Targeting Tumor Vascular CD99 Inhibits Tumor Growth

CD99 (MIC2; single-chain type-1 glycoprotein) is a heavily O-glycosylated transmembrane protein (32 kDa) present on leukocytes and activated endothelium. Expression of CD99 on endothelium is important in lymphocyte diapedesis. CD99 is a diagnostic marker for Ewing's Sarcoma (EWS), as it is highly expressed by these tumors. It has been reported that CD99 can affect the migration, invasion and metastasis of tumor cells. Our results show that CD99 is also highly expressed in the tumor vasculature of most solid tumors. Furthermore, we found that in vitro CD99 expression in cultured endothelial cells is induced by starvation. Targeting of murine CD99 by a conjugate vaccine, which induced antibodies against CD99 in mice, resulted in inhibition of tumor growth in both a tumor model with high CD99 (Os-P0109 osteosarcoma) and low CD99 (CT26 colon carcinoma) expression. We demonstrated that vaccination against CD99 is safe, since no toxicity was observed in mice with high antibody titers against CD99 in their sera during a period of almost 11 months. Targeting of CD99 in humans is more complicated due to the fact that the human and mouse CD99 protein are not identical. We are the first to show that growth factor activated endothelial cells express a distinct human CD99 isoform. We conclude that our observations provide an opportunity for specific targeting of CD99 isoforms in human tumor vasculature.

Previous studies suggest that CD99 is a promising therapeutic target. Guerzoni et al. showed that targeting of CD99 by a diabody (C7 dAbd) promoted cancer cell death of EWS tumor cells in vitro (16). In addition, knockdown of CD99 in EWS tumor cells reduced in vivo tumor growth in mouse xenograft experiments (12). Also, a monoclonal antibody against CD99 (0662 Mab) combined with doxorubicin showed enhanced inhibition of EWS tumor growth and metastasis formation in a xenograft model (17). Imaging of mice with a 64 Culabeled anti-mCD99 antibody detects subcutaneous Ewing sarcoma tumors and metastatic sites with high sensitivity (18). CD99 was also found to be highly expressed in acute lymphoblastic leukemia (ALL), acute myeloid leukemia (AML), myelodysplastic syndromes (MDS), and stem cells and anti-CD99 monoclonal antibodies show antileukemic activity in AML xenograft models (19)(20)(21). In atherosclerosis CD99 is expressed on activated endothelial cells that cover the plaque. Treatment of immunocompetent mice with an oral vaccine composed of attenuated Salmonella typhimurium transformed with pcDNA3-murineCD99 inhibited atherosclerotic plaque formation by induction of CD99 targeting cytotoxic T cells (22). Van Wanrooij et al. suggest that vaccination leads to removal of the CD99 expressing endothelial cells and thereby reduces atherosclerotic plaque formation. After vaccination a decreased expression of CD99 on leukocytes was observed and fewer leukocytes were recruited to the site of the plaque, whereas the total number of leukocytes was not affected. These observations indicate that CD99 can safely be used as a therapeutic target for vaccination.
In the current study, we explored the opportunity to use vaccination against CD99 as a treatment option against solid tumors. We showed that CD99 is heavily overexpressed on tumor endothelial cells in multiple human solid tumors. We developed a conjugate vaccine, based on a fusion protein technique published previously by Huijbers et al. (Supplementary Figure 1B) (23)(24)(25). In short, a fusion protein, consisting of the murine CD99 sequence and an engineered truncated version of bacterial thioredoxin (26), was made and used for vaccination. The vaccine induced an antibody response against CD99 by activation of specific CD99 auto-reactive B-cells. In two different immunocompetent mouse tumor models we found that vaccination against CD99 reduced tumor microvessel density and functionality, and resulted in suppressed tumor growth. No sideeffects were observed after maintaining mice hyperimmune for almost a year.
For human CD99, two different isoforms have been described (27). A long full-length isoform (185 amino acids, 32 kDa, variant I; Supplementary Figure 1C) and a short truncated isoform (161 amino acids, 28 kDa, variant II), lacking most of the cytoplasmic domain, exist. The murine CD99 only shows 46% homology with human CD99 and resembles the human short isoform (28). However, it is unclear whether the CD99 isoforms have the same function in mouse as in humans (29). In the NCBI database six different protein coding human CD99 splice variants are suggested (Gene ID: 4267) (30). In this paper we dissected the expression of these splice variants in endothelial cells and EWS tumor cells. Our results show a difference in CD99 splicing in activated endothelial cells and EWS tumor cells, which provides opportunities for specific therapeutic targeting to treat cancer.

HUVEC Starvation Assay
Six-well culture plates (VWR International, Radnor, PA, USA) were coated with 0.2% gelatin and 150,000 HUVEC were seeded in each well. Cells were allowed to adhere for 3-4 h before nutrient deprived medium was added containing only 10% FBS or 1% FBS (Lonza Biowhittaker) in RMPI-1640 medium (Lonza Biowhittaker). Cells were then harvested at 24 or 48 h and flow cytometry was performed. In addition, cell lysates were prepared from control treated cells and 48 h starved cells. To this end cells grown in a T75 culture flask were scraped off the bottom in 500 µl 2x Laemmli sample buffer (#1610737, Bio-Rad Laboratories B.V., Veenendaal, The Netherlands) plus β-mercaptoethanol (sc-202966, Santa Cruz Biotechnology Inc., Dallas, TX, USA) on ice.
Lysates were stored at −20 • C until use.
Cells were analyzed with a FACSCalibur flow cytometer (Beckton Dickinson, Franklin Lakes, NJ, USA) and CellQuest Software. Data analysis was performed with FCSalyzer (SourceForge, La Jolla, CA, US). Fold increase of the mean fluorescence intensity (MFI) was determined by dividing the MFI of CD99 stained cells by the MFI of control stained cells.
Images of different tumor types and normal tissues stained for human CD99 were retrieved from the Human Protein Atlas (32,33).

Desmin Staining
After deparaffinization, osteosarcoma tumor sections were treated with 0.3% H 2 O 2 for 15 min at RT, washed and boiled in citrate buffer. Sections were blocked with 3% BSA/PBS for 60 min at RT and incubated with primary rat anti-mouse CD31 antibody (1:50; Dianova) and goat anti-human/mouse desmin antibody

Staining of Tumor Tissue With Serum of TRXtr-mCD99 Immunized Mice
Osteosarcoma tumor sections from TRXtr immunized mice were deparaffinized, treated with 1% H 2 O 2 for 15 min at RT and boiled in citrate buffer. Sections were blocked with 20% horse serum (H-1138, heat inactivated, Sigma-Aldrich)/PBS for 1 h at RT. Consecutively, sections were incubated with goat F(ab) anti-mouse IgG H&L (ab6668, 1 mg/ml, Abcam) diluted 1:20 in PBS for 2 h at RT, to prevent non-specific binding of the mouse serum to the mouse tissue. A washing step with 0.05% PBS-T was performed and the sections were stained with serum from TRXtr immunized or TRXtr-mCD99 immunized mice diluted 1:600 in 20% horse serum/PBS overnight at 4 • C. Anti-CD99 antibodies in the serum were detected with biotinylated polyclonal goat anti-mouse Ig (E0433, Dako Cytomation) diluted 1:500 in 0.5% BSA/PBS, followed by Streptavidin-HRP (DakoCytomation) 1:200 in 0.5% BSA/PBS and DAB substrate. Sections were counterstained with Mayer's hematoxylin (Klinipath), dehydrated in an ethanol series and mounted with Quick D (Klinipath).

Immunohistochemistry Quantification
Pictures were captured with an Olympus BX50 microscope (Olympus Optical Co. GmbH, Hamburg, Germany) equipped with a CMEX DC 5000C camera (Euromex microscopes, Arnhem, The Netherlands).
Only viable tumor tissue was used for analysis. Microvessel density was assessed by manual counting of tumor tissue stained for CD31. In total 3 fields/tumor (100x magnification) and 3-10 tumors per experimental group were counted. Images were used to manually count the number of vessels with a clear lumen in osteosarcoma tumors. Images were further analyzed with ImageJ (Laboratory for Optical and Computational Instrumentation, University of Wisconsin-Madison; Version 1.51s) to determine the vessel density of osteosarcoma and CT26 tumors. For pericyte (desmin) quantification, 10 fields per tumor were chosen (magnification 200x). Images were used to manually count the number of vessels with and without desmin staining/associated pericytes. Pericyte coverage was then determined by dividing the number of vessels with pericytes by the total vessel count.

Reverse Transcriptase Quantitative Polymerase Chain Reaction (RT-qPCR)
Total RNA was isolated using TRIzol Reagent (Life technologies, Carlsbad, CA, USA) according to the manufacturer's protocol. RNA concentration and quality were measured using a NanoDrop ND-1000 spectrophotometer (Isogen Life Science, Utrecht, The Netherlands). One microgram of the obtained RNA was reverse transcribed to cDNA using an iScript cDNA synthesis kit (Bio-Rad Laboratories, Hercules, CA, USA). The obtained cDNA was diluted 1:5 and RT-qPCR was performed using iQ SYBR Green Supermix (Bio-Rad Laboratories) and 0.2 µM of each primer (Eurogentec, Seraing, Belgium) (Tables 1, 2). Primers were validated as previously described (35). Samples were run in duplicate and analyzed on the CFX96 Real Time System C1000 Thermal Cycler (Bio-Rad Laboratories). Data were analyzed with CFX Manager software (version  Primer Sequence Primer Sequence

3.1, Bio-Rad Laboratories), and further processed in MS
Excel. All samples were normalized to cyclophillin A, β-actin, and β2-microglobulin transcript expression (Tables 3, 4) to account for variations in template input (35). The following formula was used to calculate the 2 −dCt value of the gene of interest: 2 −(Ctvaluegeneofinterest−meanCtvaluereferencegenes) . Ratios of the different primer pairs and the k + l primers used to determine total CD99 were calculated by dividing the mean 2 −dCt value of each primer pair with the mean 2 −dCt value of the k + l primers.

Expression Vectors
The  Figure 1D). Murine CD99 extracellular part protein sequence (aa27-137) (36): ASDDFNLGDALEDPNMKPTPKAPTPKKPSGGFDLED ALPGGGGGGAGEKPGNRPQPDPKPPRPHGDSGGISDSDL ADAAGQGGGGAGRRGSGDEGGHGGAGGAEPEGTPQ For construction of the pET21a-mCD99 extracellular plasmid the extracellular part of murine CD99 was PCR amplified from the original pET21a-mCD99 vector (Genscript) using the following primers: After purification and restriction with NdeI and XhoI the mCD99 extracellular sequence was ligated into a pET21a vector.

Vaccine Production and Purification
The recombinant vaccine proteins were produced and purified as previously described (23,24). The pET21a expression vectors were transformed into E. Coli Rosetta gami (DE3) (Novagen; Merck Millipore, Darmstadt, Germany) for recombinant protein expression. Overnight cultures were diluted 1:3 and were grown until optical density 600 nm (OD 600 ) 0.5 was reached. Protein expression was induced with 1 mM isopropyl β-D-1-thiogalactopryanoside (IPTG, Invitrogen, Life Technologies, CA, USA) at 37 • C for 4 h for TRXtr-mCD99 extracellular and mCD99 extracellular (for simplicity the resulting proteins are named TRXtr-mCD99 and mCD99). TRXtr expression was induced at 22 • C for 15 h. Bacteria were harvested by centrifugation at 4,500 rpm (3,584 g), 10 min, 4 • C (Hettich Rotina 420R) and washed five times with PBS. Bacterial pellets were dissolved in PBS (TRXtr-mCD99 and mCD99) or in 5 M urea (TRXtr) (Acros Organics/Thermo Fisher Scientific, Landsmeer, The Netherlands). The proteins (TRXtr-mCD99 and mCD99) were released by sonication (amplitude 22-26 microns, Soniprep 150 MSE, London, UK) on ice, 12 times for 30 s with breaks of 30 s, for the TRXtr protein 15 cycles of 20 s on and 30 s off were used. After centrifugation, 20 mM imidazole (J.T. Baker, Avantor Performance Materials B.V.) was added to the supernatant to reduce non-specific binding of background proteins to the nickel (Ni) agarose. No imidazole was added to the TRXtr supernatant during the Ni-agarose incubation step. Thereafter, 500 µl 50% Ni-NTA agarose slurry (Qiagen, Venlo, The Netherlands) was mixed with 25 ml supernatant (originating from 500 ml bacteria culture) and incubated "end-over-end" at 4 • C for 3 h. After centrifugation at 3,000 rpm (Rotina 420R, Hettich), the agarose beads were washed with 250 ml wash buffer containing PBS pH 7.0/1 M NaCl /0.1% Tween-20. An additional washing step with PBS was performed to remove the detergent Tween (P1379, Sigma-Aldrich, Zwijndrecht, The Netherlands). Then, the beads were transferred to a 1 ml syringe (BD Plastipak, BD Biosciences, Madrid, Spain) with a glass filter (Sartorius Stedim Biotech, Göttingen, Germany) and washed again with PBS. The TRXtr-mCD99 protein was eluted with two 250 µl fractions of 50 mM and three fractions of 100 mM imidazole, dissolved in 20 mM Tris pH 8.0/0.1 M NaCl. The TRXtr and mCD99 protein were eluted with four 250 µl fractions of 100 mM imidazole solution. Protein content of the separate fractions was determined by SDS-PAGE using precast 4-20% gradient polyacrylamide gels (Mini-Protean TGX, Bio-Rad Laboratories). Gels were fixed and stained with home-made colloidal Coomassie brilliant blue R250 solution containing 20% methanol (VWR International). Destaining of the gels was performed with methanol for 1 min and ddH 2 O for several hours. Fractions containing most protein were pooled and dialyzed against PBS (pH 7.0). The different recombinant proteins TRXtr-mCD99 (18 kDa), TRXtr (7.5 kDa; appears as a 15 kDa dimer on reduced SDS-PAGE) and mCD99 (11 kDa) as present on an SDS-PAGE gel after purification ( Figure 1G).
For the TRXtr and TRXtr-mCD99 protein a Slide-A-Lyzer Dialysis cassette (M w cut-off (MWCO) 7,000 Da; Thermo Fisher Scientific) was used. The mCD99 protein was dialyzed in snakeskin dialysis membrane (MWCO 3,500 Da; Thermo Fisher Scientific). Final protein concentrations were measured by micro BCA protein assay (Pierce Biotechnology, Rockford, IL, USA).

Production of Rosetta Gami Extract to Block Background Binding in ELISA
Rosetta gami extract for use in ELISA was produced from uninduced pET21a-TRX transformed overnight cultures. Bacteria were harvested at 4,500 rpm, 10 min, 4 • C (Rotina 420R, Hettich) and washed 3 times with PBS. The pellet (originating from 200 ml overnight culture) was resuspended in 10 ml 0.

Western Blot
Cell lysate of Os-P0107 and CT26 was obtained using RIPA buffer (Cell Signaling Technology, Danvers, MA, USA) with the addition of HALT protease/phosphatase inhibitor 1:100 (Thermo Fisher Scientific) and stored at −20 • C until use. Protein concentration was measured by Nanodrop ND-1000 spectrophotometer (Isogen Life Science).
For staining of human CD99 the membrane was blocked with 5% BSA/PBS-T 0.05% and incubated with rabbit anti-human CD99 polyclonal antibody (ab27271, Abcam) diluted 1:200 in 1% BSA/PBS-T 0.05% overnight at 4 • C. The next day the membrane was incubated with goat anti-rabbit IRDye 800CW antibody (cat no. 926-32211, LI-COR Biosciences) diluted 1:10,000 in 1% BSA/PBS-T 0.05%. Washing steps were performed as described above. The blot was stored overnight in PBS and blocked the next day with 5% non-fat dry milk (blotting-grade blocker, cat no. 170-6404, Bio-Rad Laboratories)/PBS-T 0.1%. After that the membrane was stained overnight at 4 • C with mouse monoclonal anti-human β-actin antibody (cat no. #3700, clone 3H10D10, Cell Signaling Technology, Leiden, The Netherlands) diluted 1:1,000 in 1% non-fat dry milk/PBS-T 0.1%. As final step the membrane was incubated with donkey anti-mouse IRDye 680D (cat no. 925-6872, LI-COR Biosciences) diluted 1:10,000 in 1% non-fat dry milk/PBS-T 0.1%. All washing steps were performed with PBS-T 0.1% and performed as described above. The membrane was imaged with the Odyssey Infrared Imaging System (LI-COR Biosciences).

Animal Studies
Animal experiments were approved by the local Animal Ethics Committee of the VU University and the national Central Animal Experiments Committee (CCD) (reg. no. AngL14-01 and CCD AVD114002016576). Approximately 8-week old female C3H/HeNCrL mice (Charles River Laboratories, Leiden, The Netherlands) or BALB/c mice (Envigo, Horst, The Netherlands) were immunized four times with an interval period of 2 weeks. Each vaccine emulsion (100 µl per mouse, 50 µl per groin) contained 40 µg TRXtr (control group) or 100 µg TRXtr-mCD99 in a volume of 50 µl mixed with 50 µl Freund's complete adjuvant (F-5881, Sigma Aldrich) (ratio 1:1, aqueous phase: oil phase) for the priming immunization and Freund's incomplete adjuvant (F-5506, Sigma Aldrich) for booster immunizations. Emulsions were mixed for 30 min on a Vortex Genie (Fisher Scientific) at full speed. Four weeks after the last immunization 2 × 10 6 osteosarcoma (Os-P0107) cells were inoculated subcutaneously in the left flank of C3H mice in a total volume of 100 µl (10% culture medium/PBS). For the CT26 model 2 × 10 5 CT26 colon carcinoma cells were inoculated in the left flank of BLALB/c mice. Blood samples were taken from the tail vein 1 week after each immunization, 1 week prior to tumor cell injection, and 1 week after tumor cell injection and at the end of the experiment. Tumor growth was measured by calipers. Tumor volume was calculated by the formula: width 2 × length × π/6. At the end of the experiment mice were euthanized and tumors and organs were removed and stored in 1% PFA/PBS (Aurion, Wageningen, the Netherlands) overnight and consecutively paraffin embedded.

Long-Term Follow-Up After CD99 Vaccination
To address the safety of exposure to high antibody titers against CD99; control vaccinated (TRXtr, n = 5) and TRXtr-mCD99 vaccinated mice (CD99; n = 5) were included in the study for 45 weeks. Approximately 8-week old female C57BL/6 mice (Envigo) were immunized three times with an interval period of 2 weeks as described above. Blood samples were taken from the tail vein 1 week after each immunization. During the rest of the follow-up period monthly blood samples were taken. When antibody levels dropped below 50% of the levels after the third vaccination mice were revaccinated. In addition, body weight of the mice was monitored regularly during the whole study period. At the end of the experiment mice were euthanized and organs were removed, stored in 1% PFA/PBS (Aurion) overnight and paraffin embedded.

ELISA
Indirect ELISA was performed to determine total anti-mCD99 antibody levels. Blood samples were coagulated overnight at 4 • C and centrifuged twice at 7,000 rpm for 10 min at 4 • C in a microcentrifuge. The supernatant (serum) was stored at −20 • C until use. Volumes used per well in ELISA were 50 µl, unless indicated otherwise. 96-well ELISA plates (Nunc A/S, Roskilde, Denmark) were coated with 2 µg/ml mCD99 protein and then blocked with 100% horse serum (100 µl/well) (Sigma-Aldrich), both for 1 h at 37 • C. After a single wash with PBS (B. Braun Medical, Oss, The Netherlands) for 1 min, the plates were incubated with serum of TRXtr-mCD99 or TRXtr-vaccinated mice for 45 min at 37 • C, diluted 1:10 in 100% horse serum, which was further diluted 1:15 in 10% Rosetta Gami extract (final serum dilution 1:150) to reduce non-specific binding of the serum. Thereafter, plates were incubated with biotinylated polyclonal goat anti-mouse Ig (E0433, Dako Cytomation) for 45 min and streptavidin-horseradish peroxidase (Strep-HRP) (Dako Cytomation) for 30 min, both diluted 1:2,000 in 0.01% PBS-T at 37 • C. After each incubation step, plates were washed four times with PBS. HRP activity was detected with TMB substrate (T-8665, Sigma-Aldrich) and absorbance was measured at 655 nm after 15 min using a Biotek Synergy HT microplate reader (Biotek).

Statistical Analysis
Means were compared using a Mann-Whitney U-test or twotailed student's t-test, if Gaussian distribution could be assumed. For comparison of tumor growth, a two-way ANOVA with Bonferroni post-test was used for repeated measurements at different time points. Values are depicted as mean ± SEM. All statistical tests were executed using GraphPad Prism 5.0 (GraphPad Software Inc., La Jolla, CA, USA). Values of P < 0.05 were considered statistically significant.

CD99 Is a Marker of Tumor Blood Vessels and Is Induced by Starvation
The Ewing's sarcoma (EWS) tumor marker CD99 was found to be overexpressed in the vasculature of different solid tumor types (Figure 1A; Supplementary Figure 1A), but CD99 was not expressed in normal healthy tissues. In endothelial cells, cultured in vitro, CD99 expression was upregulated upon 24 and 48 h of starvation (p < 0.05) ( Figure 1B).

Expression of CD99 in Murine Tumor Cell Lines
In order to determine the CD99 expression in the murine osteosarcoma (Os-P0107) and colon carcinoma (CT26) cell lines qPCR primers were designed to determine the expression of endogenous CD99 (mCD99; RefSeq NP_079860.2. NM_025584.2.; GI: 125660452) ( Table 1). Expression of CD99 at the mRNA level was confirmed in both cell lines ( Figure 1C). The CT26 cell line was observed to express significantly lower levels of CD99 than the Os-P0107 cell line [( * * P < 0.01); Figure 1C, white bar]. At the protein level, flow cytometry was used to check for surface expression of CD99. These data indicated a higher surface expression of CD99 on Os-P0107 cells compared to the CT26 cell line (Figure 1D). In order to clearly distinguish any difference in CD99 protein level between the two cell lines we performed a Western blot analysis ( Figure 1E) on cell lysate. Indeed, the Western blot confirmed that the osteosarcoma cells express high levels of murine CD99 (24 kDa, green band) and that CD99 expression in the CT26 cell line is much lower. β-actin (42 kDa, red band) was used as a loading control. In addition, the ratio of the CD99 and β-actin bands was quantified with ImageJ software and the osteosarcoma cell line showed a higher CD99/β-actin ratio than the CT26 cell line (Os-P0107 ratio: 1.77; CT26 ratio: 0.46) ( Table 5).

Design/Construction of a CD99 Targeting Vaccine
The overexpression of CD99 in human tumor vasculature prompted us to study the role of the molecule in tumor growth.
To that end, we planned to vaccinate against CD99, using the conjugate vaccine technology that we recently published (26). The extracellular part of murine CD99 (mCD99 extracellular , 111aa; RefSeq NP_079860.2. NM_025584.2.; GI: 125660452) (36) was used as the self-antigen to be fused to truncated bacterial thioredoxin (TRXtr, 58aa) (26), resulting in the fusion protein TRXtr-mCD99 ( Figure 1F). The TRXtr protein was produced for vaccination of control mice and mCD99 for detection of anti-mCD99 antibodies in serum by ELISA ( Figure 1F). All proteins were soluble ( Figure 1G). Protein identity of TRXtr-mCD99 and mCD99 was confirmed by Western blot analysis (Figure 1H).

Vaccination Against CD99 Inhibits Tumor Growth
C3H mice were vaccinated with TRXtr-mCD99 (CD99; n = 10) or TRXtr (control group; n = 5). At week 10, when the mice were hyperimmune and had high antibody titers against murine CD99 in their sera, osteosarcoma (Os-P0107) tumor cells were injected into the left flank (see experimental set-up Figure 1I). Blood samples were collected 1 week after each vaccination, 1 week prior to tumor cell injection, 1 week after tumor cell injection and at the end of the experiment. In two independent experiments the mice responded with the production of anti-mCD99 antibodies (Figures 2A,B; Supplementary Figure 2A,B). Indeed, when antibodies against CD99 were present in the sera of the mice, tumor growth of osteosarcoma (Os-P0107) was significantly inhibited compared to control vaccinated mice ( * * P < 0.01; Figure 2F). We also investigated if tumor growth of the CD99 low CT26 colon carcinoma (Figure 2D, lower panel) could be inhibited after vaccination against CD99. All TRXtr-mCD99 (CD99 group; n = 5) vaccinated BALB/c mice responded with the production of anti-mCD99 antibodies ( Figure 2C, one mouse died of unrelated cause). Also, in this study a significant difference in tumor growth between the TRXtr-mCD99 group and control group (TRXtr group; n = 5) could be observed ( * * P < 0.01; Figure 2I), indicating that in this tumor model inhibition of tumor growth was mainly due to targeting of the tumor vasculature. During the study period no difference in body weight between the CD99 vaccinated and control-vaccinated mice was observed in all three different vaccination studies performed (Supplementary Figure 2C). This suggests that vaccination against CD99 is well-tolerated and safe. In addition, we investigated if the antibodies induced against CD99 would recognize native CD99 in tissue. Therefore, we stained tissue of Os-P0107 osteosarcoma tumors derived from TRXtr-vaccinated mice with serum of TRXtr-mCD99 (CD99) vaccinated mice or of control (TRXtr) vaccinated mice.

Vaccination Affects the Tumor Vasculature
Osteosarcoma tumor tissue of the first vaccination study (study I) was stained for the vascular marker CD31 to determine the effect of vaccination against CD99 on the tumor vasculature ( Figure 2E, upper panels). Vaccination with TRXtr-mCD99 seemed to have an effect on the morphology of the vasculature of osteosarcoma tumors (Figure 2E, upper panels). More specifically, tumors retrieved from TRXtr-mCD99 vaccinated mice (CD99, red bars) were found to have a significantly lower vessel density ( * P < 0.05) and lumen count ( * P = 0.05) compared to tumors retrieved from control vaccinated mice (TRXtr, blue bars) (Figure 2G). In the CD99 low CT26 tumor model also a significantly lower vessel density ( * P < 0.05) was found in tumors of CD99 vaccinated mice (Figures 2J,K). Furthermore, we stained the osteosarcoma tumors for both the vessel marker CD31 and the pericyte marker Desmin to study the effect of vaccination on vessel functionality. As illustrated in Figure 2E (lower panels) and Figure 2H, a significantly lower pericyte coverage was found in TRXtr-mCD99 vaccinated tumor tissue (red bars) compared to TRXtr vaccinated tumor tissue (blue bars) ( * P < 0.05, Figure 2H, left panel). Furthermore, tumors of CD99 vaccinated mice were found to have more vessel without pericytes than control vaccinated mice ( * P < 0.05, Figure 2H, right panel).

Induction of a Humoral Immune Response Against CD99 Is Safe
During the experimental period of 14 weeks of the tumor growth study we did not observe any toxicity of the TRXtr-mCD99 vaccine as addressed by body weight or macroscopic and behavioral characteristics between CD99 vaccinated and control-vaccinated mice. However, to further investigate the safety of the vaccine we vaccinated healthy C57BL/6 mice against CD99 and monitored their body weight over a period of 45 weeks (Figure 2L). The mice were kept with high anti-CD99 antibody levels during the whole study period. Once the anti-CD99 antibody level dropped below 50% of the starting level (the antibody level after the third vaccination) the mice were re-vaccinated. Control mice were vaccinated with the TRXtr protein. During the whole study period we did not observe any difference in body weight between the CD99 group and the control vaccinated mice (Figure 2L). In addition, all except one mouse in the CD99 group, which was lost to follow-up at week 35, were healthy during the whole study period. We also looked at the morphology of the organs of CD99 vaccinated and control mice, but did not find any changes in tissue morphology after vaccination against CD99 (Figures 2M,N;  Supplementary Figures 2D, 3). All together these observations indicate that vaccination against CD99 is safe.

A Distinct Human CD99 Isoform Is Present in Activated Endothelial Cells
In literature two different human CD99 isoforms have been described (27). A long full-length isoform (185 amino acids, 32 kDa, variant I; Supplementary Figure 1C) and a short truncated isoform (161 amino acids, 28 kDa, variant II), lacking most of the cytoplasmic domain. In the NCBI database six different protein coding human CD99 splice variants are suggested (Gene ID: 4267) (30) (Figure 3A). To distinguish the different human CD99 isoforms, as described in the NCBI database ( Table 6), The inverted "v" below exon 9 indicates the presence of an additional alanine at the start of the exon and thereby a different variant (var) (variant 5 or 6). RT-qPCR-primers used for identification: the primer pair k + l detects all human CD99 variants (pan; total CD99). Primer pair d + e identifies variant 1, 5, 2, and 6. Primer pair a + b was used to detect the full-length CD99 isoform (var 1, var 5) and the variants 7 and 4. Primer pair a + c detects variant 2 and 6, which lack exon 3 in the extracellular domain. Primer pair f + e was used to identify variant 7, lacking exon 7. Primer pair g + h detects the truncated CD99 isoform (var 4, isoform (C). (B) In activated HUVEC (HUVEC +; n = 9), EW7 (n = 11) and EWS-RDES (RDES; n = 3) relative expression (2 −dCt ) of total CD99 is downregulated on mRNA level (k + l; all variants). For native HUVEC (HUVEC; n=6) vs. EW7 this is statistically significant (**P < 0.01) and for HUVEC vs. activated HUVEC (HUVEC +) [P = 0.0905; k + l)] and RDES (P = 0.0833; k + l) there is a trend toward significance. The observed difference of a + c between native and activated HUVEC (*P < 0.05) and EW7 (*P < 0.05) is due to that there is more total CD99 present (k + l primers) in native HUVEC compared to activated HUVEC. The ratio a + c/k + l is similar for both cell types (HUVEC = 0.04327756; HUVEC + = 0.0440165). A higher signal for the a + b primer pair is observed in native HUVEC compared to activated HUVEC+ (*P < 0.05), EW7 (**P < 0.01) or RDES (*P < 0.05). However, the ratio a + b/k + l is similar for all cell types ( Table 7). The main isoform present in growth factor activated HUVEC (HUVEC+) is variant 4 (the short CD99 isoform, isoform (C), identified by primer pair g + h.  we designed RT-qPCR primers specific for the different splice variants ( Figure 3A and Table 2 In activated HUVEC (HUVEC+; n = 9), EW7 (n = 11), and EWS-RDES (RDES; n = 3) expression of total CD99 (k + l primers; all variants) is downregulated on mRNA level ( Figure 3B). For native HUVEC (HUVEC; n = 6) vs. EW7 this is statistically significant ( * * P < 0.01) and for HUVEC vs. activated HUVEC (ns P = 0.09) and RDES (ns P = 0.08) there is a trend toward significance. The difference in expression found with the a + b and a + c primers in native HUVEC is due to high expression of total CD99 (k + l primers) in native HUVEC compared to the other cell types; as determined by the ratio of a + b/k + l and a + c/k + l ( Table 7). On mRNA level however, there is a trend toward a higher expression of the short human CD99 isoform variant 4 (g + h primers; Table 7).
Western blot was performed on cell lysates of growth factor activated and serum starved HUVEC to determine if the short CD99 isoform could be detected on protein level. In activated HUVEC, after several passages (>P2) in culture, an additional protein band around 16 kDa can be observed next to the 35 kDa band of human CD99 (Figure 3C, green bands, upper panel). β-actin was used as a loading control ( Figure 3C, 45 kDa,  (Figure 3D, Table 8) and found that expression of CD99 is induced in activated HUVEC; cells in higher passages express more CD99 (35 kDa band, black bars). In addition, there is a trend toward induction of expression of the 16 kDa band (white bars, the short CD99 isoform; isoform C) with higher passage and after starvation of the cells for 48h. Peripheral blood mononuclear cells (PBMC) of healthy volunteers only express CD99 at low levels (Supplementary Figure 4B). The main CD99 variants detected in PBMC were variant 1, variant 5, variant 7, and variant 4 (primer pair a + b), of which variant 7 (primer pair f + e) was basically undetectable (data not shown).
We also isolated mRNA from human colorectal carcinoma and renal cell carcinoma and matching healthy tissue, but were not able to determine any conclusive CD99 isoform expression pattern. This is most likely due to the fact that only 1-2% of all cells present in a tumor are endothelial cells and therefore it is very difficult to pick up specific splice variants.
These results indicate that CD99 splicing is tissue specific and provide an opportunity for specific targeting of CD99 isoforms in human tumor vasculature.

DISCUSSION
It was previously described that CD99 is overexpressed in inflamed vasculature. Our study demonstrated that CD99 is also For construction of the vaccine fusion protein TRXtr-mCD99, we used the protein sequence of the extracellular part of murine CD99 as described in Bixel et al. (36). We show that it is possible to induce a polyclonal antibody response against the self-antigen CD99 in immunocompetent mice by vaccination. Vaccination induced high levels of anti-mCD99 antibodies in the sera of the mice. This confirms the findings of previous studies using the same vaccination strategy for the induction of antibodies against different self-antigens (23,24,37).
In three independent studies a significantly smaller tumor volume was measured in the TRXtr-mCD99 vaccinated mice compared to control vaccinated mice. In this context, it is important to keep in mind that our vaccination strategy induces a polyclonal antibody response that is much more effective in inducing immune system activation than monocloncal antibodies (38). A polyclonal antibody response induces antibody-dependent cellular cytotoxicity (ADCC) where the antibodies function as a recognition and binding site for nonspecific toxic cells like natural killer cells and macrophages. It also induces complement-dependent cytotoxicity (CDC) where the antibodies activate the complement system which leads to the formation of the membrane attack complex (MAC) and subsequent lysis of the target cell (39,40).
In the CT26 model only low levels of CD99 are expressed by the tumor cells as compared to the osteosarcoma model. However, in the CT26 model also a significantly lower vessel density was observed in tumors of CD99 vaccinated mice. It is therefore likely that tumor growth inhibition in the CT26 model is mainly due to targeting of the tumor vasculature by the CD99 vaccine. The osteosarcoma model highly expresses CD99 and therefore inhibition of tumor growth in this model is due to targeting of both the tumor vasculature and the tumor cells, which leads to a more pronounced anti-tumor effect.
Anti-CD99 antibodies induced by the TRXtr-mCD99 were able to detect native CD99 in osteosarcoma tumor tissue.
However, a lot of background staining was observed when the sections were stained with serum from CD99 vaccinated mice. This can be explained by the fact that staining mouse tissue with murine antibodies is difficult. We therefore used a F(ab) fragment to prevent non-specific binding of the serum. However, with this approach still a lot of background staining was observed. We have considered to purify IgG from mouse serum or to specifically purify the anti-CD99 antibodies in the serum with antigen, but we did not have sufficient mouse serum to do so.
After vaccination against CD99 we found more vessels without pericytes in the osteosarcoma tumors. This indicates that vascular targeting leading to vessel destruction occurs rather than vascular normalization after which a higher pericyte coverage is expected (41,42). Vascular normalization is characterized by neutralization of growth factors, such as vascular endothelial growth factor (VEGF) (43). Neutralization of VEGF results in a more quiescent vasculature with more pericyte coverage and improved vascular flow. Vascular targeting on the other hand, leads to killing of the tumor endothelial cells, since these are attacked and removed by the immune system. This would explain our observations of a lower number of vessels with a lumen and pericytes in the tumors of CD99 vaccinated mice. As, the target CD99 is a transmembrane molecule tissue bound frustrated phagocytosis will occur (23) and not only the endothelial cells will be destroyed but everything in their vicinity as well, including pericytes.
No toxicity of the vaccination against CD99 was observed. In the tumor growth studies, no weight loss of the mice occurred during the study period. In addition, we monitored the body weight and health condition of CD99 vaccinated mice, with constantly high anti-mCD99 antibody levels in their sera, and control vaccinated mice, for a period of 45 weeks. In this study one mouse (CD99#2) was lost to follow-up. Considering the good health of all other CD99 vaccinated mice, this was probably due to non-treatment related conditions. We scrutinized the morphology of the organs of control and CD99 vaccinated mice and did not observe any changes in tissue morphology related to the presence of anti-mCD99 antibodies, neither based on vessel staining (CD31) of kidney vasculature and hematoxylin eosin staining of heart, lung, kidney, and liver. For the mouse that was lost to follow-up both the tissue and vessel morphology were normal/comparable to control vaccinated mice, implying that the loss of the mouse was probably not due side-effects of the vaccine. Taken together, our data suggest that vaccination against CD99 is safe, and provides a vascular targeting approach that lacks the risk of current angiostatic approaches for running into drug-induced resistance (44,45).
Expression of CD99 was upregulated in growth factor activated HUVEC, resembling tumor endothelial cells, as determined by Western blot analysis. In addition, CD99 expression on cultured endothelial cells was induced by nutrient deprivation, which suggests that expression of CD99 in the tumor vasculature is most likely regulated by microenvironmental stress. We did not investigate if distinct CD99 isoforms are induced by starvation. To define which human CD99 isoform is the main variant present in the tumor vasculature we would need to perform a RT-qPCR on mRNA isolated from by flow cytometry sorted tumor endothelial cells. In literature, two distinct human CD99 isoforms have been described (27,46). The full-length isoform (variant I, 32 kDa) and the truncated isoform (variant II, 28 kDa). Variant II includes a premature stopcodon caused by an insertion in the cytoplasmic domain and therefore lacks most of the cytoplasmic domain and is thought to be non-functional. The murine CD99 only shows 46% homology with human CD99 and resembles the human short isoform (28). However, if CD99 has the same function in mouse as in humans is unclear (29).
Currently, six different protein coding human CD99 isoforms are described in the NCBI database. We found contradictory results of CD99 expression on mRNA level and protein level in growth factor activated HUVEC. On mRNA level expression of CD99 seems to be downregulated in activated HUVEC, whereas on protein level CD99 expression is upregulated in activated HUVEC. On mRNA level the main CD99 isoform identified is variant 4, the short CD99 isoform, lacking part of the cytoplasmic domain. This is consistent with appearance of a 16 kDa protein band in higher passages of HUVEC and after starvation of the cells. If the 16 kDa protein band is a true splice variant of human CD99 is difficult to determine, since the anti-human CD99 antibody that we used is a polyclonal antibody that cannot distinguish between different splice variants. The human CD99 protein is highly O-glycosylated (47) and starvation of cells changes their glycosylation pattern (48), therefore the 16 kDa band observed in the Western blot might be nonglycosylated CD99. Also, in the Western blot a protein band between 35 and 25 kDa can be observed in some of the HUVEC cell lysates. We did however not further investigate if this could be a human CD99 splice variant.
In conclusion targeting of CD99 by vaccination inhibits tumor growth in different murine tumor models and is safe. Human CD99 is overexpressed in HUVEC and expression of CD99 is induced in culture and by nutrient deprivation. Also, a distinct human CD99 isoform is induced under these conditions, which is distinct form the isoforms expressed by EWS and healthy PBMC. These observations provide an opportunity for specific targeting of CD99 isoforms in human tumor vasculature.

DATA AVAILABILITY
All datasets generated for this study are included in the manuscript and/or the Supplementary Files.

AUTHOR CONTRIBUTIONS
EH designed research, performed experiments, analyzed data, and wrote the manuscript. IvdW, LF, LS, PvdL, and HH performed experiments, analyzed data, and edited the manuscript. JvB designed research, analyzed data, and edited the manuscript. VT designed research and edited the manuscript. AC performed experiments and edited the manuscript. AG designed research and wrote the manuscript.
inserted between the restriction sites NdeI and XhoI into the multiple cloning site (MSC). Protein expression is under the control of the IPTG-inducible T7lac promoter. Amp, Ampicillin resistance gene.
Supplementary Figure 2 | Additional data osteosarcoma study I and II, CT26 study and long-term follow-up study. (A) Antibody titers of anti-CD99 antibodies in the sera of TRXtr (n = 5; left panel) and TRXtr-mCD99 (n = 10; CD99; middle and right panel) vaccinated mice at time point 9 weeks of study I Os-P0107 (C3H mice). TRXtr vaccinated mice are devoid of anti-CD99 antibodies. (B) Anti-mCD99 antibody levels in the sera of the C3H mice (Os-P0107 model) at different time points (weeks) of study II (n = 5 mice per group). (C) Body weight of CD99 vaccinated (CD99; red) and control vaccinated mice (TRXtr; blue) of the osteosarcoma study I and II (left and middle panel) and the CT26 study (right panel). No difference in body weight between the treatment groups was observed in all three different studies. Values are depicted as mean ± SEM. [study I: TRXtr (n = 5); CD99 (n = 10); study II: TRXtr and CD99 (n = 5); CT26: TRXtr and CD99 (n = 4)] (D) Kidneys stained for CD31 (brown-reddish staining) of TRXtr-mCD99 (n = 5; CD99) and control vaccinated (n = 5; TRXtr) mice from the long-term follow-up study (time point 45 weeks). Tissues were counter stained with Mayer's hematoxylin (blue) (scale bar 50 µm). No difference in tissue morphology was found between TRXtr-mCD99 vaccinated and control vaccinated mice.