Neuropilins were discovered in the neuronal system, and the first ligands identified for these receptors were the class 3 semaphorin (SEMA3) family of axonal guidance proteins [21, 33, 34]. Later NRPs were shown to be a unique class of receptors for the VEGF family of angiogenic molecules [35]. This link between the neuronal and vascular systems opened up an entirely new field comparing the molecular similarities in these two distinct yet similarly patterned systems [36]. Today it is now appreciated that NRPs can bind and mediate signals through many ligands including platelet-derived growth factor (PDGF), HGF, and TGFβ (reviewed by [37–39]). We will summarize each ligand family and its relation to NRP function below with particular interest in those processes involved in cancer progression such as angiogenesis, migration, invasion, and metastasis.

VEGF is fundamental for the development and survival of vascular supply, and therefore VEGF signaling is pivotal for physiological functions and homeostasis by maintaining a functional network of blood vessels [40]. VEGF also plays an essential role in inducing vascular permeability, which is important for angiogenesis associated with tumors and wound healing [41]. NRPs serve as receptors for growth factors in the VEGF family with differential binding affinities for each ligand. The VEGF binding sites reside in the b1 and b2 domains on the extracellular component of the NRP receptors [13, 14, 16, 42, 43]. Therefore, both membranous and secreted NRPs can bind VEGF, as they all contain intact b1/b2 domains.

The human VEGF-A gene encodes multiple isoforms that arise due to alternative splicing. The VEGF-A isoforms differ in their inclusion of exons 6 and 7, which are the domains responsible for heparin binding [44]. VEGF-A₁₂₁ is a slightly acidic form that cannot bind heparin, while VEGF-A₁₆₅, VEGF-A₁₈₉, and VEGF-A₂₀₆ all contain exon 7 and bind heparin. NRPs were initially described as isoform-specific receptors for only heparin-binding isoforms of VEGF-A, in particular, VEGF-A₁₆₅ [35, 45]. Later, crystal structures of NRPs indicated a VEGF-A exon 8a binding site [13, 46, 47]. Furthermore, a newly identified VEGF-A variant called VEGF-A_165b, which lacks exon 8a but includes exons 7 and 8b, did not bind to NRP1 [48]. Taken together, these data suggest that the b1 domain in NRP1 or NRP2 can bind VEGF-A via regions encoded by exon 7 and 8 [49]. The VEGF-A₁₆₅ ligand can bind to both NRP1 or NRP2, but recent evidence shows that binding is 50 times stronger to NRP1 than NRP2 [49]—this may explain why the initial expression cloning of the putative VEGF-A₁₆₅ receptor (now known as NRP) yielded six clones of NRP1 and only one clone for NRP2 [35].

VEGF-A isoforms bind to the canonical tyrosine kinase VEGF receptors, VEGFR1 or VEGFR2, via regions encoded by exon 4 (reviewed by [50]). Thereby, the VEGF-A₁₆₅ protein can “bridge” between NRP receptors (via exon 7–8) and VEGFRs (via exon 4) at the same time forming a complex that enhances the output of the receptor tyrosine kinase (RTK) [35, 45]. As described above, the NRP cytoplasmic domain is small (40 aa) and does not contain kinase activity; therefore, NRPs are often referred to as “co-receptors” since the business end of the signaling is performed through VEGF RTKs. Both VEGFR1 and VEGFR2 have been shown to complex with NRP1 and NRP2 in the presence of VEGF-A [45, 51].

Nearly all members of the greater VEGF family, which share NH₂-terminal cystine knot domains, have been shown to interact with either NRP1 or NRP2. Specifically, two family members that most closely resemble VEGF-A in sequence and structure are VEGF-B and PGF. VEGF-B₁₆₇ and PGF2 both contain basic COOH-terminal domains and sequences homologous to exon 7 of VEGF-A, and both have been reported to bind NRP1 [16, 52, 53] and signal through VEGFR1 [54].

Alternately, the lymphangiogenic members of the VEGF family, namely, VEGF-C and VEGF-D, primarily bind to NRP2 via its b1b2 domain (although they can also bind to NRP1) and transmit their signal through the VEGFR3 RTK [17, 55, 56]. The VEGF-C proprotein (uncleaved version) binds better to NRP2 than does the amino-cleaved version of VEGF-C. VEGF-C can bind to NRP2 in the absence of heparin, but only interacts with NRP1 in the presence of heparin [17]. VEGF-D protein requires heparin to bind to either NRP [17]. The presence of NRP2 enhances the phosphorylation of VEGFR2 in the presence of VEGF-A or VEGF-C [55].

Other ligands that are related to the VEGF superfamily include PDGF and TGFβ. Both of these proteins are capable of binding heparin and have cystine knot and beta-strand topology [57]. Although the precise binding sites on these ligands and NRP domains have not been mapped, it is clear that NRPs can form co-receptor complexes with the PDGF receptors (PDGFR) and TGFβ receptors (TGFβR). Knockdown of NRP1 in mesenchymal stem cells or vascular smooth muscle cells attenuated PDGFRα phosphorylation by PDGF-AA and/or reduced PDGFRβ activation by PDGF-BB [58, 59]. Additionally, NRP1 O-linked glycosylation at Ser612 regulated PDGF-induced smooth muscle cell migration [59].

NRP1 and NRP2 are able to bind to both the active and latent forms of TGFβ1 [19, 60]. TGFβ1 can compete with VEGF-A binding for NRP, suggesting that TGFβ also binds to the b1b2 domain of NRP. Moreover, Nrp1 expression can activate latent TGFβ1 [19]. NRP1/2 were shown to form co-receptor complexes with TGFβRI and RII. TGFβ is also a major inducer of the epithelial-to-mesenchymal (EMT) phenotype in cancer cells, and TGFβ has been shown to induce NRP2 expression in hepatocellular carcinoma cells [61].

NRP1 can bind to another heparan sulfate binding protein called hepatocyte growth factor (HGF) or scatter factor [62]. Active HGF is a heterodimeric, disulfide-linked protein composed of an alpha chain and a smaller beta chain [63]. HGF is secreted as a single-chain inactive protein and then cleaved by serine proteases to the active form. The N-terminal hairpin loop region of the HGF protein structure is strikingly similar to the heparin-binding C-terminal region of VEGF-A [64]. HGF binds and signals through the c-Met receptor tyrosine kinase. Classically, HGF stimulates mitosis in hepatocytes, but HGF also has potent angiogenic and lymphangiogenic properties in endothelial cells [65–67].

NRP1 and NRP2 both bind HGF, and HGF can compete VEGF-A binding to NRPs suggesting that HGF may bind to the b1b2 domain on NRPs [18]. NRP1 has been shown to act as a co-receptor for HGF with c-Met, and HGF likely binds to these two receptors independently or as a bridge via different domains since knockdown of NRP1 or NRP2 could not inhibit all signaling through c-Met [18]. sNRP1 also binds and sequesters HGF [27]. Since NRP1 is highly expressed in sinusoidal endothelial cells in the liver and the liver hepatocytes are a prime source of HGF, one may speculate that NRP1 in the liver is primarily an HGF receptor rather than a VEGF receptor. After hepatectomy, sNrp1 levels plummet in the liver and only increase after regeneration suggesting that this endogenous soluble receptor may regulate the bioavailability of ligands during the healing process [27].

The semaphorin (SEMA) family is a large group of proteins (more than 20 vertebrate members) which share a common structure at the amino terminus called the “sema” domain that is folded into a β-propeller structure with seven blades homologous to alpha integrins (reviewed by [68]). The SEMA family is divided into eight classes (numbered 1–7, and a viral group) containing proteins that span the plasma membrane with the exception of the vertebrate class 3 SEMA (SEMA3) proteins that are secreted. There are seven different SEMA3 secreted proteins labeled A to G that share a cysteine-rich PSI (plexin-semaphorin-integrin) domain, an immunoglobulin domain, and a C-terminal basic-charged domain (in addition to the sema domain) (reviewed by [69]). SEMA3s were originally named “collapsins” for their ability to collapse axonal growth cones [70, 71]. Subsequently, NRPs were found to be high-affinity receptors for these secreted SEMA3 mediators of neuronal guidance [21, 33, 34]. All members of the SEMA3 subfamily bind to NRPs except SEMA3E, which binds directly to Plexin D1 [72, 73]. Although NRP1 and NRP2 extracellular domains are quite homologous, there is specificity within the SEMA3 family for binding to either receptor. SEMA3A binds specifically to NRP1 [33, 34], and full-length SEMA3G binds specifically to NRP2 [74], whereas SEMA3B, C, D, and F can bind to either NRP (reviewed by [75]). In the case of SEMA3F, its affinity to NRP2 is tenfold higher than to NRP1 [21], and SEMA3F appears to only function through its interaction with NRP2 [76], suggesting that NRP1 may be a lower-affinity decoy for SEMA3F.

The functions of SEMA3 proteins are regulated by proprotein convertases (PPC)—similar to that of other proproteins like MMP, VEGF-C, and TGFβ. All SEMA3 proteins contain conserved PPC recognition sites (KRRXRR) between the sema and Ig domains and in the basic C-terminal domain. The SEMA3 proteins are secreted as 100 kD proproteins that are initially cleaved at their C-terminus to 95 kD proteins and then subsequently cleaved upstream to create two protein fragments of ~65 kD and ~30 kD (Table 14.2). The second cleavage is important for function because some SEMA3 proteins bind to NRPs as dimers, making disulfide bridges at conserved cysteines, while others function as monomers. As an example, the affinity of the SEMA3A dimer (95 kD) is 10,000 times greater for NRP1 than the cleaved monomer (65 kD) [77], and dimerization is necessary for collapsing activity of SEMA3A [78].

Full-length SEMA3G (p100) binds to NRP2, processed SEMA3G (p95) binds strongly to NRP2 and weakly to NRP1, and cleaved SEMA3G (p65) does not bind to either receptor [79]. SEMA3B and SEMA3C (either endogenous or recombinant) found in the conditioned media from tumor cells are cleaved and inactive [80, 81], likely due to the high levels of PPC found in tumor cells [82]. However full-length uncleavable SEMA3B protein engineered with a mutant PPC recognition site binds NRP1 or NRP2 and induces endothelial cell repulsion [80]. Similarly, full-length mutant uncleavable SEMA3C binds NRP2 in lymphatic endothelial cells and induces repulsion [81]. Interestingly, the cleaved SEMA3C protein (65 kD) promoted the survival of lung cancer cells, although whether this action is NRP-dependent remains unclear [81]. SEMA3E proteins have dual functions depending on their PPC cleavage status [83]. In its naturally cleaved form, SEMA3E (p61) binds Plexin D1 and associates with ErbB2 to promote lung metastasis and growth [73, 84]. However, when engineered to an uncleavable form, mutant SEMA3E (p95) activated Plexin D1 and inhibited EC migration and survival [73].

The three-dimensional structure of the SEMA3F protein is important for its proper binding and function. The “sema” domain in SEMA3F binds to the a domain of NRP2, while the basic domain of SEMA3F binds to the b domain of NRP2 [85]; therefore, it may come as no surprise that SEMA3 proteins can compete with VEGF-A and VEGF-C binding to NRPs [14, 55, 86]. Additionally, furin cleavage in the C-terminal basic region of SEMA3F creates a fragment protein called c-furSema that can bind to both NRP1 and NRP2 and inhibit VEGFA binding to these receptors [87, 88]. It is currently unclear whether SEMA3 proteins can also compete with binding of PDGF, TGFβ, or HGF to NRP, but it would seem reasonable. Since growth factor signaling is generally stimulatory for proliferation, survival, or migration and SEMA3 proteins compete with the binding of these factors, then it is understandable why SEMA3 proteins are often termed “inhibitory” proteins.

Independent of their ability to compete with VEGF family members, SEMA3 proteins trigger a chemorepulsive signal through the interaction of NRPs and large transmembrane receptors called plexins (reviewed by [89, 90]). NRPs are the SEMA3 ligand-binding part of the complex, and plexins are the signal-transducing part of the complex. For instance, the activation of Plexin A1 by SEMA3A/NRP1 or SEMA3F/NRP2 leads to the inhibition of the RhoA pathway and the subsequent depolymerization of F-actin filaments [91, 92]. This “collapsing” phenotype thus inhibits cell motility, migration, and invasion.

Although the two NRP receptors were first found in neurons, their tissue expression pattern is now appreciated to be much more widely distributed [4, 5, 38, 93]. In the embryonic neuronal system, neural crest cells express both Nrps in a non-overlapping pattern such that Nrp1-expressing cells give rise to sensory and sympathetic ganglia, while Nrp2-expressing cells primarily give rise to sensory neurons [85, 94]. Melanocytes (and melanoma cells), which are derived from neural crest cells, express high levels of NRP2 [95]. In the embryonic vascular system, NRP1 is expressed in arteries, while NRP2 is found in veins and lymphatic vessels [96–98]. NRP receptors are also involved in the guidance and patterning of blood vessels. Thus, it is not surprising that numerous studies using transgenic and mutant mice have established that any of the following can result in abnormal vascular patterning during development: overexpression of Nrp1 [99], constitutive knockout of Nrp1 [100, 101], heterozygous knockout of Nrp1 and Nrp2 [102], or deletion of the cytoplasmic domain of Nrp1 (Nrp1 ^cyto Δ/Δ) [103]. NRP expression in endothelial cells is plastic and can be affected by hemodynamic flow and ischemia/hypoxic conditions [9, 104, 105]. However, NRPs are dramatically downregulated in endothelial cells after birth when patterning is completed and homeostasis has been established [6]. That being said, both NRP1 and NRP2 can be upregulated in activated adult capillaries during phases of remodeling or angiogenesis [106–109].

Although most published reports on NRPs have focused on their roles in blood vessels or neurons, NRPs are much more strongly expressed in epithelial cells and smooth muscle cells than in endothelial cells or neurons (see Fig. 14.1– Nrp2 ^+/lacz embyro) [5, 29, 110, 111]. Nearly all epithelial cells express NRP1 including the epidermis (skin) [111, 112], mammary gland [25, 113], prostate gland, intestine, pancreas, and podocytes of the kidney [114]. Therefore, it should come as no surprise that the majority of all carcinomas, which are derived from epithelial cells, express NRP1 (discussed in more detail below). Interestingly, NRP1 expression in epithelial cells facilitates Epstein-Barr virus entry and promotes infection [115].

There is some degree of specificity between the expression of NRPs within smooth muscle subtypes such that vascular smooth muscle cells express Nrp1 [59] and visceral smooth muscle including cells from the GI tract and the bladder express predominantly Nrp2 [110]. NRPs are also strongly expressed in immune cells including T cells, macrophages, and dendritic cells [93, 116, 117]. Immunoregulatory CD4+ T cell subsets called Tregs highly express NRP1 [93, 118]. These NRP1+ Tregs may follow the gradient of VEGF in order to infiltrate a tumor. Evidence of this phenotype was seen when Nrp1 was knocked out in CD4⁺Foxp3⁺ Tregs, and tumor infiltration was reduced [119].

As mentioned above, most epithelial cells express NRP1; therefore, most carcinomas express NRP1 including carcinomas of the skin (see Fig. 14.2a) [111, 120], tongue [121], breast [122], colon [123], stomach (gastric) [124], endometrium [125], ovary [126–128], prostate [129, 130], liver [28], pancreas [131, 132], kidney [133], and lung [32, 134]. NRP1 is normally found in differentiated epithelial cells but may be upregulated in basal cells in dysplastic tissues and is therefore an early marker of tumor progression [121]. Soluble NRP1 is an early diagnostic marker for cervical cancer and cervical intraepithelial neoplasia (CIN) [135]. NRP2, on the other hand, is not normally found in cells of epithelial origin (Fig. 14.1) but is reportedly upregulated in later stages of carcinogenesis especially in aggressive or metastatic carcinomas [136–138]. NRP2 is highly expressed in cancer cells of neuronal origin such as melanoma cells (Fig. 14.2b) [106], glioblastoma cells [4], neuroblastoma [139], and medulloblastoma cells [140], as well as in some sarcomas [141]. The topic of neuropilins in cancer has been reviewed previously [4, 75, 142–147]. Herein, we focus on the current theories explaining the function of NRPs in tumor cells.

The current paradigm of NRP1/2 function in endothelial cells is that they act as receptors of VEGF family members (VEGF-A, VEGF-C, VEGF-D) and enhance VEGF receptor (VEGF-R1, VEGF-R2, VEGF-R3) tyrosine kinase signaling [17, 35, 45, 55]. The cytoplasmic portion of the NRP protein is small and does not have intrinsic kinase activity. Therefore, NRPs are ligand-binding proteins in the VEGF/NRP/VEGFR2 receptor complex but not signal-transducing proteins. Based on this information, it is unclear what role NRP is playing in carcinoma cells which typically lack expression of VEGFRs. Four potential scenarios have been suggested.

As outlined above, NRPs are expressed in cancer cells, in tumor-associated endothelial cells that participate in neoplastic angiogenesis, in tumor-associated lymphatic endothelial cells that contribute to lymphangiogenesis and metastasis , and in tumor-infiltrating immune cells. NRP1 may be a potential biomarker in many different carcinomas. In some cases the expression of NRP1 merely increases in the early stages of transformation but is still found at a lower level in the surrounding epithelial cells [111, 123], in other cases the localization of the NRP1 protein may change within the epithelium from differentiated cells to basal cells [121], and in still other tissues, normal epithelial cells may lack NRP1 expression entirely but upregulate expression in carcinoma cells—such is the case in the liver [28]. NRP1 copy number gain (CNG) as analyzed by fluorescence in situ hybridization (FISH) in non-small cell lung cancer biopsies correlated with a worse overall survival (OS) and progression-free survival (PFS) [160]. Beyond carcinomas, NRP1 also serves as a novel biomarker associated with poor prognosis in acute myeloid leukemia patients [161].

In a screen of biomarkers for lung tumors induced by benzo[a]pyrene (coal tar), serum NRP2 was found to be highly expressed [162]. NRP2 has been shown to be a novel biomarker for aggressive human melanoma [95]. Furthermore, NRP2 gene silencing in human melanoma cells decreased tumorigenicity and metastasis in a preclinical trial [163].

Based on the aforementioned studies, NRPs are excellent candidates for targeted therapies in multiple forms of cancer. Potential strategies used to target NRPs have been reviewed previously [164–166]. Most successful have been antibodies targeting the VEGF-binding domain of NRP1 or NRP2 [107, 108]. When used in combination with bevacizumab (anti-VEGFA antibodies), anti-NRP1^B antibodies were shown to inhibit tumor size, tumor angiogenesis, and tumor progression [107]. These antibodies were later humanized and tested in a phase Ib clinical trial with bevacizumab and chemotherapy [167]. Unfortunately, this combination therapy caused a high degree of proteinuria in patients—likely due to the off-target effect of anti-NRP1 on podocytes in the kidney [114]. Antibodies to NRP2 were also shown to be efficacious at blocking lymphangiogenesis and metastasis in preclinical trials [108] and have not been tested to date in human clinical trials.