Background
Breast cancer (BrCa) is a heterogeneous disease, with genomic profiling studies having identified five major human breast cancer intrinsic subtypes according to PAM50 classification (luminal A, luminal B, HER2-positive, basal-like and normal-like), which differ in their molecular profiles, incidence and prognosis [
1]. The most recently classified subtype, claudin-low, was identified in 2007 and is characterized by low expression of cell adhesion proteins such as claudins-3, -4, -7, occludin and E-cadherin, and activation of epithelial-mesenchymal transition (EMT) and mammary stem cell pathways [
2,
3]. Claudin-low tumors have the least amount of epithelial differentiation across the breast cancer subtypes and are the subtype that most closely resembles mammary epithelial stem cells [
2‐
4]. Previous studies, including our own [
5‐
7], have revealed claudin-low cell lines and tumors to be enriched in cancer stem cell (CSC) features, overexpressing CSC markers including CD44 and ALDH1, and having low expression of CD24 [
3,
8,
9]. Recent analysis suggests that rather than representing a distinct intrinsic subtype, in most cases the claudin-low signature is an acquired secondary phenotype that overlaps with established PAM50 subtypes, typically the basal-like and normal-like groups [
10]. Indeed, claudin-low tumors appear to comprise three major subgroups (claudin-low 1, 2 and 3; CL1-3) arising from distinct cells of origin; malignant transformation of normal mammary stem cells (CL1), and EMT-mediated transformation of luminal or basal-like tumors (CL2 and CL3, respectively) [
11]. Aside from mesenchymal stem cell properties, claudin-low tumors also exhibit dysregulation of p53 and RAS-MAPK pathways across all cell-of-origin subsets [
11]. Claudin-low tumors as a group have a poor response to standard chemotherapy and shorter relapse-free survival and overall survival, although the prognosis of patients with claudin-low tumors reflects the prognosis associated with their underlying intrinsic subtype [
3,
10].
The claudin-low subtype is most enriched for basal-like tumors (51.7%), and 25–39% of triple-negative tumors are claudin-low [
3,
10]. While hormone therapy and HER2-targeting drugs have improved prognosis of patients with estrogen/progesterone receptor-(ER/PR) and HER2-positive BrCa, respectively, triple-negative breast cancers (TNBCs), including the majority of basal-like tumors, lack all three of these receptors and are poorly responsive to available targeted therapies. Chemotherapy is the mainstay treatment for TNBC, but resistance develops quickly, and apart from of a minor subset of tumors with BRCA mutations that can be treated with PARP inhibitors such as Olaparib (LYNPARZA), there is no targeted therapy for TNBC. These tumors are generally more aggressive than ER/PR- and HER2-positive tumors, while also having a higher incidence in younger women [
12]. Despite advances in chemotherapy regimens, TNBC patients still have a relatively poor prognosis with higher recurrence and metastasis rates, and lower survival probability, than other subtypes [
13]. Therefore, development of new therapeutic targets for TNBC represents a major unmet clinical need.
Neuropilin-1 (NRP1) is a pleiotropic transmembrane co-receptor protein critical in embryonic development of neurological and vascular systems [
14]. It has a short cytosolic segment whose function in intracellular signaling remains unclear, while its extracellular domains mediate interactions with multiple growth factors to promote activation of their cognate receptor tyrosine kinases [
15]. Recent evidence suggests that NRP1 activates a broader spectrum of growth factor pathways than formerly thought, including EGF, VEGF, PI3K, HGF, PDGF, FGF and TGF-β1 [
16‐
18]. These interactions have implicated NRP1 in cancer progression across multiple tumor types [
15], where high NRP1 expression is associated with poor outcome in lung [
19], glioblastoma [
20], prostate [
21] and breast cancers [
22].
We investigated the prognostic value, expression and function of NRP1 across BrCa subtypes. We report that high NRP1 expression predicts shortened time to disease relapse and metastasis in ER-negative patient cohorts and is associated specifically with the claudin-low subtype. NRP1 knockdown in claudin-low cell lines led to significant reduction in cell proliferation and growth of orthotopic claudin-low xenografts. NRP1 over-expression associated with the most de-differentiated claudin-low tumors and was required to maintain high expression of ZEB1 and the mammary stem cell marker ITGA6. Targeted inhibition with an NRP1 monoclonal antibody reduced spheroid-forming capacity and potently suppressed tumor growth in an orthotopic claudin-low TNBC xenograft model. Finally, we show that NRP1 acts as a central hub for the aberrant activation of the RAS-MAPK pathway via EGFR and PDGFR activation to drive aggressive tumor progression, a hallmark of all claudin-low subgroups. These data identify NRP1 as a key driver of the claudin-low phenotype and support further testing of NRP1 inhibitors for improved control of claudin-low tumor progression.
Materials and methods
Survival analysis
NRP1 expression correlation with relapse-free survival (RFS) and distant metastasis-free survival (DMFS) was analyzed using Kaplan–Meier Plotter software (
https://kmplot.com/analysis/index.php?p=service&cancer=breast) with the following non-default settings selected; Affymetrix ID / Gene symbol; 212298_at, Split patients by; upper quartile, Survival; RFS (4934) or DMFS (2767), Probe set options; JetSet best probe set, ER status—array (
n = 7535); all, ER positive (5526) or ER negative (2009) [
23,
24]. The expression of NRP1 in the Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) data set and relevant clinical parameters was downloaded from the University of California Santa Cruz (UCSC) Xena data portal (
https://xena.ucsc.edu). For Kaplan–Meier analysis, patients were ranked by NRP1 expression from lowest-to-highest then divided into quartile groups (Q1-Q4), with Q4 being the highest expressing patients, and median months overall survival computed per quartile group of patients.
In silico analysis
Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) and The Cancer Genome Atlas (TCGA) datasets were accessed through cBioPortal and UCSC XENA [
25] [
26]. NRP1 expression was correlated to claudin-low up- and down-signature scores with RStudio (version 3.5.1) GSVA package [
27] using published claudin-low gene signatures [
3,
10]. Core claudin-low (CoreCL) signature was obtained from Fougner et al. [
10]. Heatmaps were generated using Morpheus software (Broad Institute).
The Prediction Analysis of Microarrays (PAM) R package (pamr) was used to classify the three claudin-low subtypes using the ‘nearest shrunken centroids’ algorithm as reported by Pommier et al. [
11]. The prediction model was trained based on expression of the 137 classifier genes in a subset of 45 claudin-low samples, as provided by Pommier et al. The trained model was applied to the 199 claudin-low samples from the METABRIC dataset to classify them into CL1-3 subtypes for NRP1 expression analysis. The ggplot and ggplot2 R packages were used to generate the graphical presentation for the claudin-low subtype analysis. NRP1 expression in BrCa cell lines was analyzed using transcriptomic data from Neve et al. [
28].
Cell culture
MDA-MB-231, BT-549, MCF-7, T47D, HS578T and SUM159 BrCa cell lines were sourced from the American Type Culture Collection (ATCC; Manassas, VA, USA). MDA-MB-468, BT-20, MDA-MB-361, BT474, SKBR3 and HCC1569 cells were obtained from the ATCC and transferred from Lombardi Cancer Center, USA, by Prof. Erik Thompson (QUT). All cell lines were authenticated by short tandem repeat analysis at the Genomics Research Centre (Queensland University of Technology, Australia). SKBR3, HCC1569 and T47D cells were cultured in RPMI 1630 (Gibco) supplemented with 10% FBS (Gibco). MDA-MB-231, SUM159P, HS578T, BT-549, MDA-MB-468, MDA-MB-361 and BT474 cells were cultured in DMEM (Gibco) with 10% FBS. MCF-7 cells were cultured in DMEM (Gibco) with 10% FBS and 10 ug/mL insulin (Gibco).
Western blotting
Cells were harvested in lysis buffer (1% Triton-X, 150 mM NaCl, 1 mM EDTA, 50 mM Tris, pH 8) containing protease and phosphatase inhibitors (cOmplete Protease Inhibitor Cocktail and PhosSTOP; Roche) on ice for 30 min. Protein concentration was determined by the Pierce BCA Protein Assay Kit (Bio-Rad). Protein samples were prepared using Bolt LDS Sample Buffer and Reducing Agent (Thermo Fisher). Samples were heated at 70 °C for 10 min and separated on 4–12% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gels (Thermo Fisher). Proteins were transferred onto a nitrocellulose membrane using the Mini Blot Module (Thermo Fisher Scientific). Membranes were blocked in Tris-buffered saline containing 0.05% Tween 20 (TBS-T) and 5% non-fat powdered skim milk for one hour, then incubated in primary antibody overnight as follows; NRP1 (Santa Cruz Biotechnology, C-19/sc-7239, 4 µg/ml), GAPDH (Cell Signaling Technology, D16H11/5174), EGFR (Cell Signaling Technology, D38B1/4267), phospho-EGFR (Cell Signaling Technology, D7A5/3777), γ-tubulin (Sigma Aldrich, T5326, 1 µg/ml), ZEB1 (Sigma Aldrich, HPA027524), p42/44 (Cell Signaling Technology, 137F5/4695), phospho-p42/44 (Cell Signaling Technology, D13.14.4E/4370), ITGA6 (Cell Signaling Technology, 3750), PDGFR (Cell Signaling Technology, D1E1E/3174) or phospho-PDGFR (Cell Signaling Technology, 4547) according to manufacturer’s instructions. After washing with TBS-T, horseradish peroxidase-conjugated secondary antibodies were applied for one hour at room temperature in 5% skim milk and immunodetection performed using Immobilon TM Western Chemiluminescent HRP Substrate (Millipore). Chemiluminescent signal was visualized using the Konica SRX-101A film processor or Chemidoc Gel Imaging System (BioRad). Densitometry was performed using Image Studio Lite software.
Quantitative PCR
RNA was extracted with the Direct-zol RNA miniprep kit (Zymo Research) before reverse transcription with the SensiFAST cDNA Synthesis kit (Bioline). Quantitative PCR was performed using SYBR Green (Thermo Fisher) using QuantStudio 6 Real-Time PCR System (Thermo Fisher). Gene expression was determined by the comparative Ct method, and normalized to the housekeeping gene RPL32. Expression in NRP1 knockdown samples was normalized to NT control. The following primer sequences (5’-3’) were used: NRP1 (forward; CGAGGGCGAAATCGGAAAAGG, reverse; CTTCGTATCCTGGCGTGCT), ZEB1 (forward; CAACTACGGTCAGCCCT, reverse; GCGGTGTAGAATCAGAGTC), ITGA6 (forward; CCTCTTCGGCTTCTCGCTG, reverse; CGTGGGGTCAGCATCGTTA).
Flow cytometry
For NRP1 expression analysis, cells at 80% confluence in 75 cm2 cell culture flasks were washed with ice-cold PBS and dislodged with Accutase (Gibco). Cells were neutralized with PBS/5% FBS and resuspended to 106 cells/ml. Anti-human APC-conjugated NRP1 antibody (R&D System, FAB3870A) or IgG-APC control (R&D System, IC003A) was added and cells incubated on ice for 45 min. Cells were centrifuged, resuspended in PBS/5% FBS and propidium iodide (100ug/ml) added. Fluorescence compensation and analysis was performed on a BD Accuri C6. Results were analyzed with Kaluza software. The X-axis parameter Median (‘X-Med’; the 50th percentile of a population, representing the value at which half of a measured population is above and the other half below) was used to represent fluorescence intensity. For CD44/CD24 based cell sorting, 20 × 106 cells were grown to 80% confluence, washed twice in PBS and detached with Accutase (Thermo Fisher). After washing and resuspension (106 cells/ml) in PBS/5% FBS, the cell suspension was incubated with anti-human CD44-Alexa Fluor 488 (FAB6127G, R&D Systems)) and CD24-APC (FAB5247A, R&D Systems) for 1 h on ice, then washed three times in PBS/5% FBS. Propidium iodide (3 μl, 100 μg/ml) was added to the cells immediately before loading on a MoFlo Astrios Cell Sorter (Beckman Coulter) to allow for viable cell gating.
siRNA and shRNA knockdown
BrCa cells were reverse transfected using Lipofectamine RNAiMax reagent (Thermo Fisher Scientific) and 5 nM Silencer Select siRNA (Ambion) according to manufacturer’s instructions. The following NRP1 siRNA sequences were used (sense, 5’-3’): siNRP1 [
1]; uaaccacauuucacaagaa, siNRP1 [
2]; cagccuugaaugcacuuau. A pre-designed Silencer Select non-targeting (NT) siRNA was used as a negative control (Negative Control siRNA No. 1, #4,390,844, Thermo Fisher Scientific). For western blotting, protein lysate was collected 72 h post-transfection. For constitutive knockdown, an NRP1 shRNA pLKO.1 lentiviral vector with NRP1 anti-sense sequence 5ʹ-AATACTAATGTCATCCACAGC-3ʹ was used, with the control sequence targeting firefly luciferase (shCntrl) obtained from Thermo Fisher. Viral particles were produced as previously described [
5].
Proliferation assays
Cell viability was measured at 0, 1, 2, 4 and 7 days post-siRNA treatment. A total of 5,000 cells were plated per well of a 96-well plate after mixing with transfection solution as described. Viability was assessed using the CyQuant Direct Cell Proliferation Assay (Life Technologies) or Incucyte S3 (Essenbioscience) according to manufacturer’s instructions. For day 0 measurement, cell viability was quantified 6 h after cells were plated.
Spheroid assays
Spheroid formation assays were carried as described previously [
29,
30]. Briefly, 1,200 BrCa cells in single cell suspension were seeded into ultra-low attachment 96-well plates (Corning) in standard growth medium containing 1% methylcellulose, 1 × B-27™ supplement (Thermo Fisher Scientific), EGF (20 ng/mL), bFGF (20 ng/mL) and heparin (4 μg/mL). Vesencumab or IgG was added to wells to a final concentration of 50 µM and replenished daily. Whole well images were captured (4x) every 24 h by the Incucyte® S3 Live-Cell Analysis System. Image resolution was set as 2400 × 2400 pixels with 2.8 pixel to μm ratio. Images were exported and spheroid number analyzed by CellProfiler (
https://cellprofiler.org/) based on spheroid diameter ≥ 50 μm or ≥ 18 pixels.
Experimental animals
All animal studies were carried out with approval from the University of Queensland Animal Ethics Committee (ethics approval number QUT/TRI/026/18). Eight-week-old female NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ) mice were purchased from the Animal Resources Centre (ARC), Western Australia, and acclimatized for one week at the Biological Resources Facility (BRF), Translational Research Institute, before beginning experiments. Mice were maintained on standard irradiated rat and mouse diet (Specialty Feeds) and provided a 12 h light/12 h dark cycle. Twice a week, mice received sunflower seeds as a supplement. Mouse health was monitored at least three times weekly by researchers and daily by BRF staff. All procedures were performed in a sterile laminar flow cabinet.
Orthotopic claudin-low breast cancer xenograft models
1 × 106 SUM159 cells constitutively expressing luciferase (SUM159luc) in 30 µL Matrigel (Corning) were injected into the left inguinal mammary fat pad of 8-week-old female NOD SCID gamma (NSG) mice. To test the effects of shRNA NRP1 knockdown, SUM159luc transfected with shNRP1 or shNT were used for injection (10 mice per group). Mice in the shRNA study were allowed to progress to ethical endpoint (tumor volume 1000 mm3). To test the effects of the NRP1 inhibitor Vesencumab, mice injected with SUM159luc cells were randomly allocated to IgG control or Vesencumab treatment groups (10 mg/kg by twice weekly intraperitoneal injection until study endpoint; 12–14 mice per group), for 7 weeks. Bioluminescence imaging of tumors was performed with the IVIS Spectrum In Vivo Imaging System (Perkin Elmer) one-week post-xenografting to confirm presence of live tumor cells. Tumor size was measured by digital caliper twice weekly until endpoint, when bioluminescence imaging of tumors was repeated prior to tissue collection. Vesencumab and IgG control were provided by Genentech Inc.
Immunohistochemistry
Paraffin sections were dewaxed, rehydrated and then underwent antigen retrieval in 1 mM EDTA buffer (pH 9) in a microwave oven (700 W) for 16 min. Endogenous peroxidase block was performed for 15 min in 3% hydrogen peroxide, followed by 3 washes of 10 min each with phosphate buffered saline (PBS) containing 0.1% Tween-20 (PBS-T). Sections were blocked for 30 min with CAS-block reagent (Zymed Laboratory), followed by 3 PBS-T washes. Slides were incubated with the following primary antibodies diluted in Dako antibody diluent at 4 0C for 16 h as follows; NRP1 (Sigma Aldrich, HPA030278, 4 µg/ml), Ki67 (Agilent, MIB-1/M7240), CD31 (Abcam, ab28364) and phospho-p42/44 (Cell Signaling Technology D13.14.4E/4370), with Ki67, CD31 and phospho-p42/44 antibodies diluted according to manufacturer’s instructions. Slides were washed, incubated with EnVision + Dual Link System HRP (Agilent) for 30 min, washed again and developed in Liquid DAB + Substrate Chromogen System (Agilent). Slides were counterstained with Mayer’s hematoxylin, dehydrated and mounted in DPX. Slides were scanned by the Panoramic Digital Slide Scanner (3DHISTECH) and analyzed with CaseViewer software using QuantCenter (3DHISTECH). The PatternQuant module was used to discriminate epithelial, stromal and necrotic compartments. DensitoQuant (NRP1, CD31) or NuclearQuant (Ki67, phospho p42/44) modules were used for quantification of IHC staining. For NRP1 IHC quantification, the following DensitoQuant settings were used: Detection (Brown Tolerance; 1.2, Blue Tolerance; 0.98), Score (Weak Positive Intensity; 220, Moderate Positive Intensity; 180, Strong Positive Intensity; 150). For phospho p42/44 quantification, the following NuclearQuant settings were used: Nucleus Detection (blur; 15, radius; 3–8, min area; 10), Nucleus Filters (intensity; 60, contrast; 30), Score (0/negative; 255–200, + 1/weak; 200–164, + 2/medium; 164–120, + 3/strong; 120–0).
Receptor tyrosine kinase (RTK) array and Vesencumab treatment
Cells were seeded into T25 flasks and serum-starved for 16 h when they reached 70% confluence. Cells were pre-incubated with Vesencumab or IgG control (50 μg/mL) for 1 h followed by receptor activation by addition of culture medium containing 20% FBS for 60 min. For NRP1 knockdown study, cells were transfected and seeded as described above. Two days post-transfection, cells were serum starved for 24 h followed by receptor activation by addition of culture medium containing 20% FBS for 0, 10 or 60 min. Protein lysates were then collected and antibody array binding performed according to manufacturer’s instructions (Proteome Profiler Human Phospho-RTK Array Kit, ARY001B, R&D Systems). Antibody membranes were scanned by ChemiDoc (Bio-Rad) and densitometry performed using Image Lab software (Bio-Rad).
Statistical analyses
Differences between two groups were compared using unpaired Student’s t test. For multiple group comparisons, one-way analysis of variance (ANOVA) followed by Dunnett’s post hoc test was used; for survival analysis, log-rank (Mantel-Cox) test was used to assess the significant differences among treatment and control groups. For in silico analyses, Chi-square test was used to assess the significance between subtypes, and linear regression analysis was used to calculate the significance between NRP1 expression and gene score. Tumor Control Index (TCI) was used to estimate the tumor size for one mouse (shCntrl group; shNRP1 in vivo study) which was culled early for health reasons before to reaching ethical tumor volume endpoint. The Srivastava Lab kindly provided the VBA macro with graphical user interface (SL TCI) that calculates TCI scores, based on tumor rejection, regression and stability scores [
31].
Discussion
Treatment of triple-negative breast cancer is currently limited by a lack of targeted therapies. A substantial proportion (25–39%) of triple-negative tumors classify as claudin-low. We report high NRP1 expression to be associated specifically with the claudin-low molecular subtype of breast cancer. This correlation was not due to immune or stromal cell infiltration in claudin-low tumors, as NRP1 expression correlated with a ‘core’ claudin-low signature enriched in EMT and cancer stem cell markers that characterize the claudin-low subtype [
10].
NRP1 expression associated more strongly with the claudin-low up, rather than down, gene signature score, which is enriched in mesenchymal and stem cell markers. NRP1 expression was highest in CL1 subtype tumors, which most closely resemble mammary stem cells [
11]. In the CL1 cell line Hs578T, NRP1 regulates expression of the EMT transcription factor ZEB1, which has been shown to be aberrantly over-expressed in the CL1 subtype, predisposes to malignant transformation in the absence of high levels of chromosomal instability and mediates mammary stem cell characteristics [
11,
41,
42]. NRP1 was more highly expressed in a self-renewing CD44
+/CD24
low population and conferred spheroid-initiating potential to claudin-low cells. Consistent with a known role in mediating integrin function in breast and other cell types, NRP1 was required to maintain expression of the mammary stem cell marker ITGA6 (CD49f) [
43‐
45]. These data strongly suggest a role of NRP1 in normal mammary stem cell function, which is dysregulated following neoplastic transformation, supported by the finding that NRP1 depletion induces defects in mammary epithelial cell development [
46].
The NRP1-targeted antibody Vesencumab was able to potently inhibit the growth of SUM159 orthotopic xenografts, whereby the mean volume of Vesencumab-treated tumors at the 7-week study endpoint was 12.8-fold smaller than IgG treated tumors. The main effects of Vesencumab were likely due to direct anti-tumor effects on tumor cells and largely independent of changes to angiogenesis, as no difference in CD31-positive intratumoral vasculature was observed between Vesencumab and IgG treated tumors. In our model, Vesencumab was administered at the time of tumor xenografting until study endpoint to establish the effects of NRP1 inhibition on tumor latency and growth kinetics, as suggested by its association with a mammary stem cell population. In future experiments, it will be important to determine the tumor-suppressive potential of Vesencumab when administered to animals with established tumors or after tumor resection, to more closely resemble the timing of therapeutic intervention in future clinical trial testing.
As all the claudin-low cell lines used in this study were basal-like, further studies assessing the efficacy of NRP1 inhibition across claudin-low models of different intrinsic subtypes will also be of interest. As high NRP1 expression was particularly associated with poor prognosis in ER negative tumors, we suggest that the greatest clinical utility of NRP1 inhibition is likely to be in patients with claudin-low TNBC.
Evidence is accumulating that aberrant activation of the RAS/MAPK pathway is a key mechanism capable of driving transition to a claudin-low phenotype [
11,
47]. Despite demonstrably high RAS pathway signaling in a substantial proportion of breast tumors, RAS mutations occur relatively infrequently in human breast cancers, suggesting that engagement of the RAS pathway may be driven by alternate cooperative mechanisms [
48]. Here, we show that NRP1 drives MAPK signaling in claudin-low cells via activation of upstream receptor tyrosine kinases. Changes in phospho p42/44 staining in Vesencumab-treated tumors suggested that NRP1 may regulate the RAS/MAPK axis in tumor vascular walls and endothelial-like cells. However, future experiments to validate NRP1 regulation of endothelial cell function and vascular mimicry in claudin-low tumors via the RAS/MAPK axis will be required to confirm or disprove these speculations.
NRP1 has recently been implicated in resistance to cancer therapies including oncogene-targeted therapies and chemotherapy through activation of bypass survival pathways, including receptor-tyrosine kinase pathways such as HER2, EGFR and IGF1R [
18,
21,
43]. Thus, blocking NRP1-mediated activation of a spectrum of therapy-induced bypass survival pathways beyond just RAS/MAPK, either as a single agent or alongside standard of care chemotherapy regimens, may provide superior control over adaptive resistance mechanisms and improve the durability of therapy responses.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.