No-one really understands why some women with breast cancer respond well to chemotherapy while others do not, or how to predict an individual patient's chances of survival. In the January 31 Nature, Laura van't Veer and colleagues describe a gene-expression profiling study of breast tumors (Nature 2002, 415:530-536).

They chose around 100 primary breast cancers (with and without metastases or BRCA1 mutations) and looked at the relative expression levels of 25,000 genes. They used a three-step supervised classification method to distinguish groups of tumors with good or poor prognosis. An expression signature panel of 70 genes could accurately predict 'poor prognosis' patients. This set includes genes encoding proteins involved in cell invasion, metastasis and angiogenesis. Women under 55 who are diagnosed with lymph-node-negative breast cancer have a 28-fold higher chance of developing distant metastases within 5 years if they have a 'poor prognosis signature' rather than a...