Neuron Restrictive Silencer Factor NRSF/REST Is a Transcriptional Repressor of Neuropilin-1 and Diminishes the Ability of Semaphorin 3A to Inhibit Keratinocyte Migration*

Neuropilin-1 (NRP1) is expressed by endothelial cells and neurons and serves as a receptor for both vascular endothelial growth factor (VEGF), an angiogenesis factor, and semaphorin 3A (Sema3A), a mediator of axonal guidance. We show here that NRP1 is also expressed in keratinocytes in vitro and in vivo. However, nothing has been reported about the regulation or function of keratinocyte NRP1. Using NRP1 promoter constructs in HaCaT cells, a keratinocyte cell line, we could demonstrate that a neuron restrictive silencer element (NRSE) was implicated in transcriptional repression of the NRP1 gene. Electrophoretic mobility shift assays demonstrated that the neuron restrictive silencer factor (NRSF) binds to NRSE. Overexpression of NRSF in HaCaT cells decreased NRP1 RNA and protein, whereas a dominant negative NRSF increased NRP1. Furthermore, the histone deacetylase inhibitor trichostatin A, an inhibitor of NRSF silencing activity, also increased NRP1 levels. NRP2 expression was not affected. Epidermal growth factor (EGF) and heparin-binding EGF-like growth factor (HB-EGF) strongly up-regulated NRP1 expression, concomitant with down-regulation of NRSF. Other keratinocyte mitogens such as keratinocyte growth factor (KGF) had no effect. To address function, HaCaT cells were exposed to two NRP1 ligands, VEGF165 and Sema3A. Neither had an effect on proliferation, whereas Sema3A, but not VEGF165, inhibited cell migration. Down-regulation of NRP1 by NRSF overexpression reduced Sema3A activity. It was concluded that NRSF is a transcription factor that silences NRP1 expression and thereby diminishes the Sema3A mediated inhibition of HaCaT keratinocyte migration.


Neuropilin-1 (NRP1) is expressed by endothelial cells and neu-
rons and serves as a receptor for both vascular endothelial growth factor (VEGF), an angiogenesis factor, and semaphorin 3A (Sema3A), a mediator of axonal guidance. We show here that NRP1 is also expressed in keratinocytes in vitro and in vivo. However, nothing has been reported about the regulation or function of keratinocyte NRP1. Using NRP1 promoter constructs in HaCaT cells, a keratinocyte cell line, we could demonstrate that a neuron restrictive silencer element (NRSE) was implicated in transcriptional repression of the NRP1 gene. Electrophoretic mobility shift assays demonstrated that the neuron restrictive silencer factor (NRSF) binds to NRSE. Overexpression of NRSF in HaCaT cells decreased NRP1 RNA and protein, whereas a dominant negative NRSF increased NRP1. Furthermore, the histone deacetylase inhibitor trichostatin A, an inhibitor of NRSF silencing activity, also increased NRP1 levels. NRP2 expression was not affected. Epidermal growth factor (EGF) and heparin-binding EGF-like growth factor (HB-EGF) strongly up-regulated NRP1 expression, concomitant with downregulation of NRSF. Other keratinocyte mitogens such as keratinocyte growth factor (KGF) had no effect. To address function, HaCaT cells were exposed to two NRP1 ligands, VEGF 165 and Sema3A. Neither had an effect on proliferation, whereas Sema3A, but not VEGF 165 , inhibited cell migration. Down-regulation of NRP1 by NRSF overexpression reduced Sema3A activity. It was concluded that NRSF is a transcription factor that silences NRP1 expression and thereby diminishes the Sema3A mediated inhibition of HaCaT keratinocyte migration.
Neuropilins (NRPs) 2 are 130-kDa type I transmembrane receptors that mediate neuronal guidance, angiogenesis, and the immune response (1,2). There are two genes, NRP1 and NRP2, with some distinct and some overlapping properties (3). NRPs are expressed by many cell types, including neurons, endothelial cells, tumor cells, osteoblasts, and T cells (3). NRPs bind members of the vascular endothelial growth factor (VEGF) family, including VEGF-A, VEGF-B, VEGF-C, and placenta growth factor PlGF-2 (1,3). In angiogenesis, it is thought that NRP1 expressed by endothelial cells forms a tertiary complex with VEGF and the VEGF receptor tyrosine kinase, VEGFR-2, resulting in enhancement of VEGF biological activity (4). In neuronal guidance, class 3 semaphorins bind neuronal NRP1, and a tertiary complex is formed with plexins, which are transmembrane proteins that transduce the semaphorin signal (5). There are six members in the class 3 semaphorin family, secreted proteins that bind to NRP1 and/or NRP2. An exception is Sema3E, which binds plexin D1 (6).
NRPs have important functions during development, which have been shown in knock-out mice and knockdown zebrafish experiments. Nrp1-deficient mice died in utero at E12.5 to E13.5 and exhibited both neuronal and vascular defects, including severe malfunctions in axonal pathfinding, disorganized blood vessels, lack of normal vessel branching, and missing capillary networks (7,8). Nrp2-deficient mice developed arteries and veins normally but displayed a severe reduction of small lymphatic vessels and capillaries (9). In zebrafish, knockdown of NRP1 expression by antisense morpholinos resulted in vascular defects, including a loss of circulation through the intersegmental vessels that correspond to angiogenic capillary sprouts (10). Axial vessel formation, corresponding to arteries and veins, was not affected. Neural defects also occurred in NRP1 morpholino-treated zebrafish such as aberrant axon branching of motor neurons and migration of motor neurons out of the spinal cord. 3 NRPs also contribute to tumor angiogenesis (11,12).
NRP expression is tightly regulated. This is illustrated, for example, in endothelial cells. During embryonic development NRP1 is expressed preferentially in arterial endothelial cells, and NRP2 is expressed preferentially in venous endothelial cells (13,14). Exposure of endothelial cells to shear stress (15) or transplanting a vein into an arterial site (venous arterialization) induces NRP1 expression (16,17). Furthermore, arterial differentiation is controlled by a positive feedback-loop of VEGF and NRP1 (18). Lymphatic endothelial cells express NRP2 but not NRP1 (14).
To analyze the transcriptional machinery involved in regulating NRP1 expression, we isolated the mouse and human NRP1 promoters (19). An NRP1 luciferase promoter/reporter construct demonstrated optimal activity in a region from Ϫ823 to ϩ79 base pairs relative to the transcriptional start site. It was demonstrated by mutation, deletion, and gel shift analysis that an AP-1 element, two SP-1 sites, and a CCAAT box are involved in regulating constitutive and phorbol esterinduced promoter activity in HeLa cells (19).
Only a few regulators of NRP1 expression are known, including tran-scription factors and growth factors. Overexpression of the transcription factor Prox-1 down-regulates NRP1 expression in the transition from vascular to lymphatic endothelial cells (20). On the other hand, overexpression of the transcription factor Ets-1 has been shown to directly up-regulate NRP1 expression (21). Growth factors such as tumor necrosis factor-␣ (22) and epidermal growth factor (EGF) (23) have been shown to up-regulate NRP1 in endothelial cells and tumor cells, respectively. VEGF also up-regulates NRP1 in endothelial cells (24). As reported previously by our laboratory, in situ hybridization analysis showed that NRP1 was expressed in the epidermis, especially in suprabasal keratinocytes (25), the first report that keratinocytes express NRP1. Whereas NRP1 function and binding of VEGF and semaphorins has been studied extensively in endothelial cells and neurons (1,3), the regulation of NRP1 gene expression and the response to NRP1 ligands in keratinocytes has not been documented. Therefore, we carried out regulation and response studies using HaCaT cells, a keratinocyte cell line (26) that resembles primary keratinocytes and that can be readily transfected. Closer examination of the NRP1 promoter showed that there were possible repressor elements in the promoter region between Ϫ173 and Ϫ97. The most promising candidate for a negative transcriptional regulator in this region is the neuron restrictive silencer factor (NRSF), also known as the RE-1 silencing transcription factor (REST), which was discovered in 1995 independently by two groups (27,28). NRSF is a zinc finger transcription factor that binds to a 21-bp recognition sequence known as NRSE (neuron restrictive silencer element) (29). The main function of NRSF is to repress expression of neuronal genes in nonneuronal cell types and in neuronal progenitor cells (30,31). Repression is accomplished by recruitment of histone deacetylases, which deacetylate histone lysines, rendering them more basic. As a consequence, histones become tightly associated with the DNA, making the DNA less accessible to the transcription machinery (32,33).
We report here, using overexpression of wild-type and dominant negative NRSF, that NRSF is a direct transcriptional repressor of NRP1 expression that lowers both RNA and protein levels. Furthermore, the strong up-regulation of NRP1 by EGF or the heparin-binding HB-EGF in HaCaT cells is accompanied by a down-regulation of NRSF RNA and protein levels. In addition, this is the first report to show that keratinocyte NRP1 is biologically functional. Of the two NRP1 ligands, VEGF does not affect either keratinocyte migration or proliferation, whereas Sema3A inhibits migration. The inhibitory effects of Sema3A are diminished when NRSF is overexpressed in these cells, presumably by making less NRP1 available as a Sema3A receptor. It is concluded that, in keratinocytes, NRP1 is functional as a Sema3A receptor and that NRP1 expression is negatively regulated by NRSF.

EXPERIMENTAL PROCEDURES
Cell Lines and Cell Culture-Normal human epidermal keratinocytes (NHEKs) were purchased from Cambrex (Walkersville, MD) and maintained according to the company's recommendations. Culture and preparation of porcine aortic endothelial (PAE) cells and PAE cells stably overexpressing NRP1 have been described previously (4). HaCaT cells (26) were maintained as monolayers at 37°C, 5% CO 2 in modified Eagle's medium (Cellgro Mediatech, Herndon, VA) containing 10% fetal bovine serum (Invitrogen), 100 units/ml penicillin, 100 g/ml streptomycin, and 292 g/ml L-glutamine. Trichostatin A (200 M, dissolved in ethanol, Sigma) was added to the HaCaT culture medium at a final concentration of 200 nM, and cells were cultured for 24 h. Control cells were cultured in the presence of ethanol using the same dilutions.
Stably Transfected Cell Lines-HaCaT cells (3 ϫ 10 6 ) were transfected with 6 g of pcDNA 3.1 expression plasmids (Invitrogen) containing the cDNAs for either NRP1, NRSF, or dominant-negative NRSF for 24 h using the FuGENE6 reagent (Roche Applied Science) and following manufacturer's instructions. Selection was carried out by adding Geneticin (Invitrogen) at a final concentration of 500 g/ml over a period of 30 days. Clones derived from single cells were expanded and screened for transgene expression by PCR and Western blotting.
NRP1 Promoter-Luciferase Reporter Assays-To prepare NRP1 promoter-luciferase reporter constructs, various regions of the NRP1 promoter were cloned into the pGL2-luciferase plasmid (Promega, Madison, WI) as described earlier (19). To measure promoter-reporter activity, HaCaT cells were seeded into 6-well cell culture plates and grown to a density of 5 ϫ 10 5 cells per well. Into each well 2 g of pGL2-plasmid were cotransfected with 0.5 g of pSV-␤Gal control plasmid (Promega), using 6 l of FuGENE6 reagent. After 24 h the medium was changed, and 24 h later cells were lysed in 120 l of reporter lysis buffer (Promega). Lysates (30 l) were measured for luciferase activity. To normalize for transfection efficiency, the ␤-galactosidase activity of 50 l of lysate was determined with an assay system (Promega). The NRSE element in the neuropilin-1 promoter (located at position Ϫ134 to Ϫ114 bp upstream of the transcriptional start) was mutated from CGCAGACACCCGGACCTCCC to CGCAGACAC-CTTGACCTCCC using the Quick-Change XL-Kit (Stratagene, La Jolla, CA), following the manufacturer's instructions. Successful alteration of the NRSF binding site was verified by sequencing.
Nuclear Extracts and Electrophoretic Mobility Shift Assays-The preparation of nuclear extracts from HaCaT cells was carried out as described earlier (34). Briefly, cells were lysed mechanically with a glass homogenizer. Nuclei were pelleted by centrifugation and lysed with a combination of high salt treatment and mechanical disruption in a glass homogenizer. After high speed centrifugation, soluble proteins in the supernatant were dialyzed to reduce the salt concentration, measured for protein content and stored at Ϫ80°C prior to use. Double-stranded 32 P-labeled fragments of the NRP1 promoter containing the NRSE were generated by oligonucleotide synthesis (Invitrogen, sequence: 5Ј-CAT-TGCTCGTTCCCCTCCTTCCCGCAGACACCCGGACCTCCCCT-GGGCGCCACGTCCGCGGCTC) and subsequent end-labeling with T4 polynucleotide kinase (New England Biolabs, Beverly, MA) using [␥-32 P]ATP, following the manufacturer's protocol. The resulting fragment had a length of 64 bp. For binding reactions, 300 pM of these fragments was incubated with 4 g of nuclear extracts for 30 min at room temperature in a binding buffer (Promega, 5ϫ gel shift binding buffer). For competition experiments, nuclear extracts were preincubated for 15 min at room temperature with a 100-fold molar excess of unlabeled competitor oligonucleotides or with 2 g of antibodies (11C-12 mouse monoclonal antibodies against NRSF, kindly provided by Dr. David Anderson, California Institute of Technology, Pasadena, CA, or the normal mouse IgG control antibody from Santa Cruz Biotechnology, Santa Cruz, CA) followed by addition of labeled oligonucleotides. Samples were resolved on 4 -20% nonreducing TBE polyacrylamide gels (Bio-Rad) in 0.5ϫ TBE. Following electrophoresis, gels were dried, and subjected to autoradiography.
RT-PCR and Real-time RT-PCR-Total RNA was prepared by TRIzol (Invitrogen) extraction and subsequently reverse-transcribed using Superscript II enzyme (Invitrogen) and random hexamers. Regular PCR for NRSF was performed using Taq polymerase under the following conditions: dissociation at 94°C; annealing at 55°C; and elongation for 90 s at 72°C. Number of cycles: 20 -30, depending on signal strength. The following primers were used: forward 5Ј-GTGACTACCA-GAACTCG; reverse 5Ј-CACCTCTATGGGAGGAG. The 825-bp product was visualized by 1.5% agarose gel electrophoresis and subsequent ethidium bromide staining. For semiquantitative PCR of NRP1, amplification of both NRP1 and the control gene glyceraldehyde-3phosphate dehydrogenase were performed in the same tube. The following primers and conditions were used: NRP1 (forward, 5Ј-GCA-GAGCAGTGTCTCAG; reverse, 5Ј-GCTGTCATCCATGATCATC; amplification as described for NRSF, the product has a length of 805 bp) and glyceraldehyde-3-phosphate dehydrogenase (forward, 5Ј-CCAGC-CTCGTCCCGTAGACA; reverse, 5Ј-CTGGTCCTCAGTGTAGC-CCAAGATG). To check HaCaT cell clones for stable genomic integration of the dominant negative NRSF construct, a forward primer that bound within the pcDNA3.1 plasmid was used (5Ј-TAATACGACT-CACTATAGGG), whereas the reverse primer bound to a sequence in the cDNA coding for the dominant-negative NRSF (5Ј-CCTCCAGT-GATACTCG). Conditions for the amplification were as described above for NRSF, and the product had a size of 506 bp. For real-time PCR the Quantitect SYBR-Green-based system from Qiagen (Valencia, CA) was used, with amplification cycles consisting of dissociation at 94°C for 20 s, annealing at 50°C for 20 s, and elongation at 72°C for 40 s, resulting in a PCR product of 144 bp in length. Primers were: forward, 5Ј-CGCTGTGACCGCTGCG; reverse, 5Ј-CCTCCAGTGATACTCG.
Cloning of NRSF and Dominant-negative NRSF Expression Plasmids-A full-length NRSF cDNA containing the entire coding region was amplified by RT-PCR using total RNA (extracted with TRIzol reagent, Invitrogen) from HeLa cells as a template. Reverse transcription was performed with random hexamers and Superscript II enzyme (Invitrogen). Advantage-2 polymerase (BD Biosciences, Clontech, Palo Alto, CA) was used for PCR amplification. 5Ј primer, GCCGAATACAGTTATGGC; 3Ј primer, GTTCAAAGTTTCATTACTC. The PCR product was inserted into the pcDNA3.1-TOPO-V5-His plasmid (Invitrogen) by TA cloning, and the correct sequence was verified by automated sequencing. When compared with the published reference mRNA sequence (accession number NM005612), all clones derived from different HeLa and HaCaT RNA preparations contained a single point mutation at position 1889 of the coding region, with a change of leucine into proline. This variant has been published before (accession number AB209750). A truncated NRSF form, spanning amino acids 135-446 of the 1097-amino acid full-length protein that has been reported to act as dominant-negative in vitro and in vivo (27), was generated by RT-PCR as described above. A start and a stop codon were inserted by PCR (5Ј primer, ATGTCAAATAAAGATCTTC-CCCCTGAAACACCTG; 3Ј primer, TTAAGTTTTTTCATTGGTA-ATATTATCAGG), and the PCR product was cloned into pcDNA3.1-TOPO-V5-His as described above.
Immunoprecipitation and Western Blotting-For Western blot detection of NRP1 and NRP2, HaCaT cells were lysed in a buffer containing 1% Nonidet P-40, 1 mM EDTA, 10 mM Tris-HCl, pH 7.4, and protease inhibitors (Complete Mini, Roche Applied Science). The cells were sonicated, and protein content was determined. Equal amounts of total protein were incubated with concanavalin A-Sepharose beads (Amersham Biosciences) at 4°C overnight. After washing, bound proteins were eluted in Laemmli buffer, boiled for 5 min, and subjected to gel electrophoresis (8% SDS-polyacrylamide gels). Separated proteins were transferred to nitrocellulose membranes by electroblotting, and specific proteins were detected after incubation with primary and horseradish peroxidase-coupled secondary antibodies using enhanced chemiluminescence (Amersham Biosciences), according to standard procedures. The antibodies used were A12 mouse monoclonal anti-NRP1 antibody (dilution 1:2000, Santa Cruz Biotechnology) and C9 mouse monoclonal anti-NRP2 antibody (1:2000, Santa Cruz Biotech-nology). For detection of NRSF, cell lysates were immunoprecipitated at 4°C overnight with protein G-Sepharose and 11C-12 mouse monoclonal anti-NRSF antibody (kindly provided by Dr. David Anderson). P18 goat polyclonal antibody against NRSF (1:2000, Santa Cruz Biotechnology) was used as primary antibody. Epidermal extracts were prepared by microdissection of 60-m cryosections using a binocular microscope with subsequent extraction in Laemmli buffer.
Immunohistochemical Staining-Discarded normal skin specimens were obtained from the Dana Farber Cancer Center, Boston, MA with Internal Review Board approval. Formalin-fixed, paraffin-embedded skin sections were de-paraffinized in xylene and rehydrated through ethanol to water and then PBS (pH 7.5). Endogenous peroxidases were blocked using 3% peroxide in methanol for 12 min at room temperature. After washing with PBS, proteins were incubated in protein blocking solution (3% normal goat serum Sigma and 2% normal sheep serum Sigma in PBS) for 20 min at room temperature. Sections were then incubated at 4°C overnight with anti-NRP1 antibody 44 -2 (recognizing amino acids DDSKRKAKSFEGNNNYD in the b2 domain, kindly provided by Seiji Takashima, Osaka University Hospital, Osaka, Japan) at a dilution of 1:400 in protein blocking solution. The next day, sections were washed with PBS, incubated in protein blocking solution for 10 min at room temperature, and incubated for 1 h at room temperature in peroxidase-conjugated goat anti-rabbit IgG F(abЈ) 2 antibody (Jackson ImmunoResearch, West Grove, PA). Slides were washed with PBS and visualized using stable diaminobenzidine (Vector Laboratories, Burlingame, CA). Staining was monitored under a bright-field microscope and washed with distilled water to stop the reaction. Sections were then counterstained with Gill's 3 Hematoxylin (Sigma), washed, and mounted with Permount (Fisher). Positive control for NRP1 staining included PAE/NRP1 cells grown on slides, and negative controls for staining included PAE cells grown on slides and skin sections incubated in secondary antibody alone. For immunofluorescent labeling of Sema3A a rabbit polyclonal antibody from Abcam was used (ab23393) on acetone-fixed cryosections. Sema3A was visualized by use of a rhodamine-conjugated secondary antibody. NRSF was stained with the 11C-12 mouse monoclonal anti-NRSF antibody mentioned above. Cells were permeabilized by 0.2% saponin. Bound primary antibody was detected by a FITC-conjugated secondary antibody.
Cell Proliferation and Migration-Cell proliferation was determined using the MTT (3-(4,5-dimethylthiazolyl-2)-2,5-diphenyltetrazolium bromide) assay (Sigma) in 96-well tissue culture plates in triplicates. HaCaT cells (5 ϫ 10 3 ) were seeded per well and cultured in full growth media containing 10% fetal bovine serum. After 24 h, medium was changed and VEGF 165 (1-50 ng/ml) or Sema3A (30 -300 ng/ml) was added. The medium was exchanged every 24 h, and cell numbers were determined at time points 0, 24, and 48 h after addition of growth factors. The spectrophotometric measurement of the converted dye reflects mitochondrial dehydrogenase activity of viable cells and correlates with cell number. For analysis of migration, 1 ϫ 10 4 cells in fetal bovine serum-free culture medium were seeded into the upper chambers of Millicell PCF Transwell inserts (Millipore, Bedford, MA). The filter had a pore size of 12 m. Fully supplemented medium (modified Eagle's medium) containing 10% fetal bovine serum was added to the lower compartment as the chemoattractant. Chick collapsin-1 (equivalent to mammalian Sema3A) and/or VEGF 165 were added to the media of both chamber compartments at concentrations of 150 ng/ml and 10 ng/ml, respectively. These concentrations had been shown to achieve maximal effects on dorsal root ganglia cells or endothelial cells, respectively. After 16 h cells on the lower side of the filter were stained with xanthene/thiazine dyes (Diff Quick staining set, Dade Behring, Newark, DE) and counted. On each filter five fields of a defined size (evenly distributed over the filter) were counted manually under a microscope, using a magnification of 200-fold. From these five numbers the mean was calculated. Experiments were carried out in triplicates, and the standard deviation for each treatment group was calculated from the three mean values. Migration of unstimulated HaCaT cells was defined as a reference value of 100%, and cell numbers from the VEGF-or Sema3A-treated cells were expressed relative to this group. The Student t test was used for the determination of p values.

Neuropilin-1 Is Expressed in Keratinocytes in Vivo and in
Vitro-Previous analysis by in situ hybridization showed that NRP1 is expressed by epidermal keratinocytes in human skin (25). Here we demonstrate that NRP1 is expressed in vitro in HaCaT cells, a human model keratinocyte cell line that is spontaneously immortalized but non-tumorigenic and retains most of the differentiation potential of normal keratinocytes (26), and by primary normal human epidermal keratinocytes (NHEK, Fig. 1A). Porcine aortic endothelial (PAE) cells, and PAE cells overexpressing NRP1 (4) served as negative and positive controls, respectively, for the expression of NRP1. It was concluded that keratinocytes express NRP1 in vitro and in vivo. Because NRPs depend on either plexins (for Sema3A) or VEGFRs (for VEGF) to transduce the signal into the cell, the expression of these molecules in HaCaT cells was investigated. RT-PCR analysis showed that HaCaT cells express the plexins A1, A2, A3, and B1 (Fig. 1B), whereas none of the three VEGFR kinases could be detected (Fig. 1C). We also checked the expression of NRP1 ligands and found both VEGF isoforms VEGF 121 and VEGF 165 as well as semaphorins Sema3A and Sema3F to be expressed by HaCaT cells grown in vitro (Fig.  1D, upper panel) and in tissue extracts from human epidermis (Fig. 1D,  lower panel).
NRP1 Promoter Contains a Neuron Restrictive Silencer Element That Represses NRP1 Promoter Activity-Little, if anything, is known about the regulation and function of keratinocyte NRP1. To analyze regulation of NRP1 gene expression, an NRP1 promoter-luciferase construct (19) was used to measure promoter activity (Fig. 2). The highest promoter activity resided in a region 823 bp upstream from the transcriptional start site (19). Serial truncations of the promoter at the 5Ј-end resulted in a strong decrease of transcriptional activity ( Fig. 2A). However, removal of the promoter region Ϫ173 to Ϫ97 bp upstream of the transcriptional start site actually led to an increase in reporter gene expression, suggesting that a transcriptional repressor element resided within that particular sequence. Accordingly, this promoter region was scanned with MatInspector software (Genomatix) for potential transcription factor binding elements. The program revealed several putative binding motifs, including sites for AP-2, ZF-1, or MZF-1, most of which had been described to be recognized by factors that enhance transcription. One likely candidate sequence for a repressor was a neuron restrictive silencer element (NRSE). To ascertain whether this NRSE was a regulatory element in the NRP1 promoter, this site was altered by site-directed mutagenesis, changing 2 bp within the core recognition sequence from CGGA to TTGA (Fig. 2B). When the mutated reporter construct was transfected into HaCaT cells, a statistically significant increase in luciferase activity of ϳ70% with a p value of Ͻ0.01 was measured (Fig. 2C). These results demonstrated that the NRSE was functionally active as a transcriptional repressor element.
NRSE binds the transcription factor NRSF (neuron restrictive silencer factor, also known as REST for RE-1 silencing transcription factor). To carry out electrophoretic mobility shift assays for analyzing NRSF-NRSE binding, HaCaT cell nuclear extracts were prepared and incubated with a radiolabeled, PCR-generated 64-bp fragment from the NRP1 promoter, which included the NRSE and had a molecular mass of ϳ42 kDa (Fig. 2D). The nuclear extracts contained a protein with an estimated molecular mass between 110 and 125 kDa that bound to the radioactive probe, resulting in a shifted band of ϳ160 kDa (Fig. 2D, lane  2). This size is consistent with the reported molecular mass of NRSF, which is 116 kDa (27). The binding was specific and could be competed with unlabeled probe (Fig. 2D, lane 3) or with double-stranded DNA oligonucleotides corresponding to a NRSE sequence from the SCG-10 gene (Fig. 2D, lane 4), which has been shown repeatedly to bind NRSF  (28). An unlabeled PCR fragment corresponding to the radiolabeled probe but generated from the NRP1 promoter with a mutated NRSE was not able to compete with protein binding to the labeled wild-type NRP1 promoter oligonucleotide (Fig. 2D, lane 5). Furthermore, addition of an anti-NRSF antibody resulted in the disappearance of the shifted band (Fig. 2D, lane 6) probably due to steric interference, as previously reported by other groups for this particular antibody (35). However, we could not detect the supershift reported by other investigators using this antibody (36). No disappearance of the shifted band was observed after addition of a nonspecific mouse control antibody of the same Ig-type (Fig. 2D, lane 7). It was concluded that HaCaT nuclear extracts contain NRSF protein, which binds specifically to the NRSE in the NRP1 promoter.
NRSF Overexpression Reduces NRP1 Levels, whereas Dominant-negative NRSF and Trichostatin A Increase NRP1 Levels-To determine whether NRSF binding to the NRP1 promoter results in reduced NRP1 transcription and protein levels, the full-length cDNA of NRSF was cloned by PCR and inserted into a mammalian expression vector, which was subsequently used in NRSF transfection studies. Wild-type HaCaT cells expressed NRSF and NRP1 (Fig. 3A, lane 1). On the other hand, three stable clones overexpressing NRSF had relatively lower levels of NRP1 expression (Fig. 3A, lanes 2-4). Reduction of NRP1 protein was confirmed by Western blot (Fig. 3A, right). Densitometric analysis of the Western blot revealed a 2.6-fold reduction in NRP1 protein. EGFR, which also binds to ConA-Sepharose beads, served as a control. In contrast to NRP1, NRP2 levels were not altered by NRSF overexpression. This indicates a differential regulation of the two NRPs by NRSF.
It has been reported that the N-terminal eight zinc finger domains of NRSF comprise the DNA-binding domain and act dominant-negatively, probably by competing with wild-type full-length NRSF for the NRSE binding sites (27). The corresponding N-terminal region was cloned and inserted into the same pcDNA3.1 expression plasmid as used for NRSF overexpression and transfected stably into HaCaT cells. Overexpression of this dominant-negative construct was accompanied by increased amounts of NRP1 mRNA transcripts, as demonstrated for three different clones (Fig. 3B, lanes 2-4). Induction of NRP1 was confirmed on the protein level by Western blotting (Fig. 3B, right). In this case, the densitometric measurement indicated a 3.1-fold increase in NRP1 protein.
NRSF itself acts as a repressor by recruiting histone deacetylase activity to the chromosomal region of the gene that has bound NRSF to its regulatory NRSE sequence (32,33). Trichostatin A is an inhibitor of histone deacetylase (37). Trichostatin A treatment of HaCaT cells did not change NRSF expression (Fig. 3C, left), but increased NRP1 mRNA levels (Fig. 3C, middle). The induction of NRP1 protein expression, 2.3-fold, could be confirmed by Western blotting (Fig. 3C, right). Taken together, these results indicate that NRSF acts as a transcriptional inhibitor of NRP1 gene expression.

Up-regulation of NRP1 Gene Expression by EGF and HB-EGF Correlates with Down-regulation of NRSF Expression-
To determine whether NRSF and NRP1 expression are inversely correlated in a more physiological situation, possible external regulators were examined. By Northern blotting, it was found that EGF (not shown) or HB-EGF strongly induced NRP1 mRNA expression in HaCaT cells (Fig. 4A). The increase of NRP1 mRNA started 4 h after HB-EGF stimulation and continued to rise over the next 24 h. The strong induction of NRP1 up-regulation by EGF and HB-EGF was confirmed on the protein level for both HaCaT cells and primary human keratinocytes, as demonstrated by Western blot analysis (Fig. 4B). On the other hand, other members of the EGF family of growth factors such as amphiregulin or neuregulin did not up-regulate NRP1 (Fig. 4C). Similarly, keratinocyte growth factor (KGF, FGF-7) was not able to increase NRP1 mRNA levels (Fig. 4C), even though KGF is a potent keratinocyte mitogen (38). VEGF 165 , which binds NRP1 and has been shown to enhance NRP1 expression in human FIGURE 2. Identification of a functional NRSE repressor element in the NRP1 promoter. A, various 5Ј-truncated promoter fragments of the human NRP1 gene were cloned into the pGL-2 basic vector upstream of a luciferase reporter gene and transiently transfected into HaCaT cells. Serial truncations lead to a decrease of luciferase activity, with one exception: removal of the region from Ϫ173 to Ϫ97 upstream of the transcriptional start site resulted in stronger luciferase expression. B, bioinformatic analysis revealed a putative 21-bp transcription factor binding site known as an NRSE within the respective part of the promoter. This element was inactivated by mutation, generating the reporter gene plasmid Ϫ823⌬NRSE. C, when Ϫ823⌬NRSE was transfected into HaCaT cells, a significantly higher luciferase activity could be detected, as compared with the Ϫ823-bp wildtype NRP1 promoter fragment (p Ͻ 0.01). D, electrophoretic mobility shift assays were performed using a 32 P-labeled PCR fragment generated from the NRP1 promoter, which contains the putative NRSE (lane 1). HaCaT cell nuclear extracts contained a protein with a molecular mass of ϳ120 kDa (lane 2), which bound to the 42-kDa probe. Specificity of binding was demonstrated by competition with an excess of unlabeled PCR-fragments (lane 3) or by competition with the NRSE derived from the SCG-10 gene, which is known to bind NRSF protein (lane 4). The mutated NRSE from the NRP1 promoter could not compete (lane 5). A monoclonal antibody against NRSF (11-C12) specifically blocked the formation of the DNA-protein complex (lane 6), whereas an unspecific control antibody had no effect (lane 7). umbilical vein endothelial cells (24) did not enhance NRP1 expression in keratinocytes either. On the other hand, VEGF 165 and amphiregulin down-regulated NRP1 somewhat. None of these growth factors was able to alter NRP2 expression levels (data not shown).
Because NRP1 expression is inversely correlated with NRSF expression, as described in Fig. 3A, a possible alteration of NRSF expression in response to HB-EGF stimulation was examined. When HaCaT cells were treated with HB-EGF, quantitative real-time PCR revealed an 2.2fold decrease in NRSF transcripts (Fig. 4D). This inhibition of NRSF by HB-EGF was confirmed by Western blotting (Fig. 4E), and the time course showed a strong down-regulation of NRSF protein, starting between 4 and 8 h after addition of HB-EGF. It was concluded that the down-regulation of the silencing transcription factor NRSF contributes to the strong induction of NRP1 by EGF-receptor ligands. To further demonstrate the influence of NRSF on growth factor-mediated expression of NRP1, we compared HaCaT cells and HaCaT cells stably transfected with NRSF following HB-EGF treatment (Fig. 4F). As in the parental HaCaT cells, the stable NRSF clone showed an increase in NRP1 protein after HB-EGF stimulation. Overexpression of NRSF resulted in decreased levels of basal and HB-EGF induced NRP1.
In situ hybridization showed that NRP1 is up-regulated in the epidermis along the edge of healing wounds in mice (not shown). To demon-strate the inverse correlation between NRP1 and NRSF, we prepared epidermal tissue extracts from normal human epidermis and from healing 3-day human wounds. Western blotting detected higher NRP1 and lower NRSF levels in the healing wound, as compared with the normal epidermis (Fig. 4G).
Neuropilin-1, Sema3A, and NRSF Are Expressed in the Epidermis in Vivo-Previously we showed NRP1 expression in human epidermis by in situ hybridization (25). This result was confirmed by immunohistochemistry, which showed localization of NRP1 protein in the epidermis, especially in the suprabasal layers (Fig. 5A). NRP1 expression in PAE/ NRP1 cells served as a positive control, and PAE cells or PAE/NRP2 cells served as negative controls for antibody specificity. Of the two groups of NRP1 ligands, expression of VEGF has been reported by several groups. However, there are no reports of semaphorin expression in the skin. Immunohistochemical staining for Sema3A detected a strong signal in the epidermis, with a tendency toward higher expression in suprabasal cell layers (Fig. 5B). We also immunostained for NRSF and found a nuclear expression pattern, most prominent in the spinous cell layer (Fig. 5C). These data demonstrate that both NRP1 and Sema3A are present in the epidermis in vivo.
Semaphorin 3A Inhibition of Keratinocyte Migration Is Diminished by NRSF-The function of NRP1 in keratinocytes is not known. NRP1 is a receptor for both VEGF 165 and Sema3A, and therefore it is possible that both proteins would bind to keratinocytes and have a function in these cells. Effects of these ligands on cell proliferation and migration were examined (Fig. 6). Neither Sema3A nor VEGF 165 affected HaCaT proliferation (Fig. 6A). On the other hand, effects on migration might be more plausible, because VEGF and Sema3A have been shown to stimulate and inhibit the migration of endothelial cells, respectively (39). Sema3A also inhibits neuronal migration (40). The migration of wildtype HaCaT cells, HaCaT cells overexpressing NRP1, and HaCaT cells overexpressing NRSF was examined (Fig. 6B). VEGF did not affect the migration of any of these cells (black bars). On the other hand, Sema3A inhibited the migration of HaCaT cells by ϳ30% (Fig. 6B, left, compare white to gray bar). Because NRP1 expression levels in HaCaT cells are relatively low (see control lane in Fig. 4B), HaCaT cells stably overexpressing NRP1 and thus possessing increased Sema3A receptor levels were used. Sema3A had a stronger inhibitory effect of ϳ60% on the migration of these HaCaT/NRP1 cells (Fig. 6B, center, white versus gray bar).
When HaCaT cells overexpressed NRSF so as to lower NRP1 levels (see Fig. 3A), the inhibitory activity of Sema3A was almost abolished (Fig. 6B, right, white versus gray bar). From these results it was concluded that NRP1 in keratinocytes acts primarily as a receptor for semaphorins rather than for VEGF. In addition, Sema3A is an inhibitor of keratinocyte migration via NRP1, and this inhibition is diminished when NRP1 levels are decreased by overexpression of NRSF.

DISCUSSION
Epidermal keratinocytes express NRP1 in vivo in the suprabasal epidermis and in vitro in primary epidermal keratinocytes as well as in the immortalized keratinocyte cell line, HaCaT, which closely resembles normal keratinocytes (26). This is the first demonstration that keratinocytes express functional NRP1 and that they have the capacity to respond to a NRP1 ligand, Sema3A. Previously it has been shown that in order for NRP1 to mediate angiogenesis and neuronal guidance, cells need to co-express VEGF receptor tyrosine kinases for VEGF activity and plexins for semaphorin activity, respectively (5,41). PCR and Western blotting show that the HaCaT cells did not express VEGF receptor tyrosine kinases (VEGFR-1, VEGFR-2, and VEGFR-3), whereas these cells did express plexins A1, A2, A3, and B1, predicting that Sema3A, but not VEGF 165 , would be active on keratinocytes. Consistent with this receptor profile, Sema3A treatment inhibited cell motility, whereas incubation with VEGF had no effect on cell migration. Interestingly, the proliferation rate of these cells was not altered. Thus, it appears that NRP1 in keratinocytes is primarily a semaphorin receptor. These migration and proliferation results are consistent with previous studies showing that Sema3A inhibits endothelial cell migration but not proliferation (39) and that overexpression of NRP1 in tumor cells increased their migration but not their proliferation (12). These observations are con-  4) are less responsive to the NRP1-inducing effects of HB-EGF. G, tissue extracts from healing human wounds contain higher levels of NRP1 protein and lower levels of NRSF than extracts from normal human epidermis, demonstrating an inverse correlation between these two molecules in vivo.  FEBRUARY 3, 2006 • VOLUME 281 • NUMBER 5 sistent with the nervous system where Sema3A regulates migratory events, but there are no reports showing altered cell proliferation (42).

NRSF Represses NRP1 in Keratinocytes
A role for NRP1 in keratinocytes in vivo has not been elucidated yet. In skin, NRP1 is mainly expressed in suprabasal layers of the epidermis, which consist of differentiating but not proliferating keratinocytes. The inability of VEGF to induce NRP1 expression and the restricted spatial expression of NRP1 to the suprabasal epidermis, which has no blood vessels, suggests that NRP1 doesn't necessarily affect skin angiogenesis. However, NRP1 could play a role in the regulation of keratinocyte migration, because Sema3A inhibits keratinocyte migration in vitro. It is possible that semaphorins inhibit keratinocyte migration in vivo as well. Our finding that Sema3A is present in human epidermis in vivo is consistent with this possibility. The inhibitory effect of Sema3A might be especially important during processes such as wound healing. We found that NRP1 mRNA is strongly induced by treatment with EGF or HB-EGF in vitro. HB-EGF is a potent keratinocyte mitogen and migration factor found in wound fluid (43). Because the precise regulation of keratinocyte migration is critical for the successful re-epithelialization of wounds, the NRP1/Sema3A pathway might be implicated in the finetuning of these events. This concept is supported by the observation that NRP1 mRNA is up-regulated in mice at the edges of healing wounds in vivo. 4 Additionally, we found higher levels of NRP1 protein and decreased NRSF levels in epidermis of healing wounds, as compared with normal epidermis.
In the course of analyzing NRP1 regulation in keratinocytes we noticed that the Ϫ173to Ϫ97-bp region of the NRP1 promoter, which we had previously isolated (19) appeared to have repressor elements that inhibited NRP1 expression in HaCaT cells. A number of potential transcription factor binding motifs could be found in this negative regulatory region, but these were mostly positive regulators of gene transcription, for example AP1. The only likely repressor sequence was NRSE, which interacts with the silencer NRSF. NRSF, a zinc finger transcription factor that binds to the 21-bp recognition sequence NRSE, is a transcriptional repressor of multiple neuronal genes and is normally down-regulated upon neuronal differentiation (44). It represses the expression of neuronal-specific genes in non-neuronal cells (30). Outside the neuronal system, NRSF is active in cardiomyocytes and selectively regulates expression of multiple fetal cardiac genes and maintains normal cardiac structure and function (45). We now provide evidence that NRSF also has a function in HaCaT cells, where it inhibits NRP1 expression. NRSF activity in keratinocytes has not been reported previously. Evidence for repression of NRP1 by NRSF is as follows. (i) Mutation of the NRSE element in the NRP1 promoter increased promoter activity, indicating that NRSF is a plausible NRP1 repressor. ii) Stable overexpression of NRSF in HaCaT cells decreased NRP1 RNA and protein levels. iii) A dominant-negative NRSF construct had the opposite effect. iv) Trichostatin A, an histone deacetylase inhibitor that indirectly inhibits NRSF-mediated gene silencing, increased NRP1 RNA and protein levels.
Interestingly, NRSF regulated NRP1 but not NRP2 expression. Analysis of the NRP2 promoter revealed a potential NRSE upstream of the transcriptional start site, although the homology of this 21-bp motif to the sequence of the published consensus NRSE was only 70%. Our data indicate that the NRSE in the NRP2 promoter might not be functional. NRP1 and NRP2 overlap in many properties, for example, VEGF 165 binding, but differ in other properties, for example, in endothelial cell identity. During development, NRP1 is expressed by arterial endothelial cells, and NRP2 is expressed by venous and lymphatic endothelial cells (3,13,14). Thus, NRP1 and NRP2 are regulated differently. Differential activity of NRSF on NRP1 and NRP2 regulation would be another feature distinguishing the two receptors.
NRP1 expression by tumor cells has been demonstrated to enhance tumor angiogenesis and progression (12). We have preliminary data showing that NRP1 suppression by NRSF also occurs in human melanoma cells. A recent publication identified NRSF as a human tumor suppressor gene (46). In this study, inactivation of NRSF by RNA interference initiated a transformed phenotype of human mammary epithelial cells in vitro. Furthermore, it was demonstrated that deletion or mutation of the NRSF gene is frequent in samples of human colon cancer. The mechanism for this novel function of NRSF is still unclear.
There are a number of potent keratinocyte mitogens. These include members of the EGF and FGF families. EGF and HB-EGF very strongly induce NRP1 expression. Concomitantly, NRSF expression is inhibited, both at the RNA and protein levels, by HB-EGF. These results are further evidence of the inverse relationship between NRSF and NRP1 expression. However, the mechanism for this is unclear. It should be noted that the increase of NRP1 RNA after stimulation with HB-EGF is detectable after 4 h, whereas NRSF levels decrease between 8 and 16 h after the onset of HB-EGF stimulation. This indicates that NRSF is not involved in the early induction of NRP1 by HB-EGF but, rather, helps to maintain elevated NRP1 levels. Interestingly, FGF family members that are potent keratinocyte mitogens, for example, KGFs (38), do not upregulate NRP1. Thus, it is possible to dissociate keratinocyte proliferation from NRP1 expression. VEGF has been shown previously to upregulate NRP1 in endothelial cells (24). However, we did not observe this in HaCaT cells, consistent with the lack of VEGFR-1 and VEGFR-2 in HaCaT cells. 4 D. Bielenberg, unpublished data. FIGURE 6. Sema3A inhibits migration but not proliferation of HaCaT cells. A, cell proliferation was measured using the MTT assay. Neither 10 ng/ml VEGF 165 (black bars) nor 150 ng/ml Sema3A (gray bars) was able to influence HaCaT proliferation. B, Sema3A decreased HaCaT cell migration by ϳ30%, in Transwell chambers (left). This decrease was found to be more prominent when HaCaT cell clones stably overexpressing NRP1 were used, resulting in an inhibition of ϳ60% (center, white versus gray bar). The inhibitory effect of Sema3A was greatly diminished in HaCaT cells, which stably overexpressed NRSF protein and which had therefore diminished NRP1 receptor levels (right). VEGF 165 had no effect on migration.
It is becoming increasingly clear that the transcriptional repression of genes in general is as important for cellular functions as is the induction of particular genes. During development and cell differentiation it is necessary not only to express certain genes when needed, but also to restrict the transcription of other genes at the right time. Therefore, transcriptional repressors have become a focus of gene regulation research. Prior to finding that NRSF represses NRP1, two transcription factors have been shown to down-regulate NRP1. One of these is Prox-1. Expression of Prox-1 in vascular endothelial cells results in acquisition of a lymphatic endothelial cell phenotype that includes down-regulation of NRP1 and up-regulation of NRP2 (20). The precise mechanisms responsible for these regulatory activities have not been yet identified. Recently, COUP-TFII, a member of the orphan nuclear receptor superfamily, has been shown to be specifically expressed in venous but not in arterial endothelial cells (47). Normally, NRP1 is expressed in arterial but not venous endothelial cells. COUP-TFII in venous endothelium suppresses NRP1 and inhibits downstream Notch signaling and expression of arterial endothelial-specific genes such as ephrin b2. Deletion of COUP-TFII enables venous endothelial cells to acquire arterial characteristics, including NRP1 expression. By identifying NRSF as a repressor of NRP1 gene transcription, we have described the third negative regulator of NRP1.
There are several pathological situations that are accompanied by increased NRP1 expression levels. Most notably, NRPs contribute to tumor angiogenesis, progression, and metastasis. NRP1 is overexpressed in many tumors such as human glioma, neuroblastoma, breast carcinoma, pancreatic cancer, and gastrointestinal carcinoma (48). Previously, we have shown that induced expression of NRP1 under control of an inducible promoter increased tumor vascularity and tumor size (12). Thus, inhibiting NRP1 might be a useful therapeutic strategy for these diseases. Anti-NRP1-neutralizing antibodies, soluble NRP1, and RNA interference are possible candidates to antagonize NRP-mediated effects. Enhancing NRSF expression to down-regulate NRP1 might be another way to accomplish this task.