The mechanisms that control osteosarcoma etiology and progression remain incompletely understood. This disease is predominantly seen in children and adolescents, and occurs naturally with greater frequency in dogs. Considering the similar clinical presentation, we postulated evolutionarily conserved molecular traits for this disease would be present in both species. Thus, the narrower genetic diversity of dogs would enhance our ability to define biologically and clinically significant traits.
Gene expression profiling allowed us to define two distinct molecular subgroups of canine osteosarcoma. Recent data showing that cells derived from Branch A tumors show more facile and aggressive growth in laboratory animals, including spontaneous metastasis after orthotopic invasion further support this pathological stratification [28
]. Furthermore, the clustering defined by this signature was seen repeatedly in three and five unrelated data sets from dogs and humans, respectively, although the human subsets showed more extensive branching into smaller subsets, suggesting osteosarcoma may have more complex behavior in humans. The apparent difference in tumor heterogeneity between humans and dogs could be explained by the reduced complexity of the canine genetic structure, where unlike humans, risk haplotypes are more firmly embedded in the defined populations we call breeds [29
]. Thus, the contribution of any allele to modulate the biological behavior of osteosarcoma is smaller in humans than it is in dogs, resulting in greater complexity of inherited risk factors that in turn manifest as increased inter-tumor and intra-tumor heterogeneity. Nonetheless, when we consider known differences between canine and human osteosarcoma, such as the age of disease onset and the palliative vs. curative treatment applied to these species, respectively, the similarities observed in their molecular signatures and associated biological behaviors are remarkable. Clearly, this signature was revealed in our analyses by eliminating stromal components through the use of tumor cells grown in culture, as well as by using canine samples, which reduced the intrinsic variation from heterogeneous genetic backgrounds of humans, in an unbiased (unsupervised) manner. Despite repeated attempts and the application of numerous algorithms, previous unsupervised analyses failed to segregate samples from intact tumor tissues (i.e., including tumor cells and stroma) from dogs or humans into meaningful groups. Thus, even though the gene expression signature was present in these intact tumor samples from both species, it was masked by stromal signatures can modulate the balance of expression for some of the genes [30
]. The significance of this restricted gene list is further underscored by its capacity to segregate independent cohorts into distinct branches, where each branch likely represents a molecular subtype with unique and potentially predictable biological behavior.
At this time, incomplete data annotation in the cohorts precludes definitive assessment of the prognostic value for these signatures, but the trends suggest that this signature or components thereof may be of prognostic significance both in humans and dogs, underscoring the similarities of this naturally occurring disease in both species. It is especially intriguing that the cell cycle component of this profile shows exquisite overlap with a predictive signature called CINSARC (C
omplexity Index in SARC
oma) recently defined for soft tissue sarcomas. Chibon et al showed that CINSARC predicted survival more robustly than the FNCLCC histologic grading system across multiple soft tissue sarcomas, gastrointestinal stromal tumors, breast carcinomas, and lymphomas [12
]. The CINSARC signature consists of 67 cell cycle genes, of which 40 are present as orthologs in Cluster 1 from our study, and an additional seven are members of families with retained homologs (e.g.
, three CENP family genes, CHEK1, PAK3, SGOL2, and SMC2). Most genes in this signature are coordinately regulated during the G2/M transition and/or as part of the DNA damage checkpoint. Our Gene Cluster 1 and CINSARC have many elements in common with prognostic signatures that also have been identified in urothelial cancer [31
] and melanomas with B-Raf mutations [32
]. Tumors with hyperactive DNA damage responses also possess greater chemoresistance [33
], possibly reflecting enrichment of cancer stem cells [34
]. This is not only a theoretical association derived from work in vitro, in silico, or in transgenic models; empirical data confirmed that induction of DNA damage responsive genes after radiation therapy, was associated with significantly worse survival outcomes in two independent cohorts of breast cancer patients [35
The reason why tumors with elevated cell cycle and DNA damage signatures show more aggressive progression than tumors with elevated expression of genes that modulate microenvironment interactions such as angiogenesis, cell migration, etc. is unclear. However, two non-mutually exclusive explanations are consistent with this observation. The first possibility is that the most aggressive tumors are likely to overcome constraints imposed by the tissue niche, while less aggressive tumors retain characteristics of their tissue of origin and are more dependent on the niche. In essence, the mutational profile may reflect the degree of tumor differentiation and select largely for cells that can survive and divide rapidly without much regard for their surrounding microenvironment. These cells might therefore show greater metastatic efficiency, simply measured by their ability to survive outside the original tumor [36
], and in this case, metastatic efficiency would be driven by cell-intrinsic factors and not by alterations in the microenvironment such as tissue hypoxia, pH, or the extracellular matrix. The second possibility is that the signature associated with elevated G2/M transition and DNA damage genes reflects enrichment of tumor-initiating cells, which have been documented in osteosarcoma [37
]. Either of these possibilities, or perhaps both acting in concert, might be responsible for the differential behavior and consequently the observed molecular pathological stratification of osteosarcoma patients into two groups with distinct outcome.
Additional work will be necessary to define the molecular mechanisms that underlie this gene expression signature and to further explore the role of genetic background in tumor susceptibility in both dogs and humans.