Computational Discovery of Breast Cancer Morphologic Subtypes and Their Molecular Underpinnings
Yening Xia, Marco Hefti, Nicholaus Knoblauch, Andrew H Beck. Beth Israel Deaconess Medical Center, Boston, MA
Background: The Cancer Genome Atlas Network (TCGA) recently performed comprehensive molecular profiling on ∼500 cases of invasive breast carcinoma. Several data types (mRNA expression profiling, miRNA sequencing, DNA copy number variations, DNA methylation profiling, and protein expression profiling) were analyzed by the TCGA to define breast cancer molecular subtypes. The goal of our research is: 1) to perform comprehensive and quantitative morphologic analyses of TCGA samples to define breast cancer morphologic subtypes and 2) to identify associations between morphologic and molecular subtypes of invasive breast carcinoma.
Design: We adapted the C-Path platform for use on downloaded whole slide images from 425 breast cancer cases previously analyzed by the TCGA, and quantitated a total of 2370 nuclear, cytoplasmic, epithelial-mesenchymal transition, and global features per image at 20X magnification. We performed k-means clustering on the quantitated morphologic data to identify robust morphologic subtypes. After estimating the optimal number of morphologic subtypes, we performed chi-square analyses to evaluate for associations between morphologic subtypes (identified in our analysis) and the previously determined molecular subtypes.
Results: We identified 5 robust morphologic subtypes (figure 1). Subtypes 1, 2, 3, 4, and 5 were comprised of 95 (21%), 81 (19%), 141 (33%), 65 (15%) and 47 (11%) of the breast cancer cases, respectively. On chi-square analyses, morphologic subtypes were significantly associated with molecular subtypes defined by mRNA (p =1.5 x 10-7) and protein expression (p = 0.006), DNA methylation profiling (p = 0.007) and miRNA sequencing (p = 0.04). Morphologic subtypes may also be associated with molecular subtypes defined by DNA copy number variations (p = 0.08). The standardized residuals from the chi-square analyses show that individual associations between specific morphologic and molecular subtypes were important for rejecting the null hypotheses (i.e. no association between morphologic and molecular subtypes).
Conclusions: Morphologic profiling of comprehensively annotated cancer specimens facilitates the discovery of novel morphologic subtypes and characterization of their molecular underpinnings.
Monday, March 4, 2013 9:30 AM
Poster Session I Stowell-Orbison/Surgical Pathology/Autopsy Awards Poster Session # 44, Monday Morning