Neuropilin-1/GIPC1 Signaling Regulates α5β1 Integrin Traffic and Function in Endothelial Cells

Neuropilin 1 (Nrp1) is a coreceptor for vascular endothelial growth factor A165 (VEGF-A165, VEGF-A164 in mice) and semaphorin 3A (SEMA3A). Nevertheless, Nrp1 null embryos display vascular defects that differ from those of mice lacking either VEGF-A164 or Sema3A proteins. Furthermore, it has been recently reported that Nrp1 is required for endothelial cell (EC) response to both VEGF-A165 and VEGF-A121 isoforms, the latter being incapable of binding Nrp1 on the EC surface. Taken together, these data suggest that the vascular phenotype caused by the loss of Nrp1 could be due to a VEGF-A164/SEMA3A-independent function of Nrp1 in ECs, such as adhesion to the extracellular matrix. By using RNA interference and rescue with wild-type and mutant constructs, we show here that Nrp1 through its cytoplasmic SEA motif and independently of VEGF-A165 and SEMA3A specifically promotes α5β1-integrin-mediated EC adhesion to fibronectin that is crucial for vascular development. We provide evidence that Nrp1, while not directly mediating cell spreading on fibronectin, interacts with α5β1 at adhesion sites. Binding of the homomultimeric endocytic adaptor GAIP interacting protein C terminus, member 1 (GIPC1), to the SEA motif of Nrp1 selectively stimulates the internalization of active α5β1 in Rab5-positive early endosomes. Accordingly, GIPC1, which also interacts with α5β1, and the associated motor myosin VI (Myo6) support active α5β1 endocytosis and EC adhesion to fibronectin. In conclusion, we propose that Nrp1, in addition to and independently of its role as coreceptor for VEGF-A165 and SEMA3A, stimulates through its cytoplasmic domain the spreading of ECs on fibronectin by increasing the Rab5/GIPC1/Myo6-dependent internalization of active α5β1. Nrp1 modulation of α5β1 integrin function can play a causal role in the generation of angiogenesis defects observed in Nrp1 null mice.


Introduction
In vertebrates, the development of a hierarchically organized and functional vascular tree relies on the dynamic interaction of endothelial cells (ECs) with the surrounding extracellular matrix (ECM), which is mediated by heterodimeric ab integrin adhesive receptors [1]. During evolution, vertebrates have acquired an additional set of adhesionrelated genes that regulate blood vessel assembly and function [2]. Among these genes, the ECM protein fibronectin (FN) and a5b1 integrin, the predominant FN receptor, have proven to be essential for embryonic vascular development and tumor angiogenesis [3]. Indeed, in vertebrate embryos FN is the earliest and most abundantly expressed subendothelial matrix molecule [3,4]. Endothelial a5b1 mediates cell adhesion to FN and the assembly of soluble FN dimers (sFN) into a fibrillar network [3], which has also been implicated in branching morphogenesis [5].
The biological activities of integrins depend on the dynamic regulation of their adhesive function in space and time. In cells, integrins exist in different conformations that determine their affinities for ECM proteins [6] and are continuously endocytosed, trafficked through endosomal compartments, and recycled back to the plasma membrane [7,8]. Therefore, during vascular morphogenesis, real-time modulation of EC-ECM adhesion can result from two interconnected phenomena: the regulation of integrin conformation and traffic in response to extracellular stimuli [8,9]. Indeed, there is mounting evidence that pro-and antiangiogenic cues regulate blood vessel formation by modulating integrin function [1]. In this respect, the transmembrane glycoprotein neuropilin 1 (Nrp1), which is ex-pressed in both neurons and ECs [10], is remarkable because it was originally identified as a surface protein mediating cell adhesion [11] and then found to also act as a coreceptor for both pro-and antiangiogenic factors, such as vascular endothelial growth factor A 165 (VEGF-A165, VEGF-A164 in mice) [12,13] and semaphorin 3A (SEMA3A) [14][15][16][17][18][19][20], respectively.
Nrp1 null mice display an embryonic lethal phenotype, characterized by dramatic vascular defects ascribed to impaired angiogenic sprouting [23], branching [24], or arterialization [25] that is significantly more severe than and/or qualitatively different from that of mice lacking either VEGF-A164 (VEGF 120/120 mice) [26] or Sema3A [16]. Indeed, although Nrp1 -/embryos die in utero by 13.5 days postcopulation [23], VEGF 120/120 pups are recovered at birth at a normal Mendelian frequency [27]. Moreover, a major feature of Nrp1 null mutants, i.e., the severe impairment of neural tube vascularization [23], is not phenocopied by VEGF 120/120 mouse embryos [26]. In addition, differently from Nrp1 null mice [23,24], the vascular phenotype of Sema3a null mice is significantly influenced by the genetic background [16,[28][29][30]. These findings suggest that the vascular defects caused by the loss of Nrp1 could be due to a VEGF-A164/SEMA3Aindependent function of Nrp1 in vascular cells, such as adhesion to the ECM [31,32]. However, how Nrp1 regulates integrin-dependent EC linkages to the surrounding matrix is still obscure. Here, we shed light on the molecular mechanisms by which Nrp1, via its short cytoplasmic domain and independently of VEGF-A165 and SEMA3A, specifically controls a biological function that is crucial for vascular development [3], namely, a5b1-mediated EC adhesion to FN.

Nrp1 Specifically Promotes EC Adhesion to FN and FN Fibrillogenesis
To understand the mechanisms by which Nrp1 modulates EC adhesion to different ECM proteins, we silenced the expression of Nrp1 in human umbilical artery ECs by RNA interference (RNAi). Parenthetically, Nrp1 has been found to be expressed at higher levels in arteries than in veins [33]. Endothelial cells were transfected twice with either a pool of three different small interfering RNAs (siRNAs) targeting human Nrp1 (sihNrp1) or control nontargeting siRNA (siCtl). Twenty-four hours after the second transfection, Western blot analysis revealed that, in comparison with control cells, Nrp1 protein, but neither b-tubulin nor the Nrp1 interactor GAIP interacting protein C terminus, member 1 (GIPC1), was successfully silenced in sihNrp1 ECs ( Figure 1A).
Next, we investigated the effect of Nrp1 silencing on EC adhesion to different ECM proteins. Fibronectin, vitronectin (VN), and type I collagen (COLL-I) are typical constituents of the provisional angiogenic ECM [1,3], whereas laminin (LN) isoforms are major components of the vascular basement membrane surrounding both immature and mature blood vessels [34]. Short-term (15 min) adhesion assays showed that loss of Nrp1 greatly reduced EC adhesion to FN but not to VN, COLL-I, or LN ( Figure 1B-E), suggesting that positive modulation of cell adhesion by Nrp1 is not a general phenomenon [31] but rather a function restricted to specific ECM proteins, such as FN.
Because FN polymerization by ECs has been suggested to participate in vascular morphogenesis [3], we next examined the role of Nrp1 in the fibrillogenesis of endogenous FN. During FN matrix assembly, current models envisage the binding of sFN to surface integrins, thus causing the conversion of FN to a conformational form that favors fibril formation through interactions with other integrin-bound FN dimers [3]. Endothelial cells were cultured in a medium containing FN-depleted fetal calf serum, and accumulation of endogenous FN into fibrils was then detected by confocal immunofluorescence analysis. In comparison with control cells, sihNrp1 ECs were impaired in their ability to incorporate endogenous sFN into a dense fibrillar network 3 h after plating ( Figure 1F and 1G). Time-course real-time reverse transcription PCR (RT-PCR) and Western blot analyses revealed that the endogenous FN fibrillogenesis defect observed in sihNrp1 ECs was not due to a reduction in FN mRNA ( Figure S1A) or protein ( Figure S1D). Hence, Nrp1 specifically promotes EC adhesion to FN and FN matrix formation.

Author Summary
The vascular system is a hierarchical network of blood vessels lined by endothelial cells that, by means of the transmembrane integrin proteins, bind to the surrounding proteinaceous extracellular matrix (ECM). Integrins are required for proper cardiovascular development and exist in bent (inactive) and extended (active) shapes that are correspondingly unable and able to attach to the ECM. Extracellular guidance cues, such as vascular endothelial growth factor and semaphorins, bind the transmembrane protein neuropilin-1 (Nrp1) and then activate biochemical signals that, respectively, activate or inactivate endothelial integrins. Here, we show that Nrp1, via its short cytoplasmic domain and independently of vascular endothelial growth factor and semaphorins, specifically promotes endothelial cell attachment to the ECM protein fibronectin, which is known to be crucial for vascular development. Notably, Nrp1 favors cell adhesion by associating with fibronectin-binding integrins and promoting the fast vesicular traffic of their extended form back and forth from the endothelial cell-to-ECM contacts. Binding of the Nrp1 cytoplasmic domain with the adaptor protein GIPC1, which in turn associates with proteins required for integrin internalization and vesicle motility, is required as well. It is likely that such an integrin treadmill could act as a major regulator of cell adhesion in general.
The Cytoplasmic Domain of Nrp1 Controls EC Adhesion to FN Independently of VEGF-A165 and SEMA3A To start dissecting the mechanisms by which Nrp1 controls the interaction of human ECs with FN, we sought to compare the abilities of full-length and deletion constructs of mouse Nrp1 (mNrp1) to rescue the adhesion and fibrillogenesis defects of sihNrp1 ECs ( Figure 2). In particular, we investigated the role played by the extracellular and cytoplasmic moieties of Nrp1. Indeed, the Nrp1 cytodomain, although dispensable for SEMA3A collapsing activity in neurons [22], could signal in cultured ECs [35]. Moreover, the C-terminal SEA sequence of Nrp1 interacts with the PDZ domain of the endocytic adaptor protein GIPC1 [36], whose knockdown during development results in altered arterial branching [37]. Therefore, we transduced sihNrp1 ECs with retroviral vectors carrying the hemagglutinin (HA)-tagged full-length (mNrp1) and deletion mutants of murine Nrp1 (Figure 2A), lacking either the C-terminal SEA amino acids (mNrp1dSEA) or the whole cytoplasmic domain (mNrp1dCy). The sihNrp1 pool did not target any of the mNrp1 constructs, and immunoprecipitation experiments on membrane-biotinylated cell monolayers revealed that all three transmembrane proteins were efficiently exposed on the cell surface ( Figure 2B).
In comparison to wild-type mNrp1, both mNrp1dSEA and mNrp1dCy constructs were severely impaired in their abilities to rescue sihNrp1 EC defects in adhesion to FN ( Figure 2C) and endogenous FN fibrillogenesis ( Figure 2D-F). Accordingly, only mNrp1 overexpression stimulated the adhesion of NIH 3T3 fibroblasts to FN, whereas neither mNrp1dSEA nor mNrp1dCy were active in this respect ( Figure 2G). Moreover, mNrp1 overexpression did not promote NIH 3T3 adhesion to VN ( Figure S2), further supporting the concept that Nrp1 behaves as a substrate-specific enhancer of cell adhesion.
Hence, it appears that the cytoplasmic domain of Nrp1, in particular its SEA motif, which interacts with the endocytic adaptor GIPC1 [36], is required for Nrp1 stimulation of EC spreading on FN and polymerization of endogenous FN.
Opposing autocrine loops of VEGF-A [38][39][40][41] and SEMA3A [16,19,42,43] have been found in ECs both in vitro and in vivo. Therefore, we investigated whether the SEA motif and the full cytoplasmic domain of Nrp1 could be required for the modulation of EC adhesion to FN by VEGF-A165 and SEMA3A. Consistent with previous observations [16,18,20,44], silencing Nrp1 completely blocked VEGF-A165-dependent stimulation ( Figure 2H) and SEMA3Adependent inhibition ( Figure 2I) of human EC adhesion to FN. As expected, inhibition of cell adhesion to FN by SEMA3F, which signals through Nrp2 [20,21], was not affected by Nrp1 knockdown ( Figure 2J). Moreover, similarly to what was observed for SEMA3A in neurons [22], we found that the cytoplasmic domain of Nrp1 is entirely dispensable for both VEGF-A165 ( Figure 2H) and SEMA3A ( Figure 2I) activity on EC adhesion to FN, because all three mNrp1 constructs rescued sihNrp1 EC response to these factors with a similar efficiency. Thus, the Nrp1 SEA motif and cytodomain are required for Nrp1 modulation of EC adhesion to FN and sFN incorporation into fibrils but not for Nrp1 activity as a VEGF-A165 and SEMA3A coreceptor.

Nrp1 Regulation of Cell Adhesion Depends on a5b1 Integrin
a5b1 Integrin is the main FN receptor in ECs [1,3], and by transmitting the actin-dependent tension to sFN, it triggers FN fibrillogenesis [45]. To elucidate whether Nrp1 stimulation of cell adhesion to FN was directly mediated by Nrp1 or was dependent on a5b1 integrin, CHO cells lacking (CHO B2) or expressing (CHO B2a27) the a5 integrin subunit were transfected with mNrp1 and allowed to adhere to FN. Overexpression of mNrp1 stimulated CHO cell adhesion to FN in the presence (CHO B2a27; Figure 3A) but not in the absence (CHO B2; Figure 3B) of a5b1 integrin. Therefore, Nrp19s proadhesive activity on FN is nonautonomous and mediated by a5b1 integrin. We then examined whether in ECs Nrp1 could interact physically with a5b1 integrin. Lysates from ECs adhering on endogenous ECM were immunoprecipitated with an antibody (Ab) recognizing the FN receptor a5b1 and then blotted with anti-Nrp1 Ab. Nrp1 coimmunoprecipitated with a5b1, and blotting Nrp1 immunoprecipitates with anti-a5b1-integrin Ab further confirmed the association between endogenous hNrp1 and a5b1 integrin in ECs ( Figure 3C). To better understand whether the Nrp1 cytoplasmic domain was required for the interaction with a5b1 integrin, lysates of NIH 3T3 fibroblasts overexpressing HA-tagged full-length or deletion constructs of mNrp1 and green fluorescent protein (GFP)-tagged a5 integrin subunit (a5-GFP) [46] were immunoprecipitated with anti-GFP Ab and then blotted with anti-HA Ab ( Figure 3D). We found that both the C-terminal SEA and the cytoplasmic domain of Nrp1 were fully dispensable for its interaction with a5b1 integrin.
To understand the spatial and functional relationships between Nrp1 and a5b1 integrin in ECs, we first generated a monomeric red fluorescent protein (mRFP)-tagged mNrp1 construct (mNrp1-mRFP) that was then cotransfected with a5-GFP in ECs. Fluorescent confocal microscopy showed that at the plasma membrane of ECs adhering on FN mNrp1-mRFP was enriched in close proximity to, or even tightly intermingled with, a5-GFP-containing adhesion sites ( Figure 4A, arrows). Moreover, mNrp1-mRFP and a5-GFP fully colocalized in intracellular vesicles ( Figure 4A, arrowheads). Notably, immunofluorescence analysis of endogenous endothelial proteins confirmed the spatial links between hNrp1 and vinculin ( Figure 4B) or a5b1 integrin ( Figure 4C) at either adhesion sites ( Figure 4B and 4C, arrows) or vesicular structures located in their proximity ( Figure 4C, arrowheads). The observation that Nrp1 and a5b1 colocalization was particularly apparent in intracellular vesicles indicated that these two molecules may associate at or near the point of endocytosis and that they may be internalized as a complex, which is then subsequently disassembled upon recycling to the plasma membrane. We have previously found that endosomal integrin complexes can be preserved by treating the cell with primaquine (PMQ), a receptor recycling inhibitor, prior to lysis [47]. Therefore, we immunoprecipitated a5b1 integrin or Nrp1 from cells that had been treated with PMQ for 10 min and probed for the presence of the a5b1/Nrp1 complex by Western blotting. Pretreatment of the cells with PMQ greatly increased the coprecipitation of a5b1 integrin with Nrp1 and vice versa ( Figure 3C), indicating the likelihood that this complex is more stable in endosomes than at the plasma membrane.
To further characterize the interaction between Nrp1 and a5b1 integrin, we measured fluorescence resonance energy transfer (FRET) in live NIH 3T3 cells transfected with a5-GFP alone or cotransfected with a5-GFP and mNrp1 tagged with the fluorescent protein Cherry, an improved version of mRFP (mNrp1-Cherry). Total internal reflection fluorescence (TIRF) illumination [48] was used to selectively excite a5-GFP at the basal cell plasma membrane where ECM adhesions lie. Fluorescence resonance energy transfer was measured by fluorescence lifetime imaging microscopy (FLIM) [49] and was read out as a decrease in donor (GFP) fluorescence lifetime. We found that the a5-GFP fluorescence lifetime was significantly reduced in cells that coexpressed mNrp1-Cherry, indicating that FRET, and thus a close physical interaction, was occurring between a5b1 and Nrp1 at adhesion sites with an 11.5% FRET efficiency ( Figure 5).
Taken together, these data indicate that in living cells Nrp1 physically associates with a5b1 at or near sites of cell-ECM contact and that this interaction is likely maintained following internalization of the complex.

Nrp1 Controls the Traffic of Active a5b1 Integrin
The efficiency of cell adhesion and spreading on ECM is generally thought to be proportional to the amount of either active or total (i.e., active and inactive) integrin at the cell surface [1,6]. We found that lack of Nrp1 did not alter the global amount of either total ( Figure 1A), as already reported [31], or active a5b1 integrin, as recognized by the mouse monoclonal Ab (mAb) SNAKA51 [45] ( Figure S3). Then, we analyzed whether Nrp1 could influence the amount of a5b1 integrin on the endothelial surface. Biotinylation experiments revealed that knocking down human Nrp1 did not diminish the surface levels of either total or active a5b1 Overexpression of mNrp1 stimulated CHO cell adhesion to FN in the presence but not in the absence of a5b1 integrin. (C) Immunoprecipitation of endogenous hNrp1 and a5b1 integrin from ECs preincubated for 10 min either in the absence or in the presence of 0.6 lM PMQ, followed, respectively, by Western blotting with anti-a5-integrin Ab and anti-Nrp1 Ab. In ECs, Nrp1 associates with a5b1 integrin, and the recycling inhibitor PMQ increases the stoichiometry of their interaction. (D) NIH 3T3 fibroblasts were cotransfected with the GFP-tagged a5 integrin subunit or GFP alone together with an HA-tagged version of mNrp1 fulllength (mNrp1) or mNrp1dSEA or mNrp1dCy deletion constructs. a5-GFP, but not GFP alone, coimmunoprecipitates with all HA-tagged mNrp1 constructs. doi:10.1371/journal.pbio.1000025.g003 integrin in sihNrp1 ECs ( Figure 6A), thus suggesting that a mechanism alternative to the control of integrin conformation should be responsible for Nrp1-dependent activation of a5b1 integrin function in ECs.
On the basis of our observations that Nrp1 and a5b1 integrin colocalize in intracellular vesicles ( Figure 4A and 4C) and that inhibition of recycling by PMQ increased the association of Nrp1 with a5b1 integrin (Figure 3C), we decided to monitor the effect of Nrp1 knockdown on the internalization of total and active surface a5b1 integrin. Endothelial cells were surface-labeled with cleavable biotin at 4 8C and incubated at 37 8C for different times to allow internalization, and then biotin remaining on cell-surface proteins was cleaved at 4 8C [50]. Integrin internalization was quantified by immunoprecipitation of either total ( Figure 6A and 6B) or active ( Figure 6A and 6C) a5b1 integrin, followed by Western blot analysis with streptavidin. Notably, although endocytosis of the cell-surface pool of total a5b1 integrin (i.e., active plus inactive heterodimers) was not detectably altered in sihNrp1 cells ( Figure 6A and 6B), knockdown of Nrp1 markedly reduced the quantity of active (SNAKA51-positive) a5b1 heterodimers internalized by ECs ( Figure 6A and 6C). Taken together, these data indicate that on the cell surface Nrp1 interacts with active a5b1 heterodimers at adhesion sites ( Figure 4A and 4C, arrows) and acts to promote their internalization and localization to intracellular vesicles ( Figure 4A and 4C, arrowheads).
To visualize the internalization and postendocytic trafficking of the a5b1/Nrp1 complex, we deployed the photoactivatable (PA) a5-GFP (a5-PA-GFP) probe that we had previously used to monitor a5b1 trafficking in human ovarian carcinoma A2780 cells [51]. However, the multitude of fluorescent vesicles travelling to and from the cell surface made it difficult to track the progress of individual a5b1 integrin transport vesicles. Therefore, we used TIRF to restrict the plane of activating fluorescence, such that only a5-PA-GFP present at or near the cell surface became photoactivated. Then we tracked the movement of this photoactivated fraction of a5b1 integrin using time-lapse epifluorecence microscopy. With this novel technique, a5b1 integrin was photoactivated almost exclusively at adhesion sites (mostly fibrillar adhesions), where it colocalized with mNrp1-Cherry (Figure 7, arrows). Photoactivated a5b1 was then rapidly (,6 s) internalized and cotransported with mNrp1-Cherry in small endocytic vesicles (Figure 7, empty arrowheads, and Video S1) that moved away from the fibrillar adhesions. In addition, we found that a5b1 integrin turnover in ECM adhesions was unexpectedly very rapid ( Figure S4 and Video S2), with the a5-PA-GFP signal leaving the adhesive sites, accumulating in vesicles, and disappearing by ;45 s Having established that a5b1 integrin and Nrp1 are cointernalized at fibrillar adhesions, we wished to determine whether the integrin was then recycled from Nrp1-positive vesicles back to the plasma membrane. To address this, we aimed a pulse of 405-nm laser light at a ''single point'' corresponding to Nrp1-positive vesicles, leading to the immediate photoactivation of a5-PA-GFP integrin largely within the confines of these structures (Figure 8 and Video Figure 7. Rapid Turnover of a5b1 Integrin from Adhesion Sites into Nrp1-Positive Vesicles A representative NIH 3T3 cell was cotransfected with mNrp1-Cherry and a5-PA-GFP, photoactivated in TIRF, and observed in time-lapse epifluorescence microscopy. After photoactivation, a5-PA-GFP integrin (left panels) was found in elongated adhesive structures (i.e., fibrillar adhesions; arrows), where it colocalized with mNrp1-Cherry (middle panels), as evident in merged images as well (right panels). a5-PA-GFP integrin is also present in nascent vesicles near adhesive structures (arrowheads). Over time GFP fluorescence intensity (left panels) is constant in some vesicles (solid arrowheads), while it increases in others (empty arrowheads), that become progressively enriched in a5-PA-GFP integrin deriving from membrane adhesion sites. mNrp1-Cherry (middle panels) is also present in the same vesicles (solid and empty arrowheads), as visible in merged images (right panels). At each time point, lower panels are magnifications of the corresponding upper panels (see also Video S1). doi:10.1371/journal.pbio.1000025.g007 S3). During the following 80 s, fluorescence was lost from the photoactivated vesicle, and this was accompanied by a corresponding increase in integrin fluorescence at peripheral elongated structures that look like adhesion sites (Figure 8, red circle and white arrows). On the contrary, when the activating laser was aimed at a cell region devoid of Nrp1positive vesicles, little or no photoactivation occurred ( Figure   S5 and Video S4), indicating that the a5-PA-GFP fluorescence detected in Figure 8 was indeed at mNrp1-Cherry vesicles and not at the plasma membrane above and below them.
Taken together, these data indicate that a5b1 integrin and Nrp1 are cointernalized into intracellular vesicles, which are then rapidly returned or recycled to the plasma membrane. Interestingly, both the internalization and the recycling of Nrp1-associated a5b1 integrin occur at the site of adhesion to the ECM.

Cytoplasmic Domain of Nrp1 Elicits the Endocytosis of Active a5b1 Integrin via GIPC1 in ECs
In the eukaryotic early endocytic pathway, the small GTPase Rab5 is a rate-limiting component that regulates the entry of cargoes from the plasma membrane into the early endosome [52]. Hence, we analyzed the early endocytic steps of active a5b1 integrin in ECs cotransfected with mNrp1-mRFP and Rab5-GFP, which were incubated with the a5b1 integrin activation reporter mAb SNAKA51 for 30 min at 4 8C and then at 37 8C for different time points. Fluorescent confocal analysis indicated that, after 1-3 min of internalization at 37 8C, Nrp1 and active a5b1 integrin colocalized in early Rab5-positive vesicles near the EC plasma membrane ( Figure 9A). Accordingly, immunofluorescence analysis of endogenous endothelial proteins confirmed that hNrp1 and Rab5 colocalized in vesicles, many of which were located near adhesion sites ( Figure 9B, empty arrowheads), further supporting the view that Nrp1 can induce a5b1mediated adhesion by promoting the preferential internalization of its active conformation into Rab5-positive early endosomes and the ensuing recycling to newly forming cell-ECM contacts.
Next, to characterize the molecular mechanisms by which Nrp1 regulates the traffic of active a5b1 integrin, we evaluated the abilities of mNrp1 full-length and mutant constructs to rescue the integrin internalization defects that we observed in sihNrp1 ECs. Remarkably, only wild-type mNrp1, but neither mNrp1dSEA nor mNrp1dCy construct, was able to rescue the sihNrp1 EC defects in the endocytosis of active a5b1 integrin ( Figure 6D). Therefore, in ECs the SEA motif of Nrp1, which binds the endocytic adaptor GIPC1 [36], is mandatory for Nrp1 stimulation of cell adhesion to FN ( Figure 2C), endogenous FN fibrillogenesis ( Figure 2D-F), and active a5b1 integrin endocytosis ( Figure 6D).
The N-terminal portion of GIPC1 mediates its oligomerization, whereas its central PDZ domain can bind the Cterminal consensus S/T-X-U sequence of Nrp1 [36], the a5 integrin subunit [53], and the Rab5/Rab21 interactor protein APPL1 [54,55]. Thus, we theorized that as a result GIPC1 could support the Rab5-dependent early internalization of a5b1 integrin. To test this hypothesis, we silenced the expression of GIPC1 in human umbilical artery ECs by RNAi and examined its effect on a5b1 integrin traffic. Western blot analysis showed that, 96 h after the second transfection, GIPC1 protein, but not b-tubulin, was successfully silenced in sihGIPC1 ECs in comparison with control cells ( Figure 10A). Knockdown of GIPC1 in ECs dramatically reduced the amount of internalized total ( Figure 10C and 10D) and active ( Figure 10C and 10E) a5b1 integrin by ;70% throughout the whole internalization assay, suggesting that indeed the interaction of a5b1 integrin with GIPC1 is crucial for the endocytosis and the proper functioning of this integrin. Accordingly, short-term adhesion assays showed that, in comparison with control cells, sihGIPC1 ECs adhered poorly to FN ( Figure 10B) and much less efficiently assembled endogenous sFN into a fibrillar network ( Figure S1H) in comparison with cells transfected with siCtl ( Figure S1G). The latter defect was not due to a reduction in FN mRNA or protein levels as demonstrated by real-time RT-PCR ( Figure  S1B) and Western blotting ( Figure S1E). Hence, within Nrp1 the extracellular domain mediates the association with a5b1 integrin, and the C-terminal SEA sequence allows the binding to the endocytic adaptor GIPC1 that stimulates the internalization and traffic of active a5b1 integrin, finally promoting EC adhesion to FN and FN fibrillogenesis.

GIPC1 Interacting Motor Myo6 Promotes Active a5b1 Endocytosis, EC Adhesion to FN, and FN1 Gene Transcription
Because the C terminus of GIPC1 binds to the minus-enddirected motor myosin VI (Myo6) that has also been involved in endocytosis [56], we considered the hypothesis that Myo6 could cooperate with GIPC1 in promoting a5b1 integrin internalization. Interestingly, RNAi-mediated knockdown of Myo6 in human umbilical artery ECs ( Figure 10A) resulted in a significant (;70%) impairment of active a5b1 integrin internalization ( Figure 10F and 10H), whereas the total integrin pool was only mildly affected (;25%; Figure 10F and 10G). These data, together with the fact that sihMyo6 EC adhesion to FN was severely hampered (Figure 10B), indicate that Myo6 cooperates with GIPC1 in the regulation of active a5b1 integrin endocytosis.
Similarly to what we noticed after Nrp1 and GIPC1 knockdown, ECs in which Myo6 was silenced did not efficiently assemble an endogenous FN fibrillar network ( Figure S1I) in comparison with cells transfected with siCtl ( Figure S1G). However, differently from what we observed in sihNrp1 and sihGIPC1 ECs, the endogenous FN fibrillogenesis defect seen in sihMyo6 ECs was due to an inhibition of FN1 gene transcription mRNA ( Figure S1C), which associated to a significant reduction of FN protein levels as well ( Figure S1F). Indeed, in addition to its role in cytoplasmic transporting and anchoring, Myo6 is also present in the nucleus, where it promotes the RNA-polymerase-II-dependent transcription of active genes [57]. Here we identify the FN1 gene as a new Myo6 transcriptional target and downstream effector that can bolster EC adhesion and motility.

Discussion
Defects of developing blood vessels caused by Nrp1 gene knockdown in mice [23,24] are different from vascular malformations displayed by mice lacking either SEMA3A [16] or VEGF-A165 (Vegf-a 120/120 mice) [26]. Furthermore, it has been recently reported that Nrp1 is required for EC responses to both VEGF-A165 and VEGF-A121 isoforms, the latter being incapable of binding Nrp1 on the EC surface [58,59]. Therefore, it is conceivable that the vascular abnormalities of Nrp1 -/mice could be due at least in part to the disruption of a VEGF-A165/SEMA3A-independent Nrp1 function. a5b1 Integrin and its ligand FN are key players in vascular development [3]. The data reported here support a model in which Nrp1, through its cytoplasmic domain and independently of its activity as a SEMA3A and VEGF-A165 coreceptor, stimulates GIPC1/Myo6-dependent endocytosis and traffic of active a5b1 integrin, thus promoting EC adhesion to FN and FN fibrillogenesis.
In rescue experiments, where we reintroduced full-length and mutant murine Nrp1 constructs in human ECs in which endogenous hNrp1 was simultaneously knocked down by RNAi, we showed that EC adhesion to FN and polymerization of endogenous sFN into fibrils depend on the cytoplasmic domain of Nrp1, the C-terminal SEA motif representing the minimal sequence required to exert these functions. Importantly, as already shown for SEMA3A-elicited growth cone collapse in neurons [22], we found that the cytoplasmic domain of Nrp1 is instead dispensable for VEGF-A165 stimulation and SEMA3A inhibition of EC adhesion to FN. Moreover, by using two CHO cell clones differing in the expression of a5b1 integrin, we demonstrated that Nrp1 alone does not directly mediate adhesion to FN and that it requires a5b1 integrin. Therefore, we conclude that in ECs, independently of VEGF-A165 and SEMA3A, Nrp1 stimulates a5b1-mediated adhesion to FN and endogenous FN fibrillogenesis via its cytoplasmic SEA motif [36]. This motif, similar to the C-terminal SDA sequence of the a5 integrin subunit [53], selectively and specifically binds the PDZ domain of the homomultimeric endocytic adaptor GIPC1.
It is known that conformational activation of cell-surface integrins supports cell adhesion and spreading, whereas transition of integrins toward an inactive bent conformation causes cell de-adhesion and rounding up [1,60,61]. However, we observed that lack of Nrp1 does not result in the reduction of either active or total a5b1 integrin either at the cell surface or intracellularly. Rather, by combining biochemical analysis with conventional and TIRF/FLIM confocal microscopy, we found that at the plasma membrane Nrp1 is tightly associated with adhesion sites, where it physically interacts with a5b1 integrin. The complex formed between active a5b1 integrin and Nrp1 is then rapidly internalized into Rab5-positive endosomes in an Nrp1-dependent fashion. Interestingly, the integrin is then returned to the plasma membrane from Nrp1-containing vesicles, and this recycling event appears to be targeted to adhesive structures. In addition, although the extracellular domain of Nrp1 is sufficient for its interaction with a5b1 integrin, the C-terminal GIPC1-binding SEA sequence of Nrp1 is necessary for stimulating EC adhesion to FN. Accordingly, knocking down either GIPC1 or its interacting motor Myo6 results in a significant impairment of active a5b1 integrin endocytosis and EC adhesion to FN.
Taken together, our data indicate that, during EC adhesion and spreading on FN, Nrp1, through its extracellular domain, transiently interacts with active a5b1 integrin at adhesive sites and, via its cytoplasmic association with GIPC1, enhances the early endocytosis and the ensuing recycling of active a5b1 integrin to newly forming adhesion sites ( Figure  11A). It is therefore likely that fast cycles of endocytosis from and recycling to ECM adhesions of active a5b1 integrin could allow real-time optimization of adhesion during EC spreading on FN. These conclusions are in line with the recent findings by Ivaska and colleagues [8,9] that found how endocytosis of b1 integrins, in addition to their established role in directional migration [7], regulates cell adhesion and spreading as well. In particular, they reported that class V Rab GTPases (for review, see [52]) Rab21 and Rab5 directly bind to several integrin a subunits, a5 included, by interacting with the conserved membrane proximal region GFFKR, which interestingly has been previously implicated in conformational integrin activation [61]. It is thus conceivable that GIPC1 oligomers could favor a5b1 integrin endocytosis by bridging the a5 integrin subunit and the Rab5/Rab21 interactor APPL1, finally stabilizing the interaction between these small GTPases and a5b1 integrin. This could represent a main functional feature distinguishing a5b1 from other integrin heterodimers not interacting with GIPC1. Finally, the fact that by 2 min after activation a5-PA-GFP disappeared from preexisting adhesion sites into vesicles without a concomitant cell retraction suggests the existence of a steady endo-exocytic flow of (active) a5b1 integrins from and toward existing ECM adhesions as well ( Figure 11B). This mechanism could allow adherent cells to be always ready to rapidly exchange integrins among cell-ECM contacts in response to extracellular stimuli. Such a scenario is also compatible with a previous study by Ezratty and colleagues [62] and implies that disassembly of ECM adhesions could depend on an imbalance of endocytosis over recycling.
Our observation that Myo6 siRNA severely impairs EC adhesion to FN and results in a significant reduction in the internalization of active a5b1 integrin suggests that Myo6 cooperates with GIPC1 ( Figure 11A and 11B) and is compatible with the notion that Myo6 plays a role in the formation and transport of endocytic vesicles along F-actin microfilaments [56]. The decrease in FN mRNA that we noticed in sihMyo6 ECs is likely due to the lack of the transcriptional activity displayed by Myo6 in the nucleus [57] that could depend on a still not fully characterized actin-myosin-based mechanism of transcription [63,64]. Therefore, Myo6 can support EC adhesion and motility by promoting both active a5b1 integrin traffic ( Figure 11A and 11B) and FN1 gene transcription ( Figure 11C). Additionally, these findings can have significant implications for the biology of a5b1-expressing human carcinomas [51,65], in which Myo6 can be overexpressed and promote metastatic invasion [66][67][68].
In conclusion, we propose here that Nrp1, in addition to and independently of its role as coreceptor for VEGF-A165 and SEMA3A, stimulates through its cytoplasmic domain the spreading of ECs on FN by increasing the Rab5/GIPC1/Myo6dependent internalization of active a5b1 integrin. Nrp1 modulation of a5b1-mediated adhesion can play a causal role in the generation of angiogenesis defects observed in Nrp1 null mice. We anticipate that signaling pathways controlling Nrp1 expression in ECs could ultimately modulate the activity of a5b1 integrin. In particular, Nrp1 is a major target of the inhibitory Delta-like 4-Notch signaling pathway [69] that negatively regulates the formation of endothelial tip cells [10]. Higher expression of Nrp1 in tip ECs compared with that in stalk ECs of angiogenic sprouts could differentially modulate a5b1 integrin traffic, thus favoring tip cell adhesion and spreading on FN. Finally, both Nrp1 [70] and a5b1 integrin [71,72] are expressed in pericytes and vascular smooth muscle cells, which have been implicated in vascular remodeling by intussusceptive angiogenesis [73]. Further work is needed to assess whether Nrp1 is regulating a5b1 integrin function not only in ECs but also in pericytes and vascular smooth muscle cells.
Human plasma FN was from Tebu-bio. Human plasma vitronectin, Engelbreth-Holm-Swarm murine sarcoma laminin, and calf skin collagen type I were from Sigma-Aldrich. Recombinant human VEGF-A165 was from Invitrogen. Recombinant human SEMA3A and mouse Sema3F were from R&D Systems. Sulfo-NHS-SS-Biotin was from Pierce.
DNA constructs. Hemagglutinin-tagged mNrp1 deletion constructs controlled separately by a DAC 2000 card or a uniblitz shutter operated by MetaMorph (Molecular Devices). A filter block consisting of an E480SPX excitation filter, a FF 495 dichroic mirror, and an ET 525/50M emission filter was used for activation of a5-PA-GFP with the 405-nm laser. After activation the filter was manually changed to a green/red dual filter block (ET-GFP/mCherry from AHF Analysentechnik, Germany) to allow simultaneous time-lapse acquisition of activated a5-PA-GFP and mNrp1-Cherry using 473-and 561-nm excitation. A Multi-Spec dual emission splitter (Optical Insights, NM) with a 595-nm dichroic and two bandpass filters (510-565 nm for green and 605-655 nm for red) was used to separate both emissions. All cell imaging was performed with a Cascade 512F EMCCD camera (Photometrics UK). Confocal photoactivation. Localized activation of a5-PA-GFP in mNrp1-Cherry-positive vesicles was done on a FV 1000 Olympus confocal microscope, using two-channel imaging and a separate SIM scanner for 405-nm activation [51].     Figure 7, a5-PA-GFP was photoactivated in TIRF in NIH 3T3 cells and observed in time-lapse epifluorescence microscopy. Immediately after photoactivation, the a5-PA-GFP signal starts leaving the adhesive sites and accumulating in vesicles and disappears by ;45 s in about 50% of the adhesion sites and by ;115 s in the remaining ones (see also Video S2). Found at doi:10.1371/journal.pbio.1000025.sg004 (9.6 MB TIF). Figure S5. Upon Photoactivation in Areas outside of mNrp1-Cherry-Positive Vesicles, a5-PA-GFP Does Not Recycle Back to Membrane Adhesions NIH 3T3 cells were cotransfected with mNrp1-Cherry and a5-PA-GFP. a5-PA-GFP integrin fluorescence was then locally photoactivated in an mNrp1-Cherry-positive area devoid of vesicles (blue circle) and followed in time-lapse confocal microscopy. Fluorescence intensity was measured over time outside of Nrp1-positive vesicles (blue circle) and at the plasma membrane (red circle). The time-lapse plot (lower panel) shows that, under the same experimental conditions used in the experiment shown in Figure 8, little or no photoactivation of a5-PA-GFP occurred, and no fluorescence intensity increase was detected at the plasma membrane (see also Video S4). Found at doi:10.1371/journal.pbio.1000025.sg005 (9.4 MB TIF).
Video S4. Upon Photoactivation in Areas outside of mNrp1-Cherry-Positive Vesicles, a5-PA-GFP Does Not Recycle Back to Membrane Adhesions As in Figure S5, a representative NIH 3T3 cell was cotransfected with mNrp1-Cherry and a5-PA-GFP. a5-PA-GFP was locally photoactivated in an mNrp1-Cherry-positive area devoid of vesicles and followed by time-lapse confocal microscopy. Found at doi:10.1371/journal.pbio.1000025.sv004 (2.6 MB MOV).