Although cancer is a heterogeneous disease, accumulating evidence supports the idea that signals for malignant transformation, progression and invasion are likely affected by a multitude of signaling pathways converging on several key regulators [29
]. Identifying these key regulators will be critical for the development of novel cancer therapeutics. Nonetheless, due to diverse genetic backgrounds of tumors and the intrinsic limitations of high-throughput profiling technologies, many of these key regulators remain hidden from conventional single-level or direct analyses. In this study, we used a systems biology approach to reveal and identify topologically significant nodes by integrating transcriptomic and glycoproteomic data of two human metastatic OS models.
Direct pathway analyses of the mRNA and glycoproteomic profiles revealed a low concordance of differentially expressed genes or proteins and their respective functional pathways between the models. These results may in parts reflect the unique methodologies used to develop the cell line models. The metastatic 143B subline was generated in vitro
via a Ki-RAS oncogene transformation of the HOS cell line [15
], while the LM7 subline was developed in vivo
through successive cycles of pulmonary metastases selection and re-injections in mice [14
]. The different genetic background of the parental cell lines and the in vitro
versus in vivo
development of the respective metastatic sublines, may account for the differences observed in the genomic and glycoproteomic analysis of the models. The discordance between the mRNA and proteomics profiles was not surprising because of their known limited correlation [7
], and the specific subproteome characterized in this study, namely the N-linked glycoproteins with affinity for WGA lectin. These fundamental differences and the large number of discordant pathways between the cell line models, increase the difficulty of prioritizing and identifying the key common regulators in the metastatic process of OS.
Despite of the different origins, the two OS cell line models share the same metastatic phenotype. Thus, we hypothesized that some key metastatic pathways are common to different tumors of the same cancer type, but remain hidden from the direct genomic and proteomic profiling analyses. Therefore, we performed further analysis and data integration to identify key nodes in the pathways that will reveal the critical players in common metastatic processes. For this analysis, we applied a recently developed “topological scoring” algorithm to integrate the genomic and glycoproteomic profiles and identified significant hidden common pathways between the two OS models. The “topological scoring” algorithm scores nodes in a network built from the experimentally derived, condition-specific genomic and proteomic profiles [29
]. The output provides a series of signaling proteins associated with the metastatic phenotype and helps delineate the underlying biological processes. One of the main advantages of the topological algorithm is that it assesses the relative contribution of every node in the condition-specific network under study relative to its role in the global network. Thus, the “hubs” that are commonly present in different pathways and have high connectivity in the global network, are penalized if they do not have any special role related to the experimentally derived set of differentially expressed genes or proteins. On the other hand, nodes that provide significant connectivity among the differentially expressed genes or proteins are highly scored regardless of their global network interactions, rendering the results more specific.
Pathway analysis of the identified topological nodes revealed many common and significant dysregulated pathways between the models, which were missed by the direct analyses. Remarkably, the analysis revealed pathways such as “Cytoskeleton remodeling/TGF/WNT” and “Cell adhesion/chemokines and adhesion”, which were common to both models and present at both mRNA and protein levels. All the processes represented in these common pathways have been previously associated with metastasis in various cancer types, and changes involved in these signaling networks associated to cytoskeleton remodeling, cell-cell and cell-matrix adhesion are critical for the cancer cell migration and invasion [4
]. To determine if these pathways would be identified by a random chance or nonspecifically, we performed a false discovery analysis using randomly selected genes and confirmed the specificity of these pathways in our datasets.
Our results support the notion that there are common metastatic mechanisms acting downstream of different genetic aberrations or origins. This implies the existence of common molecular therapeutic targets and disease biomarkers irrespective of the driver genetic abnormalities in the tumors. This finding is particularly significant in tumors, such as OS, in which the genomic abnormalities are known to be highly complex and it is hard to determine the driver mutations from the passengers [37
]. For instance, the most significantly overrepresented pathway of the topological analysis was the “Cytoskeleton remodeling/TGF/WNT”, which includes the interesting Wnt component. The Wnt signaling is particularly important for cancer cells as they are able to disrupt this pathway in different ways resulting in tumorigenesis and metastasis. Activation of the Wnt signaling pathway is necessary for the commitment of mesenchymal stem cells to the osteoblast lineage. In addition, aberrant Wnt signaling activity has been reported in a variety of human cancers including soft tissue sarcomas and human OS primary tissues and cell lines [36
]. Abnormal activation of the canonical Wnt pathway results in stabilized β-catenin that translocates into the nucleus. Subsequently, it binds to transcription factors and drives the uncontrolled expression of target genes implicated in cell proliferation, transformation, and tumor progression, such as Myc, matrix metalloproteinase 7, Axin-2, the cell adhesion molecule L1-CAM, the metastasis gene S100A4, and others [33
]. Therefore, targeting the Wnt signaling may have a therapeutic effect on different types of metastatic OS.
Similarly, our findings suggest that Wnt signaling is highly relevant to the metastasis of OS. Previous studies have shown that OS harbors an accumulation of β-catenin either in the cytoplasm or in the nucleus [45
], a hallmark of Wnt signaling activation. Additionally, the Wnt coreceptor LRP5 expression in OS tissue samples correlated with metastasis and a lower rate of disease-free survival in patients [46
]. Furthermore, OS cell lines have been reported to express many Wnt ligands and receptors, whereas secreted Wnt antagonists including secreted frizzled-related protein (sFRP) and Dickkopf (Dkk) families are commonly absent [46
]. Inhibition of similar mechanisms by the reintroduction of secreted Wnt antagonists in OS, such as the Wnt inhibitory factor (WIF-1), has been proposed for downregulation of Wnt signaling as a novel therapeutic approach. For instance, overexpression of WIF-1 significantly decreased tumor growth and markedly reduced the number of lung metastasis of 143B cells in vivo
]. In other studies soluble LRP5 (sLRP5), which blocks Wnt signaling, was able to reduce in vitro
cellular invasion, and transfection of sLRP5 in SaOS-2 caused a marked up-regulation of E-cadherin and down-regulation of N-cadherin suggesting a reversal of epithelial-mesenchymal transition [40
]. These findings suggest an important role for aberrant Wnt signaling in the pathobiology and progression of OS. The Wnt pathway may represent a promising source of novel therapeutic targets and disease progression biomarkers.
Another common pathway relevant in metastasis is the “Cell adhesion/chemokines and adhesion”, which was identified at both the genomic and proteomic levels. Chemokine ligands and their cognate receptors have been extensively implicated in the progression and metastasis of multiple tumors such as melanoma, breast cancer, prostate cancer, and others [49
]. In OS, research has revealed a complex interaction between chemokine ligand/receptor axis, and their role in tumor invasion, metastasis and patient prognosis [54
]. Our group previously reported that expression of two CXC chemokines were elevated in tumor and plasma of pediatric OS patients, and their levels correlated with patient outcomes [55
]. Several of the CXC chemokines were also identified by the topological analysis in this study, including CXCL12 which has been directly associated to metastasis in OS and other tumors [56
]. Identification of the significant topological nodes within such complex networks could help identify the key players and delineate their roles in cell adhesion, invasion and other important metastasis functions.
In addition to the identification of therapeutic targets, identification of tumor-derived biomarkers for early detection of disease progression and metastasis is a critical component for personalized medicine. As previously noted, characterization of key hidden nodes could facilitate the identification of candidate biomarkers for metastatic OS. Because of the diversity of the human proteome and limitations of the current proteomic methods, targeting specific subproteomes that are likely to be secreted into the blood stream, such as the glycoproteome, will improve the likelihood of identifying tumor-derived circulating biomarkers. In this study, we characterized the N-linked glycoproteome due to their involvement in metastatic OS as evidenced by the glycogene analysis, and their frequent localization in the cellular membrane and extracellular space. In spite of the limited number of glycoproteins identified in this study, the topological analysis revealed common significant pathways between the differentially regulated glycoproteins and genes in both OS models. Using this analysis, we can prioritize the up-regulated glycoproteins and up-regulated genes as well as the hidden molecules that may serve as biomarkers for the disease or metastasis in the future validation, such as proteins in the TGF beta (i.e. TGFB1, TGFBR1, TGFBR2) and MAP kinase (MAPK1, MAPK3, MAP3K1,MAP3K8) families identified by topological analysis.
Despite the encouraging results, we recognize that there are limitations in the current study. For instance, the cell line models used in our study represents a significant limitation. Although the in vitro
models provide a convenient and renewable platform that cannot be surpassed by clinical specimens, it may not faithfully reflect the behavior of tumor cells in vivo
. Therefore, future directions include the development of orthotopic xenograft mouse models to recapitulate and validate the results obtained in this study. In addition, the characterized N-linked glycoproteins with an affinity for the WGA lectin represent a restricted number of proteins, which is typical to this type of subproteome analysis [57
]. This limits the number of identifiable up-regulated glycoproteins amenable for validation as candidate biomarkers. Further studies with a wider range of proteins including additional glycoproteins by different capture methods (i.e. cell surface biotin labeling, biocytin hydrazide N-linked enrichment) [59
] and other relevant subproteomes, such as the secretome [61
] will help to alleviate this limitation.