Image and Informatics Lab, LBNL : Home
Image and Informatics Group, LBNL : Home

Molecular Predictors of 3D Morphogenesis by Breast Cancer Cells in 3D Culture

PLoS Computational Biology , 2010

    J. Han
    H. Chang
    O. Giricz
    GY Lee
    FL Baehner
    JW Gray
    MJ Bissell
    PA Kenny
    B. Parvin


    This paper aims at the development of computational methods for identify and validating molecular signature for a panel of breast cancer cell lines that are cultured in 3D and imaged with phase contrast microscopy. The panel has a significant molecular diversity that leads to a heterogenous morphometric signature as demonstrated by the growth of the colony. The key contributions of this paper are (i) automated subtying based on colony morphogenesis, (ii) molecular association with colony formation and colony subtypes, and (iii) validation of a molecular predictor in human breast tumors and in an in vitro 3D assay that indicates reversion properties. Morphometric colony signatures were imaged through phase contrast microscopy, and the system identified three subtypes, each of which correlates with a particular pathobiological phenotype. For example, morphometric properties (e.g., shape and organization) of 3D colonies cultured with ERBB2-positive cell lines were clustered together. Similarly, morphometric properties of 3D colonies cultured with triple-negative cell lines were also grouped together. Furthermore, association of gene expression data from a heterogeneous population of 3D colonies with morphometric features such as colony size (a metastatic index) revealed that PPARG is a predictor for colonies formed by triple-negative cell lines (e.g., invasive lines), a poor prognosis for breast cancer. Finally, PPARG was validated in human mammary breast cancer tissues through immunohistochemistry, and in a 3D cell culture model through an inhibitor of PPARG that indicates reversion properties. Our analysis also revealed that PPARG is not a predictor when the same cell lines are cultured in 2D.

    click the green ball to see the full version of the paper in PLoS Computational Biology