Single-Cell Sequencing

Single-cell sequencing encompasses high-throughput molecular and bioinformatics technologies that profile the genome (scDNA-seq), transcriptome (scRNA-seq), or epigenome (scATAC-seq and single-cell DNA methylation) of individual isolated cells rather than bulk tissue homogenates. Whereas conventional bulk RNA-Seq averages gene expression across millions of lysed cells—obscuring rare stem cells, infiltrating immune subsets, and clonal tumor states—single-cell sequencing resolves cell-to-cell transcriptional variability across tens of thousands to millions of individual cells in a single experiment, powering reference initiatives such as the Human Cell Atlas.

The field originated in 2009 when Fuchou Tang, Catalin Barbacioru, and M. Azim Surani manually isolated single mouse blastomeres to perform the first single-cell mRNA-Seq assay, followed by plate-based full-length transcript methods (Smart-seq2 and Smart-seq3, developed by Simone Picelli and Rickard Sandberg). In 2015, throughput scaled by three orders of magnitude with the invention of nanoliter droplet microfluidics—Drop-seq (by Evan Macosko, Aviv Regev, and Steven McCarroll) and inDrops (by Allon Klein, Marc Kirschner, and David Weitz), commercialized as 10x Genomics Chromium—and split-pool combinatorial indexing (sci-RNA-seq by Jay Shendure and Cole Trapnell, 2017; SPLiT-seq by Georg Seelig, 2018).

Plate-Based Smart-seq vs. Microfluidic Water-in-Oil Droplet Co-Encapsulation

Single-cell transcriptomic architectures fall into two experimental categories: low-throughput, full-length plate assays and high-throughput, 3′- or 5′-tagged droplet or combinatorial assays. In Smart-seq2/3, fluorescence-activated cell sorting (FACS) deposits individual cells into 96- or 384-well plates; Moloney Murine Leukemia Virus (M-MLV) reverse transcriptase synthesizes full-length cDNA and appends untemplated cytosines at the 5′ mRNA cap to trigger template-switching oligonucleotide (TSO) extension, followed by Tn5 transposase tagmentation to capture full-length alternative splicing isoforms across the transcript body.

To scale from hundreds to 10,000–100,000 cells per run, droplet-based platforms (Drop-seq, inDrops, 10x Chromium) flow a suspension of dissociated single cells (or isolated nuclei in snRNA-seq for frozen brain or solid tissues), lysis buffer, and barcoded hydrogel or polystyrene microbeads through microfluidic channels into fluorinated oil, pinching off uniform ~0.5–1.0 nanoliter water-in-oil droplets. To prevent two cells from entering the same droplet (cell doublets), cell suspensions are loaded at a dilute Poisson parameter λ ≈ 0.05 to 0.1 cells per droplet, where the probability of a droplet containing k cells follows P(k) = (λ^k e^−λ) / k!, keeping multiplet rates below 1% to 8%.

Comparison of Single-Cell Sequencing Architectures

Cell Barcodes vs. Unique Molecular Identifiers (UMIs)

A single mammalian cell contains only ~10 picograms of cellular RNA, including roughly 100,000 to 300,000 mRNA molecules—requiring PCR amplification prior to massively parallel sequencing. To identify which cell each transcript originated from while eliminating PCR amplification distortion, every capture oligonucleotide anchored to a hydrogel bead carries four functional sequence modules from 5′ to 3′:

1. Illumina Read 1 Sequencing Primer Site: Primes sequencing of the barcode and UMI.
2. Cell Barcode (14–16 nucleotides): Created by split-and-pool synthesis so that all ~10⁷ primers on a single bead share the identical 16-nt Cell Barcode sequence, whereas every bead in the pool carries a different barcode (from a diversity of >3.6 × 10⁶ combinations).
3. Unique Molecular Identifier (UMI, 10–12 random nucleotides): Synthesized by degenerate coupling (4¹² ≈ 1.67 × 10⁷ sequences) so that every individual primer molecule on the same bead possesses a separate random UMI tag.
4. 30-nt Oligo(dT)VN Tail: Captures the 3′ poly(A) tail of cellular mRNAs upon cell lysis inside the droplet.

During reverse transcription, both the Cell Barcode and the random UMI are copied into the first-strand cDNA. After droplets are broken and pooled cDNAs undergo 12–14 cycles of PCR, all sequencing reads sharing an identical (Cell Barcode + Gene + UMI) tuple are computationally collapsed into a single digital mRNA molecule count, eliminating PCR duplicate bias.

Quality Control, SCTransform Normalization, and Dimensionality Reduction

Computational analysis in Seurat (developed by Rahul Satija's group) or Scanpy (Fabian Theis's group) begins with a sparse Cell × Gene UMI count matrix (where 85%–95% of entries are zeros due to biological absence and technical "dropout" from incomplete mRNA capture). Quality control removes empty ambient-RNA droplets (via CellBender or EmptyDrops), artificial cell doublets (Scrublet, DoubletFinder), and apoptotic cells exhibiting elevated mitochondrial transcript fractions (>10%–20% mtRNA) caused by cytoplasmic mRNA leakage through permeabilized plasma membranes while mitochondrial mRNAs remain trapped inside intact mitochondria.

UMI counts are normalized by cellular sequencing depth and log-transformed, or modeled via regularized negative binomial regression (SCTransform). After selecting the top 2,000 to 3,000 Variable Genes (HVGs) that drive biological variance across cell types, Principal Component Analysis (PCA) compresses the 20,000-gene space into 30–50 orthogonal principal components.

Graph-Based Leiden Clustering, UMAP Manifolds, and RNA Velocity

Cells are embedded into a Shared Nearest Neighbor (SNN) k-nearest-neighbor graph in PCA space and partitioned into discrete cell-type clusters using the Louvain or Leiden modularity-optimization algorithm, followed by non-linear two-dimensional visualization via Uniform Manifold Approximation and Projection (UMAP).

To reconstruct dynamic differentiation trajectories from a static single-cell snapshot, pseudotime algorithms (Monocle 3, Slingshot, Palantir) order cells along developmental lineages. Furthermore, RNA Velocity (introduced by Gioele La Manno, Sten Linnarsson, and Peter Kharchenko in 2018) exploits intronic reads captured by off-target internal priming to calculate the ratio of newly transcribed unspliced pre-mRNAs (u) to mature spliced mRNAs (s) via the differential kinetic equation d*s*/d*t* = β*u* − γ*s*, predicting the future transcriptional state of each individual cell over a timescale of hours.

Why do dying or stressed cells exhibit a high percentage of mitochondrial RNA reads in scRNA-seq quality control?

In a healthy intact cell, cytoplasmic mRNAs and mitochondrial transcripts are both retained within the plasma membrane, with mitochondrial genes accounting for roughly 1% to 5% of cellular mRNA (higher in metabolically active cardiomyocytes or renal tubular cells). When a cell undergoes mechanical dissociation stress or apoptosis, pores form in the outer plasma membrane, allowing soluble cytoplasmic mRNA molecules to leak out into the ambient suspension buffer prior to droplet encapsulation. Because the double-membrane-bound mitochondria remain structurally intact inside the dying cell body, mitochondrial transcripts stay trapped and account for >15%–50% of the remaining captured UMI reads.

When is Single-Nucleus RNA Sequencing (snRNA-seq) preferred over whole-cell scRNA-seq?

Whole-cell scRNA-seq requires enzymatic and mechanical dissociation of fresh, living tissue into a viable single-cell suspension. For snap-frozen archival clinical biobank tissues (where freeze-thaw ice crystals rupture plasma membranes), fragile interconnected cells such as adult human brain neurons (whose axons and dendrites shear during dissociation, triggering immediate-early stress gene expression), or multinucleated skeletal myofibers and fibrotic tumors, whole-cell dissociation fails or biases cell-type recovery. Single-nucleus RNA-seq (snRNA-seq) homogenizes frozen tissue directly in detergent lysis buffer to release intact nuclear membranes, profiling nuclear pre-mRNAs (enriched in intronic reads) without dissociation-induced transcriptional artifacts.

How does combinatorial indexing (sci-RNA-seq and SPLiT-seq) barcode millions of single cells without microfluidic droplets?

In split-pool combinatorial indexing (sci-RNA-seq, SPLiT-seq, Parse Biosciences), formaldehyde- or methanol-fixed permeabilized cells (or nuclei) function as their own reaction compartments. Cells are distributed across a 96- or 384-well plate where well-specific barcoded primers perform in situ reverse transcription inside the intact cells (Barcode 1). All cells are then pooled together, randomly redistributed across a second 96/384-well plate to ligate a second well-specific barcode (Barcode 2), and repeated for a third or fourth round (Barcode 3 + UMI). Across three split-pool rounds in 384-well plates, 384³ = 5.66 × 10⁷ barcode combinations are generated, enabling cubic N³ scaling of 10⁵ to >10⁶ cells using only standard pipettes and multiwell plates.

1. Tang, F., Barbacioru, C., Wang, Y., Nordman, E., Lee, C., Xu, N., Wang, X., Bodeau, J., Tuch, B. B., Siddiqui, A., Lao, K., & Surani, M. A. (2009). "mRNA-Seq Whole-Transcriptome Analysis of a Single Cell." Nature Methods, 6(5), 377–382. https://doi.org/10.1038/nmeth.1315
2. Macosko, E. Z., Basu, A., Satija, R., Nemesh, J., Shekhar, K., Goldman, M., Tirosh, I., Bialas, A. R., Kamitaki, N., Martersteck, E. M., et al. (2015). "Highly Parallel Genome-Wide Expression Profiling of Individual Cells Using Nanoliter Droplets." Cell, 161(5), 1202–1214. https://doi.org/10.1016/j.cell.2015.05.002
3. Klein, A. M., Mazutis, L., Akartuna, I., Tallapragada, N., Veres, A., Li, V., Peshkin, L., Weitz, D. A., & Kirschner, M. W. (2015). "Droplet Barcoding for Single-Cell Transcriptomics Applied to Embryonic Stem Cells." Cell, 161(5), 1187–1201. https://doi.org/10.1016/j.cell.2015.04.044
4. La Manno, G., Soldatov, R., Zeisel, A., Braun, E., Hochgerner, H., Petukhov, V., Lidschreiber, K., Kastriti, M. E., Lönnerberg, P., Furlan, A., et al. (2018). "RNA Velocity of Single Cells." Nature, 560(7719), 494–498. https://doi.org/10.1038/s41586-018-0414-6