BACR Translational Research Award: Molecular taxonomy of breast cancer: is there a gold standard?
Britta Weigelt 1,Alan Mackay 2,Roger A hern 3,Rachael Natrajan 2,David SP Tan 2,Mitch Dowsett 2,Alan Ashworth2,Jorge S Reis-Filho2
Cancer Research UK London Research Institute, UK 1,The Breakthrough Breast Cancer Research Centre, The Institute of Cancer Research, London, UK 2,Cancer Research UK Clinical Trials and Statistics Unit, The Institute of Cancer Research, Sutton, UK 3,Academic Department of Biochemistry, Royal Marsden Hospital, London, UK 4
Background
Microarray gene expression profiling has led to a working model for a molecular taxonomy of breast cancer comprising five molecular subtypes: luminal A, luminal B, basal-like, HER2 and normal breast-like. In the past decade, three distinct microarraybased single sample predictors (SSPs) have been described to determine the molecular subtype class of individual samples. It has been assumed that assignments made by different SSPs are synonymous. We analysed the agreement between these three SSPs in the identification of the molecular subtypes, and investigated whether each SSP identifies molecular subtypes with similar associations with outcome.
Method
Breast cancers derived from four microarray datasets (total n=832) were assigned to the molecular subtype classes using the three previously described SSPs. We used Kappa score analysis to determine the agreement between SSPs for the whole classification system and for each molecular subtype separately. Associations with outcome were assessed by Kaplan-Meier analysis.
Results
The agreement between each pair of SSPs in each cohort was modest (Kappa scores 0.238 to 0.740). The assignment of samples to the luminal A, luminal B, HER2 and normal breast-like subtype classes was strongly dependent on the SSP employed, whereas basal-like cancers were robustly identified irrespective of the SSP (Kappa scores >0.810). This was not altered by different microarray-centring methodologies. Different SSPs produced similar survival curves, however the number and identity of cases assigned to the five molecular subtypes in each cohort differed depending on the SSP used, with the exception of the basal-like subtype. Analysis of each molecular subtype individually revealed that the associations with outcome of each molecular subtype other than basal-like and luminal A varied by the SSP employed.
Conclusion
Before introduction of the molecular subtypes into clinical practice, clear definitions and standardised methods for the identification of the molecular breast cancer subtypes are required.
