A lung cancer genomic risk prediction model derived from paraffin-embedded tissue

Abstract

10059 Background: We previously developed a validated fresh tissue-based genomic risk model in patients with early stage non-small cell lung cancer (NSCLC) using the Affymetrix U133 plus 2.0 Genechip. Limitations of this fresh tissue-based model include the need for immediate processing and limited availability; however, formalin-fixed, paraffin-embedded (FFPE) tissue is readily available and archived on every patient resected in North America. We investigated the ability of gene expression profiles generated on DNA microarrays using RNA isolated from FFPE NSCLC specimens to distinguish short-term and long-term survivors. Methods: Five to ten 5 um sections of FFPE tumor were collected from 61 NSCLC patients consisting of equal numbers of long- (+5-year) and short-term (<2 year cancer death) survivors. Fifty-five samples were microdissected (6 samples contained no tumor tissue) and RNA was extracted using a proprietary procedure of Response Genetics, Inc. For this feasibility study, Actin 300 < 30 cTs was chosen as a threshold for adequate RNA quantity for amplification to the GeneChip. Amplification and labeling of RNA were done using the Affymetrix two cycle amplification kit. The resulting cRNA was successfully hybridized to the U133 plus 2.0 GeneChip in 54/55 samples (98%). Data were analyzed using the SAM statistical software with Kaplan Meier survival analyses. Results: All analyses were performed using unsupervised hierarchical clustering and blinded duplicate samples had nearly identical gene expression profiles, indicating reproducibility. Adenocarcinoma segregated from squamous cell carcinoma with 98% accuracy (p=0.00004). A differentially expressed gene list between long and short survivors was determined. Distinct gene clusters were observed within each histological type segregating the tumors according to outcome. Kaplan Meier survival analysis stratifying on these clusters revealed significant differences in survival (cluster 1 and cluster 2 median survival>75 mos. vs. 30 mos., respectively; p<0.001). Conclusions: We have demonstrated the feasibility of creating a preliminary genomic risk prediction model using FFPE NSCLC tissue. Data will be presented on a larger training set (100+ patients) and a separate validation cohort of 100 patients. [Table: see text]