Tracking Pluripotent Stem Cell Differentiation with FLOW-MAP, a Graph-Based, Force-Directed Layout Algorithm for Single-Cell, Time Course Datasets

Identifying the molecular mechanisms that control the progression of cell differentiation and the branch points of cell fate is an important goal for stem cell biology. Molecular analysis at the single-cell level can provide critical insight toward this goal, because stem cell differentiation occurs in complex and heterogeneous environments where the cell type of interest may be rare and/or transitory. High-content single-cell analysis methods such as flow cytometry, mass cytometry, and single-cell RNA-seq have proven useful for identifying and molecularly characterizing rare cell populations, but using this information to reconstruct cell lineage hierarchies remains challenging. The FLOW-MAP algorithm was developed to serve this need by tracking cell populations as they change over time within single-cell datasets, and combining these trajectories into a global network graph that can be used to infer lineage relationships between cell types. First, the algorithm identifies unique cell populations by density-dependent downsampling and hierarchical clustering. Second, the algorithm connects these cell populations across time points into a sparse graph structure. Third, a force-directed layout function is applied to the graph in order to aid visualization and analysis of the cell lineage trajectories. To investigate the molecular mechanisms that drive cell fate decisions during germ layer formation, this FLOW-MAP approach was applied to an in vitro differentiation time course experiment. At Day 0, 2i/LIF-cultured mESCs were dissociated and then cultured as embryoid bodies with rotary suspension in DMEM/FBS medium. At Day 2.5, the embryoid bodies were plated onto gelatin in basal N2B27 medium, with either no supplement (producing ectoderm), BMP4 (producing mesoderm), or Activin/EGF (producing endoderm and epiblast stem cells). Every 24 hours from Day 0 to Day 11, cell samples were collected with rapid dissociation followed by immediate paraformaldehyde fixation. Mass cytometry analyses was then performed to simultaneously measure the following markers at the single-cell level in samples from every timepoint and culture condition: pluripotency markers (Oct4, Nanog, Sox2, Lin28, SSEA1, Klf4, Sall4), germ layer and cell type markers (Nestin, TUJ1, Flk-1, PDGFR-α, Desmin, GATA4, FoxA2, Cdx2, ICAM-1, CD44, CD24, CCR9, CD45, CD41), epithelial vs. mesenchymal status (EpCAM, Vimentin), cell signaling and cell cycle status (phospho-Stat3, β-Catenin, c-Myc, p53, cleaved-Caspase 3, phospho-Rb, Ki67, IdU), global epigenetic status (acetyl-Lysine 9 Histone H3, pan-acetyl Histone H4). Global, time-resolved, FLOW-MAP analysis of these high-dimensional single-cell datasets reveals continuous cellular trajectories from Nanog-positive/CD54-positive “ground-state” pluripotent stem cells to TUJ1-positive neurons (ectoderm), GATA4-positive/FoxA2-negative cardiomyocytes (mesoderm), and GATA4-positive/FoxA2-positive (endoderm) cell types, as well as an Oct4-positive/EpCAM-positive epiblast stem cell-like population. Molecular characterization of the intermediate cell types along these trajectories and analysis of their branching structure reveals two alternative paths for ectoderm formation, and similarity-based entanglement between early mesoderm and endoderm progenitors. In addition to in vitro pluripotent stem cell differentiation, this single-cell time course-based analysis strategy with the FLOW-MAP algorithm can be applied to other dynamic cell systems in vitro and in vivo, including direct reprogramming, embryonic development, immune response, oncogenesis, and drug resistance.