Abstract

It is now well appreciated that intra-tumor heterogeneity is of critical importance. There is remarkable molecular variability within populations of tumor cells, driven by both genetic and epigenetic variation. We address the challenge of identifying and characterizing tumor sub-populations through a combined experimental and computational approach. We employ mass cytometry, which accurately measures the expression and phosphorylation states of more than forty proteins in thousands of single cells, including surface proteins and signaling molecules. The data is then analyzed with our vi-SNE dimensionality reduction algoritm (based on t-SNE) that maps 40 dimensions down to two, revealing the “shape” of the tumor and identifying subpopulations within. Application of vi-SNE to healthy immune-cells automatically separates cells based on their known immune subtypes, providing confidence in our approach. We applied our approach to Acute Myeloid Leukemia (AML) identifying distinct subpopulations that differ in surface markers, signaling and drug response. A striking signal in this data is dysregulated development trajectories in these cancers, hence we set to map normal development on healthy bone marrow, where we can observethe full gamut of developmental stages from progenitors to mature B-cells. We developed a graph based trajectory algorithm (wanderlust) that can trace a continuous progression from the hematopoietic stem cells, through the progenitor cells, to the final, committed B-cells. Our derived map of healthy B-cell development revealed the order and timing of developmental events at unprecedented resolution. Both algorithmic details and biological findings will be discussed.