Abstract: Introduction & Objectives: Prostate cancer is heterogeneous and multifocal. Although we know that metastatic spread is the most common cause of death, little is known about the clonal origins of micro-metastatic disease. Conventional approaches to establishing clonal phylogenetic hierarchies rely on genomic (DNA-level) information. However, transcriptomic data (RNA-level) is often more informative in determining cell function and is a better tool for spatial analysis. We asked whether transcriptomic data could also be used to generate sample trees that were concordant with phylogenetic trees generated from genomic data, and whether transcriptomic phylogenies could aid in the identification of tumor-initiating- clones. Materials & Methods: We obtained paired genomic and transcriptomic data, derived from the same regions, from primary PCs and PC metastases as reported in previous publications. We repeated previously reported analysis for genomic data, and then separately generated transcriptomic sample trees. In brief, by discretizing continuous transcriptomic data, one can generate a transcriptomic data-matrix, which is comparable to traditional inputs for phylogenetic analyses. This data can then be used as inputs for standard phylogenetic approaches. Results: We analyzed 37 samples from 8 men across 3 publications and found that transcriptomic phylogenies were largely concordant with genomic phylogenies. We were able to infer the distinct clonal lineages of differing regions giving rise to higher grade disease in primary PC patients (Figure 1). Additionally, transcriptomic phylogenies were able to decode the complex pattern of metastatic spread of distant PC metastases. Conclusions: To our knowledge, this is the first report of attempting to model tumor transcriptomic phylogenies. Our work demonstrates that transcriptomic data can be modeled in a similar manner to conventional phylogenetic trees to derive new biological information and relationships in tumors. This methodology could be used in any setting with multifocal or multi-regional transcriptomic data. We will now use this approach to interrogate spatial transcriptomic data at almost single cell resolution.