Dissociation-Induced Stress Artifacts in scRNA-Seq

Team Dynamic Matrices | 6 July 2026

Single-cell RNA-seq is meant to capture a transcriptional snapshot of a tissue. The problem is that the tissue has to be turned into a single-cell suspension first, and the cells notice. Within minutes of warm enzymatic digestion they start transcribing a stress program that can look, in the data, like a real cell state. We have seen "activated fibroblast" clusters and "stressed epithelium" subtypes that were artefacts of a 40-minute collagenase step. This post covers what the artefact is, how to see it, and what to do about it.

The mechanism

Dissociation exposes cells to three things they never experience in situ: 37 °C incubation without their normal contacts, mechanical shear from trituration, and loss of adhesion to the matrix. Cells respond to acute stress the way they respond to any acute stress: by rapidly transcribing immediate early genes (FOS, FOSB, JUN, JUNB, EGR1, ATF3) and heat-shock proteins (HSPA1A, HSPA1B, DNAJB1, HSP90AA1). van den Brink et al. (Nature Methods, 2017) showed this in muscle stem cells, with the signature appearing in a subset of cells and creating an artificial cluster. Temperature and duration drive it: most enzymes work best at 37 °C, which is when transcription is most active, and a 20 to 45 minute digestion is enough for the program to accumulate.

Why it is worse than it looks

Two features make the artefact hard to handle after the fact.

It is not uniform. Cell types differ in their sensitivity. O'Flanagan et al. (Genome Biology, 2019) characterised a conserved collagenase-associated stress response across many human tumour types and found it varied by cell type, so the artefact masquerades as biological heterogeneity. A "stressed" subcluster within one lineage may be nothing more than the cells that were dissociated last.

It changes who survives. Denisenko et al. (Genome Biology, 2020) compared dissociation and storage protocols in mouse kidney and found that method changed apparent cell-type proportions, because fragile populations die during warm digestion. Populations that appear rare may simply have poor dissociation survival. Proportions from a single protocol should be treated as protocol-specific, not biological.

Recognising the signature

Score every dataset against a curated stress gene set. Both van den Brink and O'Flanagan published lists; a minimal one is FOS, FOSB, JUN, JUNB, EGR1, ATF3, IER2, HSPA1A, HSPA1B, DNAJB1. Look for a score that is high in a subset of cells, that does not track a known biology, and that correlates with capture time or batch. If a cluster is defined mainly by these genes, it is a handling artefact until proven otherwise. Regressing the score out during clustering is a partial fix; preventing it is better.

Mitigations that work

Cold-active protease. Adam, Potter and Potter (Development, 2017) used a protease from Bacillus licheniformis at 6 °C and dramatically reduced stress gene induction in kidney. O'Flanagan confirmed it in tumours. Transcription at 6 °C is slow enough that the program does not accumulate even over a longer digestion. This is our first recommendation.

Shorten the warm phase. If you must use collagenase or TrypLE at 37 °C, cut every minute you can. Check digestion every 5 minutes and stop early. Keep everything on ice from the moment digestion ends.

Block transcription. Actinomycin D during dissociation prevents new transcription and eliminates the induced signature (Wu et al., JASN, 2019). It has to be validated for your application because it affects all transcription, but for a pure snapshot it is effective.

Skip whole cells: single-nucleus RNA-seq. Isolating nuclei from flash-frozen tissue avoids warm digestion entirely. Nuclear transcriptomes are enriched for pre-mRNA and under-represent cytoplasmic and mitochondrial transcripts, so it is a different measurement, but for stress-sensitive tissues (brain, kidney, muscle) it is often the better one.

Reduce the matrix step. For organoids and 3D cultures, a large fraction of the warm phase is spent dissolving the matrix, not the cells. Basement membrane extract needs 30 to 60 minutes of cold depolymerisation or enzymatic digestion; collagen needs collagenase. A DNA-crosslinked gel dissolves in minutes with a nuclease that does not touch cells, so the enzymatic step can be reserved for cell-cell junctions and kept short. In our organoid work this removes 20 to 40 minutes from the warm phase and is the single largest reduction in stress score we have measured.

A pilot diagnostic worth running

Before a large study, dissociate the same tissue three ways (standard warm, shortened warm, cold protease), run a small capture on each, and compare stress scores and cell-type proportions. It costs one extra day and three small libraries. It tells you which populations you would otherwise have lost and how much of your "biology" was handling.

References

van den Brink SC et al. Single-cell sequencing reveals dissociation-induced gene expression in tissue subpopulations. Nat Methods 14, 935–936 (2017).
O'Flanagan CH et al. Dissociation of solid tumor tissues with cold active protease for single-cell RNA-seq minimizes conserved collagenase-associated stress responses. Genome Biol 20, 210 (2019).
Adam M, Potter AS, Potter SS. Psychrophilic proteases dramatically reduce single-cell RNA-seq artifacts. Development 144, 3625–3632 (2017).
Denisenko E et al. Systematic assessment of tissue dissociation and storage biases in single-cell and single-nucleus RNA-seq workflows. Genome Biol 21, 130 (2020).
Wu H, Kirita Y, Donnelly EL, Humphreys BD. Advantages of single-nucleus over single-cell RNA sequencing of adult kidney. J Am Soc Nephrol 30, 23–32 (2019).
Peng Y-H et al. Dynamic matrices with DNA-encoded viscoelasticity support the development of organoids and stem cells. Nat Nanotechnol (2023).