Introduction: Endometrial cancer is the most common gynecologic cancer in the United States and the fourth most common cancer in women. Incidence and mortality are both rising, and five-year survival rates have decreased since the 1970s. Most endometrial cancers are estrogen receptor alpha (ER) positive and are related to excess estrogen signaling. ER is an estrogen-activated transcription factor that mostly binds to sequences in the DNA known as estrogen response elements (ERE) to regulate gene expression. It is also a known contributor to many breast cancer pathologies, and many therapeutics targeting ER in breast cancer have been successfully developed. However, these successes in breast cancer have not translated to endometrial cancer, likely because ER exhibits tissue-specific behavior. ER has been shown to bind to different genomic locations and to have different cofactors in breast and endometrial cancer cells. However, much is still unknown about the extent to which ER behavior differs between the tissues. We seek to better characterize the mechanisms controlling ER behavior in both breast and endometrial cancer cells. To do this, we will use computational techniques to determine which genomic features of an ERE predict its ability to recruit ER and affect gene expression. We will also use a CRISPR-based experimental technique to knock out potential cofactors of ER and measure the changes in ER binding and gene expression. This will allow us to better understand the mechanisms of ER binding control, and how they differ between the cell types.
Materials and
Methods: To determine which features of an ERE predict its ability to recruit ER and affect gene expression, genomic data for each ERE was compiled. This data includes sequence information, binding locations of transcription factors and histone modifications, chromatin accessibility data, and RNA polymerase II activity data. The data was then evaluated by the feature selection algorithm, Boruta, which identified features that are predictive of an ERE binding ER and affecting gene expression using random forest classification. The Boruta analysis was done separately on data from two cell lines, an endometrial cancer and a breast cancer cell line, and then the results were compared to determine what features were predictive in both cell lines, neither cell line, or only one cell line. To experimentally determine the cofactors of ER in a breast and an endometrial cancer cell line, the CRISPR/Cas9 system was used with a library of guide RNAs targeting ~2500 potential cofactors. We created cell lines where the expression of GFP, a reported gene, is driven by multiple EREs, which correlates fluorescence with ER activity. The reporter cell lines are being transduced with the guide RNA library such that one potential cofactor is knocked out per cell. The cells are being treated with estrogen and then sorted for fluorescence and sequenced to determine which cofactors affected ER activity when knocked out. These experiments together will allow us to compile a list of potential shared and cell type-specific mechanisms that regulate ER activity in both breast and endometrial cancer cells.
Results, Conclusions, and Discussions: From our computational analysis, we have identified a few non-cell type-specific factors that predict whether an ERE will bind ER. These include the sequence of the ERE, which predicts its relative affinity for ER, the activity of RNA polymerase II, and chromatin accessibility. Factors found to be important in both cell lines are circled in green in the attached figure. We also identified factors that were only important in predicting ER binding in one cell line. The presence of the transcription factors JUN and FOXA1 (circled in orange) were found to be more important in breast cancer cells, and the transcription factors MAX, TAF1, SIN3A, and LSD1 along with the histone modification H3K4 monomethylation (circled in blue) were found to be more important in our endometrial cancer cells. For predicting whether a bound ERE would affect transcription, the two cell lines were very similar, with RNA polymerase II activity and 3D genomic interactions, measured via Hi-ChIP loops, being important in both. Because there are so few differences in which factors predict whether a bound ERE will impact expression, it is likely that the differences in ER activity between cell types arises during binding site selection, and that once ER has bound to the DNA, the mechanisms controlling its transcriptional effect are very similar between cell types. As the above results were obtained using solely computational methods, we hope to be able to verify them using our CRISPR screening technique for identifying potential cofactors of ER. We expect that transcription factors found to be important in predicting an ERE’s ability to bind ER and affect gene expression will decrease ER binding and, therefore, decrease the fluorescence of the reporter cell lines. The combination of the results from the CRISPR screen and Boruta analyses will allow us to plan further experiments investigating the role of cofactors, chromatin modifications, and other aspects of the genomic environment in regulating ER activity. These experiments will elucidate the differences and similarities between ER control in breast and endometrial cancer and help pave the way for translating the therapeutic successes in breast cancer to endometrial cancer.
