© 2017. Published by The Company of Biologists Ltd | Development (2017) 144, 17-32 doi:10.1242/dev.

133058

REVIEW

Understanding development and stem cells using single

cell-based analyses of gene expression
Pavithra Kumar, Yuqi Tan and Patrick Cahan*

ABSTRACT that are weighted averages of the constituent cell types rather than
In recent years, genome-wide profiling approaches have begun to an accurate reflection of an individual cell. Third, although single
uncover the molecular programs that drive developmental processes. cell profiling can help to define cell types with higher resolution, it
In particular, technical advances that enable genome-wide profiling of can also be used to discover previously unappreciated cell types in
thousands of individual cells have provided the tantalizing prospect heterogeneous populations and complex tissues. For example, the
of cataloging cell type diversity and developmental dynamics in a single cell ‘profiling’ (using Southern hybridization) of cDNA
quantitative and comprehensive manner. Here, we review how single- from nine genes in 15 single pyramidal neurons of the rat
cell RNA sequencing has provided key insights into mammalian hippocampus led to the discovery of two neuronal subtypes
developmental and stem cell biology, emphasizing the analytical distinguished by their K+ to Ca2+ channel gene expression ratios
approaches that are specific to studying gene expression in single (Eberwine et al., 1992). Fourth, and finally, it is becoming evident
cells. that single-cell profiling will allow us to address a wide range of
questions and hypotheses concerning the co-occurrence of
KEY WORDS: RNA-Seq, Computational biology, Gene regulatory molecular events in individual cells. This exploration need not be
networks, Pseudotime, Single cell, Stem cells limited to gene expression. For example, the simultaneous
interrogation of DNA copy number variation (CNV) and gene
Introduction expression in single cells (Dey et al., 2015; Macaulay et al., 2015)
To characterize the diversity of cell types in multicellular could be used to uncover the extent to which CNVs contribute to
organisms, to investigate the mechanisms that give rise to this functional heterogeneity in the developing nervous system
diversity in development and how they go awry in disease, and to (McConnell et al., 2013), and to determine the extent to which
understand how dynamic intercellular interactions contribute to this is mediated by alterations in gene expression. Likewise, the
these processes, we need technologies that allow us to make integration of DNA methylation state with transcription will reveal
genome-wide measurements of many single cells. Over the past the extent to which this epigenetic modification contributes to
16 years, a number of genome-wide profiling techniques (e.g. ‘stochastic’ expression (Angermueller et al., 2016). More
RNA sequencing and chromatin immunoprecipitation sequencing generally, the incorporation of other data on a per cell basis will
or ChIP-Seq; see Glossary, Box 1) have been developed and used compound the amount of knowledge that can be gleaned from
to study global changes in, for example, gene expression, single-cell molecular profiling.
chromatin occupancy by transcription factors and epigenetic In the relatively brief time since the first description of the mRNA
marking. However, in general, these approaches require more content of single cells (Tang et al., 2009), a staggering array of
starting material than is available in an individual cell, limiting single-cell genome-wide profiling techniques and applications have
their application to cell populations. Thus, while such studies have been reported (Fig. 1). The surfeit of methods to quantify RNA in
provided important advances, it is becoming clear that the profiling single cells, including Smart-Seq (Ramskold et al., 2012), CEL-Seq
of individual cells would be highly advantageous. There are many (Hashimshony et al., 2012) and Quartz-Seq (Sasagawa et al., 2013),
reasons for this. First, especially in developmental contexts, the reflects the relative ease with which mRNA can be captured,
rarity of some cell types means that large numbers of animals need amplified and sequenced in order to provide a molecular readout of
to be used in order to acquire sufficient cells for profiling. For cell state. In addition to single-cell gene expression, methods to
example, profiling the transcriptome of hematopoietic stem cells assess DNA variation (Navin et al., 2011), chromatin organization
(HSCs) across seven distinct stages of development required the (Nagano et al., 2016), chromatin accessibility (Cusanovich et al.,
manual dissection of greater than 2500 mouse embryos, a 2015; Buenrostro et al., 2015), DNA-protein interactions (Rotem
painstaking feat accomplished over the course of 3 years et al., 2015) and DNA methylation (Smallwood et al., 2014) have
(McKinney-Freeman et al., 2012). Second, even highly robust been developed for single cells.
and functionally verified isolation strategies do not reach 100% Here, we review single-cell genome-wide studies of mammalian
purity. For example, at best, only one in two CD150+CD48- development and stem cells, focusing on single-cell RNA
DEVELOPMENT

Sca-1+Lineage-c-kit+ bone marrow cells can reconstitute the sequencing (scRNA-Seq; see Box 2), its applications and the
hematopoietic system of irradiated mice (Kiel et al., 2005). This insights that have been gleaned from this technique. We do not
impurity is problematic because it generates molecular signatures discuss the tantalizing progress being made in single-cell
proteomics (Bandura et al., 2009) or in situ RNA-Seq (Ke et al.,
Department of Biomedical Engineering, Institute for Cell Engineering, Johns
2013; Lee et al., 2014; Lovatt et al., 2014). We also refer the reader
Hopkins University School of Medicine, Baltimore, MD 21205, USA. to several other reviews that provide a more in-depth discussion of
the technical and molecular details of single-cell methods (Etzrodt
*Author for correspondence ( patrick.cahan@jhmi.edu)
et al., 2014; Kolodziejczyk et al., 2015a; Macaulay and Voet, 2014;
P.C., 0000-0003-3652-2540 Wang and Navin, 2015).

17
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

(Brennecke et al., 2013; Grün et al., 2014; Kharchenko et al.,

Box 1. Glossary 2014; Pierson and Yau, 2015).
Bayesian network: A probabilistic graph in which each node represents scRNA-Seq can be used to determine the various cell types
a random variable and each edge represents a conditional dependence within a population or tissue, including rare cell types. Commonly
between two random variables (or nodes). used approaches to identify sub-structure in scRNA-Seq data and to
Chromatin immunoprecipitation sequencing or ChIP-Seq: A method
to determine the genomic regions with which a protein interacts.
identify distinct cell types include principal component analysis
Drop-out: A false negative in scRNA-Seq data. In other words, when a (PCA; see Glossary, Box 1) and t-distributed stochastic neighbor
gene is expressed in a cell but is not detected by scRNA-Seq. embedding (t-SNE; see Glossary, Box 1), both of which aim to
Gaussian mixture model: A class of probabilistic models that represent reduce the number of variables required to represent the total
clusters of data points using Gaussian densities. variation in the data (Maaten and Hinton, 2008). After running the
Gene regulatory network (GRN): The complete set of regulatory data through these dimensionality reduction techniques, the results
relationships between genes and gene products.
are visualized and subsequently used as input for secondary
K-means clustering: An algorithm that assigns entities (e.g. samples or
cells) to K distinct groups, where K is an integer specified by the user. algorithms, such as K-means clustering (see Glossary, Box 1) and
K-means seeks to find the set of group assignments that minimize the Gaussian mixture modeling (see Glossary, Box 1), to identify the
distances within all of groups. number of clusters and to assign cells to clusters, sometimes in a
Minimal spanning tree (MST): An algorithm to connect vertices of a probabilistic fashion (Fig. 2A). Owing to the low sensitivity of
weighted-edge graph, such that the resulting graph has the minimal total scRNA-Seq, it has been challenging to use these approaches ‘as is’
edge weight.
to identify rare sub-populations and distinguish them from technical
Principal component analysis (PCA): A linear projection of data from
high to low dimensions constrained by maximizing the variance between
outliers. However, an analytical pipeline called RACE ID was
components. Good at preserving large distances between points (cells) recently developed to address this problem (Grün et al., 2015).
in the original space. RACE ID first estimates the number of clusters (cell types or states)
Pseudotime: An artificial ordering of cells based upon a statistically using k-means. Second, it statistically models the expression of each
inferred trajectory often interpreted as time. Such an approach is useful gene within each cluster and uses these models to identify outlier
when sampling from a population or populations in which single cells are cells, which are defined as those with highly unlikely expression of
at distinct stages of a process.
Simpson’s paradox: The loss or reversal of statistical associations
two or more genes. Finally, it assigns outliers to new clusters,
between variables, as determined in more than one group, when those defining these as new cell types or states, that are visualized using
groups are combined. t-SNE. Although this approach has several parameters that require
Synthetic RNA spike-ins: Poly-adenylated mRNA synthesized and tweaking, it has been used successfully for identifying rare Paneth
provided at known copy number used to estimate absolute abundance of progenitor cells in intestinal organoids (Grün et al., 2015). Other
target mRNA, and to estimate and correct for technical noise in scRNA- similar approaches have also been described, including GiniClust
Seq. Commonly used spike-in sets are designed to have no similarity to
(Jiang et al., 2016), and predictions generated with these methods
the transcriptomes of commonly studied species but to have similar
sequence composition and lengths. can be tested by searching for genes encoding cell-surface markers
t-distributed stochastic neighbor embedding (t-SNE): a projection of that distinguish the new cell clusters, prospective isolation by
high dimensional data into lower dimensions by preserving fluorescence-activated cell sorting (FACS) and subsequent
probabilistically determined pairwise distances between points. Good at functional assessment.
preserving smaller distances between points (cells) in the original space. In addition to cell type heterogeneity, cells within a population
Transcriptional noise: Random fluctuations in the transcription of a
can exhibit temporal heterogeneity. They may, for example, differ
single gene, quantified as the standard deviation divided by the mean.
primarily with regard to the stage (e.g. of a developmental
process) at which they are sampled. Another simple variable is
the stage of the cell cycle but the concept is extendable to
The basics of scRNA-Seq analysis developmental trajectories, or even to stages of disease
The technique of scRNA-Seq involves isolating and lysing single progression. Several approaches have recently been developed
cells, producing cDNA in such a way that material from a cell is to reconstruct major trajectories from single-cell molecular
uniquely marked or barcoded, and generating next-generation profiling data and to place cells along these trajectories
sequencing libraries that are subjected to high-throughput (Fig. 2B). The first of these to be developed were Wanderlust
sequencing (see Box 2). The ultimate output of this process is a and Monocle (Bendall et al., 2014; Trapnell et al., 2014). Monocle
series of sequence reads that are attributed to single cells with the relies on the minimal spanning tree (MST; see Glossary, Box 1)
barcode, aligned to a reference genome or transcriptome, and algorithm to find trajectories in data, which are interpreted as a
transformed into expression estimates. After sequencing, libraries temporal progression or ‘pseudotime’ (see Glossary, Box 1). Cells
are subjected to quality control to remove low-quality samples can then be placed along pseudotime based on their distance from
(e.g. material from incompletely lysed cells), and normalized the major trajectories defined by the MST, and the data can be
expression estimates are then used as input for an ever-increasing analyzed using standard approaches for temporal data. Such an
battery of algorithms tailored for scRNA-Seq. We briefly describe approach is typically used to identify regulators of developmental
DEVELOPMENT

the approaches currently used to analyze scRNA-Seq data (Fig. 2). progression or bifurcation points. By contrast, Wanderlust (which
We refer the reader to other reviews that discuss the many was implemented to order single cell mass-cytometry data) creates
pre-processing and quality-control steps that are required to an ensemble of nearest neighbor graph and determines an average
produce ‘clean’, informative single-cell data (Bacher and path based on the trajectories defined as the shortest path starting
Kendziorski, 2016; Stegle et al., 2015), and that describe from a defined starting point. A multitude of new algorithms have
methods to detect and account for uninteresting confounding been described more recently to achieve a similar aim. These
effects, such as the stage of cell cycle (Buettner et al., 2015; include Wishbone (Setty et al., 2016), Sincell (Juliá et al., 2015),
Vallejos et al., 2015), and to analyze and account for technical time variant clustering (Huang et al., 2014), SCUBA (Marco
noise and the so-called ‘drop out’ (see Glossary, Box 1) effect et al., 2014), Waterfall (Shin et al., 2015), probabilistic Boolean

18
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Fig. 1. The growth of single cell genome-

Arabidopsis
wide profiling techniques. A surge in
Mouse
C. elegans Rat
scRNA-Seq applications can be observed.
Human Zebrafish
The cumulative number of cells that have
Macosko (2015) been subjected to scRNA-Seq is shown,
Klein (2015) separated by species. Landmark studies
Number of cells sequenced (⫻1000)

60 are highlighted. Tang et al. (2009), Tang

(2010b), Islam (2011), Ramskold (2012)
1
and Hashimshony (2012) are the first five
Picelli (2013) scRNA-Seq studies. They introduced the
0.75 Yan (2013) major varieties of scRNA-Seq: Tang
Xue (2013) protocol, STRT-Seq, CEL-Seq and Smart-
40 Hashimshony (2012) Seq. Yan (2013) and Xue (2013) leverage
0.5
scRNA-Seq to explore and the dynamics of
Ramskold (2012) human zygotic genome activation. Picelli
Islam ((2011) (2013) introduces Smart-Seq2 with
0.25 Tang (2010)
Tang (2009) increased sensitivity. Macosko (2015) and
Klein (2015) introduce high-throughput low-
0 cost droplet-based methods that have
20
2009 2010 2011 2012 2013 2014 vastly increased the number of cells that
can be sequenced.

2010 2012 2014 2016

Year

networks (Chen et al., 2015), diffusion maps (Haghverdi et al., ‘Embryomics’: using scRNA-Seq to understand
2015), TSCAN (Ji and Ji, 2016), SLICER (Welch et al., 2016) and embryogenesis
SCOUP (Matsumoto and Kiryu, 2016). These various types of As we have summarized above, a host of approaches and techniques
pseudotime analyses allow the identification of regulators of have been developed in recent years to study gene expression in
temporal processes and of transient events that are obscured by single cells and to then analyze this data so as to provide meaningful
bulk-derived data. Models generated from these types of analyses datasets. Importantly, such methods have been used successfully to
can be tested by live-cell tracking, by modulating the expression gain insights into various aspects of embryogenesis and early
of candidate transcriptional regulators or by perturbing the development (summarized in Table 1). Below, we highlight just
identified signaling pathways. some of these advances.
Similar to the concept of placing cells along a temporal axis,
several algorithms have been developed to place cells into spatial Lineage segregation in the pre-implantation embryo
contexts. Such spatial reconstruction methods (Satija et al., 2015; Before it implants, the mammalian embryo consists of three
Achim et al., 2015) use prior information about localized marker lineages: the epiblast (EPI), which gives rise to three germ layers;
gene expression to place single cells from scRNA-Seq into a spatial the trophectoderm (TE), which mediates implantation; and
representation of an anatomical context (Fig. 2C). When temporal the primitive endoderm (PE), which provides nutrition to the
and spatial axes coincide, Sinova (a method similar in concept to developing embryo (Rossant et al., 2009). A first hint of the power
Monocle) can be used to place cells spatially without prior of single cell techniques was provided by a single-cell qPCR
knowledge of marker gene expression (Li et al., 2016c). study that uncovered transcriptional differences between these early
Finally, there is much excitement around the prospect of using embryonic lineages in mice (Guo et al., 2010). In this study, ∼450
scRNA-Seq to reconstruct gene regulatory networks (GRNs; see single cells at seven developmental stages (from the zygote to
Glossary, Box 1) that more faithfully predict transcriptional state 64-cell blastocyst) were manually isolated and the expression of 48
and dynamics than those produced from the profiling of bulk genes representing, for example, developmental signaling pathways
populations. In theory, GRNs constructed from single-cell data (e.g. Bmp4) or transcription factors known to regulate pluripotency
should be better because they will not be confounded by population (e.g. Utf1) and gastrulation (e.g. Gata2) were analyzed. Using the
substructure, which can lead to Simpson’s Paradox (Trapnell, 2015) expression of markers characteristic of cells constituting the
(see Glossary, Box 1), and because gene-to-gene correlations (from blastocyst, Guo et al. were able to group the cells from the 64 cell
which GRNs are reverse engineered) are elicited by stochastic embryos as EPI, TE, or PE and identify genes that mark fate
DEVELOPMENT

variation rather than non-physiological overexpression or knockdown. decisions. For example, Sox2 expression marked the first fate
(Bian and Cahan, 2016). However, the low sensitivity of scRNA- decision – the choice to form inner or outer cells of the morula.
Seq is problematic for detecting correlations, especially for genes Notably, it was shown that lineage specification also involves a
that are transcribed at very low rates. Thus, although GRNs have reduction in the expression of some TFs in cells of opposing
been reconstructed from single-cell quantitative PCR (qPCR) data lineages, as well as lineage-specific increases in some TFs. For
using Bayesian networks (see Glossary, Box 1) (Moignard et al., example, Gata6 expression is reduced in EPI progenitors, whereas
2015), and formal methods have been devised in this context factors such as Klf2 are reduced in TE progenitors.
(Ocone et al., 2015), no large-scale GRN reconstruction from Given that the above study was based on the targeted analysis of
scRNA-Seq data has been described to date. just a few genes using qPCR, the identification of novel genes that

19
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Box 2. Single-cell RNA sequencing: how does it work?

scRNA-Seq Cell isolation Reverse Library preparation

Amplification
method and lysis transcription and sequencing
Drop-Seq

Cells: 50,401
Genes: 6177
Studies: 1
inDrop Droplets
Cells: 5212
Genes: NA
Studies: 1
Smart-Seq
PolyA tailing and
Cells: 513 SSS and UMI
Genes: 6832
Studies: 40 Library prep
and Seq
Tang
Cells: 107
TS and UMI
Genes: 10,874 PCR
Studies: 8
Microfluidics
SCRB-Seq
Cells: 851
Genes: 3400
Studies: 1 Tubes

Quartz-Seq
Cells: 47
Genes: NA
Studies: 1
TS
CEL-Seq
Cells: 85
Genes: 5867
Studies: 2

STRT-Seq IVT-PCR
PolyA tailing
Cells: 1467
and SSS
Genes: 8822
Studies: 5

MARS-Seq
Cells: 2042
Genes: 533
Studies: 3

Some of the most widely used protocols for scRNA-Seq are listed; shown in boxes are the number of studies in which the approach has been used, the
average number of single cells subjected to scRNA-Seq and the average number of genes reported as detected. Although all techniques follow a similar
outline, they vary in their methods. The first step in scRNA-Seq is the efficient capture and lysis of single cells. This can be achieved via manual isolation
of cells using FACS or micropipetting into tubes containing lysis solution (tubes), via commercial microfluidics-based platforms such as Fluidigm’s C1
(microfluidics), or by capturing cells into nanoliter droplets that contain lysis buffer (droplets). Once cells are lysed, the mRNA population is bound by primers
containing a polyT region that allows them to bind to the polyA tail of mRNA. These primers can also have other unique features such as unique molecular
identifiers (UMIs), cell barcodes or sequences that serve as PCR adapters. The captured mRNA is subsequently converted to cDNA using a reverse
transcriptase to generate the first cDNA strand. Historical techniques then use polyA tailing of the 3′ end of the newly synthesized strand followed by second-
strand synthesis (SSS) to produce double-stranded DNA (ds-cDNA). However, recently, template switching (TS) is carried out prior to generation of the
second strand, using a custom oligo called the template switch oligo (TSO) that binds the 3′ end of the newly synthesized cDNA and serves as a primer for
the generation of the second strand, thus resulting in identical sequences on both ends of the ds-cDNA. This ensures efficient amplification of the full-length
ds-cDNA. PolyA tailing and TS can be carried out both with or without UMIs. After successful second-strand synthesis, most techniques use PCR-based
amplification to amplify the ds-cDNA obtained from a single cell, in order to generate enough starting material for sequencing. However, techniques such as
MARS-Seq, CEL-Seq and inDrop perform in vitro transcription (IVT) followed by another round of cDNA synthesis, before PCR amplification. After this point,
all techniques converge, such that the amplified ds-cDNA is used as starting material to generate a collection of short, adapter-ligated fragments called a
DEVELOPMENT

library, that is fed into a sequencer of choice to generate sequencing reads. NA, not applicable.

play key roles in these developmental stages was not possible. multiple isoforms of the same gene – information that is
However, the first scRNA-Seq study began to address this issue indeterminable from bulk samples.
by characterizing the complete transcriptomes of individual The most recent and comprehensive transcriptional portrait of
blastomeres from four-cell stage murine embryos and from mature human pre-implantation embryos, using 1529 individual cells from
oocytes (Tang et al., 2010a). In addition to acting as a proof of 88 pre-implantation embryos, substantiated many observations of
principle, this study documented that a single cell expresses the earlier molecular characterizations (Petropoulos et al., 2016).

20
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

A Identifying cell types

Cluster
annotation

scRNA-Seq

Dimension Cluster
reduction detection

Cluster 2
Differential
expression

Cluster 1
B Pseudotime analysis
scRNA-Seq
Pseudotime Identify
Dimension and bifurcation Trajectory 2 candidate
TF3

Expression
reduction analysis regulators TF2 TF4
TF1 TF4
Trajectory 1

Trajectory 1

C Spatial reconstruction
In situ hybridization
data of landmark
genes
Region
Gene A
Region I A
Predicted location
Genes

Gene B of cells
Region II B

Gene C C
Region III
I II III
Compare region
and cell profiles

scRNA-Seq Single cells

A +
others
Genes

B+
others
C+
others

Fig. 2. Typical approaches for analyzing scRNA-Seq datasets. Several types of analyses are popular for analyzing scRNA-Seq datasets. (A) When trying to
identify cell types, dimension reduction techniques such as independent component analysis, principal component analysis, t-distributed stochastic neighbor
embedding, ZIFA (Pierson and Yau, 2015) or weighted gene co-expression network analysis (Langfelder and Horvath, 2008) are first used to project high-
dimensional data into a smaller number of dimensions to ease visual evaluation and interpretation. Clusters of similar cells can be identified using generally
applicable methods, such as Gaussian mixture modeling (Fraley and Raftery, 2002) or K-means clustering, or methods devised specifically for single cell data,
such as StemID (Grü n et al., 2016), SCUBA, SNN-Cliq (Xu and Su, 2015), Destiny (Angerer et al., 2015) or BackSpin (Zeisel et al., 2015). Clusters can then be
DEVELOPMENT

annotated based on domain-specific knowledge of the expression of a few genes, or automatically based on gene set enrichment. Finally, specific genes that are
differentially expressed between clusters can be identified using scRNA-Seq-specific methods such as SCDE (Kharchenko et al., 2014) and MAST (Finak et al.,
2015). (B) Most pseudotime analyses (which place each cell on a statistically derived axis that represents progression along a process, such as developmental
time) start by performing dimension reduction. They then determine trajectories through the reduced dimensionality data; some algorithms identify bifurcation
points and generate a distinct trajectory. The trajectories can then be used to order single cells along the process and to identify candidate regulators of stage
transitions, for example, by finding stage-specific transcription factors (TF1-TF5). (C) One of the major drawbacks of scRNA-Seq is the loss of spatial context
information when cells are dissociated and/or isolated. Spatial reconstruction methods attempt to ameliorate this issue by leveraging prior knowledge of landmark
gene expression. Typically, localized expression of select genes is generated from in situ hybridization. Spatial reconstruction algorithms then compare scRNA-
Seq profiles to discretized in situ hybridization profiles, and cells are placed in silico in the anatomical region with a matching profile. Machine-learning approaches
can be used to estimate the expression of landmark genes to overcome the noisy nature of scRNA-Ssq data.

21
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Table 1. Single-cell RNA-Seq-based studies of early mammalian development

Number
Study Cell type(s) Primary Species Cells of genes Method Summary
Tang et al. Early embryo Yes Mouse 7 11,920 Tang The first scRNA-Seq study; characterized
(2009) blastomere expression state
Tang et al. Early embryo, mESC, ICM Yes Mouse 34 10,815 Tang Discovered a metabolic switch from ICM
(2010b) outgrowth cells to ESCs
Islam et al. mESC, MEF No Mouse 85 4250 STRT- STRT-Seq is described and can pinpoint
(2011) Seq the exact location of the 5′ end of
transcripts
Ramskold et al. Cancer cell lines, oocyte, Yes Mouse, 38 10,000 Smart- Identified candidate biomarkers of
(2012) CTC, melanocytes, hESC Human Seq circulating tumor cells
Hashimshony Embryo, MEF, mESC Yes C. elegans, 52 5500 CEL- CEL-Seq is described, representing
et al. (2012) mouse Seq advances in processivity and cost
effectiveness
Pan et al. (2013) K562, dorsal root ganglia Yes Human, 3 4706 Custom Optimizes two protocols for sequencing
mouse low-abundance material
Sasagawa et al. mESC, ESC-derived primitive No Mouse 47 NA Quartz- Describes Quartz-Seq
(2013) endoderm Seq
Yan et al. (2013) Oocyte, zygote, 2-cell, 4-cell, Yes Human 124 11,006 Tang A comprehensive transcriptomic profiling of
8-cell, morula, late blast, human pre-implantation embryos and
hESC ESCs
Xue et al. (2013) Oocyte, pronucleus, zygote, Yes Human, 37 10,231 Tang Discovers that paternal-specific single
2-cell, 4-cell, 8-cell mouse nucleotide polymorphisms can be
detected as early as the 2-cell stage
Islam et al. ESC No Mouse 41 7595 STRT- Introduction of UMIs to enable mRNA to
(2013) Seq ameliorate the issue of PCR duplications
during amplification
Deng et al. Zygote, 2-cell, 4-cell, 8-cell, Yes Mouse 298 NA Smart- Global analysis of allelic expression on
(2014) 16-cell, early blast, mid Seq mouse pre-implantation embryos;
blast, late blast, M-II revealed that random monoallelic
oocyte, fibroblasts, liver expression results from stochastic allelic
transcription
Grün et al. ESC No Mouse 118 6235 CEL- Proposed a noise model to correct for
(2014) Seq sampling noise and global cell-to-cell
variation in sequencing efficiency
Kumar et al. ESC No Mouse 415 NA Smart- Showed that transcriptional heterogeneity
(2014) Seq is regulated and associated with
expression of lineage specifiers; loss of
mature miRNA pushes ESCs to a low-
noise state
Satija et al. Embryo Yes Zebrafish 851 3400 SCRB- Description of Seurat to computationally
(2015) seq reconstruct the spatial organization of
zebrafish embryos
Klein et al. ESC No Mouse 5212 NA inDrop High-throughput droplet-microfluidic
(2015) approach applied to RNA-Seq thousands
of single cells
Cacchiarelli Fibroblasts, PSC No Human 52 NA Smart- Suggested that reprogramming reflects
et al. (2015) Seq aspects of development in reverse
Kim et al. (2015) ESC No Mouse 54 7385 Smart- Described a generative statistical model to
Seq quantify technical noise using spike-ins
CEL-Seq, cell expression by linear amplification and sequencing; CTCs, circulating tumor cells; ESC, embryonic stem cell; hESC, human embryonic stem cell;
MEF, mouse embryonic fibroblast; mESC, mouse embryonic stem cell; ICM, inner cell mass; scRNA-Seq, single-cell RNA sequencing; STRT-Seq, single-cell
tagged reverse transcription sequencing; NA, not applicable; PSC, pluripotent stem cell; UMIs, unique molecular identifiers.

Indeed, similar to the findings based on qPCR analyses, it was shown et al., 2004). This, however, has changed with the development of
that lineage-specific markers exhibit promiscuous co-expression prior single-cell-based approaches. Indeed, to more finely map human
to lineage maturation between E3 and E5. For example, co-expression ZGA, scRNA-Seq was carried out on 33 cells isolated from human
of TE- (GATA2 and GATA3), PE- (GATA4 and PDGFRA) and EPI- pre-implantation embryos, ranging from the zygote to the 8-cell
DEVELOPMENT

(SOX2 and TDGF21) indicative genes was observed before the three stage, all of which had been derived by intra-cytoplasmic sperm
distinct groups of cells were labeled at late E5. injection from a single sperm donor (Xue et al., 2013). Using
this approach, maternal and paternal transcripts in single cells
Human zygotic genome activation could be distinguished based on paternal-specific single-nucleotide
The dynamics of human zygotic genome activation (ZGA, also polymorphisms (SNPs), and it was found that the expression of
referred to as embryonic genome activation or EGA) have remained paternal alleles occurs as early as the 2-cell stage, followed by major
elusive for many years because it is difficult to obtain the numbers ZGA in the 4- to 8-cell stages. These findings were corroborated
of precisely timed human embryos that would be required for in a scRNA-Seq-based analysis of 124 human embryonic cells,
traditional, bulk molecular profiling (Braude et al., 1988; Dobson including zygotes and cells from the 2-cell, 4-cell, 8-cell, morula

22
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

and late blastocyst stages (Yan et al., 2013). Based on the sheer shown that X-chromosome genes exhibit lingering bi-allelic
number of genes that are differentially expressed between the 4-cell expression, which is absent at later stages (Petropoulos et al., 2016).
and the 8-cell stage, and because the genes upregulated are enriched
in ribosome and RNA metabolism functions, it was concluded that Using scRNA-Seq to gain insights into the biology of stem
the major phase of ZGA occurs at this stage. This is in contrast to cells
ZGA dynamics in the mouse, where the major phase of ZGA was The scRNA-Seq studies discussed above focused primarily on gene
found by scRNA-Seq to occur between the zygote and late 2-cell expression in early embryos, but it soon became clear that such an
stage (Blakeley et al., 2015). In spite of this difference in the timing approach could also be used to further understand the biology of
of ZGA, a high degree of conservation between the human and different types of stem cells. Indeed, and as we highlight briefly
mouse pre-implantation development genetic programs was below, scRNA-Seq studies carried out in just the past few years have
observed (Xue et al., 2013). By performing network and gene begun to answer some key questions in the stem cell field.
enrichment analysis, it was shown that the genetic networks
coinciding with the three waves of ZGA/EGA in human and mouse The relationship between stem cell states
embryos share analogous cellular functions. For example, networks Embryonic stem cells (ESCs) are derived by explant culture of day 4.5
activated in the early ZGA wave are enriched in protein transport (murine) or day 8 (human) embryos; however, the precise relationship
and GTPase signaling genes (at the 1- to 4-cell stage in human, the between ESCs and the cells from which they originate in vivo has
1- to 2-cell stage in mouse), networks activated in the major ZGA remained ill-defined (Nichols and Smith, 2011). To address this issue,
wave are highly enriched in RNA processing and ribosome scRNA-Seq has been used to elucidate the precise changes that
biogenesis genes (at the 8-cell stage in human, and the 2- to 4-cell accompany the transition of human inner cell mass cells to human
stage in mouse), and networks activated in the final wave are ESCs (Tang et al., 2010a). This study revealed a switch in the levels of
enriched in translation and mitochondrial genes (at the 16-cell stage genes encoding metabolic factors, as well as increases in the levels of
in human, and the 8- to 16-cell stage in mouse). This suggests that genes encoding epigenetic repressors, although it should be noted that
the regulation of these conserved genetic programs is decoupled (to the study was limited in the number of cells profiled. Similarly,
some extent) from the number of cell cycles post-fertilization, scRNA-Seq has leverage to identify transcriptional changes that occur
raising the issue of how the waves of ZGA/EGA are timed. during the reprogramming of cells to induced pluripotent stem cells
(iPSCs). A pioneering single-cell qPCR study discovered that the
Blastomere asymmetry reprogramming process is divided into an early, rate-limiting stochastic
Another elusive facet of early human embryogenesis is the timing of phase followed by a deterministic phase (Buganim et al., 2012).
the first symmetry-breaking event – the moment at which seemingly Subsequent studies that have applied scRNA-Seq to reprogramming
equivalent blastomeres start to exhibit differences. Using scRNA- have refined this model, finding that reprogramming follows
Seq, it was demonstrated that the transcriptional profile of the zygote development ‘in reverse’ (Cacchiarelli et al., 2015).
was distinct when compared with that of other cleavage-stage
embryos (Xue et al., 2013), an observation that was further Transcriptional heterogeneity and pluripotency
substantiated in 2015 by a computational meta-analysis of scRNA- Related to how ESCs are derived is the question ‘how is this artificial
Seq data (Shi et al., 2015). Comparing scRNA-Seq data with a state is maintained in culture?’. The role of transcriptional
theoretical prediction of a biased distribution of transcripts suggested heterogeneity in pluripotency has been a subject of debate since
that the asymmetric distribution of transcripts occurs at the point fluctuations in the levels of Nanog and other pluripotency factors in
immediately after the first embryonic cleavage. This biased mouse ESCs were first reported (Chambers et al., 2007; Niwa et al.,
distribution of transcripts follows a binomial pattern, which means 2009; Toyooka et al., 2008). One hypothesis is that transcriptional
that when the RNA copy number is low, the transcript will be less heterogeneity in lineage regulators or signaling components affords
evenly distributed to daughter cells after the first cleavage. In the stem cells reversible opportunities to exit the pluripotent state if
subsequent 2- to 16-cell stages, asymmetrically distributed transcripts conditions are permissive, resulting in a meta-stable state (reviewed
diverge into either being minimized through negative-feedback loops by MacArthur et al., 2009; Cahan and Daley, 2013). This hypothesis
or enhanced through positive-feedback loops, suggesting that has been explored by applying scRNA-Seq to 183 mouse ESCs
transcriptional noise (see Glossary, Box 1) is initially important but cultured in traditional conditions (LIF and serum on feeders) (Kumar
then has progressively minimal impact during lineage specification. et al., 2014). The authors indeed found that the expression of some
pluripotency regulators (e.g. Essrb) is bimodal. Most interesting was
Allele-specific gene expression the discovery of genes that are sporadically expressed, i.e. that are
Although allelic exclusion has been linked to diverse biological expressed at a high level in a few cells but not detected in the rest.
functions, including T-cell receptor expression and antigen Both Polycomb targeted genes, which define lineage regulators, and
recognition in B cells (Brady et al., 2010), its genome-wide components of developmental signaling pathways are enriched in
prevalence has been unclear. However, in 2014, this issue was these sporadically expressed gene sets. Furthermore, the expression
addressed by determining allele-specific expression in 269 single of Polycomb target genes as a whole correlates with the expression of
DEVELOPMENT

cells isolated from pre-implantation stage mouse embryos (Deng variably (both positively and negatively) expressed pluripotency
et al., 2014; Ramskold et al., 2012). By crossing mice of different factors, implying the presence of genetic circuits that regulate
backgrounds (CAST and C57), allele-specific expression could be transitions among distinct pluripotent states, thereby offering access
quantified on a per cell basis with scRNA-Seq by using SNPs to to stochastically selected lineage fate choices. The above study also
distinguish alleles. Using this approach, it was estimated that the examined whether transcriptional heterogeneity is affected by
extent of monoallelic expression is surprisingly high (54% of culturing in the presence of GSK and ERK inhibitors (‘2i’), which
genes), a figure that subsequently has been revised to 17.8% after had been reported to reduce heterogeneous expression of some
accounting for technical noise (Kim et al., 2015). More recently, by pluripotency factors, and in mouse ESCs lacking mature
observing SNPs in male pre-implantation human embryos, it was microRNAs, which fail to differentiate. Indeed, it was found that

23
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

the expression of pluripotency-associated genes is substantially less model of continuous transition from RG to IPC to neuron that had
heterogeneous in mouse ESCs cultured in 2i than in either mouse been proposed based on FACS-isolated bulk RNA-Seq and scRNA-
ESCs cultured in serum or mouse ESCs lacking mature microRNAs. Seq of fewer cells (Johnson et al., 2015). A separate study (Thomsen
This result was subsequently corroborated by a study that used et al., 2015) reported a novel method for sequencing RNA from
scRNA-Seq of 250 mouse ESCs in serum and LIF, 295 in standard 2i fixed and stained single cells (‘fixed and recovered single cell RNA’
and 159 in alternative ground state conditions (Kolodziejczyk et al., or FRISCR) and applied this to corroborate the distinct vRG and
2015b). Although there was no difference in global transcriptional oRG signatures. Regrettably, none of these studies applied temporal
heterogeneity between conditions, gene sets that included reconstruction to their data, which might have provided new data-
pluripotency factors were more heterogeneous if cells had been driven hints to the temporal relationships between RG cells, IPCs,
cultured in serum and LIF than in either of the ground state neuroblasts and neurons.
conditions. Taken together, these results suggest a model whereby
mouse ESCs are afforded the opportunity to access lineage Pseudotime: understanding lineage progression using
specification programs through stochastic expression of scRNA-Seq
pluripotency factors, which is perhaps facilitated by lower Inferring temporal trajectories from ‘snap-shots’ of single cells has
H3K27me3 at these lineage regulators. However, the extent to already proven to be so attractive that it has created a virtual cottage
which this model is applicable to early fate decisions in transiently industry of computationalists dedicated to devising and improving
pluripotent cells of the blastocyst has not been addressed. new methods. One of the most powerful outcomes of these
methods is the identification of signaling pathways and genetic
Defining and refining cell identity using scRNA-Seq circuits that contribute to cell state transitions, which thereby
molecular profiles generates specific and testable hypotheses. Some notable and
A number of recent studies have also applied scRNA-Seq to study recent examples of applying pseudotime analytics to diverse
post-implantation development and beyond, focusing primarily on developmental contexts include the specification of human
defining the cellular composition of diverse tissues and populations mesoderm (Loh et al., 2016) and endoderm (Chu et al., 2016)
at different developmental time points. These investigations range derivatives from pluripotent stem cells, and the specification of
from defining the molecular profiles of hematopoietic stem cells tissue-resident macrophages from erythroid-myeloid progenitors
(HSCs) from mouse embryos (Zhou et al., 2016) and the (Mass et al., 2016). Here, we discuss the application of temporal
transcriptional landscape of heart development (DeLaughter et al., inference, or pseudotime, methods to explore the progression of
2016; Li et al., 2016a), to comparing cell type diversity in the quiescent neural stem cells (NSCs) to neurons.
embryonic midbrain between human and mouse (La Manno et al., In adult murine brains, NSCs can be found in the subventricular
2016) and obtaining initial cellular censuses of the murine spleen zone (SVZ) and the subgranular zone (SGZ) of the dentate gyrus,
(Jaitin et al., 2014), cortex and hippocampus (Zeisel et al., 2015). although there is functional and phenotypic heterogeneity within NSC
We list many of these studies in Table 2 but note that more studies pools from either region, i.e. individual NSCs differ in their
are being published every month. Here, we focus our discussion on proliferative tendency and their expression of selected NSC-marker
human radial glial (RG) cell diversity in the developing neocortex, genes. This issue of heterogeneity was recently examined by applying
which has been the subject of several distinct studies. scRNA-Seq to NSCs and their progeny isolated from the dentate gyrus
RG cells, which give rise to most of the neurons of the neocortex (at one time point), as marked by Nestin-CFP (Shin et al., 2015). Six
and, at subsequent stages of development, to astrocytes (Kriegstein different states were identified and the pseudotime algorithm Waterfall
and Alvarez-Buylla, 2009), reside in the ventricular and outer was used to place cells from five of these states onto a continuous
subventricular zone of the neocortex. RG cells from these regions progression from quiescent NSCs to intermediate progenitor cells.
(termed vRG and oRG cells, respectively) have distinct functional Enrichment analysis across these states uncovered a gradual decrease
and morphological characteristics, but the molecular profiles that in expression of Acyl-CoA synthetases and components of the
determine these traits have remained elusive, as have their glycolytic metabolism machinery, and a concomitant upregulation of
relationship to each other and to intermediate progenitor cells ribosomal and spliceosome genes, and genes involved in oxidative
(IPCs), owing to the inability to prospectively isolate pure and phosphorylation. Based on these findings, it was proposed that the
relatively unharmed vRG and oRG cells. This challenge was sequential reduction of signaling pathway genes reflects the importance
overcome by a pair of studies, the first of which was a proof-of- of the niche role served by NSCs of both maintaining the NSC state and
principle analysis demonstrating the feasibility of single-cell allowing it to respond rapidly to perturbations therein.
profiling of single human neocortex-derived primary cells (Pollen In a different study, scRNA-Seq was applied to prospectively
et al., 2014). In this study, the authors applied scRNA-Seq to 24 isolated populations of NSCs and neuroblasts from the SVZ
cells from the developing (gestational week 16-21) human (Llorens-Bobadilla et al., 2015). Dimension reduction analysis
neocortex and were able to find expression signatures that clearly identified three distinct populations corresponding to
distinguish RG cells from newborn neurons. They presented data oligodendrocytes, NSCs and neuroblasts. Unsupervised hierarchical
suggesting that even low-coverage sequencing (∼50,000 reads per clustering of the cells revealed four distinct NSC states. By ordering
DEVELOPMENT

cell) can be sufficient for gross cell type classification. By profiling these cells using Monocle, the authors were able to attribute each
393 human cortical germinal zone cells using scRNA-Seq, the same NSC state to a developmental time point spanning from a dormant
group later distinguished oRG from vRG cells. They found that stage to a primed quiescent stage, to an early activated stage and
oRG cells are enriched for genes related to cellular migratory finally to a dividing stage. Similar to the metabolic and ribosomal
behavior and extracellular matrix, such as HOPX and TNC, whereas dynamics of NSC differentiation in the SGZ, SVZ NSCs also express
vRG express CRYAB, PDGFD, TAGLN2, FBXO32 and PALLD relatively higher levels of glycolytic and fatty acid metabolism genes
(Pollen et al., 2015). They also described a transcriptional state that in the quiescent stages and lower levels of ribosomal genes. However,
characterized putative intermediate progenitors, but found that the unlike the situation with SGZ NSCs, it was found that subsets of SVZ
distinct nature of this state was not readily compatible with the NSC transcription factors reflective of distinct neuronal sub-types

24
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Table 2. Single-cell RNA-Seq analyses of differentiated cell types

Study Cell type(s) Primary Species Cells Genes Method Summary
Brennecke et al. Root quiescent center cells, root Yes Arabidopsis, 104 NA Tang Method to model gene-specific
(2013) epidermis and murine immune Human biological and technical
transcriptional noise
Picelli et al. HEK293T, DG-75, C2C12, MEF No Human 252 10,000 Smart- Smart-Seq2 protocol introduced;
(2013) Seq resulted in increased cDNA length
and yield
Wu et al. (2013) HCT116 No Human 109 4750 Smart- Evaluation of sensitivity and accuracy
Seq of various single-cell RNA-Seq
methods
Grindberg et al. Dentate gyrus neurons, NPC Yes Mouse 8 NA Tang Demonstration of nuclear isolation
(2013) nucleus, NPC cell followed by scRNA-Seq in tissues
where intact cell isolation is difficult
Marinov et al. Lymphoblastoid cell line No Human 15 NA Smart- Absolute quantification of RNA
(2014) Seq molecules per gene using spike-in-
based quantification
Jaitin et al. Spleen Yes Mouse 4590 553 MARS- Automated massively parallel single-
(2014) Seq cell RNA sequencing for in vivo
sampling of multiple cells and
FACS-based sorting into wells
Trapnell et al. Myoblast Yes Human 279 5273 Smart- Describes Monocle: one of the first
(2014) Seq pseudotime algorithms
Treutlein et al. Lung epithelia Yes Mouse 198 3500 Smart- Characterized cell types and
(2014) Seq developmental hierarchies in the
developing lung
Brunskill et al. Kidney Yes Mouse 235 NA Smart- Created an atlas of gene expression
(2014) Seq profiles in different stages of kidney
development and provided
evidence for multi-lineage priming
Shalek et al. BMDC Yes Mouse 1775 6313 Smart- Highlighted the importance of
(2014) Seq intercellular communication in
establishing cell heterogeneity, and
showed modes of establishment of
complex dynamic responses of
multicellular populations
Patel et al. Glioblastoma Yes Human 430 NA Smart- Revealed cellular heterogeneity in
(2014) Seq regulatory programs pertinent to the
biology and hence treatment of
glioblastoma tumors
Pollen et al. CML line, hiPSC, keratinocytes, Yes Human 301 5000 Smart- Established a strategy to compare
(2014) ductal carcinoma, Seq heterogeneous cell populations in
lymphoblastoid cells, APL cells, an unbiased manner using
fibroblast, neural progenitors, microfluidics-based cell capture
fetal neurons followed by low coverage
sequencing
Zeisel et al. Neocortical and hippocampus Yes Mouse 3315 15,310 STRT- Characterized the cell types present in
(2015) neurons Seq the mouse cortex and
hippocampus, and identified novel
cell types and their corresponding
marker genes
Macosko et al. HEK293T, fibroblasts and retinal Yes Human, 50,401 6177.5 Smart- Description of Drop-Seq: a
(2015) cells mouse Seq microfluidics-based platform that
enables capture, lysis and
barcoding of thousands of single
cells
Llorens- Adult radial glial cells, neuroblasts Yes Mouse 130 NA Smart- Identified lineage-specific neural stem
Bobadilla et al. Seq cells in the subventricular zone
(2015)
Shin et al. (2015) Neural progenitors Yes Mouse 168 NA Smart- Characterized adult hippocampal
Seq quiescent neural stem cells;
DEVELOPMENT

described Waterfall, another

pseudotime algorithm
Pollen et al. Cortical germinal zone cells Yes Human 393 NA Smart- Revealed that radial glia located in the
(2015) Seq outer subventricular zone support
brain expansion by increasing
proliferative potential at the niche
during cortical development
Continued

25
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Table 2. Continued
Study Cell type(s) Primary Species Cells Genes Method Summary
Thomsen et al. Radial gial, intermediate Yes Human 255 NA Smart- Developed FRISCR (fixed and
(2015) progenitor cells Seq recovered intact single-cell RNA),
which can profile transcriptomes of
individual cells
Hanchate et al. Olfactory sensory neurons Yes Mouse 85 NA Smart- Discovered that, unlike mature
(2015) Seq olfactory neurons [which express
only one of the 1000 odorant
receptors (Olfrs)], immature
neurons can express multiple Olfrs
Camp et al. iPSC and ESC-derived cerebral No Human 508 NA Smart- Comparison of cerebral organoids and
(2015) organoid cells Seq fetal neocortex with regards to cell
composition and progenitor-to-
neuron lineage relationships
Li et al. (2016b) Pancreatic islet cells Yes Human 64 NA Smart- Identified TFs specific to islet
Seq subtypes
Tasic et al. Cortical neurons Yes Mouse 1679 NA Smart- Constructed a cellular taxonomy of the
(2016) Seq primary visual cortex in adult mice
Angermueller ESC No Mouse 61 5000 G&T- Developed scM&T-Seq, which allows
et al. (2016) Seq transcriptome and methylome
profiling of single cells
Macaulay et al. Hematopoietic progenitors Yes Zebrafish 363 3500 Smart- Refined the conventional lineage tree
(2016) Seq of hematopoiesis to thrombocytes
Xin et al. (2016) Pancreatic islet cells Yes Mouse 341 NA Smart- Assessed the Fluidigm C1 system
Seq using islets as the cell source and
discovered limitations in the cell
capture microfluidic device
Zhou et al. Pre-HSC and HSC Yes Mouse 99 5875 Tang Dissected the molecular mechanisms
(2016) involved in the stepwise generation
of hematopoietic stem cells and
characterized purified nascent pre-
HSCs
Eltahla et al. T cells Yes Human 56 NA Smart- Proposed a novel method
(2016) Seq (VDJpuzzle) to study T-cell
heterogeneity by linking gene
expression profiles; reconstructed
TCRαβ using scRNA-Seq of Ag-
specific T cells
Gao et al. (2016) Dentate gyrus neurons Yes Mouse 84 NA Smart- Classified postnatal immature neurons
Seq into distinct developmental lineages
as they show diverging expression
profiles
Liu et al. (2016) Neocortical cells Yes Human 226 NA Smart- Profiled lncRNA expression in human
Seq neocortical cells by performing
strand-specific scRNA-Seq during
various developmental stages
Nelson et al. Placenta cell Yes Mouse 448 15,402 Tang Provided insights into the various cell
(2016) types present in the maternal-fetal
interface
Petropoulos Pre-implantation embryonic Yes Human 1529 8500 Smart- Provided a comprehensive
et al. (2016) tissue Seq transcriptional map of human pre-
implantation development,
revealing lineage and X-
chromosome dynamics
Nowakowski Cerbral organoid cell No Human 210 NA Smart- Explored putative Zika virus entry
et al. (2016) Seq proteins in neural stem cells
Li et al. (2016c) Growth plate cells Yes Mouse 217 9000 Smart- Developed Sinova, a spatial
Seq reconstruction method; used the
pipeline to analyze growth-plate
development with high temporal and
DEVELOPMENT

spatial resolution
Loh et al. (2016) ESC-derived mesoderm No Human 651 NA Smart- Provided a stepwise map of
Seq developmental pathways that
specify diverse mesoderm-derived
lineages
Gokce et al. Striatum Yes Mouse 1208 NA Smart- Constructed the cellular taxonomy of
(2016) Seq the mouse striatum and revealed the
Continued

26
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Table 2. Continued
Study Cell type(s) Primary Species Cells Genes Method Summary
diversity between the various striatal
cell types
Mass et al. Tissue-resident macrophages Yes Mouse 408 NA MARS- Analyzed the specification of tissue-
(2016) Seq resident macrophages and
proposed their subsequent
differentiation to be an integral part
of organogenesis
Chu et al. (2016) Pluripotent stem cell differentiated No Human 1776 NA Smart- Elucidated novel regulators of
endoderm derivatives Seq mesendoderm transition to
definitive endoderm by combining
scRNA-Seq and genetic
approaches
Tintori et al. Early embryo Yes C. elegans 219 8575 Smart- Provided a resource and a
(2016) Seq visualization tool for the
transcriptional profiles of each cell
until the 16-cell stage of the C.
elegans embryo
Gury-BenAri Intestinal innate lymphoid cells Yes Mouse 1129 NA MARS- Identified diversity in innate lymphoid
et al. (2016) Seq cells of the gut that results from
signaling from the local microbiome
population
Habib et al. Hippocampal neurons Yes Mouse 1367 5100 Smart- Described a method to sequence
(2016) Seq individual dividing cells by
combined single-nucleus RNA-Seq
(sNuc-Seq) with EdU pulse labeling
Olsson et al. Multipotent progenitor cells; Yes Mouse 382 NA Smart- Combined iterative clustering and
(2016) common myeloid progenitor Seq guide-gene selection with scRNA-
cells; granulocyte monocyte Seq to dissect mixed lineage states
progenitor cells of a multipotent progenitor
population into macrophage or
neutrophil lineage specification
Nath et al. ALA neuron Yes C. elegans 9 8133 STRT- Investigated downstream
(2016) Seq mechanisms of a neuro-secretory
cell that promotes sleep
Yu et al. (2016) Bone marrow cells Yes Mouse 497 8758 Smart- Identified a molecular marker for a
Seq novel innate lymphoid cell precursor
that could potentially be
manipulated for use in
immunotherapy
La Manno et al. Ventral midbrain cell Yes Mouse, 3884 NA STRT- Analyzed the time-course of ventral
(2016) Human Seq midbrain development and provided
a method to assess the fidelity of
iPSC-derived dopaminergic
neurons
Kee et al. (2016) Lmx1a neuron Yes Mouse 550 NA Smart- Revealed a relationship between
Seq differentiating dopamine and sub-
thalamic nucleus lineages, which
could have implications in the
treatment of Parkinson’s disease
DeLaughter 129SV cardiac cells Yes Mouse 1133 NA Smart- Obtained dynamic spatiotemporal
et al. (2016) Seq gene expression profiles for distinct
cardiomyocyte populations across
development
Li et al. (2016a) Murine heart cells Yes Mouse 2233 NA Smart- Uncovered chamber-specific genes in
Seq the embryonic mouse heart
BMDC, bone marrow-derived dendritic cell; CML, chronic myeloid leukemia; FACS, fluorescence-activated cell sorting; FRISCR, fixed and recovered single cell
RNA; HSC, hematopoietic stem cell; iPSC, induced pluripotent stem cell; MARS-Seq, massively parallel single-cell RNA sequencing; NA, not applicable; NPC,
neural progenitor cell; MEF, mouse embryonic fibroblast; scM&T-Seq, single-cell genome-wide methylome and transcriptome sequencing; STRT-Seq, single-cell
DEVELOPMENT

tagged reverse transcription sequencing; TFs, transcription factors.

(e.g. dorsal, ventral, dorsolateral) were correlated, consistent with the progenitor stage during which the cell type-specific expression
notion that both active and quiescent NSCs are predisposed or even programs of related but distinct cell types are simultaneously
committed towards specific lineages. co-expressed (Fig. 3). Evocative of Janus – the two-faced Roman
deity of gateways and transitions – this state would comprise
Insights into the ‘Janus’ progenitor state multiple transcriptional programs that, later in development, are
It is possible that, between the ∼40 rounds of cell divisions through uniquely attributable to a single cell type. Such a state was identified
which the zygote gives rise to all mature cell types, there is a when studying the events that regulate the developmental

27
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

A was hypothesized that this population represented a bipotent

E16.5 alveolar progenitor of AT1 and AT2 cells. This hypothesis was consistent
progenitor with single-cell qPCR data of E16.5 alveolar progenitors, which
also express markers of both alveolar progenitors, and implies that
the AT1/AT2 fate choice entails the active repression of alternative
lineages rather than selective activation. Notably, bipotent
AT1 AT2 progenitors were shown by immunofluorescence to co-express
AT1 and AT2 marker genes, making it unlikely that they represent
expression profiles of doublets, as has been reported in the context
of pancreatic islets (Xin et al., 2016) and in species-mixing
experiments (Macosko et al., 2015).
The observation of this dual state has raised several provocative
B
questions. First, by what mechanisms are the genetic programs of
E11.5 alternate lineages repressed? Hopx, a transcriptional repressor that
Six2 has been implicated in maturation in a wide range of lineages, was
metanephric
Foxd1
mesenchyme found to mark AT1 cells in this study, but it is also expressed in
bipotent progenitors, so its expression cannot be the initiating cell
fate event. In fact, no AT-specific lineage factor appears to be
Six2 Nephron induced. Therefore, it is possible that transcriptional repressors of
Stromal cells
Foxd1
the alternate lineage were undetectable using scRNA-Seq due to low
copy number, or that alternative mechanisms of repression, such as
microRNAs, which currently are not profiled in scRNA-Seq, are
major contributors to the differentiation of these alveolar lineages.
Alternatively, post-transcriptional events could be the major drivers
of this fate decision.
C Gfi1
Second, how pervasive are dual states in development? As
Bipotent more scRNA-Seq studies are performed, we will gain a better
progenitors sense of this, but there are already some hints that it is not an
Irf8
idiosyncrasy of the lung. Reminiscent of the bipotent progenitor
dual-state is the observation that Foxd1, which marks stromal-
Neutrophil
committed cells, and Six2, which marks nephron-committed
>?", Macrophage cells, are co-expressed in single cells of E11.5 metanephric
Gfi1
Irf8 mesenchyme (Brunskill et al., 2014). This observation was made
originally using single-cell microarrays and scRNA-Seq, and co-
expression was also confirmed at the protein level, albeit at a
lower frequency. At the time, this observation was attributed
Fig. 3. The ‘Janus’ progenitor state. scRNA-Seq has enabled the to stochastic expression because, unlike the lung bipotent
identification of embryonic progenitors that simultaneously express genes that progenitor cells, the dual-expressing metanephric mesenchyme
were previously suspected of being lineage specific. (A) The PCA analysis of cells do not otherwise reflect an ensemble of stromal and
scRNA-Seq profiles of 198 developing murine lung cells has identified a cluster nephron progenitor profiles. However, it is possible that the
that expresses markers for both AT1 and AT2 cells, corroborating with single- co-expressing metanephric mesenchyme cells represent the tail
cell qPCR data of E16.5 alveolar progenitors. (B) scRNA-Seq profiles of
end of a fate-decision process and, therefore, sampling more cells
E11.5 metanephric mesenchyme has identified cells that co-express Foxd1
and Six2, which mark stromal-committing cells and nephron-committing cells,
at earlier time points would clarify this issue.
respectively. (C) A binary cell fate decision between the macrophage lineage Another example of dual-expressing progenitors was uncovered
and the neutrophil lineage was unveiled when bipotent progenitors were shown by the application of scRNA-Seq to 85 developing olfactory sensory
to co-express Irf8 and Gfi1, which regulate macrophage and neutrophil neurons (Hanchate et al., 2015). In this approach, Monocle was
specification, respectively. applied to reconstruct a temporal trajectory of the 85 cells and to
assign them to four distinct classes: progenitors, precursors, and
transformation of the lung bronchial tree into alveolar air sacs immature and mature neurons. The authors found that almost half of
(Treutlein et al., 2014). Analysis of this stage of lung development the immature neurons expressed more than one receptor, and as the
had previously been impaired by the paucity of cells involved and cells matured they exhibited increased expression of a selected
the absence of markers that could be used to isolate pure populations receptor and repressed the alternative receptor genes, a finding
of progenitors. However, these issues were avoided by applying validated by single-molecule FISH. In general, a better
DEVELOPMENT

scRNA-Seq to 198 different cells of the developing murine lung: understanding of when dual states are employed, and of the
E14.5 bronchial progenitors, E16.5 cells undergoing sacculation, molecular basis by which they are initiated, permitted and resolved,
E18.5 distal lung epithelial cells and mature alveolar type 2 (AT2) will enable us to speculate on a more fundamental question: what is
cells (Treutlein et al., 2014). The authors used PCA to define the purpose of this duality? Does it allow progenitors to perform
distinct cell types within the 80 E18.5 cells to find five major functions during development that are later distributed to more-
clusters or groups of cells. By examining the expression of genes specialized cell types? Does it provide greater robustness to
representative of distinct lung cell types, they could annotate four of environmental perturbations during development? Or is it simply
these groups of cells as either epithelial, ciliated, AT1 or AT2. a neutral consequence of how cell type-specific circuitry is encoded
Because the fifth group expressed markers of both AT1 and AT2, it and elaborated during development?

28
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Conclusions Braude, P., Braude, P., Bolton, V., Bolton, V., Moore, S. and Moore, S. (1988).
Human gene expression first occurs between the four- and eight-cell stages of
As we have summarized here, scRNA-Seq-based approaches are preimplantation development. Nature 332, 459-461.
being used increasingly to provide insights into various aspects of Brennecke, P., Anders, S., Kim, J. K., Kołodziejczyk, A. A., Zhang, X.,
developmental and stem cell biology. Such studies have defined Proserpio, V., Baying, B., Benes, V., Teichmann, S. A., Marioni, J. C. et al.
the transcriptional programs of the earliest stages of mammalian (2013). Accounting for technical noise in single-cell RNA-seq experiments. Nat.
Methods 10, 1093-1095.
development, have implicated regulated transcriptional heterogeneity Brunskill, E. W., Park, J.-S., Chung, E., Chen, F., Magella, B. and Potter, S. S.
as a contributor to the pluripotent state, and have uncovered an (2014). Single cell dissection of early kidney development: multilineage priming.
unexpected yet widespread pattern of dual identity in embryonic Development 141, 3093-3101.
Buenrostro, J. D., Wu, B., Litzenburger, U. M., Ruff, D., Gonzales, M. L.,
progenitors. However, there are several substantial obstacles to Snyder, M. P., Chang, H. Y. and Greenleaf, W. J. (2015). Single-cell
realizing the full potential of scRNA-Seq when applied to chromatin accessibility reveals principles of regulatory variation. Nature 523,
development. First, the isolation of intact, unperturbed single cells 486-490.
Buettner, F., Natarajan, K. N., Casale, F. P., Proserpio, V., Scialdone, A., Theis,
from dissociated tissue remains a major challenge. Methods such as
F. J., Teichmann, S. A., Marioni, J. C. and Stegle, O. (2015). Computational
FRISCR, in which cells are fixed rapidly, promise to ameliorate some analysis of cell-to-cell heterogeneity in single-cell RNA-sequencing data reveals
of these issues. Second, what is a biological replicate of a single cell? hidden subpopulations of cells. Nat. Biotechnol. 33, 155-160.
Several groups have attempted cell splitting to assess technical Buganim, Y., Faddah, D. A., Cheng, A. W., Itskovich, E., Markoulaki, S., Ganz,
K., Klemm, S. L., van Oudenaarden, A. and Jaenisch, R. (2012). Single-cell
variability but this has not become widespread in practice, likely due expression analyses during cellular reprogramming reveal an early stochastic and
to technical challenges in evenly splitting cells and maintaining a a late hierarchic phase. Cell 150, 1209-1222.
modicum of sensitivity. In general, distinguishing technical noise Cacchiarelli, D., Trapnell, C., Ziller, M. J., Soumillon, M., Cesana, M., Karnik, R.,
Donaghey, J., Smith, Z. D., Ratanasirintrawoot, S., Zhang, X. et al. (2015).
from true biological noise is an area of very active research, and Integrative analyses of human reprogramming reveal dynamic nature of induced
synthetic RNA spike-ins (see Glossary, Box 1) have so far proven to pluripotency. Cell 162, 412-424.
be the most common approach to deal with this issue (Marinov et al., Cahan, P. and Daley, G. Q. (2013). Origins and implications of pluripotent stem cell
2014). Third, most scRNA-Seq methods are polyA-centric, thus variability and heterogeneity. Nat. Rev. Mol. Cell Biol. 14, 357-368.
Camp, J. G., Badsha, F., Florio, M., Kanton, S., Gerber, T., Wilsch-Brä uninger,
limiting our ability to measure non-polyA RNA at a single cell level. M., Lewitus, E., Sykes, A., Hevers, W., Lancaster, M. et al. (2015). Human
Perhaps the most significant barrier is the limited efficiency of reverse cerebral organoids recapitulate gene expression programs of fetal neocortex
transcription, which leads to limited sensitivity. A final substantial development. Proc. Natl. Acad. Sci. USA 112, 15672-15677.
Chambers, I., Silva, J., Colby, D., Nichols, J., Nijmeijer, B., Robertson, M.,
challenge in the field will be the development of computational and Vrana, J., Jones, K., Grotewold, L. and Smith, A. (2007). Nanog safeguards
experimental approaches that enable some level of data integration pluripotency and mediates germline development. Nature 450, 1230-1234.
across studies. As these issues are tackled, and as scRNA-Seq is Chen, H., Guo, J., Mishra, S. K., Robson, P., Niranjan, M. and Zheng, J. (2015).
applied more broadly, we anticipate a time when there will be enough Single-cell transcriptional analysis to uncover regulatory circuits driving cell fate
decisions in early mouse development. Bioinformatics 31, 1060-1066.
data to develop quantitative definitions of cell type identity Chu, L.-F., Leng, N., Zhang, J., Hou, Z., Mamott, D., Vereide, D. T., Choi, J.,
throughout development. Kendziorski, C., Stewart, R. and Thomson, J. A. (2016). Single-cell RNA-seq
reveals novel regulators of human embryonic stem cell differentiation to definitive
Competing interests endoderm. Genome Biol. 17, 173.
The authors declare no competing or financial interests. Cusanovich, D. A., Daza, R., Adey, A., Pliner, H. A., Christiansen, L.,
Gunderson, K. L., Steemers, F. J., Trapnell, C. and Shendure, J. (2015).
Multiplex single-cell profiling of chromatin accessibility by combinatorial cellular
Funding indexing. Science 348, 910-914.
The authors’ research is funded by the National Institutes of Health (National DeLaughter, D. M., Bick, A. G., Wakimoto, H., McKean, D., Gorham, J. M.,
Institute of Diabetes and Digestive and Kidney Diseases) (K01DK096013). Kathiriya, I. S., Hinson, J. T., Homsy, J., Gray, J., Pu, W. et al. (2016). Single-
Deposited in PMC for release after 12 months. cell resolution of temporal gene expression during heart development. Dev. Cell
39, 480-490.
References Deng, Q., Ramskold, D., Reinius, B. and Sandberg, R. (2014). Single-cell RNA-
Achim, K., Pettit, J.-B., Saraiva, L. R., Gavriouchkina, D., Larsson, T., Arendt, D. seq reveals dynamic, random monoallelic gene expression in mammalian cells.
and Marioni, J. C. (2015). High-throughput spatial mapping of single-cell RNA- Science 343, 193-196.
seq data to tissue of origin. Nat. Biotechnol. 33, 503-509. Dey, S. S., Kester, L., Spanjaard, B., Bienko, M. and van Oudenaarden, A.
Angerer, P., Haghverdi, L., Bü ttner, M., Theis, F. J., Marr, C. and Buettner, F. (2015). Integrated genome and transcriptome sequencing of the same cell. Nat.
(2015). Destiny: diffusion maps for large-scale single-cell data in R. Bioinformatics Biotechnol. 33, 285-289.
32, 1241-1243. Dobson, A. T., Dobson, A. T., Raja, R., Abeyta, M. J., Taylor, T., Shen, S., Haqq,
Angermueller, C., Clark, S. J., Lee, H. J., Macaulay, I. C., Teng, M. J., Hu, T. X., C. and Pera, R. A. R. (2004). The unique transcriptome through day 3 of human
Krueger, F., Smallwood, S. A., Ponting, C. P., Voet, T. et al. (2016). Parallel preimplantation development. Hum. Mol. Genet. 13, 1461-1470.
single-cell sequencing links transcriptional and epigenetic heterogeneity. Nat. Eberwine, J., Yeh, H., Miyashiro, K., Cao, Y., Nair, S., Finnell, R., Zettel, M. and
Methods 13, 229-232. Coleman, P. (1992). Analysis of gene expression in single live neurons. Proc.
Bacher, R. and Kendziorski, C. (2016). Design and computational analysis of Natl. Acad. Sci. USA 89, 3010-3014.
single-cell RNA-sequencing experiments. Genome Biol. 17, 63. Eltahla, A. A., Rizzetto, S., Pirozyan, M. R., Betz-Stablein, B. D., Venturi, V.,
Bandura, D. R., Baranov, V. I., Ornatsky, O. I., Antonov, A., Kinach, R., Lou, X., Kedzierska, K., Lloyd, A. R., Bull, R. A. and Luciani, F. (2016). Linking the T cell
Pavlov, S., Vorobiev, S., Dick, J. E. and Tanner, S. D. (2009). Mass cytometry: receptor to the single cell transcriptome in antigen-specific human T cells.
technique for real time single cell multitarget immunoassay based on inductively Immunol. Cell Biol. 94, 604-611.
coupled plasma time-of-flight mass spectrometry. Anal. Chem. 81, 6813-6822. Etzrodt, M., Endele, M. and Schroeder, T. (2014). Quantitative single-cell
Bendall, S. C., Davis, K. L., Amir, E.-A. D., Tadmor, M. D., Simonds, E. F., Chen, approaches to stem cell research. Cell Stem Cell 15, 546-558.
T. J., Shenfeld, D. K., Nolan, G. P. and Pe’er, D. (2014). Single-cell trajectory Finak, G., McDavid, A., Yajima, M., Deng, J., Gersuk, V., Shalek, A. K.,
DEVELOPMENT

detection uncovers progression and regulatory coordination in human B cell Slichter, C. K., Miller, H. W., McElrath, M. J., Prlic, M. et al. (2015). MAST: a
development. Cell 157, 714-725. flexible statistical framework for assessing transcriptional changes and
Bian, Q. and Cahan, P. (2016). Computational tools for stem cell biology. Trends characterizing heterogeneity in single-cell RNA sequencing data. Genome
Biotechnol. 34, 993-1009. Biol. 16, 278.
Blakeley, P., Fogarty, N. M., del Valle, I., Wamaitha, S. E., Hu, T. X., Elder, K., Fraley, C. and Raftery, A. E. (2002). Taylor & Francis Online: model-based
Snell, P., Christie, L., Robson, P., Niakan, K. K. et al. (2015). Defining the three clustering, discriminant analysis, and density estimation. J. Am. Stat. 97,
cell lineages of the human blastocyst by single-cell RNA-seq. Development 142, 611-631.
3613-3613. Gao, Y., Wang, F., Eisinger, B. E., Kelnhofer, L. E., Jobe, E. M. and Zhao, X.
Brady, B. L., Brady, B. L., Steinel, N. C., Steinel, N. C., Bassing, C. H. and (2016). Integrative single-cell transcriptomics reveals molecular networks defining
Bassing, C. H. (2010). Antigen receptor allelic exclusion: an update and neuronal maturation during postnatal neurogenesis. Cereb. Cortex 143,
reappraisal. J. Immunol. 185, 3801-3808. 1649-1654.

29
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Gokce, O., Stanley, G. M., Treutlein, B., Neff, N. F., Camp, J. G., Malenka, R. C., Kolodziejczyk, A. A., Kim, J. K., Svensson, V., Marioni, J. C. and Teichmann,
Rothwell, P. E., Fuccillo, M. V., Sü dhof, T. C. and Quake, S. R. (2016). Cellular S. A. (2015a). The technology and biology of single-cell RNA sequencing. Mol.
taxonomy of the mouse striatum as revealed by single-cell RNA-seq. Cell Rep. 16, Cell 58, 610-620.
1126-1137. Kolodziejczyk, A. A., Kim, J. K., Tsang, J. C. H., Ilicic, T., Henriksson, J.,
Grindberg, R. V., Yee-Greenbaum, J. L., McConnell, M. J., Novotny, M., Natarajan, K. N., Tuck, A. C., Gao, X., Bü hler, M., Liu, P. et al. (2015b). Single
O’Shaughnessy, A. L., Lambert, G. M., Araú zo-Bravo, M. J., Lee, J., Fishman, cell RNA-sequencing of pluripotent states unlocks modular transcriptional
M., Robbins, G. E. et al. (2013). RNA-sequencing from single nuclei. Proc. Natl. variation. Cell Stem Cell 17, 471-485.
Acad. Sci. USA 110, 19802-19807. Kriegstein, A. and Alvarez-Buylla, A. (2009). The glial nature of embryonic and
Grü n, D., Kester, L. and van Oudenaarden, A. (2014). Validation of noise models adult neural stem cells. Annu. Rev. Neurosci. 32, 149-184.
for single-cell transcriptomics. Nat. Methods 11, 637-640. Kumar, R. M., Cahan, P., Shalek, A. K., Satija, R., Jay DaleyKeyeser, A., Li, H.,
Grü n, D., Lyubimova, A., Kester, L., Wiebrands, K., Basak, O., Sasaki, N., Zhang, J., Pardee, K., Gennert, D., Trombetta, J. J. et al. (2014).
Clevers, H. and van Oudenaarden, A. (2015). Single-cell messenger RNA Deconstructing transcriptional heterogeneity in pluripotent stem cells. Nature
sequencing reveals rare intestinal cell types. Nature 525, 251-255. 516, 56-61.
Grü n, D., Muraro, M. J., Boisset, J.-C., Wiebrands, K., Lyubimova, A., La Manno, G., Gyllborg, D., Codeluppi, S., Nishimura, K., Salto, C., Zeisel, A.,
Dharmadhikari, G., van den Born, M., van Es, J., Jansen, E., Clevers, H. Borm, L. E., Stott, S. R. W., Toledo, E. M., Villaescusa, J. C. et al. (2016).
et al. (2016). De novo prediction of stem cell identity using single-cell Molecular diversity of midbrain development in mouse, human, and stem cells.
transcriptome data. Cell Stem Cell 19, 266-277. Cell 167, 566-580.e19.
Guo, G., Huss, M., Tong, G. Q., Wang, C., Sun, L. L., Clarke, N. D. and Robson, P. Langfelder, P. and Horvath, S. (2008). WGCNA: an R package for weighted
(2010). Resolution of cell fate decisions revealed by single-cell gene expression correlation network analysis. BMC Bioinformatics 9, 559.
analysis from zygote to blastocyst. Dev. Cell 18, 675-685. Lee, J. H., Lee, J.-H., Daugharthy, E. R., Daugharthy, E. R., Scheiman, J.,
Gury-BenAri, M., Thaiss, C. A., Serafini, N., Winter, D. R., Giladi, A., Lara- Scheiman, J., Kalhor, R., Kalhor, R., Yang, J. L., Yang, J. L. et al. (2014).
Astiaso, D., Levy, M., Salame, T. M., Weiner, A., David, E. et al. (2016). The Highly multiplexed subcellular RNA sequencing in situ. Science 343,
spectrum and regulatory landscape of intestinal innate lymphoid cells are shaped 1360-1363.
by the microbiome. Cell 166, 1231-1246.e13. Li, G., Xu, A., Sim, S., Priest, J. R., Tian, X., Khan, T., Quertermous, T., Zhou, B.,
Habib, N., Li, Y., Heidenreich, M., Swiech, L., Avraham-Davidi, I., Trombetta, Tsao, P. S., Quake, S. R. et al. (2016a). Transcriptomic profiling maps
J. J., Hession, C., Zhang, F. and Regev, A. (2016). Div-Seq: single-nucleus anatomically patterned subpopulations among single embryonic cardiac cells.
RNA-Seq reveals dynamics of rare adult newborn neurons. Science 353, Dev. Cell 39, 491-507.
925-928. Li, J., Klughammer, J., Farlik, M., Penz, T., Spittler, A., Barbieux, C., Berishvili,
Haghverdi, L., Buettner, F. and Theis, F. J. (2015). Diffusion maps for high- E., Bock, C. and Kubicek, S. (2016b). Single-cell transcriptomes reveal
dimensional single-cell analysis of differentiation data. Bioinformatics 31, characteristic features of human pancreatic islet cell types. EMBO Rep. 17,
2989-2998. 178-187.
Hanchate, N. K., Kondoh, K., Lu, Z., Kuang, D., Ye, X., Qiu, X., Pachter, L., Li, J., Luo, H., Wang, R., Lang, J., Zhu, S., Zhang, Z., Fang, J., Qu, K., Lin, Y.,
Trapnell, C. and Buck, L. B. (2015). Single-cell transcriptomics reveals receptor Long, H. et al. (2016c). Systematic reconstruction of molecular cascades
transformations during olfactory neurogenesis. Science 350, 1251-1255. regulating GP development using single-cell RNA-seq. Cell Rep. 15, 1467-1480.
Hashimshony, T., Wagner, F., Sher, N. and Yanai, I. (2012). CEL-Seq: single-cell Liu, S. J., Nowakowski, T. J., Pollen, A. A., Lui, J. H., Horlbeck, M. A., Attenello,
RNA-Seq by multiplexed linear amplification. Cell Rep. 2, 666-673. F. J., He, D., Weissman, J. S., Kriegstein, A. R., Diaz, A. A. et al. (2016). Single-
Huang, W., Cao, X., Biase, F. H., Yu, P. and Zhong, S. (2014). Time-variant cell analysis of long non-coding RNAs in the developing human neocortex.
clustering model for understanding cell fate decisions. Proc. Natl. Acad. Sci. USA Genome Biol. 17, 67.
111, E4797-E4806. Llorens-Bobadilla, E., Zhao, S., Baser, A., Saiz-Castro, G., Zwadlo, K. and
Islam, S., Kjä llquist, U., Moliner, A., Zajac, P., Fan, J.-B., Lö nnerberg, P. and Martin-Villalba, A. (2015). Single-cell transcriptomics reveals a population of
Linnarsson, S. (2011). Characterization of the single-cell transcriptional dormant neural stem cells that become activated upon brain injury. Cell Stem Cell
landscape by highly multiplex RNA-seq. Genome Res. 21, 1160-1167. 17, 329-340.
Islam, S., Zeisel, A., Joost, S., La Manno, G., Zajac, P., Kasper, M., Lö nnerberg, Loh, K. M., Chen, A., Koh, P. W., Deng, T. Z., Sinha, R., Tsai, J. M., Barkal, A. A.,
P. and Linnarsson, S. (2013). Quantitative single-cell RNA-seq with unique Shen, K. Y., Jain, R., Morganti, R. M. et al. (2016). Mapping the pairwise choices
molecular identifiers. Nat. Methods 11, 163-166. leading from pluripotency to human bone, heart, and other mesoderm cell types.
Jaitin, D. A., Kenigsberg, E., Keren-Shaul, H., Elefant, N., Paul, F., Zaretsky, I., Cell 166, 451-467.
Mildner, A., Cohen, N., Jung, S., Tanay, A. et al. (2014). Massively parallel Lovatt, D., Ruble, B. K., Lee, J., Dueck, H., Kim, T. K., Fisher, S., Francis, C.,
single-cell RNA-seq for marker-free decomposition of tissues into cell types. Spaethling, J. M., Wolf, J. A., Grady, M. S. et al. (2014). Transcriptome in vivo
Science 343, 776-779. analysis (TIVA) of spatially defined single cells in live tissue. Nat. Methods 11,
Ji, Z. and Ji, H. (2016). TSCAN: pseudo-time reconstruction and evaluation in 190-196.
single-cell RNA-seq analysis. Nucleic Acids Res. 44, e117. Maaten, L. V. D. and Hinton, G. (2008). Visualizing Data using t-SNE. J. Mach.
Jiang, L., Chen, H., Pinello, L. and Yuan, G.-C. (2016). GiniClust: detecting rare Learn. Res. 9, 2579-2605.
cell types from single-cell gene expression data with Gini index. Genome Biol. MacArthur, B. D., Ma’ayan, A. and Lemischka, I. R. (2009). Systems biology of
17, 1. stem cell fate and cellular reprogramming. Nat. Rev. Mol. Cell Biol. 10, 672-681.
Johnson, M. B., Wang, P. P., Atabay, K. D., Murphy, E. A., Doan, R. N., Hecht, Macaulay, I. C. and Voet, T. (2014). Single cell genomics: advances and future
J. L. and Walsh, C. A. (2015). Single-cell analysis reveals transcriptional perspectives. PLoS Genet. 10, e1004126.
heterogeneity of neural progenitors in human cortex. Nat. Neurosci. 18, 637-646. Macaulay, I. C., Haerty, W., Kumar, P., Li, Y. I., Hu, T. X., Teng, M. J., Goolam, M.,
Juliá , M., Telenti, A. and Rausell, A. (2015). Sincell: an R/Bioconductor package Saurat, N., Coupland, P., Shirley, L. M. et al. (2015). G&T-seq: parallel
for statistical assessment of cell-state hierarchies from single-cell RNA-seq. sequencing of single-cell genomes and transcriptomes. Nat. Methods 12,
Bioinformatics 31, 3380-3382. 519-522.
Ke, R., Mignardi, M., Pacureanu, A., Svedlund, J., Botling, J., Wä hlby, C. and Macaulay, I. C., Svensson, V., Labalette, C., Ferreira, L., Hamey, F., Voet, T.,
Nilsson, M. (2013). In situ sequencing for RNA analysis in preserved tissue and Teichmann, S. A. and Cvejic, A. (2016). Single-cell RNA-sequencing reveals a
cells. Nat. Methods 10, 857-860. continuous spectrum of differentiation in hematopoietic cells. Cell Rep. 14,
Kee, N., Volakakis, N., Kirkeby, A., Dahl, L., Storvall, H., Nolbrant, S., Lahti, L., 966-977.
Bjö rklund, Å. K., Gillberg, L., Joodmardi, E. et al. (2016). Single-cell analysis Macosko, E. Z., Basu, A., Satija, R., Nemesh, J., Shekhar, K., Goldman, M.,
reveals a close relationship between differentiating dopamine and subthalamic Tirosh, I., Bialas, A. R., Kamitaki, N., Martersteck, E. M. et al. (2015). Highly
nucleus neuronal lineages. Cell Stem Cell 20, 1-12 parallel genome-wide expression profiling of individual cells using nanoliter
DEVELOPMENT

Kharchenko, P. V., Silberstein, L. and Scadden, D. T. (2014). Bayesian approach droplets. Cell 161, 1202-1214.
to single-cell differential expression analysis. Nat. Methods 11, 740-742. Marco, E., Karp, R. L., Guo, G., Robson, P., Hart, A. H., Trippa, L. and Yuan,
Kiel, M. J., Yilmaz, Ö . H., Iwashita, T., Yilmaz, O. H., Terhorst, C. and Morrison, G.-C. (2014). Bifurcation analysis of single-cell gene expression data reveals
S. J. (2005). SLAM family receptors distinguish hematopoietic stem and epigenetic landscape. Proc. Natl. Acad. Sci. USA 111, E5643-E5650.
progenitor cells and reveal endothelial niches for stem cells. Cell 121, 1109-1121. Marinov, G. K., Williams, B. A., McCue, K., Schroth, G. P., Gertz, J., Myers, R. M.
Kim, J. K., Kolodziejczyk, A. A., Illicic, T., Teichmann, S. A. and Marioni, J. C. and Wold, B. J. (2014). From single-cell to cell-pool transcriptomes: stochasticity
(2015). Characterizing noise structure in single-cell RNA-seq distinguishes in gene expression and RNA splicing. Genome Res. 24, 496-510.
genuine from technical stochastic allelic expression. Nat. Commun. 6, 8687. Mass, E., Ballesteros, I., Farlik, M., Halbritter, F., Gü nther, P., Crozet, L.,
Klein, A. M., Mazutis, L., Akartuna, I., Tallapragada, N., Veres, A., Li, V., Jacome-Galarza, C. E., Hä ndler, K., Klughammer, J., Kobayashi, Y. et al.
Peshkin, L., Weitz, D. A. and Kirschner, M. W. (2015). Droplet barcoding for (2016). Specification of tissue-resident macrophages during organogenesis.
single-cell transcriptomics applied to embryonic stem cells. Cell 161, 1187-1201. Science 353, aaf4238.

30
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Matsumoto, H. and Kiryu, H. (2016). SCOUP: a probabilistic model based on the sequencing method, reveals non-genetic gene-expression heterogeneity.
Ornstein-Uhlenbeck process to analyze single-cell expression data during Genome Biol. 14, R31.
differentiation. BMC Bioinformatics 17, 232. Satija, R., Farrell, J. A., Gennert, D., Schier, A. F. and Regev, A. (2015). Spatial
McConnell, M. J., Lindberg, M. R., Brennand, K. J., Piper, J. C., Voet, T., reconstruction of single-cell gene expression data. Nat. Biotechnol. 33, 495-502.
Cowing-Zitron, C., Shumilina, S., Lasken, R. S., Vermeesch, J. R., Hall, I. M. Setty, M., Tadmor, M. D., Reich-Zeliger, S., Angel, O., Salame, T. M., Kathail, P.,
et al. (2013). Mosaic copy number variation in human neurons. Science 342, Choi, K., Bendall, S., Friedman, N. and Pe’er, D. (2016). Wishbone identifies
632-637. bifurcating developmental trajectories from single-cell data. Nat. Biotechnol. 34,
McKinney-Freeman, S., Cahan, P., Li, H., Lacadie, S. A., Huang, H.-T., Curran, 637-614.
M., Loewer, S., Naveiras, O., Kathrein, K. L., Konantz, M. et al. (2012). The Shalek, A. K., Satija, R., Shuga, J., Trombetta, J. J., Gennert, D., Lu, D., Chen, P.,
transcriptional landscape of hematopoietic stem cell ontogeny. Cell Stem Cell 11, Gertner, R. S., Gaublomme, J. T., Yosef, N. et al. (2014). Single-cell RNA-seq
701-714. reveals dynamic paracrine control of cellular variation. Nature 510, 363-369.
Moignard, V., Woodhouse, S., Haghverdi, L., Lilly, A. J., Tanaka, Y., Wilkinson, Shi, J., Chen, Q., Li, X., Zheng, X., Zhang, Y., Qiao, J., Tang, F., Tao, Y., Zhou, Q.
A. C., Buettner, F., Macaulay, I. C., Jawaid, W., Diamanti, E. et al. (2015). and Duan, E. (2015). Dynamic transcriptional symmetry-breaking in pre-
Decoding the regulatory network of early blood development from single-cell gene implantation mammalian embryo development revealed by single-cell RNA-seq.
expression measurements. Nat. Biotechnol. 33, 269-276. Development 142, 3468-3477.
Navin, N., Kendall, J., Troge, J., Andrews, P., Rodgers, L., McIndoo, J., Cook, Shin, J., Berg, D. A., Zhu, Y., Shin, J. Y., Song, J., Bonaguidi, M. A., Enikolopov,
K., Stepansky, A., Levy, D., Esposito, D. et al. (2011). Tumour evolution inferred G., Nauen, D. W., Christian, K. M., Ming, G.-L. et al. (2015). Single-cell RNA-Seq
by single-cell sequencing. Nature 472, 90-94. with waterfall reveals molecular cascades underlying adult neurogenesis. Cell
Nagano, T., Lubling, Y., Stevens, T. J., Schoenfelder, S., Yaffe, E., Dean, W., Stem Cell 17, 360-372.
Laue, E. D., Tanay, A. and Fraser, P. (2016). Single-cell Hi-C reveals cell-to-cell Smallwood, S. A., Lee, H. J., Angermueller, C., Krueger, F., Saadeh, H., Peat, J.,
variability in chromosome structure. Nature 502, 59-64. Andrews, S. R., Stegle, O., Reik, W. and Kelsey, G. (2014). Single-cell genome-
Nath, R. D., Chow, E. S., Wang, H., Schwarz, E. M. and Sternberg, P. W. (2016). wide bisulfite sequencing for assessing epigenetic heterogeneity. Nat. Methods
C. elegans stress-induced sleep emerges from the collective action of multiple 11, 817-820.
neuropeptides. Curr. Biol. 26, 2446-2455. Stegle, O., Teichmann, S. A. and Marioni, J. C. (2015). Computational and
Nelson, A. C., Mould, A. W., Bikoff, E. K. and Robertson, E. J. (2016). Single-cell analytical challenges in single-cell transcriptomics. Nat. Rev. Genet. 16, 133-145.
RNA-seq reveals cell type-specific transcriptional signatures at the maternal- Tang, F., Barbacioru, C., Wang, Y., Nordman, E., Lee, C., Xu, N., Wang, X.,
foetal interface during pregnancy. Nat. Commun. 7, 11414. Bodeau, J., Tuch, B. B., Siddiqui, A. et al. (2009). mRNA-Seq whole-
Nichols, J. and Smith, A. (2011). The origin and identity of embryonic stem cells. transcriptome analysis of a single cell. Nat. Methods 6, 377-382.
Development 138, 3-8. Tang, F., Barbacioru, C., Bao, S., Lee, C., Nordman, E., Wang, X., Lao, K. and
Niwa, H., Ogawa, K., Shimosato, D. and Adachi, K. (2009). A parallel circuit of LIF Surani, M. A. (2010a). Tracing the derivation of embryonic stem cells from the
signalling pathways maintains pluripotency of mouse ES cells. Nature 460, inner cell mass by single-cell RNA-Seq analysis. Cell Stem Cell 6, 468-478.
118-122. Tang, F., Barbacioru, C., Nordman, E., Li, B., Xu, N., Bashkirov, V. I., Lao, K. and
Nowakowski, T. J., Pollen, A. A., Di Lullo, E., Sandoval-Espinosa, C., Surani, M. A. (2010b). RNA-Seq analysis to capture the transcriptome landscape
Bershteyn, M. and Kriegstein, A. R. (2016). Expression analysis highlights of a single cell. Nat. Protoc. 5, 516-535.
AXL as a candidate Zika virus entry receptor in neural stem cells. Cell Stem Cell Tasic, B., Menon, V., Nguyen, T. N., Kim, T.-K., Jarsky, T., Yao, Z., Levi, B., Gray,
18, 591-596. L. T., Sorensen, S. A., Dolbeare, T. et al. (2016). Adult mouse cortical cell
Ocone, A., Haghverdi, L., Mueller, N. S. and Theis, F. J. (2015). Reconstructing taxonomy revealed by single cell transcriptomics. Nat. Neurosci. 19, 335-346.
gene regulatory dynamics from high-dimensional single-cell snapshot data. Thomsen, E. R., Mich, J. K., Yao, Z., Hodge, R. D., Doyle, A. M., Jang, S.,
Bioinformatics 31, i89-i96. Shehata, S. I., Nelson, A. M., Shapovalova, N. V., Levi, B. P. et al. (2015). Fixed
Olsson, A., Venkatasubramanian, M., Chaudhri, V. K., Aronow, B. J., single-cell transcriptomic characterization of human radial glial diversity. Nat.
Salomonis, N., Singh, H. and Grimes, H. L. (2016). Single-cell analysis of Methods 13, 87-97.
mixed-lineage states leading to a binary cell fate choice. Nature 537, 698-702. Tintori, S. C., Osborne Nishimura, E., Golden, P., Lieb, J. D. and Goldstein, B.
Pan, X., Durrett, R. E., Zhu, H., Tanaka, Y., Li, Y., Zi, X., Marjani, S. L., (2016). A transcriptional lineage of the early C. elegans embryo. Dev. Cell 38,
Euskirchen, G., Ma, C., LaMotte, R. H. et al. (2013). Two methods for full-length 430-444.
RNA sequencing for low quantities of cells and single cells. Proc. Natl. Acad. Sci. Toyooka, Y., Shimosato, D., Murakami, K., Takahashi, K. and Niwa, H. (2008).
USA 110, 594-599. Identification and characterization of subpopulations in undifferentiated ES cell
Patel, A. P., Tirosh, I., Trombetta, J. J., Shalek, A. K., Gillespie, S. M., Wakimoto, culture. Development 135, 909-918.
H., Cahill, D. P., Nahed, B. V., Curry, W. T., Martuza, R. L. et al. (2014). Single- Trapnell, C. (2015). Defining cell types and states with single-cell genomics.
cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Genome Res. 25, 1491-1498.
Science 344, 1396-1401. Trapnell, C., Cacchiarelli, D., Grimsby, J., Pokharel, P., Li, S., Morse, M.,
Petropoulos, S., Edsgä rd, D., Reinius, B., Deng, Q., Panula, S. P., Codeluppi, S., Lennon, N. J., Livak, K. J., Mikkelsen, T. S. and Rinn, J. L. (2014). The
Reyes, A. P., Linnarsson, S., Sandberg, R. and Lanner, F. (2016). Single-cell dynamics and regulators of cell fate decisions are revealed by pseudotemporal
RNA-seq reveals lineage and X chromosome dynamics in human preimplantation ordering of single cells. Nat. Biotechnol. 32, 381-386.
embryos. Cell 165, 1012-1026. Treutlein, B., Brownfield, D. G., Wu, A. R., Neff, N. F., Mantalas, G. L., Espinoza,
Picelli, S., Bjö rklund, Å. K., Faridani, O. R., Sagasser, S., Winberg, G. and F. H., Desai, T. J., Krasnow, M. A. and Quake, S. R. (2014). Reconstructing
Sandberg, R. (2013). Smart-seq2 for sensitive full-length transcriptome profiling lineage hierarchies of the distal lung epithelium using single-cell RNA-seq. Nature
in single cells. Nat. Methods 10, 1096-1098. 509, 371-375.
Pierson, E. and Yau, C. (2015). ZIFA: dimensionality reduction for zero-inflated Vallejos, C. A., Marioni, J. C. and Richardson, S. (2015). BASiCS: Bayesian
single-cell gene expression analysis. Genome Biol. 16, 241. Analysis of Single-Cell Sequencing Data. PLoS Comput. Biol. 11, e1004333-e18.
Pollen, A. A., Nowakowski, T. J., Shuga, J., Wang, X., Leyrat, A. A., Lui, J. H., Li, Wang, Y. and Navin, N. E. (2015). Advances and applications of single-cell
N., Szpankowski, L., Fowler, B., Chen, P. et al. (2014). Low-coverage single-cell sequencing technologies. Mol. Cell 58, 598-609.
mrNA sequencing reveals cellular heterogeneity and activated signaling pathways Welch, J. D., Hartemink, A. J. and Prins, J. F. (2016). SLICER: inferring branched,
in developing cerebral cortex. Nat. Biotechnol. 32, 1053-1058. nonlinear cellular trajectories from single cell RNA-seq data. Genome Biol. 17,
Pollen, A. A., Nowakowski, T. J., Chen, J., Retallack, H., Sandoval-Espinosa, C., 106.
Nicholas, C. R., Shuga, J., Liu, S. J., Oldham, M. C., Diaz, A. et al. (2015). Wu, A. R., Neff, N. F., Kalisky, T., Dalerba, P., Treutlein, B., Rothenberg, M. E.,
Molecular identity of human outer radial glia during cortical development. Cell 163, Mburu, F. M., Mantalas, G. L., Sim, S., Clarke, M. F. et al. (2013). Quantitative
55-67. assessment of single-cell RNA-sequencing methods. Nat. Methods 11, 41-46.
Ramskold, D., Luo, S., Wang, Y.-C., Li, R., Deng, Q., Faridani, O. R., Daniels, Xin, Y., Kim, J., Ni, M., Wei, Y., Okamoto, H., Lee, J., Adler, C., Cavino, K.,
DEVELOPMENT

G. A., Khrebtukova, I., Loring, J. F., Laurent, L. C. et al. (2012). Full-length Murphy, A. J., Yancopoulos, G. D. et al. (2016). Use of the Fluidigm C1 platform
mRNA-Seq from single-cell levels of RNA and individual circulating tumor cells. for RNA sequencing of single mouse pancreatic islet cells. Proc. Natl. Acad. Sci.
Nat. Biotechnol. 30, 777-782. USA 113, 3293-3298.
Rossant, J., Rossant, J., Tam, P. P. L. and Tam, P. P. L. (2009). Blastocyst lineage Xu, C. and Su, Z. (2015). Identification of cell types from single-cell transcriptomes
formation, early embryonic asymmetries and axis patterning in the mouse. using a novel clustering method. Bioinformatics 31, 1974-1980.
Development 136, 701-713. Xue, Z., Huang, K., Cai, C., Cai, L., Jiang, C.-Y., Feng, Y., Liu, Z., Zeng, Q.,
Rotem, A., Ram, O., Shoresh, N., Sperling, R. A., Goren, A., Weitz, D. A. and Cheng, L., Sun, Y. E. et al. (2013). Genetic programs in human and mouse early
Bernstein, B. E. (2015). Single-cell ChIP-seq reveals cell subpopulations defined embryos revealed by single-cell RNAsequencing. Nature 500, 593-597.
by chromatin state. Nat. Biotechnol. 33, 1165-1172. Yan, L., Yang, M., Guo, H., Yang, L., Wu, J., Li, R., Liu, P., Lian, Y., Zheng, X.,
Sasagawa, Y., Nikaido, I., Hayashi, T., Danno, H., Uno, K. D., Imai, T. and Ueda, Yan, J. et al. (2013). Single-cell RNA-Seq profiling of human preimplantation
H. R. (2013). Quartz-Seq: a highly reproducible and sensitive single-cell RNA embryos and embryonic stem cells. Nat. Struct. Mol. Biol. 20, 1131-1139.

31
REVIEW Development (2017) 144, 17-32 doi:10.1242/dev.133058

Yu, Y., Tsang, J. C. H., Wang, C., Clare, S., Wang, J., Chen, X., Brandt, C., Kane, Cell types in the mouse cortex and hippocampus revealed by single-cell RNA-seq.
L., Campos, L. S., Lu, L. et al. (2016). Single-cell RNA-seq identifies a PD-1(hi) Science 347, 1138-1142.
ILC progenitor and defines its development pathway. Nature 539, 102-106. Zhou, F., Li, X., Wang, W., Zhu, P., Zhou, J., He, W., Ding, M., Xiong, F., Zheng, X.,
Zeisel, A., Muñoz-Manchado, A. B., Codeluppi, S., Lö nnerberg, P., La Manno, Li, Z. et al. (2016). Tracing haematopoietic stem cell formation at single-cell
G., Juré us, A., Marques, S., Munguba, H., He, L., Betsholtz, C. et al. (2015). resolution. Nature 533, 487-492.

DEVELOPMENT

32