Towards a systems approach to understanding breast cancer

by Mike Hallett (McGill Centre for Bioinformatics, McGill University, Montreal, Canada)

Wednesday 7 May 2008, 16:00 → 17:00 Europe/Stockholm

RB35 (RB35)

It is increasingly evident that breast cancer outcome is strongly influenced by signals emanating from tumor-associated stroma. However, little is known about how gene expression changes in this tissue affect tumor progression. In this talk, we compare gene expression profiles from laser capture-microdissected tumor-associated versus matched normal stroma, and derive transcriptional profiles strongly associated with clinical outcome. We present a stroma-derived predictor that generates new information to stratify disease endpoint, independent of standard clinical prognostic factors and previously published predictors. Our predictor selects poor-outcome patients from multiple clinical subtypes, including node-negative patients, and predicts outcome in multiple published expression datasets generated from whole tumor tissue. Our predictor has increased accuracy compared to previously published predictors, and prognostic accuracy increases when these predictors are integrated using graphical models. Genes represented in the stroma-derived predictor reveal the strong prognostic capacity of differential immune responses as well as angiogenic and hypoxic responses. The computational and statistical aspects underpinning this work are built upon a new approach to analyzing gene expression data that in some sense is "orthologonal" to traditional clustering based tools, and is general in the sense that a wide range of data types can be easily integrated into the system.