Breast cancer is a very heterogeneous disease. Although most types respond well to therapy, and five-year survival rates in many countries are at 80% or higher, a significant minority of women still have poor prognoses.

Breast tumours are already classified according to the receptors they express, but within any receptor phenotype there is a subset of tumours with poor responses. Identifying those patients who are likely to respond poorly to standard treatments and who might benefit from more invasive chemotherapy or from enrolment in a clinical trial is now an important clinical priority.

A large group of researchers led by Fraser Symmans from the M.D. Anderson Cancer Center, Houston, Texas, USA has now published the results of a clinical study to develop a predictor of therapy response and survival in herceptin-receptor negative breast cancer based on gene expression signatures. Patients diagnosed with this disease were recruited into two independent cohorts, a discovery cohort comprising 310 patients and a validation cohort comprising 198.

All patients were treated with sequential taxane and anthracycline–based chemotherapy, and those whose tumours were oestrogen receptor positive were then given endocrine therapy. Biopsies were taken from all tumours and gene expression profiles obtained using Affymetrix microarrays. Patterns of gene expression (signatures) were derived from the discovery cohort for oestrogen receptor positive and negative cases separately, and these were used to develop an algorithm to predict response to treatment.

This algorithm uses three separate gene signatures: for sensitivity to endocrine therapy; for resistance to chemotherapy (based on prediction of early or distant relapse or of extensive residual disease); and of sensitivity to chemotherapy (based on prediction of complete response or minimal residual disease). All patients were then classified as either predicted treatment sensitive or predicted treatment resistant.

The resulting algorithm was then applied firstly to all samples in the discovery cohort and then, separately, to all samples in the validation cohort (which had not been used during development). The main outcome tested was distant relapse free survival (DRFS) at a median follow-up time of three years. Twenty-eight percent of patients in the validation cohort were predicted to be treatment sensitive.

For all patients in this cohort, the distant relapse free survival at three years was 92% in the predicted treatment sensitive group and 75% in the predicted insensitive group. This corresponds to an odds ratio for relapse of 4.1 for those patients predicted to be treatment insensitive (p=0.002). A slightly higher percentage (30%) of patients with oestrogen receptor positive tumours were predicted to be treatment sensitive, and this sub-group of patients had an excellent DRFS of 97% at three years.

Patients with oestrogen receptor negative tumours also did much better if they were predicted to be treatment sensitive, although the overall figures were lower.

Several other methods for predicting response in breast cancer, based on clinical and pathological characteristics and/or on genetic profiling are in clinical use or under investigation. The researchers evaluated their algorithm alongside some of these, and found that adding predictions based on staging and oestrogen receptor status to their model improved its predictive value. The other genomic predictors evaluated tended to predict a worse outcome for patients who Symmans and co-workers predicted would respond well to chemotherapy. The researchers concluded that the algorithm developed would already be a useful addition to diagnostic practice, although additional validation studies are still necessary.

Article: Hatzis, C., Pusztai, L., Valero, V. and 27 others (2011). A Genomic Predictor of Response and Survival Following Taxane-Anthracycline Chemotherapy for Invasive Breast Cancer. JAMA 305(18), 1873-1881