Molecular identification of carcinoma of unknown primary (CUP) with gene expression profiling

Abstract

21024 Background: We previously reported a method for determining the site of tumor origin for CUP by comparing a 92- gene expression profile to that in a database created from 600 primary and metastatic tumor bank specimens of known origin. K-nearest neighbor analysis was used to determine the likelihood of an unknown patient specimen originating from a particular site with the likelihood assigned as a confidence level and reported as high, medium, low, or unclassified. Herein, we report the analysis of the gene expression profiling results from our initial series of clinical CUP specimens. Methods: We reviewed the results of 76 consecutive de-identified patient samples submitted to our laboratory for routine CUP testing. RNA was extracted from the formalin-fixed, paraffin-embedded (FFPE) tissue blocks and cDNA products used in a semi-quantitative real-time PCR to detect 87 tumor-associated genes and 5 reference genes. Gene expression data were then compared with our database and k-nearest neighbor analysis used to identify the 5 closest neighbors. If all 5 or 4/5 were the same, the result was classified as “high likelihood”, 3/5 = “moderate likelihood”, 2/5 = “low likelihood” and none matching was “unclassifiable”. Results: For the 76 clinical CUP samples tested, gene profiling analysis yielded high-likelihood predictions for 34 (45%), moderate for 12 (16%), low for 12 (16%), and unclassified for 14 (18%); amplification was inadequate for 4 (5%) samples. Overall, gene profiling analysis yielded classifiable predictions in 58 (76%) of clinical CUP samples. An occult carcinoid, metastatic melanoma and adenocarcinoma of the endocervix were identified and then found clinically using this assay. Conclusions: Our previous findings indicate that gene expression profiling can correctly identify the site of tumor origin in a high percentage of tumor bank samples. Data from the present study suggests that this approach can identify a primary site of tumor origin in 76% of actual clinical specimens from pathologist-submitted CUP cases. No significant financial relationships to disclose.