[1039] Molecular Classification Helps Discriminate between Oncocytomas and Chromophobe Renal Carcinomas Using Meta-Analysis of Gene Expression Microarrays

Vladimir A Valera Romero, Beatriz A Walter Rodriguez, Maria J Merino. National Cancer Institute, National Institutes of Health, Bethesda, MD

Background: Renal oncocytoma and chromophobe renal cell carcinomas are tumors that may share some morphologic features. Given the differences in prognosis, several attempts have been made to identify unique molecular markers based on gene expression studies. The results have been limited by the reduced number of cases included in each study. In this study, we attempted to find unique gene expression profiles of renal oncocytomas and chromophobe RCC by systematically combining and analyzing publicly available microarray data.
Design: Microarray datasets containing the MeSH identifiers “Chromophobe” and “Oncocytoma” were queried in the NCBI Gene Expression Ommibus (GEO) database. Original, raw intensity files from single-platform (Affymetrix®) studies were retrieved. Only non-redundant samples were used for analysis. Built-in quality control probes were used to ensure comparability. After data preprocessing, gene expression signatures were investigated by unpaired T-test analysis with stringent conditions for false discovery rate and gene fold-changes. A classifier based on gene expression pattern was built and externally validated. Analyses were conducted with Genespring® and R/Bioconductor software.
Results: Seventeen databases were initially identified. From these, 10 databases were based on Affymetrix® gene microarray technology, and 4 non-redundant databases were used for analysis, comprising 34 chromophobe tumors, 40 oncocytomas, and 38 normal samples. Uniform array data preprocessing, 5291 genes were found differentially expressed. After data reduction by PCA, 121 genes responsible for the highest variability among samples were used for building a classifier. Apolipoprotein E (APOE, 3.9-fold), Aquaporin 6 (AQP6, 9.4-fold) and Phosphoenolpyruvate Carboxykinase 1 (PCK1, 15.1-fold) showed higher expression in oncocytomas. Osteoactivin (GPNMB) and Claudin 8 (CLDN8) showed increased expression in chromophobe tumors (9.2 fold and 14.2-fold respectively). The combination of genes helped to accurately classify tumors with an overall accuracy of 94%.
Conclusions: In this study, we have identified unique gene expression patterns comparing chromophobe RCC tumors and renal oncocytomas reanalyzing original, publicly available raw microarray data. The results demonstrated that gene expression profiling can be used to differentiate benign and malignant renal oncocytic tumors with high accuracy.
Category: Genitourinary (including renal tumors)

Tuesday, March 20, 2012 9:30 AM

Poster Session III # 158, Tuesday Morning

 

Close Window