Abstract
Single-cell RNA-seq can precisely resolve cellular states, but applying this method to low-input samples is challenging. Here, we present Seq-Well, a portable, low-cost platform for massively parallel single-cell RNA-seq. Barcoded mRNA capture beads and single cells are sealed in an array of subnanoliter wells using a semipermeable membrane, enabling efficient cell lysis and transcript capture. We use Seq-Well to profile thousands of primary human macrophages exposed to Mycobacterium tuberculosis.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$259.00 per year
only $21.58 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Accession codes
Change history
21 June 2017
The doi, link, and publication date of the Seq-Well protocol in the Protocol Exchange were updated; the doi and link have been corrected from http://dx.doi.org/10.1038/protex.2017.006 to http://dx.doi.org/10.1038/protex.2017.006a, and the date has been changed from 2016 to 2017. The grant number U24 AI11862 was incorrect; it has been corrected to U24 AI118762. These errors have been updated in the HTML and PDF versions of the article.
References
Shalek, A.K. et al. Nature 498, 236–240 (2013).
Lohr, J.G. et al. Nat. Biotechnol. 32, 479–484 (2014).
Shalek, A.K. et al. Nature 510, 363–369 (2014).
Tirosh, I. et al. Science 352, 189–196 (2016).
Hashimshony, T., Wagner, F., Sher, N. & Yanai, I. Cell Rep. 2, 666–673 (2012).
Bendall, S.C. et al. Science 332, 687–696 (2011).
Buenrostro, J.D. et al. Nature 523, 486–490 (2015).
Smallwood, S.A. et al. Nat. Methods 11, 817–820 (2014).
Zeisel, A. et al. Science 347, 1138–1142 (2015).
Treutlein, B. et al. Nature 509, 371–375 (2014).
Klein, A.M. et al. Cell 161, 1187–1201 (2015).
Macosko, E.Z. et al. Cell 161, 1202–1214 (2015).
Fan, H.C., Fu, G.K. & Fodor, S.P. Science 347, 1258367 (2015).
Bose, S. et al. Genome Biol. 16, 120 (2015).
Yuan, J. & Sims, P.A. Sci. Rep. 6, 33883 (2016).
DeKosky, B.J. et al. Nat. Biotechnol. 31, 166–169 (2013).
Ilicic, T. et al. Genome Biol. 17, 29 (2016).
Yamanaka, Y.J. et al. Integr. Biol. (Camb.) 4, 1175–1184 (2012).
Han, Q. et al. Proc. Natl. Acad. Sci. USA 109, 1607–1612 (2012).
Steinberg, G., Stromsborg, K., Thomas, L., Barker, D. & Zhao, C. Biopolymers 73, 597–605 (2004).
Gierahn, T. et al. Seq-Well protocol. Protocol Exchange http://dx.doi.org/10.1038/protex.2017.006a (2017).
Roederer, M. Current Protocols in Cytometry (eds. Robinson, J.P. et al.) Unit 1.14 (Wiley, 2002).
Satija, R., Farrell, J.A., Gennert, D., Schier, A.F. & Regev, A. Nat. Biotechnol. 33, 495–502 (2015).
Buettner, F. et al. Nat. Biotechnol. 33, 155–160 (2015).
Levine, J.H. et al. Cell 162, 184–197 (2015).
Xu, C. & Su, Z. Bioinformatics 31, 1974–1980 (2015).
Acknowledgements
We thank K. Shekhar, T. Tickle, and M. Xie for fruitful discussions. This work was supported by the Searle Scholars Program (A.K.S.), the Beckman Young Investigator Program (NIH, A.K.S.), NIH New Innovator Award DP2 OD020839 (A.K.S.), NIH U24 AI118762 (A.K.S.), P50 HG006193 (A.K.S.), the Bill and Melinda Gates Foundation grant 03629000189 (A.K.S., J.C.L. and S.F.), the Ragon Institute (A.K.S. and S.F.), the Burroughs Wellcome Foundation (S.F.), NIH P30 AI060354 (S.F.), DP3 DK09768101 (NIH, J.C.L.), P01 AI045757 (J.C.L.), NIH R21 AI106025 (J.C.L.), NIH R56 AI104274 (J.C.L.), the W.M. Keck Foundation (J.C.L.), and the US Army Research Office through the Institute for Soldier Nanotechnologies, under contract number W911NF-13-D-0001 (J.C.L.). This work was also supported in part by the Koch Institute Support (core) NIH Grant P30-CA14051 from the National Cancer Institute. J.C.L. is a Camille Dreyfus Teacher–Scholar.
Author information
Authors and Affiliations
Contributions
T.M.G., M.H.W. II, T.K.H., J.C.L. and A.K.S. developed the concepts and designed the study. T.M.G., M.H.W. II, T.K.H. and B.D.B. performed the experiments. T.M.G., M.H.W., T.K.H., A.B. and R.S. performed bioinformatic analysis. S.F. helped design and interpret the M. tuberculosis experiments. All authors analyzed and interpreted the data. T.M.G., M.H.W. II, T.K.H., J.C.L. and A.K.S. wrote the manuscript with feedback from all authors.
Corresponding authors
Ethics declarations
Competing interests
T.M.G., M.H.W. II, T.K.H., J.C.L., A.K.S., and the Broad Institute and the Massachusetts Institute of Technology have filed a patent application (patent no. PCT/US17/13791) that relates to Seq-Well, compositions of matter, the outlined experimental and computational methods and uses thereof.
Integrated supplementary information
Supplementary Figure 1 Open Array Gene and Transcript Capture
a) An open array format results in decreased gene and transcript capture, and increased cross-contamination, relative to the membrane sealing implemented in Seq-Well. (b) Species mixing experiments with reversible membrane sealing using Seq-Well provides increased gene/transcript capture and improved single-cell resolution.
Supplementary Figure 2 Seq-Well Experimental Workflow (see Seq-Well Protocol – Ref. 21)
Cells are obtained from complex tissues or clinical biopsies, and digested to form a single-cell suspension. Barcoded mRNA capture beads are added to the surface of the microwell device, settling into wells by gravity, and then a single-cell suspension is applied. The device is sealed using a semi-permeable membrane that, upon addition of a chemical lysis buffer, confines cellular mRNAs within wells while allowing efficient buffer exchange. Liberated cellular transcripts hybridize to the bead-bound barcoded poly(dT) primers that contain a cell barcode (shared by all probes on the same bead but different between beads) and a unique molecular identifier (UMI) for each transcript molecule. After hybridization, the beads are removed from the array and bulk reverse transcription is performed to generate single-cell cDNAs attached to beads. Libraries are then made by a combination of PCR and tagmentation, and sequenced. After, ssingle-cell transcriptomes are assembled in silico using cell barcodes and UMIs.
Supplementary Figure 3 Bead And Cell Loading Efficiency
(a) Two arrays were loaded with barcoded beads through intermittent rocking. After washing, arrays were imaged in transmitted light and AF488 channel to capture bead autofluorescence. A plot of the frequency of the 75th percentile AF488 well intensity across the array (Panel 1) and the frequency of wells containing zero, one and multiple beads is displayed (Panel 2). (b) 200 μL of a 1:1 mix of fluorescently labeled human (HEK 293) and mouse (3T3) cell solution was loaded into 3 arrays and 12 wells of a 96 well plate. The number of cells loaded into each array and well as enumerated by fluorescent imaging is plotted, normalized to the average number of cells/well in the 96 well plate. Mean and standard error are denoted by line and error bars respectively. (c) 2x102, 2x103, and 2x104 total cells of a 1:1 mixture of fluorescently labeled HEK 293T and 3T3 cells were loaded onto three functionalized arrays each. All arrays were fluorescently imaged to enumerate the number of each cell line in each array microwell. The mean ± standard deviation of the number of empty, single and multiple occupancy wells across the three replicate arrays for each loading density is displayed along with the mean ± standard deviation of the percentage of occupied wells containing a cell from each species
Supplementary Figure 4 PDMS Surface Chemistry Functionalization Protocol and Differential Functionalization of Microwell Arrays
(a) The surface of the PDMS device is initially treated with an air plasma under mild vacuum, terminating the surface in hydroxyls. This PDMS surface is aminated using (3-Aminopropyl)triethoxysilane (APTES). The amine surface is then activated with PDITC to create an isothiocyanate surface. The isothiocyanate on the top surface of the array (negative space) is covalently linked to chitosan polymers through their amine group. The hydrophobicity of the isothiocyanate surface prevents solvation of the microwells with the aqueous chitosan solution, preventing chitosan from reacting with the inner well surfaces (positive space). These surfaces are subsequently reacted with the free amine of poly(glutamic) acid polymers under vacuum to drive the solvation of the wells. (b) The top surface of a PDITC-activated array was coated with streptavidin-PE (red) and the inner well surfaces were coated with streptavidin-AF488 (green) using same method used to functionalize with chitosan and poly(glutamate). (c) Two chitosan/poly(glutamate) bifunctionalized arrays were submerged in MES buffer without (Panel 1) or with (Panel 2) 100 μg/mL EDC and 10 μg/mL NHS for 10 minutes. The arrays were washed and then submerged in PBS solution containing 1 μg/mL AF568-labeled antibody overnight. After washing, arrays were imaged for AF568 fluorescence.
Supplementary Figure 5 Microwell Sealing With Semipermeable Membrane
PBMCs labeled with αCD45-AF647 were loaded into two BSA-blocked arrays and one array functionalized with chitosan and poly(glutamate). A semipermeable membrane was attached to one of the BSA-blocked arrays and the chitosan:polyglutamate functionalized array prior to addition of lysis buffer. (a) Example images of transmitted light and AF647 fluorescence of the arrays before, and 5 and 30 minutes after addition of lysis buffer are displayed for each array. (b) The total fluorescence intensity (FI) of all pixels associated with cells within a well is plotted against the median fluorescent intensity (MFI) of the volume of the same well 5 minutes after lysis for 12,100 wells from each array. (c) The MFI of the well volume 5 minutes after lysis is plotted against the MFI of the volume of the same well 30 minutes after lysis for the same 12,100 wells from each array.
Supplementary Figure 6 Read Mapping Quality
Read mapping quality matrices were generated for each sample for human (blue) and mouse (red) cells, aligned to hg19 and mm10, respectively. High quality samples had relatively higher percentages of annotated genomic (genic) and exonic transcripts and low percentages of annotated intergenic and ribosomal transcripts (Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR).
Supplementary Figure 7 Comparison of Gene and Transcript Capture and Percent Contamination Among Massively-Parallel scRNA-Seq Methods Using Mouse and Human Cell Lines
Histograms of the percent cross-species contamination in (a) Seq-Well, (b) Drop-Seq (Ref. 12), and (c) Yuan and Sims (Ref. 15). In each plot, cells with greater than 90% of human transcripts are displayed in blue and cells with less than 10% human transcripts are displayed in red. (d) Transcript capture in human (blue) and mouse (red) cell lines across three massively-parallel, bead-based single-cell sequencing platforms (Seq-Well, Drop-Seq, and 10X Genomics, with downsampling to an average read-depth of 80,000 reads per cell, consistent with 10X genomics data (Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR). We detect an average of 32,841 human transcripts and 29,806 mouse transcripts using Seq-Well compared to an average of 39,400 human transcripts and 24,384 mouse transcripts using Drop-Seq, an average of 24,751 human transcripts and 22,971 mouse transcripts using 10X Genomics (available from http://support.10xgenomics.com/single-cell/datasets/hgmm). (e) Gene detection across human and mouse cell lines across the same three single-cell sequencing platforms with down-sampling to the average read-depth of 80,000 reads per cell, consistent with 10X genomics (Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR). We detect an average of 6,174 human genes and 5,528 mouse genes using Seq-Well, an average of 5,561 human genes and 4,903 mouse genes using Drop-Seq and an average of 4,655 human genes and 3,950 mouse genes using 10X Genomics. (f) Downsampling to an average of 42,000 reads per cell consistent with data published in Yuan and Sims 2016, results in average detection of 23,061 mouse transcripts using Seq-Well compared to an average of 24,761 mouse transcripts using the Yuan and Sims platform (Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR). (g) Downsampling to an average of 42,000 reads per cell results in average detection of 4,827 mouse genes using Seq-Well compared to an average of 4,569 mouse genes using the Yuan and Sims platform (Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR).
Supplementary Figure 8 Transcript Cutoff For Species-Mixing Validation
We sequenced two arrays (a & b) to confirm single-cell resolution and minimal cross-contamination between mouse and human cells. We called cells by plotting the cumulative distribution of transcripts and making a cutoff at the elbow in the curve. In the first experiment (a), which was used to validate our single-cell resolution, we shallowly sequenced the array and made the cutoff at 2,000 transcripts. In the second experiment (b), where we sequenced the array deeply to allow a competitive comparison to Drop-Seq, we made our cutoff at 10,000 transcripts.
Supplementary Figure 9 Comparison Of In-Silico HEK293 Populations With Bulk Populations
Scatterplots showing the correlation between gene expression estimates from bulk populations (40,000 HEK cells and 40,000 mRNA capture) and populations generated in-silico from 1, 10, 100, and 1,000 randomly-sampled single HEK293 cells (1 Cell: R = 0.751 ± 0.0726; 10 Cells: R = 0.952 ± 0.008; 100 Cells: R = 0.980 ± 0.0006; 1000 Cells: R = 0.983 ± 0.0001).
Supplementary Figure 10 Mapping Lineage Defining Transcripts To PBMC Clusters
(a) Clusters identified through graph-based clustering (Methods) correspond to major immune cell populations. (b,e) CD4 T cells are characterized by expression of CD3D and T-cell receptor expression without pronounced expression of cytoxic genes NKG7 and PRF1. (c,f) CD8 T cells are defined by expression of NKG7 and PRF1. (d,g) Monocytes are defined by expression of cathepsin B (CTSB) and SOD2. (e) Natural killer cells are characterized by expression of cytotoxic genes in the absence of T cell receptor expression. (h) B cells are marked by elevated expression of MS4A1 (CD20) transcripts. (i) Dendritic cells are enriched for expression of BIRC3.
Supplementary Figure 11 Heatmap Of PBMCs
(a) Genes enriched in each cluster were identified using an “ROC” test in Seurat, comparing cells assigned to each cluster to all other cells. A heatmap was constructed using enriched genes found to define each cluster. One cluster of 602 cells that demonstrated exclusive enrichment of mitochondrial genes was removed as these likely represent low-quality or dying cells. (b) We generated a t-SNE projection of 4,296 cells with greater than 10,000 reads, 1,000 transcripts, 500 genes, and 65% transcript mapping. We removed a total of 602 cells from the final analysis found to be strongly enriched for expression of mitochondrial genes. The remaining 3,694 cells form distinct clusters enriched for lineage-defining that distinguish cells types from one another.
Supplementary Figure 12 Read Mapping Quality In PBMCs
(a-c) Violin plots depicting (a) reads (a), (b) transcripts (b), and (c) genes (c) per cell, separated by cell type. (d) Percent mRNA bases per cell, separated by cell type.
Supplementary Figure 13 Comparison Of Human PBMC Gene And Transcript Capture With Other Massively-Parallel scRNA-Seq Methods
(a) Comparison of transcript capture (top) and gene detection (bottom) between Seq-Well and 10X Genomics within PBMC cell types prior to downsampling (colored as in Figure 2; Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR). Among B cells (orange), an average of 1,315 genes and 3,632 transcripts were detected using Seq-Well and an average of 710 genes and 1,910 transcripts were detected in 10X Genomics data. Among CD4 T cells (blue), an average of 861 genes and 2,444 transcripts were detected using Seq-Well and an average of 815 genes and 2,370 transcripts were detected in 10X Genomics data. Among CD8 T cells (yellow), an average of 885 genes and 2,574 transcripts were detected using Seq-Well and an average of 809 genes and 2,029 transcripts were detected in 10X Genomics data. Among Monocytes (green), an average of 1,288 genes and 3,568 transcripts were detected using Seq-Well and an average of 974 genes and 2,835 transcripts were detected in 10X Genomics data. Among NK cells (red), an average of 902 genes and 2,338 transcripts were detected using Seq-Well and an average of 907 genes and 1,943 transcripts were detected in 10X Genomics data. (b) Transcript capture (top) and gene detection (bottom) upon downsampling of Seq-Well data to an average read depth 69,000 reads per cell (Center-line: Median; Limits: 1st and 3rd Quartile; Whiskers: +/- 1.5 IQR; Points: Values > 1.5 IQR). Upon downsampling, in Seq-Well, an average of 1,048 genes and 3,103 transcripts were detected among B cells, 735 genes and 2,221 transcripts among CD4 T cells, 763 genes and 2,353 transcripts among CD8 T cells, 1,052 genes and 3,105 transcripts among monocytes, and 789 genes and 2,041 transcripts among NK cells.
Supplementary Figure 14 T-SNE Visualization Of Exposed And Unexposed Macrophages Using A 5,000 Transcript Cutoff
(a) Using a threshold of 5,000 detected transcripts, we identified 4,638 macrophages. (b) Among these 4,638 cells, we identified 5 distinct clusters of macrophages by performing graph-based clustering over 5 principal components (377 variable genes). (c) Clusters 1-3 are defined by unique gene expression signatures, while Clusters 4 and 5 are defined by expression of mitochondrial genes, suggesting low-quality cells. (d) Following removal of cells within Clusters 4 and 5, there remain a total of 2,560 cells in Clusters 1-3.
Supplementary Figure 15 Quality By Cluster Among TB Macrophages
(a-c) Violin plots depicting (a) reads (a), (b) transcripts (b), and (c) genes (c) per cell, separated by cluster. (d) Percent mRNA bases per cell, separated by cluster.
Supplementary information
Supplementary Text and Figures
Supplementary Figures 1–15 and Supplementary Tables 1–2 (PDF 3154 kb)
Supplementary Protocol
Seq-Well protocol (PDF 2705 kb)
Supplementary Table 3
Gene Expression Matrix for PBMCs (ZIP 33984 kb)
Supplementary Table 4
PBMC Cluster Enrichments (XLSX 184 kb)
Supplementary Table 5
Gene Expression Matrix for Mtb-Exposed Monocyte-Derived Macrophages and Unexposed Control Cells (ZIP 123454 kb)
Supplementary Table 6
TB Cluster Enrichments (XLSX 523 kb)
Supplementary Table 7
Cluster Enrichments between Exposure Groups (XLSX 70 kb)
Supplementary Table 8
Differentially Expressed Genes between TB Exposed and Unexposed Cells within Each Cluster (XLSX 1775 kb)
Supplementary Table 9
TB Infection by Cluster Enrichments (XLSX 605 kb)
Supplementary Table 10
GSEA Comparisons of Exposed and Unexposed Cells within Each Cluster (XLSX 4966 kb)
Bead Loading
Video demonstration of bead loading on microwells arrays. (MP4 76581 kb)
Rights and permissions
About this article
Cite this article
Gierahn, T., Wadsworth, M., Hughes, T. et al. Seq-Well: portable, low-cost RNA sequencing of single cells at high throughput. Nat Methods 14, 395–398 (2017). https://doi.org/10.1038/nmeth.4179
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/nmeth.4179
This article is cited by
-
scDrugPrio: a framework for the analysis of single-cell transcriptomics to address multiple problems in precision medicine in immune-mediated inflammatory diseases
Genome Medicine (2024)
-
Pro-inflammatory feedback loops define immune responses to pathogenic Lentivirus infection
Genome Medicine (2024)
-
Single-cell transcriptomics enable the characterization of local extension in retinoblastoma
Communications Biology (2024)
-
Comparative analysis of cell–cell communication at single-cell resolution
Nature Biotechnology (2024)
-
FastCAR: fast correction for ambient RNA to facilitate differential gene expression analysis in single-cell RNA-sequencing datasets
BMC Genomics (2023)
