This project is based on the premise that genes that are co-regulated in a variety of cell types are likely to be functionally related. We developed that concept utilizing our data mining software (""""""""CellMiner"""""""") to analyze gene expression data in the NCI-60 human tumor cell lines. In our first study using this approach, we derived a set of mutually expression-correlated genes relating to tumor cell migration. We found that most of those genes were involved in a molecular interaction network that controls cell migration and invasion by actions on extracellular matrix required for cell mobility, i.e. to allow cells to penetrate through the extracellular matrix. We developed a molecular interaction map of the network that suggested target sites for potential inhibition of tissue invasion/metastasis. In a second study, we identified mutually correlated genes that function specifically in epithelial-like or mesenchymal-like human tumor cells, based on correlated expression with a set of tight-junction genes. More explicitly, we devised a new approach to the characterization of tumor cells having epithelial character, based on the expression of genes for the proteins involved in tight-junctions, structures that are required to maintain adherence and polarity of epithelial cells. Since tight-junctions are definitive structures in epithelial cells, we expected that the expression of those genes would gage epithelial character of different types of tumor cells. We obtained a consensus subset of NCI-60 cell lines and a corresponding consensus of selective gene expressions specific for epithelial tight junctions. and employed them as signatures to identify other candidate epithelial-specific genes having epithelial character. We used that signature to identify a large set of genes that were selectively expressed in the NCI-60 epithelial consensus cell lines. Similarly, we identified genes that were selectively NOT expressed in those cell lines. Most of the selectively expressed genes had known epithelial functions relevant to networks governing cell-cell adhesion structures (tight junctions, adherens junctions, desmosomes) and their connections to cytoskeletons (actin, microtubules, intermediate filaments), as well as effects on cell proliferation and differentiation. Most of the selectively NOT expressed genes had known functions in mesenchymal cell types. The selectively expressed and selectively NOT expressed genes (including certain microRNAs) came together as a molecular interaction map of the controls on the transitions between epithelial and mesenchymal phenotypes. This provided a new and more detailed and integrated picture of those transitions, which are critical to the invasion and metastasis capabilities of most malignant tumors. We then investigated whether or to what extent the epithelial/mesenchymal gene expression correlations that we observed in the NCI-60 cell lines are preserved in the much larger gene expression database of the Cancer Cell Line Encyclopedia (CCLE) of the Broad institute of MIT and Harvard, which includes 1000 human tumor cell lines of a wider variety of tissues of origin. We found excellent agreement between gene-gene expression correlations (z-scores) in the NCI-60 cell lines (all tissues of origin) and the CCLE cancer cell lines derived from particular epithelial tissues (breast, colon, lung, pancreas, stomach, ovary). This indicates that the extensive NCI data on NCI-60 sensitivity profiles for drug, chemical compound, and natural products may carry over to the CCLE epithelial tumor cell lines, which would provide new tissue-specific information of various chemical agents, as well as enhanced information of gene expression signatures for tumor cell chemosensitivity. Our findings suggest that an NCI-60 gene expression pattern that correlates with sensitivity/resistance of a given compound would correspond to a similar gene expression pattern in CCLE cell lines (e.g. from certain tissues of origin) having similar sensivity/resistance to that compound. This hypothesis can be tested as a step towards prediction of chemosensivity of clinical tumors on the basis of gene expression pattern. We plan a similar investigation of gene expression data on human tumor tissues of various tissues of origin, data currently available from The Cancer Genome Atlas (TCGA) project.
Showing the most recent 10 out of 14 publications
