Gene expression signatures that predict outcome of tamoxifen-treated estrogen receptor-positive, high-risk, primary breast cancer patients: a DBCG study

Abstract

Background: Tamoxifen significantly improves outcome for estrogen receptor-positive (ER+) breast cancer, but the 15-year recurrence rate remains 30%. The aim of this study was to identify gene profiles that accurately predicted the outcome of ER+ breast cancer patients who received adjuvant Tamoxifen mono-therapy.

Methodology/principal findings: Post-menopausal breast cancer patients diagnosed no later than 2002, being ER+ as defined by >1% IHC staining and having a frozen tumor sample with >50% tumor content were included. Tumor samples from 108 patients treated with adjuvant Tamoxifen were analyzed for the expression of 59 genes using quantitative-PCR. End-point was clinically verified recurrence to distant organs or ipsilateral breast. Gene profiles were identified using a model building procedure based on conditional logistic regression and leave-one-out cross-validation, followed by a non-parametric bootstrap (1000x re-sampling). The optimal profiles were further examined in 5 previously-reported datasets containing similar patient populations that were either treated with Tamoxifen or left untreated (n = 623). Three gene signatures were identified, the strongest being a 2-gene combination of BCL2-CDKN1A, exhibiting an accuracy of 75% for prediction of outcome. Independent examination using 4 previously-reported microarray datasets of Tamoxifen-treated patient samples (n = 503) confirmed the potential of BCL2-CDKN1A. The predictive value was further determined by comparing the ability of the genes to predict recurrence in an additional, previously-published, cohort consisting of Tamoxifen-treated (n = 58, p = 0.015) and untreated patients (n = 62, p = 0.25).

Conclusions/significance: A novel gene expression signature predictive of outcome of Tamoxifen-treated patients was identified. The validation suggests that BCL2-CDKN1A exhibit promising predictive potential.

Figure 1. Genes identified and their expression pattern.

A) The 2-, 8- and 9-gene signatures identified by various statistical analyses. B) ΔΔCt of the genes present in the 2-, 8- and 9-gene signatures. BCL2 overlap in all three, whereas CDKN1A is in the 2- and 9-gene signatures, and PRKCE and EGFR are in both the 8- and 9-gene signatures. A positive ΔΔCt_median value denote that the expression of the gene is highest in the tumor sample from patients without recurrence, whereas a negative value means the expression is higher in the tumor samples from patients with recurrence.

Figure 2. Joint distribution of the ΔΔCt values of BCL2 and CDKN1A.

The diagonal line corresponds to the rule determined by conditional logistic regression. Pairs to the right of the line are correctly classified with respect to their outcome (recurrence/non-recurrence) (accuracy of 75%), whereas pairs left of the line are classified incorrectly.

Figure 3. Performance of the identified genes.

The capabilities of the identified 2-, 8- and 9-gene signatures to predict recurrence was evaluated in 6 independent gene expression datasets. A) Summarized results of accuracy (%), along with sensitivity/specificity in parenthesis (both given as %), of the identified signatures to predict recurrence. B–G) Dot-plots of the identified 2-gene signature (BCL2-CDKN1A) illustrating the probability of recurrence. The vertical line separates the cases, i.e. patients with recurrence (left of the line) from controls (right of the line). The horizontal line refers to the cut-point used, hence the upper left and lower right corners includes the correctly classified patients. X-axis denotes the patient index in the study (same random order as original study). The Y-axis is the SVM probability of recurrence. B) GSE1378 C) GSE1379 D) GSE9893 E) GSE12093 F) GSE6532-GPL96 and G) GSE6532-GPL570.

Figure 4. Survival analysis of the 2-gene signature.

Kaplan-Meier curves of recurrence-free survival according to model-based prediction of outcome using the 2-gene signature (BCL2-CDKN1A) for the independent gene expression dataset GSE2990. Grey line (top) indicates the good outcome signature, whereas the black line (bottom) indicates the poor outcome signature. Only data from post-menopausal (>50 years) and ER+ breast cancer patients were included in the analysis. Data was adjusted for clinical variables. A) Tamoxifen-treated patient samples (N = 58). B) Untreated patient samples (N = 62).