1033

Background: The development of cancer is a multi-step process, where the accumulation of genetic changes such as rearrangements, amplifications and deletions drives tumorigenesis. These copy number changes can now be profiled at the whole genome level with single gene resolution using array-CGH. Breast cancer is the most common cause of cancer death in women but current methods are unsatisfactory in predicting disease outcome and response to therapy. The aim of this study was to use array-CGH to generate a breast cancer taxonomy that correlates with clinical features and is predictive of outcome. Methods: The samples studied were 134 frozen primary breast cancers and 20 breast cancer cell lines (used primarily to validate the technique and to characterise differences between primary tumours and cell lines). DNA was isolated, labelled by random prime labelling and hybridised onto the Vysis Genosensor 300 DNA-array chip with 287 loci (274 on somatic chromosomes and 13 on sex chromosomes). The array contains known tumour suppressor genes, oncogenes and regions that have been reported to be gained or lost in cancers. Results: Changes known to be associated with breast cancers such as amplification of 8q24 (78%), 11q13 (82%), 17q12 (50%), 20q13 (70%) and loss of 9p21 (53%) were identified in both tumours and cell lines. In addition, a number of loci/genes frequently gained/lost were identified but not previously reported in breast cancer. Unsupervised clustering analysis has identified different subgroups and correlations with clinic-pathological variables. Supervised clustering is being used to identify potential predictors of outcome. Comparison of changes in cell lines and primary tumours indicate that there is a greater accumulation of abnormalities in cell lines (75% of primary tumours have changes in less than 15% of loci studied but in contrast, 75% of cell lines have changes in more than 15% of loci studied). Conclusion: Array-CGH molecular profiling identifies distinct subsets of breast cancer and has the potential to generate predictive classifications.

[Proc Amer Assoc Cancer Res, Volume 45, 2004]