Exploration of Cell Development Pathways through High-Dimensional Single Cell Analysis in Trajectory Space

Abstract

High-dimensional single cell profiling coupled with computational modeling is emerging as a powerful tool to elucidate developmental programs directing cell lineages. We introduce tSpace, an algorithm based on the concept of "trajectory space", in which cells are defined by their distance along nearest neighbor pathways to every other cell in a population. Graphical mapping of cells in trajectory space allows unsupervised reconstruction and exploration of complex developmental sequences. Applied to flow and mass cytometry data, the method faithfully reconstructs thymic T cell development and reveals development and trafficking regulation of tonsillar B cells. Applied to the single cell transcriptome of mouse intestine and C. elegans, the method recapitulates development from intestinal stem cells to specialized epithelial phenotypes more faithfully than existing algorithms and orders C. elegans cells concordantly to the associated embryonic time. tSpace profiling of complex populations is well suited for hypothesis generation in developing cell systems.

Keywords: Developmental Biology; In Silico Biology; Systems Biology.

Graphical abstract

Figure 1

tSpace Concept and Application on Thymic T cells: tSpace Reveals Developmental Trajectories and Recovers Expression Patterns of Markers of T Cell Differentiation (A) The schematic example illustrates the concept of trajectory space. The “cells” are marked with the letters A–I, and their developmental sequences are marked with arrows. Matrix of cell-to-cell distances along developmental paths is created (each cell is one unit from its nearest neighbor). Visualization of cell positions in this “trajectory space,” here using PCA, recapitulates the branches. Note that E and H, although similar phenotypically, are most distant in trajectory space, reflecting their developmental pathways. (B) Unsupervised tSpace analysis of thymic mouse thymocytes accurately recapitulates thymic T cell development. (C) t-SNE of thymic T cells defines clusters but not developmental relationships. (D) Isolated trajectory from DN2 precursors to CD4 thymic emigrants. (E) Smoothed expressions of measured markers along isolated trajectory (shown in D) reveals patterns of protein regulation during T cell differentiation. The identities of manually defined cell subsets as well as cell density along the isolated trajectory are shown for reference above the heatmap. DN, double-negative T cells; CD4 emig., CD4⁺ T cell poised emigrants; CD8 emig., CD8⁺ T cell poised emigrants.

Figure 2

tSpace Analysis of B Cell Differentiation in Tonsils and Inter-organ Trajectories with Blood (A) tSpace unravels maturation paths of B cells starting from naive B cells in tonsil throughout GC into memory B cells and plasmablasts (PB). Magenta arrows mark suggested directionalities based on known biology. (B) tPC1 and tPC2 reveal branches and potential developmental relationships in tonsillar B cell maturation. (C) tPC3 and tPC4 reveal branches and potential developmental relationships in tonsillar B cell maturation. Ellipses show 80% confidence intervals for indicated clusters. (D) Blood (BL) PB align as an extension of tonsillar PB trajectories, whereas recirculating blood memory B cells overlap with the major tonsil memory cell clouds. Tonsil B cells are in light gray.

Figure 3

tSpace Analysis of Mouse Small Intestinal Epithelial Cells (A) tSpace separates trajectories to enterocytes, enteroendocrine (EE), Paneth, and goblet cells. CBC and TA subsets were defined by our analysis, as described in Figure S7; other subsets are labeled as in Yan et al. (2017). Shaded rectangle highlights the position of short-lived EE progenitors (slEEP) cells. (B) Isolated enterocyte trajectory. (C) Isolated EE trajectory. (D) Expression patterns of selected genes (Clevers, 2013) (known markers or regulators of intestinal crypt development; expanded gene list in Figure S8A) along the isolated trajectories. (E) Four detected transcription factor modules in early trajectories, identified by comparing gene expression between cells at similar stages in the two trajectories (Transparent Methods): M1 comprises TFs involved in cell cycle and genome integrity expressed in precursor populations (early in the shared trajectory). M2 and M3-M4 differentiate the two lineages and comprise TF's that may determine cell fate or specialization (see text). Cell stage (Transparent Methods) and cell identities defined in this study (Cell type) or in Yan et al. (Orig. labels) are indicated above the heatmap. ND, fully differentiated enterocytes, not used in trajectory alignment. (F) Summary of differences between two branches suggested by gene regulation along the trajectories. Different genes in the transforming growth factor-β and circadian rhythm pathways are expressed in the two lineages (genes in blue above the cartoon). TFs enriched in the EE branch are involved in endocrine secretory cell development, whereas TFs associated with enterocyte commitment include regulators of lipid/cholesterol metabolism. Expression of Dll1 and Sox4 in EE development and Alpi in enterocyte differentiation mark specific progenitor cells located within the +4/+5 position in the intestinal crypt according to the literature (van Es et al., 2012, Gracz et al., 2018, Tetteh et al., 2016); clear peaks are seen in their expression along the trajectories (Figures 3D and 3E) near the TA to differentiated cell transitions, likely representing these specific progenitor populations.

Figure 4

Benchmarking of tSpace Performance Using a Developmentally Timed C. elegans Dataset (A) UMAP of 75 PCs of gene expression matrix, re-creating the UMAP from Packer et al., showing major cell types. (B) tSpace analysis using 1,000 trajectories, showing the first three tPCs and raw embryonal time; inset shows cell types. (C–F) (C) The same tSpace analysis as in (B) visualized in tPC1, tPC3, and tPC4: this combination of tPCs allows separation of all major lineages, and interestingly places pharyngeal intestinal valve (PIV) cells in close proximity to intestine and intestine close to rectal cells, relations that resemble spatial cell positions in C. elegans body: PIV cells link the posterior bulb of the pharynx to the anterior cells of the intestine, and the intestine is eventually connected to the rectum. The first three tPCs and raw embryonal time using (D) 50 trajectories, (E) 100 trajectories, and (F) 500 trajectories. Arrows and ellipsoids mark intestinal cell type. (G) tSpace analysis using 500 trajectories, 3D representation of cell types in tPC1, tPC3, and tPC4. tSpace defines branches for all main lineages.