Prognostic value of a 92-probe signature in breast cancer.

Salima Akter,Joohun Ha,Jayoung Kim,Liaquat Ali,Wonchae Choe,Abel Matondo,Ara Jo,Muhammad Shahid,Yen Ngoc Nguyen Ngo ,Manh Tuan Nguyen ,Jin Hwan Kim ,Seong Wook Seo ,Ju Seog Lee ,Sung Soo Kim,Dae Young Jun,Yunhee Jo ,In‐Cheol Kang ,Tae Gyu Choi,Ji Youn Yoo,Kyung‐Sik Yoon

doi:10.18632/oncotarget.3525

Abstract

Clinical applications of gene expression signatures in breast cancer prognosis still remain limited due to poor predictive strength of single training datasets and appropriate invariable platforms. We proposed a gene expression signature by reducing baseline differences and analyzing common probes among three recent Affymetrix U133 plus 2 microarray data sets. Using a newly developed supervised method, a 92-probe signature found in this study was associated with overall survival. It was robustly validated in four independent data sets and then repeated on three subgroups by incorporating 17 breast cancer microarray datasets. The signature was an independent predictor of patients' survival in univariate analysis [(HR) 1.927, 95% CI (1.237-3.002); p < 0.01] as well as multivariate analysis after adjustment of clinical variables [(HR) 7.125, 95% CI (2.462-20.618); p < 0.001]. Consistent predictive performance was found in different multivariate models in increased patient population (p = 0.002). The survival signature predicted a late metastatic feature through 5-year disease free survival (p = 0.006). We identified subtypes within the lymph node positive (p < 0.001) and ER positive (p = 0.01) patients that best reflected the invasive breast cancer biology. In conclusion using the Common Probe Approach, we present a novel prognostic signature as a predictor in breast cancer late recurrences.

Highlights

Breast cancer is the leading cause of cancer-related deaths amongst women worldwide [1] and it is recognized to be a molecularly heterogeneous disease [2]
There is a need to identify a prognostic signature that would solve the problems of small patient data and preserve the predictive strength without combining microarray data, to accurately predict the patient’s outcome as well as the biology of the disease
We explored a significant gene signature related to prognosis of breast cancer patients by investigating three independent microarray datasets of heterogeneous primary breast cancer

Summary

Introduction

Breast cancer is the leading cause of cancer-related deaths amongst women worldwide [1] and it is recognized to be a molecularly heterogeneous disease [2]. Meta-analysis is considered to be a promising approach to overcome this limitation by combination of microarray data sets [8] This might have some common problems such as challenges of different probes in individual microarray chips with varying in precision, different relative scales, and diverse dynamic ranges [9, 10]. It has been shown that robust identification of prognostic signature is performed either by the combination of identical [11] or different microarray chip [12] In both approaches, confined probe sets are used because the former method incorporates limited number of probes while the latter excluded majority of the genes required to predict the patient’s outcome. There is a need to identify a prognostic signature that would solve the problems of small patient data and preserve the predictive strength without combining microarray data, to accurately predict the patient’s outcome as well as the biology of the disease

Methods

Results

Conclusion