The mechanisms underlying the origin and progression of hemangiosarcoma remain unclear. In humans, angiosarcoma can be associated with exposure to DNA-damaging agents or, in the case of Kaposi sarcoma, with infection by HHV8 in immunosuppressed patients [1
]. In mice, hemangiosarcoma can develop in susceptible strains treated with both genotoxic and non-genotoxic agents [2
]. In dogs, however, the disease occurs sporadically (not as a heritable condition) and with alarming frequency in the absence of known mutagens; a canine gamma herpes virus also has not been characterized [50
]. Indeed, preliminary experiments using well established methods to detect gamma herpes viruses [51
] yielded no amplification products when applied to these canine hemangiosarcoma samples (M. Duckett and M. Cannon, unpublished results), suggesting the etiology of canine hemangiosarcoma does not involve infection by a gamma herpes virus.
Angiosarcoma in humans and hemangiosarcoma in dogs are rapidly progressive diseases that are poorly responsive to conventional therapy. An improved understanding of their pathogenesis is needed to develop effective strategies for prevention and treatment. Our data suggest that inflammation and angiogenesis (defined by enrichment of cytokines and adhesion molecules that may be downstream effectors of a single molecule, like IL8, IL6, or IL1, as well as by robust upregulation of VEGF, MMPs and TIMPs, PDGF and PDGFRs, and others) are two general processes that are central to the pathogenesis of canine hemangiosarcoma.
Here, we first tested the hypothesis that known genes that regulate VEGF, including VHL and members of the Ras family were targets of mutation in canine hemangiosarcoma. We previously showed that the PTEN/Akt pathway that appears to be essential in other vascular tumor models is intact in hemangiosarcoma cells [23
]. Our observations that every hemangiosarcoma sample tested had wild type sequence for VHL, N-Ras, K-Ras, and H-Ras, no significant elevations of HIF1α, and no constitutive activation of Erk1 and Erk2 suggest that dysregulated VEGF production and the aggressive proliferation seen in these tumors are probably mediated by mechanisms that are independent from abnormalities of VHL and Ras genes. Recently, Pressler reported similar findings with regard to mutations of VHL in sporadic canine renal cell carcinoma [52
]. When considered along with the estimate that solid tumors from humans (colon and breast carcinomas) carry an average of ~100 gene mutations, these results suggest the probability to identify recurrent abnormalities by candidate gene approaches based on lineage similarity or dysregulation of a single known pathway is low. Indeed, even the recurrent mutations of the C-terminal domain of PTEN that we characterized previously were unlikely to be singularly responsible for the behavior of hemangiosarcoma or tractable for therapy. We hence tested the hypothesis that canine hemangiosarcoma would show characteristic gene expression profiles that would be informative for etiology and progression.
Genes that regulate cellular metabolism, cell cycle and cell signaling, cell-cell interactions, survival and apoptosis, angiogenesis, transcription, and the immune response were among those dysregulated in hemangiosarcoma cells when compared to non-malignant proliferating endothelial cells. Our data specifically highlight pathways that are important in response to hypoxia or angiogenesis, malignant transformation, and inflammation. However, altered expression of genes within these functional categories could describe virtually any solid tumor (where cyclins, glucose transporters, and other genes associated with the mitotic cell cycle commonly show elevated expression) from its normal counterparts. For example, we recently showed that elevated expression of CDKN1a (p21) confers chemoresistance to renal cell carcinomas, which share common hypoxia-induced, pro-angiogenic signatures with vascular tumors [53
]. The presence of hypoxia-inducible genes, including HGF, VEGF, bFGF, ADM, and PTGER4 was especially predictable [54
]; in most tumors, cells become hypoxic and upregulate genes that promote blood vessel outgrowth and that control metabolic processes such as vasodilation that can make cells normoxic. Clonal evolution in the tumor, and perhaps the environment in cell culture might favor selection of cells that upregulate such hypoxia response genes, although expression of these genes might be inherent to tumors of blood vessel forming cells.
Recently, Antonescu et al reported that human angiosarcomas have unique gene expression profiles when compared to other soft tissue sarcomas [56
], with notable enrichment of expression of vascular-specific receptor tyrosine kinases TIE1, KDR (VEGFR2), SNRK, TEK, and FLT1 (VEGFR1), and other genes that are prototypical endothelial markers including EPHA2 and PDGFβ. The overlap in the gene lists from Antonescu et al [56
] and from our data supports the similarities between human angiosarcoma and canine hemangiosarcoma. However, an interesting contrast is Antonescu's observation that VEGF expression was lower in angiosarcoma than other soft tissue sarcomas, versus our observation showing enriched expression of VEGF in hemangiosarcoma cells as compared to non-malignant endothelial cells from splenic hematomas. This may be due simply to the relative comparisons of tumor vs. tumor (by Antonescu et al) and tumor vs. non-tumor (in this study), which also could explain the lack of enrichment for vascular-specific receptors in our study, as these molecules would be expressed both in hemangiosarcoma cells and in non-malignant endothelial cells. In this respect, an intriguing finding from our data was the specific enrichment of VEGFR1 in hemangiosarcoma cells derived from golden retriever tumors compared to hemangiosarcoma cells derived from tumors of dogs from other breeds [40
], highlighting the potential utility of the organization and the relative homogeneity of dog breeds to understand how heritable factors might influence tumor pathogenesis.
We also were not surprised to find alterations in the expression of genes that contribute to malignant transformation and inflammation, including those that mediate cellular adhesion, stromal degradation or invasion (metalloproteinases), and the pathogenesis of leukemia [57
]. For example, among the receptor tyrosine kinases found by Antonescu [56
], TIE1 governs expression of inflammation-associated genes by endothelial cells [61
]. IL8 and IL6, both of which were enriched in our hemangiosarcoma samples, are recurrently associated with inflammation that "benefits" tumors (Figure S3), and PTGS2 (a.k.a., COX-2) is the single most common tumor-associated pro-inflammatory mediator [16
]. It is especially interesting that expression of PTGS2/COX-2 was enriched in our samples, as it was previously reported that the enzyme was undetectable by immunohistochemistry in formalin-fixed samples from canine hemangiosarcomas [62
]. There are several non-mutually exclusive explanations for this finding, including greater sensitivity in the expression microarray platform than immunohistochemistry, inefficient translation or relatively short protein half-life for Cox-2 in these tumors, or induction of the gene when cells are removed from the tumor microenvironment. Additional work will be required to clarify the role of PTGS2/COX-2 in hemangiosarcoma.
As may be true for hypoxia response genes, upregulation of proinflammatory genes in hemangiosarcoma also could result from selective pressures to create a favorable microenvironment for growth and survival [63
]. Tumors have been likened to "wounds that never heal" [64
], which is reflected by shared expression of genes mediating breakdown of the extracellular matrix, productive chronic inflammation, and angiogenesis. For example FN-1, which was overexpressed in all hemangiosarcomas evaluated in this set of experiments, is involved in wound healing, blood coagulation, and cancer metastasis. FN-1 also increases MMP9 activity, which together with urokinase is involved in tumor cell invasion through the extracellular matrix [65
]. The correlation between FN-1 and urokinase is interesting, since the survival rate of patients and dogs with angiosarcoma and hemangiosarcoma, respectively, is exceptionally poor due to its exceedingly high metastatic potential. The role of urokinase and FN-1 to promote the metastatic phenotype has been the subject of intense study in other tumors [46
], but it remains to be examined in hemangiosarcoma.
On the other hand, angiogenic and inflammatory signatures might reflect the ontogeny of hemangiosarcoma, rather than selection in the tumor microenvironment. Inflammatory infiltrates are commonly seen in canine hemangiosarcoma, but rather than reflecting recruitment of tumor-associated macrophages and myeloid cells due to inflammation, perhaps leukocytes may actually be derived from a population of multipotent progenitor cells that give rise to hemangiosarcoma. Recent data suggest that classical cell markers for endothelial and myeloid origin cells are less tissue specific than historically thought [67
], possibly due to a the existence of a shared hematopoietic/endothelial progenitor (the putative angioblast). We proposed recently that hemangiosarcomas might arise from such a cell [36
], while Yoder et al described a similar population of myeloid cells that are intimate participants in blood vessel formation [37
]. This cell is a "vascular mimic" that can express a variety of cell surface proteins associated with endothelial precursor cells (CD133, CD34, VEGFR2), but also proteins that belie hematopoietic origin (CD45, CD14, and CD115, the CSF1 receptor), that has phagocytic activity, and that does not contribute to the capillary endothelial layer in transplanted matrix. The enrichment of the adhesion molecule CD44, which in combination with PGE and Wnt-mediated signals may maintain slow cycling stem cell populations [68
], support the possibility that tumor-initiating cells in hemangiosarcomas share properties that have been ascribed to "cancer stem cells" in other tumors.
We conclude that one single lineage may give rise to both endothelial and hematopoietic progenitors, or alternatively, that multiple lineages contribute to blood vessel formation, including one originating from a restricted angioblastic progenitor that gives rise to the endothelial lining cells and one originating from a myeloid progenitor that is responsible for creating (but not lining) vascular channels. In this latter scenario, plasticity of adult hematopoietic and mesenchymal stem cells would be limited, differentiation of myeloid progenitors into endothelial-like cells would have to reflect functional rather than ontogenetic plasticity, and we should consider the possibility that canine hemangiosarcoma, and by extension, human angiosarcoma, might represent a subtype of myeloid sarcomas. This interpretation is supported by the general enrichment of genes overexpressed in dendritic cells and in leukemia, as well as by enrichment of the patterning gene HOXA10 and by the specific downregulation of PLZF in hemangiosarcomas; both of which are involved in hematopoietic differentiation [45
]. These results are especially significant in light of the poor response to treatments that presume canine hemangiosarcomas are tumors of blood vessels, and it may signal the need to revise the therapeutic approach to hemangiosarcoma and angiosarcoma as tumors of hematopoietic origin.
Preliminary data from our laboratories support the existence of rare progenitor cells in hemangiosarcoma that are responsible for propagating our cell lines in vitro
. There also is evidence for cancer stem cells that are capable of differentiating along different developmental paths to give rise to endothelial cells in chronic myelogenous leukemia and Burkitt lymphoma [70
]. Nevertheless, the possibility of vascular mimicry rather than true vascular differentiation cannot be excluded because the reverse outcome (endothelial tumors to hematopoietic cells) has not been documented.
To overcome the potential limitation from use of cell lines vs. intact tumors, we compared expression of a restricted, recurrently enriched signature from the hemangiosarcoma cell lines to whole tumor tissues. The data suggest that stromal and inflammatory cells can explain observed difference between these types of samples. Tumor cells modify the microenvironment and are themselves responsive to environmental cues. Nevertheless, to understand the contribution of the tumor cells to biological and pathological processes, it is important to examine the response in isolated cells. Microdissection of malignant cells from vascular tumors is difficult without retaining blood elements and normal angiogenic components that can be morphologically indistinguishable from the tumor cells. Conversely, cell lines provide a homogeneous, unlimited resource that can be extensively characterized with regard to ontogeny. The potential limitation of cell lines is further mitigated by the use of non-malignant controls to filter adaptation to ex vivo growth and by use of multiple samples. Finally, cell lines derived using our protocols retain the unique properties of the sample specimens and provide biologically relevant information.