Aligning Single-Cell Developmental and Reprogramming Trajectories Identifies Molecular Determinants of Myogenic Reprogramming Outcome

Abstract

Cellular reprogramming through manipulation of defined factors holds great promise for large-scale production of cell types needed for use in therapy and for revealing principles of gene regulation. However, most reprogramming systems are inefficient, converting only a fraction of cells to the desired state. Here, we analyze MYOD-mediated reprogramming of human fibroblasts to myotubes, a well-characterized model system for direct conversion by defined factors, at pseudotemporal resolution using single-cell RNA-seq. To expose barriers to efficient conversion, we introduce a novel analytic technique, trajectory alignment, which enables quantitative comparison of gene expression kinetics across two biological processes. Reprogrammed cells navigate a trajectory with branch points that correspond to two alternative decision points, with cells that select incorrect branches terminating at aberrant or incomplete reprogramming outcomes. Analysis of these branch points revealed insulin and BMP signaling as crucial molecular determinants of reprogramming. Single-cell trajectory alignment enables rigorous quantitative comparisons between biological trajectories found in diverse processes in development, reprogramming, and other contexts.

Epigenetic adaptation prolongs photoreceptor survival during retinal degeneration

“…We used the Monocle 2 package (ver.2.6.1, http://cole-trapnell-lab.github.io/monocle-release/) to generate pseudo-time trajectories for 1, 3, and 6 month old rod cells from Spata7 mutants and wildtype animals which were extracted from the integrated analysis to discover developmental transitions (Cacchiarelli et al, 2018). We excluded low-quality cells using the detectGenes function of Monocle2 with min_expr =0.1.…”

Alignment of time-course single-cell RNA-seq data with CAPITAL

“…Human fibroblast MyoD-mediated (hFib-MyoD) reprogramming data [9] were obtained from Gene Expression Omnibus (GEO) database (GSE105211). 658 cells were filtered in a way that cells with at most 200 genes expressed were removed according to Scanpy's instructions, and outliers revealed with uniform manifold approximation and projection (UMAP) [19] were also removed, resulting in 538 cells.…”

“…Human skeletal muscle myoblast (HSMM) differentiation data [20] were available at GEO (GSE52529). 372 cells were filtered to remove unrelated cells on the basis of instructions in the literature [9] , leading to 111 cells. Genes were filtered in the same way as described above, but the top 120 highly variable genes were used.…”

“…To test how CAPITAL can compare two pseudotime trajectories with potential branchings generated from real time-course data, we used two public scRNA-seq data on human fibroblast MyoD-mediated (hFib-MyoD) reprogramming [9] and human skeletal muscle myoblast (HSMM) differentiation [20], which was first used to quantitatively compare these two models with the early trajectory alignment method [9]. In the following, we will refer to these reprogramming and differentiation data as hFib-MyoD and HSMM, respectively.…”

“…Recently, computational methods for aligning pseudotime trajectories across different experimental conditions have been proposed [8,9]. They aim to align two linear trajectories with dynamic time warping [10,11], which is an analogy to the classical problem of pairwise sequence alignment, but differs in that multiple cells in one condition may be matched with one cell in the counterpart at a time point.…”