[297] Multivariate Analysis of Immunohistochemical Data from 6925 Breast Cancer Patients: Molecular Classification Using 6 Well Known Markers

HP Sinn, S Aulmann, P Schirmacher. University Hospital Heidelberg, Heidelberg, Germany

Background: Molecular classification of breast cancer using cluster analysis of gene expression data was shown to be a good indicator of the biology of these tumors, and correlates with the histopathology, the treatment response, and the patients outcome. Therefore, it is desirable to define these clusters routinely by immunohistochemical methods.
Design: Data from 4952 patients with invasive breast cancers from a 12 year period (1995-2007) were re-analyzed. All patient data sets include information on the Ki67, bcl2, ER, PR, and p53 immunohistology, and the tumor stage. Where available, also the lymph node status was taken into account. These data were analyzed with regard to intrinsic clustering and relationship to lymph node status. Based on these results, rules for IHC clustering were developed to yield results comparable to clustering of gene expression data.
Results: Principal component analysis (PCA) revealed the presence of three distinct clusters of breast cancers: ER/PR/bcl2 positive tumors, HER2 positive tumors, and tumors with accelerated proliferation characterized by high scores of p53 and Ki67. HER2 positive tumor could be further subdivided into a classical type (no p53 overexpression, no ER positivity), an ER positive type (ER > 2%, no p53 overexpression), and a p53 overexpression type (>50% p53 pos. nuclei). The metastatic behaviour of the HER2+++/ER+ cluster was identical to other ER+ tumors, and 10%, but the other types of HER2+++ tumors showed a more highly metastatic phenotype. Therefore, HER2+++/ER+ tumor were clustered in the Luminal B group together with other tumors having a Ki-67 score of > 30% and separated from the Luminal A group without these features. HER+++ tumors with ER-/PR- formed seperate cluster. It was not possible to distinguish basal- and non-basal phenotypes of triple negative tumors using this set of markers.
Conclusions: IHC clustering of breast cancer data that correlates with the biological behavior is possible and yields roughly the same numerical distribution as could be expected from molecular clustering. An expanded set of markers, including basal cytokeratins, will be needed to get a more refined clustering.
Category: Breast

Monday, March 9, 2009 1:00 PM

Poster Session II # 47, Monday Afternoon