Abstract B48: Cross sectional analysis of copy loss in Barrett’s Esophagus

Abstract

Abstract B48 Introduction Barrett’s Esophagus (BE) is a pre-malignant neoplasm that increases the risk of developing esophageal adenocarcinoma (EA) (Paulson et. al. Cancer Cell 2004;6(1):11-6). To identify groups of BE patients at high risk of cancer progression, we sought to identify common chromosomal aberrations across the full risk spectrum of the condition. We implemented a meta-analysis of three studies from the Seattle Barrett’s Esophagus Project. The goal of this analysis was to combine SNP and array-CGH datasets of chromosomal loss from BE and EA samples to pinpoint regions of common loss across patients. Methods The three datasets included Illumina 33k SNP arrays on whole biopsies (34 patients) and surgical resections specimens (8 patients), an Illumina 317K SNP array on 12 flow purified biopsies (1 patient) and a 4,500 spot bacterial artificial chromosome (BAC) hybridization array on 157 flow purified samples (72 patients). When there were multiple samples from a patient, we included the union of all detected lesions across those samples but only counted a lesion once per patient for the purposes of analysis. All SNP arrays were run on both BE and normal (gastric or lymphocyte) samples from the same patients for comparison. All BAC arrays were run on BE samples and compared against a common reference sample. Illumina’s BeadStudio software was used to call genotypes and produce signal intensity data in log2(Rsub/Rref) format that represents the difference in copy number of BE versus normal samples, where we assume normal samples have no aberrations. We then processed the SNP data to call regions of copy number loss using GLAD (Hupe et. al. Bioinformatics 2004;20(18):3413-22) setting logR ratio thresholds of -0.2 for single and -1.5 for double copy loss. BAC data was processed by a wavelet method (Hsu et. al. Biostatistics 2005;6(2):211-26) to call copy loss, copy gain or no aberration for every BAC. Regions of copy number loss, for the combination of both SNP and BAC datasets, were analyzed using STAC (Diskin et. al. Genome Res. 2006;16(9):1149-58) to identify statistically significant areas of loss across samples. The STAC analysis was performed at 0.5Mb resolution using 500 permutations. Results The combined STAC analysis identified 78 regions that were significant at the 95% confidence level, after multiple testing correction, including some previously known losses at chr. 3: 59-61MB (FHIT, FRA3B), chr.16: 77-77.5Mb (WWOX, FRA16D), chr. 9p: 21-32Mb (p16/CDKN2A/INK4a), and some newly discovered losses at chr. X: 31.5-32Mb (DMD), chr. 22: 22.5-23Mb (SMARCB1, DERL3, SLC2A11, MIF, GSTT1, GSTT2, DDT, CABIN1, SUSD2, GGT5) and chr. 18: 57-57.5Mb (CDH20). Conclusions Combining copy number data across studies in STAC increases sample size that may increase power to detect statistically significant regions of copy number loss across samples. We are currently working to extend the same analysis to loss of heterozygosity and copy gain in the SNP array data. Citation Information: Cancer Prev Res 2008;1(7 Suppl):B48.