Small Molecule Neuropilin-1 Antagonists Combine Antiangiogenic and Antitumor Activity with Immune Modulation through Reduction of Transforming Growth Factor Beta (TGFβ) Production in Regulatory T-Cells

We report the design, synthesis, and biological evaluation of some potent small-molecule neuropilin-1 (NRP1) antagonists. NRP1 is implicated in the immune response to tumors, particularly in Treg cell fragility, required for PD1 checkpoint blockade. The design of these compounds was based on a previously identified compound EG00229. The design of these molecules was informed and supported by X-ray crystal structures. Compound 1 (EG01377) was identified as having properties suitable for further investigation. Compound 1 was then tested in several in vitro assays and was shown to have antiangiogenic, antimigratory, and antitumor effects. Remarkably, 1 was shown to be selective for NRP1 over the closely related protein NRP2. In purified Nrp1+, FoxP3+, and CD25+ populations of Tregs from mice, 1 was able to block a glioma-conditioned medium-induced increase in TGFβ production. This comprehensive characterization of a small-molecule NRP1 antagonist provides the basis for future in vivo studies.


■ INTRODUCTION
Neuropilin-1 (NRP1) is a cell-surface coreceptor for a number of different growths factors, including several different isoforms of vascular endothelial growth factor (VEGF), transforming growth factor-β1 (TGF-β1), PLGF, HGF (also known as scatter factor) as well as Semaphorins 3A, 4F. 1 As such, NRP1 plays key roles in both vascular and neuronal development. 2,3 It has also been shown that NRP1 has an important immunological function. 4 NRP1 is expressed on several types of immune cells, including T cells and dendritic cells, where it is one of the components of the immunological synapse. 5 NRP1 is implicated in potentiating the function and survival of regulatory T cells (Tregs). 6 This T cell fragility is linked to responses to PD1 checkpoint inhibitors. 7 NRP1 expression can be used to distinguish Treg subsets arising in vivo, thus NRP1 is present on thymus derived Tregs (natural Tregs), 8 whereas it is not present on Foxp3 + positive inducible Tregs. 9, 10 The Ikaros family protein Helios has been suggested as an additional and more general marker for thymic derived Tregs. 11 NRP1 is also important in the control of the M2 shift in tumor associated macrophages/microglia in gliomas. 12 NRP1 interacts with TGFβR1 to activate SMAD2/3 and drive secretion of TGF-β1, which results in expansion of Treg subsequent immune suppression. 13−15 As the role of the immune system in cancer development becomes better understood, 16 NRP1 is emerging as an attractive anticancer target. 17 Novel drug compounds which act as NRP1 antagonists could therefore exhibit their anticancer effects in three different ways: blocking tumor angiogenesis by blocking the NRP1/ VEGF-A interaction, 18 preventing tumor cell migration by binding to NRP1, 19 and reducing Treg or macrophage mediated suppression of the immune response. 20 A number of peptide antagonists of neuropilin are known: ATWLPPR 21 is a low affinity linear peptide, whereas a bicyclic disulfide bonded peptide, EG3287, is derived from the Cterminal domain of VEGF-A 22 (Scheme 1). N-Terminal modification (N-octanoyl) resulted in a high affinity antagonist EG00086 (K D = 76 nM). 23 EG00086 was also shown to inhibit VEGF-A mediated cell signaling, including cell adhesion, through reduction in p130Cas tyrosine phosphorylation. Its usefulness for in vivo studies was limited, however, by its low plasma stability (t 1/2 < 5 min). 23 In addition, NRP1 antibodies 18 and a mini-protein based on the kalata cyclotide have been reported. 24 Development of a potent, small molecule NRP1 antagonist, with increased in vivo stability, would therefore be attractive. Despite the interest in this area, only a small number of molecules have been identified 25−27 These molecules are reported to have micromolar potencies, and some antitumor effects have been claimed in vivo. The best characterized of these is (S)-2- (3-(benzo[c][1,2,5]thiadiazole-4-sulfonamido)thiophene-2-carboxamido)-5-((diaminomethylene)amino)pentanoic acid (EG00229), which has been previously identified as a specific inhibitor of the NRP1/VEGF-A interaction. 27 Other compounds, such as the benzimidazolebased inhibitor exemplified by N-((5-(1H-benzo[d]imidazol-2yl)-2-methylphenyl)carbamothioyl)-2,3-dihydrobenzo[b][1,4]dioxine-6-carboxamide were identified through screening approaches (Scheme 1). 25 EG00229 was shown to inhibit the binding of biotinylated VEGFA (bt-VEGF-A) to NRP1 with an IC 50 of 8 μM. It was also demonstrated to have functional effects on cell-migration and VEGF-R2 phosphorylation. 27 Further studies have shown EG00229 to reverse an immune phenotype elicited by the immunomodulatory peptide tuftsin by blocking canonical TGFβ signaling through SMAD3/ AKT. 28 When delivered locally, the compound also inhibits glioma proliferation in vivo, replicating genetic ablation studies. 12 In squamous cell carcinoma, the compound suppresses epidermal stem cell function and tumor formation in vivo. 29 The binding mode of EG00229 has been confirmed by NMR and crystallographic studies, providing a useful starting point for the development of new NRP1 antagonists. 27 Herein, we utilize EG00229 as a starting point for the discovery of potent and bioavailable inhibitors of the NRP1/VEGF-A interaction, resulting in the identification of 1 (EG013777) as a new lead.

■ RESULTS AND DISCUSSION
Structure-Based Design of New NRP1 Antagonists. The crystal structure of EG00229 bound to human NRP1-b1 was previously solved to 2.9 Å resolution (PDB 3I97), and two different binding poses for the ligand were identified ( Figure  1A,1B). 27 The crystal structure revealed a close fit of the arginine portion of the molecule into the NRP1 binding pocket with near identical conformations observed for this part of the molecule. As previously noted, the 3-aminothiophene-2carboxylic acid displays an H-bonded constrained conformation in the bound molecule, indicating the presence of an alternate tautomeric form of this substructure. In contrast, the positions of the benzothiadiazole group were markedly different. Although the two binding poses of EG00229 were distinct, the positions of the protein side chains were identical when overlapping the two chains, except for E348. We hypothesized that through further modification and elaboration of the EG00229 scaffold, more potent NRP1 antagonists could be synthesized. Using chain B as an illustration ( Figure 1B), two key amino acids were identified which could be targeted either to improve existing H-bond interactions (S298) or to introduce new ones (E348). 27

Scheme 1. Previously Identified Small Molecule and Peptidic Antagonists of NRP1
Journal of Medicinal Chemistry Article Chemistry. To target outer pocket residues, such as S298 and E348, a range of substituted dihydrobenzofurans were prepared. These were designed to be able to make potential hydrogen bonding or salt bridge contacts with the S298 and E348 residues. The first part of the general synthetic route for the 5-substituted dihydrobenzofuran series is shown (Scheme 2). The synthesis began with sulfonamide formation between 5bromo-2,3-dihydrobenzofuran-7-sulfonyl chloride 2 and methyl 3-aminothiophene-2-carboxylate, to give sulfonamide 3. Hydrolysis of the methyl ester with LiOH gave acid 4, which was then coupled with the Pbf-protected arginine methyl ester to give 5. Subsequent hydrolysis of 5 gave the key intermediate 6, which was suitable for Suzuki−Miyaura couplings with a range of arylboronic acids.
From this common brominated intermediate 6, a range of azaheterocycles were prepared as shown in (Scheme 3). First, intermediate 6 was coupled to either 2-or 3-formylphenyl boronic acid to give 7a,b, and then reductive aminations were carried out using the desired amine to furnish substituted analogues 8 (for definitions of R, see Tables 1−3). Removal of the Pbf-protecting group in acidic conditions gave the final products 9a−e, 10a−e (Table 1), 11a−e, 12a−e, and 13a ( Table 2).
The synthesis of primary methylaminoaryl analogues was achieved by the use of preformed boronic acids (Scheme 4). Thus, 6 under Suzuki−Miyaura conditions with the 2-or 3methylaminoboronic acids gave the intermediates 14a−c, which with Pbf removal gave the target molecules 15a, 15b, and 1. For large scale batches of 1, a slightly modified synthetic route was employed with a Boc protected methylaminoboronic

Scheme 3. Synthesis of Aryl Substituted
Dihydrobenzofurans a acid (Supporting Information, Scheme S1). For the synthesis of cyclized isoindolyl analogues, 6 was transformed into the functionalized boronic acid 16 using bispinacolato diboron and Pd(dppf)Cl 2 , and this was used directly for the Suzuki− Miyaura coupling using potassium acetate as the base and Pd(PPh4) as the palladium catalyst (Scheme 5). In this case, cesium carbonate was preferred as the base. Final deprotection of the Pbf group furnished the isoindolyl analogues 18a and 18b.
Crystallographic Studies of 1. To further investigate the binding of 1 to NRP-1, X-ray crystallography studies were carried out. The differences in binding modes between 1 and EG00229 were then analyzed. The structure of 1 bound to NRP1-b1 was determined in two conformations: a high (0.9 Å) and a low (2.8 Å) resolution structure (Figure 1). The highresolution crystal structure provides us with the most detailed view of the ligand-binding site to date. The refined model includes residues 273−427 of NRP1-b1, 39 non-hydrogen atoms of 1 and 472 water molecules. High resolution allowed us to observe multiple conformations of the side chains; 24 side chains were refined with at least two alternative rotamers. Comparison of NRP1-b1/compound 1 complexes indicate that the ligand can bind in two different conformations. In the lowresolution 2.8 Å structure (PDB 6FMF), the ligand's bulky aromatics extend out of the back of the binding pocket. In the high resolution 0.9 Å structure (PDB 6FMC), they extend out of the top of the binding pocket ( Figure 2). The difference in ligand conformation originates from a rotation about the carbon−carbon bond axis of the carboxyl group, which forms hydrogen bonds to S346 and T349. There is an approximate 77°rotation along this bond ( Figure 2B and Supporting Information, Figure S1), resulting in more than 1 Å separation between the two different conformations. By exiting out of the top of the binding pocket, the ligand in the high-resolution structure forms additional interactions with the N-terminal residues (in particular G271−M276) of a symmetry mate

Journal of Medicinal Chemistry
Article located above the NRP1-b1 binding pocket. It is likely these interactions improve the crystal contacts, increasing crystal order, which is necessary to produce the higher resolution data explaining the difference in resolution between the two conformations. These contacts are however a crystallographic artifact, with the lower resolution structure more likely to represent the true conformation of 1 bound to NRP1-b1. The difference in ligand conformations results in a significant change in the side chain rotamer of E348. In the low resolution structure, E348 points away from the binding pocket and forms a hydrogen bond with the aryl-NH 2 of 1, which may help to explain the compound's increase in potency. In the high resolution structure, the aryl-NH 2 of 1 does not interact with E348 changing the side chain rotamer such that it now faces toward the center of the binding pocket. The detection of the hydrogen-bond to E348 in the low resolution structure confirmed our modeling predictions.
Biological Evaluation. All the compounds were evaluated using an SPR binding assay (Biacore) where recombinant NRP1-b1b2 protein was immobilized on a dextran coated chip. 23 Selected compounds were then evaluated in competitive binding assay systems using biotinylated VEGF. As part of an extensive structure−activity investigation, the binding of 9a and 10a to the NRP1-b1 domain was assessed by SPR and

Journal of Medicinal Chemistry
Article promising activity noted for the morpholine extended analogue 10a with binding affinity of 3.76 ± 0.52 μM by SPR as opposed to 14.43 ± 3.76 μM for the unsubstituted compound 9a. This encouraging result prompted us to conduct a more focused structure−activity study around the 10a structure.
The first group of analogues examined heteroaryl substituents on the 2′ and 3′ positions. All of the synthesized compounds 9b−e and 10a−e showed binding to the NRP1-b1 domain, with some compounds demonstrating nanomolar K D values. Substitution at the 3-position seemed generally favorable, with all of the 3-substituted compounds, 10a−e, showing higher binding affinities than the 2-substituted analogues.
A further range of analogues, 11a−e and 12a−e, were designed which contained a functionalized piperidine linker to add length and flexibility to further explore outer-pocket interactions. Binding affinities to the NRP1-b1 domain were again assessed by SPR ( Table 2).
The resulting compounds once again showed binding to NRP1, although the binding affinities were generally lower than had been observed for the previous azaheterocyclic compounds. The highest binding affinity for this series was obtained for the nonfunctionalized piperidine 12a (K D = 1.17 μM). The generally lower binding affinities for 11a−e suggested that the addition of the piperidine linker was not an effective strategy to introduce specific interactions with any additional surface amino acid residues, and so this series was not pursued further. With these results in hand, a compound set with smaller methylamino substituents that could be accommodated at the 4-position was synthesized (Table 3). Compounds 1 and 13a−e showed consistent activity although this declined with the methylated analogue 13c. Compound 1 showed reasonable affinity ( Figure 3A,B), which was encouragingly maintained in both cell-based and cell-free competition assays with bt-VEGF (Table 4). Isothermal calorimetry data for 1 fitted to a one-site binding model and provided an orthogonal assay system ( Figure 3B). Evaluation of 1 against NRP2, a closely related receptor to NRP1, showed no detectable binding (Supporting Information, Figure S2), indicating very good selectivity. These results prompted us to investigate the pharmacokinetic profiles of some selected analogues.
Pharmacokinetics. Both of the compounds from the dihydrobenzofuran series exhibited improved PK profiles over the historical compound EG00229, which has a relatively short half-life of 0.5 h. 12 Compound 10d had a longer half-life (1.2 h) with an improved V d of 1103 mL/kg (Table 4). Compound 1 also exhibited an encouraging half-life of 4.29 h, sufficient to sustain once per day dosing. The methylated analogue 13c showed less favorable parameters with a notably higher clearance and lower AUC than 1. With this data in hand demonstrating 1 to be a reasonably potent and stable inhibitor, we undertook a thorough biological characterization of 1 examining its antiangiogenic, antitumor, and immune effects.
Compound 1 Inhibits VEGF-A Stimulated Tyrosine Phosphorylation of VEGF-R2/KDR. VEGF-A signaling through VEGF-R2/KDR plays an important role in cell function in endothelial, tumor, and other cell types. 30 We investigated the effect of 1 on VEGF-R2/KDR tyrosine phosphorylation induced by VEGF-A in HUVECs. VEGF-A (1 ng /mL) stimulated a significant increase in VEGF-R2/KDR tyrosine phosphorylation at 10 min, which was inhibited by 50% on treatment with 1 at 30 μM ( Figure 4). Studies with 1 had previously shown a 20% inhibition at 30 μM. 27 These results once again indicate the importance of NRP1 for optimal VEGF function and signaling 1 and confirmed the higher potency of 1 compared to EG00229 as indicated by its higher affinity for NRP1-b1 and higher potency in a cell-free binding assay.
Angiogenesis, Inhibition of VEGF-Induced Migration in HUVEC Cells. To investigate the importance of blocking

Scheme 5. Synthesis of Isoindoyl Substituted
Dihydrobenzofurans a NRP-1 in HUVEC cells, we performed transwell assays of chemotaxis and in vitro scratch assays of wound closure (chemokinesis). The transwell assay examines cell chemotaxis, the directional cell migration toward the chemo-attractant. To understand if 1 could inhibit VEGF-A-induced migration of HUVEC cells, 2 × 10 5 HUVEC cells were plated in serum-free medium (EBM) with the addition of either 0.1% DMSO, 25 ng/mL VEGF-A, 1 (30 μM), or a combination of VEGF-A and 1 on the bottom chamber. Cells were allowed to migrate through the pores of the insets for 4 h. Data collected was consistent with previous reports, 31 with VEGF-A being able to induce HUVEC cells migration by almost 3 times more compared to DMSO control ( Figure 5A,B). Treatment of HUVEC cells with 1 alone did not influence the migratory ability of these cells but the administration of 1 at 30 μM in the presence of VEGF-A significantly reduces, by more than 60%, the ability of cells to migrate toward VEGF-A stimulus ( Figure  5B). These results suggest that 1 has a higher potency than the previously reported compound, EG00229, 27 that only displayed significant inhibition (≈34% reduction) once used at 100 μM in combination with VEGF-A.
Wound Healing Scratch Assay. HUVEC cells were plated and once confluent a scratch was made as described in the methods. Cells were kept in culture for 5 days in 1% EGM with 0.1% DMSO, 25 ng/mL VEGF-A, 1 (30 μM), or a combination of VEGF-A and 1. Data shows that 1 can delay the VEGFinduced wound closure ( Figure 5C).
Compound 1 in Combination with VEGF-A Reduces Network Area, Length, and Branching Points. Next, we used an organotypic endothelial−fibroblast coculture assay to recapitulate the endothelial tube formation characteristic of VEGF-A stimulated angiogenesis. The coculture assay of angiogenesis is a simple in vitro assay where HUVEC cells are cultured with human embryonic fibroblasts (HDF). The layer of fibroblasts secretes a complex extracellular matrix that contains collagen I with fibronectin, tenascin-C, decorin, and versican, mimicking the composition of tissue stroma. This matrix becomes remodelled into a 3D environment, allowing HUVECs to reorganize into a network of tubes.
This assay is particularly suited to test factors that promote or inhibit angiogenesis. Thus, we next analyzed endothelial tubulogenesis in coculture HUVEC cells treated with either VEGF-A or VEGF-A + 1 during 4 days ( Figure 6). Data collected shows that HUVEC cells stained for the endothelial marker Von Willebrand factor (VWF), have a ≈41% reduction in the number of VEGF-induced branch points in tubular networks upon NRP1 inhibition with 1 ( Figure 6B). This reduction was also observed when overall network area (≈50%) and length (≈40%) ( Figure 6B) were assessed. Results suggest that NRP1 inhibition can significantly influence the angiogenic properties of endothelial cells, thus being an attractive target to test on highly metastatic cancers that express NRP1.
Reduced VEGF-Induced Angiogenesis after Treatment with 1. To further analyze the effect of blocking NRP1 during angiogenesis, we next used an ex vivo mouse aortic rings assay. The aortic ring assay provides a more complete picture of angiogenic processes compared with traditional cell-based assays. In this model, endothelial cells are able to proliferate and migrate, forming network tubes and branching points without the need for cellular dissociation. 32 This allows us to assess different steps that occur during the angiogenic process, which we aim to target. Rings were

Journal of Medicinal Chemistry
Article embedded in collagen and kept in culture in medium containing 0.1% DMSO, 25 ng/mL VEGF-A, 1 (30 μM), or a combination of VEGF-A and 1. As expected, after 7 days in culture, VEGF-A increased vessel sprouting from WT aortic rings, but this response was significantly suppressed (≈7-fold reduction) by the administration of 30 μM of 1 ( Figure 7A,B). Studies have described that NRP1 up-regulation is associated with the tumor invasive behavior and metastatic potential, for instance, in melanoma and breast cancer. 17,33 Thus, our data reinforces the importance of targeting NRP1 and suggest a possible attractive therapeutic approach for cancers that are so far resistance to the traditional angiogenic therapies.
Antitumor: Blocking NRP1 Reduces Melanoma Invasion in a 3D Spheroid Assay. To further investigate the effects of NRP1 blockade on cancer cells, we used a threedimensional (3D) spheroid assay. 3D spheroids are a useful tool to replace the commonly used 2D cell culture systems. By using this system, we aimed to recapitulate how cells grow in vivo in three dimensions.
NRP1 expression is associated with melanoma progression and invasiveness. In addition, studies have shown that these properties can be inhibited by the use of anti NRP1 antibodies or shRNA constructs. 34 Thus, we hypothesize that NRP1 is a potential target for the treatment of the metastatic melanoma.
In our study, we have used A375P (malignant melanoma) cells that express NRP1 (data not shown).
A375P spheroids were embedded in collagen and treated with medium supplemented with 0.1% DMSO, 25 ng/mL VEGF-A, 1 (30 μM), or a combination of VEGF-A and 1 ( Figure 8A). Data collected shows that treatment with 1 in A375P cells significantly inhibited invasion induced by VEGF-A, whereas 1 treatment on its on its own had no significant effect on radial invasion compared to the DMSO control ( Figure 8B). These results further establish an important role for blocking NRP1 in regulating VEGF-A mediated signaling, which are essential for cell motility and invasion in melanoma cells.
Blocking NRP1 on Regulatory T Cells (Treg) with 1 Reduces Their Production of TGFβ in the Presence of Glioma-Conditioned Media (GCM). Nrp1 is upregulated on the surface of Treg and is vital to their maintenance. Nrp1 + Treg populations have also been shown to induce allograft tolerance and limit potential antitumorigenic responses in murine models. Depletion of Nrp1 + Treg leads to enhancement of antitumoral immune responses, making them a favorable population of cells to target pharmacologically. 6,35 To determine whether 1 had the potential to block the protumorigenic polarization of Nrp1 + Treg, we isolated and purified Nrp1 + , FoxP3 + , and CD25 + populations of Treg from mice ( Figure 9A) and exposed them to glioma conditioned media (GCM) 12 for 12 h after pretreating the cells with 1. TGFβ is normally present in GCM 36,37 and contributes to the immunosuppressive tumor microenvironment because interference with TGFβ expression has been shown to strongly promote recognition of glioma cells by cytotoxic T cells and NK cells. 37 Treatment of the Nrp1 + Tregs with GCM alone activated the Tregs, which resulted in further increased TGFβ

Journal of Medicinal Chemistry
Article cytokine production, while pretreatment of the cells with 1 inhibited the GCM-induced production of TGFβ ( Figure 9B).

■ CONCLUSION
A focused set of novel NRP-1 antagonists were designed using structure-based drug design to allow targeting of specific residues located close to the binding pocket of arginine. X-ray crystallography was able to confirm that these interactions were being formed and enabled the design of further analogues. Compounds were tested in several different assays to confirm binding to NRP1 and inhibition of NRP1−VEGF complex formation. Of these new inhibitors, compound 1 shows consistent biological activity and good stability in vivo. It exhibits submicromolar potency in inhibition of VEGF-A binding to NRP1 and good functional inhibition of VEGF driven angiogenesis, cell migration, tumor invasiveness, and notably Treg cell activation. The compound also demonstrates a sustained IV PK profile, making it an exciting new proof-ofconcept molecule for the investigation of NRP-1 antagonists as anticancer therapies (Table 5).

■ EXPERIMENTAL SECTION
Materials and Methods. Chemistry. All materials were obtained from commercial suppliers and used without further purification unless otherwise noted. Anhydrous solvents were either obtained from Aldrich or Fisher Scientific and used directly. All reactions involving air-or moisture-sensitive reagents were performed under a nitrogen atmosphere. Routine analytical thin layer chromatography was performed on precoated plates (Alugram, SILG/UV254). Reaction analyses and purity were determined by reverse-phase LC-MS using an analytical C18 column (Phenomenex Luna C18 (2) 50 mm × 4.6 mm, 5 μm for 4.5 and 13 min methods), using a diode array detector and an A:B gradient starting from 95% A:5% B at a flow rate of 2.25 mL/min or 1.5 mL/min, where eluent A was 0.1% formic acid/H 2 O and eluent B was 0.1% formic acid/MeOH or eluent A was 10 mM NH 4 HCO 3 (aq) and eluent B was MeOH. Silica gel chromatography was performed with prepacked silica gel Biotage SNAP (KP-Sil) cartridges. Ion exchange chromatography was performed using Isolute Flash . Compound 1 inhibits VEGF-A stimulated tyrosine phosphorylation of VEGF-R2/KDR. HUVECs were grown to confluence and serum-starved with medium containing 0.5% serum for 16 h. Cells were preincubated for 30 min with medium containing 0.1% DMSO (Veh) 3, 10, and 30 μM 1 or medium alone followed by stimulation with 1 ng mL −1 VEGF-A or with no further treatment (control) for 10 min. Cell lysates were then prepared, blotted, and probed with the indicated antibodies. The data shown are representative of three independent experiments. Quantitation of pVEGF-R2/KDR phosphorylation was performed by densitometry using ImageJ; see Materials and Methods. Data are presented as pVEGF-R2/KDR phosphorylation relative units (RU; means ± SEM) normalized to total VEGF-R2/KDR; p < 0.05 = * and p < 0.001 = ***.

Journal of Medicinal Chemistry
Article SCX-2 cartridges. Reverse-phase preparative HPLC was carried out on a Waters ZQ instrument using mass-directed purification on a preparative C18 column (Phenomenex Luna C18 (2), 100 mm × 21.2 mm, 5 μm). Depending upon the retention time and the degree of separation of the desired compound from any impurities, an A:B gradient was employed starting from high %A/low %B at a flow rate of 20 mL/min. The following combinations of A and B were typically used: A = H 2 O + 0.1% formic acid and B = MeOH + 0.1% formic acid, or A = 10 mM NH 4 HCO 3 (aq) and B = methanol. 1 H and 13 C spectra were measured with Bruker NMR spectrometers as indicated. All observed protons are reported as parts per million (ppm) and are aligned to the residual solvent peak, e.g., for DMSO-d 6 at δ H 2.50 and δ C 39.5 and for CDCl 3 at δ H 7. 26. Data are reported as follows: chemical shift, multiplicity (s = singlet, d = doublet, t = triplet, br = broad, m = multiplet), coupling constants (J) recorded in Hz, and a number of protons. Low-resolution mass spectrometry data were determined on Waters ZQ4000 single quadrupole, Micromass Ultima triple quadruple mass spectrometers or Agilent 6100 single quadrupole/1200 series. High-resolution mass spectroetry was determined using an Orbitrap. All compounds tested (bioassays) were determined to be at least 95% pure by LC-MS unless otherwise stated.

Journal of Medicinal Chemistry
Article diluted with water (30 mL) and then acidified to pH 1 with 6 M hydrochloric acid. Ethyl acetate (200 mL) was added to the resulting suspension and, after thorough mixing, the organic layer separated. The aqueous layer was further extracted with ethyl acetate (150 mL), and the organic extracts were combined, washed with brine (saturated aqueous solution; 3 × 75 mL), dried over magnesium sulfate, filtered, and the solvent removed in vacuo. The product (pale-yellow foam, 2.42 g, 100%) was used without further purification. LC 25.

Journal of Medicinal Chemistry
Article (200 mL). The organic extracts were combined, washed with brine (saturated, aqueous solution; 2 × 100 mL), dried over magnesium sulfate, filtered, and the solvent removed in vacuo. The crude product (typically a yellow solid; approx 1.5 g) was purified by flash column chromatography on silica gel (eluent: dichloromethane increasing to dichloromethane/methanol; 75:25) to afford the desired product.

Journal of Medicinal Chemistry
Article Data Collection and Processing. Low resolution (2.8 Å) data was collected at the Institute for Structural and Molecular Biology X-ray crystallography facilities equipped with Rigaku Micromax 007 generator and a Rigaku Saturn 944+ CCD detector. The 600 images were collected with a 0.5°oscillation which were processed using d*TREK. 38 High resolution (0.9 Å) data was obtained at Soleil, Paris. Diffraction data was collected at the PROXIMA 1 beamline with a DECTRIS PILATUS 6 M detector. The 1800 images of 0.1°o scillations were collected and processed using XDS software. 39 Molecular replacement was performed using Phaser 40 with the NRP1-b1 domain structure (PDB 3I97) as a model. Refinement of the lower resolution structure was performed using Phenix. 41 The high resolution structure was initially refined with Phenix before completing the refinement using ShelXL. 42 Model building of both structures was performed using COOT. 43 Data quality and refinement statistics are shown in Table S1 (Supporting Information).
SPR Experimental. Surface plasmon resonance experiments were performed using a Biacore 4000 instrument at 25°C. Sensor chips, buffer stock solutions, and immobilization reagents were from GE Healthcare. Recombinant human NRP1-b1 was as above. Other reagents were obtained from Sigma. Immobilization: PBS (containing 0.05% surfactant P20) was used as the running buffer. The four flow cells were treated in the same way to optimize throughput. In summary, using a CM5 chip spots 1 and 2 were activated with the coupling reagents EDC and NHS for 10 min. NRP1-b1 at a concentration of 20 μg/mL in 10 mM sodium acetate pH 5 was injected onto the surface for 10 and 5 min in spots 1 and 2, respectively, to generate surfaces with high and low density. The immobilization levels ranged from 2302 to 1823 RU on spot 1 and from 948 to 1112 RU in spot 2. The unmodified spot 3 was used as a reference. Kinetics and affinity measurements: PBS containing 0.05% surfactant P20 and 3% DMSO was used as the running buffer and sample dilution buffer. Dose−responses were obtained using a 2-fold sample dilution from 16 μM to 31 nM, using an injection time of 60 s. Surface regeneration between injections was not necessary, but a wash step with 1 M NaCl was included after injection of the highest concentration sample for each compound. Data processing: Binding curves were corrected for variations in DMSO concentration and normalized by molecular weight. Binding results to high and lowdensity surfaces were processed independently and the average ± SD is presented. K D s reported are derived from steady-state binding responses and therefore correspond to the equilibrium binding affinity of the compounds.
Isothermal Titration Calorimetry. Isothermal titration calorimetry (ITC) experiments were carried out in a reaction buffer (20 mM Tris pH 7.9, 50 mM NaCl) using a MicroCal iTC200 system (Malvern) at 20°C. Prior to the experiments, the Nrp1-b1 sample was dialyzed in the reaction buffer and all solutions, including the buffer that was used for heat dilution measurements, were degassed and filtered just before loading into the calorimeter. Then  Pharmacokinetics. To test compound drug-like properties, selected compounds with low IC 50 were further tested for pharmacokinetics (PK) profile. First, 6−8 week-old BABL/c female mice were used, then 2 mg/kg of compounds was formulated in 7.5% DMSO and 92.5% solution and intravenously dosed from the tail vein as a bolus. Blood samples were collected by cardiac puncture at 5, 15, 30, 60, 180, and 240 min post dosing. Plasma samples were prepared by centrifugation at 7000 rpm for 5 min, and supernatants were collected, immediately snap-frozen on dry ice, and stored at −20C. Samples were analyzed by liquid chromatography−tandem mass spectrometry using electrospray ionization and data was analyzed by WinNonlin software.
btVEGF-A Cell-Free Binding Assay. The 96-well plates were precoated with NP1 protein at 3 μg/mL overnight at 4°C. On the following day, the plates were treated with blocking buffer (PBS containing 1% BSA) and washed three times with wash buffer (PBS containing 0.1% Tween-20). The various concentrations of compounds diluted in PBS containing 1% DMSO were added, followed by addition of 0.25 nM of bt-VEGF-A 165 . After 2 h of incubation at room temperature, the plates were washed three times with wash buffer. The bound bt-VEGF-A 165 to NP-1 was detected by streptavidin− horseradish peroxidase conjugates and the enzyme substrate, and measured using a Tecan Genios plate reader at A 450 nm with a reference wavelength at A 595 nm. Nonspecific binding was determined in the absence of NP-1 coated wells of the plates.
Immunoblotting. Cells were lysed with RIPA buffer (Sigma) supplemented with protease inhibitor (Roche) and phosphatase inhibitors II and III (Sigma) and analyzed by SDS-PAGE with 4− 12% Bis·Tris gels (Nupage, Invitrogen), followed by electrotransfer onto Invitrolon PVDF membranes (Invitrogen). Membranes were blocked with nonfat dry milk (5% w/v) and Tween-20 (0.1% v/v) in tris-buffered saline (TBS-T), for 1 h at room temperature before being probed with the primary antibody by overnight incubation at 4°C, followed by incubation for 1 h at room temperature with a horseradish-peroxidase-linked secondary antibody (Cell Signaling) and detection with the aid of Clarity Western ECL Substrate (BioRad), by the manufacturer's protocol. Immunoblots were quantified by scanning the films and then performing densitometry by use of ImageJ (US National Institutes of Health; http://rsb.info. nih.gov/ij/). Differences between three concentrations of 1 were evaluated by the two-way analysis of variance (ANOVA) with Sidak's multiple comparisons test. Values represent means ± SEM determined from the results of three independent experiments. A value of p < 0.05 was considered statistically significant.
Cell Culture. Human umbilical vein endothelial cells (HUVECs) were obtained from TCS Cell Works (Buckingham, UK) and were cultured on tissue culture-coated flaks in endothelial basal medium (EBM; Cambrex BioScience Ltd., Nottingham, UK) supplemented with gentamycin−ampicillin, epidermal growth factor, and bovine brain extract (Singlequots; Cambrex) and 10% fetal bovine serum (FBS) (complete EBM). For experimental purposes, fully confluent HUVECs at passages 1−3 were preincubated overnight with 1% FBS in EBM prior to addition of factors and other treatments. DU145/ cells.Ad.NRP1 cells: DU145 prostate cancer cells were from ATCC and infected with an adenovirus expressing WT NRP1 as previously described 27 (cells were grown in 10% FCS in DMEM). A375P melanoma cells were from ATCC. Cells were grown in 10% FCS in DMEM.
Human dermal fibroblasts (HDF) were obtained from TCS cells Works Buckingham, UK) and were cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin (Gibco). For experimental purposes, fully confluent HDF at passages 1−3 were preincubated overnight with serum-free DMEM prior to other procedures. GL261 cells were purchased from ATCC and cultured as described. 44,45 Transwell Migration. Transwell cell culture inserts made of transparent, low pore density polyethylene terephthalate (PET) with 8 μm pore size (Falcon; BD Biosciences, Oxford, UK) were inserted into a 24-well plate. Serum-free media supplemented with or without 25 ng/mL VEGF-A, 0.1% DMSO, or 1 (30 μM) were placed in the bottom chamber, and HUVECs in suspension (1.5 × 10 5 cell/well in serum-free EBM) were added to the top chamber of a 24-well plate and incubated at 37°C for 4 h. HUVECs that had not migrated or had only adhered to the upper side of the membrane were removed before membranes were fixed and stained with a Reastain Quik-Diff kit (IBG Immucor Ltd., West Sussex, UK) using the manufacturer's protocols. Plates were allowed to dry overnight and HUVECs that had migrated to the lower side of the membrane were counted in three random fields per well.

Journal of Medicinal Chemistry
Article Coculture. In vitro angiogenesis was determined by using a coculture assay. Briefly, HDF cells were grown to confluence in 24well plates in 10% DMEM, 1% penicillin/streptomycin. Medium was replaced, and 10000 HUVECs were plated on top of the fibroblast layer cultured in complete endothelial growth medium supplemented with 1% FBS. HUVECs, or medium with 0.1% DMSO, VEGF 25 ng/ mL, 1 (30 μM), and 1 (30 μM) + VEGF 25 ng/mL and fibroblasts were propagated in coculture for 4 days at 37°C and 5% CO 2 . Cells in coculture were fixed in absolute ethanol for 1 h at room temperature, washed twice in PBS, and blocked using PBS 5% milk. HUVECs were identified by incubating with anti-von Willebrand factor antibody (Dako, 1:1000) in PBS 5% milk overnight at 4°C. Antibody was removed and cells washed with PBS. The secondary antibody, goat anti-rabbit IgG, Alexa Fluor 488 conjugate (Life Technologies, 1:1000), was added on cells and left for 1 h in the dark. Solution was removed and plate was scanned using IncuCyte.
Photomicrographs of von Willebrand factor-stained cocultures were analyzed using ImageJ software. The Network area, length of all tubular structures, and the number of branching points were measured in four representative microscopic fields per well.
Scratch Assay. The experiment was performed using the ESSEN IncuCyte system. Graduated 96-well plates from ESSEN were used to seed HUVECs. When cells reached 95% confluence, a wound was made on every well using the Wound Maker 96 instrument (ESSEN instruments). Wells were supplemented with 1% EGM containing 0.1% DMSO, 25 ng/mL VEGF-A, and/or 1 (30 μM). Cell migration toward the wounds was monitored every 2 h and analyzed by the IncuCyte software.
Ex Vivo Aortic Ring Sprouting Assay. This protocol was adapted from previous studies. 32 All animal and tissue procedures were carried out in accordance with United Kingdom Home Office regulations and guidance. Briefly, female wild-type C57Bl/6 mice were killed in accordance with United Kingdom Home Office regulations. Thoracic aorta was harvested from aortic arch. Aorta was placed into a sterile Petri dish containing Opti-MEM Glutamax (Life technologies) and 1% penicillin/streptomycin. Under the dissection micrioscope, aortas were cleaned by sharp dissection and the vessel sliced into 0.5 mm rings with a scalpel. Rings were serum starved overnight at 37°C in 5 mL of OptiMEM Glutamax supplemented with 1% penicillin/streptomycin. On ice, 1.37 mL of purified type 1 rat-tail collagen (Millipore, Watford, United Kingdom) was mixed with 0.5 mL of 10× DMEM (Gibco) and 3.13 mL of dH 2 O before adding 2 μL/ml of 5 M NaOH. A 55 μL amount of this embedding matrix was pipetted per well into a 96-well plate and aortic ring submerged within. Plates were left for 15 min at room temperature before incubation for 60 min at 37°C. A 150 μL amount of OptiMEM Glutamax containing 2.5% FBS and 1% penicillin/streptomycin was added per well with medium containing 0.1% DMSO, 25 ng/mL VEGF-A, and/or 1 (30 μM). Aortic rings were incubated at 37°C for 7 days with a medium change on day 3 and 5. Wells were washed with 150 μL of PBS containing 2 mM CaCl 2 and 2 mM MgCl 2 and fixed in 4% formalin for 30 min. The collagen was permeabilized with three 15 min washes with PBS buffer containing 2 mM MgCl 2 , 2 mM CaCl 2 , and 0.25% Triton X-100. Rings were blocked in 30 μL of 1% BSA in PBLEC (PBS containing 100 μM MnCl 2 , 1% Tween-20, 2 mM CaCl 2 , 2 mM MgCl 2 ) for 30 min at 37°C . Then 100 μL of a mix containing IsolectinB4 (1:100, Vector Laboratories) and anti smooth muscle, SMA (1:100, Sigma) was added per well, followed by overnight incubation at 4°C. Wells were washed three times with 100 μL of PBS containing 0.1% Triton X-100 and then with 100 μL of sterile water. Aortic rings were imaged and the area of sprout growth was quantified using ImageJ software.
Spheroid Assay. Spheroids were generated using the methocellulose technique as described previously. Briefly, cells were trypsinized and counted. A mix containing 10 mL of methocellulose-containing medium (30% metho-cellulose, 70% culture medium) was prepared and kept on ice. Approximately 5000 cells were added to the mix. Spheroids were produced by pipetting 100 μL of the cell suspension into a well of a 96-well round-bottomed nontissue-culture plate and incubating for 24 h (37°C, 5% CO 2 ). Spheroids were collected and embedded in a mix containing 700 μL of collagen type I (3.1 mg/mL), 200 μL of 5× DMEM, 100 μL of H 2 O, and supplemented with 0.1% DMSO, 25 ng/mL VEGF-A, and/or 1 30 μM. Spheroids were allowed to invade for 7 days, followed by fixation in 4% formaldehyde. Spheroid invasion was determined by measuring the circular area of the spheroid core and the rim of invasion using ImageJ. The rim of invasion was determined as the circular distance from the edge of the core to the edge of contiguous invading cells.
Isolation and Treatment of Nrp1 + Regulatory T Cells. Primary murine splenocytes were isolated from 6-week-old C57/Bl6 mice. Briefly, mice were heavily anesthetized with avertin (0.02 mg/g ip) and perfused with 1× PBS. Spleens were dissected, chemically digested in papain for 15 min at 37°C, minced, and dissociated in PBS supplemented with 1% FBS and 1 mM EDTA. Cell suspensions were filtered through 40 μm filters, washed thoroughly, and a Dynabeads Untouched Mouse CD4 Cells kit (Invitrogen) was used to purify CD4+ cells following the manufacturer's protocol. The cells were then purified further for Nrp1 expression using positive selection for Nrp1 using anti-Nrp1 antibody (Biolegend) and then using magnetic Dynabeads (Invitrogen) to pull out this fraction. Cells were plated at a density of 200000 cells per well. Plates were precoated with anti-CD3ε (clone 145-2C11, BD Biosciences) overnight at 4°C. Cells were cultured in 250 μL of RPMI supplemented with 10%FBS, 1× penicillin/streptomycin, and anti-CD28 antibody (clone 37.51, BD Biosciences) for costimulation. Cells were allowed to expand for 72 h, at which point they were treated with either DMSO control (untreated) or 500 nM 1 for 2 h. Media was then supplemented with 50% glioma-conditioned media (GCM) isolated from confluent plates of GL261 cells (ATCC) 44,45 that were serum starved for 24 h. Cultures were allowed to sit for 12 h, after which cells were isolated for flow cytometric analysis of CD4 (eBioscience), CD25 (Biolegend), Nrp1 (Biolegend), and FOXP3 (BD Biosciences) expression using a BD LSR Fortessa flow cytometer (BD Biosciences). Medium was isolated and TGFβ release by cells was quantified using a TGFβ Ready Set Go ELISA kit (eBioscience) following the manufacturer's protocol.

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10

Journal of Medicinal Chemistry
Article Notes