Abstract We, and others, have recently shown that the vast majority of primary tumors are mosaics of clonal populations of varying sizes, different genetic makeup and distinct phenotypes. If subsets of these clones evolve the ability to migrate from the primary tumor and to survive in blood or lymphatic circulation, these clones can seed distant metastasis. For this study, we have two goals. First, we identify clones with metastatic phenotypes and characterize the somatic mutations that distinguish them from non-metastatic clones. Second, we use these mutation signatures to learn to recognize metastatic clones and to calculate the likelihood that a primary tumor will metastasize. The identification of clones with metastatic phenotypes among heterogeneous tumor populations has so far been limited due to the availability of only single samples from tumors. We overcome this limitation using our previously published algorithm, EXPANDS, to identify clones present at >10% cell frequency within single tumor samples across eight different types of carcinomas. We use TCGA's exome-sequencing data to characterize the size and genetic content of clones in 453 primary and 23 metastatic tumors. To quantify the metastatic potential of clones, we compare clone size between primary and metastatic tumors. Next, we model the metastatic potential of clones as a function of their specific point mutations and copy number variations and use principal component analysis to select metastasis gene candidates. Finally, we calculate the likelihood that a primary tumor will metastasize, from the number and size of primary tumor clones with high metastatic potential, further referred to as metastatic clones. We validate the prognostic significance of metastatic clone presence in two independent exome-sequencing datasets: a cross-sectional cohort, consisting of 683 primary tumors and a longitudinal cohort, consisting of six matched primary and metastatic tumors from three patients. We identify 102 metastatic driver gene candidates mutated in clonal populations that demonstrate significantly larger expansion in metastatic as compared to primary tumors. These candidates were enriched for genes associated with cell junction, sympathetic nervous system development, cell adhesion and fibronectin function. Metastatic clones were larger and more frequently observed within stage IV primary tumors as compared to stage I-III primary tumors (p-value = 3E-4). Validation in the independent dataset confirmed increased prevalence and penetrance of metastatic clones in the primary tumors of patients who had metastasis compared non-metastatic patients (p-value = 2E-3). In a univariate Cox model, the presence of metastatic clones was a significant risk factor, independent of cancer type (p-value 1E-2). Overall, we identify gene candidates that may be responsible for the metastatic phenotype of certain clones and the potential utility of clonal quantitation as a biomarker of metastatic progression. Citation Format: Noemi Andor, Hanlee P. Ji. Primary tumor subclones carry somatic mutation signatures of metastasis. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 5164. doi:10.1158/1538-7445.AM2015-5164