Expression analysis, typified by next-generation RNA sequencing (RNA-seq), has given unprecedented insights into a tissue’s transcriptional activities, allowing comparisons between strains, for example, tracking of responses to interventions, or following changes across disease and development. Yet distinctions between individual cells are lost by averaging across the hundreds, thousands, or even millions typically used as input.

Single-cell transcriptome analysis (scRNA-seq) facilitates comparison of the transcriptomes of individual cells. It’s a fairly new technology—the first paper describing its use was published almost ten years ago—but commercial scRNA-seq platforms are increasingly available as are bioinformatics solutions. Here we look at some recent research enabled by scRNA-seq.

Pipette, split, and pool

Genome-wide, single-cell analysis is very much in-demand, says Bosiljka Tasic, associate director, molecular genetics at the Allen Institute for Brain Science. Instead of getting a smoothie as you would with bulk analysis, “you’re actually getting individual ingredients of whatever your system is—that’s the single cells.” And genome-wide analysis, as opposed to techniques such as PCR and in situ hybridization (ISH), provides an unbiased look into what the cell is expressing: “you don’t have to choose which genes you’re analyzing.”

There are a number of platforms and techniques available to prepare single-cell RNA for sequencing. These generally come down to either segregating individual cells in individual wells of a plate, or having droplets that act as reaction chambers for individual cells. Either way the key, explains Tasic, is to separate and barcode cells at some point during the analysis—at either the reverse transcription level or the library preparation level—allowing the RNA sequences to be assigned to the unique cell from whence they came.

Georg Seelig’s team at the University of Washington (UW) developed a technique in which the cells themselves act as reaction chambers, termed split-pool ligation-based transcriptome sequencing (SPLiT-seq). Here, the cells or nuclei are fixed so that the RNA stays trapped, but reagents can be washed in and out. A series of pooling and splitting steps interspersed with reverse transcription and ligation (with barcoded labels), and culminating in lysis and PCR (using barcoded primers), “basically allows you to barcode as many cells as your barcode library allows,” adds Tasic, who collaborated with UW for a recent paper in Science. The library preparation cost is on the order of one cent per cell, and “what’s really powerful is the fact that it’s almost instrumentation-free.”

The scientists looked at nuclei from the brains and spinal cords of two- and eleven-day-old mice. They were able to distinguish more than 100 different cell types on the basis of “gene-expression patterns corresponding to cellular function, regional specificity, and stage of differentiation,” according to the abstract. The data were used to create an atlas of gene expression, complementing other Allen Institutes reference atlases.

The nuclear option

Kun Zhang’s group at the University of California, San Diego, focuses on the single cellome in human archived tissue. “You need to separate [the cells from each other] before you can do all these single-cell analyses,” he says. While it’s difficult to dissociate brain tissue—“you might have very biased representation because some cells are breaking down, whereas other cells are integrated with each other”—extracting intact nuclei from it is relatively straightforward.

They put the extracted nuclei through a modified Drop-seq protocol, termed snDrop-seq, designed to disrupt the nuclear membrane in the micro-droplets with minimal degradation of the RNA. “If you just throw nuclei into a regular Drop-seq or 10X Genomics protocol, it doesn’t work because the membrane won’t break,” Zhang explains. There are currently several ways to accomplish the task, including altering the microfluidic chip so that the membrane will break down under mechanical force. “We actually raised the temperature to break the nuclear membrane.”

They performed both snDrop-seq and a chromatin accessibility assay (transposome hypersensitive-site sequencing, scTHS-seq) they also developed on the same batch of nuclei, from the same brain region. “That allowed us to compare these single-cell profiles at the RNA level versus the chromatin level,” Zhang points out. They were able to reconstruct the epigenetic profiles of fine-grained brain cell types, and used their single-cell multi-omic approach to map risk factors onto specific cell types and ask, for example, what the relative contributions of neurons, microglia, and oligodendrocytes to Alzheimer’s disease, autism spectrum disorder, or schizophrenia are.

Not so smooth

Adam Reid and his colleagues at the Wellcome Sanger Institute wanted to understand genetic control of the malarial lifecycle. By sequencing unsynchronized single cells and arranging the transcriptomes in pseudotime, they found that “there are actually big shifts in development” between the parasite’s stages, the senior staff scientist says. “That wasn’t at all apparent from the bulk RNA seq [of supposedly synchronized cells] because there was some blending at the edges, and that meant that you saw a continuous progression. Whereas we have uncovered a more stepwise progression.”

They used a modified version of the comparatively low-throughput Smart-seq2 protocol, aiming for about 100 cells per stage. “So you get a lot more information about which genes are expressed and how highly expressed they are” than the higher-throughput 10X or Drop-seq platforms, Reid says. The malaria-causing Plasmodium parasites are very small, with very small amounts of RNA, and have an extremely biased genome—about 20% GC content, compared to the mammalian 35-40%— “so the molecular biology tends not to work so well.” But by increasing the number of PCR cycles and trying different enzymes, among other tweaks, “we managed to get it to work quite well.”

The researchers hope that having identified these step changes in development—“we also have hints about what some of the regulators of these step changes”—will pave the way toward intervention with drugs or vaccines.

It’s about time

With its complex instrumentation to purchase or assemble, and physical and bioinformatics workflows to master, scRNA-seq has sometimes been viewed as a specialist’s realm. Oftentimes communication between biologists and computational scientists is “very poor,” recalls Bart Deplancke, group leader of the Swiss Institute of Bioinformatics. While preparing for a single-cell transcriptomic study of adipose tissue with many datasets to process—which led to the discovery of a new cell type responsible for shaping the form of fat tissue—his group found that the type of analyses discussed in bioinformatics community forums were generally inaccessible to their collaborators.

They set out to enable collaboration in which both types of researchers could visualize and process the data in a more intuitive fashion. Enter the Automated Single-cell Analysis Pipeline, a.k.a. ASAP.” Deplancke explains that ASAP was built as a complete, sequential, web-based workflow providing standard tools including filtering, normalization, dimension reduction, clustering, differential expression, and functional enrichment. It can interact with various databases, and display results in 2D or 3D. “For each step we provide a basic tutorial—it will tell you what each of the analysis tools will do to your data.”

He notes that “even the bioinformaticians love it because it’s a very easy way to quickly process and look at the data and then, together with the biologists, visualize it, look at it, formulate some new hypotheses which then can be investigated either experimentally or computationally a little bit further down the road.”