The relevance of naturally occurring canine tumors to improve our understanding of cancer biology and genetics has been increasingly recognized in recent years 
. Canine tumors can be utilized as a system to understand how genetic background can influence the susceptibility of an individual to non-inherited cancers. Due to the homogeneity among dog breeds, we can study frequently occurring cancers within groups in a way that would be difficult within the genetically diverse human population or in laboratory animals, where most tumors are induced chemically or by genetic manipulation.
We studied naturally occurring canine hemangiosarcoma to test the hypothesis that patterns of gene expression could outline biological differences between tumor cells originating from dogs of a distinct breed that have a higher lifetime risk for hemangiosarcoma. Hemangiosarcoma is ontogenetically related to human angiosarcoma and Kaposi sarcoma, as all three are presumed to arise from hemangioblastic or endothelial progenitors and they share signaling abnormalities 
. The highly metastatic behavior and modest response to chemotherapy distinguish canine hemangiosarcoma and human angiosarcoma from other common soft tissue sarcomas that are locally invasive and generally unresponsive to chemotherapy. We uncovered a set of hemangiosarcoma-associated genes peculiar to a single dog breed suggesting these are modulated by (or with) heritable traits that may influence risk for this cancer.
We considered carefully the choice of low passage cell lines vs. intact tumors for these experiments. Tumors are in essence tissues 
. 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 be able to examine the response on isolated cells. One approach to do this is microdissection, but in a vascular tumor, it is difficult to microdissect malignant tissue without retaining normal angiogenic components, which are morphologically indistinguishable in many cases, and blood elements. On the other hand, cell lines provide a homogeneous, unlimited resource that can be extensively characterized with regard to ontogeny. The potential limitations of cell lines such as their restricted origin, possible in vitro
evolution or drift, and adaptation for growth in culture, can be mitigated by use of controls that replicate culture conditions so that adaptation to ex vivo
growth is filtered from responsive transcript lists, and by use of more than one sample. Our results show that despite the different origin, isolation, and establishment of the cell lines we used for these experiments, hemangiosarcomas retained unique characteristics that distinguished them from other cultured (or primary) cells, and that the recurrent finding of genes that are over- or under-expressed in the samples is significant and represents differences that can be traced to the developmental process of the sample (ontogeny or pathological progression), rather than to selection in culture. Ongoing experiments are designed to define the correlation of these findings in intact tumor samples where extracellular matrix associations are maintained.
Among genes whose expression differed between Golden Retrievers and non-Golden Retrievers, a disproportionately high number of genes encode transcription factors. This suggests that transcriptional regulation might play a key role in disease susceptibility and progression. Upregulation of SMARCA1 in Golden Retrievers with hemangiosarcoma was intriguing since changes in expression of a single transcriptional regulator can create genome-wide disruption of a variety of genes, possibly resulting in faster progression of the disease. It is thus feasible that deregulation of SMARCA1 potentiates susceptibility and/or heritability of hemangiosarcoma in Golden Retrievers. The downregulation of MHC class I genes in hemangiosarcoma from Golden Retrievers added a level of confidence, as these genes represent the likely targets to define individual or breed-specific differences. Preliminary assessment of MHC class I expression by flow cytometry generally support the gene expression data, with Frog (Golden Retriever) cells having no detectable MHC class I, and Dal-4 (non-Golden Retriever) cells expressing MHC class I molecules. This pattern is rather unique to hemangiosarcoma, as normal blood leukocytes and other tumors from Golden Retrievers (for example, leukemias) show robust expression of MHC class I. The organization and control of genes in the canine MHC class I locus remains poorly understood, and our data will undoubtedly spur further study of how genetic variants within breed and transforming factors might influence MHC class I expression. In fact, breed-related polymorphisms or changes in expression level have not been identified in normal canine somatic cells; thus, downregulation of MHC class I genes (at least MHC DLA-88 and DLA-64) in hemangiosarcoma cells from Golden Retrievers might reflect selective pressure to evade immune responses, or perhaps a response to autocrine or paracrine factors such as interferons or other inflammatory mediators. This illustrates the potential benefit of studies in dogs where a suitable experimental design could help distinguish whether T-cell-mediated therapies that elicit productive responses in non-Golden Retrievers might be less successful in Golden Retrievers 
, and similarly whether tumors of Golden Retrievers provide suitable targets for natural killer cell-mediated immunotherapy.
The specificity of these findings to one breed and one disease were further illustrated when we compared Golden Retrievers with hemangiosarcoma to Golden Retrievers with osteosarcoma and non-Hodgkin lymphoma. In this case, we found acid ceramidase was overexpressed in hemangiosarcomas, but not osteosarcoma or non-Hodgkin lymphoma. Acid ceramidase belongs to a family of anti-apoptotic genes that promote ceramide production. At least one inhibitor of acid ceramidases, B13, increased ceramide content selectively in tumor cells, inducing apoptosis 
, suggesting acid ceramidase inhibitors may hold therapeutic potential. It is thus possible that overexpression of this gene is a consequence of interaction among factors that underlie the observed predisposition of Golden Retrievers to hemangiosarcoma.
Another gene that was underexpressed in Golden Retrievers with hemangiosarcoma compared to non-Golden Retrievers is TSP-3, a member of the Thrombospondin family. A different member of this family, TSP-1, has potent anti-angiogenic activity 
and has been a template for mimetics designed to treat cancer 
. Two of these mimetics, ABT-510 and ABT-526, have yielded promising results in pet dogs with a variety of tumors, albeit they were ineffective in dogs with hemangiosarcoma 
. TSP-3 and TSP-1 are both calcium-binding proteins, but the physiological role of TSP-3 is unknown 
. The downregulation of TSP-3 should be explored further in light of these clinical results.
Despite these differences, the precise cause for increased risk to develop hemangiosarcoma in Golden Retrievers remains unclear. At least part of this perceived “risk” may be due to more rapid disease progression. In other words, it is possible that transformation of hemangiosarcoma-initiating cells does not occur with significantly greater frequency in Golden Retriever, but once it occurs, progression to clinical disease is faster, thus leading to a higher frequency of hemangiosarcoma diagnoses in Golden Retriever. An interesting correlation along these lines was the enrichment of VEGFR1 in tumors from Golden Retrievers, which generally seemed to occur at the expense of VEGFR2. It is important to note that the enrichment of VEGFR1 in tumors from Golden Retrievers was not absolute, but rather occurred in concert with various other genes that were preferentially expressed in a coordinated fashion in these cells. We tested the possibility that the “Golden Retriever background” might create a phenotype that was responsive to VEGFR1. It seemed reasonable to assume that growth of hemangiosarcoma cells, which are presumed to be of endothelial origin, was driven by VEGF. In fact, hemangiosarcoma cells make their own VEGF 
, resulting in systemic elevation of this cytokine in affected dogs 
. The prevailing dogma states that VEGFR2 activates biochemical cascades that result in proliferation and prevent programmed cell death 
, whereas the action of VEGFR1 is less clear. VEGFR1 may transmit bona fide
growth signals 
, or it may oppose VEGFR2 signals directly or act as a decoy receptor 
. In some cases, VEGFR1 may even promote tumor growth and metastasis 
. Our data reveal two important points. The first is that inhibition of VEGFR2 has little if any effect on proliferation of canine hemangiosarcoma cells in culture. While this may seem surprising, it is consistent with previous results in other hemangiosarcoma cell lines 
and suggests the VEGFR2 pathway may be an ontogenic relic in these cells. That is, VEGF production and VEGFR2 expression may remain as part of the differentiation program, but the cells are not “addicted” to, or rely on, growth and survival signals transmitted through this prototypical VEGF receptor. Instead, hemangiosarcoma cells rely on other pathways for growth and survival. The second is that, at least in hemangiosarcoma cells from Golden Retrievers that express VEGFR1, this receptor may be more than simply a “decoy”, and instead, signals transmitted by VEGFR1 may dampen proliferation and/or differentiation.
These observations also are consistent with our findings that, unlike what is seen in some sporadic vascular tumors in humans, mutations of VHL are absent or infrequent in hemangiosarcoma, suggesting this disease entity may represent a distinct or specialized subset of blood vessel forming cells. Yoder et al 
recently described a myeloid cell that is a major participant in blood vessel formation. This cell is a “vascular mimic” that can express a variety of cell surface proteins associated with endothelial precursor cells (CD133, CD34, VEGFR2), but it also has proteins that belie hematopoietic origin (CD45, CD14, CD115), has phagocytic activity, and does not contribute to the capillary endothelial layer in transplanted matrix. These findings suggest that plasticity of adult hematopoietic and mesenchymal stem cells is limited, and differentiation of myeloid progenitors into endothelial cells reflects functional rather than ontogenetic plasticity, raising the possibility that canine hemangiosarcoma is in fact a myeloid sarcoma. In this context, the inhibitory effects of VEGFR1 would be predictable, as they mirror functions of this receptor as an inhibitor of differentiation in human and murine dendritic cells. It is worth noting that enrichment for VEGFR1 and other genes may be causally related to the incidence and biological behavior of hemangiosarcoma in Golden Retrievers, but it just as likely could be an effect of other risk factors in the breed that are upstream regulators of these pathways, as our data do not distinguish between these possibilities. Nevertheless, we interpret the reproducibility of the results as an indicator that these are not simply epiphenomena.
In conclusion, our data show that gene expression profiles are informative to identify differences in tumor progression that may be influenced by heritable factors. As important, our results indicate these differences must be interpreted carefully and in the context of biological pathways. Specifically, gene expression profiling suggests that inflammation and angiogenesis are two general processes that may be sensitive to modulation by a dog's genetic background in hemangiosarcoma. Inflammation, defined by enrichment of cytokines such as IL8, IL5, IL18, and several molecules that mediate adhesion and cell-cell interactions, might reflect the action of a single aberrantly regulated molecule (for example, IL1). Angiogenesis, defined by preferential enrichment of VEGFR1 in tumor cells from Golden Retrievers might reflect engagement of unique growth (inhibitory) pathways. However, some of these differences also might reflect the ontogeny of the cells, so we must consider the possibility that the cell of origin in hemangiosarcoma retains moderate or extensive plasticity and the heritable influence is manifested based on the stage of differentiation achieved by the tumor cells. We should bear in mind, then, that part of the “susceptibility” for this disease in Golden Retrievers could be due to different biological behavior in the early stages of the disease, and also to different sensitivity of intrinsic tumor surveillance and/or chemotherapy. That is to say, upregulation of VEGFR1, downregulation of MHC class I, and downregulation of TSP-3 may underscore important differences that explain susceptibility, pathogenesis, and response to therapy. An alternative interpretation is that, regardless of the ontogeny of the tumor-initiating cell, the transformation events responsible for hemangiosarcoma involve pathways that render VEGF signals mostly inconsequential and other pathways controlled at the level of transcriptional regulation (e.g., by SMARCA1) and/or survival (e.g., acid ceramidase) are important determinants of the breed-dependent phenotype. Overall, this study emphasizes potential benefits of gene expression analysis and bioinformatics to study sporadic disease, and highlights the unique contribution that studies of naturally occurring cancer in man's best friend can make into disease susceptibility, heritability and progression.