99P - Consensus on breast cancer cell lines classification for an effective and efficient clinical decision-making

Date 07 May 2015
Event IMPAKT 2015
Session Welcome reception and Poster Walk
Topics Basic Science
Bioethics, Legal, and Economic Issues
Breast Cancer
Presenter Heloisa Milioli
Citation Annals of Oncology (2015) 26 (suppl_3): 31-33. 10.1093/annonc/mdv121
Authors H.H. Milioli1, I. Tishchenko1, C. Riveros2, J. Sakoff3, R. Berretta2, P. Moscato1
  • 1Centre For Bioinformatics, Biomarker Discovery And Information-based Medicine, The University of Newcastle, 2305 - Newcastle/AU
  • 2School Of Electrical Engineering And Computer Science, The University of Newcastle, Newcastle/AU
  • 3Department Of Medical Oncology, Calvary Mater Hospital Newcastle, Newcastle/AU



Background: Cell lines are widely used to investigate the breast cancer clinical pathology and molecular heterogeneity. The stratification of tumour lineages according to intrinsic subtypes (luminal A, luminal B, HER2-enriched, normal-like and basal-like) has also changed the functional management of laboratory models. In particular, cancer cells cultures are explored to test and validate molecular drivers for group targeted therapies. In this study, we aim to provide a robust classification of the in vitro models by comparing the gene expression profile of cell lines and approximately 2,000 primary breast tumours.

Methods: Microarray data sets from three distinct platforms (Affymetrix HG-U133A, Agilent A-AGIL-28, and Illumina Human WG-6v) are considered for the cell lines subtype classification. Here we perform an ensemble learning approach based on 24 classifiers trained with the METABRIC breast cancer transcriptomic data. The proposed method assesses the 50-genes signature from the PAM50 method to classify 75 cell lines.

Results: Overall, we found a consensus on the subtype's assignment within breast cancer cell lines. However, some preliminary results show a disagreement on the subtype labels attributed to widely used cell lines such as BT474, HCC1500, HCC1954, and SKBR3. These cell lines had different labelling across studies and platforms. Our achievements also corroborate with controversial phenotypes published in the literature, normally defined by single markers.

Conclusions: Using ensemble learning we were able to predict intrinsic subtypes considering the similarities between in vitro and in vivo specimens. The integration of different microarray platforms also evidenced the strength of the gene expression profile shared among cell lines. This approach adds a new perspective to effective experimental models that are used to investigate intrinsic markers for improving the therapeutic decisions and the clinical outcomes.

Disclosure: All authors have declared no conflicts of interest.