Massively parallel digital transcriptional profiling of single cells

Abstract

Characterizing the transcriptome of individual cells is fundamental to understanding complex biological systems. We describe a droplet-based system that enables 3' mRNA counting of tens of thousands of single cells per sample. Cell encapsulation, of up to 8 samples at a time, takes place in ∼6 min, with ∼50% cell capture efficiency. To demonstrate the system's technical performance, we collected transcriptome data from ∼250k single cells across 29 samples. We validated the sensitivity of the system and its ability to detect rare populations using cell lines and synthetic RNAs. We profiled 68k peripheral blood mononuclear cells to demonstrate the system's ability to characterize large immune populations. Finally, we used sequence variation in the transcriptome data to determine host and donor chimerism at single-cell resolution from bone marrow mononuclear cells isolated from transplant patients.

Figure 1. GemCode single-cell technology enables 3′ profiling of RNAs from thousands of single cells simultaneously.

(a) scRNA-seq workflow on GemCode technology platform. Cells were combined with reagents in one channel of a microfluidic chip, and gel beads from another channel to form GEMs. RT takes place inside each GEM, after which cDNAs are pooled for amplification and library construction in bulk. (b) Gel beads loaded with primers and barcoded oligonucleotides are first mixed with cells and reagents, and subsequently mixed with oil-surfactant solution at a microfluidic junction. Single-cell GEMs are collected in the GEM outlet. (c) Percentage of GEMs containing 0 gel bead (N=0), 1 gel bead (N=1) and >1 gel bead (N>1). Data include five independent runs from multiple chip and gel bead lots over >70k GEMs for each run, n=5, mean±s.e.m. (d) Gel beads contain barcoded oligonucleotides consisting of Illumina adapters, 10x barcodes, UMIs and oligo dTs, which prime RT of polyadenylated RNAs. (e) Finished library molecules consist of Illumina adapters and sample indices, allowing pooling and sequencing of multiple libraries on a next-generation short read sequencer. (f) CellRanger pipeline workflow. Gene-barcode matrix (highlighted in green) is an output of the pipeline.

Figure 2. Demonstration of technical performance of GemCode single-cell technology with cell lines and ERCC.

(a) Scatter plot of human and mouse UMI counts detected in a mixture of 293T and 3T3 cells. Cell barcodes containing primarily mouse reads are colored in cyan and termed ‘Mouse only'; cell barcodes with primarily human reads are colored in red and termed ‘Human only'; and cell barcodes with significant mouse and human reads are coloured in grey and termed ‘Human:Mouse'. A multiplet rate of 1.6% was inferred. Median number of genes (b) and UMI counts (c) detected per cell in a mixture of 293T (red) and 3T3 (cyan) cells at different raw reads per cell. Data from three independent experiments were included, mean±s.e.m. (d) Mean observed UMI counts for each ERCC molecule is compared with expected number of ERCC molecules per GEM. A straight line was fitted to summarize the relationship. (e) Principal component analysis was performed on normalized scRNA-seq data of Jurkat and 293T cells mixed at four different ratios (100% 293T, 100% Jurkat, 50:50 293T:Jurkat and 1:99 293T and Jurkat). PC1 and PC3 are plotted, and each cell is colored by the normalized expression of CD3D. (f) SNV analysis was performed, and 293T- and Jurkat-enriched SNVs were plotted for each sample. A 3.1% multiplet rate was inferred from the 50:50 293T: Jurkat sample.

Figure 3. Distinct populations can be detected in fresh 68k PBMCs.

(a) Distribution of number of genes (left) and UMI counts (right) detected per 68k PBMCs. (b) tSNE projection of 68k PBMCs, where each cell is grouped into one of the 10 clusters (distinguished by their colours). Cluster number is indicated, with the percentage of cells in each cluster noted within parentheses. (c) Normalized expression (centred) of the top variable genes (rows) from each of 10 clusters (columns) is shown in a heatmap. Numbers at the top indicate cluster number in (b), with connecting lines indicating the hierarchical relationship between clusters. Representative markers from each cluster are shown on the right, and an inferred cluster assignment is shown on the left. (d–i) tSNE projection of 68k PBMCs, with each cell coloured based on their normalized expression of CD3D, CD8A, NKG7, FCER1A, CD16 and S100A8. UMI normalization was performed by first dividing UMI counts by the total UMI counts in each cell, followed by multiplication with the median of the total UMI counts across cells. Then, we took the natural log of the UMI counts. Finally, each gene was normalized such that the mean signal for each gene is 0, and standard deviation is 1. (j) tSNE projection of 68k PBMCs, with each cell coloured based on their correlation-based assignment to a purified subpopulation of PBMCs. Subclusters within T cells are marked by dashed polygons. NK, natural killer cells; reg T, regulatory T cells.

Figure 4. Genotype analysis of in silico and in vitro mixing of PBMCs.

(a) Sensitivity versus percentage of minor population, where sensitivity is evaluated against the true labelling of in silico mixed PBMCs from Donors B and C. Red line indicates that the major population comes from Donor B PBMCs. Blue line indicates that the major population comes from Donor C PBMCs. (b) Positive predictive value (PPV) versus percentage of minor population, where PPV is evaluated against the true labelling of in silico mixed PBMCs from Donors B and C. Red line indicates that the major population comes from Donor B cells. Blue line indicates that the major population comes from Donor C cells. (c) Called mix fraction versus actual mix fraction in in silico mixing of PBMCs from Donors B and C. Fifty per cent actual mix fraction is correctly called, but omitted from the plot so that the rest of the ratios can be clearly displayed.

Figure 5. Genotype and single-cell expression analysis of transplant BMMCs.

(a) tSNE projection of scRNA-seq data from a healthy control, AML027 pre- and post-transplant samples (post-transplant sample is separated into host and donor) and AML035 pre- and post-transplant samples. tSNE projection was also performed on a second healthy control, but the plot is not included here as it is very similar to that of the first healthy control. Each cell is coloured by their classification, which is labelled next to the cell clusters. (b) Proportion of subpopulations in each sample.