Bioengineers at EPFL have found a way to radically increase the efficiency of single-cell RNA-sequencing, a powerful tool that can “read” the genetic profile of an individual cell.
Single-cell RNA sequencing, or “scRNA-seq” for short, is a technique that allows scientists to study the expression of genes in an individual cell within a mixed population – which is virtually how all cells exist in the body’s tissues. Part of a larger family of “single-cell sequencing” techniques, scRNA-seq involves capturing the RNA of a single cell and, after multiple molecular conversion reactions, sequencing it. Since RNA is the intermediate step from gene (DNA) to protein, it provides an overview about which genes in that particular cell are active and which are not.
Because scRNA-seq captures the activity of all genes in the cell’s genome – thousands of genes at once – it has become the gold standard for defining cell states and phenotypes. This kind of data can reveal rare cell types within a cell population, even types never seen before.
Cost and efficiency
But scRNA-seq isn’t just a tool for basic cell biology; it has been widely adopted in medical and pharmacological research as it is capable of identifying which cells are actively dividing in a tissue, or which are reacting to a particular drug or treatment.
“These single-cell approaches have transformed our ability to resolve cellular properties across systems,” says Professor Bart Deplancke at EPFL’s School of Life Sciences. “The problem is that they are currently tailored toward large cell inputs.”
This isn’t a trivial problem, as scRNA-seq methods require over a thousand cells for a useful measurement. Dr Johannes Bues, a researcher in Deplancke’s group, adds: “This renders them inefficient and costly when processing small, individual samples such as small tissues or patient biopsies, which tends to be resolved by loading bulk samples, yielding confounded mosaic cell population read-outs.”
The DisCo solution
Bues, with Marjan Biočanin and Joern Pezoldt, also in Deplancke’s group, have now developed a new method that allows scRNA-seq to efficiently process samples with fewer cells. Published in Nature Methods, the method is dubbed “DisCo” for “deterministic, mRNA-capture bead and cell co-encapsulation dropleting system”.
Unlike usual single-cell methods that rely on passive cell capture, DisCo uses machine-vision to actively detect cells and capture them in droplets of oil and beads. This approach allows for continuous operation, and also renders scaling and serial processing of cell samples highly cost efficient.
As shown in the study, DisCo features precise particle and cell positioning, and controls droplet sorting through combined machine-vision and multilayer microfluidics. All this allows for continuous processing of low-input single cell suspensions at high capture efficiency (over 70%) at speeds that can reach 350 cells per hour.
Overview and critical feature assessment of the DisCo system
a, Schematic diagram of the DisCo microfluidic device, which contains three inlet channels for cells, beads and oil (shown twice for illustration purposes); two outlets for waste and sample liquids, and several Quake-style microvalves (green boxes): 1, cell valve; 2, bead; 3, dropleting; 4, oil; 5, waste; 6, sample. Particles are detected by a camera and are placed at the Stop point. b, DisCo co-encapsulation process on the DisCo device (red, closed; green, open; light brown, dropleting pressure (partially closed)). c, The co-encapsulation process of two beads as observed on-chip. Dyed liquids were used to examine the liquid interface of the carrier liquids. Channel sections with white squares are 100 μm wide. d, The droplet capture process as observed on-chip. Valves are highlighted according to their actuation state (red, closed; green, open). e, Image of DisCo droplet contents. Cells (blue circles) and beads (red circles) were co-encapsulated and the captured droplets were imaged. Mean bead-size is approximately 30 μm. f, Droplet occupancy of DisCo-processed cells and beads (total encapsulations, n = 1,203). Bars represent the mean, and error bars represent ±s.d. g, Cell capture efficiency and speed for varying cell concentrations (2–20 cells per μl, total encapsulations, n = 1,203). h, DisCo scRNA-seq species separation experiment. HEK 293T and murine IBA cells were processed with the DisCo workflow for scRNA-seq, the barcodes merged and the species separation visualized as a Barnyard plot. i, Comparison of detected transcripts (UMIs) per cell of conventional Drop-seq experiments. UMIs per cell from HEK 293T data for conventional Drop-seq experiments are compared with the HEK 293T DisCo data. Drop-seq datasets were down-sampled to a similar sequencing depth. Box plot elements showing UMI counts per cell represent the following values: center line, median; box limits, upper and lower quartiles; whiskers, 1.5-fold the interquartile range; points, UMIs per cell. j, Total cell processing efficiency of DisCo at low cell inputs. Input cells (HEK 293T) ranging from 74 to 170 were quantified by impedance measurement. Subsequently, all cells were processed with DisCo, sequenced and quality filtered (>500 UMIs). The red line represents 100% efficiency, and samples were colored according to the recovery efficiency after sequencing.
To further showcase DisCo’s unique capabilities, the researchers tested it on the small chemosensory organs of the Drosophila fruit fly, as well as on individual intestinal crypts and organoids. The latter are tiny tissues grown in culture dishes closely resembling actual organs – a field that EPFL has been spearheading for years.
The researchers used DisCo to analyze individual intestinal organoids at different developmental stages. The approach painted a fascinating picture of heterogeneity in the organoids, detecting various distinct organoid subtypes of which some had never been identified before.
“Our work demonstrates the unique ability of DisCo to provide high-resolution snapshots of cellular heterogeneity in small, individual tissues,” says Deplancke.
Availability – The source code for the machine-vision software is available on github (https://github.com/DeplanckeLab/DisCo_source)