Dissociation: from tissue to a cell suspension
Single-cell sequencing does not start at the sequencer. It starts by taking a piece of tissue apart into intact, living, separate cells. This step decides which cells make it into the data, and it leaves the bench’s fingerprint on every cell’s transcriptome. The example dataset, pbmc3k, is peripheral blood mononuclear cells (PBMCs) isolated by density-gradient centrifugation, with no enzymatic digestion; almost every solid tissue, however, goes through this step.
Enzymes and force
Cells in a solid tissue are held together by extracellular matrix and cell junctions. A typical protocol digests the matrix at 37 °C with enzymes such as collagenase or trypsin while pipetting, mincing or an automated dissociator adds mechanical force. The suspension is then filtered to remove clumps, red cells are lysed where needed (leftover red cells show up in the data as the haemoglobin fraction), and debris and dead cells are removed.
The process has two side effects:
- Stress transcription. Cells stay alive and keep transcribing during digestion, so immediate-early and heat-shock genes such as FOS, JUN, EGR1 and HSPA1A are induced quickly. What you sequence is a cell after dissociation, not exactly the cell as it was in the tissue.
- Selective loss. Fragile or very large cells (neurons, adipocytes, cardiomyocytes, some epithelial cells) break or get filtered out more easily, while immune cells tolerate the process better. Cell-type proportions in single-cell data are therefore not the proportions in the tissue.
Viability: the mitochondrial signal of broken cells
Dissociation, freeze–thaw and waiting before loading all damage the cell membrane. Once the membrane is breached, cytoplasmic mRNA leaks out, but the mitochondria have membranes of their own and their transcripts stay behind. The cell ends up with fewer UMIs and fewer genes, and a higher share of mitochondrial reads.
The leaked mRNA does not disappear. It floats in the suspension and is later packed into every droplet along with the cells, becoming ambient RNA (the “soup”) that adds the same background profile to all cells.
The viability measured before loading is a forecast of all this: the lower it is, the more cells you will remove in analysis and the heavier the background.
Common pitfalls
- Treating percent.mt as a fixed cut-off. 5–10% is common for PBMCs, but metabolically active cells such as cardiomyocytes or kidney tubule cells run higher at baseline; a single threshold can delete a real cell type.
- Reading cell proportions as tissue composition. Dissociation and cell death are selective; when comparing composition between groups, both groups must go through the same protocol.
- Calling a stress-gene cluster a new subtype. First check whether its top markers are FOS, JUN or HSP genes.