Tremendous energy and resources have been expended to sequence tumors in an effort to identify drivers of metastasis or the lethal phenotype. We applied genome and transcriptome sequencing to directly address this problem in patient derived prostate tumor xenografts well characterized model systems. Extensive analysis of the tumor contents failed to reveal any mutations or gene expression linked to the metastatic phenotype. However, analysis of the stromal compartment revealed a gene expression signature predictive of metastasis in the xenograft models and in human cohorts. This is particularly interesting because in these patient derived xenografts the human stoma is replaced by mouse stromal cells. The implication is that the human tumors actively educate the mouse stroma to provide a permissive environment for metastasis to occur. A further implication is that tumor focused genome studies may have missed half the story and that therapies targeting only the tumor may be missing the mark. These and other problems with precision oncology will be discussed.