CD31 Expression Determines Redox Status and Chemoresistance in Human Angiosarcomas

Angiosarcomas are rare vasoformative sarcomas arising in the skin, soft tissues, and various other locations, including breast, liver, spleen, and bone (1). Patients with long-standing lymphedema (so-called Stewart–Treves syndrome; refs. 2, 3) or with a history of irradiation (4), for example, for breast cancer, are at increased risk. The median overall survival rates for angiosarcoma drop from 51 months in localized to only 12 months in metastatic tumors (5–7). Current therapeutic approaches include surgical excision for localized tumors and systemic therapy using agents such as doxorubicin or paclitaxel (8). However, most angiosarcomas soon become resistant to chemotherapy, and no treatment has so far been shown to prolong overall survival (7, 8). Clinical trials using antiangiogenic agents such as bevacizumab or multikinase inhibitors targeting angiogenesis pathways such as sorafenib, imatinib, or sunitinib were also mostly disappointing (9, 10).

The pathohistologic diagnosis of angiosarcoma rests on the morphologic and IHC identification of endothelial differentiation using antibodies against, for example, CD34, CD31, D2-40/podoplanin, and ERG. For largely unknown reasons, platelet-endothelial cell adhesion molecule 1 (PECAM1/CD31) is a very robust marker of endothelial differentiation in angiosarcoma and is usually expressed even in rare variants (11–13), pointing to an important role of PECAM1/CD31 in the biology of these tumors, whereas other endothelial markers, such as CD34, are frequently lost (11–13). CD31 expression is normally restricted to endothelial cells, circulating platelets, monocytes, and circulating T-cell subsets. PECAM1/CD31 is an adhesion receptor that has several important functions, including immune cell adhesion and transendothelial migration, coagulation, angiogenesis, and integrin activation (14, 15). CD31 has been shown to play an important role in the regulation of apoptosis as well as cell adhesion–mediated cell growth and in the maintenance of vascular permeability and integrity (16, 17). It was recently demonstrated that genetic depletion of CD31 in endothelial cells resulted in reduced VE-cadherin expression, increased survivin expression, and nuclear localization of Yes-associated protein (YAP), an important mediator of the Hippo pathway. This led to increased proliferation and decreased apoptosis through transcriptional activation of genes required to promote cell growth and inhibit apoptosis (17).

To study the biological function of CD31 in angiosarcomas, we analyzed the expression of CD31 in human angiosarcoma tissue samples and in angiosarcoma-derived cell lines. We here show that suppression of CD31 protects angiosarcoma from oxidative DNA injury and ROS production. Mechanistically, this redox protection is achieved through nuclear accumulation of YAP. Our data suggest activation of the Hippo pathway as a general mechanism in malignant endothelial tumors with potential relevance for the development of more effective treatments for these aggressive neoplasms.

All angiosarcoma samples were taken from a previously published cohort (18). IHC staining of CD31 was performed on 3-μm paraffin sections using standard procedures on a Dako Omnis platform [CD31 (JC70A), ready-to-use, Dako]. All experiments with human angiosarcoma samples were obtained with full patient consent and carried out in compliance with the ethical regulations approved by the Ethics Committee of the University Medical Center Göttingen (Gö 26/8/15) and of the University Medical Center Mannheim (Ma 2012-289N-MA).

The following primary human endothelial cell lines were used: human umbilical vein endothelial cells (HUVECs) and human dermal microvascular endothelial cells (HDMECs) were purchased from Promocell and passaged for seven times maximum. The angiosarcoma cell lines ASM (provided from Vera Krump-Konvalinkova, University Mainz, Germany; ref. 19), HAMON (provided by Daichi Hoshina; ref. 20), ISO-HASc.1 (subclone of the ISO-HAS cell line generated and provided by James Kirkpatrick, University Mainz, Germany; ref. 21) and the transformed fusion cell line EA.hy926 cell line (DMSZ). Murine endothelial cells from CD31^+/+ and CD31^−/− mice were extracted as previously described (22). HUVEC, HDMEC, and ASM cell lines were cultured in endothelial cell growth medium, advanced (Provitro), HAMON cells were cultured in EGM-2 BulletKit endothelial cell medium (Lonza), ISO-HASc.1, EA.hy926, and murine endothelial cells were cultured in DMEM supplemented with 10% serum, 1% l-glutamine, and 1% penicillin/streptomycin (all from Thermo Fisher Scientific).

The following antibodies were used: CD31 antibody clone JC70A from DAKO (used for IHC and immunoblot analysis), clone D8V9E from Cell Signaling Technology (immunoblot analysis). vWF (clone A0082) was purchased from DAKO. Antibodies against YAP (#14074), p-H2AX (γ-H2AX; #9718), caspase-3 (#9665), caspase-7 (#12827), PARP (#9542), Beclin-1 (#3495), LC3B (#12741), total GSK3β (#12456) and p-GSK3β (Ser9; #5558) and p-AKT (Ser473; #4060), Catalase (#12980) were all purchased from Cell Signaling Technology, mouse monoclonal β-actin (clone AC-74) from Sigma. Antioxidants N-Acetyl-l-cysteine (NAC), Ferrostatin-1 (Fer-1), and vitamin E (trolox); sodium arsenite (AsN), proteasome blocker MG-132 (MG), HSP90 inhibitor 17-AAG, MEK inhibitor U0126 and p38^MAPK inhibitor SB203580 were all purchased from Sigma-Aldrich. JNK inhibitor SP600125 and multikinase inhibitor pazopanib were from Selleckchem and selective p38α inhibitor JX401 from Tocris.

Cells at 60% to 70% density were treated with indicated conditions and time points. After subsequent washing in PBS, cells were scraped in RIPA lysis buffer. Total protein concentrations were analyzed using Bio-Rad protein assay. Cell extracts were immunoblotted as described elsewhere (23). For determination of differential expression of pro- and antiangiogenic factors, total protein extracts from indicated cell lines (200 μg total) were analyzed using the Human Angiogenesis Antibody Array (R&D Systems). All arrays were analyzed and quantified using ImageJ software (NIH, Bethesda, MD).

Cells were cultivated on glass coverslips to a confluence of 70% to 80% and after washing fixated in 4% buffered paraformaldehyde (PFA) for 10 minutes at room temperature. After PBS washing, nuclei were permeabilized in 0.3% Triton-X for 10 minutes at room temperature. After washing, unspecific binding sites were blocked with 1% BSA for 30 minutes at room temperature. After an additional washing, cells were subjected to primary YAP antibody (#14074, Cell Signaling Technology) in a 1:100 dilution at room temperature for 1 hour. Cells were then incubated with a fluorescent secondary antibody Alexa 488 (1:600) and DAPI (1:2,500; both Thermo Fisher Scientific) for 1 hour at room temperature. Coverslips were then upside down embedded in Mowiol and fluorescent signals were scanned on a confocal laser scan microscope with a ×60 optical enlargement (Olympus). Immunofluorescence stainings on formalin-fixed, paraffin-embedded tissues were performed on 3-μm sections. After deparaffinization, rehydration, and antigen retrieval in citrate buffer (pH 6.1), the rabbit monoclonal anti-ERG (EP111) and the mouse anti-CD31 mAb (PECAM-1; JC70A; both Dako) were incubated in the Autostainer (Dako) according to the recommended protocol. Thereafter, the sections were sequentially incubated for 30 minutes with a secondary anti-rabbit antibody labeled with Alexa488 (Thermo Fisher Scientific) and a secondary anti-mouse antibody labeled with CY3 (Thermo Fisher Scientific). Samples were embedded in Mowiol, and fluorescent signals were detected using a confocal laser scanning microscope with a ×60 optical and ×1.5 digital enlargement (Olympus).

In brief, 100 μL of Matrigel (BD Biosciences) was added to each well of a 96-well plate and allowed to polymerize for a minimum of 30 minutes at 37°C. Indicated cell lines were resuspended in endothelial cell medium and seeded in each well at a concentration of 1 × 10⁵ cells/well. Cells were monitored under a light microscope to determine the formation of tube-like structures and representative pictures were taken at indicated time points. HUVECs were used as a positive control–forming capillary-like structures as early as 4 to 6 hours and matured after 24 hours.

Cells were transfected with 80 nmol/L siRNA against either CD31 (Hs_PECM1_7, catalog. No. SI05098863) or YAP (Hs_YAP1_5, catalog No. SI0266254; Qiagen) using HiPerFect Transfection Reagent (Qiagen) according to the manufacturer's protocol. Briefly, 100 μL transfection medium containing 12 μL HiPerFect, 9.6 μL siRNA/negative control (20 μmol/L), and 88.4 μL RPMI160 medium without FCS was incubated at room temperature for 20 minutes and added to 3 × 10⁵ cells in 2.3-mL medium just after seeding. Cells were incubated for 24 hours and further processed for immunoblot analysis.

For determination of CD31 expression, cells were incubated with CD31-PE (monoclonal, 390, Biolegend) for 30 minutes, washed twice with ice-cold PBS, and analyzed and sorted by a BD FACS Aria. For analyses of intracellular cytoplasmic ROS, cells were incubated in FCS-free medium containing 20 μmol/L 2′,7′ –dichlorofluorescin diacetate (DCFDA) for 10 minutes at 37°C, for assessing lipid peroxidation incubated with 20 μmol/L C11-BODIPY581/591 dye for 30 minutes and for mitochondrial ROS incubated with 20 μmol/L MitoSOX dye for 30 minutes (all from Invitrogen), washed twice and analyzed via flow cytometry. For cell tracking, indicated cell lines were incubated with CellTrace Violet dye (final concentration 5 μmol/L) for 20 minutes at 37°C protected from light. After quenching unbound dye, stained cells were combined with unstained cells and left untreated or incubated with increasing doxorubicin doses. After 24-hour incubation, all cells were washed in PBS and analyzed via flow cytometry using 405-nm laser and 450-nm emission filter. Viable marked cells were quantified and presented as the percentage of labeled cells to untreated controls.

CD31^high and CD31^low cells were tested for catalase activity using the OxiSelect Catalase Activity Assay (Cell Biolabs) according to the manufacturer's protocol. In brief, cells were harvested and sonicated in cold PBS with 1 mmol/L EDTA. Samples were centrifuged at 10,000 × g for 15 minutes at 4°C and supernatant was used for catalase activity measurement. Twenty microliters of the diluted catalase standards or unknown samples were incubated exactly for 1 minute with 50 μL hydrogen peroxide working solution (12 mmol/L). Then, the reaction was stopped with 50 μL catalase quencher. Five μL of each reaction were mixed with 250 μL of the chromogenic working solution. After 40-minute incubation, absorbance was measured at 520 nm. Data are presented as relative values.

For RNA isolation, cells were harvested by trypsination and pelleted by centrifugation at 800 × g. The dry pellet was lysed in 1 mL TRIzol (Invitrogen) and processed according to the manufacturer's protocol. Fifty nanograms of total RNA was used to measure gene expression. Expression profiles were established using a nCounter custom made Panel for oxidative genes and the nCounter Digital Analyzer (NanoString Technologies) according to the manufacturer's protocol.

Unsorted wild-type ASM cells with a previous determined high (48%) CD31^low fraction was used. Nontargeting scrambled shRNA (plasmid #1864) was obtained from Addgene. Lentiviral particles were produced in HEK293T cells with plasmids pCMV-Δ-R8.91 (containing gag, pol, and rev genes) and pMD.G (VSV-G-expressing plasmid) following standard protocols. After transduction, cells were kept for 72 hours under 2 μg/mL puromycin (InvivoGen) resulting in more than 70% dead cells. Surviving clones (initial 27) were picked, expanded, and six representative clones were immunblotted for CD31.

For cell viability analysis, equal numbers of cells were seeded into 96-well flat-bottom plates and incubated for indicated time points. Cell viability was determined by adding MTS (Promega) to each well for 2 hours at 37°C, and then absorbance was measured at 490 nm. For colony formation, cells were diluted in medium and plated in methylcellulose (from R&D Systems) at 1 × 10³ cells/30-mm plates. For treatment experiments, cells were incubated for 24 hours with indicated compounds. After treatment, cells were trypsinized, pelleted, and seeded in methylcellulose as outlined above. After incubation for at least 7–16 days at 37°C in 5% humidified CO₂, formed compact colonies were counted using inverted microscopy.

Chick chorioallantoic membrane (CAM) assay was used to determine in vivo tumorigenicity with small modifications as described elsewhere (24). In brief, equal cell numbers (2.5 × 10⁶/egg) of indicated ASM sublines were mixed with Matrigel (BD Matrigel) and implanted on the CAM on embryonic day 10 (E10) chicken embryos. On day 10 (E20) after implantation, formed tumors were explanted, weighed, and photographed.

All experiments were performed at least in triplicates unless specified otherwise. Differences between treatment groups were either evaluated using the unpaired t test, one- or two-way ANOVA followed by Bonferroni post hoc analysis, as indicated. Statistical analyses were performed using GraphPad Prism Software (GraphPad Software Inc.)

CD31 expression was assessed in 17 full-section paraffin samples of human angiosarcomas. Of these, the majority (n = 11; 65%) showed a small population of CD31^low tumor cells (Fig. 1A; Supplementary Fig. S1A). Representative nuclear ERG immunreactivity revealed the endothelial nature of CD31^low tumor cells (Fig. 1B).

To analyze CD31 expression in healthy and malignant endothelial cell lines, we used primary endothelial cells derived from HUVEC and HDMEC cell lines, a hybrid immortalized cell line (EA.hy926) in which normal HUVEC cells had been fused with the lung adenocarcinoma cell line A549 (25), and the human angiosarcoma cell lines ASM, ISO-HASc.1 (a cell line subcloned from the original ISO-HAS cell line; refs. 19, 21) and the newly established HAMON cell line (20). Western blotting showed strong CD31 expression in all cell lines, but higher levels in normal endothelial cells compared to the hybrid cells and angiosarcoma cell lines (Fig. 1C and D). To further investigate whether CD31 heterogeneity was also present in these cell lines, we performed FACS using a primary phycoerythrin (PE)-labeled CD31 antibody and isotype-specific control antibodies. In the non-neoplastic HUVEC and HDMEC cells, 99.6% of the cells showed high levels of CD31. In contrast, analysis of the tumor cell lines showed a small but consistent CD31^low population (ranging from 1.02% in HAMON to 60% in ASM), which was reproducible in repetitive tests independent of the passage number or cultivation method although the population size was variable (Fig. 1E).

We conclude that both human angiosarcoma tissue samples as well as cell lines contain variable numbers of CD31^low tumor cells.

Because ASM cells contained the highest fraction of CD31^low cells, we mainly focused on this cell line for further experiments. To establish stable sublines of CD31 high versus low cells, we transduced ASM cells with the empty lentiviral vector pLKO.1 puro under puromycin selection, and subsequently picked resistant single-cell colonies (Fig. 2A). Six representative clones were expanded and subjected to immunoblot analysis for CD31. For following experiments, ASM subclones #3 (termed CD31^low hereafter) and #6 (termed CD31^high hereafter) were used. Both subfractions preserved their initial CD31^high versus CD31^low phenotype even after multiple passages, whereas the expression level of von Willebrand factor (vWF) was similar, suggesting that both cell fractions were endothelial in nature in spite of different CD31 levels (Fig. 2B).

Next, we sought to characterize the ASM CD31^low and CD31^high sublines. Under normal culture conditions containing 20% serum and low cell density, CD31^high cells appeared oval-shaped and built a cobblestone pattern as monolayer, similar to primary HUVECs and HDMECs. In strong contrast, CD31^low cells were more spindly, elongated, and flattened, resembling fibroblasts (Fig. 2C).

To study their capacity to form vascular structures, we performed tube formation assays and included HUVECs as a positive control. As expected, HUVECs formed organized networks of vascular structures after 6 hours that matured after 18 hours. In CD31^high cells, pretubular structures could be detected after 6 hours and mature tubes after 18 hours. In strong contrast, CD31^low cells did not show tubular structures at any time point, even after extended observation for up to 24 hours (Fig. 2D).

To further characterize CD31^high and CD31^low cells at the molecular level, we measured 55 angiogenesis-related proteins in whole-cell lysates of both ASM sublines. Compared with CD31^high ASM cells and HUVECs, many angiogenic factors such as angiopoietin-2, PDECGF (thymidine phosphorylase), and endothelin-1 were strongly suppressed in CD31^low cells, whereas established angiogenic inhibitors such as TIMP-1 were upregulated. Surprisingly, several other potent factors involved in angiogenesis such as HGF, amphiregulin AR, and VEGF were strongly activated along with suppression of the potent angiogenic inhibitor thrombospondin-1, suggesting a complex molecular counter-regulation to compensate for the loss of CD31 function (Fig. 2E).

To further study the biological differences between CD31^high and CD31^low cells, we compared their cell growth under normal culture conditions and under serum-reduced starvation. CD31^low cells showed a significantly higher proliferation rate than CD31^high cells under normal culture conditions (Fig. 3A) and higher cell survival under serum deprivation (Fig. 3B). Moreover, CD31^low cells survived up to 16 days in semi-solid methylcellulose media and were even able to form conglomerates of two-dimensional colonies. In strong contrast, CD31^high cells were mainly dispersed as single cells after 7 days that did not survive up to 16 days under these conditions (Fig. 3C). Similar results could be also observed in normal HUVEC cells (data not shown). According to the initial report and own previous observations, ASM and ISO-HASc.1 cells failed to establish tumors when xenografted into immunodeficient mice. One reason could be lack of an established microenvironment needed for tumor formation and propagation (19). Therefore, we used the CAM xenograft assay to analyze in vivo tumorigenicity (24). The CAM is a highly vascularized extraembryonic membrane providing optimal delivery of growth supplements, notably with an immature immune system. Indeed, under these conditions, both sublines formed detectable tumors 10 days after implantation. In line with our in vitro results, inoculated CD31^low cells formed significantly larger and heavier tumors than their CD31^high counterparts (Fig. 3D).

Taken together, these results indicate that CD31^low cells represent a highly proliferative, stress-resistant, and tumorigenic subpopulation that outcompetes vasculogenic CD31^high cells.

Anthracycline-based chemotherapy is the backbone of current angiosarcoma therapy by improving local disease control, but does not result in any survival advantage (8, 26). We treated both CD31 sublines with increasing concentrations of doxorubicin for 24 hours and measured cell survival using MTS assay. At concentrations >500 nmol/L, CD31^low cells survived significantly better than CD31^high cells with only a 20% to 30% decrease in cell viability at 10 μmol/L doxorubicin (peak plasma concentration achieved in patients ranging between 5 and 15 μmol/L; ref. 27; Fig. 4A). In accordance with this observation, Western blot analysis showed increasing levels of cleaved PARP, and effector caspases-3 and -7 as indicators of apoptosis only in CD31^high, but not in CD31^low cells (Fig. 4B; Supplementary Fig. S2A). Because suppressed CD31 levels propagated chemoresistance, we next asked whether doxorubicin treatment results in selection of CD31^low cells. We therefore used unsorted wild-type ASM cells with a predominant CD31^high (92.8%) and a smaller CD31^low (3.7%) subpopulation. Indeed, 1 μmol/L doxorubicin efficiently killed the majority of cells after 24 hours. However, the residual cells that fully recovered after 12 days had a fibroblast-like morphology (Supplementary Fig. S2B) and had lost their vasculogenic capability (Supplementary Fig. S2C). In agreement with this phenotype, Western blot analysis demonstrated low CD31 protein levels in doxorubicin-surviving cells and flow cytometric analysis clearly revealed a shift toward CD31^low cells as the predominant subpopulation (Fig. 4C and D). To further elucidate this selection for the CD31^low phenotype under chemotherapy more precisely, we first labeled CD31^high cells with CellTrace Violet dye, mixed them at a 1:1 ratio with unstained CD31^low counterparts, and challenged the mixture with increasing doxorubicin concentrations (Fig. 4E). After 24-hour treatment, flow cytometric tracking of the cell trace dye showed a dose-dependent depletion of the CD31^high population. In strong contrast, the reverse experiment with CD31^low cells labeled and mixed with equal numbers of unlabeled CD31^high cells clearly showed that doxorubicin treatment left the population size of CD31^low cells nearly unaffected (Fig. 4F and G).

In conclusion, these results indicate that treatment with doxorubicin depletes CD31^high cells and selects for CD31^low cells rather than leading to downregulation of CD31 in CD31^high cells.

To get a deeper insight into the molecular effects of CD31, we first analyzed the steady-state expression level of various specific cell death markers in both CD31 cell lines. No differences in basal activation of apoptosis (caspase-3) and autophagy (beclin-1 and LC3) could be observed. Interestingly, CD31^high cells had markedly higher levels of phosphorylated histone variant H2AX (γ-H2AX), a reliable marker of double-strand DNA breaks, than CD31^low cells (Supplementary Fig. S3A). In a parallel experiment, we performed a focused pharmaceutical screen for compounds that could restore the diminished colony formation capacity of CD31^high cells. Surprisingly, potent antioxidants, such as N-acetyl-l-cysteine (NAC) that replenishes intracellular levels of the antioxidant glutathione (GSH) or the water-soluble vitamin E analogue trolox and ferrostatin-1 (Fer-1), both known lipid-ROS scavengers, were capable to significantly increase the colony formation capacity in semi-solid methylcellulose (Supplementary Fig. S3B and S3C). Vice versa, treatment with the potent ROS inducer sodium arsenite (AsN) dramatically decreased the colony formation of CD31^low cells (Supplementary Fig. S3D). These experiments hinted toward marked differences in the redox status between both CD31 subpopulations. To test this assumption, we evaluated the endogenous levels of cytoplasmic reactive oxygen species (ROS) via dihydrodichlorofluorescein diacetate (DCF), lipid peroxidation levels via C11 undecanoic acid (C11-BODIPY), and mitochondrial superoxide level using MitoSox dyes. All three ROS indicators conclusively revealed significantly lower levels of cytoplasmic, lipid, and mitochondrial-derived ROS entities in CD31^low compared with CD31^high cells (Fig. 5A). Furthermore, exogenous addition of the ROS inducer H₂O₂ (500 μmol/L over 20 minutes) resulted in a significant ROS release in CD31^high but not in CD31^low cells (Fig. 5B). In a similar manner, serum reduction potently evoked starvation-induced oxidative DNA damage only in CD31^high cells (Fig. 5C). These results suggest that CD31^low cells have a much greater ROS-scavenging capacity than CD31^high cells.

We next performed a focused gene-expression analysis using a custom-designed NanoString array. This technique quantifies specific RNA molecules by dual probe hybridization and allows direct measurement and comparison of absolute transcript levels. Compared with CD31^high cells, CD31^low cells showed overexpression of 10 previously described YAP/TAZ target genes (Fig. 5D; Supplementary Table S1). As a transcriptional coactivator, YAP binds to promoters of multiple target genes, including the antioxidative enzyme catalase (CAT; ref. 28; Fig. 5D; Supplementary Table S1). CD31^low cells showed an 80% greater steady-state enzymatic catalase activity compared with CD31^high cells (Fig. 5E). It was previously demonstrated that Akt activation propagates higher YAP phosphorylation causing its enhanced proteasomal degradation (29). In agreement with these findings, CD31^high cells showed increased phosphorylation levels of Akt (Ser473) and its downstream target kinase GSK3β (Ser9; ref. 30; Fig. 5F). Moreover, CD31^low cells not only contained markedly higher YAP protein levels (Fig. 5F), but also showed higher nuclear YAP accumulation compared with CD31^high cells (Fig. 5G).

We confirmed the redox-regulating role of CD31 by incubating CD31^high cells with a selective antagonizing CD31 antibody (clone JC/70A) that not only effectively blocked tube formation (Fig. 6A), but also resulted in an increased YAP expression along with reduced steady-state ROS and less oxidative DNA damage as assessed by γH2AX (Fig. 6B and C). In a similar fashion, genetic CD31 depletion in CD31^high cells using an siRNA approach resulted in YAP accumulation and suppressed oxidative stress and DNA damage levels (Fig. 6D and E). Moreover, siRNA-mediated YAP suppression in CD31^low cells consequently induced ROS and DNA damage (Fig. 6F and G). To confirm the molecular link between CD31 expression, endothelial cell morphology, YAP signaling, and ROS detoxification as a general mechanism, a number of control experiments were performed. First, we demonstrated increased YAP and reduced ROS as well as γH2AX levels in the previously established doxorubicin-surviving ASM cells compared with their parental cells (Supplementary Fig. S4A and S4B). Second, ISO-HASc.1 cells were sorted for the CD31^high and CD31^low population. Again, ISO-HASc.1 CD31^low cells grew better in semi-solid methylcellulose, had more YAP and lower ROS and less DNA damage compared with CD31^high cells (Supplementary Fig. S4C–S4H). Finally, we compared previously established murine endothelial cells isolated from wild-type (CD31^+/+) or CD31 knockout (CD31^−/−) mice (22). Morphologic analysis showed that cells from CD31^+/+ animals preferentially grew in cobblestone patterns and formed small colonies, whereas cells from CD31^−/− animals had a more elongated, fibroblastic morphology (Supplementary Fig. S4I). The CD31^−/− cells had lower Akt activation and higher YAP expression than their wild-type counterparts. Basal ROS and DNA damage levels were significantly lower in CD31^−/− cells compared with their wild-type controls (Supplementary Fig. S4J and S4K). In summary, the aggressive and redox-protective phenotype of CD31-depleted angiosarcoma can be attributed to YAP activation and subsequent induction of YAP-regulated antioxidative target genes.

In a recent study, the FDA-approved multikinase inhibitor pazopanib was described to enhance proteasomal YAP degradation in cancer cells (31). We therefore analyzed the effects of pazopanib on the doxorubicin-resistant CD31^low cell fraction in angiosarcoma. Increasing doses of pazopanib over 24 hours decreased YAP protein and induced γH2AX levels in a concentration-dependent fashion (Fig. 7A). Moreover, pazopanib dramatically reduced catalase activity (Fig. 7B) resulting in a marked induction of total ROS levels (Fig. 7C). In contrast, the same treatment had no significant effect on catalase activity and ROS or γH2AX levels in CD31^high cells (Fig. 7D–F). Importantly, combined treatment of CD31^low cells with pazopanib plus doxorubicin markedly suppressed cell viability and colony formation by inducing apoptosis (Fig. 7G–I). In conclusion, our results indicate that pazopanib helps to deteriorate the redox status preferentially in “YAP-addicted” CD31^low cells and thereby resensitizes them for chemotherapy (Fig. 7J).

The identification of mechanisms by which malignant neoplasms evade current chemotherapies is a major goal to improve patient outcome. Intratumorous heterogeneity is believed to be one major factor that helps tumors to escape.

We here show for the first time that most if not all tested human angiosarcoma specimens as well as angiosarcoma-derived cell lines contain a small population of CD31^low endothelial cells that we did not observe in non-neoplastic endothelial cell cultures. This subpopulation has fewer endothelial phenotypic properties, is resistant to starvation-induced oxidative stress, is more tumorigenic, and more resistant to anthracycline-based chemotherapy. It is therefore tempting to speculate that this subpopulation could be responsible for the chemoresistance seen in most angiosarcoma patients.

One major downstream effect of CD31 seemed to be degradation versus nuclear accumulation of the putative oncogene and cotranscription factor YAP. YAP and WWTR1 (also called TAZ, for transcriptional coactivator with PDZ-binding motif) are prime mediators of the Hippo pathway, essentially contributing to cell proliferation, stem cell function, tissue regeneration, and tumorigenesis. YAP/TAZ are known to function as signaling nexus and integrators of several other prominent signaling pathways such as the Wnt, G protein–coupled receptor (GPCR), EGF, and others (reviewed in refs. (32–34). In addition, it was recently found in cardiomyocytes that YAP in conjunction with the transcription factor FoxO1 positively regulate the transcription of antioxidative genes, such as catalase and MnSOD (28). Moreover, YAP overexpression in pancreatic-β-cells conferred apoptosis-resistance via upregulation of the thioredoxin (TXN) antioxidant pathway (35). In good agreement with these data, our gene expression profiling demonstrated that CD31^low cells, besides showing marked induction of canonical YAP target genes (ANKRD, CTGF, and CYR61), expressed higher levels of FoxO1, TXN, and CAT. This resulted in improved protection from ROS and subsequent DNA damage in CD31^low cells (Fig. 5D; Supplementary Table S1).

The finding that deregulation of the Hippo pathway plays an important role in neoplastic endothelial cells is not without precedence: epitheloid hemangioendotheliomas, another important group of malignant vascular tumors, are characterized by a common t(1;3) chromosomal translocation that fuses the Hippo downstream target WWTR1 (also called TAZ, for transcriptional coactivator with PDZ-binding motif), to CAMTA1, a transcription activator in cell-cycle regulation (36–38). More recently, a small subset (10%) of hemangioendotheliomas was shown to carry a YAP–TFE3 fusion gene (39). WWTR1/TAZ and YAP are both oncogenes that drive proliferation and tumorigenesis (40, 41). It has recently been estimated that activation of WWTR1 and YAP occurs in about 50% of soft tissue sarcomas, including angiosarcomas, and is associated with poor prognosis (42).

The nuclear translocation and activity of YAP are determined by posttranslational modifications, such as phosphorylation. YAP phosphorylation at its Ser-127 residue causes its cytoplasmic sequestration. Recently, it was shown that Akt can lead to Ser-127 phosphorylation independently from Hippokinases LATS1/2 (29, 43). We found that high CD31 expression correlates well with activation of Akt and subsequent sequestration of YAP, whereas low levels of CD31 and Akt lead to increased nuclear translocation and activation of YAP with transcriptional induction of ROS-detoxifying enzymes such as catalase (see scheme in Fig. 7J). Frequent activation of the PIK3CA/AKT pathway in angiosarcomas has been previously noted (44). This advantage in oxidative stress coping was most likely the explanation for the observed resistance of CD31^low cells against doxorubicin treatment, whereas CD31^high cells were doxorubicin sensitive. Given the apparent dependence of CD31^low cells on YAP signaling, we reasoned that YAP was a good target for reversing chemoresistance in angiosarcoma. We therefore chose the multitargeted tyrosine kinase inhibitor pazopanib for two main reasons. First, pazopanib was recently identified as a potent YAP inhibitor causing its cytoplasmic retention and proteasomal degradation in breast cancer cells (31). Second, pazopanib has been approved for the treatment of soft-tissue sarcomas including angiosarcoma with promising results in ongoing phase II trials alone (NCT01462630) or in combination with the endoglin/CD105 blocker TRC0105 (NCT01975519). We showed that pazopanib treatment in vitro at therapeutically relevant doses not only diminished YAP levels, but also suppressed the YAP downstream target catalase resulting in increased ROS levels selectively in CD31^low cells. Interestingly, despite higher oxidative stress levels, pazopanib alone had no effect on apoptosis compared with untreated CD31^low controls. However, combination with low-dose doxorubicin resensitized the cells for apoptosis resulting in a synergistic cytotoxicity and inhibition of colony formation. A recent study (although not in the context of YAP signaling) also observed that pazopanib induced a shift toward a pro-oxidative state in induced pluripotent stem cells (45).

An important question that needs to be addressed in future studies is how population homeostasis is maintained between CD31^high and CD31^low cells. We found that CD31^low cells contained higher levels of the transcription factor Slug and lower VE-cadherin levels than CD31^high cells (Supplementary Fig. S5). Slug is a key transcriptional repressor of CD31 and VE-cadherin, coordinating the transition of endothelial into mesenchymal fibroblast-like cells (46). Recent studies demonstrated the critical role of Notch signaling that drives Slug induction, independently of TGFβ signaling (47). However, although these findings could explain the differences in CD31 expression and morphology, they cannot answer the question whether transitions between the two populations occur.

In summary, our results show that CD31 plays an important biological role in human angiosarcoma by regulating endothelial cell function and redox status via the Hippo signaling pathway member YAP. Specifically, most if not all human angiosarcoma contain a small population of tumor cells with a CD31^low phenotype and increased YAP levels and improved redox protection. This intrinsic resistance represents a universal mechanism that shields tumor cells against the first contact with cytotoxic agents and provides cells with survival time to acquire longer-lasting resistance mechanisms. The presented data point to an emerging pivotal role of the Hippo pathway in malignant endothelial tumors (hemangioendotheliomas and angiosarcoma) with new clues for the development of more effective treatments of these aggressive neoplasms. The proof-of-concept that combination of pazopanib and doxorubicin could specifically resensitize CD31^low cells toward doxorubicin indicates that YAP inhibition together with conventional chemotherapy may be a valid new therapeutic perspective in angiosarcoma.

No potential conflicts of interest were disclosed.

Conception and design: V. Venkataramani, P. Ströbel

Development of methodology: V. Venkataramani, G.G. Wulf

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Venkataramani, S. Küffer, K. Cheung, G.G. Wulf, P. Ströbel

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Venkataramani, S. Küffer, X. Jiang, G.G. Wulf, P. Ströbel

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Venkataramani, L. Trümper, G.G. Wulf, P. Ströbel

Study supervision: V. Venkataramani, P. Ströbel

We wish to thank Vera Krump-Konvalinkova (Ludwig-Maximilians-Universität, Munich, Germany) for the ASM cell line, James Kirckpatrick (University Mainz, Germany) for providing the ISO-HAS subclone ISO-HASc.1 and Daichi Hochina (Hokkaido University Graduate School of Medicine) for the HAMON cell line. We furthermore thank Sabrina Becker for flow cytometric analysis, Christina Kiecke for colony formation analysis, Verena Pape for experimental support and Nuno Raimundo (University Medicine Göttingen, Department of Cellular Biochemistry) for providing the MitoSOX dye. V. Venkataramani was supported by the intramural research grant (Forschungsförderung) of the University Medical Center Göttingen and the Else-Kröner-Fresenius Foundation. X. Jiang was supported by the NCI cancer center core grant P30 CA008748. PS was supported by the Deutsche Krebshilfe (KO.SAR network).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.