| Current markers for breast cancer, ER, PR, EGFR and HER-2/neu imperfectly predict prognosis or therapy response. Studies by my mentor revealed distinct molecular subtypes with different clinical characteristics using gene expression profiles from LCM-procured carcinoma cells. Subtypes A and B, although exhibiting ER-positive cancers, had vastly different clinical outcomes. The goal was to derive a clinically relevant subset of genes that predicts breast cancer behavior. Using ERTargetDB and literature review, a ten gene subset of estrogen receptor-associated genes among 100 candidates was validated by qPCR and correlated with biochemical and clinical parameters in primary breast cancer biopsies. These results were combined with expression of genes for conventional biomarkers and assessed for prediction of prognosis in a population of ER-positive, early stage primary breast cancers. All genes except PTGDS exhibited a positive relationship with ER and PR levels. Expression of subtype A genes (BCL2, CAXII, ERBB4, LIV1 and RERG) was significantly decreased in cancers exhibiting increased EGFR protein, although this relationship was not observed for HER-2/neu. Subtype B genes were not altered as a function of EGFR or HER-2/neu levels. Expression of subtype A genes was significantly decreased in cancers with positive lymph nodes, higher grade and larger tumors compared to cancers with subtype B gene expression.Surprisingly, PGR gene expression independently predicted prognosis. Survival analyses of ER+/PR+ cancers revealed multi-gene models classifying risk of recurrence and mortality. LIV1, CD34, EDG1 and NQO1 expression distinguished prognosis differently in node-negative breast cancers compared to the node-positive population. Compared to standard clinical indicators of prognosis using the Adjuvant! Online algorithm, multi-gene models provided superior assessment of risk of recurrence and mortality. Survival analyses of patients with ER+/PR+ breast cancers treated with adjuvant Tamoxifen revealed three multi-gene models significantly predicting recurrence and mortality even after adjusting for age, nodal status, chemotherapy and radiation therapy. Bayesian modeling applied in silico using microarray data identified regulatory networks involving interactions among the ten candidate genes suggesting relationships with cancer differentiation and growth. In summary, a novel subset of 14 ER-associated genes was derived that both predicts risk of recurrence of breast carcinoma and response to Tamoxifen treatment. |
