Abstract: Background: The targeting of new cancer therapeutics will be driven by genomic analysis of an individual’s tumor. It is questionable that standard sampling approaches that utilize slices from single ‘tissue blocks’ or clinical biopsies would always reflect the full spectrum of clinically relevant changes in the breast tumor landscape. Methods: We explore the diversity between multiple samples (n=294) from 49 patient’s primary breast tumors. Using a combination of targeted gene screens (n=285) and/or whole genome sequencing (n=25) mutation catalogues were compiled for each tumor. Results: To assess geographical heterogeneity within primary tumors we obtained multiple spatially separated samples (7-17) from 13 patients’ primary surgical specimens. Somatic point mutation and insertion/deletion analysis identified branching heterogeneity in 11/13 surgical samples (including ER positive, triple negative and HER2+ subtypes). In 4 cases, potentially ‘druggable’ targets such as PIK3CA mutations occur in a single sampled region – indicating frequent sub-clonality, and spatial confinement, of cancer driver mutations. We observe divergent and convergent evolution involving cancer drivers within individual tumors. We exploit the resolution of whole genome sequencing to demonstrate that even when homogeneity of driver mutations is observed, mutational processes such as kataegis continue to operate independently in spatially distinct regions. The analysis is extended to clinical samples: Pre-treatment biopsies and post-surgical blocks from 36 patients who underwent neo-adjuvant chemotherapy were analyzed. Initial findings reveal genomic heterogeneity within cases of both complete response (n=10) and residual disease (n=26), but interestingly, sub-clonal driver mutations so far appear to be limited to those with residual disease. Our ongoing analysis incorporates copy number analysis, and explores the relationship between chemotherapy response and heterogeneity and compares the genomic landscape pre and post neo-adjuvant chemotherapy. Conclusions: Current sampling approaches can be expected to under-report clinically actionable genomic events in a significant proportion of breast cancers.