 ligands VEGF-C and VEGF-D interact with NRP1 and NRP2 [52,53]. PlGF-2 also binds to NRP1 . There is some evidence that other growth factors may bind to NRP1. Thus FGF (fibroblast growth factor)-2 binds to NRP1 in vitro , and PDGF-BB has been co-immunoprecipitated with NRP1 in VSMCs, an observation that is suggestive of direct binding . However, the evidence for high-affinity functional binding of extracellular factors other than VEGFs and semaphorins to NRP1 is so far inconclusive, and awaits further investigation.
GENOMIC ORGANIZATION AND EXPRESSION
NRPs are present in all vertebrates so far examined, including mammals, chicken (Gallus gallus) and the zebrafish and are highly conserved between species. Although homologues of the VEGFs and semaphorins have been found in Drosophila, Caenorhabditis elegans and other invertebrates, NRPs have so far not been identified in non-vertebrate organisms. Human NRP1 and NRP2 genes are located on chromosomes 10p12 and 2q34 respectively [57,58], and both encode full-length proteins with predicted molecular masses of 130–140 kDa (923 amino acids for NRP1 and 926 amino acids for NRP2a). The NRP1 gene spans over 120 kb and is composed of 17 exons . The NRP2 gene is >112 kb, also comprising 17 exons . The strong similarities in the exon–intron organization of the NRP1 and NRP2 genes (Figure 1), their exon and intron sizes and positions of many of the splice sites suggests that they may have originated from a gene duplication event.
Alternative splicing results in the generation of several isoforms of NRP1. One membrane-associated NRP1 isoform has been identified, called NRP1(Δexon16), which lacks the 51 nucleotides corresponding to exon 16 [57,59]. NRP1(Δexon16) does not differ from the common full-length NRP1 in its binding to VEGF165, dimerization with VEGFR2 or regulation of VEGF165 signalling . Four NRP1 mRNA isoforms have so far been reported, all predicted to encode soluble proteins containing the a1/a2 and b1/b2 domains, but lacking the MAM (c), transmembrane and cytoplasmic domains (Figure 1). These isoforms vary in size from 551 to 704 amino acid residues owing to alternative splicing. However, only two soluble NRP1 (sNRP1) isoforms, s12NRP1 (NRP1 isoform b) and sIVNRP1 (NRP1 isoform c), have so far unambiguously been shown to be expressed in protein form and are found in protein databases. Soluble isoform sIIINRP1 contains a1/a2 and b1/b2 domains missing 48 amino acids at the C-terminus of the b2 domain, but contains 13 extra amino acids resulting from a shift in the reading frame of exon 12 . sIIINRP1 (551 amino acids) and sIVNRP1 (609 amino acids) occur in both normal and cancerous human tissues and both bind to VEGF165 and Sema3A . Soluble NRP1s may act as decoys, competitively binding and sequestering ligands such as VEGF165 and Sema3A, and therefore negatively regulating functions mediated by these cytokines. In support of such a role, sNRP1 inhibits tumour cell growth in vivo and triggers tumour- cell apoptosis, mimicking the effect of VEGF165 withdrawal . However, the effects of sNRP1s may be more complex than suggested by a straightforward decoy role. Thus whereas an sNRP1 monomer sequesters VEGF165 and inhibits its activity, sNRP1 dimers appear to deliver VEGF165 to endothelial-cell VEGFR2, thereby promoting angiogenesis .
Membrane-bound NRP2 exists in two major isoforms, NRP2a, which shares 44% overall homology at the amino acid level with NRP1, and NRP2b, which is identical with NRP2a in its extra-cytoplasmic domain, but exhibits only 11% homology with NRP2a in its transmembrane and cytoplasmic regions (see Figure 1). In the mouse, a total of four NRP2a isoforms are generated by alternative splicing resulting in the insertion of 0, 5, 17 and 22 (17+5) amino acids after residue 809, situated between the MAM and the transmembrane domains . In humans, only two forms of NRP2a have been cloned, NRP2a(17) and NRP2a(22), homologous with the corresponding mouse isoforms. The NRP2a(22) isoform (931 amino acids) results from the insertion of the five amino acids GENFK within the 17-amino-acid insertion of NRP2a(17) (926 amino acids; Figure 1). The NRP2a(0) and NRP2a(5) isoforms have not so far been found in human tissues. These isoforms do not appear to differ in their ligand-binding properties, but insertion of residues between the MAM and transmembrane domains might potentially alter its ability to form complexes with VEGFR2 or to homodimerize. NRP2b displays little homology with NRP2a after residue 808  with a distinct cytoplasmic domain lacking the C-terminal PDZ domain recognition sequence, SEA, required for interaction with synectin. Similar to NRP2a, NRP2b(0) and NRP2b(5) isoforms result from alternative splicing and the insertion of zero or five amino acids after amino acid 808. The marked differences in their cytoplasmic domains suggest that NRP2a and NRP2b isoforms may have divergent functions, a possibility supported by their differential tissue expression. NRP2a and NRP2b are both highly expressed in the brain, but NRP2a is preferentially expressed in the liver, lung, small intestine, kidney and heart, whereas NRP2b is present in heart and skeletal muscle .
A soluble NRP2 isoform is also generated by alternative splicing (see Figure 1), s9NRP2 (1785 bp, 555 amino acids, 62.5 kDa) consisting of the two a1/a2 domains the b1 domain and a truncated b2 domain followed by the nine amino acids VGCSWRLPL encoded by intron 9 .
NEUROPILIN FUNCTION IN DEVELOPMENT
Targeted disruption of NRP genes has demonstrated an essential dual role of these molecules in neurogenesis and cardiovascular development (see Table 1). NRP1-null mice die between E12 and E13.5 (where E is embryonic day) with a spectrum of cardiovascular and neuronal defects. Both the CNS (central nervous system) and PNS (peripheral nervous system) are severely affected with severe abnormalities in the trajectory and connection of efferent fibres of the PNS . Both small and large vessels of the yolk sacs are disorganized and the capillary network sparse, whereas aberrant embryonic macrovascular development is characterized by lack of development of the branchial arch, related great vessels and dorsal aorta, transposition of the aortic arch and insufficient septation of the truncus arteriosus . NRP1 overexpression also results in embryonic lethality with excess capillary growth, haemorrhage in the head and neck and a malformed heart, in addition to anarchic sprouting and defasciculation of nerve fibres . These transgenic embryos also appear redder than normal, suggesting that the blood vessels are leaky, perhaps due to an enhanced vascular permeability activity of VEGF-A165. In contrast with NRP1 mutant mice, NRP2-null mice survive to adulthood, with no obvious cardiovascular abnormalities, but exhibit a severe reduction of small lymphatic vessels  and capillaries, as well as abnormal guidance and fasciculation of cranial and spinal nerves [65,66]. Doubly deficient NRP1−/−/NRP2−/− mice exhibit earlier embryonic mortality than the NRP1 knockout (E8 compared with E12–13.5) and have a more severe vascular phenotype resembling the VEGF-A165 and VEGFR2 [KDR (kinase insert domain-containing receptor)] knockouts , marked by large avascular areas in the yolk sacs, and head and trunk regions, and a lack of connections between blood vessel sprouts.
An essential role of NRP1 in vascular development has also been demonstrated in other vertebrate species. In the zebrafish embryo, NRP1a is expressed in the neural tube, NRP1b is expressed in the nose and the cranial neural crest cell, NRP2a is expressed in the telencephalon and anterior pituitary, and NRP2b is expressed in the telencephalon, thalamus, hypothalamus and epiphysis [68,69]. Morpholino NRP1-knockdown in the developing zebrafish  produced a severe vascular phenotype characterized by a loss, or anarchic sprouting, of new capillaries from pre-existing intersomitic vessels, but had no effect on the formation of the major axial vessel, suggesting that, in zebrafish, NRP1 is not implicated in vasculogenesis . Defects were also observed in the developing zebrafish nervous system after NRP1 knockdown, characterized by aberrant migration and branching of motor neurons [70,71].
Important insights into the relationship between the vascular and neural functions of NRPs have come from analysis of tissue-specific knockouts and mutant knockin mice. An essential role in cardiovascular development for vascular NRP1 has been demonstrated by the generation of endothelial-specific NRP1 knockout mice. These mice exhibit mid-to-late embryonic lethality, a poorly branched vasculature and multiple defects in the major arteries and failure of septation of the major cardiac outflow tract . In contrast, knockin mice expressing a mutant NRP1 with a seven-amino-acid deletion in the a1 domain required for Sema3A binding, but retaining the ability to bind VEGF-A, survive to birth and possess no obvious cardiovascular defects, but exhibit aberrant pathfinding of sensory afferent nerves to synapses in the CNS and defasciculation of spinal and cranial nerves, most of the mice dying by P7. Specifically, the semaphorin/NRP1 signalling axis is essential for formation of cranial and spinal nerve projections, guidance of peripheral projections of bipolar neurons of the vestibular ganglion and the central projections of a subset of axons of cutaneous sensory neurons, and for basal-cortical neuron dendrite development . These findings indicate a critical role for semaphorin binding to NRP1 in axonal pathfinding, but also demonstrate the compartmentalization of the neuronal and vascular roles of NRP1. The importance of semaphorin–NRP interactions in axonal homing is further emphasized by the close similarities of the neuronal pathfinding defects in NRP1 and Sema3A mutant mice [4,5,72], and in mice lacking Sema3F and NRP2 [65,66,73–75]. The guidance of motor axons from the spinal cord during vertebrate limb bud development serves as a striking illustration of how the expression patterns of NRPs and their specific semaphorin ligands exquisitely choreograph the homing of specific subsets of axons. As they leave the spinal cord, both lateral and medial axons of the LMC (lateral motor column; LMCl and LMCm respectively) express NRP1, whereas NRP2 is restricted to the LMCm; chemorepulsive Sema3A is expressed throughout the limb bud, whereas Sema3F is localized to the dorsal limb bud. Targeted disruption of either NRP1 or Sema3A causes premature and disorganized invasion of the limb bud by motor axons, and misprojection of both LMCl and LMCm axons. Deficiency of either Sema3F or NRP2 results in selective misprojection of LMCm axons, usually expressing NRP1 and NRP2, to the dorsal limb, whereas LMCl axons normally expressing only NRP1 track normally to the dorsal limb [73,75]. However, NRPs do not mediate all functions of their class 3 semaphorin ligands. For example, the findings that loss of either Sema3E or plexin D1 causes a similar disorganization of the intersomitic vasculature [76–78] whereas genetic ablation of NRP1 has no such effect, are explained by generation of repulsive cues to endothelial cells as a result of Sema3E binding directly to plexin D1 independently of NRP1. Interestingly, Sema6D, which does not bind NRPs, utilizes a complex between a VEGFR2 co-receptor and plexin A1 to mediate signalling essential for cardiac development .
These findings indicate that, in key respects, the cardiovascular and neuronal guidance functions of NRP1 are not reliant on mutual dependence or cross-talk between the development of these networks, mediated by ligands for a shared receptor, but result from spatially distinct and divergent functions of NRP1 expressed in the vasculature or neurons. However, there is some co-operation between Sema3A and VEGF binding in cardiovascular development because knockin NRP1 mice deficient in Sema3A binding and also null for NRP2 (see Table 1), mice in which VEGF is only able to bind to NRP1, exhibit cardiovascular defects similar to those in endothelial-specific NRP1-null mice .
NRP expression may also play an important role in the early specification of arterial and venous fate during vascular development. In 1-day-old chick embryos, NRP1 and NRP2 are co-expressed in the early extra-embryonic blood islands, but by the 13-somite stage, expression of NRP1 and NRP2 has become restricted to, respectively, the arterial and venous regions of the primary vascular plexus before blood has started to flow . In 26-somite embryos, which have a functioning vasculature, NRP1 and NRP2 are differentially expressed in arteries and veins, whereas the NRP2 ligand, Sema3F, was bound to NRP2-expressing cells [80,81]. Similar to the expression patterns in the chicken embryo, NRP1 is preferentially expressed in the dorsal aorta of zebrafish embryos, and NRP2 transcripts are localized to the posterior cardinal vein . It is unclear at present whether differential NRP expression is crucial for the early embryonic segregation of arterial and venous cells in mammalian embryos, although the relatively mild phenotype in NRP2-deficient mice (Table 1), suggests that NRP2 expression may be less important in this respect.
The most characteristic biological function mediated by NRPs in neuronal cells is chemorepulsion. NRP1 is essential for mediating Sema3A stimulation of growth-cone collapse and axon repulsion in DRG neuronal cultures [31,83], whereas NRP2 is responsible for Sema3F-induced repulsion of superior cervical ganglia [15,84]. Sema3A and 3F have also been reported to promote chemorepulsion in PAE (porcine aortic endothelial) cells expressing NRP1 and NRP2 respectively, [85,86], and in human endothelial cells [87–89], although other investigators failed to find an inhibitory effect of Sema3A on endothelial-cell migration . Sema3A also appears to be required for the correct orientation and chemoattraction of cortical apical dendrites . This chemoattractant role of Sema3A is mediated at least in part via NRP1, as judged by the effects of function-blocking NRP1 antibody. Remarkably, the conversion of the Sema3A/NRP1 axis from chemorepulsion into chemoattraction is determined by the distribution of soluble guanylate cyclase: guanylate cyclase is localized to the dendrite, and inhibition of either guanylate cyclase or protein kinase G disrupted dendrite outgrowth in response to Sema3A [90,91]. Most evidence points to a role of NRP1 in endothelial-cell migration and adhesion. Co-expression of NRP1 and VEGFR2 in PAE cells enhances VEGF binding and chemotaxis , and NRP1 mediates endothelial cell attachment to the extracellular matrix . VEGF-A165 also stimulates morphogenetic responses in renal epithelial cells, including sheet migration and tubulogenesis, through a mechanism dependent on VEGFR2 and which is also blocked by either neutralizing antibody to NRP1 or by the chemorepulsive NRP1 ligand, Sema3A . Antibodies which specifically block VEGF binding to NRP1, inhibited the migratory response to VEGF, in vitro endothelial-cell sprouting and neovascularization in vivo . Interestingly, the same study reported that blocking NRP1 antibodies prevent pericyte recruitment to new vessels in mouse neonatal retinal vascularization and tumour vascularization models, suggesting a role of endothelial NRP1 in the maturation and stabilization of developing vessels. The cellular mechanism underlying such a role of NRP1 is unclear, but could occur via one or both of two mechanisms, one in which endothelial NRP1 is important for adhesion or chemoattraction of pericytes to developing vessels, and/or a role for pericyte NRP1 in adhesion and migration of immature VSMCs . The fact that NRP1 is expressed in some mature VSMC cell types suggests a role for this molecule in migratory and adhesive functions of these cells . Proteoglycans play important roles in VSMCs in vivo, and it is tempting to speculate that the extensive and preferential CS-GAG modification of NRP1 in VSMCs may play a role in migratory functions of this molecule in developing or remodelling arteries. It is important to note, however, that there is a lack of information regarding expression of NRPs in arterial VSMCs in vivo either during development or in adult vessels, and determination of a biological role of NRP1 in VSMCs will await analysis of appropriate animal models.
Analysis of vascularization in the developing hindbrain of NRP1−/− mice shows that this molecule plays a key role in the guidance of specialized endothelial tip cells in newly sprouting vessels. Tip cells extend long filopodia, which sense gradients of VEGF-A, and undergo a stereotypic series of movements during development. In NRP1−/− mice, tip cell filopodia remain associated with radial glia in the subventricular zone of the hindbrain and fail to move laterally across this region, forming characteristic tufts . These findings indicate that NRP1 may not be essential for endothelial cell migration, or for elaboration of the cellular migratory apparatus, but rather for determining the trajectories of migrating cells, similar to its pathfinding and homing role in neuronal patterning.
Several studies indicating a role for NRP1 in cell adhesion to extracellular matrix, are suggestive that such a role of NRP1 may be independent of known ligand interactions [92,95]. Shimuzu et al.  found that recombinant NRP1 proteins supported adhesion of a variety of cell lines, including L cells, HEK (human embryonic kidney)-293T, COS-7, HeLa, p19, KB and NIH 3T3, and identified specific regions in the NRP1 b1 and b2 domains for adhesion, but showed that neither Sema3A nor VEGF-A165 interfered with this activity . Furthermore, siRNA (small interfering RNA)-mediated NRP1-knockdown disrupted endothelial-cell adhesion to fibronectin, laminin or gelatin, whereas silencing of VEGFR2 had little effect, suggesting that an NRP1-mediated adhesive function is independent of VEGF-A signalling through VEGFR2 . Blocking NRP1 antibodies appeared to have no effect on adhesion of endothelial cells to fibronectin , consistent perhaps with the independence of any adhesive function from VEGF binding.
RECEPTORS AND SIGNALLING MECHANISMS
NRPs are not thought to be able to transduce a biological signal or response in the absence of another signalling receptor, but to function as co-receptors that bind extracellular ligands with high affinity and complex with other transmembrane molecules (e.g. VEGFR2 or plexins) to form a holoreceptor. However, this model does not fully explain all the evidence relating to the functions of NRPs in endothelial cells and other cell types. Furthermore, NRP1 has been demonstrated to associate with at least one intracellular PDZ protein, synectin, raising the possibility that NRPs may be able to modulate intracellular signalling through protein–protein interactions, despite possessing a small cytoplasmic domain with no defined signalling role.
NRP1 mediates the chemorepulsive effects of Sema3A, but does so by acting as the ligand-binding module of a complex or holoreceptor between NRP1 and specific members of a family of transmembrane receptors called plexins [96–98]. NRP1 and 2 form complexes with several plexins independently of the presence of semaphorins, including plexins A1, A2, A3 and B1 [97–99]. The human plexin gene family comprises nine members divided into four subfamilies: four plexin As (1–4), three plexin Bs (1–3), plexin C1 and plexin D1 . Plexins are large transmembrane receptors with extracellular regions comprising a sema domain (homologous with sema domains in semaphorins), two or three plexin cysteine-rich regions called MRS (Met-related sequences) domains also found in the MET oncogene family, and three or four immunoglobulin-like domains shared by plexins and transcription factors (termed IPT), a transmembrane domain, and an intracellular region containing two conserved regions, the functions of which are not yet clearly defined, but show some homology with ras GAP (GTPase-activating protein) domains, and which, in the case of plexin B1, has been shown to act as a GAP for R-Ras .
Sema3A contains both a NRP-binding site, and potential sites of interaction with plexins. Antipenko et al.  have proposed a model based on the crystal structure of the Sema3A sema domain and mutagenic analysis, in which Sema3A-binding results in a 2:2:2 complex between Sema3A, plexin A1 and NRP1 (see Figure 4), and involving relief of autoinhibition of plexin A1 . This has been proposed to cause activation of the plexin intracellular domain and the transduction of signals essential for chemorepulsion. Deletion of the cytoplasmic domain of NRP1 did not impair Sema3A signalling , indicating that NRPs may function solely as the binding entity of the complex, whereas plexins mediate signalling.
The cytoplasmic domain of plexins is responsible for downstream signalling induced by semaphorin and resulting in the collapse of neurons [96–98]. Monomeric GTPases of the Rho family are thought to play a key role in regulating actin-based motility in neuronal cells , and have been implicated in semaphorin-mediated growth-cone collapse [102,103]. Recruitment of the small GTPase Rnd to the cytoplasmic plexin A1 domain triggers cytoskeletal collapse, but this effect is antagonized by RhoD, which is also able to bind to plexin A1, but blocks repulsion of sympathetic neurons  (see Figure 5). Sema3A-induced cytoskeletal collapse in Cos-7 cells was dependent on co-transfection of plexin A1 and NRP1, required activation of Rac, but not Rho, and involved a direct interaction between Rac and the plexin A1 cytoplasmic domain [105–107]. If and how NRP1 regulates plexin-dependent signalling through small GTPases is unclear.
CAM (cell-adhesion molecule) L1
The immunoglobulin superfamily CAM L1 is also a potential partner for NRP1 in mediating chemorepulsive Sema3A signals (Figure 5). L1-deficient mice are smaller, less sensitive to touch and pain, exhibit lack of hindlimb co-ordination compared with littermate controls and display a striking reduction in the size of the corticospinal tract and in the association of Schwann cells with axons . That at least part of the phenotype in these mice results from defective axonal guidance orchestrated by Sema3A is suggested by the inability of Sema3A to repel L1-deficient cortical axons  and the formation of stable complexes of L1 and NRP1 mediated by interactions between their extracellular domains . NRP2, which is not required for Sema3A signalling, was unable to associate with L1-CAM . Several L1 mutations located in the NRP1-binding region are associated with some human neurological disorders, including X-linked hydrocephalus and MASA (mental retardation, aphasia, shuffling gait, adduced thumbs) syndrome. Interestingly one such mutation, L120V, also disrupts the L1 association with NRP1, suggesting a causal role of defective L1/NRP1 signalling in human neurological disease . The soluble L1 extracellular domain is able to convert repulsive Sema3A signals into chemoattraction by binding in-trans to NRP1 and this conversion is mediated by nitric-oxide-dependent activation of guanylate cyclase and consequent cyclic GMP synthesis , reminiscent of the role of asymmetric guanylate cyclase distribution in determining chemoattractant properties of Sema3A in cortical apical dendrites . The effects of L1 deficiency do not phenocopy the neural defects of Sema3A-deficient mice or NRP1sema3A−/− mice expressing mutant NRP1 unable to bind Sema3A, indicating that L1 regulates Sema3A signalling via NRP1 in a more restricted set of neurons. It is also unclear yet whether the L1–NRP1 complex is either distinct from, or linked with, the plexin/NRP1 holoreceptor.
In endothelial cells, NRPs act as co-receptors for VEGFs by forming complexes with the VEGF protein tyrosine kinase receptor VEGFR2/KDR. Soker et al.  identified NRP1 as a high-affinity receptor for VEGF-A in endothelial and tumour cells and found that NRP1 co-expression with VEGFR2 enhanced VEGF-induced chemotaxis in comparison with cells expressing only VEGFR2 . Co-expression of NRP1 with VEGFR2 also enhances VEGF binding to VEGFR2, VEGFR2 phosphorylation and VEGF-induced signalling and migration [111–113]. Whether or not NRP1 increases the affinity of VEGFR2/KDR for VEGF-A remains uncertain. However, the enhanced function of VEGFR2 in the presence of NRP1 has been attributed to enhanced VEGFR2 signalling when it complexes with NRP1, rather than to an increase in the intrinsic affinity of VEGF for its receptors . There are also differing reports as to whether or not VEGF binding is required for NRP1–VEGFR2 complexation. Two groups showed that the association between NRP1 and VEGFR2 was dependent on VEGF-A treatment [11,112], whereas Whitaker et al.  found that co-immunopreciptation of VEGFR2 and NRP1 in Cos-1 cells co-expressing the two receptors and in HUVECs (human umbilical-vein endothelial cells) occurred independently of VEGF-A.
Although the precise mechanism mediating the modulation of VEGFR2 signalling by NRP1 is unclear, complex formation does appear to play a crucial role. Complexation between NRP1 and VEGFR2 enhances VEGF binding, and inhibition of complex formation is associated with reduced VEGFR2 phosphorylation, intracellular signalling, mitogenesis, cell migration and angiogenesis [11,45,111,112,114,115]. Since VEGF stimulates biological activities in PAE cells expressing only VEGFR2 , signalling via VEGFR2 can be triggered by VEGF-A independently of NRP1. Furthermore, selective inhibition of VEGF-A binding to NRP1 and NRP1–VEGFR2 complex formation does not prevent either VEGFR2 signal transduction or the stimulation of endothelial biological responses by VEGF-A [11,45]. Thus inhibition of VEGF binding to NRP1 using a specific peptide antagonist (EG3287) that mimics the VEGF NRP1-binding domain, attenuated rather than inhibited VEGF-induced VEGFR2 tyrosine phosphorylation, and activation of ERK (extracellular-signal-regulated kinase) and phospholipase Cγ, and reduced prostacyclin production, but had less effect on Akt activation and did not reduce the cell survival or proliferative responses to VEGF . Antibodies that specifically block VEGF binding to NRP1 prevented NRP1 complexation with VEGFR2, but had limited effects on VEGFR2 phosphorylation and signalling . It seems therefore that NRP1 is not essential for the full spectrum of signalling pathways and biological responses stimulated by VEGF, but is instead required for optimal VEGF-induced VEGFR2 signalling through some pathways, and for maximum function in the case of certain biological processes, particularly migration.
Several mechanisms could theoretically account for the ability of NRP1 to enhance VEGFR2 signalling. NRP1 could increase the affinity of VEGFR2 for VEGF-A, although previous studies have concluded that NRP1–VEGFR2 complexation does not change the affinity of VEGF for VEGFR2 . However, selective antagonism of VEGF binding to NRP1 also greatly reduced VEGF-A cross-linking to KDR in human endothelial cells co-expressing the two receptors, although the antagonist had no effect on VEGF-A binding to PAE cells expressing only KDR , suggesting that NRP1–VEGFR2 complex formation may stabilize binding of VEGF-A to VEGFR2, therefore increasing the longevity of VEGFR2 signalling. A second mechanism is that NRP1–VEGFR2 complex formation may stabilize VEGFR2 at the cell surface, by rendering it less labile and/or less prone to receptor-mediated endocytosis and therefore able to increase the duration and amplitude of receptor activation and downstream signalling. In support of this notion, siRNA-mediated NPR1-knockdown was recently reported to both reduce KDR expression and attenuate VEGF-induced gene expression in human endothelial cells . Thirdly, as discussed in more detail below, NRP1 may itself either transduce intracellular signals, or participate in protein–protein interactions, which enhance the signalling function of the NRP1–VEGFR2 complex compared with VEGFR2 alone.
In contrast with the well-defined signalling role of VEGFR2 in VEGF angiogenic activity, the function of VEGFR1 has not been fully characterized. The VEGFR1-knockout is lethal , but a normal phenotype is observed in transgenic mice with Flt-1 lacking the intracellular domain . These findings indicate that VEGFR1 plays a decoy function by sequestering VEGF and therefore limiting its availability for binding to VEGFR2. NRP1 has been found to bind in vitro to the VEGFR1/Flt-1 extracellular immunoglobulin-like domains 3 and 4 , and it is possible that such a complex may compete for VEGF binding to VEGFR2–NRP1 heterodimers.
Recent findings indicate that VEGF-C interacts with NRP2 in a mainly heparin-independent manner, whereas VEGF-D binding to NRP2 is heparin-dependent . Moreover, VEGFR3 and NRP2 displayed co-localization and co-internalization following stimulation by VEGF-C and -D . The formation of complexes between NRP2 and VEGFR2 or VEGFR3 resulted in a lowering of the activation threshold of VEGFR2 and an enhancement of cell survival and migration induced by VEGF-A and the VEGFR3 ligand VEGF-C . VEGFR3 and VEGF-C are both strongly implicated in lymphatic vascular development , and complex formation between VEGFR3 and NRP2 may help to explain the involvement of NRP2 in lymphangiogenesis as suggested by the phenotype of NRP2-knockout mice ( and Table 1). Co-immunoprecipitation of cross-linked 125I-VEGF also revealed the existence of a NRP2–VEGFR1 complex , but the biological relevance of such a complex is not clear.
Although there is strong evidence that NRP1 functions primarily as a co-receptor without an independent signalling role, some findings suggest that NRPs have functions that are not dependent on the known NRP ligands and interacting receptors, and further raise the possibility that NRP1 is able to support functional cellular signalling. For example, VEGF elicits biological responses in some cell types that are NRP-positive but express little or no KDR, such as VSMCs  and some cancer cells (see below). As discussed above, antibodies and antagonists that selectively block VEGF binding to NRP1 have restricted effects on endothelial VEGF signalling and in in vivo models of angiogenesis, which do not replicate the effects of function-blocking antibodies directed against VEGF [11,45], suggesting that NRP1 has VEGF-independent roles in endothelial cells. Although the cytoplasmic domains of NRP1 and NRP2 are small, they do contain a C-terminal consensus PDZ-binding motif that associated with the PDZ protein NIP1 (also called synectin or GIPC1) in a yeast two-hybrid screen . A functional role for the association of NRP1 with synectin in angiogenesis is supported by the finding that expression of NRP1 lacking the C-terminal SEA motif disrupted vessel formation in zebrafish, and that knockdown of either synectin or NRP1 in zebrafish produced similar vascular phenotypes . Furthermore, synectin was found to associate with NRP1 in human endothelial cells, and synectin knockdown inhibited NRP1-mediated endothelial migration . In contrast, the C-terminal PDZ-binding domain does not appear to be important for Sema3A-mediated neuronal pathfinding functions of NRPs .
Synectin/GIPC was originally found to associate with one of the regulators of G-protein signalling (RGS) proteins, called RGS19 (also known as GAIP) . The GIPC–RGS19 complex is anchored to the cell membrane, localizes to clathrin-coated vesicles and has been implicated in endocytosis and intracellular membrane trafficking . Synectin/GIPC binds to G-protein-coupled receptors and modulates their signalling , and can also interact with up to 20 other proteins, including the proteoglycan syndecan-4 , integrin α5 and α6 subunits , the transmembrane semaphorin M-SemaF  and rho GEF (rho guanine-exchange-factor or syx1) . These findings suggest that synectin has the potential to participate in multimeric protein complexes or scaffolds able to link surface receptors and integrins with intracellular signalling networks. In vivo studies show the importance of synectin in the development of a functional vascular system in both the zebrafish and the mouse . Synectin-deficient mice are viable, but the mice are smaller (30% less than littermate controls), and the number of small arteries were significantly reduced, resulting in impaired vascular functions . Arterial cells from synectin−/− mice exhibited reduced angiogenic and endothelial responses in cell culture models, and aberrant cellular distribution of rac1 . The mainly microvascular arterial defects in synectin-knockout mice do not phenocopy the embryonic lethality or aberrant macrovascular development of either global or endothelial-specific NRP1 deficiency, indicating that not all developmental functions of NRP1 are dependent on synectin association. However, it is possible that other related molecules, such as GIPC2 or GIPC3, may compensate for loss of synectin . Nevertheless, the fact that synectin binds to the C-terminus of NRP1 raises the possibility that NRP1 participates in independent cytoplasmic signalling networks and suggests the existence of other as yet unidentified NRP1 interacting proteins.
NEUROPILIN FUNCTIONS IN DISEASE AND ADULT TISSUES
NRP1 and NRP2 are expressed by a wide variety of human tumour cell lines and diverse human neoplasms [8,42,133] and are implicated in mediating the effects of VEGF and semaphorins on the proliferation, survival and migration of cancer cells [134–136]. Table 2 summarizes the protein expression of NRPs and other VEGF receptors in a panel of representative carcinoma cells obtained from published work and our recent studies (H. Jia, A. Bagherzadeh, L. Cheng, P. Frankel and I. Zachary, unpublished work). Overexpression of NRP1 in Dunning rat prostate AT2.1 carcinoma cells increased tumour growth in vivo , whereas NRP1-knockdown using siRNA inhibited breast carcinoma cell migration , and a peptide targeted to the VEGF-binding site of NRP1 induced breast tumour cell apoptosis . NRP1 is expressed in patient specimens from lung, breast, prostate, pancreatic and colon carcinomas, but not in corresponding normal epithelial tissues [23,139–144]. NRP1 has also been found in several other tumours, including melanoma , astrocytoma  and neuroblastoma . NRP2 expression was reported in lung cancer [139,140], neuroblastoma , pancreatic cancer , osteosarcoma  and bladder cancer . It has been suggested that NRP1 is more prevalently expressed in carcinomas (mainly of epithelial origin), whereas NRP2 may be more frequently expressed in non-carcinoma neoplasms such as melanomas, leukaemias and neuroblastomas [8,133,151]. However, as Table 2 indicates, there is no sharp distinction between the types of neoplasms expressing NRPs 1 and 2 , and often they are co-expressed. Furthermore, different cell lines derived from the same tumour types, such as glioma , may exhibit divergent patterns of NRP1 and NRP2 expression.
Clinical studies suggest that NRP1 plays a role in tumour growth and disease progression [8,10,133]. Overexpression of NRPl has been demonstrated to be positively associated with the metastatic potential, advanced stage and clinical grade of prostate carcinoma . NRP1 up-regulation in gastrointestinal carcinomas appears to correlate with invasive behaviour and metastatic potential . Co-expression of NRP1 and NRP2 also increased in the progression from dysplasia to microinvasive lung carcinoma, and correlated significantly with tumour progression and poor prognosis in patients with non-small-cell lung carcinoma . NRP1 also appears to be preferentially expressed in metastatic cells, and is found, for example in the metastatic breast cancer MDA-MB-231 and melanoma MDA-MB-435 cell lines but not in the non-metastatic cell line MDA-MB-453 or some non-metastatic tumours [42,134].
Although most studies have indicated a pro-tumorigenic role of NRPs, some reports suggest that NRP1 plays a more complex role in some tumour types. Thus NRP1 over-expression in Panc-1 cells was found to reduce tumour volume and incidence , whereas the same group showed that FG pancreatic carcinoma cells expressing NRP1 have increased resistance to anoikis and cytotoxic drugs . Furthermore, some findings suggest that not all effects of VEGF in tumour cells may be dependent on NRP1. Lee et al.  found that survival effects of VEGF in the breast carcinoma MDA-MB-231 and MCF-7 cell lines were mediated by internally expressed VEGFR1/Flt1, and were unaffected by NRP1-knockdown.
There is also some evidence pointing to potentially differential or antagonistic roles of NRP1 and NRP2 in tumour-cell regulation. Sema3F, the best-characterized ligand for NRP2, induces a poorly vascularized non-metastatic phenotype in melanoma xenografts in mice  and, suggestively, Sema3F and a second NRP2 ligand, Sema3B, are both localized to the 3p21.3 chromosomal region, which is commonly deleted in human lung cancers [158,159]. These findings suggest that Sema ligands for NRP2 are potential tumour suppressors. Confirmation of such an anti-tumorigenic role of NRP2 awaits further work.
Recent studies provide direct evidence that NRP1 contributes to tumour cell growth and tumour neovascularization in vivo. A peptide that inhibits VEGF binding to NRP1 has been reported to inhibit angiogenesis and growth of tumour xenografts . An antibody targeted to the b1 domain that specifically blocks VEGF-A binding to NRP1 causes a range of effects in endothelial cell cultures, including inhibition of VEGFR2 complex formation, VEGF-induced migration and vascular sprouting, reduces angiogenesis in a neonatal retinal neovascularization model and inhibits tumour growth and tumour vascularization in mouse xenograft models. A blocking NRP1 antibody alone had a relatively small effect on tumour growth, but produced a strong additive effect when used in combination with the blocking VEGF antibody, bevacizumab or Avastin , now approved for clinical use in patients with late-stage colorectal carcinoma. Interestingly, Pan et al.  reported that the anti-tumour activity of blocking NRP1 antibodies was not dependent on NRP1 expression in the tumour cell line used in the xenograft model. Furthermore, this study also found no evidence of direct effects of NRP1 antibodies on tumour cell proliferation, suggesting that their anti-tumorigenic effects were mainly due to inhibition or destabilization of the tumour vasculature. However, any conclusion that the role of NRP1 in tumorigenesis is restricted to vascularization should be tempered in the light of studies, discussed above, implicating NRP1 in cell migration and adhesion. Furthermore, NRP1 plays a role in potentiating the effect of HGF (hepatocyte growth factor)/Scatter factor signalling through the c-Met receptor in both glioma and pancreatic cancer cell lines, regulating tumour progression and invasion [161,162]. These findings taken together with the expression of NRPs in diverse neoplasms, suggests a possible role for this molecule in tumour invasion and metastasis in addition to its involvement in tumour vascularization, and future work should explore this possibility.
NRP1 was found to be expressed in naive T-cells and immature dendritic cells (antigen-presenting cells), cell types that interact during the primary immune response in the secondary lymphoid organs , and are essential for triggering the proliferation and differentiation of mature T-cells that will later interact again with antigen-presenting cells to mediate antigen elimination . NRP1 expressed on naïve T-cells also mediated their clustering with NRP1-expressing Cos-7 cells, and the stimulation of resting T-cell proliferation by dendritic cells was reduced by approx. 50% by blocking NRP1 antibodies . T-cell activation was also reduced by Sema3A [163,164]. NRP1 distribution in T-cells was polarized, and NRP1 co-localized with the T-cell marker, CD3, at the interface between dendritic cells and naïve T-cells . Although the biological relevance of these findings for the immune response in vivo is unclear, they are consistent with the formation of homophilic NRP1 interactions between dendritic cells and immature T-cells, which contribute to an early step in formation of the immunological synapse, essential for T-cell maturation .
There is evidence that NRP1 is up-regulated in response to tissue injury and may be involved in regeneration and repair. In Xenopus, when the optic nerve was crushed and allowed to regenerate, the level of NRP1 increased and remained elevated for weeks before finally declining after healing . NRP1 was also found to be strongly expressed in the neovasculature during wound angiogenesis in a murine model of dermal wound healing, whereas blocking anti-NRP1 antibody reduced wound vascularization . Optic nerve injury in a rat model resulted in cell invasion at the site of injury by microglia, oligodendrocytes and astrocytes associated with induction of Sema3A and NRP1 . NRP1 may protect neuronal cells against damage resulting from stress or injury. Thus IFNγ (interferon γ) activation of microglia, a cell type that becomes activated following neuronal injury, caused up-regulation of NRP1 and plexin A1 and was associated with induction of microglial apoptosis by Sema3A, suggesting that a NRP1-mediated pathway may protect neurons against damage caused by activated microglia . Expression of NRPs, VEGFs and class 3 semaphorins are also up-regulated by cerebral ischaemic injury in different animal models [168–176]. For example, in a mouse model of cerebral ischaemia, NRP1 mRNA expression increased rapidly and remained elevated for at least a month after the ischaemic event . Interestingly, NRP1 was not only localized to the ischaemic neurons, but also to the endothelial cells of the brain vessels. These findings are suggestive of a dual role for NRP1 in the response to cerebral injury, in the promotion of both neuronal growth and in cerebral angiogenesis.
CONCLUSIONS AND PERSPECTIVES
Neuropilins are receptors for class 3 semaphorins and for members of the VEGF family, with essential roles in neuronal patterning and cardiovascular development respectively. Much evidence points to a key function of NRP1 in the migratory guidance of both axons and endothelial cells, a role that is presumably controlled by gradients of extracellular ligands and the co-expression of signalling receptors, plexins in neuronal cells and VEGFRs in endothelial cells. Although complex formation between NRPs and VEGFR2 or plexins seem to be key to understanding the role of NRPs, the mechanisms underlying NRP regulation of either VEGFR and plexin signalling or cell guidance remain unknown. In this context, associations between the NRP C-terminus and the PDZ domain protein synectin may offer a tantalising clue, though as yet it is still unknown if or how synectin participates in VEGFR and plexin signalling pathways. Further work delineating the role of NRP protein–protein associations, and the identification of novel protein interactions with the NRP intracellular domain is likely to reveal novel insights into the mechanisms mediating NRP functions.
The mechanisms mediating specific functions of NRP1 and NRP2 are also not well understood. The different expression patterns and phenotypes of the respective mutant mice indicate that they play distinct roles in the development of both the neuronal and vascular systems, although they also have the ability to interact. Several basic questions remain to be answered. Although NRP2 is important for guidance of some classes of neuronal axons, the cellular function of NRP2 is much less well studied than for NRP1, and it is unclear what its primary functions are in endothelial, lymphatic or cancer cells. If some ligands with selectivity for NRP2 are also antagonistic towards NRP1, as has been proposed for semaphorins, what are the consequences for regulation of cell function? In cells co-expressing NRP1 and NRP2, how, if at all, do they interact and what is the impact on VEGFR2 and plexin signalling? It is interesting to speculate that in some contexts NRP2 may play a modulatory or inhibitory role in NRP1 functions.
Despite the wealth of information regarding NRP functions from mutant mouse and other animal models (Table 1), several important features of NRPs remain enigmatic. Although the most widely accepted model for NRP function is that NRPs are ligand-binding but non-signalling co-receptors, unable to function independently of their partner receptors in neuronal and endothelial cells, some evidence, particularly from studies in endothelial, cancer and T-cells, does not readily fit this mechanistic paradigm, and suggests additional modes of NRP1 function. NRP1 is expressed in some cell types, immature T-cells and VSMCs in particular, which are not typical targets for either semaphorins or VEGFs, and it remains unclear whether functions of NRP1 in some cancer cells are dependent on co-expression of either VEGFRs or plexins. There is evidence that NRP1 may participate in homophilic adhesive interactions between cells in, for example, the formation of early contacts, the so-called ‘immune synapse’, between naïve T-lymphocytes and immature dendritic cells. The details of these cell–cell interactions, and the precise role that NRP1 plays, remain to be elucidated, but they are suggestive that NRP1 plays roles independent of its known complement of ligands and receptor partners, particularly in adhesion. The recent identification of high-molecular-mass NRP1 species modified by O-linked CS-GAGs suggests additional modalities for interactions between NRP1 and extracellular constituents with potential relevance for NRP1 functions in VSMCs and tumour cells. NRPs may also form higher-order structures through homodimerization and oligomerization independent of ligands, raising the possibility of a further way in which these molecules may influence cell signalling and behaviour.
In addition to their major biological roles in neurogenesis and cardiovascular development, recent evidence has identified novel functions for NRPs in other physiological processes, such as T-cell maturation and in disease. An exciting recent development has been the emergence of NRPs as a novel therapeutic target for cancer. Since the role of NRP1 in angiogenesis appears to be distinct from that of VEGF, targeting NRP1 may complement and potentiate the anti-tumour effects of therapies targeted at VEGF, such as bevacizumab. The ability of anti-NRP1 antibodies to synergize with bevacizumab in animal models of tumour growth provides some early proof-of-principle for such a combined approach. NRPs are expressed on many tumour cells and are implicated in cancer cell functions and tumour development, although their precise role in tumorigenesis is unclear. Given the importance of NRPs for axonal and endothelial tip cell guidance, it is tempting to think that they may also play a role in tumour cell adhesion and migration in metastatic cancer spread. If this were so, then therapeutically targeting NRPs may offer a novel approach to the inhibition of both tumour vascularization and metastasis.
C. P.-M. and I. Z. are supported by grant funding from the British Heart Foundation and P. F. and H. J. are supported by funding from Ark Therapeutics Plc.
↵1 Ian Zachary, Paul Frankel and Haiyan Jia have an interest in Ark Therapeutics Plc, which is developing therapies targeted at inhibition of neuropilin.
Abbreviations: BMP1, bone morphogenetic protein 1; CAM, cell-adhesion molecule; CNS, central nervous system; CS, chondroitin sulfate; CUB, complement binding factors C1s/C1r, Uegf, BMP1; DRG, dorsal root ganglia; ERK, extracellular-signal-regulated kinase; GAG, glycosaminoglycan; GAP, GTPase-activating protein; GIPC, RGS-GAIP-interacting protein; IPT, immunoglobulin-like domains shared by plexins and transcription factor; KDR, kinase insert domain-containing receptor; LMC, lateral motor column; MAM, meprin, A5 antigen, receptor tyrosine phosphatase μ; HUVEC, human umbilical-vein endothelial cell; MRS, Met-related sequences; NIP1, neuropilin-interacting protein-1; NRP, neuropilin; PAE, porcine aortic endothelial; PDGF, platelet-derived growth factor; PDZ, post synaptic density protein (PSD95), Drosophila disc large tumor suppressor (DlgA), and zonula occludens-1 protein; PlGF, placental growth factor; PNS, peripheral nervous system; siRNA, small interfering RNA; VEGF, vascular endothelial growth factor; VEGFR, VEGF receptor; VSMC, vascular smooth muscle cell
- © The Authors Journal compilation © 2008 Biochemical Society