Peripheral nerve-derived VEGF promotes arterial differentiation via neuropilin 1-mediated positive feedback

Proper vascularization of organs and tissues requires not only angiogenesis and arteriovenous differentiation, but also appropriate patterning of the vascular network. Emerging evidence has suggested that the nervous system may play a role in coordinating these processes (reviewed by Carmeliet, 2003), but precisely how is unclear. Previously, we have found that in embryonic limb skin,arteries but not veins are aligned with peripheral nerves, and their branching pattern follows that of the nerves(Mukouyama et al., 2002). In embryonic mouse mutants lacking such nerves, the primitive capillary plexus forms normally but arterial differentiation fails to occur. Moreover, in mutants where the pattern of nerve branching is disrupted, the arterial branching pattern still follows the trajectory of the disorganized nerves. Thus, in limb skin, the pattern of peripheral nerve branching provides a template that instructively guides the branching of the emerging arterial vascular network.

These observations raised the question of the nature of the signal(s) that control nerve-vessel alignment and arterial differentiation, and their cellular source(s). In culture, VEGF-related factors derived from sensory neurons or Schwann cells can induce arterial differentiation in endothelial cell (EC) precursors (Mukouyama et al.,2002). These studies, however, left unresolved the issue of whether nerve-derived VEGFA is actually required for arterial differentiation in the limb skin in vivo. Furthermore, the lack of an in vitro assay for nerve-vessel alignment made it unclear whether VEGF, or rather some other molecule, is involved in this patterning process.

Two recent studies have provided evidence of a role for VEGFA in promoting arterial differentiation in vivo. In the zebrafish, Vegf is essential for formation of the dorsal aorta, the major midline arterial vessel(Lawson et al., 2002). However, this vessel develops by de novo assembly of angioblasts, rather than by remodeling of a capillary network. Therefore, these data do not address whether Vegf is required to induce arterial differentiation from a pre-existing capillary plexus, the mechanism by which most small arterial vessels form in higher vertebrates. In mice, transgenic overexpression of Vegfa in the heart promoted an increased number of cardiac arterial vessels (Visconti et al.,2002). Although these gain-of-function data suggest that VEGFA can promote or enhance arteriogenesis in vivo, they did not address whether it is actually required for this process during normal development.

The early embryonic lethality of Vegfa mutants has made it difficult to address whether nerve-derived VEGFA is required for nerve-vessel alignment and arterial differentiation in vivo(Carmeliet et al., 1996; Ferrara et al., 1996). To circumvent this problem, we have now used several different lines of transgenic mice expressing Cre recombinase, singly or in combination, to specifically delete Vegfa in the major cell types comprising peripheral nerve. In parallel, we have examined the role of neuropilin 1(NRP1), a co-receptor for VEGF¹⁶⁴ that is specifically expressed in arteries. Our data indicate that nerve-derived VEGFA is indeed required for arterial differentiation in vivo, and further suggest that arteriogenesis may be promoted by an NRP1-mediated positive feedback loop. Surprisingly,nerve-vessel alignment occurs normally in these conditional mutants, despite the defect in arteriogenesis. This suggests that such alignment is mediated either by a residual low level of VEGFA, or by a distinct nerve-derived signal.

The characterization of Vegf floxed mice(Gerber et al., 1999), Nrp1 floxed mice (Gu et al.,2003), Wnt1-Cre mice(Danielian et al., 1998), Isl1-Cre mice (Srinivas et al.,2001; Yang et al.,2001), Tie2-Cre mice(Gu et al., 2003; Kisanuki et al., 2001), HB9-eGFP mice (Wichterle et al.,2002) and R26R mice(Soriano, 1999) have been reported elsewhere. Wnt1-Cre; Isl1-Cre; Vegf^flox/flox,Wnt1-Cre; Vegf^flox/flox or Isl1-Cre;Vegf^flox/flox embryos were produced by mating Wnt1-Cre;Vegf^flox/+ with Isl1-Cre; Vegf^flox/+ mice. Offspring were genotyped by PCR using primers that specifically detect the transgene for Wnt1-Cre or Isl1-Cre and the floxed Vegf allele (see `recombination assay by genomic PCR'). Tie2-Cre;Nrp1^flox/- embryos were produced by mating Nrp1^flox/- (harboring one floxed allele and one Nrp1 null allele) with Tie-2-Cre; Nrp1^flox/+mice.

Cre-mediated recombination was assayed using genomic DNA essentially as described (Inoue et al.,2002). Ten dorsal root ganglia were isolated from one E15.5 embryo and dissociated with Proteinase K (Sigma). Skin from both limbs was isolated from one E15.5 embryo, and dissociated by digestion with type 3 collagenase(Worthington) and dispase (Gibco BRL). The cells were incubated with rat anti-mouse P75 monoclonal antibody (M. Rao and D.J.A., unpublished) and then goat anti-rat IgG magnetic beads (Dynal). The magnet-selected P75⁺cells were dissociated with Proteinase K. Genomic DNA was extracted and analyzed for recombination by genomic PCR. using expand Taq DNA polymerase(Roche) for 45 cycles. The PCR products were fractionated by electrophoresis. The sequences of the PCR primers are: VEGF419F(5′-CCTGGCCCTCAAGTACACCTT-3′) and VEGFc5R2(5′-ACATCTGCTGTGCTGTAGGAAG-3′). For genotyping, genomic DNA was isolated from embryonic tail tissue and the presence of Cre transgenes was detected by PCR using PLATINUM Taq DNA polymerase (GibcoBRL) for 28 cycles. The sequence of the PCR primers were VEGF419F(5′-CCTGGCCCTCAAGTACACCTT-3′) and VEGF567R(5′-TCCGTACGACGCATTTCTAG-3′) for floxed Vegf allele;Cre5′a (5′-ACCTTCCTCTTCTTCTTGGG-3′) and Wnt1s(5′-TAAGAGGCCTATAAGAGGCG-3′) for Wnt1-Cre transgene;IRES-N (5′-GCAAGGTCTGTTGAATGTCGTTGA-3′) and IRES-C(5′-GTACCTTCTGGGCATCCTTCAGC-3′) for Isl1-Cre transgene;and Cre1 (5′-GTTCGCAAGAACCTGATGGACA-3′) Cre2(5′-CTAGAGCCTGTTTTGCACGTTC-3′) for Cre transgene.

Staining was performed essentially as described previously(Mukouyama et al., 2002). Embryos or limb skin tissue was fixed with 4% paraformaldehyde/PBS at 4°C overnight, sunk in 30% sucrose/PBS at 4°C and then embedded in OCT compound for frozen sectioning (15 μm). Staining was performed using anti-β-galactosidase antibodies (rabbit polyclonal antibody, 5-prime 3-prime, 1:1000, overnight at 4°C or goat polyclonal antibody, Bio Trend,1:1000, overnight at 4°C) to detect lacZ expression; anti-GFP antibody (rabbit polyclonal antibody, Molecular Probes; 1:500, overnight at 4°C); anti-BFABP antibody (rabbit polyclonal antibody, T. Müller,1:1000, 3 hours at room temperature) to detect glial cells; Alexa green-conjugated HuD antibody (mouse monoclonal antibody, clone 16A11,Molecular Probe; 1:50, 1 hour at room temperature); anti-neuronal class IIIβ-tubulin antibody (mouse monoclonal antibody, clone Tuj1, Covance,1:500, 1 hour at room temperature) to detect neurons; anti-PECAM-1 antibody(rat monoclonal antibody, clone MEC 13.3, BD Pharmingen, 1:300, overnight at 4°C) to detect endothelial cells; Cy3-conjugated anti-αSMA antibody(mouse monoclonal antibody, clone 1A4, Sigma, 1:500, 1 hour at room temperature) to detect smooth muscle cells; and anti-VEGF antibody (goat polyclonal antibody, R&D, 1:200, overnight at 4°C). For immunofluorescent detection, FITC-, Cy3-, Alexa-488-, Alexa-568-, or Cy5-conjugated secondary antibodies (Jackson, 1:300, Southern Biotechnology Associations, 1:300, and Molecular Probes, 1:250, 1 hour at room temperature)were used. All confocal microscopy was carried out on a Leica SP confocal(Leica).

Whole-mount immunohistochemical staining of limb skin was performed as described previously (Mukouyama et al.,2002). Embryos at E15.5 were dissected, fixed overnight in 4%paraformaldehyde/PBS at 4°C, and dehydrated in 100% methanol at -20°C. Staining was performed using anti-PECAM-1 antibody to detect endothelial cells; anti-β-galactosidase antibody to detect lacZ expression;anti-NRP1 antibody (rabbit polyclonal antibody, A.L. Kolodkin, 1:3000, 3 hours at room temperature); and anti-CX40 antibody (rabbit polyclonal antibody,Alpha Diagnostic International, 1:300, overnight at 4°C) as arterial markers; Cy3-conjugated anti-αSMA antibody to detect smooth muscle cells; anti-neurofilament antibody (mouse monoclonal antibody, clone 2H3,Developmental Studies Hybridoma Bank, 1:200, 1 hour at room temperature);anti-peripherin antibody (rabbit polyclonal antibody, Chemicon, 1:1000, 1 hour at room temperature) to detect nerve fibers; anti-BFABP antibody to detect Schwann cells; and anti-VEGF antibody. FITC-, Cy3-, Alexa-488-, Alexa-568- or Cy5-conjugated secondary antibodies were from Jackson, Southern Biotechnology Associations and Molecular Probes. TUNEL labeling was performed according to the manufacturer's protocol (In Situ Cell Death Detection, Roche). The average mean fluorescence (pixels/area) was analyzed using ImageJ software, and statistical significance of samples (n>3) was assessed using Student's t-test.

Whole-mount in situ hybridization was carried out essentially as described previously (Mukouyama et al.,2002; Wang et al.,1998). E15.5 limbs were hybridized with a cRNA probe against the ephrin B2 (Efnb2 - Mouse Genome Informatics) extracellular domain.

The isolation and culture of ephrin B2-negative endothelial cells from E10.5 Efnb2^lacZ/+ embryos was performed as described previously (Mukouyama et al.,2002). All sorts and analyses were performed on a FACS Vantage dual laser flow cytometer (BD Biosciences). The culture medium contained EMB-2(Clonetics) with 15% FBS (Hyclone Laboratories), Penicillin/Streptomycin(BioWhittaker) and 10 ng/ml bFGF (R&D). The freshly isolated endothelial cells were challenged with serial dilutions (1-100 pg/ml) of VEGF¹²⁰ or VEGF¹⁶⁴ isoforms (R&D), used at equal mass/volume concentrations according to the manufacturer's technical specifications. Cultures were incubated for 2 days in a reduced oxygen environment to more closely approximate physiological oxygen levels(Mukouyama et al., 2002; Studer et al., 2000). X-gal staining and immunohistochemistry on these cultures were performed as described (Mukouyama et al.,2002).

The ephrin B2-lacZ-positive or negative cells were counted in PECAM1-positive cells and statistical significance was assessed using the Student's t-test.

To test directly whether peripheral nerve-derived VEGFA is required for arteriogenesis of nerve-associated vessels in vivo, we sought to conditionally knock out Vegfa, specifically in these neural elements. At the stages of development investigated (E13.5-E15.5), peripheral nerves are comprised of Schwann (glial) cells and sensory, but not autonomic, axons(Mukouyama et al., 2002). We have previously shown that both Schwann cells and sensory neurons express mRNAs for various VEGF isoforms, in vivo(Mukouyama et al., 2002). As both of these cell types are neural crest-derived, deletion of Vegfain them can be accomplished by employing a Wnt1-Cre driver, which is active in the pre-migratory neural crest(Danielian et al., 1998). We confirmed deletion in both sensory neurons and peripheral glia by Wnt1-Cre,using the Rosa26 Cre-dependent lacZ reporter (R26R)line (Soriano, 1999)(Fig. 1A,G,I,K).

Wnt1-Cre does not delete in motoneurons(Fig. 1I, open arrow). However,motor axons are present in the peripheral nerves innervating limb skin at E14.5, as revealed by an enhanced green fluorescent protein (eGFP) reporter driven by the motoneuron-specific HB9 promoter (see Fig. S1A-E,arrows, in the supplementary material)(Wichterle et al., 2002). To determine whether such motor axons might be an additional source of VEGFA, we carried out an RT-PCR analysis of Vegfa mRNA expression on EGFP⁺ motor neurons isolated by fluorescence-activated cell sorting(FACS), from the spinal cord of Hb9::EGFP embryos. This analysis revealed that motor neurons do express mRNAs for all isoforms of VEGFA (see Fig. S1F in the supplementary material). These data therefore indicated that elimination of all nerve-derived sources of Vegfa in the limb skin should require deletion of the gene in Schwann cells, sensory and motoneurons.

As there is no single promoter-driven Cre line that simultaneously and specifically targets all three of these cell types, we combined the Wnt1-Cre driver with an Isl1-Cre driver, which is active in both sensory and motoneurons, but not in Schwann cells (T. M. Jessell, unpublished)(Srinivas et al., 2001; Yang et al., 2001). Analysis of Isl1-Cre; R26R embryos using cell type-specific markers confirmed that this Cre driver deletes in both sensory and motoneurons, but not in Schwann cells (Fig. 1B; Fig. 1H, inset; Fig. 1J,K). Thus, the combination of Wnt1-Cre and Isl1-Cre should suffice to delete the conditional Vegfa allele (Vegf^flox) in sensory neurons, motoneurons and Schwann cells(Fig. 1K; see Fig. S2 in the supplementary material).

To create embryos carrying both the Wnt1-Cre and Isl1-Cretransgenes, as well as both `floxed' alleles of the Vegf gene(Gerber et al., 1999), we generated and then inter-crossed Wnt1-Cre; Vegf^flox/+ with Isl1-Cre; Vegf^flox/+ mice. On average, one in 16 embryos derived from this inter-cross inherited both the Wnt1-Cre and Isl1-Cre transgenes, as well as both Vegf alleles(Vegf^flox/flox). The limb skin of such embryos was subjected to whole-mount analysis of the expression of three independent arterial markers: neuropilin 1 (NRP1), connexin 40 (CX40; GJA5 - Mouse Genome Informatics) and ephrin B2. The expression of all three markers was greatly reduced in small-diameter vessels of embryos of this genotype(Fig. 2D,F,G,K,M, arrows; Fig. 2N; data not shown). The expression of smooth muscle markers was also reduced, albeit to a lesser extent than that of the arterial endothelial markers, in such embryos (data not shown). Embryos carrying either of the individual Cre transgenes also showed a clear, but less extensive, reduction of arterial marker expression (Wnt1-Cre; Fig. 2E,G,L, arrows; Fig. 2N); this was particularly evident for Isl1-Cre (data not shown). TUNEL-labeling experiments indicated that the reduction of arterial marker expression does not reflect a selective death of arterial endothelial cells in these Vegf conditional knockout embryos(Fig. 2O-U).

In contrast to the defective arterial differentiation of smaller-diameter vessels, the expression of NRP1 and ephrin B2 was relatively unaffected in the major, large-diameter vessels in the skin(Fig. 2D-F; arrowheads; data not shown). These data are consistent with our previous observation that expression of these arterial markers is not diminished in the large-diameter limb vessels of mutants lacking peripheral nerve fibers or Schwann cells(Mukouyama et al., 2002). Similarly, no reduction in the expression of arterial markers in other large-diameter vessels, such as the carotid and thoracic arteries, was observed in the mutants (see Fig. S3 in the supplementary material).

These data suggested that the arterial differentiation of small-diameter nerve-associated vessels, but not of large-diameter vessels, is compromised in Wnt1-Cre; Isl1-Cre; Vegf^flox/flox embryos. One possible explanation for this dissociation is that autocrine or paracrine secretion of VEGFA by endothelial or smooth muscle cells in large-diameter vessels may compensate for the absence of nerve-derived VEGFA. In support of this,antibody staining revealed that VEGFA is expressed in both endothelial and smooth muscle cells of large-diameter blood vessels (see Fig. S4E-H in the supplementary material, arrows). Furthermore, expression of mRNA for VEGF¹⁶⁴ was detected by RT-PCR in endothelial cells that were freshly isolated from E15.5 limb skin (data not shown).

The foregoing data suggested that nerve-derived VEGF is selectively required for arteriogenesis in small-diameter nerve-associated blood vessels within limb skin. To determine whether other aspects of neuro-vascularization might also depend on peripheral-nerve-derived VEGF, we examined blood vessel development in the dorsal root sensory ganglia (DRG) of the Vegfaconditional mutants. In control (Vegf^flox/flox) embryos,blood vessels both surround the DRG (Fig. 3C, open arrowhead) and branch into it(Fig. 3C, arrow). By contrast,there was virtually a complete failure of such internal ganglionic vascular branching in the DRG of Wnt1-Cre; Isl1-Cre; Vegf^lox/lox or Wnt1-cre; Vegf^lox/lox embryos(Fig. 3B,D,G; data not shown),while the vessels peripheral to the ganglia appeared unaffected(Fig. 3C,D; open arrowheads). These data suggest that VEGFA derived from sensory neurons and/or glia is required for proper vascularization of the DRG, as well as for arteriogenesis in limb skin.

In addition to their requirement for arterial differentiation, peripheral nerves instructively guide the pattern of arterial branching in limb skin(Mukouyama et al., 2002). Surprisingly, in contrast to the clear defects in arterial marker expression,no evident perturbation of nerve-blood vessel association was observed in the nerve-derived Vegfa conditional mutants(Fig. 4A-I; data not shown). The extent of innervation of the skin, and the coverage of nerves by Schwann cells (detected by expression of BFABP), also appeared to be normal in the mutants (Fig. 4E-H).

The foregoing experiments indicated that conditional inactivation of Vegfa in peripheral nerve caused a failure of arterial differentiation, but not of nerve-vessel alignment. To examine directly the extent of VEGF reduction in the conditional mutants, Cre-dependent excision of exon3 of the floxed Vegf allele was first examined by genomic PCR in both the DRG (Fig. 5J) and limb skin (data not shown). This analysis revealed a high degree of recombination in DRG from Wnt1-Cre, Isl1-Cre; VEGF^flox/flox embryos(non-recombined:recombined=30%:70%).

The residual non-recombined Vegf^flox DNA(Fig. 5J, upper band) may reflect incomplete excision by Cre recombinase in cell types expressing Wnt1-Cre and Isl1-Cre, or else contaminating non-neural cells (fibroblasts, endothelial cells, smooth muscle cells, etc.) that do not express these Cre drivers. To distinguish these alternatives, we analyzed the level of VEGFA protein expression in such embryos by anti-VEGF antibody staining (Fig. 5A-I). Anti-VEGFA immunoreactivity was detected in both the DRG and spinal cord of control Vegf^flox/flox embryos(Fig. 5A, arrows and arrowheads, respectively). Such immunoreactivity was markedly reduced in both the DRG and spinal cord of Wnt1-Cre; Isl1-Cre;Vegf^flox/flox embryos(Fig. 5C,I; arrows and arrowheads), although a few DRG cells still exhibited staining(Fig. 5C,I; open arrows). Quantification, by measurement of fluorescent pixel intensity per unit area,indicated an approximately threefold reduction in staining levels in the mutant embryos (Fig. 5I). A similar analysis in limb skin revealed a strong reduction of VEGFA expression in peripheral nerves of embryos harboring both Cre drivers (data not shown). Other sites of VEGF expression, where Wnt1-Cre and Isl1-Cre are not active, were unaffected. Taken together, these DNA- and protein-level analyses validated the peripheral nerve-specific deficiency of VEGFA. However, there appeared to be a residual level of VEGFA in these neural tissues, despite the use of two different Cre drivers with overlapping cellular specificities. This may reflect the requirement for both Vegf^flox alleles to undergo recombination in each cell, in order to achieve reduction to homozygosity.

The residual low level of VEGFA in the peripheral nerves of our conditional knockout embryos left open the possibility that such levels were sufficient for the persistence of nerve-vessel alignment in these mutants. To address this issue by an independent approach, we sought to selectively reduce the VEGF sensitivity of endothelial cells. Because deletion of the core VEGFA receptor FLK1 (VEGFR2) in endothelial cells would be expected to compromise their viability (M. Hirashima and J. Rossant, personal communication), we examined embryos with an endothelial cell-specific deletion of the VEGF¹⁶⁴ co-receptor NRP1(Kawasaki et al., 1999; Soker et al., 1998).

Such a deletion was achieved by using the endothelial-specific Tie2- promoter to drive Cre expression in a genetic background containing a floxed allele of Nrp1 over a conventional null allele(Gu et al., 2003). In Tie2-Cre; Nrp1^flox/- embryos, expression of arterial markers, including CX40 and ephrin B2, was strongly reduced in limb vessels(compare Fig. 6C with 6D, arrows; Fig. 6E; data not shown). Surprisingly, in these mutants even the large-diameter vessels exhibited a strong reduction of arterial marker expression, as well as in smooth muscle coverage (compare Fig. 6C with 6D; compare Fig. 6H with 6I, arrowheads). The defects in limb skin arterial differentiation observed in mutants lacking endothelial NRP1 were, therefore, similar but even more pronounced than those observed in embryos deficient in peripheral nerve-derived VEGFA. Despite this more severe failure of arteriogenesis in limb skin, the association of nerves with blood vessels appeared normal in Tie2-Cre; Nrp1^flox/- embryos(Fig. 6B,G, compare arrows with open arrowheads).

The requirement of Nrp1 for arteriogenesis in vivo is surprising,given that NRP1 is a co-receptor for only the VEGF¹⁶⁴ isoform, and that peripheral nerve elements express mRNAs for multiple VEGF isoforms(Mukouyama et al., 2002) (see Fig. S1F in the supplementary material). This observation raised the issue of whether VEGF¹⁶⁴ is more potent than other VEGFA isoforms in its ability to induce arterial marker expression in undifferentiated endothelial cells. To address this, we compared the ability of VEGF¹⁶⁴ and a non-NRP1-binding isoform, VEGF¹²⁰, to induce arterial marker expression in vitro. We employed a previously described system in which PECAM1⁺, β-gal^- embryonic endothelial cells are isolated by flow cytometry from Efnb2^lacZ/+ knock-in mice and cultured in bFGF (Mukouyama et al.,2002). Under these conditions, the effects of VEGF in inducing arterial marker expression can be examined in the absence of any confounding effects on cell survival.

In such cultures, both VEGF¹⁶⁴ and VEGF¹²⁰ induced expression of Efnb2-lacZ (Fig. 7A-F). At 100 pg/ml, both isoforms induced ephrin B2 expression to a similar extent (Fig. 7M, left panel). However, at lower concentrations (1-10 pg/ml), VEGF¹⁶⁴ was at least twice as effective as VEGF¹²⁰ at inducing ephrin B2-lacZ expression (Fig. 7M, green bars). There was no effect on cell survival at any of the concentrations tested, for both isoforms(Fig. 7M, right panel). These data suggest that NRP1-binding isoforms of VEGF may preferentially promote arterial differentiation, in comparison with non-NRP1-binding isoforms, at very low concentrations of the growth factor.

We have shown that the artery-specific expression of NRP1 is dependent on nerve-derived VEGFA in vivo (Fig. 2E-G). Moreover, NRP1 itself is required in endothelial cells for arterial differentiation (Fig. 6D), and an NRP1-binding isoform of VEGF preferentially upregulates the arterial marker ephrin B2 in vitro. These data suggest that a positive-feedback loop involving NRP1 and VEGF may contribute to arteriogenesis. To address more directly this possibility, we examined the ability of VEGFA isoforms to upregulate NRP1 expression in vitro. Both VEGF¹²⁰ and VEGF¹⁶⁴ upregulated expression of NRP1 in cultured embryonic endothelial cells (Fig. 7G-L). The proportion of NRP1⁺ cells was similar to that of ephrin B2⁺ cells, in the presence of either VEGF isoform,although direct double-labeling for both markers was precluded by antibody incompatibility. These data demonstrate that NRP1-binding VEGFA isoforms can upregulate the expression of NRP1 itself, as well as that of other arterial-specific markers, in embryonic endothelial cells.

In this study, we have used a conditional gene inactivation approach to investigate the signaling systems that control nerve-dependent arteriogenesis and blood vessel patterning in developing limb skin. Our data provide the first in vivo evidence that VEGF, which is derived from local tissue sources,is required to induce arteriogenesis during angiogenic remodeling of a primitive capillary network. The results also imply an important role for the heparin-binding isoform of VEGF in arterial differentiation in vivo. Consistent with this, we provide evidence that the NRP1-binding isoform of VEGF is a more potent inducer of arterial markers in vitro, compared with other VEGF isoforms. These data suggest a model in which a `winner-takes-all'competition for VEGF may control arterial differentiation, with the outcome biased by a VEGF¹⁶⁴-NRP1 positive-feedback loop. The requirement for NRP1 may also help to explain how arteriogenesis is restricted to those vessels located in close proximity to the nerve.

Despite their profound effects on arterial differentiation, neither the Vegfa nor the Nrp1 conditional mutants disrupted nerve-vessel alignment. This genetic uncoupling of arteriogenesis and nerve-vessel association suggests that guidance of the vessel branching pattern by the nerve requires either a very low concentration of VEGF, or more likely a second, as yet unidentified, signaling system(Fig. 8A).

Previous studies have indicated that VEGF is required for de novo differentiation of the dorsal aorta in zebrafish(Lawson et al., 2002), and is sufficient to increase the density of cardiac arterial vessels when overexpressed in the heart (Visconti et al., 2002). However, these experiments did not address whether VEGF is required for arterial differentiation from a pre-existing capillary network, during normal development in vivo. The present results indicate that nerve-derived VEGFA is required for arteriogenesis in the primitive capillary plexus of murine limb skin. These data provide additional evidence that VEGF functions as an arterial-inducing signal in vivo, in addition to its classical role as a growth and survival factor for endothelial cells(Ferrara, 2000). Interestingly, the degree of arterial differentiation in the nerve-specific Vegfa mutants depended upon the extent of Vegfa deletion (as observed in different Wnt1- or Isl1-Cre-containing genotypes; Fig. 2). This suggests a dose-sensitive requirement for VEGFA in nerve-dependent arteriogenesis,consistent with our in vitro observations(Mukouyama et al., 2002).

Recent data have demonstrated that the endothelial specific Notch ligand Delta-like 4 (Dll4) is essential for arteriogenesis in vivo(Duarte et al., 2004; Gale et al., 2004; Krebs et al., 2004). Studies in zebrafish have suggested that dorsal aorta formation is controlled by VEGF and Notch signaling, acting in series(Lawson et al., 2001; Lawson et al., 2002). In preliminary experiments, we have found that arterial markers are induced in cultured murine embryonic endothelial precursor cells not only by VEGF, as previously reported (Mukouyama et al.,2002), but also by soluble Notch ligands. Interestingly, the ability of VEGF to induce arterial marker expression is blocked by peptide inhibitors of Notch activation, while a soluble VEGF receptor antagonist does not block induction of such markers by Notch ligands (Y.M. and D.J.A.,unpublished). These data suggest that in mouse, as in zebrafish, VEGF induces arterial differentiation by acting through the Notch pathway. How the interaction between these two signaling systems occurs is not yet clear.

What cell type(s) within peripheral nerve are the most important source(s)of VEGFA? We observed the strongest reduction of arterial marker expression when Vegfa was deleted in all three neural cell types(Wnt1-Cre+Isl1-Cre), and a weaker phenotype using either Wnt1-Cre or Isl1-Cre alone(Fig. 2 and data not shown). One explanation for this higher expression is that it reflects a requirement to eliminate Vegfa expression in all three peripheral nerve cell types, as Isl1-Cre is not expressed in Schwann cells, while Wnt1-Cre is not expressed in motoneurons. Motoneurons are not the primary source of VEGF in the nerve, however, as arterial marker expression is normal in Olig1; Olig2 double mutants(Zhou and Anderson, 2002),which lack all spinal motoneurons, but have a normal complement of sensory neurons (Y.M., Q. Zhou and D.J.A., unpublished). A Schwann cell-specific knockout of Vegfa is currently not possible, because all available Cre driver lines active in these peripheral glia, including Po-Cre(Feltri et al., 1999; Giovannini et al., 2000),desert hedgehog (Dhh)-Cre(Lindeboom et al., 2003) and BFABP-Cre (Anthony et al.,2004), also promote recombination in peripheral neurons (Y.M. and D.J.A., unpublished). This most probably reflects the early expression of these glial promoters in multipotent neural crest stem or progenitor cells(Kim et al., 2003).

A second explanation is that, since both Wnt1-Cre and Isl1-Cre are expressed in sensory neurons, the stronger phenotype of Wnt1-Cre, Isl1-Cre embryos reflects a higher level of Cre expression in these cells, leading to more efficient recombination at the Vegfalocus. The structure of the conditional Vegfa allele makes it impossible to determine whether individual sensory neurons have undergone Cre-mediated recombination in one or both alleles. However, the reduction of VEGFA immunoreactivity in the DRG appeared greater in Wnt1-Cre;Isl1-Cre mutants than in Wnt1-Cre mutants (data not shown),suggesting a more efficient recombination with both Cre drivers.

An endothelial-specific knockout of Nrp1 produced an even more severe defect in arteriogenesis than was observed in the nerve-specific Vegfa conditional knockouts. There are several possible explanations for this difference. First, cardiac defects in the endothelial Nrp1mutants (Gu et al., 2003)might affect peripheral arteriogenesis via impaired blood flow; by contrast cardiac development in the nerve-Vegfa mutants appears to be normal. We think this is unlikely, because dorsal aorta arteriogenesis is normal in conventional Nrp1-null mice (Y.M., D.J.A. and H. Fujisawa,unpublished). Second, the Tie2-Cre; Nrp1 phenotype could reflect a requirement for a ligand of NRP1 other than VEGF¹⁶⁴, such as a semaphorin. In avian embryos, implantation of Sema3a protein or of a neutralizing anti-Sema3a antibody, caused a failure of proper vascularization(Bates et al., 2003). Others have reported angiogenesis defects in Sema3a mutant mice on a CD-1 genetic background (Serini et al.,2003). However, we observed no defects in limb skin arteriogenesis in Sema3a-null mice (Taniguchi et al., 1997), on the C57Bl6 background used here(Mukouyama et al., 2002). Third, the stronger arteriogenesis defects in Tie2-Cre; Nrp1embryonic limb skin vasculature may reflect a more complete interruption of VEGF signaling than is obtained in Wnt1-Cre; Isl1-Cre;Vegf^flox/flox embryos. This could occur either because deletion of a crucial coreceptor in endothelial cells reduces responsiveness to VEGF¹⁶⁴ derived from all local tissue sources, neural or otherwise, and/or because Cre-mediated recombination of Nrp1 is required in only one allele to achieve reduction to homozygosity, whereas recombination of both Vegf^flox alleles is required.

What might be the function(s) of NRP1 in arteriogenesis? Because NRP1 is arterial specific (Moyon et al.,2001; Mukouyama et al.,2002) and is itself induced by VEGF (Y.M., H.-P.G., N.F. and D.J.A., unpublished) (Fig. 7)(Oh et al., 2002), it could mediate a positive-feedback loop, that increases the sensitivity of nascent arterial cells to VEGF. This feedback loop could be initiated by non-NRP1-binding isoforms of VEGF (e.g. VEGF¹²⁰), which we have shown induce NRP1 in vitro, and maintained or further amplified by the NRP1-binding isoform. These considerations suggest a model in which a`winner-takes-all' competition for VEGF may control arterial differentiation,with the outcome biased by a VEGF¹⁶⁴-NRP1 positive-feedback loop(Fig. 8B). In addition, because NRP1 is selective for the heparin-binding isoform of VEGF, it could also function to restrict induction of arterial markers to those vessels in close proximity to nerves.

This model predicts that nerve-dependent arteriogenesis should be compromised in mice selectively lacking the NRP1-binding isoform. But in such mice, e.g. those expressing only VEGF¹²⁰, arterial differentiation is normal in both the retina (Stalmans et al., 2002) and developing limb (Y.M., D.J.A. and P. Carmeliet,unpublished). The levels of VEGF¹²⁰ expression in these mutant mice, however, are similar to the combined levels of VEGF¹²⁰,VEGF¹⁶⁴ and VEGF¹⁸⁸ expression in wild-type mice, i.e. almost three-fold higher than normal(Carmeliet et al., 1999). As we have shown in vitro, at higher concentrations (>10 pg/ml),VEGF¹²⁰ is equally effective as VEGF¹⁶⁴ at inducing arterial differentiation. Therefore the increased levels of VEGF¹²⁰in Vegf^120/120 mice may obscure the requirement for the VEGF¹⁶⁴ isoform in arteriogenesis. Such an explanation would be consistent with the fact that in zebrafish, overexpression of either Vegf¹²¹ or Vegf¹⁶⁵ can rescue arterial differentiation blocked by a deficiency of Shh signaling(Lawson et al., 2002).

Although a commercial source of both VEGF isoforms was used according to the manufacturer's technical specifications (R&D), we recognize the possibility that these two preparations might have differences in specific activity that could account for their differential ephrin B2-inducing activities in our assay. It is difficult to compare the activities of these two isoforms independently of the presence of NRP1, because both isoforms induce this co-receptor at concentrations well below those at which they promote survival and proliferation. However, using porcine aortic endothelial cells, which do not express NRP1, Soker et al.(Soker et al., 1998)demonstrated that both VEGF¹²¹ and VEGF¹⁶⁵ had equal activities in a chemotaxis assay when the cells were reconstituted with exogenous VEGFR2 (Soker et al.,1998). Furthermore the concentrations of VEGF¹²⁰ used in our assay were 1.4-fold higher, on a molar basis, than the concentration of VEGF¹⁶⁴ used at each serial dilution. The fact that the latter isoform exhibited more potent ephrin B2-inducing activity, under conditions where there was a bias in favor of VEGF¹²⁰, makes it more likely that the difference observed is indeed intrinsic to the two isoforms.

During normal development, arteriogenesis is immediately preceded by nerve-vessel alignment and is dependent on the presence of nerves(Mukouyama et al., 2002). Our results demonstrate that it is possible to genetically uncouple these sequential processes: in mutants that disrupt VEGF signaling from the nerve to the vessels, arteriogenesis is disrupted, while nerve-vessel alignment is apparently unperturbed. The simplest explanation for this uncoupling is that a nerve-derived factor distinct from VEGF mediates the alignment process. However, the residual VEGF detected in the conditional Vegf knockout embryos leaves open the possibility that nerve-vessel alignment simply requires a lower threshold level of VEGF, than does arteriogenesis. Nevertheless, the fact that the arteriogenesis defect is even stronger in the endothelial-specific Nrp1 knockout, while nerve-vessel alignment remains intact, argues against the dual-threshold model. At the very least,for such a model to be tenable, the Tie2-Cre; Nrp1 data would require that alignment depend only on the non-heparin-binding, more diffusible VEGF isoforms. This is reasonable given that alignment presumably requires action-at-a-distance to attract the vessels to the nerves. However, we did not see any significant defects in nerve-vessel alignment in the limb skin of mutants that lack the VEGF¹²⁰ isoform (Y.M., D.J.A. and P. Carmeliet, unpublished). Therefore, the available data favor the notion that VEGF and a separate, as yet unidentified signal, mediate arteriogenesis and nerve-vessel alignment, respectively.

What determines which blood vessels in skin become aligned with nerves and undergo arteriogenesis? One explanation is that the process is stochastic: all vessels are equally capable of becoming aligned with nerves, and the selection of a subset for arteriogenesis is simply determined by their initial proximity to nerves at the time of innervation. A mechanism involving a`winner-takes-all' competition for limiting amounts of nerve-derived VEGF,biased by the VEGF¹⁶⁴-NRP1 positive-feedback loop discussed earlier, would be well suited to such a mechanism(Fig. 8B). Alternatively, a subset of vessels may be pre-specified for association with the nerve. For example, if nerves release a signal that recruits vessels to align with them,then a subset of vessels might express higher levels of a receptor for this alignment signal, before the nerves ever arrive. Such a pre-specification mechanism would, however, require precise matching to ensure a sufficient supply of presumptive arterial vessels for developing nerves. The identification of nerve-derived signals for vessel alignment may help to distinguish between these alternatives.

We thank Andrew McMahon for Wnt1-Cre mice; Tom Jessell, Yasuto Tanabe and Chris William for Isl1-Cre mice; Alex Kolodkin and David Ginty for Nrp1^lox/- mice and anti-NRP1 antibody; and T. Müller for providing anti-BFABP antibody. Thanks to R. Diamond, P. Koen and S. L. Adams for FACS assistance; S. Pease, B. Kennedy and the staff of the Transgenic Animal Facility at Caltech for assistance with mouse breeding and care; G. Mosconi for laboratory management; J. S. Chang and M. Martinez for technical support; and G. Mancuso for administrative assistance. Thanks also to D. H. Shin for invaluable help and discussion, C. Hochstim for helpful comments on the manuscript, and other Anderson lab members for technical help and discussion. D.J.A. is an Investigator of the Howard Hughes Medical Institute.