Image-based screening of mammary tumors

David W. Knowles

1 R33 CA118479-01 National Institutes of Health
October 2006-2009

Research Plan 

Part A) Specific Aims:
Under a current exploratory Department of Defense Breast Cancer Research Program Award (see: which has supported a multidisciplinary collaboration with Sophie Lelièvre, a cancer biologist at Purdue University, we have developed image analysis methods to quantitatively describe the organization of specific nuclear chromatin-associated proteins. By applying these Local Bright Feature (LBF) methods to three-dimensional culture models that mimic normal and malignant breast epithelial tissue we have demonstrated that the distribution of Nuclear Mitotic Apparatus protein (NuMA) and heterochromatin related protein histone 4 methylated on lysine 20 (H4-K20m) are biomarkers capable of distinguishing non-neoplastic and malignant human mammary epithelial cells.
The goal of this project is to develop our current technology to produce a method capable of turning high resolution fluorescence images of human mammary epithelial tissue into tissue-maps which report the probable nonneoplastic, premalignant and malignant phenotype at cellular resolution. Our long term goal is to aid the treatment decision process of breast cancer patients by providing pathologists with a phenotype tissue-map, based on nuclear protein organization, to aid and support the histological classification of biopsied breast tissue.
Our working hypothesis is that the distribution of chromatin-related proteins will permit a novel imaging-based phenotype screening of individual nuclei and the recognition of subtle differences in tissue morphology and behavior, which would enable better detection of benign and malignant lesions. Our rationale is that chromatin organization and associated redistribution of chromatin-related proteins reflect the changes in gene expression that accompany alterations in cell phenotype. Thus, a wide range of distinct distributions of chromatin-related proteins characteristic of different stages of breast cancer and/or degrees of cell malignancy should be recognized. The goal of this project will be achieved in three specific aims.

Aim 1: To develop and assess the capability of our Local Bright Feature (LBF) analysis methods to identify mixed populations of phenotypically different human mammary epithelia. Currently, our LBF analysis methods have been developed using phenotipically homogenous populations of both nonneoplastic and malignant cells. In this phase of the project we will demonstrate that we can detect phenotypically different cells within a heterogeneous population. We will expand our previous work by using cluster analysis on the measured distributions of NuMA and H4-K20m to identify and group phenotypically similar cells from heterogeneous populations of premalignant cultured cells, mixed populations of premalignant and malignant cultured cells and premalignant human mammary tissue biopsies.

Aim 2:  To develop and assess the capability of our Local Bright Feature (LBF) analysis methods to automatically analyze heterogeneous populations of human mammary epithelia. Currently our image analysis techniques have been developed to automatically analyze the nuclear organization in homogeneous populations of epithelia with nuclei of similar volume and “well behaved” shape. One major challenge of working with heterogeneous populations of epithelia is identifying those cells belonging to common tissue-structures and working with nuclei with variations in size and shape. In this phase of the project spatial statistical methods will be developed to identify neighbouring cells comprising a common tissue structure and novel improvements will be made to our LBF analysis to maintain automation and analysis accuracy when dealing with morphologically heterogeneous populations of epithelia.

Aim 3: To develop and assess the capability of an image-based classification system, that uses the nuclear organization of specific proteins to define new sub-classes of various graded lesions. In this phase of the project we will use the cluster analysis results from non-neoplastic, premalignant and malignant cells to define a set of features that characterize these cell phenotypes. Using these we will develop a classification system which will assign the probable tissue phenotype at cellular resolution. This phenotype tissue-map technology will be tested on needle-core biopsies of a variety of premalignant tumors with the aim of defining sub-classes of graded lesions. The results will be correlated with the histopathology of the initial needle-core and the follow-up surgical biopsies with the hope of predicting more aggressive phenotypes missed by the initial screen.