In this study, we aimed to compare gene expression profiles of osteosarcoma biopsies with cell lines and xenografts, in order to investigate whether these model systems are representative for the primary tumor. We have determined differential gene expression for different clinical parameters important in high-grade osteosarcoma on a dataset consisting of 76 conventional high-grade osteosarcoma samples. Importantly, pre-treatment biopsies were used instead of resected specimens, because pre-operative chemotherapy may cause tumor necrosis in responsive patients, thus altering gene expression and hampering the generation of high quality mRNA. We intended to generate a gene expression profile that could not only predict a specific clinical parameter in biopsies, but in osteosarcoma cell lines and xenografts as well. The metastasis/survival profile is described previously and may serve as a tool to predict prognosis and as a target for therapy. However, since most of the genes associated with osteosarcoma metastasis were macrophage associated, and no stroma or infiltrate is present in cell lines, this profile could not be applied to osteosarcoma cell lines. We therefore compared gene expression profiles of these different sample sets based on other clinical parameters. No significant differentially expressed genes were found between poor and good responders to chemotherapy. Several reports on genome-wide expression profiling in osteosarcoma have been published describing detection of differential expression between poor and good responders of pre-operative chemotherapy[26–29]. However, the cohorts used in these studies are all relatively small (n = 13-30), and, more importantly, the reported P-values were not corrected for multiple-testing in these studies. Remarkably, only two of the genes that were found to correlate with response to chemotherapy in these studies overlap, and one of these two genes was upregulated in poor responders in one study, whereas it was upregulated in good responders in the other study[26, 29]. Another report described differential expression between a metastatic and a non-metastatic cell line, for which metastatic capacity correlates with response to chemotherapy. In that particular study, four genes out of 252 were found to overlap with a patient study by Mintz et al. . However, the up- and downregulation of these four genes were not consistent between the two studies. We clearly show in a large cohort that there are no differences between these groups of patients, as the most significant probe had an adjP of 0.9998. These results are in line with our previous findings obtained by analyzing an osteosarcoma cohort on a different platform. The parameter 'histological subtypes' resulted in a considerable number of differentially expressed genes. Our prediction profile is not directly applicable to other platforms, but there is no real need to have a prediction profile for primary osteosarcoma histological subtype, since pathologists are very well able to assess this on an HE-section, even on a biopsy, with a concordance of 98% between histological subtype of biopsies and corresponding resections. Yet, we here show a quite important use of this profile, i.e. to determine the histological subtype of cell lines and xenografts. In vitro 2-dimensional growing cells lack extra cellular matrix formation, which is the characteristic feature to distinguish histological subtypes in high-grade central osteosarcoma.
The gene expression profiles as detected by analyzing osteosarcoma biopsy data show a large number of subtype-specific differentially expressed genes. In particular, fibroblastic osteosarcoma differed most from the two other main subtypes. Using gene set enrichment, we detected a network of genes upregulated in fibroblastic osteosarcoma, with a role in cellular growth and proliferation, and connection to the NF-κB pathway. This may be a readout of the high cellularity and low matrix composition of fibroblastic osteosarcoma in comparison with osteoblastic and chondroblastic osteosarcoma. GO term enrichment analysis confirmed these results. These pathways may explain why it is comparatively easy to culture fibroblastic osteosarcoma cells, which also may explain why the percentage of fibroblastic osteosarcoma is relatively high in our cell line dataset (31%, compared to 11% in the biopsy dataset). Next to this link to cellular growth and proliferation, the most significant network with fibroblastic-specific upregulated genes showed a connection to the immune system. GO analysis of the five most significant GO terms pointed to involvement of the immune system as well (GO term GO:0006955, P = 3.9E-3, see Additional File 3 for GO term subgraphs). Forty-four genes in this GO term were significant, of which 43 were upregulated in fibroblastic osteosarcoma. An elevated immune response might be the reason why patients with fibroblastic osteosarcoma tend to have better survival profiles, as a pro-inflammatory environment has an important role in osteosarcoma metastasis-free survival. This profile is different from the previously found macrophage-specific profile which was associated with better metastasis-free survival of osteosarcoma patients. The overrepresentation of pathways involved in chondrogenesis in the chondroblastic profile is in line with the high amount of chondroid matrix in this subtype. We only detected one osteoblastic-specific gene, UNQ1940, or FAM180A, with a yet unknown function. Since 50 osteoblastic osteosarcoma samples were compared with only 9 chondroblastic and 7 fibroblastic osteosarcoma samples, we suggest that fibroblastic and chondroblastic osteosarcoma have specific characteristics that distinguishes these tumors from osteoblastic osteosarcoma, and that the latter does not have such an extra feature in comparison with chondro- and fibroblastic osteosarcoma.
Our histological subtype prediction profile consists of 24 probes encoding for 22 genes, all with a specific score which depends on the significance of each gene. The genes that make up the chondroblastic-specific part of this expression profile are mostly chondroid matrix-associated genes, such as ACAN, COL2A1, and MATN4, and are all upregulated in chondroblastic osteosarcoma. Fibroblastic-specific genes that make up the prediction profile are up- or downregulated. An example of a gene upregulated in fibroblastic osteosarcoma is NFE2L3, a transcription factor which heterodimerizes with small musculoaponeurotic fibrosarcoma factors and for which a protective role was suggested in hematopoietic malignancies. DLX5, a transcription factor involved in bone formation, is downregulated in fibroblastic osteosarcoma, and reflects the lower amounts of matrix present in fibroblastic osteosarcoma. No known function is yet available for the two osteoblastic-specific genes. The misclassification error of the prediction profile in the training set of biopsies was 15%. Cell lines and xenografts were predicted with misclassification errors of 15% and of 11%, respectively. It seems most likely that these prediction errors are inherent to the error rate of the prediction profile, which is also 15%. Thus, because these misclassification errors are in the same range, we suggest that gene expression of these model systems is highly similar to gene expression of the tumor from which they are derived. This is especially of interest for studies in cell lines, since no stroma is present on which subtyping can be performed, but repeatedly passaged xenografts often lose stroma as well. Most genes of the prediction profile are matrix-associated genes. Even though these cell lines do not secrete any matrix, and xenografts have diminished amounts of matrix, we can still detect histological subtype markers on an mRNA level, and are able to distinguish different histological subtypes of cell lines and xenografts using this profile.