This project is based on the results reported in Bradford et al., 2016, “Whole transcriptome profiling of patient-derived xenograft models as a tool to identify both tumors and stromal specific biomarkers”. The tumor stroma, or the microenvironment around a tumor, is a complex system made up of endothelial cells, fibroblasts, immune cells and more. Its interactions with the tumor can enable many hallmarks of cancer such as resisting cell death and metastasis. This project’s approach focused on comparing several different breast cancer types using RNA-seq and machine learning methods, and included 79 PDX mouse models with human primary tumors. PDX models are human tumor samples implanted into immune-deficient mice. The T-BioInfo platform was used to perform transcriptomic analysis of raw data sequences and a number of machine learning methods.
The identification of tumor-stroma crosstalk, or the complex cell signaling that occurs between tumor cells and its outside microenvironment, is a challenging problem. The experimental study performed by Bradford et al. provided the data which allowed an essential step forward in this direction. However, the analysis presented by Bradford et al. was limited to a one-to-one correlations study. Furthermore, the study was compromised by uncorrected batch effect. Subsequently, alternative analysis of experimental data provided deeper insight into the problem and identified new biologically meaningful group-wise associations between tumor and stroma genes. A group of tumors were identified which appeared to be enriched by immune processes, and it was hypothesized that they were lymphomas that were associated with Epstein-Barr virus.This study is the first comprehensive analysis across PDX models that focused on identifying the specific stromal cell type, investigated the relationship between human tumor and mouse stroma, and identified specific biomarkers for both tumor and stroma.