Abstract: Background: Cancers arise through branching evolution. When applying genomic analysis in the clinic we are faced with several basic questions such as what to sample and when. The existence of heterogeneity within a tumour could confound the use of targeted therapies or the reliability of prognostic indicators. Understanding the branching patterns that underlie cancers can therefore help us to address these basic questions. Furthermore, exploring the specific features of mutations and when they occur across space and time may help us to understand what is driving ongoing evolution at cancer relapse and metastasis. Material and Methods: Evolutionary histories were inferred from the analysis of whole genome massively parallel sequencing data generated from triplet samples from 17 breast cancers. Samples for each subject were derived from 1) germline (normal breast tissue or blood), 2) primary tumour and 3) a local relapse or metastatic deposit. Mutational types and mutation signatures within individual branches of the phylogenetic tree were analysed to assess for differences across the different stages of evolution. Relationships between mutation features in each branch and clinical factors such as time to relapse and treatment exposures were assessed. Results: In most cases around 90% of mutations present in the primary tumour were also present in the relapse or metastatic sample. This finding of late branching, indicates that the genomic landscape of the primary tumour is a reasonable surrogate for that of relapse seeding subclones at the time of adjuvant treatment. Relapses after several months or years however, have on average 40% additional mutations compared to the primary tumour or synchronous metastases. This provides strong rationale for re-sampling at relapse. Mutational signature analysis identifies that the mutational forces driving cancer evolution change over time. The contribution from a mutational signature characterized by C to A transversions frequently increases or appears de novo late in disease − this is likely to reflect a change in biology late in disease. APOBEC activity including bursts of kataegis operate at multiple points in evolution but does not show a single pattern in breast cancer. Mutagenic treatments such as radiotherapy can leave detectable imprints in the genome of relpased cancers. Conclusions: Genome-wide genomic analyses using multiple samples has the power to reveal the dynamics and mutational forces at play during individual cancer’s evolutionary histories. Although currently experimental these techniques can help us to rationalize sampling approaches particularly within clinical trials. Extending our focus from simple driver mutation analyses to the mutational forces operating during disease may expose new therapeutic avenues and a potential solution to driver mutation heterogeneity.