Variability Of Gene Expression Profile In Breast Cancer Core Biopsy: Impact On Prognostic And Predictive Multigene Signatures

e21049 Background: Gene expression profiling has been shown to be effective in the analysis of postoperative tumor samples in breast cancer. However, when small specimens like core biopsy samples are considered, the small size of the collected material makes it difficult to obtain high tumor percentage in the specimen. Thus, we tested the strategy of obtaining three independent cores for molecular analysis and assessed the influence of the stability of different prognostic and predictive signatures. Methods: We prospectively collected breast core biopsies from 82 consecutive patients treated by pre-operative chemotherapy. In each patient, three cores from different tumor areas were sampled for molecular testing, independent of the cores collected for histopathological analysis. RNA was isolated by QIAGEN RNeasy kits, gene expression profiling was carried out using Affymetrix HG-U133 Plus 2.0 microarrays. Results: In 27 patients, three independent specimens were analyzed by microarray gene expression profiling (in total 81 samples). When this training set was analyzed by unsupervised Principal Component Analysis, the overall concordance between the three tumor cores was observed, but no such concordance was seen for three tumors (for each with one significant outlier). When we assessed the set of chosen 20 prognostic and predictive signatures, we confirmed that the homogeneity of gene expression within the signatures is significantly better than for random sets of genes, but we observed the notable differences between the signatures. The heterogeneity decreased when genes with low between-patient variance were excluded; they also showed a higher intra-patient variability of gene expression. We also assessed the heterogeneity of single expression markers and furher validated our findings on an independent set of 55 patients. Conclusions: The use of three independent cores for gene expression testing in breast cancer may be a successful strategy to overcome tumor heterogeneity and sampling error and result in more stable results of prognostic and/or predictive signatures. Supported by Polish National Science Center grant N402 6861 40