Date of Award




Document Type


Degree Name

Doctor of Philosophy (PhD)


Department of Biomedical Sciences

Content Description

1 online resource (xiii, 227 pages) : illustrations (some color)

Dissertation/Thesis Chair

Stewart Sell

Committee Members

Gennadi Glinsky, Bruce Herron, Erasmus Schneider, Douglas Conklin


Bmi-1, breast cancer, microarray, mouse models, tumor initiation, Breast, Stem cells

Subject Categories

Cell Biology | Genetics | Medicine and Health Sciences


Cancer stem cells are the seeds of tumor growth, but there are limited cell-based methods that exist to study the properties of these cells. To create a model of breast cancer stem cells, we isolated tumors from MMTV-PyMT mice. Two out of the four different cell types isolated survived in culture, CD44+CD24- and CD24+CD49f+CD44low. We found that we could initiate tumors with as few as 10 cells injected subcutaneously in the hind leg or orthotopically in the cleared fat pad with CD24+ cells. However, we could not initiate tumors with injection of CD24- cells. We found a requirement for TICs to express CD49f and Bmi-1 to initiate tumors. Subsequent analysis of the CD24- cells indicated that these cells were tumor-derived mesenchymal stem cells (TDMSCs). When these cells were co-injected with tumor initiating cells (TICs), we observed an increase in tumor formation. Those tumors generated from the TICs recapitulated the proportions of cell surface marker phenotypes and the different histopathological subtypes observed in the primary tumors. We also observed an increase in metastasis from the co-injections of TICs with TDMSCs compared to TIC injection alone, and the appearance of sarcomas. In primary MMTV-PyMT tumors, sarcomas are not observed, and we hypothesized that the sarcomas were the result of epithelial to mesenchymal transition of the TICs. To validate our breast cancer stem cell model, we collected gene expression data by microarray analysis of several cell cultures, including positive and negative control cell types. After extensive analysis, we created a 192-human gene signature, which we screened for genes that were significant for relapse-free survival. We identified two genes, PNRC1 and RRM2 that when the ratio of gene expression values for these two genes were calculated, could significantly striate patient data into good and poor prognosis groups. Further analysis allowed us to develop a 3-signature predictor, which we used to assign scores to patients based on calculations using gene expression data. We concluded that our mouse model was relevant to human breast cancer, based on our ability to generate prognostic signatures to identify risk of relapse in human breast cancer patients.