Analysis of Transcriptomic Similarity between Osteosarcoma Cell Lines and Primary Tumors

Abstract

<b><i>Background:</i></b> Osteosarcoma (OS) cell lines are commonly used to mimic tumors for in vitro experiments. The present study explores the resemblance of OS cell lines to OS primary tumors in regard to gene expression. <b><i>Methods:</i></b> Transcriptomic data were retrieved from published data sets for 18 primary tumor samples and 13 commonly used OS cell lines. Tumor purity was accounted for when correlating tumor and cell line gene expression. Differentially expressed genes between tumors and cell lines were discovered and gene ontology analysis was performed. <b><i>Results:</i></b> Certain commonly used cell lines, including NY, NOS1, and U2OS, display less resemblance to OS tumors than do other cell lines. For genes overexpressed in tumors, and consequently underexpressed in cell lines, gene ontology analysis enriched pathways related to cell-cell adhesion and stimulus detection. <b><i>Conclusion:</i></b> The pathways dysregulated between cell lines and tumors have been implicated in OS pathogenesis. Therefore, the findings suggest that the transcriptome of OS cell lines may not be completely representative of OS primary tumors’ gene expression and the disease process.

Systems approach for congruence and selection of cancer models towards precision medicine

“…However, existing evaluation tools mostly belong to two major categories, congruence (correlation-based) analysis and authentication (machine-learning-based) analysis, and do not sufficiently serve the purpose of identifying appropriate models for precision medicine. In congruence analysis, correlation/association measures are usually applied to quantify similarity of a cancer model to the target tumor cohort in a genome-wide scale [ 5 , 10 , 11 , 14 , 15 , 18 ]. In contrast, authentication analysis develops machine learning models, such as suitability score [ 16 ], random forest, ridge regression and nearest template prediction, for accurate assignment of cancer models to human cancer types.…”

“…Many cancer models may be potentially mis-annotated from their origins or the quality of congruence may vary or decay over time [4][5][6]. Due to increasing availability of new cancer models and associated comprehensive omics data, evaluation and comparison of cancer models with human tumors using transcriptomic and multi-omics data have drawn increasing attention in recent years [4,5,[7][8][9][10][11][12][13][14][15][16][17][18]. However, existing evaluation tools mostly belong to two major categories, congruence (correlation-based) analysis and authentication (machine-learning-based) analysis, and do not sufficiently serve the purpose of identifying appropriate models for precision medicine.…”