Ultrasonography of Osteosarcoma in Dogs
Sonography Atlas of Dogs

Abstract

Although the small population size in the present study is unlikely to show significant differences, it is important to identify any possible confounding factors that may influence survival/prognosis and should be considered when interpreting the expression data at biological levels. In both survival groups, dogs were subjected to a variety of single agent therapies, including lobaplatin, doxorubicin or carboplatin; or a combination of carboplatin and doxorubicin. Some were treated with postoperative chemotherapy and some not, and Kaplan-Meier analysis showed that chemotherapy did not significantly prolong survival in the overall population (log rank score 1.353, P value 0.245), neither among short survivors (log rank score 0.460, P value of 0.498) nor among long survivors (log rank score 0.033, P value 0.857) separately (Figure 2). Although these survival curves only represent bivariable analyses (treated vs. not treated) that do not include all confounding variables, a separate univariable and subsequent multivariable statistical model with inclusion of variables such as histological grade, alkaline phosphatase, neuter status, gender and age upon chemotherapy stratification revealed that none of the variables appeared to have influenced survival significantly (Additional file 2). An additional important insight from the long survival group is that all 4 dogs that were censored from the study (death due to other causes and no apparent metastatic disease until death) received postoperative chemotherapy, which suggests possibilities that these dogs had responded reasonably well to the chemotherapeutic regime. Finally, since both survival groups consist of comparable heterogeneous populations of dogs and poor survivors tend to have higher ALP measurements, the arbitrary but defendable approach to a binary distinction of 'good' and 'poor' outcome based on 6 month survival time is supported for further gene expression profiling.

Microarray data analysis

Comparison of short and long survivors using two-class unpaired Significance Analysis of Microarrays (SAM) revealed fifty-one genes that were differentially regulated at a false discovery rate (FDR) of 10%. All these genes were found to be upregulated in the SS group where 37 of them were upregulated with a fold change of more than 1.4 (Table 3). Of the 37 genes, 8 were not functionally annotated for the dog and 5 others were unknown Expressed Sequence Tags (ESTs). To further visualize the gene expression data, hierarchical clustering was performed on all 32 arrays based on the differentially expressed genes (Figure 3). A gene tree dendrogram revealed two distinct clusters, where Cluster 2 appears to distinguish long from short survivors or, in other words, this signature defines prognosis based on the 6 month survival binary outcome. Within Cluster 1, particular attention should also be given to a further separation into 2 different subgroups (A and B). Although Fisher's exact test revealed no significant differences among the variables assessed with respect to histological grade, sex, neuter status and postoperative chemotherapy (Additional file 3), this could primarily be due to the low number of subjects within each subgroup (n = 8) which most likely will not allow statistical significance. We did find a significant difference in terms of overall survival time between the two subgroups, with subgroup A: having a shorter survival time than subgroup B. The log rank score was 5.82, with a corresponding P value of 0.0158 (Figure 4).

Pathway analyses

The number of annotated genes in the present study is too low to form any relevant protein networks. Therefore, we used the 'Shortest path analyses' by MetaCore, which allows for one intermediate gene to be linked to the differential genes in order to form a more informative gene-protein network. These intermediate genes may be a transcription factor, receptor or ligand. Transcripts with limited information available on network or associated with a pathway were excluded, leaving a final gene regulatory network generated based on 11 genes that were overexpressed in short survivors of canine OS with the addition of seven transcription factors, one receptor and a ligand (Figure 5). Transcription factors p53, c-myc, SP1 and NF-kB seemed to play roles as 'central hubs' connecting these transcripts. All three heat shock proteins (HSP90-alpha, HSP60 and MOT/HSAP9/GRP75) appeared to be linked to a common transcription factor c-myc.

Because the interaction among genes is definitely more complex, we performed a PANTHER^® pathway analysis in which functional and pathway data available on human transcripts were used as reference to analyze the transcripts from the present study in the dog. PANTHER^® analysis on the 51 differentially expressed genes by SAM yields too few genes associated with a cellular signaling pathway, undermining the power of pathway analysis. This is mainly because a majority of genes are not associated with a signaling pathway, but also because the dog lacks functional annotation at present. Therefore, we excluded the FDR correction for multiple testing and selected a longer list of genes with a threshold of P < 0.05 from EDGE (a microarray analyzing software). After removal of non-annotated genes and transcript replicates, there were 533 genes mapped on PANTHER^® which permitted further analysis. Identified pathways included the top four commonly associated with cancer and OS pathogenesis: Wnt signaling, inflammation mediated by chemokine and cytokine signaling, integrin signaling, and ubiquitin proteasome (Table 5). Our pathway list was then compared with three microarray datasets of human osteosarcoma [24, 25]. Two gene lists were obtained from Srivastava A., et al., 2006: one of genes overexpressed in human osteogenic sarcomas in comparison to normal bones, the other a partial list of genes upregulated in a highly metastatic human OS cell line, and a gene list from Rajkumar T., et al., 2008 identifying genes that were differentially expressed in a doxorubicin drug resistant human OS cell line. These gene lists, obtained either from the authors themselves or via the supplementary tables provided in their articles, were subsequently subjected to PANTHER^® pathway analysis as similarly conducted in the present study to define pathways pertaining to drug resistance/metastasis/tumorigenesis of OS. Comparison of the top 20 pathway hits from these 3 studies in human OS (Table 5) revealed seven pathways overlapping with the present study on canine OS, including Wnt, chemokine/cytokine, Alzheimer disease-presenilin pathway, fibroblast growth factor (FGF), platelet derived growth factor (PDGF), apoptosis and interleukin signaling pathways.

Discussion

Canine OS gene expression profiling as a model for human OS

Osteosarcoma (OS) is a devastating disease in both human and dogs. The clinical presentation, characteristics and disease progression is similar in people and dogs except for the age of clinical onset, where 75% of human cases affect young adolescents while onset is predominantly observed in middle-aged to older dogs. For dogs, the decision to initiate treatment is often made by the owners. The standard therapy for canine OS primarily involves amputation or limb sparing surgery, followed by adjuvant chemotherapy. Some dogs tend to develop metastases within 4 months regardless of the therapy modality, while others survive for longer periods of time [17]. Similarly, despite the advance standard of care for children with OS, only 60% reach a 5-year disease free interval and 20% will not survive beyond 5 years [26].

The similarities seen in OS disease progression and survival rates in humans and dogs provides a reasonable justification to compare gene expression profiles based on clinical outcomes once these are adjusted for comparative lifetimes. The similarities in the genetic expression and biological behavior of canine and human OS also makes the dog a suitable model to study this disease [27, 28]. Gene expression profiling of spontaneous tumors in the dog offers a unique translational opportunity to identify prognostic biomarkers and signaling pathways common to both species. This is further supported by a report on the strong similarity in the gene expression profile found in both canine and human pediatric OS, suggesting that specific genes and pathways are commonly involved in these two species (Paoloni, M. et al., 2005, personal communication).

Further, various clinical and pathological methods are being used as prognostic indicators for canine OS, but to our knowledge, this is the first research conducted using canine specific cDNA microarray to analyze gene expression profiles associated with survival in a panel of thirty-two primary canine OS. The ability to identify gene markers of tumor aggressiveness at the time of primary tumor removal would provide prognostic values for a tailored therapy which can be translated to human medicine.

Clinicopathological relevance for OS prognostic gene expression studies in dogs

Gene expression profiling on the basis of segregating and then identifying poor and favorable outcome markers is not novel, but this is the first study conducted in dogs with OS. Selection of appropriate tumor candidates based on clinicopathological relevance for gene profiling is crucial for the overall experimental design. Although advanced histological grade and postoperative chemotherapy are known confounders to influence survival and prognosis in dogs with OS [29], these factors lack to significantly influence survival in the present study population, mainly due to the small population size. Elevation of serum alkaline phosphatase is a known negative prognosticator in dogs with OS, which is true for the present study population where poor survivors had significantly higher alkaline phosphatase levels than long survivors. Some dogs, despite receiving postoperative chemotherapy, did not live beyond 6 months and can be characterized as 'poor responders' to chemotherapy. On the other hand, the majority of dogs that received chemotherapy within the long survivors was censored which suggests that either these tumors responded to the chemotherapy or the primary tumor did not harbor the aggressive phenotype. These data further justify distinguishing between good and poor outcome: dogs can be assessed at the gene-biological level of primary tumors regardless of post-operative chemotherapy. In addition, the current gene profiling was carried out on tumors prior to chemotherapy, where 'clues' at gene expression levels will help to determine which animals will probably live longer regardless of either the type of regime or decision to include postoperative chemotherapy (Figure 6). To further improve the efficacy of therapy, it is necessary to identify dogs with OS with an increased risk of treatment failure, as well as to identify those that might not need aggressive chemotherapeutic protocols.

To address issues concerning applicability for tailored therapies, microarray technologies have identified new molecular subclassifications in various human tumors, but to date there are no such reports for prognostic signatures in human and canine OS. The current research thus provides new insights into possible prognostic molecular subclassifications in canine OS. Hierarchical clustering of the differentially expressed genes revealed that long survivors have a similar gene expression profile, which is distinctly different from that of short survivors: these genes appear to be consistently downregulated among the long survivors. In contrast, clustering the genes based on expression was found to separate the short survivors into 2 subgroups. Cluster 1 genes were found to be overexpressed in subgroup B in comparison to the rest of the tumors, making this subgroup unique among the 32 tumors. Intriguingly, subgroup A appeared to have a significantly decreased survival compared to subgroup B where it shares common downregulation of genes with cluster 1. The mechanism behind this evident contradiction can only be speculated. Possibly, upregulation of both cluster 1 and cluster 2 genes provides a somewhat protective mechanism that improves survival as compared to the combination seen for subgroup A. On the other hand, the data can also be interpreted such that Cluster 1 genes are upregulated in 25% of canine OS, where its overexpression is related to poor prognosis. Cluster 2 genes appear to be true prognostic gene signature, defining the gene expression that segregates short (<6 months) from long survivors. The importance of these findings is they provide insights into possible existence of molecular subclassifications among short survivors that may be regulated by different signaling pathways whereby, due to the complex interaction among genes, these subgroups may respond differently to certain targeted therapies. These interpretations must be addressed cautiously given the small number of samples and differentially expressed genes selected for the generation of the present dendrogram. As has been noted previously, "a replicable classification is not necessarily a useful one; but a useful one that characterizes some aspect of the population must be replicable" [30]. Therefore, insights into molecular sub-classifications among the canine OS from the present study require further validation on a larger and independent set of tumors.

Differentially expressed genes between short and long survivors: similarities with human OS

For a more biologically meaningful approach to data analysis, we looked at the level of common biological processes that differentiates short from long survivors. Genes highly expressed among short survivors of canine OS are likely to contribute to the aggressive nature of the disease in terms of increased cell cycle and proliferation, drug resistance and metastasis-associated properties. Among the transcripts that were differentially expressed between the two survival groups that have been implicated in human OS were HSP70/MOT, HSP60 and NCOR1, while HMGB1, HSP90 and MGST1 have been reported for both human and canine OS. Following discussion will focus on the relevance of these genes to human and canine OS pathogenesis and disease progression.

The main goal for discovering new therapeutic interventions for OS is to inhibit metastasis and eradicate micrometastases. The difference in growth patterns of metastases may be detected by differences in gene expression of primary tumors at the point of amputation. Due to their gene expression differences, some dogs develop metastasis rapidly and others have delayed metastatic disease, which contributes significantly to their survival. In the overall clinical data comparison in this study, there were long survivors who did not develop metastases at all. We did not exclude these tumors from the analysis because they appear to be true survivors, where there was no metastatic disease detected after primary tumor removal. Among the differentially expressed genes between short and long survivors, CFL2, MOT and V-ATPase C have been associated with invasion and metastasis in various human tumors [31–33]. The metastatic potential of the stated genes in OS has not been reported previously; therefore additional research into their exact functions in OS metastasis is required.

Over the past decade, besides metastasis, there has been growing interest in the development of multidrug resistance in OS, where various genes have been reported to contribute to the drug resistance phenotype [24, 34, 35]. The two drug resistance related genes overexpressed in poor survivors found in the present research in canine OS include HMGB1 and MGST1. HMGB1 expression levels have been previously investigated in 5 canine OS where its expression was suggested to be a potential marker for cisplatin therapy based clinical outcome [36]. This gene may play a role in protecting OS cells, making them less susceptible to the cisplatin where it has been similarly associated with activation of p53 [37, 38] and drug resistance in other human tumors [37, 39]. Similarly, MGST1 was found to be overexpressed in several human malignant tissues where it was shown to protect cells from several cytotoxic drugs [40, 41] as well as by direct detoxification and downstream protection of tumor cells from oxidative stress [40]. Preliminary evidence suggests overexpression of another family member, glutathione S-transferase π, is significantly related to poor histological response to preoperative chemotherapy and poorer prognosis in human OS [41]. Similarly, an in vitro model of canine OS (COS31) with high glutathione-S-transferase activity was found to be resistant to cytotoxic effects of cisplatin [42]. Drug resistance issue is of interest in the present study population because despite given postoperative chemotherapy there were dogs unable to extend survival beyond 6 months while others received no chemotherapy and yet lived much longer. This suggests that screening through gene profiles can help with therapeutic decisions by segregating dogs based on key targets for personalized chemotherapy, and hence ensure treatment success.

Genes associated with proliferation, cell cycle and differentiation revealed at present have not been described for OS pathogenesis except for Mortalin (MOT/HSP70), HSP90 and HSP60. MOT/HSP70 has been previously shown to form complexes with p53 tumor suppressor, causing inactivation of the wild type p53 leading to excessive proliferation [43]. Knockdown of MOT/HSP70 by shRNA expression plasmids have proven to cause growth arrest in human OS [44]. Since MOT has been found upregulated in short survivors in this study and since it is known that p53 mutations are common feature of poor prognosis in canine and human OS, extensive in depth investigation should be done to elucidate both the role of MOT/HSP70 as a prognostic marker and its association with p53 in OS. The other two heat shock protein family members revealed by the present study include HSP90 and HSP60, both having been strongly implicated in multiple stages of tumorigenesis from proliferation to impaired apoptosis and angiogenesis, invasion and metastasis. High expressions of HSP90-beta have been reported in 3 human OS cell lines as well as in 3 primary tumors from microarray analysis in previous research [20]. In primary human OS removed at surgery, HSP90α was found to be 40% overexpressed and HSP60 60% [45]. In another, expression of HSP60 was detected in 83% of human OS biopsy specimens, and 43% of these OS patients had increased levels of anti-hsp60 antibodies in their serum [46]. Recently, in vitro studies have demonstrated that inhibition of HSP90 exhibits selective cytotoxicity in canine OS cells by downregulation of Met, Akt and p-STAT3, key players for multiple oncogenic signaling pathways [47]. A growing body of evidence suggests that heat shock proteins provide protection to a large group of client proteins: they help protect highly aggressive and metastatic cells from cellular stressors in the tumor microenvironment. The overexpression of all these HSP family members in poor survivors of canine OS makes them interesting in the context of drug development for the next generation of HSP inhibitors.

Another candidate gene of interest to both human and canine OS is NCOR1. We found that NCOR1 was overexpressed in 8 out of 32 (25%) of tumors, all from subgroup B (among the candidates from gene Cluster 1) and there was a clear downregulation among all long survivors. In human OS, NCOR1 was amplified in 22.6% of tumors where it did not correlate with disease free interval, and amplification status correlated significantly with tumor size [48]. The mechanistic role of NCOR1 as an independent negative prognosticator in both human and canine OS is currently unknown, although the overexpression of NCOR1 in 25% of canine tumors could be related to its possible amplification status, similar to what is seen for human OS. Here, we have confirmed the overexpression of MGST1 & NCOR1 in poor survivors as has been previously described for both canine and human OS, and identified two new candidates, ANKRD17 and MRPS31, choosing to validate expression data using quantitative real-time PCR. Those dogs that lived less than 6 months were found to highly express these genes in parallel to the microarray expression data, which suggests these transcripts could be attractive therapeutic targets and markers for future prognosis and disease progression for canine OS.

Share this article / Teilen Sie diesen Artikel