The fibrinolysis inhibitor α₂-antiplasmin restricts lymphatic remodelling and metastasis in a mouse model of cancer

Abstract

Remodelling of lymphatic vessels in tumours facilitates metastasis to lymph nodes. The growth factors VEGF-C and VEGF-D are well known inducers of lymphatic remodelling and metastasis in cancer. They are initially produced as full-length proteins requiring proteolytic processing in order to bind VEGF receptors with high affinity and thereby promote lymphatic remodelling. The fibrinolytic protease plasmin promotes processing of VEGF-C and VEGF-D in vitro, but its role in processing them in cancer was unknown. Here we explore plasmin’s role in proteolytically activating VEGF-D in vivo, and promoting lymphatic remodelling and metastasis in cancer, by co-expressing the plasmin inhibitor α₂-antiplasmin with VEGF-D in a mouse tumour model. We show that α₂-antiplasmin restricts activation of VEGF-D, enlargement of intra-tumoural lymphatics and occurrence of lymph node metastasis. Our findings indicate that the fibrinolytic system influences lymphatic remodelling in tumours which is consistent with previous clinicopathological observations correlating fibrinolytic components with cancer metastasis.

Keywords:

• Cancer
• lymphatic remodelling
• metastasis
• plasmin
• protease
• VEGF-D

Introduction

Lymphatic remodelling, including lymphangiogenesis (the formation of new lymphatic vessels) and enlargement of lymphatics, is associated with tumour development in humans and has been shown to facilitate the metastatic spread of cancer via the lymphatic vasculature in animal models (Alitalo, 2011; Dadras et al., 2003; He et al., 2005; Mandriota et al., 2001; Shayan et al., 2012; Skobe et al., 2001; Stacker et al., 2001, 2014). Hence the remodelling of lymphatic vessels is considered a potential therapeutic target for restricting metastasis (Achen et al., 2006, 2005; Alitalo, 2011; He et al., 2002; Heckman et al., 2008; Jain & Padera, 2002; Lin et al., 2005). The secreted protein growth factors VEGF-C and VEGF-D induce lymphatic remodelling and metastatic spread in animal models of cancer and have been correlated with metastasis in a range of human tumour types (Karnezis et al., 2012; Karpanen et al., 2001; Mandriota et al., 2001; Skobe et al., 2001; Stacker et al., 2001, 2002, 2004). Furthermore, signalling for lymphatic remodelling in tumours and lymph nodes may regulate the immune microenvironment in cancer (Lund et al., 2016), and VEGF-C has been reported to promote immune tolerance to tumours (Hendry et al., 2016; Lund et al., 2012; Swartz, 2014). Hence VEGF-C signalling pathways may be relevant targets for novel agents aimed at enhancing immunotherapeutic approaches to cancer. Given the potential importance of VEGF-C and VEGF-D for promoting cancer metastasis, immunomodulation and tissue oedema (Sato et al., 2016), a range of agents targeting signalling by these growth factors has recently completed early phase clinical trials in cancer and other disease settings (Saif et al., 2016).

VEGF-C and VEGF-D are initially produced as full-length forms which can be proteolytically processed to remove N- and C-terminal propeptides, thus generating mature forms consisting of dimers of the central VEGF homology domain (Baldwin et al., 2001; Joukov et al., 1997; Stacker et al., 1999). These mature forms can bind the endothelial cell surface receptors VEGFR-2 and VEGFR-3 with high affinity, in contrast to the full-length forms (Joukov et al., 1997; Stacker et al., 1999). By activating VEGFR-2 and VEGFR-3, mature forms of VEGF-C and VEGF-D drive proliferation and migration of endothelial cells (Achen et al., 1998; Davydova et al., 2016; Makinen et al., 2001), and remodelling of lymphatics and blood vessels (Byzova et al., 2002; Paquet-Fifield et al., 2013; Rissanen et al., 2003; Wise et al., 2003), which in the setting of cancer facilitate tumour growth and spread (Harris et al., 2011, 2013). These studies indicated that the proteolytic activation of VEGF-C and VEGF-D can be a key control point in the molecular regulation of lymphatic remodelling and metastasis in cancer. However, the proteases important for activation of VEGF-C and VEGF-D in cancer are unknown.

We previously conducted a proximity assay-based molecular screen for proteases capable of activating VEGF-D, which indicated that the fibrinolytic serine protease plasmin could cleave both propeptides from full-length VEGF-D, thereby activating this growth factor, and could activate VEGF-C (McColl et al., 2003). These findings were intriguing given that the fibrinolytic system, which involves a variety of proteases and protease inhibitors that ultimately regulate conversion of plasminogen to plasmin (Longstaff & Kolev, 2015), can also participate in sculpting the tumour microenvironment (Heissig et al., 2016; Kwaan & McMahon, 2009). Importantly, components of this system, such as urokinase-type plasminogen activator (uPA) which converts plasminogen to plasmin, can be associated with metastatic spread and poor patient outcome in a range of human cancers [for review, see Mekkawy et al. (2014)]. However, our proximity protease assay was conducted in vitro, so the capacity of plasmin to activate VEGF-C and VEGF-D in the setting of cancer in vivo remained unknown. Moreover, evidence for a mechanistic link between the fibrinolytic system and lymphatic remodelling in cancer, based on in vivo studies, was lacking. Here we explore the role of plasmin in activating VEGF-D in a tumour model in mice by co-expressing α₂-antiplasmin (α₂-AP), the principal physiological inhibitory regulator of plasmin in vivo (Law et al., 2008; Schaller & Gerber, 2011), with VEGF-D in tumour cells. We report that α₂-AP restricts the processing of VEGF-D, remodelling of tumour lymphatics and metastasis in this model which indicates a novel mechanism by which the fibrinolytic system could regulate lymphatic remodelling and the metastatic spread of cancer.

Materials and methods

Cell lines

A derivative of the 293-EBNA cell line stably harbouring an APEX-3 plasmid construct expressing a form of human full-length VEGF-D tagged at the N-terminus with the FLAG octapeptide (this protein was previously designated VEGF-D-FULL-N-FLAG) has been described and was designated ‘VEGF-D-293’ (Stacker et al., 1999, 2001). An alternative 293-EBNA derivative harbouring the APEX-3 expression vector lacking DNA for VEGF-D was previously designated ‘293’ (Stacker et al., 1999, 2001). VEGF-D-293 and 293 cells were stably transfected, using FuGENE 6 transfection reagent (Promega, Madison, WI), with a pcDNA3.1Zeo expression plasmid derivative containing a cDNA for mouse α₂-AP (Law et al., 2008), or with pcDNA3.1Zeo (Thermo Fisher, Waltham, MA). Colonies were selected in Zeocin (0.4 mg/ml), and clones strongly expressing α₂-AP were identified by immunoprecipitation followed by Western blotting (see section below entitled Immunoprecipitation and Western blotting). The clonal cell line stably expressing human VEGF-D-FULL-N-FLAG and mouse α₂-AP was designated ‘VEGF-D/α₂-AP’; the line expressing VEGF-D-FULL-N-FLAG but not mouse α₂-AP was designated ‘VEGF-D/Zeo’; the line expressing mouse α₂-AP but not VEGF-D-FULL-N-FLAG was designated Apex/α₂-AP. The VEGF-D/α₂-AP and Apex/α₂-AP cell lines secreted comparable levels of α₂-AP which exhibited comparable bioactivity (Figure 1(A,B)).

Figure 1. Expression of α₂-AP in cell lines and tumours. (A) 293-EBNA cell lines expressing VEGF-D and α₂-AP (VEGF-D/α₂-AP), VEGF-D but not α₂-AP (VEGF-D/Ζeo) or α₂-AP but not VEGF-D (Apex/α₂-AP) were analysed for α₂-AP (top) and VEGF-D (bottom) by immunoprecipitation from conditioned media and Western blot. Expected size of α₂-AP is 67 kDa under these reducing conditions; α₂-AP band is indicated by arrow and position of molecular weight marker (in kDa) is shown. For VEGF-D blot, ‘1’ denotes conditioned media were undiluted for immunoprecipitation; ‘0.1’ denotes media were diluted 1/10. Identities of bands are schematically indicated to right of blot. ‘F’ denotes FLAG tag; ‘N-pro’ and ‘C-pro’, N- and C-terminal propeptides, respectively; ‘VHD’, VEGF homology domain. VEGF-D was not detected in Apex/α₂-AP cells (data not shown). (B) Conditioned media were assessed for α₂-AP activity. In the assay, plasmin cleaves chromogenic substrate causing increase in absorbance over time; α₂-AP prevents this increase. Capacity of pure recombinant α₂-AP (rα₂-AP) and conditioned media from VEGF-D/α₂-AP or Apex/α₂-AP cells to prevent increase in absorbance is apparent. DMEM is negative control. (C) Tumours were established in flanks of mice (n = 7 for VEGF-D/α₂-AP; n = 9 for VEGF-D/Zeo; n = 8 for Apex/α₂-AP) – morphology of primary tumours was similar between study groups (one example for each of VEGF-D/α₂-AP and VEGF-D/Zeo study groups is shown after skin was pealed back from body wall). (D) α₂-AP was immunoprecipitated from lysates of primary tumours and detected by Western blot (left). Results for three tumours from the VEGF-D/α₂-AP and VEGF-D/Ζeo study groups are shown as examples. Arrow indicates α₂-AP band and dotted line indicates where irrelevant tracks were deleted from image. The range of apparent molecular weight shown in the image spans from ∼82 to ∼57 kDa. For further comparison of α₂-AP levels, relative levels of α₂-AP were determined based on analysis of all tumours from the three study groups by immunoprecipitation and Western blotting. Given the large number of tumours involved, multiple Western blots were conducted and scanned, each of which included a positive control track with the same amount of recombinant α₂-AP. Graph (right) shows mean intensity ± SEM for α₂-AP bands with intensity of positive control defined as 100%. * indicates statistically significant difference as assessed by one-way analysis of variance with Tukey’s post hoc test.

Mice

Female SCID/NOD mice (6–8 weeks of age) were obtained from the Animal Resource Centre, Perth Australia. All experiments performed on mice were in accordance with the Animal Ethics Committee of the Peter MacCallum Cancer Centre and with the guidelines set by the National Health and Medical Research Council of Australia.

Assays of α₂-AP activity

VEGF-D/α₂-AP, VEGF-D/Zeo and Apex/α₂-AP cells were maintained in culture under identical conditions, and matched volumes of conditioned media from these cell lines were harvested, snap-frozen in dry-ice and stored at −80 °C. After thawing, matched volumes of conditioned media were incubated with human plasmin (catalogue number HCPM-0140, Haematologic Technologies Inc., Essex Junction, VT) at 2.5 nM and chromogenic substrate S-2251 (Chromogenix/Diapharma Group Inc., West Chester, OH) at 0.12 mM for 100 min at 37 °C in an automated spectrophotometer (FLUOstar Optima plate reader, Ortenberg, Germany), and the absorbance was measured continuously at 405 nm. Recombinant human α₂-AP, purified as described previously (Lu et al., 2011), was included at 12.5 nM as positive control. Dulbecco’s Modified Eagle’s medium (DMEM) was used as a negative control, and assays were conducted three times.

Tumour model

Female SCID/NOD mice, 10 weeks of age, were injected subcutaneously in the left flank with VEGF-D/α₂-AP, VEGF-D/Zeo or Apex/α₂-AP cells (2.5 × 10⁷ cells in a volume of 200 μl of PBS). The resulting primary tumour xenografts were measured twice per week with digital callipers and tumour volumes were calculated as follows: volume = length × width² × 0.52, as described previously (Stacker et al., 2001). Primary tumours were harvested when their size reached 2500 mm³; a part of each tumour was snap-frozen in Tissue-Tek O.C.T. (Sakura Fineteck, Torrance, CA) for generation of frozen tissue sections, another part was fixed in 4% paraformaldehyde for generation of paraffin-embedded tissue sections, and the rest was frozen on dry-ice for generating tumour lysates. Axillary lymph nodes were also excised from mice and fixed in 4% paraformaldehyde for paraffin embedding. Each lymph node was sectioned (4 μm sections) through its entirety and every 10th section was collected and assessed for the presence of tumour cells by staining with haematoxylin and eosin (H&E). Tumour cells were detected based on their distinct cell morphology, i.e. they were larger and had a higher ratio of nuclear area to cytoplasmic area compared to cells normally resident in lymph nodes, as described previously (Harris et al., 2011). The number of mice analysed for the presence of tumour cells in lymph nodes varied slightly between study groups because a small proportion of mice did not develop primary tumours – mice lacking primary tumours were excluded from the analysis of lymph nodes.

Immunoprecipitation and western blotting

Immunoprecipitation of α₂-AP or VEGF-D was conducted from the conditioned media of VEGF-D/α₂-AP, VEGF-D/Zeo or Apex/α₂-AP cells that had been cultured under identical conditions. Matched volumes of conditioned media from the cell lines were used for immunoprecipitations. Immunoprecipitation of α₂-AP was conducted using a polyclonal antibody to mouse α₂-AP (catalogue number AF1239, R&D Systems, Minneapolis, MN). The α₂-AP antibody was incubated at 25 μg/ml in conditioned media overnight at 4 °C, and 50 μl of a slurry of Protein A agarose (Sigma Aldrich, St. Louis, MO) (approximately 50% Protein A and 50% PBS) was then added, and the mixture incubated on a rotating wheel at 4 °C for 2 h. The agarose was collected by centrifugation at 9600 g for 2 min at 4 °C, washed twice in 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.1% Triton X-100, twice in 0.5 M NaCl and twice in 50 mM Tris-HCl, pH 8.0, before resuspension in NuPAGE denaturing loading sample buffer (Thermo Fisher) and reducing agent (Novex by Life Technologies, Waltham, MA) and heating at 70 °C for 10 min. Samples were then electrophoresed in Bolt 4–12% Bis-Tris Plus gels (Novex by Life Technologies). Transfer of proteins to nitrocellulose was carried out with an iBlot 2 Dry Blotting System (Life Technologies, Waltham, MA), Western blotting was performed with α₂-AP antibody (AF1239; R&D Systems) at 0.2 μg/ml, and detection was with a donkey anti-goat antibody conjugated to 800 IR dye (LI-COR Bioscience, Lincoln, NB) and an Odyssey Infrared Imaging System (LI-COR Biosciences). For VEGF-D, immunoprecipitation was with A2 antiserum (Stacker et al., 1999), which targets the VEGF homology domain, and Western blotting was conducted with anti-VEGF-D monoclonal antibody MAB286 (R&D Systems) that had been coupled to 800 IR dye (LI-COR Biosciences) according to the manufacturer.

For analysis of α₂-AP or VEGF-D in tumours, lysates were prepared on ice by homogenizing up to 0.75 g of tissue in 500 μl of lysis buffer, consisting of 10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Triton X-100 with protease inhibitors (cOmplete, Minin EDTA-free, Roche Diagnostic, Basel, Switzerland) included as described by the manufacturer, using an electric Ultra-Turrax homogenizer (IKA-WERK, Breisgau, Germany). Post-homogenization, lysates were gently shaken at 4 °C for 30 min and centrifuged at 17,000g for 15 min at 4 °C. The concentration of protein in lysates was determined using a Pierce BCA Protein Assay Kit (ThermoFischer Scientific, Waltham, MA). Lysates containing 9 or 4 mg of total protein for analysis of α₂-AP or VEGF-D, respectively, were used for immunoprecipitations and subsequent Western blotting which were conducted as described above.

Comparison of band intensities for α₂-AP or different forms of VEGF-D was conducted as follows. After scanning with an Odyssey Infrared Imaging System, ‘regions of interest’ (RoI) of the same size were placed to encompass each relevant band, and the Odyssey software was used to quantify the Raw Intensity (RI) in each RoI. In addition, the background RI for a given blot was measured for an RoI that was distant from any bands. The background RI for a given blot was subtracted from the RI measurements for all relevant bands on that blot to generate corrected RIs that were used for comparing band intensities. All immunoprecipitation/Western blotting analyses, involving comparison of band intensities for α₂-AP or VEGF-D, were conducted at least twice with very similar results observed in each experiment.

Immunohistochemistry

Tissue sections (4 μm thick) were cut from paraffin-embedded tumour tissue or lymph nodes, and stained for LYVE-1 to identify lymphatic vessels. Tissue sections were stained with rabbit polyclonal anti-mouse LYVE-1 antibody (catalogue number 70 R-LR003, Fitzgerald Industries, North Acton, MA) as follows, or for VEGF-A or VEGFR-2 as described previously (Achen et al., 1997). Heat-induced antigen retrieval was carried out with Dako Target Retrieval Solution, pH 6.0, in a Dako programmable pressure cooker (Dako, Glostrup, Denmark) at 125 °C and 14 psi for 3 min. Tissue sections were allowed to cool to room temperature and endogenous peroxidase activity was then blocked by incubation in 3% H₂O₂ in methanol for 20 min at room temperature. Tissue sections were then blocked with 10% goat serum in Protein Block (catalogue number X0909, Dako) overnight at 4 °C in a humidified chamber. After discarding the goat serum/Protein Block, tissue sections were incubated with LYVE-1 antibody in Antibody Diluent (catalogue number S3022, Dako) for 1 h at room temperature, as described by the antibody manufacturer. Tissue sections were incubated with biotinylated goat anti-rabbit secondary antibody (catalogue number E0432, Dako) for 30 min. A Vectastain Elite ABC kit (Vector Laboratories, Burlingame, CA) was then used for signal amplification as described by the manufacturer. The chromogenic signal was then developed using DAB substrate (catalogue number SK 4100, Vector Laboratories) according to the manufacturer. Images were acquired using an Olympus BX61 microscope with a Spot RT3 camera and Spot Advanced Version 5 software (Spot Imaging Solutions, Sterling Heights, MI). The average size of LYVE-1-stained lymphatic vessels in tissue sections was determined by computer-assisted morphometric vessel analysis as described previously (Paquet-Fifield et al., 2013). This approach was conducted on one representative tissue section from each tumour which consisted of a longitudinal slice through the centre of the tumour in the direction if it’s longest dimension.

Bioinformatics

The Cancer Genome Atlas (TCGA) RNAseqV2 gene expression data for the genes encoding α₂-AP (gene designation is SERPINF2) and VEGF-D (gene designation is FIGF) were analysed and downloaded via cBioPortal (www.cBioPortal.org) (Cerami et al., 2012; Gao et al., 2013) for a range of different solid tumour types. Data pertaining to metastases were removed to ensure that analysis of transcript levels for α₂-AP and VEGF-D was based on primary tumours. Lymph node metastasis and patient clinical information was downloaded from the TCGA data portal (https://portal.gdc.cancer.gov).

Statistical analyses

Results were evaluated using Excel 2003 (Microsoft, Los Angeles, CA) or Graph Pad Prism 4.0 (GraphPad Software, San Diego, CA) software. Data are shown as mean ± SEM as indicated, and were analysed with a two-tailed unpaired Student’s t-test, Fisher’s exact test or one-way analysis of variance with Tukey’s post hoc test. For all analyses, p < 0.05 was considered statistically significant.

Results

A novel tumour model to study the role of plasmin in VEGF-D-mediated events in cancer

Our overall strategy was to explore the role of plasmin in modifying the activity of VEGF-D in cancer by expressing mouse α₂-AP, the major physiological inhibitor of plasmin (k_a is 4.0 × 10⁶ M⁻¹s⁻¹; stoichiometry of inhibition is 1.0) (Law et al., 2008), in tumours for which VEGF-D drives lymphatic remodelling and metastasis. α₂-AP rapidly inactivates plasmin resulting in the formation of a stable inactive complex, plasmin-α₂-AP (Carpenter & Mathew, 2008). Co-expression of α₂-AP with VEGF-D in tumour cells would allow us to monitor the contribution of plasmin to VEGF-D processing and subsequent metastasis and lymphatic remodelling in an in vivo cancer model. To this end, we exploited the VEGF-D-293 cell line (Stacker et al., 2001) that expresses a full-length form of human VEGF-D designated ‘VEGF-D-FULL-N-FLAG’ (see ‘Materials and methods’) (Stacker et al., 1999). VEGF-D-293 cells have been used previously to generate epitheliod-like tumour xenografts in immunocompromised mice which exhibited extensive intratumoural lymphangiogenesis, enlargement of lymphatics and high rates of lymph node metastasis (Karnezis et al., 2012; Stacker et al., 2001). These effects on lymphatic vessels and metastasis were shown to be dependent on VEGF-D-FULL-N-FLAG protein produced by the VEGF-D-293 cells (Stacker et al., 2001).

We stably transfected VEGF-D-293 cells with an expression construct encoding mouse α₂-AP – the secretion of α₂-AP by these cells was confirmed by immunoprecipitation and Western blotting, and the bioactivity of the α₂-AP in the conditioned medium of these cells was confirmed in plasmin inhibition assays (Figure 1(A,B)). These cells stably expressing both VEGF-D-FULL-N-FLAG and α₂-AP were designated ‘VEGF-D/α₂-AP’ cells. Further, we also generated control cells expressing VEGF-D-FULL-N-FLAG and stably harbouring the expression construct lacking DNA for α₂-AP (designated ‘VEGF-D/Zeo’ cells), and another control expressing α₂-AP, at comparable levels to VEGF-D/α₂-AP cells (Figure 1(A,B)), but which did not express VEGF-D-FULL-N-FLAG (designated ‘Apex/α₂-AP’ cells). α₂-AP was not detected in conditioned media from VEGF-D/Zeo cells (Figure 1(A)). The levels of VEGF-D-FULL-N-FLAG secreted by VEGF-D/α₂-AP and VEGF-D/Zeo cells were comparable (Figure 1(A)), whereas VEGF-D was not detected in the conditioned media of Apex/α₂-AP cells (data not shown). Moreover, the degree of proteolytic processing of VEGF-D-FULL-N-FLAG in the conditioned media of VEGF-D/α₂-AP and VEGF-D/Zeo cells was also comparable, given the similar relative abundances of the full-length (∼53 kDa), partially processed (∼31 kDa) and mature (∼21 kDa) forms of VEGF-D that were detected (Figure 1(A)). This suggests that plasmin does not play a significant role in the proteolytic activation of VEGF-D under the conditions in which these cells were maintained in vitro, and is consistent with previous findings indicating that proprotein convertases can be important for activation of VEGF-D in vitro (McColl et al., 2007).

We generated tumour xenografts with VEGF-D/α₂-AP, VEGF-D/Zeo and Apex/α₂-AP cells in SCID/NOD mice. The primary tumours grew at similar rates with no statistically significant differences in tumour size between these study groups over a 50-day period (N.B. tumours were allowed to grow to 2500 mm³ upon which tumours were harvested – most mice were sacrificed between 45 and 55 days post-injection of tumour cells). The tumours from these study groups also exhibited similar morphologies in terms of colour, shape and intratumoural or peritumoural fluid accumulation, and were similarly located in the skin of flanks (Figure 1(C)). Tumours were assessed for levels of α₂-AP by immunoprecipitation from tumour lysates and Western blotting which revealed that VEGF-D/α₂-AP tumours contained approximately 90% more α₂-AP than VEGF-D/Zeo tumours (Figure 1(D)). The α₂-AP that was detected in VEGF-D/Zeo tumours was likely derived from the blood plasma of the mice which is known to contain this protein; the concentration of α₂-AP in mouse plasma was previously reported to be 51 μg/ml (Vercauteren et al., 2012). In summary, this model system is appropriate for our study as it provides significantly higher levels of α₂-AP in VEGF-D/α₂-AP tumours, in comparison to the control VEGF-D/Zeo tumours, in a tumour background which is known to be dependent on VEGF-D for lymph node metastasis and lymphatic remodelling.

α₂-AP restricts processing of VEGF-D in vivo

VEGF-D/α₂-AP and VEGF-D/Zeo tumours were compared for VEGF-D processing by immunoprecipitation of VEGF-D from tumour lysates and detection of the distinct differentially processed forms by Western blotting (Figure 2). VEGF-D is a dimeric protein; each subunit can be in the full-length form, consisting of N- and C-terminal propeptides flanking the central VEGF homology domain containing receptor binding sites, which has an apparent molecular weight of ∼53 kDa as assessed by reducing SDS-PAGE (Stacker et al., 1999). Previous studies have also shown that a VEGF-D subunit consisting of the N-terminal propeptide and central domain (∼31 kDa) is commonly detected and that the mature subunit lacking both propeptides is ∼21 kDa (Stacker et al., 1999). Analysis of the tumours generated here revealed that the 31 and 21 kDa forms of the VEGF-D subunit were both less abundant, relative to the full-length 53 kDa form, in VEGF-D/α₂-AP tumours compared to VEGF-D/Zeo tumours (Figure 2(A)), with the ratio of the intensity of the 21 kDa band to the 53 kDa band being much lower than for VEGF-D/Zeo tumours (Figure 2(B)). Likewise, the ratio of the intensity of the 31 kDa band to the full-length 53 kDa band was much lower in VEGF-D/α₂-AP tumours (Figure 2(B)). VEGF-D was not detected in Apex/α₂-AP tumours (data not shown). These findings show that α₂-AP restricted the degree of proteolytic processing of VEGF-D in this tumour model, indicating that plasmin can play a role in activating VEGF-D in tumours in vivo.

Figure 2. Processing of VEGF-D in tumour xenografts. (A) VEGF-D was immunoprecipitated from tumour lysates and detected by Western blotting. Two examples are shown for VEGF-D/α₂-AP and VEGF-D/Zeo tumours. Identities of detected bands are schematically indicated to right of blots, and numbers in brackets to right of schematics denote apparent molecular weights for these proteins as reported previously (Stacker et al., 1999) and are consistent with blots shown here. Positions of molecular weight markers (in kDa) are shown to left. ‘F’ denotes FLAG tag; ‘N-pro’ and ‘C-pro’, N- and C-terminal propeptides, respectively; ‘VHD’, VEGF homology domain; slashed line indicates where irrelevant tracks were excised from image. Decreased abundance of smaller (i.e. 31 and 21 kDa) forms of VEGF-D subunit, relative to full-length 53 kDa form, in VEGF-D/α₂-AP tumours indicates less proteolytic processing occurred than in VEGF-D/Zeo tumours. (B) Relative abundance of different forms of VEGF-D in lysates of every tumour (n = 7 for VEGF-D/α₂-AP; n = 9 for VEGF-D/Zeo) was assessed by scanning of blots. A ratio of intensity of band for 21 kDa mature form to intensity of band for 53 kDa full-length form (21/53 ratio) was generated for each tumour; likewise, ratios for intensity of band for 31 kDa partially processed form to intensity of band for full-length form (31/53 ratio) were also generated. Graphs show mean ± SEM. * denotes p < 0.05; ** p < 0.005, Student’s t-test.

α₂-AP modulates VEGF-D-driven metastasis and lymphatic remodelling

Axillary lymph nodes associated with tumour xenografts were inspected for the presence of tumour cells which could be detected histologically, upon staining with H&E, based on their large size and distinct morphology compared to cells normally resident in the nodes (Figure 3(A,B)). This analysis revealed that lymph node metastasis was far more common in mice harbouring VEGF-D/Zeo tumours than in those harbouring VEGF-D/α₂-AP tumours – this difference was statistically significant (Figure 3(C)). Apex/α₂-AP tumours did not spread to lymph nodes which is not surprising given they do not produce VEGF-D. These findings suggest that the restriction of VEGF-D processing by α₂-AP in VEGF-D/α₂-AP tumours was sufficient to significantly compromise the capacity of VEGF-D to promote lymph node metastasis in this model.

Figure 3. Effect of α₂-AP on VEGF-D-driven metastasis to lymph nodes. Axillary lymph nodes from mice harbouring VEGF-D/α₂-AP or VEGF-D/Zeo tumours were assessed for tumour cells as described in Materials and Methods. (A) Axillary lymph nodes in mice harbouring VEGF-D/α₂-AP tumours were not typically swollen and not readily discernible at macroscopic level whereas a swollen axillary lymph node (indicated by white arrow) from a mouse harbouring a VEGF-D/Zeo tumour is clearly visible. Bottom panels show higher-power images of regions encompassed by black rectangles in upper panels. (B) Histology (H&E staining) showing absence of tumour cells in axillary lymph node from mouse with VEGF-D/α₂-AP tumour in contrast to large cluster of tumour cells (indicated by black arrows) in lymph node from mouse with VEGF-D/Zeo tumour. Bottom panels show higher-power images of regions encompassed by black rectangles in upper panels. Large tumour cells located to the right of the arrows in the higher power VEGF-D/Zeo image are apparent. Scale-bars in upper panels denote 400 μm; those in lower panels denote 20 μm. (C) Graph shows proportion of mice from each study group which exhibited lymph node metastasis, including the Apex/α₂-AP control study group. * denotes p < 0.01, Fisher’s exact test.

Lymphatic vessels in tumours can provide direct entry for tumour cells into the lymphatic vasculature as well as a means of transport to lymph nodes. The enhanced lymph node metastasis exhibited by VEGF-D/Zeo tumours compared to VEGF-D/α₂-AP tumours could therefore be a consequence of altered abundance or size of lymphatic vessels in the primary tumours. We previously showed that the formation of lymphatic vessels inside tumours, by lymphangiogenesis, in the VEGF-D-293 model system was largely dependent on VEGF-D, as VEGF-D-293 tumours had abundant lymphatics located within the tumours whereas few lymphatics were detected inside 293EBNA tumours which did not express VEGF-D (Stacker et al., 2001). To assess tumour lymphatics in VEGF-D/α₂-AP and VEGF-D/Zeo tumours, we stained tissue sections of primary tumours for LYVE-1, a protein expressed by initial lymphatic vessels (Banerji et al., 1999; Makinen et al., 2005), a subtype of lymphatic vessel that can give rise to tumour lymphatics in response to VEGF-D (Shayan et al., 2013). Intra-tumoural lymphatic vessels could be readily detected in both study groups, but the density of lymphatics was higher in VEGF-D/Zeo tumours compared to VEGF-D/α₂-AP tumours, although this difference did not reach statistical significance (Figure 4). The abundance of intra-tumoural lymphatics in Apex/α₂-AP control tumours was very low, as expected of tumours that did not produce VEGF-D. We next determined the size of intra-tumoural lymphatics by combining the area of LYVE-1-positive endothelium with the area of the lumen of each vessel. This analysis revealed that lymphatics in VEGF-D/Zeo tumours were significantly larger than in VEGF-D/α₂-AP tumours (Figure 4), suggesting that the restriction of VEGF-D processing in VEGF-D/α₂-AP tumours was sufficient to limit the enlargement of intra-tumoural lymphatic vessels. The lymphatics in Apex/α₂-AP tumours were small, similar to VEGF-D/α₂-AP tumours. The enhanced size of lymphatics in VEGF-D/Zeo tumours, compared to VEGF-D/α₂-AP tumours, provides a greater surface area for interaction of tumour cells with lymphatic endothelium – this may facilitate entry of tumour cells into lymphatics which in turn could increase the probability of metastatic spread to lymph nodes.

Figure 4. α₂-AP modulates intra-tumoural lymphatics induced by VEGF-D. Intra-tumoural lymphatic vessels in primary VEGF-D/α₂-AP and VEGF-D/Zeo tumours (n = 7 and n = 9, respectively) were identified by immunostaining for lymphatic marker LYVE-1. LYVE-1 staining (dark grey/black) is localized on endothelium lining lymphatics (top panels). Scale-bars denote 100 μm. Density of LYVE-1-positive lymphatic vessels, expressed as number of vessels per high-power field (hpf), is shown in left-side graph. Size of LYVE-1 positive lymphatics, assessed by combining area of LYVE-1-stained endothelium with area of vessel lumens, is shown in right-side graph. Graphs include results for the Apex/α₂-AP study group (n = 8). Data are mean ± SEM; * indicates statistically significant difference as assessed by one-way analysis of variance with Tukey’s post hoc test.

Relationship between α₂-AP and lymph node metastasis in human tumours

We carried out bioinformatics analysis using The Cancer Genome Atlas (TCGA) tumour datasets to explore the potential role of α₂-AP in regulating lymph node metastasis of solid human cancers expressing VEGF-D. All primary pancreatic, breast, thyroid, colorectal, bladder, prostate, liver and renal clear cell carcinomas, as well as primary head and neck squamous cell carcinomas, lung squamous cell carcinomas, lung adenocarcinomas and cutaneous melanomas represented in the RNAseqV2 gene expression data were included in the analysis (the total number of tumours was 5903). For tumours with VEGF-D mRNA levels in the upper quartile of all tumours regardless of tumour type, the subset of tumours with low levels of α2-AP mRNA were significantly more likely to spread to lymph nodes than the subset with high levels of α2-AP mRNA (Figure 5). In contrast, for tumours with lower levels of VEGF-D mRNA (i.e. those tumours in the lowest three quartiles, the lowest half or the lowest quartile for VEGF-D mRNA levels), there was no significant difference in the propensity for lymph node metastasis between the subsets with low or high levels of α2-AP mRNA. These findings are consistent with the hypothesis that α2-AP can restrict the capacity of VEGF-D to promote lymph node metastasis in human cancer, by inhibiting the proteolytic activation of this growth factor by plasmin, when levels of VEGF-D in the primary tumour are relatively high.

Figure 5. Bioinformatic analysis of α₂-AP mRNA levels and propensity for lymph node metastasis of human primary tumours. All primary pancreatic, breast, thyroid, colorectal, bladder, prostate, liver and renal clear cell carcinomas, as well as primary head and neck squamous cell carcinomas, lung squamous cell carcinomas, lung adenocarcinomas, and cutaneous melanomas represented in the RNAseqV2 gene expression data of TCGA were included in the analysis (total number of tumours = 5903). Tumours were categorized based on VEGF-D mRNA level: in the upper quartile (>75%), the lower three quartiles (<75%), the lower half (<50%) or the lowest quartile (<25%) of all tumours regardless of tumour type. Tumours in each category were classified as having either greater or less than median α₂-AP mRNA level of all tumours (‘α₂-AP up’ or ‘α₂-AP down’, respectively). For each of these eight groups, the proportions that were positive or negative for lymph node metastasis (‘Node positive’ or ‘Node negative’, respectively) are shown. For the group of tumours with VEGF-D mRNA levels in the upper quartile, the subset with low α₂-AP mRNA levels has a significantly greater propensity for lymph node metastasis than the subset with high α₂-AP mRNA levels. Asterisks indicate statistically significant difference as assessed by Fisher’s exact test, ‘ns’ denotes not statistically significant and p values are shown.

Discussion

Our findings demonstrate that the plasmin inhibitor α₂-AP, a member of the serpin family of enzyme inhibitors, restricted proteolytic activation of VEGF-D and the associated effects of intra-tumoural lymphatic enlargement and lymph node metastasis in a tumour model in mice. This indicates that plasmin can activate VEGF-D in vivo. Given that plasmin was previously shown to activate the closely related lymphangiogenic factor VEGF-C in vitro (McColl et al., 2003), it is likely that this enzyme can also activate VEGF-C in cancer and other pathological settings. A range of clinicopathological studies of human cancers has shown that VEGF-C, VEGF-D and uPA (an enzyme which converts plasminogen to plasmin), can correlate with tumour progression, metastasis and poor patient outcome (Achen & Stacker, 2012; Mekkawy et al., 2014; Stacker et al., 2004). Our findings suggest a mechanistic link for these observations in which the production of plasmin in a tumour by uPA is a key step in the activation of VEGF-C and VEGF-D, which in turn promote lymphatic remodelling thereby facilitating metastatic spread. This may be one of multiple molecular mechanisms by which the uPA/plasmin axis can contribute to tumour development and metastasis (Mekkawy et al., 2014). Further, our findings show that the level of α₂-AP mRNA in solid human primary tumours that have high levels of VEGF-D mRNA is inversely correlated with lymph node metastasis. This is consistent with the hypothesis that α₂-AP can restrict proteolytic activation of VEGF-D by plasmin in human cancer, thereby inhibiting remodelling of tumour lymphatics and metastasis via the lymphatic vasculature.

Although our study is consistent with the well-known link between the activation of VEGF-D and both the remodelling of tumour lymphatics and lymph node metastasis (Harris et al., 2011, 2013), the diversity of effects of the uPA/plasmin axis means we cannot completely exclude the possibility that the effects of plasmin in our model were in part due to mechanisms not involving the proteolytic processing of VEGF-D. For example, it was previously shown that driving expression of α₂-AP in oral squamous cell carcinoma cells led to inhibition of proteolysis of E-cadherin resulting in reduced cell motility and suppression of tumourigenicity in mice (Hayashido et al., 2007). In the tumour model, we employed here, it is possible that plasmin enhanced the motility of tumour cells, thereby facilitating their access and entry to intra-tumoural lymphatics induced by VEGF-D.

Our findings suggest there could be a link between fibrinolysis and lymphatic remodelling in healing wounds based on the capacity of plasmin to activate lymphangiogenic growth factors, such as VEGF-C and VEGF-D, in vivo. For example, plasmin might co-ordinately induce fibrinolysis and lymphangiogenesis, or other forms of lymphatic remodelling, which occur in wound healing (Paavonen et al., 2000; Shimamura et al., 2009). However, the relative timing of these processes in different types of healing wounds is currently not well understood – this needs to be established in order to assess if plasmin may promote both fibrinolysis and VEGF-C- or VEGF-D-induced lymphatic remodelling during the later phases of wound healing. The effect of α₂-AP administration on lymphatic remodelling in wound healing also requires assessment.

The level of α₂-AP in the VEGF-D/α₂-AP tumours studied here was approximately 90% higher than in VEGF-D/Zeo tumours which is a relatively moderate increase. An issue which arises is how VEGF-D processing, lymphatic remodelling and metastasis were significantly reduced in VEGF-D/α₂-AP tumours by this moderate increase in α₂-AP level. One possible explanation is that the level of free plasmin (i.e. plasmin that was not bound to endogenous α₂-AP and thus inactivated) in VEGF-D/Zeo tumours was low, such that the relatively moderate increase in α₂-AP in VEGF-D/α₂-AP tumours was sufficient to inhibit much of the free plasmin in these tumours, thereby significantly restricting activation of VEGF-D and subsequent lymphatic remodelling and metastasis. Alternatively, the explanation may lie in the localization of VEGF-D and α₂-AP in the tumours. In the model, VEGF-D is secreted from tumour cells into tumour interstitium where it can be processed prior to binding and activating VEGF receptors on endothelial cells. In contrast, the host-derived α₂-AP in both VEGF-D/α₂-AP and VEGF-D/Zeo tumours would have come from the liver and other tissues via the bloodstream (Carpenter & Mathew, 2008), and would have been localized principally within the blood vessels of tumours rather than in tumour interstitium. Hence much of the host-derived α₂-AP would not have been in the correct location to activate VEGF-D. Importantly, the α₂-AP produced by VEGF-D/α₂-AP tumour cells was secreted directly into tumour interstitium so it was co-localized with VEGF-D. It is possible that the concentration of α₂-AP in the tumour interstitium, i.e. the relevant α₂-AP with regard to VEGF-D processing, may have been much more than 90% higher in VEGF-D/α₂-AP than in VEGF-D/Zeo tumours. However, a major increase in interstitial α₂-AP might only lead to a 90% increase in total α₂-AP in the tumours due to the relatively high concentration of α₂-AP in the bloodstream of tumours (Carpenter & Mathew, 2008).

Our finding that α₂-AP can restrict the proteolytic activation of VEGF-D in tumours, and thereby inhibit lymphatic remodelling and lymph node metastasis, warrants comparison to other approaches that have been employed to target components of the VEGF-D signalling axis in cancer. Notably, use of neutralizing monoclonal antibodies (mAbs) to VEGF-D (Achen et al., 2000; Davydova et al., 2011, 2012, 2016) or VEGFR-3 (Pytowski et al., 2005) in the VEGF-D-293 model system led to significant decreases in the abundance of LYVE-1-positive lymphatic endothelium in tumours and to reduced metastatic spread to lymph nodes (Matsumoto et al., 2013; Stacker et al., 2001). Significant reductions in intra-tumoural lymphatic endothelium and lymph node metastasis were also observed when processing of VEGF-D was prevented by using a variant full-length form of VEGF-D with mutations in the cleavage sites of the N- and C-terminal propeptides (Harris et al., 2011). Our findings are therefore consistent with previous work showing that VEGF-D can promote remodelling of tumour lymphatics and metastasis, and that proteolytic processing of VEGF-D is important for these processes, but they extend our understanding of the signalling mechanisms involved by showing that the fibrinolytic protease plasmin can contribute to VEGF-D activation in cancer.

It is noteworthy that the α₂-AP produced by VEGF-D/α₂-AP tumours did not completely block processing of VEGF-D. This may be because there was not sufficient α₂-AP in tumour interstitium to inactivate all plasmin at this location, but an alternative explanation is that other proteases contributed to VEGF-D activation in our model. It has been reported previously that members of the proprotein convertase (PC) family of serine proteases, particularly furin, PC5 and PC7, can activate VEGF-D, as well as VEGF-C (Khatib et al., 2010; McColl et al., 2007; Siegfried et al., 2003). Notably, treatment of VEGF-D-293 cells with a small molecule inhibitor of PCs was previously shown to restrict proteolytic processing of full-length VEGF-D in vitro (McColl et al., 2007) so it is likely that PCs contributed to VEGF-D activation in our model given that both VEGF-D/α₂-AP and VEGF-D/Zeo cells were derived from VEGF-D-293 cells. More recently, it was shown that the A disintegrin and metalloprotease with thrombospondin motifs-3 (ADAMTS3) metalloprotease can activate VEGF-C (Bui et al., 2016; Jeltsch et al., 2014). Clearly, there is a variety of enzymes that can contribute to activation of VEGF-C and VEGF-D, and the proteases involved in this process in human cancer may vary in different tumours based on the profile of proteases produced by tumour cells or host infiltrating cells. Hence it will be important to characterize the proteases prevalent in human tumours, and the degree of activation of key protease targets, for supporting drug development programs based on inhibitors of such proteases as furin (Basak et al., 2009), plasmin (Swedberg & Harris, 2012) and uPA (Heinemann et al., 2013; Mekkawy et al., 2009). A range of systems biology approaches (e.g. ‘terminomics’) are being developed for determining the profiles of proteases and the degree of proteolytic activation of protease targets in clinical settings (Doucet & Overall, 2008; Rogers & Overall, 2013) which may facilitate drug targeting of key proteases that drive remodelling of lymphatic vessels in tumours and the metastatic spread of cancer.

Acknowledgements

This work was supported by grants and fellowships from the National Health and Medical Research Council of Australia (to MGA and SAS) and the Operational Infrastructure Support Program of the Victorian Government.

Disclosure statement

M.G.A. and S.A.S. are shareholders in Opthea Ltd. and are Inventors on patents assigned to Vegenics Ltd.

Additional information

Funding

This work was supported by grants and fellowships from the National Health and Medical Research Council of Australia (to MGA and SAS) and the Operational Infrastructure Support Program of the Victorian Government.