A meta-analysis of gene expression-based biomarkers predicting outcome after tamoxifen treatment in breast cancer

Abstract

To date, three molecular markers (ER, PR, and CYP2D6) have been used in clinical setting to predict the benefit of the anti-estrogen tamoxifen therapy. Our aim was to validate new biomarker candidates predicting response to tamoxifen treatment in breast cancer by evaluating these in a meta-analysis of available transcriptomic datasets with known treatment and follow-up. Biomarker candidates were identified in Pubmed and in the 2007-2012 ASCO and 2011-2012 SABCS abstracts. Breast cancer microarray datasets of endocrine therapy-treated patients were downloaded from GEO and EGA and RNAseq datasets from TCGA. Of the biomarker candidates, only those identified or already validated in a clinical cohort were included. Relapse-free survival (RFS) up to 5years was used as endpoint in a ROC analysis in the GEO and RNAseq datasets. In the EGA dataset, Kaplan-Meier analysis was performed for overall survival. Statistical significance was set at p<0.005. The transcriptomic datasets included 665 GEO-based and 1,208 EGA-based patient samples. All together 68 biomarker candidates were identified. Of these, the best performing genes were PGR (AUC=0.64, p=2.3E-07), MAPT (AUC=0.62, p=7.8E-05), and SLC7A5 (AUC=0.62, p=9.2E-05). Further genes significantly correlated to RFS include FOS, TP53, BTG2, HOXB7, DRG1, CXCL10, and TPM4. In the RNAseq dataset, only ERBB2, EDF1, and MAPK1 reached statistical significance. We evaluated tamoxifen-resistance genes in three independent platforms and identified PGR, MAPT, and SLC7A5 as the most promising prognostic biomarkers in tamoxifen treated patients.