Studying temporal dynamics of single cells: expression, lineage and regulatory networks

Abstract

Learning how multicellular organs are developed from single cells to different cell types is a fundamental problem in biology. With the high-throughput scRNA-seq technology, computational methods have been developed to reveal the temporal dynamics of single cells from transcriptomic data, from phenomena on cell trajectories to the underlying mechanism that formed the trajectory. There are several distinct families of computational methods including Trajectory Inference (TI), Lineage Tracing (LT), and Gene Regulatory Network (GRN) Inference which are involved in such studies. This review summarizes these computational approaches which use scRNA-seq data to study cell differentiation and cell fate specification as well as the advantages and limitations of different methods. We further discuss how GRNs can potentially affect cell fate decisions and trajectory structures.

Supplementary information: The online version contains supplementary material available at 10.1007/s12551-023-01090-5.

Keywords: Gene regulatory network inference; Lineage tracing; Single-cell RNA sequencing; Trajectory inference.

Fig. 1

Inferring developmental trajectories from single-cell omics data. a Trajectory inference using a single batch of scRNA-seq data. b Trajectory inference using RNA velocity information. With the input of both single-cell spliced and unspliced RNA counts, RNA velocity can be calculated and can be used to infer the developmental trajectory. c Trajectory inference using scRNA-seq time series data. Given cell-by-gene matrices measured at different time points, the developmental trajectory of cells covering more developmental stages is inferred. d Trajectory inference using single-cell multimodal data such as scRNA-seq and scATAC-seq data. With multiple cell-by-feature (gene expression, chromatin accessibility, protein abundance, etc.) matrices, a joint developmental trajectory that combines the different modalities is inferred

Fig. 2

Lineage reconstruction from CRISPR/Cas9 induced barcodes. Step A The lineage tracing system uses Cas9 proteins to generate double-stranded breaks that result in heritable insertions or deletions (mutations) after repair. Indels are induced at specific target sites of the barcode. Step B At the root, an unedited barcode, together with the Cas9 proteins and guide RNAs, is injected into the starting cell. Throughout generations of cell divisions, the Cas9 protein can bind to the designed barcode and induce mutations that are inherited and accumulated. Step C With the scRNA-seq experiment, the mutated barcodes of the present-time cells (leaf cells on the lineage tree) are sequenced. Step D Inferring the hidden lineage tree topology given the mutated barcodes of the leaf cells using computational methods

Fig. 3

Theoretical analysis of connections between GRN, developmental trajectory, and gene expression. In the GRNs, each node represents a gene (gene names are A, B, C) and the edges denote up-regulation (arrow) and down-regulation (block). For each gene, we use a total expression DE (differential equation) to model the changes of the gene expression, while also considering the regulation effects of promoters based on the GRN. We use a two-phase framework: an initial warm-up Burn Phase that only part of the network is active; a Transcription Phase that all network components are active. From the DE solution of the equations, discrete states can be defined. a A GRN that generates a bifurcation trajectory. Three cell states are defined from the DE solution: $S_1$ where only gene A is highly expressed; $S_2$ where gene A, B are highly expressed and $S_3$ where gene A, C are highly expressed. b A GRN that generates a bifurcation convergence trajectory. Three cell states are defined from the DE solution: $S_1$ where only gene A is highly expressed; $S_2$ where gene B is highly expressed and $S_3$ where gene C is highly expressed