Systematic Comparison of High-throughput Single-Cell and Single-Nucleus Transcriptomes during Cardiomyocyte Differentiation

Abstract

A comprehensive reference map of all cell types in the human body is necessary for improving our understanding of fundamental biological processes and in diagnosing and treating disease. High-throughput single-cell RNA sequencing techniques have emerged as powerful tools to identify and characterize cell types in complex and heterogeneous tissues. However, extracting intact cells from tissues and organs is often technically challenging or impossible, for example in heart or brain tissue. Single-nucleus RNA sequencing provides an alternative way to obtain transcriptome profiles of such tissues. To systematically assess the differences between high-throughput single-cell and single-nuclei RNA-seq approaches, we compared Drop-seq and DroNc-seq, two microfluidic-based 3' RNA capture technologies that profile total cellular and nuclear RNA, respectively, during a time course experiment of human induced pluripotent stem cells (iPSCs) differentiating into cardiomyocytes. Clustering of time-series transcriptomes from Drop-seq and DroNc-seq revealed six distinct cell types, five of which were found in both techniques. Furthermore, single-cell trajectories reconstructed from both techniques reproduced expected differentiation dynamics. We then applied DroNc-seq to postmortem heart tissue to test its performance on heterogeneous human tissue samples. Our data confirm that DroNc-seq yields similar results to Drop-seq on matched samples and can be successfully used to generate reference maps for the human cell atlas.

Figure 1

Experimental design and preliminary data analyses. (A) Two cell lines of iPSCs differentiating into CMs over a 15-day time period underwent mRNA sequencing with Drop-seq and DroNc-seq. (B) Boxplots showing the distribution of number of genes in each day and cell line for Drop-seq (top) and DroNc-seq (bottom). (C) Number of cells present after applying quality control cut-offs. (D) Percentage of counts for the top 15 genes in Drop-seq (left) and DroNc-seq (right).

Figure 2

(A) Distribution of reads across the genome in Drop-seq and DroNc-seq. (B) Incorporating intronic reads in quantifying gene expression increases each cell’s gene detection rate by ~1.5X on average for DroNc-seq, enabling detection of more genes per cell, compared with using exon reads only. (C) Mesoderm and cardiac genes with expression detected when incorporating intronic reads. (D) Differential expression analysis between methods, days, and cell lines. Genes with adjusted p-value < 0.05 and log-fold-change >4 were kept. (E) Proportion of differentially expressed genes (DEGs) between Drop-seq and DroNc-seq associated with different gene categories.

Figure 3

Cell type and single-cell trajectory analysis. (A,B) Clustering results visualized with UMAP and colored by inferred cell type for Drop-seq and DroNc-seq. (C,D) Expression of marker genes overlaid on UMAP plots from A and B for Drop-seq and DroNc-seq. (E) Pearson correlation of DroNc-seq and Drop-seq pseudo-bulk against bulk RNA-seq from iPSCs (n = 18), iPSC-Cardiomyocytes (n = 51), and primary heart tissue (n = 22). (F,G) Distribution of cell types per time-point in Drop-seq and DroNc-seq, respectively. (H,I) Inferred trajectories using Monocle with color representing inferred cell types. A total of 3500 cells were used for the trajectory corresponding to 700 per time-point.

Figure 4

Application of DroNc-seq on human heart tissue. (A) Cell type analysis visualized with UMAP. (B) Distribution of marker genes identified with differential expression analysis. All genes listed have p-values < 10⁻²⁹. (C) Pearson correlation of primary heart pseudo-bulk against bulk RNA-seq from iPSCs (n = 18), iPSC-Cardiomyocytes (n = 51), and primary heart tissue (n = 22). (D) Bi-clustering on Pearson correlation values of primary heart nuclei with nuclei from iPSCs and iPSC-derived cardiomyocytes.