Neuropilin 1 binds PDGF-D and is a co-receptor in PDGF-D–PDGFRβ signaling

Platelet-derived growth factor (PDGF)-D is the most recently discovered member of the PDGF family of growth factors. PDGFs fulfill important functions during embryonic development as well as in adult physiology (Andrae et al., 2008; Bergsten et al., 2001; LaRochelle et al., 2001).

The PDGF family consists of four ligands, PDGF-A to -D. They are secreted as disulfide-bonded homo- or heterodimers, denoted PDGF-AA, -AB, -BB, -CC and -DD, which bind to and signal through two tyrosine kinase receptors, PDGF receptor (PDGFR)α and PDGFRβ (Fredriksson et al., 2004; Heldin and Westermark, 1999). The PDGF ligands share a conserved growth factor domain and, in addition, PDGF-A and -B also contain C-terminal retention motifs important for their physiological function (Andrae et al., 2013; Lindblom et al., 2003). PDGF-A and PDGF-B are secreted in their active form, while secreted PDGF-C and -D polypeptide chains contain a N-terminal CUB (homology to complement components C1r/C1 s, Uegf, Bmp1) domain that needs to be proteolytically removed to enable receptor binding, and are thus produced and released as latent, inactive growth factors (Bergsten et al., 2001; Li et al., 2000; Ustach and Kim, 2005). The activation of PDGF-D is executed by serine proteases, such as urokinase-type plasminogen activator (uPA, also known as PLAU) or matriptase (Ehnman et al., 2009; Ustach et al., 2010).

PDGF-D is a ligand for PDGFRβ, suggesting that it has a functional overlap with PDGF-B, which also binds to the same receptor (Andrae et al., 2008; Bergsten et al., 2001). PDGF-D and PDGFRβ are both expressed in the vasculature. PDGFRβ is persistently expressed in pericytes (Lindahl et al., 1997), while PDGF-D expression changes in a spatial and temporal manner. The main sites of PDGF-D expression appear to be arterial endothelial cells, but it can also be expressed in vascular smooth muscle cells and presumably other subpopulations of mural cells (Gladh et al., 2016). PDGF-D, in contrast to PDGF-B, is dispensable for murine embryonic development (Gladh et al., 2016), but has been shown to, for example, stimulate the proliferation of PDGFRβ-positive cells in fibrotic processes and various cancers (Andrae et al., 2008; Cortez et al., 2016; Ponten et al., 2005; Wang et al., 2010) and is associated with cardiovascular disease (Hsu and Smith, 2012). However, the biology of PDGF-D remains largely enigmatic. Therefore, it is of interest to better understand the role of PDGF-D in physiological as well as pathological conditions.

Increasing evidence shows that PDGFR signaling can be modified by neuropilin 1 (NRP1), a multifunctional trans-membrane receptor protein (Kofler and Simons, 2016). NRP1 is well described as a co-receptor for members of the class 3 semaphorins (SEMA3) or vascular endothelial growth factors (VEGFs) (Koch, 2012; Zachary, 2014). Structurally, NRP1 is comprised of seven sub-domains whereof the first five are extracellular; two CUB domains (a1 and a2), two coagulation factor V/VIII domains (FV/VIII; b1 and b2) and a meprin, A5 μ-phosphatase domain (MAM; c). NRP1 contains only a short cytosolic tail with a PDZ-binding domain lacking internal signaling activity (Pellet-Many et al., 2008). The different ligand families bind to different sites of NRP1; SEMA3A binding requires the first three sub-domains of NRP1 (a1, a2, and b1) while binding of VEGF-A requires the b1 and b2 domains (Gu et al., 2002; Mamluk et al., 2002). Despite reports of NRP1 interaction with PDGFs (Ball et al., 2010; Banerjee et al., 2006; Cao et al., 2010; Pellet-Many et al., 2011), a specific binding site for PDGFs in NRP1 is not known. In addition, interactions of NRP1 with PDGFRα or PDGFRβ have been suggested (Pellet-Many et al., 2015), but its potential role in PDGF–PDGFR signaling remains elusive.

Here we demonstrate that PDGF-D directly binds to NRP1 and induces PDGFRβ–NRP1 complex formation. These NRP1–PDGFRβ complexes can form between proteins expressed in the same cell (in cis) or between these proteins in two adjacent cells (in trans). We identified the PDGF-D C-terminal arginine residue (Arg³⁷⁰), and the b1 and b2 domains of NRP1 to be crucial for the direct interaction of PDGF-D with NRP1. Furthermore, PDGF-D induces NRP1 translocation to endothelial cell junctions independently of PDGFRβ, possibly changing the availability of NRP1 for other endothelial signaling pathways. With these novel insights into the co-receptor function of NRP1 for PDGF-D–PDGFRβ signaling, we reveal differential signaling between PDGF-D and PDGF-B and provide the underlying mechanism. This knowledge may contribute to a better understanding of the physiological and pathological role(s) of PDGF-D as well as NRP1.

An amino acid sequence alignment of PDGF-B and PDGF-D (core) showed the conserved cysteine-knot structure and a high overall sequence similarity (Fig. 1A). Sequence alignment of PDGF-D with the VEGF-A isoform VEGF-A₁₂₁ and VEGF-E also showed the conserved cysteine-knot structure. Additionally, a common C-terminal arginine (Arg) residue was revealed (Fig. 1A, red box), which has previously been shown to be important for binding of specific VEGF-A isoforms and VEGF-E to NRP1 (Cebe-Suarez et al., 2008; Delcombel et al., 2013; Parker et al., 2012). This suggests a possible interaction of PDGF-D with NRP1, similar to that described for some members of the VEGF family. Importantly, PDGF-B lacks a C-terminal Arg residue, indicating a putative difference with respect to co-receptor engagement between PDGF-D and PDGF-B.

To test whether the C-terminal motif of PDGF-D enables an interaction with NRP1, we performed protein-binding studies. The results showed a dose-dependent binding of PDGF-D to NRP1, while PDGF-B, the other PDGFRβ ligand, showed no binding to NRP1 (Fig. 1B). Previously, it was shown that the FV/FVIII domains (b1 and b2) of NRP1 are important for the binding of VEGF-A to NRP1 (Gu et al., 2002; Mamluk et al., 2002). To assess whether PDGF-D binds in a similar way to NRP1 as does VEGF-A, we produced domain-deleted NRP1 mutant proteins and tested their binding to PDGF-D. NRP1 proteins lacking either the b1 or b2 domain (Δb1 or Δb2, respectively) showed impaired ability to bind to PDGF-D (Fig. 1C), while Δa1+7 and Δa2 bound with an efficiency comparable to wild-type (wt) NRP1, suggesting a similar binding mode of PDGF-D and VEGF-A to NRP1 (NRP1 deletion constructs are described in Gu et al., 2002). To demonstrate the requirement of the C-terminal Arg³⁷⁰ residue in PDGF-D for NRP1 binding we produced PDGF-D mutant proteins with either the truncation of the C-terminal Arg³⁷⁰ (-R), or the exchange of the C-terminal Arg³⁷⁰ for a glycine residue (R→G) and compared those to wt PDGF-D (Fig. S1A). Overexpression of the recombinant PDGF-D proteins in PDGFRβ overexpressing porcine aortic endothelial (PAE-Rβ) cells resulted in similar phosphorylation of PDGFRβ, demonstrating comparable binding and activation characteristics to PDGFRβ among the different mutant and wt PDGF-D proteins (Fig. S1B). To test the binding abilities of the mutant proteins to NRP1, we performed a ligand blot assay with recombinant NRP1 and concentrated supernatant from COS-1 cells overexpressing mutant PDGF-D proteins. Only recombinant PDGF-D with the wt C-terminal sequence exhibited binding to NRP1 (Fig. 1D), while removal of the C-terminal Arg³⁷⁰ (-R), or the exchange to Gly (R→G), resulted in the inability to bind NRP1. This confirmed that the C-terminal Arg³⁷⁰ residue is required for binding of PDGF-D to NRP1 and distinguishes between the co-receptor-binding characteristics of PDGF-D and PDGF-B.

The common sequence features hint at a similar binding mechanism for PDGF-D and VEGF-A to NRP1. As heparan sulfate proteoglycans (HSPG) or heparin can function as co-factors for VEGF-A–NRP1 interaction (Mamluk et al., 2002), we compared PDGF-D to VEGF-A isoforms that exhibit different binding characteristics to NRP1 and HSPG. We confirmed that heparin promoted the interaction of the VEGF-A isoforms VEGF-A₁₆₅ and VEGF-A₁₈₉ with NRP1 (Fig. 1E). Interestingly, the addition of heparin augmented the binding of NRP1 to PDGF-D core protein (the activated form), but not to full-length (FL-)PDGF-D (Fig. 1E). Heparin did not induce a binding of PDGF-B to NRP1 (Fig. 1E). These results suggest that PDGF-D, upon activation by removal of the CUB domain, can bind to heparin, which we could confirm by a direct binding assay of biotin-labeled heparin to immobilized PDGF-D (core and FL) and other growth factors (Fig. 1F). The PDGF-D core protein bound heparin, while FL-PDGF-D showed only weak binding to heparin, VEGF-A₁₂₁ failed to bind heparin and, as expected, VEGF-A₁₈₉ bound especially strong to heparin (Fig. 1F). PDGF-B proteins failed to bind heparin (Fig. 1F). The observation that only active PDGF-D could bind heparin led us to the hypothesis that cleavage of PDGF-D exposes its heparin-binding motif. Sequence analysis revealed an enrichment of positive charged amino acids (Arg or Lys) within the N-terminal sequence of active PDGF-D as a possible heparin-binding motif, located downstream (C-terminal) of the activating cleavage site (Ustach and Kim, 2005) (Fig. S1D). Modeling of the three-dimensional structure of core PDGF-D supported the hypothesis for a heparin-binding motif located between the activating cleavage site and the first cysteine in the cysteine-knot structure (Fig. S1D).

To investigate whether NRP1 might have relevance for the PDGF-D–PDGFRβ signaling cascade, we stimulated fibroblasts (BJ-hTERT) with PDGF-D or PDGF-B (Fig. 2). Immunofluorescence analysis of stimulated BJ-hTERT cells showed strong PDGFRβ clustering and internalization in response to both growth factors (Fig. 2Ab″,c″). However, only PDGF-D promoted the colocalization of PDGFRβ and NRP1 and concomitant clustering and internalization of NRP1 (Fig. 2Ab′-b′″). Quantification followed by Pearson's correlation showed that PDGF-D, but not PDGF-B, induced colocalization of PDGFRβ and NRP1, while both PDGF-D and PDGF-B caused co-clustering of PDGFRβ and EEA1 (early endosome antigen 1) (Fig. 2B), highlighting equally efficient activation and internalization of PDGFRβ in response to PDGF-D and PDGF-B.

Co-immunoprecipitation experiments further confirmed a PDGF-D-dependent interaction of PDGFRβ with NRP1 (Fig. 2C). BJ-hTERT cells stimulation with PDGF-D or PDGF-B followed by a NRP1 pull-down showed that PDGFRβ co-precipitated with NRP1 upon stimulation with PDGF-D, but not with PDGF-B. Both growth factors efficiently stimulated PDGFRβ-phosphorylation, as seen in the input samples, demonstrating no obvious difference between PDGF-D and PDGF-B with respect to PDGFRβ binding and activation (Fig. 2C). Nevertheless, in response to PDGF-D stimulation, NRP1 co-immunoprecipitated PDGFRβ was also phosphorylated, demonstrating that the NRP1-associated PDGFRβ is activated (Fig. 2D). This suggests that NRP1 can participate in active PDGF-D–PDGFRβ signaling.

Stimulation of BJ-hTERT cells followed by analysis of whole-cell lysates revealed that only PDGF-D-dependent internalization of NRP1 together with PDGFRβ results in degradation of NRP1 (Fig. 3), while both PDGF-D and PDGF-B induce degradation of PDGFRβ and activate downstream signaling via AKT1 (also known as PKB), and ERK-1 and ERK-2 (ERK-1/2, also known a MAPK3 and MAPK1, respectively). Despite the engagement of NRP1 by PDGF-D, no difference in the signaling characteristics between PDGF-D and PDGF-B could be observed (Fig. 3), which is in accordance with previous reports (Borkham-Kamphorst et al., 2015).

In order to test whether the C-terminal Arg³⁷⁰ of PDGF-D is required for the PDGF-D-induced colocalization of PDGFRβ and NRP1, we used proteins with mutated C-terminal Arg³⁷⁰ residues. Stimulation of BJ-hTERT cells with conditioned medium from COS-1 cells overexpressing wt PDGF-D efficiently induced PDGFRβ–NRP1 co-clustering, compared to conditioned medium from COS-1 cells transfected with the empty vector (Fig. 4Aa,b). In contrast, conditioned medium from COS-1 cells expressing PDGF-D proteins with either a truncated C-terminus, without the C-terminal Arg³⁷⁰ residue (-R) (Fig. 4Ac) or with an exchange of the C-terminal Arg³⁷⁰ residue to Gly (R→G) (Fig. 4Ad) did not induce PDGFRβ–NRP1 colocalization. Nevertheless, these mutant PDGF-D proteins could efficiently activate PDGFRβ, as indicated by perinuclear PDGFRβ accumulation (Fig. 4Ab–d). In line with these observations, we could show by separate siRNA-mediated knockdown experiments of either PDGFRB or NRP1 in BJ-hTERT cells, that PDGFRβ is required for PDGF-D-induced NRP1 clustering, internalization and degradation, while NRP1 is dispensable for efficient PDGFRβ activation, internalization and degradation (Fig. S2). This is different to the reported impact of NRP1 on the VEGF-A–VEGFR2 signaling cascade (Koch, 2012; Plein et al., 2014).

Based on the well-established function of NRP1 in regulation of VEGF-A–VEGFR2 signaling, we investigated whether the interaction of PDGF-D with NRP1 could affect VEGF-A–VEGFR2–NRP1 signaling mechanisms. Stimulation of endothelial cells [human umbilical vein endothelial cells (HUVECs); high NRP1 expression and PDGFRβ negative] with VEGF-A induced co-clustering of VEGFR2 with NRP1 (Fig. 5A,B). The receptor co-clustering was effectively prevented by pre-treatment of HUVECs with PDGF-D (Fig. 5Ac,B), but not PDGF-B (Fig. 5Ad,B). Interestingly, PDGF-D treatment of HUVECs resulted in NRP1 translocation to endothelial cell junctions, indicated by colocalization of NRP1 with the tight junction protein zona occludens 1 (ZO1, also known as TJP-1) (Fig. S3A) or the gap junction protein connexin 43 (CX43, also known as GJA1) (Fig. S3B). PDGF-D stimulation rendered translocation of NRP1 exclusively to endothelial cell–cell junctions, while VEGF-A mainly induced NRP1 internalization, as seen by colocalization with EEA1-positive endosomes, presumably in complex with VEGFR2 (Fig. S3C). These observations raise the possibility that PDGF-D directly modulates the cellular localization of NRP1 in endothelial cells independently of PDGFRβ. Indeed, stimulation of HUVEC with PDGF-D followed by NRP1 immunoprecipitation revealed that PDGF-D could disrupt the VEGFR2–NRP1 complexes that were apparently present under basal cell culture conditions (Fig. 5C,D). This data suggests that PDGF-D can function as a modulator of VEGF-A–VEGFR2 signaling by influencing NRP1 distribution and availability.

To implement our observation that PDGF-D binds NRP1 and thereby possibly influences PDGFRβ as well as VEGFR2 signaling pathways, we compared the effect of PDGF-D versus PDGF-B in three-dimensional ex vivo models of sprouting angiogenesis, mediating analysis of pericyte and endothelial cell behavior. Embryo explants or aortic ring cultures from mice with endothelial cell- and pericyte-specific endogenous fluorescence (GFP–GPI and NG2DsRed, respectively), allowing for simultaneous imaging of endothelial cells and pericytes, were investigated (Fig. 6). In embryo explant cultures, PDGF-D and PDGF-B increased the sprouting speed in a comparable manner (Fig. 6A,B). Both PDGF-D and PDGF-B, led to detachment of pericytes from the endothelial sprouts. However, PDGF-D also mediated a relative increase in pericyte coverage, which was not observed with PDGF-B (Fig. 6B). This data was confirmed in aortic ring cultures, obtaining a similar outcome (Fig. 6C). More in-depth analysis of pericyte behavior revealed that the migration speed of individual pericytes was increased in response to stimulation with PDGF-D, as well as PDGF-B, suggesting a similar magnitude of cellular activation. However, when separately analyzing either pericytes that were in contact with the endothelial sprout or those that had no contact with endothelial cells, the migration speed differed significantly in case of PDGF-D stimulation, but not when stimulated with PDGF-B (Fig. 6C). The data may suggest that PDGF-D–PDGFRβ signal transduction governed by complex formation with NRP1 in trans, might contribute to intercellular communication between endothelial cells and pericytes that in turn results in differential signaling outcomes from PDGF-D- or PDGF-B-induced PDGFRβ activation.

The data from the ex vivo experiments showed a differential effect of PDGF-D stimulation on pericyte behavior depending on their contact with endothelial cells. To evaluate the possibility that PDGF-D would induce PDGFRβ–NRP1 interaction between pericytes and endothelial cells (in trans) we applied in vitro co-culture models, where we stimulated PAE-Rβ and HUVEC, or human micro-vascular pericyte (HBVP) and HUVEC co-cultures with PDGF-D or PDGF-B. In this experimental set-up HUVECs served as the source of NRP1, while PAE-Rβ or HBVPs were the source for PDGFRβ. PAE-Rβ and HUVEC co-culture stimulation with PDGF-D, or PDGF-B, followed by NRP1 immunoprecipitation revealed the potential of PDGF-D to also induce clustering of PDGFRβ together with NRP1 in trans (Fig. 7A). Similar to the BJ-hTERT cell experiments, PDGF-D promoted the co-immunoprecipitation of PDGFRβ with NRP1, and activation of PDGFRβ. In contrast, PDGF-B stimulation efficiently induced the phosphorylation of PDGFRβ, but did not induce the association of PDGFRβ with NRP1 (Fig. 7A). Importantly, only in the co-culture of PAE-Rβ cells together with HUVECs was a co-immunoprecipitation of PDGFRβ with NRP1 observed, highlighting the trans-interaction of PAE-Rβ PDGFRβ with HUVEC NRP1. Likewise, staining of HBVP–HUVEC co-cultures for PDGFRβ and NRP1 showed PDGFRβ–NRP1 co-clustering in response to PDGF-D stimulation, which appeared to be especially pronounced in HBVP protrusions in contact with HUVECs (Fig. 7B,C). Stimulation with PDGF-B did not induce PDGFRβ–NRP1 colocalization in HBVP–HUVEC co-cultures (Fig. S4A). Additionally, the NRP1 translocation to cell–cell junctions between endothelial cells could also be seen in the PDGF-D-stimulated HBVP–HUVEC co-culture (Fig. S4Ae). As expected, PDGF-D only induced downstream signaling in PDGFRβ-positive HBVPs (alone or in co-culture with HUVECs) but not in HUVEC monocultures (Fig. S4B). This suggests an exclusive PDGF-D-specific communication route between endothelial cells and pericytes mediated by the combined interaction of PDGF-D with PDGFRβ and NRP1.

Most studies on PDGF-D do not compare PDGF-D and PDGF-B biology, or when compared, both growth factors display rather identical signaling outcomes (Borkham-Kamphorst et al., 2015), leaving PDGF-D-specific functions undiscovered. This prompted us to search for features that would differentiate PDGF-D from PDGF-B.

Here, we show that PDGF-D binds to NRP1 and delineate the structural basis underlying this interaction. A required attribute found in PDGF-D, but not present in PDGF-B, is the C-terminus with the Arg³⁷⁰ residue, which is similarly seen in NRP1-binding VEGF-A isoforms. We further demonstrate that PDGF-D binding to NRP1 resembles that of VEGF-A. As for VEGF-A₁₆₅ (Gu et al., 2002; Mamluk et al., 2002), the b1 and b2 domains of NRP1 appear to be important for the interaction of PDGF-D with NRP1. Parker et al. (2012) characterized the binding pocket of NRP1 for the C-terminus of exon 8a-containing VEGF-A isoforms, which is located within the NRP1-b1 subdomain, and subsequent animal studies employing NRP1 mutations could confirm the lack of VEGF-A–NRP1 interaction when a single amino acid of this binding pocket was mutated (Fantin et al., 2014; Gelfand et al., 2014). We suggest that PDGF-D binds directly to NRP1, employing the same binding pocket as VEGF-A (containing the exon 8a-encoded C-terminus) and this interaction together with the NRP1-b2 domain confers correct binding of PDGF-D to NRP1 (Delcombel et al., 2013; Parker et al., 2012).

Others have reported that NRP1 can modulate PDGF-A- and PDGF-B-induced receptor signaling (Ball et al., 2010; Pellet-Many et al., 2011). However, whether these effects of NRP1 are direct or indirect remain open questions. Post-translational modification of NRP1 with chondroitin sulfate has been described to be an important aspect of NRP1-dependent modulation of PDGF-A- and PDGF-B-induced downstream signaling (Pellet-Many et al., 2011). For the protein-binding studies, we used non-modified recombinant NRP1 and thus can conclude that post-translational modification of NRP1 is not important for the interaction of PDGF-D with NRP1, although this could explain why we did not observe an interaction of PDGF-B with NRP1, as was suggested by earlier studies.

Matrix retention has previously been shown for activated (core) PDGF-D, while unprocessed FL-PDGF-D remains freely diffusible (Ehnman et al., 2009; Huang and Kim, 2015). Accordingly, we show that heparin only promotes the binding of active, core PDGF-D to NRP1, but not that of FL-PDGF-D. Likely, the heparin-binding domain is covered by the CUB domain and thus inaccessible in FL-PDGF-D. We suggest that the heparin-binding domain in PDGF-D is located between the CUB domain and the growth factor domain. Hence, proteolytic cleavage of FL-PDGF-D does not only enable PDGFRβ engagement, but potentially also determines spatial deposition of PDGF-D within the extracellular matrix. Because FL-PDGF-D is able to bind to NRP1, a cell surface retention by this interaction is possible and a contribution of NRP1 to the proteolytic processing of FL-PDGF-D awaits investigation.

Interestingly, in endothelial cells that express NRP1 but not PDGFRs, PDGF-D stimulation induces a robust translocation of NRP1 to cell–cell junctions. Thereby, PDGF-D has the potential to sequester NRP1 away from the available pool for other signaling pathways, such as VEGF-A–VEGFR2 (Koch, 2012), or transforming growth factor (TGF)-β to activin A receptor type II like 1 or TGF-β receptor I (Aspalter et al., 2015; Kofler and Simons, 2016), suggesting a possible regulatory role of PDGF-D that would be dependent on availability as well as concentration of the competing NRP1 ligands. The localization of NRP1 at endothelial cell–cell junctions was recently described to induce endothelial cell permeability in a VEGFR2-independent fashion (Roth et al., 2016). The precise mechanism for this VEGFR2-independent function of NRP1 is still incompletely understood, and whether the PDGF-D-induced translocation of NRP1 is associated with increased permeability remains to be investigated. Our observation that in fibroblasts devoid of PDGFRβ, NRP1 did not translocate to cell–cell contacts suggests another yet undefined endothelial factor that promotes NRP1 trafficking and cell junction localization after PDGF-D binding.

NRP1 recruitment to VEGFR2 signaling complexes modulates intracellular sorting and trafficking of VEGFR2, thereby altering the signaling outcome of VEGF-A–VEGFR2 complexes (Ballmer-Hofer et al., 2011; Lanahan et al., 2010). Similar mechanisms have been described for PDGFRβ (Hellberg et al., 2009; Karlsson et al., 2006), and NRP1, as well as NRP2, has been suggested to be involved in PDGFR downstream signal conduction (Pellet-Many et al., 2015). Our initial analysis did, however, not reveal changes in PDGF-D–PDGFRβ downstream signaling dependent on the presence or absence of NRP1. This is not surprising since clear differences between PDGF-D- and PDGF-B-induced PDGFRβ downstream signaling have not been described (Borkham-Kamphorst et al., 2015). This might be different when PDGFRβ and NRP1 interact in trans, between two adjacent cells or even two different cell types. In the trans situation, physical hindrance would prevent internalization of both PDGFRβ and NRP1 and would lock them at the cell surface, possibly changing signaling kinetics from fast and transient to slow and persistent, as was recently shown for VEGFR2 signaling when NRP1 is presented in trans (Koch et al., 2014). Such a scenario would result in a spatially restricted change of PDGFRβ downstream signaling in a PDGF-D-specific manner.

Pdgfd is expressed during murine embryonic development and into adulthood. It has a mainly arterial vascular expression pattern (Gladh et al., 2016), suggesting that its interaction with NRP1 and activation of PDGFRβ might be involved in spatial regulatory mechanisms in mural cells or between endothelial cells and adjacent pericytes. Many fundamental vascular processes depend on proper endothelial-to-pericyte communication, such as basement membrane formation, vascular stabilization, maintenance and remodeling (Armulik et al., 2011; Geevarghese and Herman, 2014). The need for PDGF-B signaling through PDGFRβ in angiogenesis and vascular homeostasis is well established, as is the role of NRP1 (Andrae et al., 2008; Koch, 2012; Zachary, 2014). Here, we show in an ex vivo system sensitive for intercellular crosstalk, that PDGF-D induces a pericyte behavior different from PDGF-B. We thus provide a first clue that PDGF-D–PDGFRβ signaling can be involved in intercellular crosstalk between endothelial cells and pericytes through the formation of PDGFRβ and NRP1 complexes. Detailed investigation of the PDGFRβ–NRP1 interaction in in vitro co-culture systems confirmed that occurrence of PDGFRβ–NRP1 complexes was primarily in response to PDGF-D stimulation. These clusters of PDGFRβ and NRP1 were predominantly found in pericyte protrusions that are in contact with endothelial cells, which could hint towards a potential spatially restricted signaling hub for intercellular crosstalk downstream of PDGF-D. This idea is further strengthened by our previous observation showing that NG2⁺ cells display a mispatterning phenotype in the adult heart of Pdgfd^−/− mice (Gladh et al., 2016). Thus, PDGF-D–PDGFRβ signaling by recruitment of NRP1 may actively modulate endothelial-cell–pericyte communication in temporally and spatially restricted events. Recently, we could show that PDGF-D, but not PDGF-B, promotes neuroendocrine tumor growth by activation of a subpopulation of PDGFRβ-positive tumor cells (Cortez et al., 2016). NRP1 is a good candidate to be a co-receptor for PDGF-D-induced PDGFRβ-signaling in this context, a hypothesis that waits to be investigated.

Taken together, we demonstrate that PDGF-D directly binds to NRP1, thereby altering the cellular distribution and availability of NRP1 for other signaling pathways, or recruiting NRP1 as a co-receptor for PDGF-D–PDGFRβ signaling (summarized in Fig. 8). These novel findings are the basis for further investigations to understand the biology of PDGF-D as well as NRP1, which have implications for clinical conditions, such as cardiovascular disorders or cancer.

Commercially available recombinant proteins, human PDGF-D (1159-SB), human PDGF-B (220-BB), human VEGF-A₁₆₅ (293-VE) human VEGF-A₁₈₉ (8147-VE), and mouse NRP1 (5994-N1) were purchased from R&D Systems (Minneapolis, MN). Recombinant human VEGF-A₁₂₁ (100-20A), VEGF-A₁₆₅ (100-20) and PDGF-B (100-14B) were purchased from Peprotech (Rocky Hill, NJ) Recombinant human full-length (FL-)PDGF-D (TP760442) was purchased from OriGene Technologies Inc. (Rockville, MD). Biotinylated heparin from porcine intestinal mucosa was purchased from Merck Millipore (Darmstadt, Germany). Heparin (unfractionated) was purchased from Sigma-Aldrich (Schnelldorf, Germany). Recombinant rat NRP1 deletion constructs (Gu et al., 2002) were transfected into PAE cells using Lipofectamine 2000 reagent (Thermo Fisher Scientific). After 48 h, cells were harvested and lysed in lysis buffer (see Immunoprecipitation) to prepare protein solutions. Commercially available primary antibodies are presented in Table S1, and were used according to manufacturers' instructions.

For the alignment of protein sequences, the EMBL-EBI online portal (www.ebi.ac.uk/tools/msa/) for pairwise or multiple sequence alignments was utilized. The Clustal Omega operation was used to generate the sequence alignments. Sequences from mature human growth factor proteins, without signal sequence and post transcriptionally removed latent peptide sequences were used for the study.

The amino acid sequence of core PDGF-D was loaded into the online portal of Swiss model (https://swissmodel.expasy.org/interactive) to model PDGF-D structure using VEGF-B (PDB ID: 2c7wA) as template. Graphic representation was prepared using Swiss Pdbviewer software (http://spdbv.vital-it.ch/). For surface electrostatic potential visualization, the color indication was set to 0=red, 1=white and 2=blue.

Standard in vitro cell culture conditions were applied. In brief, immortalized human foreskin fibroblasts (BJ-hTERT; ATCC, CRL-4001) and porcine aortic endothelial cells (PAE) stably transfected with human PDGFRβ (PAE-Rβ) (Bergsten et al., 2001) were cultured using Dulbecco's modified Eagle's medium (DMEM) with F12 (Thermo Fischer Scientific Inc., Waltham, MA) supplemented with 10 U/ml penicillin, 10 µg/ml streptomycin and 10% fetal calf serum (FCS) (all three Thermo Fisher Scientific Inc.). African green monkey kidney fibroblasts (COS-1, ATCC, CRL-1650) were cultured using DMEM (Thermo Fischer Scientific Inc.) supplemented with 10 U/ml penicillin, 10 µg/ml streptomycin and 10% FCS. Human umbilical vein endothelial cells (HUVECs; PromoCell, Heidelberg, Germany) and human brain micro-vascular pericytes (HBVPs; ScienCell Research Laboratories, Carlsbad, CA) were cultured on cell culture plates coated with gelatin (1% w/v in PBS) (Sigma Aldrich) and in endothelial cell growth medium 2 (EGM2) (PromoCell) including growth supplements (2.5% FCS) and 10 U/ml penicillin, 10 µg/ml streptomycin.

BJ-hTERT cells were cultured as described above (see Cell culture). BJ-hTERT were transfected with control siRNAs (siCONTROL) or siRNA against PDGFRβ or NRP1 (siPDGFRβ or siNRP1, respectively) (all Santa Cruz Biotechnology Inc.) using Lipofectamine RNAiMAX (Thermo Fisher Scientific) according to the manufacturer's recommendations. Following overnight incubation with the respective siRNA, the cells were plated for the specific assays, and further processed as described in the Direct cell stimulation or Immunofluorescence analysis section.

96-well microtiter plates were coated with 50 µl of PDGF-D or PDGF-B solution at increasing amounts (0.2–25 ng/well) in 100 mM sodium carbonate buffer (pH 9.5) (Sigma Aldrich) for 16 h at 4°C. Wells were washed and non-specific binding sites blocked with 3% BSA/TBS. Recombinant mouse NRP1 (1 µg/ml) or rat NRP1 deletion mutant proteins (Gu et al., 2002) were allowed to bind for 1 h at room temperature. To test the effects of heparin on the NRP1 binding to growth factors, 10 µg/ml heparin (Sigma Aldrich) was added simultaneously with soluble mouse NRP1 to one half of the growth factor-coated wells. Bound NRP1 was detected using goat anti-mouse and rat NRP1 antibody (cat. no. AF566) or rabbit anti-NRP1 (clone no. EPR3113, cat. no. ab81321), and the corresponding horseradish peroxidase (HRP)-tagged secondary antibody (GE Healthcare, Uppsala, Sweden) and immunopure TMB substrate kit (Thermo Fisher Scientific Inc.). For direct binding of biotinylated heparin to growth factors, microtiter plates were coated with 50 µl of growth factor proteins (25 ng/well) and blocked as described above. Biotinylated heparin was added to the wells and bound heparin was detected using streptavidin conjugated to HRP and a TMB substrate kit as described above. A plate reader (POLARstar Omega; BMG Labtec GmbH, Ortenberg, Germany) was used to measure absorbance at 450 nm.

PDGF-D core protein cDNA (Ser²⁵⁰–Arg³⁷⁰) was amplified from vector plasmids (Ponten et al., 2005) using specific primers to introduce single base mutations in the C-terminal region. The common forward primer 5′-TAGGATCCTCATACCATGATCGGAAG-3′ together with three different reverse primers to amplify PDGF-D cDNA containing the wild type (wt) sequence 5′-TAGAATTCTTATCGAGGTGGTCTTGAGC-3′, PDGF-D with a truncated C-terminal Arg³⁷⁰ (-R) sequence 5′-TAGAATTCTTATCAAGGTGGTCTTGAGCT-3′ and PDGF-D with a substitution of the C-terminal arginine to glycine (R→G) 5′-TAGAATTCTTATCCAGGTGGTCTTGAGC-3′ were used for construct amplification. The amplified constructs were cut and ligated into a modified pSecTag vector, using BamHI and EcoRI restriction enzymes, FastAP and T4 ligase (all Thermo Fisher Scientific Inc.). Ligated plasmid vectors were transformed into TOP10 E. coli, and amplified and purified with the respective preparation kits (Macherey-Nagel, Dueren, Germany) according to the manufacturer's recommendations. Plasmids containing the PDGF-D constructs were sent for sequencing using Sanger sequencing (GATC Biotech AG, Constance, Germany). Obtained sequences were visualized using 4Peaks software (by A. Griekspoor and Tom Groothuis, nucleobytes.com/4peaks/) and analyzed by using Sequence analysis software (www.informagen.com/SA/) for the correct base sequence. PDGF-D mutant protein constructs were transfected into PAE cells stably transfected with human PDGFRβ (PAE-Rβ) (Bergsten et al., 2001) or COS-1 cells. Cell lysates were prepared by direct addition of SDS sample buffer (63 mM Tris-HCl pH 6.8, 10% glycerol and 2% SDS). Conditioned medium was collected for direct application, or concentrated using 30 kDa cut-off concentrating columns (Merck Chemical and Life Sciences AB, Solna, Sweden).

Recombinant human PDGF-D (R&D Systems) or concentrated supernatant from COS-1 cells overexpressing recombinant PDGF-D proteins were separated on a 10% SDS-PAGE gel and transferred onto a nitrocellulose membrane (both Thermo Fisher Scientific Inc.). After blocking of non-specific binding sites with 3% BSA/TBS-T the membrane was incubated with recombinant mouse NRP1 solution (1 µg/ml). Bound NRP1 was detected by incubation with goat anti-mouse and -rat antibody (cat. no. AF566) followed by the corresponding HRP-tagged secondary antibody. As loading control, PDGF-D proteins were detected by sequential incubation with goat anti-PDGF-D antibody (cat. no. AF1159) followed by corresponding HRP-tagged secondary antibody, and development was performed as described below (Direct cell stimulation).

BJ-hTERT cells were cultured in 24-well cell culture plates, serum-deprived for 4–6 h in DMEM/F12 with 0.2% FCS prior to stimulation with PDGF-D or PDGF-B. Cells were stimulated for 10 to 120 min at 37°C. Thereafter, cells were washed once with PBS and directly lysed in SDS sample buffer. Proteins were separated on a 4–12% SDS-PAGE gradient gel, transferred to nitrocellulose membranes and detected by sequential incubation with primary antibodies and corresponding HRP-tagged secondary antibodies according to the manufacturer's recommendations. For repeated detection of different proteins, the membranes were washed with stripping buffers (1: 200 mM glycine, 0.1% Tween 20, pH 2.8; and 2: 200 mM glycine, 500 mM NaCl, 0.1 % Tween 20 pH 9.6; Sigma Aldrich) and re-probed with new primary and secondary antibodies. Signals were detected by using an ECLprime reagent kit (GE Healthcare) and pictures were taken with the imaging system FluoroChem Q (Alpha Innotec GmbH, Kasendorf, Germany) and graphically handled using Image J software. All images were appropriately adjusted in brightness and contrast and processed with ‘despeckle’ noise reduction and ‘smoothen’ operations.

BJ-hTERT cells were cultured on gelatin-coated glass coverslips in six-well cell culture plates, and serum deprived for 4–6 h in DMEM/F12 with 0.2% FCS prior to stimulation with PDGF-D or PDGF-B (each 20 ng/ml). Cells were stimulated for 20 min at 37°C, thereafter medium was removed and cells fixed with ice-cold 4% paraformaldehyde (PFA) in PBS for 10 min at room temperature. Likewise, HUVECs and HBVPs were cultured alone or together on gelatin-coated glass coverslips in six-well cell culture plates, serum deprived for 4–6 h in EGM2 with 0.5% FCS and stimulated as described above. After fixation with 4% PFA in PBS cells were blocked with blocking buffer [serum-free protein blocking solution (DAKO, Glastrop Denmark)], supplemented with 0.2% Triton X-100 (Sigma-Aldrich) for >1 h at room temperature. Thereafter, cells were sequentially incubated with primary antibodies and corresponding fluorescently tagged secondary antibodies (Molecular Probes, Eugene, OR) in blocking buffer, according to the manufacturer's recommendations. Cells were mounted with ProLong Gold mounting medium containing 4,6-diamidino-2-phenylindole (DAPI; Molecular Probes). Evaluation of the staining and image capture of immunofluorescence micrographs was carried out by using an upright laser scanning, confocal microscope (LSM 700; Carl Zeiss GmbH, Göttingen, Germany).

Cells were cultured as described above. After serum deprivation, cells were stimulated with PDGF-D or PDGF-B (each 25 ng/ml) for 5 min at 37°C, transferred to 4°C and stimulation continued for 1 h. Cells were washed with PBS, and cross linker (DTSSP; Thermo Fisher Scientific Inc.) was added at a concentration of 2 mM for 2 h at 4°C. Excess cross linker was blocked by washing with 100 mM glycine in PBS (Sigma-Aldrich) for 15 min at 4°C. Thereafter, cells were lysed with lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% Ipegal CA-630 and 1% Triton X-100), supplemented with PhosSTOP phosphatase inhibitor and Complete protease inhibitor cocktail (both Roche Diagnostics, Indianapolis, IN) for 1 h at 4°C, scraped and unresolved cell-debris was pelleted by centrifugation for 10 min at 15,000 g. For immunoprecipitation, goat anti-human NRP1 antibody (C19) was pre-bound to protein-G–Sepharose (GE Healthcare, Uppsala, Sweden) at a concentration of 2 µg/sample, for >30 min at 4°C. Antibody-bound Sepharose was then added to cell lysates and incubated with end-over-end rotation for 16 h at 4°C. Thereafter, the Sepharose beads were washed with TBS, then the bound proteins were denatured in β-mercaptoethanol (Sigma Aldrich) containing SDS sample buffer, and SDS-PAGE was performed according to manufacturer's recommendations, and the proteins were detected by western blotting as described above (Direct cell stimulation).

Mice expressing membrane-bound GFP (Tg(CAG-GFP*)Hadj/J (CAG:GFP-GPI) (Rhee et al., 2006) and mice expressing dsRED under the control of the Cspg4 (gene of NG2) promoter NG2DsRedBAC tg (Zhu et al., 2008) were cross bred to generate double-positive reporter mice (hereafter GFP–GPI/NG2DsRed). All mice had always ad libitum access to water and chow. All mouse strains were housed in accordance with local guidelines and regulations for animal research (Swedish Animal Welfare Board, Stockholm, Sweden). For the sprouting assay using embryo explants, embryos at embryonic day (E)13.5 were isolated and the limbs, head and lower body were removed. The upper body was cut into 1 mm³ pieces and embedded between two layers of rat-tail collagen type I (Thermo Fischer Scientific Inc.) and cultivated in DMEM supplemented with 10 U/ml penicillin, 10 µg/ml streptomycin and 10% FCS (all three Thermo Fisher Scientific Inc.) including 30 ng/ml VEGF-A₁₆₅ (Peprotech). For specific growth factor stimulation, the medium was changed to DMEM with 2.5% FCS and 30 ng/ml VEGF-A, 30 ng/ml PDGF-D (R&D Systems) or 30 ng/ml PDGF-B (Peprotech). After live imaging, samples were fixed with 4% PFA in PBS. For aortic ring sprouting assay, mice were killed between 4 to 6 weeks of age and thoracic aortas were isolated, cleaned from connective tissue, cut into 1 mm rings, embedded and cultivated as described above. For specific growth factor stimulation, the medium was changed to DMEM with 10% FCS and 30 ng/ml VEGF-A, 10 µg/ml control IgG (BioLegend, San Diego, CA), 30 ng/ml PDGF-D (R&D Systems) or 30 ng/ml PDGF-B (Peprotech).

For the embryo explant culture assay, 13 z-slices were acquired at each position every 23 min using an inverted laser-scanning confocal microscope (SP8, Leica Microsystems GmbH, Wetzlar, Germany), with a 10× (NA 0.3, air) objective and maintained at 37°C and 5% CO₂ with a humidifier system. For the aortic ring culture assay, 75 z-slices were acquired at each position every 20 min using the same microscope set-up with a 20× (NA 0.75, water immersion) objective.

Vessel sprouting speed was measured as the vessel elongation divided by time. Pericyte coverage was measured as the number of pericytes divided by vessel length at the start and the end frame of live imaging. The changes in pericyte coverage were plotted as percentages. Pericyte migration speed was calculated by tracking of individual pericytes (NG2⁺ cells) over the period of live imaging. Pericytes with a defined distance to the endothelial sprout were considered as detached.

All presented experiments have been performed at least twice, except for embryo explant or aortic ring sprouting assays, which each have been performed once. Results from one representative experiment or the mean±s.d. of the replicate experiments are shown as indicated in the respective figure legend. Specific P-values and statistical tests used are indicated in the respective figure legend. A P-value<0.05 was considered as statistically significant.

The authors would like to thank Associate Professor Chenghua Gu (Harvard Medical School, MA, USA), and Alessandro Mega and Professor Arne Östman (Karolinska Institutet, Stockholm, Sweden) for their generous gifts of crucial reagents. The authors also like to thank Dr Yi Jin (Karolinska Institutet, Stockholm, Sweden) for his excellent technical assistance on time-lapse imaging and Zacharias Iredahl for his technical assistance.