Abstract Driver genes and mutations responsible for prostate cancer tumorigenesis have been identified, but research has not previously quantified the effects of prostate cancer driver mutations, or how the magnitude of those effects vary with cancer stage and epistatic interactions among genes. Using 2699 prostate cancer tumor exomes, genomes, and panel sequences (1648 primary tumors and 1051 metastatic samples), we calculated the cancer effect size (CES)—the each mutation confers to lower-risk (Grade Group 1/2) and higher-risk (Grade Group 3/4/5) tumorigenesis and metastasis. The lower-risk, higher-risk, and metastatic tumors exhibited similar neutral rates at which mutations were introduced into DNA, with these underlying gene mutation rates generally increasing from lower-risk to higher risk to metastatic tumors. However, lower risk, higher risk, and metastatic tumors noticeably differed in the genes and specific somatic variants that were most highly selected for within each group. Therefore, to assess whether differences in natural selection that we observed among low-risk, high-risk, and metastatic tumor types were at least partly due to epistatic interactions, we applied a model of pairwise epistasis that quantified the effect of the presence or absence of a mutation in one gene on the selection for mutations in another gene. We found significant evidence of epistatic effects among driver genes. Our results indicate a very early role for SPOP in the development of prostate cancer: strength of selection for SPOP mutations is tightly associated with the degree to which the mutation ablates contact with the BRD3, and these disruptive mutations in turn increase the selection for mutations in several other tumor suppressors and oncogenes such as RHOA. Previously identified gene-gene interactions such as that between SPOP and PIK3CA were supported by this analysis, and we also identified novel gene-gene interactions such as those between AR and MUC16 occurring during metastasis. In conclusion, we have demonstrated that rates of mutation and strengths of selection on these driver mutations vary dynamically along the trajectories to lower-risk, higher-risk, and metastatic prostate cancer, and that powerful selective epistasis among driver genes provides an explanation for these differences. Because precision therapeutic targeting of somatic variant genes is fundamentally targeting the selective benefit obtained from that variant at the time of treatment, understanding these changes in stage-specific and somatic genotype-specific selective strengths will enable better therapeutic targeting in translational and clinical science. Citation Format: Moein Rajaei, Alexander Yang, Christopher N. Cross, J Nick Fisk, Elizabeth B. Perry, Jeffrey D. Mandell, Stephen G. Gaffney, Takafumi N. Yamaguchi, Julie Livingstone, Vincent L. Cannataro, Peter Humphrey, Jose Costa, Paul C. Boutros, Jeffrey P. Townsend. Persistence and dynamics of mutation, selection, and epistasis during the somatic evolution of low-risk, high-risk, and metastatic prostate cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1627.