scFTD-seq: freeze-thaw lysis based, portable approach toward highly distributed single-cell 3' mRNA profiling

Abstract

Cellular barcoding of 3' mRNAs enabled massively parallel profiling of single-cell gene expression and has been implemented in droplet and microwell based platforms. The latter further adds the value for compatibility with low input samples, optical imaging, scalability, and portability. However, cell lysis in microwells remains challenging despite the recently developed sophisticated solutions. Here, we present scFTD-seq, a microchip platform for performing single-cell freeze-thaw lysis directly toward 3' mRNA sequencing. It offers format flexibility with a simplified, widely adoptable workflow that reduces the number of preparation steps and hands-on time, with the quality of data and cost per sample matching that of the state-of-the-art scRNA-seq platforms. Freeze-thaw, known as an unfavorable lysis method resulting in possible RNA fragmentation, turns out to be fully compatible with 3' scRNA-seq. We applied it to the profiling of circulating follicular helper T cells implicated in systemic lupus erythematosus pathogenesis. Our results delineate the heterogeneity in the transcriptional programs and effector functions of these rare pathogenic T cells. As scFTD-seq decouples on-chip cell isolation and library preparation, we envision it to allow sampling at the distributed sites including point-of-care settings and downstream processing at centralized facilities, which should enable wide-spread adoption beyond academic laboratories.

Figure 1.

scFTD-seq platform and workflow. (A) Microwell array devices used in scFTD-seq method. For closed-environment format, microwell arrays are bonded to a microfluidic channel. For open-surface format, microwell arrays are used without a channel on top. (B) Cell and bead capture in microwell arrays before (top) and after (bottom) freeze-thaw cell lysis. (C) Open-surface format workflow for scFTD-seq. (D) Closed-environment format workflow for scFTD-seq. (E) Schematic of the scFTD-seq workflow with distributed sampling and modular operation. Library preparation after freeze-thaw lysis follows the protocol described in DropSeq method (17).

Figure 2.

Technical performance of scFTD-seq. (A) Mixed-species experiments reveal single-cell resolution and minimal cross-well contamination for scFTD-seq in closed-environment format. (B) Similar results are also obtained for open-surface format but with higher doublet frequency due to the inherent limitation in cell loading. (C) Sequencing performance comparison with 10× Genomics with V1 chemistry (n = 538 cells), 10× Genomics with V2 chemistry (n = 50 cells), Dropseq (n = 27 cells), SeqWell (n = 172 cells) and scFTD-seq (n = 238 cells) platforms at varying read depths (mouse 3T3 cells). Curves exhibit near-saturation profiles after 100 million reads per cell. (D) Sequencing performance comparison with 10× Genomics platform (V2). At an average of ∼35 000 reads per cell, an average of 5094 genes and 26 192 transcripts (n = 238 cells) are detected in scFTD-seq platform, comparable to 4233 genes and 27 189 transcripts (n = 50 cells) detected in 10× platform in the same cell line (mouse, 3T3). (E) Sequencing performance comparison with Dropseq and SeqWell platforms at saturating read depths using same cell line (mouse, 3T3). An average of 5469 genes and 31 017 transcripts are detected in scFTD-seq platform (n = 238 cells; average of 221 622 reads/cell), comparable to 6113 genes and 33 586 transcripts (n = 172 cells; average of 217 227 reads/cell) in SeqWell and 5753 genes and 26 700 transcripts (n = 27 cells, average of 194 428 reads/cell) in Dropseq.

Figure 3.

Identifying cell types in mixed samples. (A) t-SNE plot showing the clustering results from a mixture of human K562 and mouse NIH3T3 cells mixed at 1:10 ratio. 5062 cells were identified after alignment and filtering, of which 451 cells were inferred as human and 4511 as mouse. (B) t-SNE plot showing the sequencing results of a mixture of three human cell lines, HEK-K562- HUVEC, mixed at 1:1:2 ratio. Unsupervised clustering analysis identifies three major clusters. (C) Each cluster is identifiable as the corresponding cell type based on expression of cell-type specific genes.

Figure 4.

Inferring cell types in heterogeneous samples—whole tumor. (A) t-SNE plot of cells dissociated from melanoma tumor generated in YUMMER1.7 syngeneic mouse model in which tumor cells express Gfp. Two major clusters are identified where the left cluster is melanoma cells inferred by Gfp expression. (B) Immune cell cluster is inferred by Ptprc (Cd45) expression. (C) Unsupervised clustering analysis identified 4 clusters where clusters 1 (n = 337), 3 (n = 56) and 4 (n = 12) were identified as immune cells, and cluster 2 (n = 237) as tumor cells. (D) Differential gene expression analysis between the identified clusters.

Figure 5.

Single cell RNA profiling of circulating CD4⁺CXCR5⁺ T cells and activation states in SLE. (A) Schematic depicting experiment design to profile CD4⁺CXCR5⁺ T cells. (B) t-SNE plot showing the clustering results of stimulated versus untreated CD4⁺CXCR5⁺PD1^lowCCR7^high T cells (Tcm). Two major clusters are identified that overlap with stimulated (n = 601 cells) and untreated samples (n = 946 cells). (C) Differential gene expression analysis of stimulated and untreated Tcm cells. (D) Comparison of cytokine gene expression between stimulated and untreated Tcm cells. (E) t-SNE plot showing the clustering results of stimulated vs untreated CD4⁺CXCR5⁺PD1^highCCR7^low T cells (cTfh). Two major clusters are identified that overlap with stimulated (n = 926 cells) and untreated samples (n = 738 cells). (F) Differential gene expression analysis of stimulated and untreated cTfh cells. (G) Comparison of cytokine gene expression between stimulated and untreated cTfh cells. (H) t-SNE plot of stimulated Tcm and cTfh cells. (I) Differential gene expression analysis of stimulated Tcm and cTfh cells. (J) Comparison of cytokine gene expression between stimulated Tcm and cTfh cells.