Pilot assessment of vascular endothelial growth factor receptors and trafficking pathways in recurrent and metastatic canine subcutaneous mast cell tumours

Abstract Canine subcutaneous mast cell tumour (scMCT) shows less aggressive biological behaviour than cutaneous MCT. Vascular endothelial growth factor receptor 2 (VEGFR2) is expressed by neoplastic cells in canine scMCT, but the relevance of this signalling pathway for disease pathobiology is not clear. The objective of this study was to quantify VEGF‐A, VEGFR2, pVEGFR2, the VEGF co‐receptor Neuropilin 1 (NRP‐1) and the E3 ubiquitin protein ligase c‐Cbl in canine scMCT, and to evaluate their association with disease outcome. Immunohistochemical staining for biomarkers was quantified from 14 cases of canine scMCT using manual and computer‐assisted methods. Kaplan–Meier curves were generated for disease‐free survival (DFS) and compared using Mantel–Cox log‐rank analysis. Cases with high levels of neoplastic cell VEGFR2, pVEGFR2 or c‐CBL immunoreactivity had significantly reduced DFS. All cases displayed neoplastic cells positive for VEGF‐A, which was significantly associated with pVEGFR2 immunoreactivity. There were also significant positive correlations between VEGFR2 and pVEGFR2, and between c‐CBL and pVEGFR2 levels. This pilot study demonstrates the potential utility of these markers in a subset of scMCT in dogs.


Introduction
Mast cell tumours (MCTs), account for 7-21% of all canine skin cancer (Welle et al. 2008) and there are subcutaneous and dermal (also called cutaneous) types. The events resulting in MCT development are unknown, but breed predisposition is well documented and may indicate an underlying genetic component (Warland & Dobson 2013). Activation of the receptor tyrosine kinase KIT plays a role in normal mast cell growth and development, as well as in malignant transformation (London et al. 1999;Webster et al. 2007;Warland & Dobson 2013). Studies with the c-KIT gene have indicated a key role in dermal MCT development and progression (London et al. 1999;Jones et al. 2004;Webster et al. 2007;Warland & Dobson 2013;Patruno et al. 2014). The KIT protein, encoded by the proto-oncogene c-KIT, stimulates the proliferation, maturation and activation of normal mast cells (Yavuz et al. 2002). Its role in subcutaneous MCT is less well understood (Thompson et al. 2011a(Thompson et al. , 2016. Toceranib and masitinib are receptor tyrosine kinase inhibitors specifically designed to treat canine MCT and represent the first targeted therapies for canine cancer patients (London 2013). Toceranib targets KIT, VEGFR, PDGFR and Flt-3, while masitinib targets KIT and PDGFR (London 2009). As these tyrosine kinase inhibitors have 1 These authors contributed equally to this work. multiple targets, interpretation of their activity in a clinical setting is challenging. For instance, masitinib significantly improved long-term survival in dogs with non-resectable MCTs, regardless of KIT mutation status (Hahn et al. 2010). Up to 60% of dogs with MCT given a grade of III do not have detectable KIT mutations and no KIT mutations have been reported in subcutaneous MCT (Hahn et al. 2010; Thompson et al. 2011a). Given the complex interplay between signalling pathways in these neoplastic cells, further investigation of other potential prognostic and therapeutic targets would be informative.
Vascular endothelial growth factor (VEGF) is an important regulator of endothelial cell proliferation, vasculogenesis, angiogenesis and vascular permeability, mediated primarily by the tyrosine kinase receptor VEGFR2 (Ferrara et al. 2003;Folkman 2006). Canine MCT, both dermal and subcutaneous, are reported to express VEGFR2 (Rebuzzi et al. 2007;Thompson et al. 2016) and mast cell tumour cells express both VEGF ligand and VEGFRs (Sekis et al. 2009) although whether autocrine signalling pathways are active in mast cell tumours is controversial. Neuropilin-1 (NRP-1) is a VEGF co-receptor expressed on endothelial cells and some neoplastic cells, acting to modulate VEGF signalling by regulating VEGFR2 recycling, at least in endothelial cells (Soker et al. 1998(Soker et al. , 2002Kawasaki et al. 1999;Hong et al. 2007;Adham et al. 2010;Zhang & Simons 2014).
Casitas B-lineage lymphoma proto-oncogene (Cbl) proteins are a highly conserved family of E3 ubiquitin ligases which mono-ubiquitinate their targets KIT, VEGFRs and PDGFRs (along with other receptor molecules), routing them for destruction via multivesicular bodies/lysosomes (Masson et al. 2006;Ryan et al. 2006;Swaminathan & Tsygankov 2006;Meyer et al. 2011). Loss of c-CBL led to development of mastocytosis in transgenic mice associated with defective ubiquitination and subsequent accumulation of KIT and disrupted signalling (Bandi et al. 2009;Orinska et al. 2010). KIT expression and signalling was also found in human acute myeloid leukaemia cells with c-Cbl mutations .
Subcutaneous MCT develops in the subcutaneous region, and in most cases, is well defined and delimited by fatty tissue (Newman et al. 2007). While the majority of canine subcutaneous MCT are low grade and biologically benign, some show local recurrence and metastasis (Newman et al. 2007;Thompson et al. 2011a,b;Risselada et al. 2015). Due to the limited information on prognostic factors in canine subcutaneous MCT, and the potential for RTK inhibition as a therapeutic strategy, this research aims to evaluate the prognostic significance of VEGFR2, NRP-1 and c-CBL protein markers in a subset of canine subcutaneous MCT.

Case selection and case characteristics
Paraffin blocks from 14 canine subcutaneous MCTs were used for this pilot study. There were 11 female spayed and three male neutered dogs in this study. The age at diagnosis ranged from 5 to 13 years, with a median age of 7 years. There were 13 purebred dogs (six Labrador Retrievers and seven other purebred dogs) and one mixed breed dog. Neoplastic cells were within 1 mm of the surgical margins in each tumour. Three dogs developed metastasis to lymph node and all dogs eventually developed local recurrence at the original surgical site. No dogs received chemotherapy and one dog received radiation for recurrent disease.

Histology and immunohistochemistry
Paraffin sections (5 lm thickness) were placed on Superfrost+ slides for histology and immunohistochemistry (IHC). Slides were deparaffinized in xylene and subsequently rehydrated in a graded isopropanol series. For immunohistochemistry, antigen retrieval was performed in sodium citrate buffer (pH 6.0) in a 95°C water bath for 15 min. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide for 30 min and non-specific binding sites were blocked with a 1:1 ratio of 5% normal goat serum (Life Technologies, Burlington, ON, Canada) and DAKO block (Dako, Carpinteria, CA, USA) for 2 h at room temperature followed by overnight incubation at 4°C with the anti-VEGFR2 (1:1600); anti-phospho-VEGFR2 (1:125); anti-NRP-1 (1:11 600) or anti-c-CBL (1:200), followed by washing and incubation in biotinylated anti-rabbit IgG (1:200; Vector Laboratories, Burlington, ON, Canada) for 1 h, at room temperature. Peroxidase-streptavidin complex (1:200; Cedarlane, Burlington, ON, Canada) was then applied for 1 h at room temperature, and peroxidase activity visualized with 3,3 0 -diaminobenzidine tetrahydrochloride (DAB; Vector Laboratories). Negative controls consisted of slides incubated with no primary antibody and slides incubated with an irrelevant primary antibody (rabbit polyclonal anti-pan-cytokeratin; Fig. S1). Positive controls consisted of vascular endothelial cells for VEGFR2, pVEGFR2 and NRP-1 (internal control) and canine testis for c-CBL (Fig. S1). Slides were counterstained with Harris haematoxylin, dehydrated and coverslipped. Immunohistochemical labelling for VEGF-A was conducted by the Michigan State University Diagnostic Center for Animal and Population Health (Lansing, MI, USA). Additional slides were stained with haematoxylin and eosin, and used to assess mitotic count, numbers of multinucleate cells and tumour growth pattern as previously reported (Thompson et al. 2011b).

Image analysis
The slides were viewed using a compound light microscope (DMLB100 T, Leica Microsystems, Wetzlar, Germany) and five images were captured from each section with QCapture software (QImaging, Surrey, BC, Canada) at high-power (40X objective) fields. Representative areas were systematically selected for immunostaining analysis, avoiding regions of necrosis, edges of sections and regions lacking neoplastic cells. The images were captured and analysed in a blinded fashion, and clinical information was not available until after image analysis was completed. Staining was evaluated by establishing a 'threshold' using Image J software (ImageJ v1.47; NIH, Bethesda, MD, USA), and determining the percent positively stained area for each image. The threshold was established for each case based on staining of positive control cells. Reaction product associated with vascular endothelium was subtracted from images immunostained for VEGFR2, pVEGFR2 or NRP-1 as appropriate. The percent positive area score for each case was calculated by taking the average of all five fields, and the median positive area was determined for all 14 cases. Based on that median, cases were divided into high expression tumours and low expression tumours and the outcome was compared between both groups. For VEGF-A immunoreactivity, slides were scored in a blinded fashion using a modification of published methods (Rebuzzi et al. 2007): +++, > 90% neoplastic cells strongly reactive; ++, > 90% neoplastic cells positive; +, > 50% neoplastic cells positive; +/À, < 50% neoplastic cells positive.
Kaplan-Meier curves were generated and Mantel-Cox log-rank analysis was used to analyse diseasefree survival (DFS) defined as interval between diagnosis and detection of disease progression (local recurrence and/or metastasis). Linear regression analysis was performed to compare quantification for VEGFR2 and pVEGFR2, and for c-CBL and pVEGFR2 immunoreactivity, and differences in pVEGFR2 expression according to VEGF expression were compared by Mann-Whitney test for unpaired analysis. The cut off points for mitotic count and multinucleate cells were established as previously reported (Thompson et al. 2011b). For all analysis, a value of 95% (P ≤ 0.05) was used for statistical significance.

Results
Western blotting analysis was performed to validate the cross reactivity of the primary antibodies with canine proteins (Fig. 1). VEGFR2 and pVEGFR2 antibodies identified a major band at approximately 250 kDa, and minor bands of slightly lower molecular weight, as has been reported for human and canine neoplastic cells Thompson et al. 2016). The c-CBL blots showed a single specific band of approximately 100 kDa (Feshchenko et al. 1998), while NPR-1 blots demonstrated full length and truncated isoforms, as has been reported for human NRP-1 (Gagnon et al. 2000). The same antibodies were then employed for immunolocalization of antigens in FFPE samples of canine subcutaneous MCT.
The immunohistochemical protocols developed were successfully used on all 14 MCT samples with minimal background staining (Figs. 2, 3). Negative controls involved replacing the primary antibody with PBS, and use of an irrelevant immunogen (pancytokeratin) (Fig. S1). VEGFR2 and pVEGFR2 immunostaining of neoplastic MCT cells was both membranous (green arrows) and cytoplasmic, and vascular endothelial cells in blood vessels (asterisks) were strongly positive (Fig. 2a,b). Positive nuclear immunolocalization of neoplastic cells was also occasionally seen (black arrows). VEGF immunolocalization was predominantly cytosolic (Fig. 2c). The VEGF co-receptor NRP-1 showed predominantly cytoplasmic with some membranous immunostaining in neoplastic cells (green arrow); nuclear immunolocalization was also occasionally present (black arrow, Fig. 2d). Comparison of manual counting (number of positive cells/field) vs. Image J analysis revealed a significant positive correlation (P = 0.0014; Fig. S2), thus we proceeded to use the higher throughput image analysis. Quantification of IHC staining revealed a significant positive correlation between neoplastic cell VEGFR2 and pVEGFR2 immunoreactivity (P = 0.0305; Fig. 2e). VEGF scores were grouped into 'low' (+/À and +) and 'high' (++ and +++), and pVEGFR2 expression levels were compared for each group. Cases with high VEGF scores had significantly higher pVEGFR2 levels (P = 0.0293; Fig. 2f).
Scatter plots showing range of positive immunoreactivity scores for VEGFR2, pVEGFR2, NRP-1 and c-CBL are shown in Figure S3. Cases were stratified into 'Low' and 'High' based on median values for percent positive neoplastic cell pixel immunoreactivity. Low expression of VEGFR2 or pVEGFR2 by MCT neoplastic cells was significantly associated with increased DFS (P = 0.0476 and P = 0.0289, respectively; Table 1). There were no statistical differences between NRP-1 expression and DFS (Table 1). The E3 ubiquitin ligase enzyme c-CBL showed diffuse to intense cytoplasmic immunolocalization, with occasional nuclear staining (arrows) in MCT neoplastic cells (Fig. 3a). Low expression of c-CBL by MCT neoplastic cells was significantly associated with increased DFS (P = 0.0029; Table 1). There was also a significant positive correlation between neoplastic pVEGFR2 and c-CBL immunolocalization (P = 0.0117; Fig. 3b). Significant differences in DFS of this case set were not seen with tumour growth pattern or presence of multinucleated cells (Table 1). Mitotic count >4 in 10 hpf was significantly associated with reduced DFS (P = 0.0100; Table 1), as has been reported Thompson et al. (2011b), but outcome stratified by c-CBL staining achieved higher statistical significance and displayed greater Hazard Ratios (Table 1).

Discussion
Here, we report that a subset of recurrent canine subcutaneous MCTs were heterogeneous in their expression of VEGFR2, with membranous, cytoplasmic, or nuclear immunolocalization in neoplastic cells. In parallel, we detected phosphorylated (activated) VEGFR2 in similar cellular compartments. In vascular endothelial cells, membrane-associated VEGFR2 undergoes tyrosine phosphorylation upon ligand binding, leading to the activation of signalling pathways that affect the proliferation, survival, permeability and migration of the cell (Domingues et al. 2011). Nuclear localization of pVEGFR2 in endothelial cells regulates its own transcription and potentially increases the angiogenic response (Domingues et al. 2011). Cytosolic localization of VEGFR2 may be indicative of high levels of VEGFR2 stored in intracellular vesicles, awaiting signalling to direct its trafficking to the plasma membrane (Manickam et al. 2011). Neoplastic cells from a variety of cancer types use VEGFR2 as an autocrine growth factor, and may employ a similar process to that observed in endothelial cells (von Marschall et al. 2000;Strizzi et al. 2001;Sher et al. 2009;Adamcic et al. 2012;Darrington et al. 2012). Levels of neoplastic cell VEGFR2 expression were positively associated with lymph node status and a trend towards shorter  survival time in canine mammary carcinoma (Diessler et al. 2016). Here, we found a significant positive correlation between neoplastic cell expression of VEGFR2 and of pVEGFR2. Since positive VEGF-A immunoreactivity was seen in all cases, it suggests a possible role for VEGFR autocrine signalling in the pathobiology of canine subcutaneous MCT. Our results are also in line with a recent study that reported VEGFR2 expression (detected via western blot) in neoplastic cells isolated from some canine cutaneous and subcutaneous MCT (Thompson et al. 2016). These findings may help explain the success of KIT directed TKIs in tumours that lack mutated KIT ).
Recently, alterations in the KDR (VEGFR2) gene have been reported for some human cancers. Over one-third of human malignant melanomas are reported to contain a Q472H germline variation, located in the fifth IgG-like repeat (Silva et al. 2016). BLAST alignment of the human VEGFR2 amino acid sequence with the putative canine version [EMBL nucleotide sequence database #AAEX03009081] shows 93% homology, with the relevant glutamine residue conserved between human and canine. Amplification of the KDR gene (≥4 copies) is also reported in human non-small cell lung cancer (Nilsson et al. 2016). Cancer cell lines displaying KDR copy number gain showed enhanced VEGFR2 protein levels and activation of alternative signalling pathways compared to NSCLC cells lacking copy number amplification. The KDR mutation analysis and copy number status and their relationship with VEGFR expression in canine cutaneous MCTs is worth evaluating in future studies.
Neuropilins (NRPs) belong to the large group of axon guidance molecules, playing key roles in migration of neural crest cells and axonal processes during development. In addition to their involvement as coreceptors for angiogenic signalling via VEGF, NRPs are also implicated in human cancer progression, including leukaemia (Grandclement & Borg 2011). High NRP-1 expression by neoplastic cells was associated with lymph node metastasis and poor prognosis in human oral squamous cell carcinoma (Chu et al. 2014). Overexpression of NRP-1 induced enhanced invasiveness in malignant melanoma cells (Bai et al. 2015) and induction of epithelial to mesenchymal transition (EMT) and cancer stem cell phenotype in oral squamous cell carcinoma (Chu et al. 2014). The group of MCT studied expressed NPR-1, but this did not correlate with disease-free survival, suggesting this is not a good prognostic factor for subcutaneous MCT.
c-CBL mutations are found in human leukaemia where they are implicated in disrupted KIT signalling (Tefferi 2010;Makishima et al. 2012;Schwaab et al. 2012;Traina et al. 2012). Acting as a tumour suppressor, loss of CBL expression could confer a more aggressive phenotype; yet, in our study, we found high c-CBL levels were associated with worse DFS. CBL proteins can also be involved as adaptor molecules independent of their E3 ligase activity, leading to activation of PI3K, AKT and MAPK and cytoskeletal pathways, among others (Swaminathan & Tsygankov 2006;Liyasova et al. 2015). While most CBL mutations lead to loss of E3 ligase activity, such mutant proteins can maintain their signal-enhancing functions (Schwaab et al. 2012). In support of this, we found a significant positive correlation between levels of neoplastic cell pVEGFR2 and c-CBL in these canine subcutaneous mast cell tumours. Interestingly, it was recently reported that high levels of c-CBL expression were significantly associated with poor outcome in human glioma (Jing et al. 2016).
Canine subcutaneous mast cell tumours are distinct from dermal lesions in both their location and biological behaviour. The prognosis of subcutaneous MCT can be better determined using recognized histological parameters such as mitotic count, Ki67 and AgNOR score and immunohistochemical KIT expression pattern (Thompson et al. 2011a). In this pilot study, we extend the evaluation of canine subcutaneous mast cell tumours and determine that VEGFR2 and c-CBL expression could further stratify aggressive lesions for improved prognosis. The study is limited by the small sample size and retrospective nature of the case selection. However, our findings support exploration of VEGFR as a potential target in recurrent or metastatic subcutaneous MCTs. Future approaches could potentially include use of current KIT targeted inhibitors that may have anti-VEGFR activity, and/or therapies that directly target VEGFR signalling. The impact of the biomarkers evaluated here on disease pathobiology, and their potential utility in less aggressive subcutaneous mast cell tumours or for cutaneous mast cell tumours, is currently under investigation.

Source of funding
This study was supported by grants from the Ontario Veterinary College Pet Trust Fund (#051425) and the São Paulo Research Foundation (FAPESP #2014/25583-1).

Supporting information
Additional Supporting Information may be found online in the supporting information tab for this article: Figure S1. Control procedures for immunohistochemistry. Figure S2. Regression analysis illustrating the correlation between scoring. Figure S3. Scatter plots of Mean Pixel area for parameters evaluated.