Procoagulant and immunogenic properties of melanoma exosomes, microvesicles and apoptotic vesicles

Extracellular vesicles (EV) are lipid particles released from eukaryotic cells into the extracellular fluid. Depending on the cell type or mechanism of release, vesicles vary in form and function and exert distinct functions in coagulation and immunity. Tumor cells may constitutively shed vesicles known as exosomes or microvesicles (MV). Alternatively, apoptosis induces the release of apoptotic blebs or vesicles (ApoV) from the plasma membrane. EV have been implicated in thrombotic events (the second highest cause of death in cancer patients) and tumor vesicles contribute to the anti-cancer immune response. In this study, we utilized the well characterized B16 melanoma model to determine the molecular composition and procoagulant and immunogenic potential of exosomes, MV and ApoV. Distinct patterns of surface and cytoplasmic molecules (tetraspanins, integrins, heat shock proteins and histones) were expressed between the vesicle types. Moreover, in vitro coagulation assays revealed that membrane-derived vesicles, namely MV and ApoV, were more procoagulant than exosomes–with tissue factor and phosphatidylserine critical for procoagulant activity. Mice immunized with antigen-pulsed ApoV and challenged with B16 tumors were protected out to 60 days, while lower protection rates were afforded by MV and exosomes. Together the results demonstrate distinct phenotypic and functional differences between vesicle types, with important procoagulant and immunogenic functions emerging for membrane-derived MV and ApoV versus endosome-derived exosomes. This study highlights the potential of EV to contribute to the prothrombotic state, as well as to anti-cancer immunity.


Mass spectrometry and flow cytometry reveal a distinguishable panel for ApoV, MV, and exosomes
To further characterize B16-F1-derived EV, we analyzed their surface for a number of proteins including tetraspanins, adhesion molecules such as integrins and CD44, sialic acids, and the clotting factors TF and PS by flow cytometry (Figure 2A). Sucrose cushionpurified exosomes, MV and ApoV were dialyzed (10,000 Da cut off) and analyzed by mass spectrometry using a LTQ Orbitrap XL mass spectrometer (Table 1 and  Supplementary Table 1). As illustrated in Figure 2D, EV differed in the abundance of a number of proteins, with particular enrichment of histones and heat shock proteins in exosomes, as compared to MV and ApoV. Notably, the ten most abundant ion scores in exosomes included the histones (H2A, H2B, H3.1 and H4), heat shock proteins (GRP78 and HSC71) and the tetraspanin CD81. Only ApoV showed enrichment for the melanoma tumourassociated antigen PMEL ( Figure 2D and Supplementary Table 1). The raw data for the total 553 proteins identified by mass spectrometry are represented in Supplementary  Table 1. By both flow cytometry and mass spectrometry, ApoV showed low expression of the tetraspanin protein CD9, while MV exhibited intermediate CD9 expression and exosomes the highest CD9 expression. All three EV were positive for the tetraspanin protein CD81 ( Figure  2A), but with higher CD81 ion intensities obtained for exosomes, as compared to MV and ApoV (Table 1). In terms of integrin molecules, EV did not express detectable αV subunits. Exosomes showed the highest expression for the α4 subunit followed by MV, then ApoV. The β1 integrin subunit was highest in MV followed by exosomes, then ApoV (Figure 2A and Supplementary  Table 1). While low levels of α6 were detected in all three vesicles, MV showed the highest expression for α6 (Figure 2A and Supplementary Table 1). Although positive, no differences in CD44 expression levels were detected between the three EV. All three EV showed high level sialic acid expression, which may contribute to the capture of extracellular vesicles in lymphoid tissue [39]. ApoV showed the highest level of PS expression followed by MV, then exosomes. Because the levels of TF were below the detectable limit for three EV, we confirmed the functionality of the TF-specific polyclonal antibody by labeling TF-transfected EL4, as well as the parental B16-F1 line, in the presence or absence of soluble TF-Fc protein. The results showed detectable TF expression on EL4-TF and parental B16-F1 cells and the specificity of the labeling was demonstrated by a loss in binding of the antibody to cells in the presence of excess, soluble TF-Fc protein ( Figure 2B). The vesicular nature of our preparations was confirmed by the presence of CD147 and the coat protein clathrin by western blot and proteomic analysis. As expected from previous reports [25,26,40], there was a preferential association of CD147 and clathrin with vesicles, as compared to the B16 parental cell line ( Figure 2C, Table 1 and Supplementary Table 1).

Fibrin generation potential of EV
Since tumor-derived EV have been implicated in cancer-related thrombosis, we next determined the procoagulant potential of the three types of EV using a fibrin generation assay ( Figure 3A). ApoV generated significantly more fibrin as compared to MV and exosomes. Based on normalized protein content, exosomes were the least coagulative of the EV types. Despite the inability to detect TF by flow cytometry (Figure 2A), the activity of procoagulant ApoV was inhibited by anti-TF antibody and annexin V ( Figure 3B). Furthermore, the importance of TF/ extrinsic pathway, was confirmed using coagulation factor VII depleted (FVII -) plasma ( Figure 3C).

Density fractionation of ApoV
To rule out that contaminating macromolecules were contributing to the procoagulant activity of melanoma (MV) or (c) apoptotic vesicles (ApoV) were purified by differential centrifugation of cell supernatant from live or doxorubicin-treated B16-F1 cells and extracellular vesicles (EV) visualized by cryo-electron microscopy. Original magnification ×20,000. Bars represent 20 nm (for Exo), and 100 nm (for MV and ApoV). Diameter of all three vesicle types determined by (d) cryo-electron microscopy (n = 150 vesicles) with a bin width of 5 nm and by (e) dynamic light scattering. (F) Yield of Exo, MV, and ApoV per mL of tissue culture; error bars represent mean ± SD. One-way ANOVA with Bonferroni post-correction test performed: ns = not significant; **P < 0.01, ***P < 0.001. Results are representative of three experiments.
Oncotarget 56282 www.impactjournals.com/oncotarget Sucrose cushion purified B16F1-derived apoptotic vesicles (ApoV), microvesicles (MV), and exosomes (Exo) were subjected to mass spectrometry and data processed and searched against the mouse reference sequence database using the MASCOT, Sequest HT, and MS Amanda search engines. The TOP3 precursor ion intensities [69] normalised to β-actin of the highest 50 protein intensities present in the exosome samples are represented for all EV types. Additional proteomic data for the total 553 proteins identified in the three EV types are shown in Supplementary Table 1.
Oncotarget 56283 www.impactjournals.com/oncotarget EV, we further purified the highly procoagulant ApoV fraction using a 30% sucrose/D 2 O cushion and assayed fractions by fibrin generation assay ( Figure 4A). Only the ρ ≤ 0.21 g/ml interface exhibited procoagulant activity making it unlikely that non-vesicle associated proteins, polyphosphates or nucleic acids contributed to the observed fibrin generation initiated by ApoV. We next subjected ApoV to a continuous sucrose gradient and tested different density fractions for fibrin generation ( Figure 4B). Although there was some heterogeneity within the coagulative ApoV fractions, the 1.12-1.15 g/ cm 3 density range encompassed the most procoagulant fractions.

Thrombin generation potential of EV
To further confirm the procoagulant ability of the three B16-F1-derived EV types, we subjected purified vesicles to the thrombin generation assay (TGA; Figure 5A). Although results indicate that ApoV and MV were faster than exosomes at generating thrombin, significance between samples was not seen (Mean lag time: ApoV 15.89 min ± 1.34 SD; MV 15.89 min ± 0.16 SD; exosomes 20.22 ± 4.96 SD). Similar to our fibrin generation assay (Figure 3), thrombin generation was retarded for all EV by the inclusion of neutralizing anti-TF antibody in the TGA ( Figure 5B). To ensure that phospholipid was not limiting for the activity of TF, we supplemented the reaction with phospholipid microparticles (MP; Figure 5C and 5D). Although only a small decrease in lag time to peak thrombin production was noted following the addition of phospholipid, blocking TF in the presence of excess MP significantly reduced thrombin generation capability of ApoV.

Anti-cancer responses induced by EV
To determine if the three EV types could play a role in inducing immunity against the B16 tumor we first immunized mice subcutaneously (s.c.) with EV  microspheres were analyzed by flow cytometry using biotinylated (bio) antibodies for the indicated tetraspanins, adhesion molecules, and clotting factors; bio-annexin V for phosphatidylserine (PS) and goat anti-mouse tissue factor (TF). (b) TF expression on B16-F1 cells, EL4, or TF-transfected EL4 (EL4-TF) as analyzed by flow cytometry. Soluble TF (sTF) was used to neutralise the anti-TF polyclonal antibody. Biotin was detected using streptavidin-allophycocyanin (SA-APC), and TF was detected using rabbit anti-goat IgG Alexa Fluor ® 594. Grey shaded peaks represent BSA-bead control, goat IgG control for TF bead samples; black lines represent EV-beads or cells; dotted lines represent TF antibody neutralized cells. (c) Vesicle lysates were subjected to PAGE and Western blotted with goat anti-mouse CD147 (detected with anti-goat IgG-horse radish peroxidase (HRP), mouse anti-mouse clathrin heavy chain, and mouse anti-mouse β actin IgG-HRP (detected with anti-mouse IgG-HRP). MW in kDa are shown. Results are representative of at least two experiments. (d) TOP3 precursor ion intensities [69] normalised to β-actin (y-axis) are represented in rank order (x-axis) in the exosome proteome for the three vesicle types.
Oncotarget 56284 www.impactjournals.com/oncotarget derived from B16-ovalbumin (B16-OVA) in the flank. We then challenged all mice, seven days later, with B16-OVA cells at the opposite flank. Although the B16-OVA cell line expresses ovalbumin at sufficient levels to act as a surrogate tumor antigen for protection in ovalbumin-vaccinated mice [41,42], no protection was observed when exogenously supplied ovalbumin was omitted from the vaccine B16-OVA ApoV preparations (data not shown). B16-OVA cells were therefore treated with additional soluble ovalbumin (200 µg/ml) prior to the isolation of vesicles. Mice immunized with OVA-pulsed ApoV showed the highest protection with only one mouse developing a B16-OVA tumor that reached maximum size at day 69 ( Figure 6A-6B). This level of protection was followed by the mice immunized with OVA-pulsed MV, where three mice reaching maximum tumor size at days 36, 46, and 57. All mice immunized with OVA-pulsed exosomes reached maximum B16-OVA tumor sizes at 16,22,28,36, and 44 (Mantel-Cox analysis; MV vs. exosomes: P < 0.0022; ApoV vs. exosomes: P < 0.0005; ApoV vs. MV; no significance). Surprisingly, the weakly protective exosomes contained a greater quantity of ovalbumin, as compared to MV and ApoV ( Figure 6C). Therefore, the superior protection afforded by ApoV was not a result of enhanced ovalbumin loading into this EV subtype.

dIscussIon
EV released from living and dying tumor cells contribute to the outcome of cancer progression in the host. For example, EV have been proposed to induce a pre-metastatic niche for cancer metastasis [4,20,22,43] and contribute to thrombotic events, such as pulmonary Oncotarget 56285 www.impactjournals.com/oncotarget embolism deep vein thrombosis in cancer patients [10,21,33,34,[36][37][38]44]. EV-associated TF is responsible for the prothrombotic effect of EVs in a mouse model of thrombosis [34]. Less well studied is the role of tumor-derived EV in the induction of immune responses to the tumor itself.
Our results offer clarity of the distinction between vesicles released from the endosome and plasma membrane both from living cells and under chemotherapeutic stress. The murine B16 cell line was chosen for this study as it is the most frequently used syngeneic tumor mouse model studied and recapitulates features of human melanoma including immune suppression and metastasis, and a variable response to immunotherapy [45].
Our study identified distinct morphological and phenotypic features of EV. Exosomes, the smallest vesicle type analyzed, displayed a similar size range to those previously reported for the B16 cell line [46]. ApoV displayed the largest size and greatest range in diameter, as compared to MV and exosomes ( Figure 1). B16 melanoma-derived ApoV were significantly larger than ApoV released from the EL4 lymphoma cell line, as determined in our recent study [14], demonstrating that parental cell type may predetermine certain physical attributes of EV. DLS generates a monomodal distribution and the software assumes that particles are spherical and non-aggregated, while the cryoEM sizing technique employed was simple diameter measurement of individual vesicles at the widest point. While a small difference in vesicle sizing was noted between the DLS and cryoEM techniques, the geometric means were in fact quite similar. The reason for the size difference between the two plasma membrane-derived EV (MV and ApoV) is not entirely clear, but likely relates to caspase activation during the apoptotic response to doxorubicin. The mechanism of release of MV and ApoV are thought to be fundamentally similar, starting with asymmetric redistribution of membrane phospholipids, including the translocation of phosphatidylserine to the outer leaflet, followed by the budding process via actin-myosin interactions [11]. However, in apoptotic vesicle formation, actin-myosin interactions are dependent on caspase cleavage of Rhoassociated kinases 1 (ROCK1) which in turn phosphorylates the myosin light chain for bleb expansion [47].
Phenotypically, major distinctions in terms of molecular profiles were noted between membrane-derived (MV and ApoV) vesicles as compared to exosomes. Melanoma exosomes were strikingly enriched in histones and heat shock proteins, as compared to the other two EV types. Remarkably, this finding is similar to an earlier comparison of apoptotic vesicles and exosomes released from mouse dendritic cells, and is also consistent with the exosome proteome of human dendritic cells [5,12] or murine and human melanoma-derived exosomes [43,48]. Historically, the histone content of EV was thought to be via contamination with apoptotic bodies [12]. However, histones are present in both cytoplasmic and nuclear pools [49] and frequently appear in exosome preparations [5,12,43,48]. Furthermore, we consider that contamination of apoptotic vesicles into our exosome preparations was quite unlikely for the following reasons: (i) our exosome preparations were pre-depleted of cells, debris and MV using sequential 450 × g, 3200 × g and 25,000 × g steps then filtered through a 0.2 µm filter and floated over a sucrose/D 2 O cushion prior to dialysis, and  Table 1). It is surprising, given the highly positive charge of histones, that more links have not been investigated between the RNA and histone cargo of exosomes. We suggest that the enrichment of histones in exosomes may reflect a chaperone role histones for the nucleic acid content of exosomes [1,2] and it is likely that RNA-histone interactions will become an active area of future research. Indeed, a direct association of miRNA and histones has been reported for breast cancer cell linederived exosomes [50] and histone H3 modification was suggested to be essential for exosome release [51].
In cancer patients, tumor vesicles have been implicated in thrombosis, metastatic spread and immune suppression [10, 20-22, 25, 29, 33, 37, 38, 43, 44]. Our investigation into the relative procoagulant activity of vesicles revealed a superior ability of ApoV to induce fibrin generation in platelet poor plasma, compared to MV and exosomes. However, in the TGA, the difference between ApoV and MV was less marked and failed to reach significance. Nevertheless, in both the fibrin generation and TGA, the plasma membrane-derived ApoV and MV displayed higher levels of procoagulant activity compared to exosomes. Although the fibrin generation and TGA differ dramatically in duration (approximately 8 min vs. 25 min to peak fibrin or thrombin respectively) the same pattern was observed; MV and ApoV were more procoagulant than exosomes (Figures 3 and 5). It should be noted that that fibrin clot time precedes the peak thrombin production, with < 1% of total thrombin production required for clot formation [49,52]. Interestingly, although extracellular histones have been shown to enhance thrombin generation in platelet-poor plasma, exosomes, which contained the highest content of histones, were the least procoagulant of the three EV types [53].
Despite the fact that EV TF was below the detectable limit on EV analyzed by flow cytometry (Figure 2), we were nevertheless able to show its importance by neutralization of TF function, or by removing its critical ligand FVII / FVIIa in our fibrin generation assay ( Figure 3B and 3C). Fibrin generation was not completely inhibited in FVII-depleted plasma. However, the commercial source does not guarantee complete removal of FVII by affinity chromatography. Thus residual clotting in commercial FVII plasma may reflect residual FVII  Oncotarget 56288 www.impactjournals.com/oncotarget activity, rather than the initiation of alternate pathways of coagulation by ApoV. Although TF was not detectable by flow cytometry, it was still highly functional. It should be noted that only picomolar concentrations of TF are required for high level activity in the TGA [54]. Therefore, the activity of TF resembles that of cytokines, which are able to induce large biological activity at low molar concentrations. For example, only 3-10 TF molecules per µm 2 are sufficient to induce fibrin deposition under flow conditions [55,56]. Interestingly, ApoV procoagulant activity far exceeds that of parental tumor cell lines, including those overexpressing TF, when normalized for protein content (manuscript in preparation). Therefore, the activity of TF is likely critically dependent on the context of anionic phospholipids, particularly phosphatidylserine [57]. which was enriched on ApoV relative to MV and exosomes (Figure 2). More controversially, TF activity may require decryption through dimerization or disulfide bond formation [58]. A caveat with our conclusions on the relative quantity and activity of TF on vesicles is the semi-quantitative nature of the highly sensitive flow cytometric assay employed. For example, the same ligand density on the membranes of larger flow cytometric events (e.g. tumor cells) would generate higher fluorescent signals, as compared to smaller particles (i.e. 4 µm beads with immobilized EV). In addition, autofluorescence and non-specific binding of isotype controls may mask detection on antibody ligands on the EV preparations.
Due to genomic and epigenetic alterations, tumor can express neoantigens, or overexpress antigens, that are recognized by the immune system [30]. Additionally, tumor associated antigens and endogenous adjuvants are translocated to vesicles [15,28,59]. Notably, the release of tumor antigen on EV is one of the major pathways of the induction of immune responses against tumors [28]. In our experiments, challenge of mice with B16-OVA demonstrated that ApoV afford the highest anti-tumor protection, as compared to MV and exosomes. The exceptional ability of ApoV to protect against melanoma challenge could relate to products of "immunogenic cell death", such as high mobility group box protein B1 (HMGB1) and calreticulin translocating to the ApoV pathway [59,60]. However, neither HMGB1 or calreticulin were present in the ApoV proteome (Supplementary Table 1), suggesting that other as yet unidentified factors mediate the immunogenicity of ApoV. We cannot rule out a partial contribution of the PMEL tumour-associated antigen enriched in the ApoV proteome ( Figure 2D) to the observed immune response, however ApoV failed to induce measurable anti-tumour effects in the absence of exogenously supplied ovalbumin. In addition, the products of apoptotic cell death are efficiently phagocytosed and processed by antigen presenting cells [61], with the likelihood that the abundant release of ApoV with high level expression of phosphatidylserine contributes to enhanced uptake and processing [24,47,60,62]. Although less protective than ApoV, MV were significantly more protective than exosomes. It is possible that distinct factors mediate the protection observed with MV and ApoV, in which case a combination of MV and ApoV might induce even higher levels of protection compared to either vesicle fraction alone. However, it seems equally likely that ApoV and MV harbour the same immunogenic factors, but in different quantities, given the lower levels of protection induced by MV immunization, as compared to ApoV immunization. The poor level of protection afforded by exosomes is surprising given their well characterised ability to capture exogenously supplied antigens (see Figure 6C) and stimulate the immune response [1-3, 39, 63]. Nevertheless, our results with melanoma-derived exosomes closely match the level of protection afforded using ovalbumin-pulsed dendritic cell-derived exosomes with B16-OVA challenge at day 7 [64]. This may reflect a general lack of efficacy for exosomes within this particular melanoma experimental setting.
A weakness in our study was the inability to exclude the possibility that ApoV preparations might contain MV and exosomes released during the induction of cell death by doxorubicin. Conversely, MV preparations might be contaminated with ApoV resulting from normal cell turnover and death during cell culture. Such contamination might mask antigen phenotypic and functional differences between these EV preparations. Nevertheless, the techniques employed allowed for sufficient enrichment of the different EV preparations for the discernment of distinct yields, morphological, molecular, procoagulant and immunogenic properties between the three EV populations.
Our study is the first to directly compare the procoagulant and immunogenic properties of exosomes, MV and ApoV. In particular, this study highlights the contribution of abundantly produced ApoV to pathological states, and their contribution to the anti-cancer response. The greater yield of ApoV released from the cell, as compared to exosomes and MV, further emphasizes the potential of ApoV to contribute to the pro-thrombotic state of cancer patients. Cytoablative anti-cancer therapy may therefore enhance the risk of thrombotic events, but also enhance T cell and natural killer cell-mediated anti-cancer responses via immunogenic EV release.

Vesicle preparation
B16-F1 or B16-OVA cells were incubated for 48 hours either with (ApoV) 25 μM doxorubicin (Baxter Healthcare Ltd, NZ) or without (MV and exosomes) doxorubicin at 70% confluence (approximately 1 × 10 5 cells/mL) in exosome-depleted R5 at 37°C with 5% CO 2 . In order to obtain vesicle-rich fractions, the supernatant was first depleted of cells and debris using differential centrifugation at 450 × g for five minutes, followed by 3200 × g for 20 minutes at 4°C in a 225 mL conical tube (Falcon, In Vitro Technologies, Auckland, NZ). MV (from untreated cultures) or ApoV (doxorubicin treated cultures) enriched fractions were then pelleted by centrifugation at 25,000 × g for one hour at 4°C. For exosomes, the supernatant was first depleted of larger vesicles at 25,000 × g for 1 hour. The remaining supernatant was then filtered using a 0.22 μm nitrocellulose filter (Cole-Parmer ® #EW-02915-52) and exosomes pelleted at 100,000 × g for 60 minutes at 4°C. All vesicles were washed twice in phosphate buffered saline (PBS; Gibco #21600-010). Where stated, vesicles were further purified by either a discontinuous sucrose cushion or a linear sucrose gradient. Briefly, for the sucrose cushion, vesicles were resuspended in 10 mL of PBS and overlaid onto 4 mL of 30% sucrose, 200 mM Tris/deuterium oxide; D 2 O, centrifuged at 100,000 × g for 75 minutes at 4°C and vesicles at the interphase aspirated (approximately 2 mL). Alternatively, for continuous gradient separation, the ApoV pellet was resuspended in 1 mL of 2.5 M sucrose/20 mM HEPES, pH 7.2, and then a linear sucrose gradient (0.25-2 M sucrose/20 mM HEPES, pH 7.2) was overlaid onto the ApoV suspension and centrifuged at 100,000 × g for 18 hours at 4°C. Fractions (1 mL) were removed, and their density determined with an Abbe refractometer (Tokyo, Japan). The protein content of purified vesicles was quantified using a Bradford assay [14].

Sizing and cryo-electron microscopy analysis of vesicles
Purified vesicles (4 µL) were loaded onto plasmaglowed Quantifoil 2/2 grid and blotted to remove excess liquid. The grid was frozen by plunging into −180°C liquid ethane within a Reichert KF80 plunge freezing device (C. Reichert, Austria) and then stored in liquid nitrogen. The grids were then mounted into a pre-chilled Gatan 914 cryo holder and viewed in a JEOL 2200FS cryotransmission electron microscope with an omega filter. Zero-loss images were acquired using a filter width of 20-25 eV via a TVIPS F416 camera. The diameters of the vesicles were measured using the software IMOD. Alternatively, vesicle size was measured using DLS (Zetasizer Nano-ZS; Malvern Instruments, UK).

Proteomic analyses
EV prepared by differential centrifugation were further purified by 30% sucrose/D 2 O and dialyzed (10,000 Da cut off) against PBS to remove sucrose and D 2 O. Vesicle lysates were buffer exchanged and purified by the FASP (filter-aided sample preparation) method [66] using 0.5 ml ultrafiltration units with a molecular weight cut-off of 3 kDa (Millipore). Reduction, alkylation and tryptic protein digestion were performed on filter. The recovered tryptic peptides of each sample were then subjected to liquid chromatography coupled tandem mass spectrometry using an Ultimate 3000 uHPLC-system inline coupled to the nanospray source of a LTQ Orbitrap XL mass spectrometer. Raw data were processed through the Proteome Discoverer software (Thermo Scientific) and searched against the mouse reference sequence database (download Nov 2015 from http://www.ncbi.nlm.nih. gov/refseq; 57928 sequence entries) using the MASCOT (matrixscience.com), Sequest HT (Thermo Scientific) and MS Amanda [67] search engines. The Percolator algorithm [68] was used to adjust the score threshold to an estimated peptide false discovery rate of 1%. Only proteins with two significant peptide hits were considered as significantly identified. Relative protein abundances between the different samples were estimated through using the TOP3 approach [69]. The TOP3 intensity values were calculated using the Proteome Discoverer software and normalized to the β-actin ion intensity of each sample.

Fibrin generation assay
Purified vesicles (quantities as stated) in 20 μL of PBS were added to 100 μL of citrated platelet-poor plasma (Siemens #10446238, or obtained from normal donors with the approval of the regional Human Ethics committee) in a 96-well plate. Coagulation was initiated by adding 10 mM CaCl 2 and the absorbance at OD 405nm was measured every 30 seconds over 30 minutes at 23°C using a plate reader (Varioskan Flash, Thermo Scientific). All samples were performed in triplicates. Where stated, 20 μg/mL goat anti-mouse TF polyclonal antibody (R&D #8686) or 100 μg/mL annexin V (eBioscience #BMS306) were added for blocking experiments.

Vesicle immunization and tumor implantation
B16-OVA was cultured in R5, pulsed for 48 hours with 200 μg/mL OVA protein (Sigma-Aldrich #A5503) and vesicles harvested as stated above. For generation of ApoV, doxorubicin was added simultaneously with the OVA protein. C57BL/6 mice were obtained from the Jackson Laboratory (Bar Harbor, ME) and bred and housed under specific pathogen-free conditions at the University of Otago Hercus-Taieri Research Unit as described [70]. All experiments were approved by the University of Otago regional animal ethics committee. Mice were immunized s.c. with 25 µg (in 50 µL of PBS) of B16-OVA-derived ApoV, MV, or exosomes in the flank. Seven days later, B16-OVA cells were harvested from logarithmically growing cultures using a cell scraper, filtered through a 70 μm cell strainer, and resuspended in PBS. Mice were then challenged with 1 × 10 5 B16-OVA cells s.c. in the opposite flank to the immunization site. Tumor growth was determined by measuring the length and width using calipers every 1-2 days. Results are expressed as the mean product of the tumor diameters. Mice were removed from the study when tumors reached 150 mm 2 . Data are represented as tumor growth curves or as Kaplan-Meier survival plots using Graphpad Prism 6 (GraphPad, San Diego, CA).

Statistical analysis
All statistical analyses were performed with GraphPad Prism 6 (GraphPad, San Diego, CA). Fibrin generation assays were analyzed using one-way ANOVA with Bonferroni post-correction test. Survival data were represented as tumor growth curves or as Kaplan-Meier survival plots and analyzed using the Mantel-Cox test.