Single-cell and spatial transcriptomics help to unlock our understanding of the subtleties of cellular diversity

Differences in the transcriptomic behaviour of individual cells are not visible when heterogeneous cell populations are analysed in bulk. If the contents of a cell are labelled before analysis in such a way that molecules from each cell are distinct from one another then it is possible to compare the gene-expression level of single cells. One way of achieving this is to use water-in-oil droplets. Single cells are encapsulated in separate droplets along with a bead that is coated with reverse transcription (RT) primers. The primers surrounding any one bead contain the same cell-barcode sequence, and cell lysis and the RT reaction take place within the droplets. The result is that all cDNAs derived from the same cell share a cell barcode, and cDNAs from different cells have different barcodes (Fig. 1a). Following RT and strand switching, all cDNAs can be pooled and amplified, before attachment of sequencing adapters (Fig. 1b). During data analysis, cell barcodes are first identified by a shortlisting and error-correction process, and this label is retained throughout subsequent analyses (Fig. 1c). A similar labelling concept is used for spatial transcriptomics. Here the aim is to interpret the transcriptional activity of cells from a tissue slice in the context of their 2D location within the slice. A tissue slice is positioned over an array of RT primers, where oligos within each spot on the array have a different cell barcode. After cell permeabilisation, RT and strand-switching takes place in situ. The expression level of genes can be compared to the histological features of the slice (Fig. 1d)

PCR artefacts are frequently produced during amplification of the barcoded single-cell cDNAs, and this limits the proportion of full-length transcript reads, to around 50% of the total number. The major PCR artefact consists of a truncated cDNA flanked by copies of the strand-switching oligo (Fig. 2a). It is possible to deplete these strands by biotin capture, which results in a far higher proportion of full-length reads per run: approximately 78% (Fig. 2b). Typical yields from a PromethION (TM) Flow Cell are shown. Although the protocol is still under optimisation, it is currently typical to generate around 60 million full-length reads per PromethION Flow Cell (Fig. 2c). By removing this major library artefact, we see the correlation between our single-cell expression levels and those from a short-read dataset increase substantially (Fig. 2d).

We enriched 6,515 B cells from an individual who had recently been vaccinated against influenza and prepared sequencing libraries using the method shown in Figs. 1a) and 1b). We clustered transcripts from each cell based on the expression levels of specific transcripts using Seurat (Fig. 3a). Cells expressing full-length antibody transcripts are present in all clusters other than cluster 9. The darker the colour, the more antibody transcripts produced by that cell. Fig. 3b shows the top 100 antibody-expressing cells by percentage, calculated as the sum of unique transcripts in each group against the total UMI count per cell. Antibody transcripts correspond to those of immunoglobulin heavy-chain (IGH), kappa light-chain (IGK) and lambda light-chain (IGL) transcripts. A mitochondrial gene expression of <5% usually indicates a healthy cell.