90-gene signature assay for tissue origin diagnosis of brain metastases

Abstract

Abstract Background Brain metastases (BM) are the most common intracranial tumors. 2–14% of BM patients present with unknown primary site despite intensive evaluations. This study aims to evaluate the performance of a 90-gene expression signature in determining the primary sites for BM samples. Methods The sequence-based gene expression profiles of 708 primary brain tumors (PBT) collected from The Cancer Genome Atlas (TCGA) database were analyzed by the 90-gene expression signature, with a similarity score for each of 21 common tumor types. We then used Optimal Binning algorithm to generate a threshold for separating PBT from BM. Eighteen PBT samples were analyzed to substantiate the reliability of the threshold. In addition, the performance of the 90-gene expression signature for molecular classification of metastatic brain tumors was validated in a cohort of 48 BM samples with the known origin. For each BM sample, the tumor type with the highest similarity score was considered tissue of origin. When a sample was diagnosed as PBT, but the similarity score below the threshold, the second prediction was considered as the primary site. Results A threshold of the similarity score, 70, was identified to discriminate PBT from BM (PBT: > 70, BM: ≤ 70) with an accuracy of 99% (703/708, 95% CI 98–100%). The 90-gene expression signature was further validated with 18 PBT and 44 BM samples. The results of 18 PBT samples matched reference diagnosis with a concordance rate of 100%, and all similarity scores were above the threshold. Of 44 BM samples, the 90-gene expression signature accurately predicted primary sites in 89% (39/44, 95% CI 75–96%) of the cases. Conclusions Our findings demonstrated the potential that the 90-gene expression signature could serve as a powerful tool for accurately identifying the primary sites of metastatic brain tumors.

Keywords

• Primary brain tumor
• Brain metastases
• Gene expression profiling
• Tumor origin identification
• Quantitative real-time PCR