Single-cell transcriptome conservation in cryopreserved cells and tissues

Abstract

A variety of single-cell RNA preparation procedures have been described. So far, protocols require fresh material, which hinders complex study designs. We describe a sample preservation method that maintains transcripts in viable single cells, allowing one to disconnect time and place of sampling from subsequent processing steps. We sequence single-cell transcriptomes from >1000 fresh and cryopreserved cells using 3'-end and full-length RNA preparation methods. Our results confirm that the conservation process did not alter transcriptional profiles. This substantially broadens the scope of applications in single-cell transcriptomics and could lead to a paradigm shift in future study designs.

Keywords: Conservation; Cryopreservation; MARS-Seq; PBMC; PDOX; Patient-derived orthotopic xenograft; Peripheral blood mononuclear cells; RNA sequencing; Single-cell genomics; Smart-seq2; Transcriptomics.

Fig. 1

Comparative analyses of MARS-Seq-derived single-cell transcriptome data from fresh (red) and cryopreserved (–80 °C: blue; liquid nitrogen: green) HEK293 cells. a Mapping distribution of sequencing reads after MARS-seq library preparation. Each line represents a single cell and transcript sizes are scaled from 0 to 100. b Cumulative gene counts split by fresh and cryopreserved cells and analyzed using randomly sampled cells (average of 100 permutations). c, d Comparative analysis of the number of sequencing reads and detected transcripts (c) or genes (d) per cell using a linear model. The slope of the regression line was calculated separately for fresh and cryopreserved cells. e, f Gene expression profile variances between fresh and cryopreserved cells displayed as principal component analysis (PCA, e) or as t-distributed stochastic neighbor embedding (t-SNE) representation (f) using the 100 most variable genes

Fig. 2

Correlating MARS-Seq-derived single-cell transcriptomes from fresh (red) and cryopreserved (blue) HEK293 cells identify subpopulations. a Pearson’s correlation analysis between 20 randomly selected fresh and cryopreserved cells displaying the correlation coefficient (r²). b Distribution of Pearson’s correlation coefficients (r²) within and between processing conditions. The median coefficients are indicated. c Pearson’s correlation analysis between 20 randomly selected fresh and cryopreserved HEK293 or K562 cells displaying the correlation coefficient (r²). d–f Linear regression model comparing average gene expression levels of (d) expressed, (e) cell cycle (G2/M checkpoint, and (f) apoptosis genes. The coefficient of determination (r²) is indicated. g Hierarchical clustering of single cells based on transcriptional programs (defined by Gene Ontology) and correlating gene sets [21]. Transcriptional programs and gene clusters are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). Cells are assigned to condition (fresh: red; cryopreserved: blue) and clusters. h A t-SNE representation of similarities between cells using previous defined distances and cluster identity (as in g). Conditions are indicated (fresh: circle; cryopreserved: triangle). i Hierarchical cluster of single cells (as in g) displaying the 25 most variable cell cycle genes (G2/M checkpoint). Expression levels of the cell cycle signature are summarized (first panel; high: orange, low: green) and conditions (second panel; fresh: red; cryopreserved: blue) and clusters are indicated

Fig. 3

Comparative analyses of Smart-seq2-derived single-cell transcriptomes from fresh (red) and cryopreserved (blue) HEK293 cells. a Sequencing read distribution following RNA library preparation of full-length transcripts. Each line represents a single cell and transcript sizes are scaled from 0 to 100. b Cumulative gene counts split by fresh and cryopreserved cells and analyzed using randomly sampled cells (average of 100 permutations). c, d Gene expression variances of single cells displayed as PCA (c) or t-SNE representation (d) using the 100 most variable genes. e Pearson’s correlation analysis between 20 randomly selected fresh and cryopreserved cells displaying the correlation coefficient (r²). f Distribution of Pearson’s correlation coefficients (r²) within and between processing conditions. The median coefficients are indicated. g Hierarchical clustering of single cells based on transcriptional programs (defined by Gene Ontology) and correlating gene sets [21]. Transcriptional programs and gene clusters are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). Cells are assigned to conditions (fresh: red; cryopreserved: blue) and clusters. h A t-SNE representation of similarities between cells using previous defined distances and cluster identities (as in g). Conditions are indicated (fresh: circle; cryopreserved: triangle). i Hierarchical clustering (as in g) displaying the 25 most variable cell cycle genes (G2/M checkpoint). Expression levels of the cell cycle signature are summarized (first panel; high: orange, low: green) and conditions (second panel; fresh: red; cryopreserved: blue) and clusters are indicated

Fig. 4

Comparative analyses of Smart-seq2-derived single-cell transcriptomes from fresh (red) and cryopreserved (blue) K562 cells. a Sequencing read distribution following library preparation of full-length transcripts. Each line represents a single cell and transcript sizes are scaled from 0 to 100. b Cumulative gene counts split by fresh and cryopreserved cells and analyzed using randomly sampled cells (average of 100 permutations). c, d Gene expression variances between single cells displayed as PCA (c) or t-SNE representation (d) using the 100 most variable genes. e Pearson’s correlation analysis between 20 randomly selected fresh and cryopreserved cells displaying the correlation coefficient (r²). f Distribution of Pearson’s correlation coefficients (r²) within and between processing conditions. The median coefficients are indicated. g Hierarchical clustering of single cells based on transcriptional programs (defined by Gene Ontology) and correlating gene sets [21]. Transcriptional programs and gene clusters are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). Cells are assigned to conditions (fresh: red; cryopreserved: blue) and clusters. h A t-SNE representation of similarities between cells using previous defined distances and cluster identities (as in g). Conditions are indicated (fresh: circle; cryopreserved: triangle). i Hierarchical clustering (as in g) displaying the 25 most variable cell cycle genes (G2/M checkpoint). Expression levels of the cell cycle signature are summarized (first panel; high: orange, low: green) and conditions (second panel; fresh: red; cryopreserved: blue) and clusters are indicated

Fig. 5

Correlating single-cell transcriptome data from fresh (red) and cryopreserved (blue) samples determines cell subtypes in PBMC. a, b Comparative analysis of the number of sequencing reads and detected transcripts (a) or genes (b) per cell using a linear model. The slope of the regression line was calculated separately for fresh and cryopreserved cells. c Cumulative gene counts split by fresh and cryopreserved cells and analyzed using randomly sampled cells (average of 100 permutations). d Linear regression model comparing average gene expression levels of expressed genes. The coefficient of determination (r²) is indicated. e Gene expression variances displayed as t-SNE representation using the 100 most variable genes. f Hierarchical clustering of single cells based on transcriptional programs (see “Material and methods”) and correlating gene sets [21]. Transcriptional programs and gene clusters are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). Cells are assigned to condition (fresh: red; cryopreserved: blue) and clusters. g A t-SNE representation of similarities between cells using distances and cluster identities (as in f). Conditions are indicated (fresh: circle; cryopreserved: triangle). Cell types were annotated based on marker gene expression (BC B-cells, CytoTC cytotoxic T-cells, MemTC memory T-cells, Myd myeloid cells). h Hierarchical clustering of single cells (as in f). Displayed are the expression levels of the 25 most variable genes implicated in cluster formation. Cells are assigned to conditions (first panel: fresh: red; cryopreserved: blue) and clusters

Fig. 6

Single-cell transcriptome data from fresh (red) and cryopreserved (blue) mouse colon cells. a, b Comparative analysis of the number of sequencing reads and detected transcripts (a) or genes (b) using a linear model. The slope of the regression line was calculated separately for fresh and cryopreserved cells. c Cumulative gene counts split by fresh and cryopreserved cells and analyzed using randomly sampled cells (average of 100 permutations). d Linear regression model comparing average gene expression levels of expressed genes. The coefficient of determination (r²) is indicated. e Gene expression variances displayed as t-SNE representation using the 100 most variable genes. f Hierarchical clustering of single cells based on transcriptional programs (defined by Gene Ontology) and correlating gene sets [21]. Transcriptional programs and gene clusters are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). Cells are assigned to condition (fresh: red; cryopreserved: blue) and clusters. g A t-SNE representation of similarities between cells using distances and cluster identities (as in f). Conditions are indicated (fresh: circle; cryopreserved: triangle). Cell types were annotated based on marker gene expression [9] (TA transit amplifying, ECpr enterocytes precursors, EC enterocytes, SC secretory cells). h Hierarchical clustering of single cells (as in f). Displayed are the expression levels of the 25 most variable genes implicated in cluster formation. Cells are assigned to condition (first panel: fresh: red; cryopreserved: blue) and clusters

Fig. 7

Comparative analyses of single-cell transcriptome data from fresh (red) and cryopreserved (blue) patient-derived orthotopic ovarian tumor xenograft cells. a, b Comparative analysis of the number of sequencing reads and detected transcripts (a) or genes (b) using a linear model. The slope of the regression line was calculated separately for fresh and cryopreserved cells. c Gene expression variances displayed as t-SNE representation using the 100 most variable genes. d Linear regression model comparing average gene expression levels of expressed genes. The coefficient of determination (r²) is indicated. e Hierarchical clustering of single cells based on transcriptional programs (defined by Gene Ontology) and correlating gene sets [21]. Transcriptional programs and gene clusters are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). Cells are assigned to condition (fresh: red; cryopreserved: blue) and clusters. f A t-SNE representation of similarities between cells using distances and cluster identities (as in e). Conditions are indicated (fresh: circle; cryopreserved: triangle). g, h Hierarchical clustering of single cells (as in e). Displayed are the expression levels of the 25 most variable ribosomal genes (g) and genes implicated in cell cycle (G2/M checkpoint, h). Gene set expression levels are summarized (first panel: high: orange; low: green) and cells are assigned to condition (second panel: fresh: red; cryopreserved: blue) and clusters