Abstract: Background and purpose: Cancer of unknown primary (CUP) represents approximately 5%~10% of malignant neoplasms. For CUP patients, identification of tumor origin allows for more specific therapeutic regimens and improves outcomes. Methods: By retrieving the gene expression data from ArrayExpress and Gene Expression Omnibus data repositories, we established a comprehensive gene expression database of 5 800 tumor samples encompassing 22 main tumor types. The support vector machine-recursive feature elimination algorithm was used for feature selection and classification modelling. We further optimized the RNA isolation and real-time quantitative polymerase chain reaction (RTQ-PCR) methods for candidate gene expression profiling and applied the RTQ-PCR assays to a set of formalin-fixed, paraffin-embedded tumor samples. Results: Based on the pan-cancer transcriptome database, we identified a list of 96-tumor specific genes, including common tumor markers, such as cadherin 1 (CDH1), kallikrein-related peptidase 3 (KLK3), and epidermal growth factor receptor (EGFR). Furthermore, we successfully translated the microarray-based gene expression signature to the RTQ-PCR assays, which allowed an overall success rate of 88.4% (95%CI: 83.2%-92.4%) in classifying 22 different tumor types of 206 formalin-fixed, paraffin-embedded samples. Conclusion: The 96-gene RTQ-PCR assay represents a useful tool for accurately identifying tumor origins. The assay uses RTQ-PCR and routine formalin-fixed, paraffin-embedded samples, making it suitable for rapid clinical adoption.

Key words: Cancer of unknown primary, Tumor tissue origin, Gene expression profiling, Real-time quantitative polymerase chain reaction, Immunohistochemistry