Content uploaded by Hailiang Liu
Author content
All content in this area was uploaded by Hailiang Liu on Aug 04, 2021
Content may be subject to copyright.
Article
Single-Cell Transcriptome Analyses Reveal Signals
to Activate Dormant Neural Stem Cells
Graphical Abstract
Highlights
dSingle-cell RNA-seq reveals a signature gene program in
ependymal CD133
+
cells
dVEGF is a mitogen for ependymal CD133
+
cells and bFGF for
subependymal B cells
dCD133
+
cells lining the fourth ventricle show NSC activity
upon VEGF/bFGF treatment
Authors
Yuping Luo, Volkan Coskun, ...,
Yi Eve Sun, Siguang Li
Correspondence
yi.eve.sun@gmail.com (Y.E.S.),
siguangli@163.com (S.L.)
In Brief
Using single-cell transcriptome and
network analyses, Luo et al. identify a
subset of quiescent neural stem cells in
non-neurogenic brain regions and show
that these cells could be mitotically
activated and differentiated into neurons
and glia upon stimulation.
Accession Numbers
GSE61288
Luo et al., 2015, Cell 161, 1175–1186
May 21, 2015 ª2015 Elsevier Inc.
http://dx.doi.org/10.1016/j.cell.2015.04.001
Article
Single-Cell Transcriptome Analyses Reveal Signals
to Activate Dormant Neural Stem Cells
Yuping Luo,
1,2,10
Volkan Coskun,
4,10
Aibing Liang,
1,10
Juehua Yu,
1,10
Liming Cheng,
1,8,10
Weihong Ge,
4
Zhanping Shi,
1
Kunshan Zhang,
1
Chun Li,
6
Yaru Cui,
2
Haijun Lin,
2
Dandan Luo,
1
Junbang Wang,
1
Connie Lin,
4
Zachary Dai,
4
Hongwen Zhu,
7
Jun Zhang,
1
Jie Liu,
1
Hailiang Liu,
1
Jean deVellis,
4
Steve Horvath,
9
Yi Eve Sun,
1,3,4,5,
*and Siguang Li
1,3,
*
1
Stem Cell Translational Research Center, Tongji Hospital, Tongji University School of Medicine, Shanghai 200065, China
2
College of Life Sciences, Nanchang University, Nanchang 330031, China
3
Collaborative Innovation Center for Brain Science, Tongji University, Shanghai 200092, China
4
Department of Psychiatry and Biobehavioral Sciences, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles,
CA 90095, USA
5
Yunnan Key Laboratory of Primate Biomedical Research, Kunming 650500, China
6
Shanghai Stem Cell Institute, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
7
Tianjing Hospital, Tianjin Academy of Integrative Medicine, Tianjin 300211, China
8
Department of Spine Surgery, Tongji Hospital, Tongji University School of Medicine, Shanghai 200065, China
9
Department of Human Genetics, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA 90095, USA
10
Co-first author
*Correspondence: yi.eve.sun@gmail.com (Y.E.S.), siguangli@163.com (S.L.)
http://dx.doi.org/10.1016/j.cell.2015.04.001
SUMMARY
The scarcity of tissue-specific stem cells and the
complexity of their surrounding environment have
made molecular characterization of these cells partic-
ularly challenging. Through single-cell transcriptome
and weighted gene co-expression network anal-
ysis (WGCNA), we uncovered molecular properties
of CD133
+
/GFAP
ependymal (E) cells in the adult
mouse forebrain neurogenic zone. Surprisingly, prom-
inent hub genes of the gene network unique to epen-
dymal CD133
+
/GFAP
quiescent cells were enriched
for immune-responsive genes, as well as genes en-
coding receptors for angiogenic factors. Administra-
tion of vascular endothelial growth factor (VEGF) acti-
vated CD133
+
ependymal neural stem cells (NSCs),
lining not only the lateral but also the fourth ventricles
and, together with basic fibroblast growth factor
(bFGF), elicited subsequent neural lineage differentia-
tion and migration. This study revealed the existence
of dormant ependymal NSCs throughout the ventricu-
lar surface of the CNS, as well as signals abundant
after injury for their activation.
INTRODUCTION
Tissue-specific stem cells reside in highly complex cellular envi-
ronments (Doetsch, 2003; Beckervordersandforth et al., 2010;
Coskun et al., 2008; Ming and Song, 2011; Morrison and Spra-
dling, 2008; Merkle et al., 2007) and are in close contact with
stem cell niches and progenies to maintain homeostasis,
balancing between quiescent and activated states (Lugert
et al., 2010; Li and Clevers, 2010). While a cell can be, for the
most part, defined by the pattern of genes it expresses, a major
challenge for genome-wide transcriptome analyses of tissues
with heterogeneous cellular composition is that the readout is
the sum or average of all of the different cells in that particular tis-
sue. Unfortunately, the transcriptome of such an ‘‘averaged cell’’
does not truly reflect any particular cells in the tissue, and when it
comes to tissue-specific quiescent stem cells or any rare but
important cell types in the tissue, population-based transcrip-
tome analyses become nearly impractical and may provide unin-
tentional misleading results (Shapiro et al., 2013; Shalek et al.,
2013; Nolan et al., 2013; Meacham and Morrison, 2013; Wu
and Tzanakakis, 2013; Snippert and Clevers, 2011). To circum-
vent such a problem, single-cell-based transcriptome analyses
become imperative.
The field of single-cell transcriptome analyses has developed
quickly in recent years (Shalek et al., 2013; Xue et al., 2013; Yan
et al., 2013; Tang et al., 2009, 2010). Major challenges for single-
cell transcriptome analyses lie in technicality. One of the bottle-
necks is to maintain the authenticity of gene expression levels
during cDNA conversion and amplification. Single-cell RNA
sequencing (RNA-seq) analyses are different from single-cell
DNA sequencing analyses. The focus of the latter is the exact
nucleotide sequence, while transcriptome deals with gene
expression levels (mRNA levels); therefore, efficient reverse tran-
scription (cDNA conversion) and linear amplification to keep the
relative abundance of different transcripts constant are very
important for transcriptome analysis but not for genomic
sequencing. Another bottleneck is related to bioinformatics ana-
lyses (i.e., big-data processing). Since no detection system is
perfect, one must know the nuts and bolts of the single-cell
transcriptome analyses, including the detection limit and sys-
tem-generated variations, which is different from true biological
variations, in order to apply the technology well into solving
the aforementioned heterogeneity-related difficult biological
problems.
The ependymal/subependymal regions of the adult mouse
forebrain have been reported to harbor neural stem cells
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. 1175
(NSCs), which give rise to olfactory bulb interneurons throughout
life. This region contains the previously described four cell types
related to adult NSC activities: (1) ependymal E cells; (2) sube-
pendymal GFAP
+
B cells, some of which are also referred to as
mono-ciliated, CD133 (encoded by the prominin1 gene), and
GFAP double-positive NSCs (Beckervordersandforth et al.,
2010); (3) transit-amplifying C cells; and (4) neuroblast A cells
(Figure S1). It has been postulated and widely accepted that
GFAP
+
B cells contain NSC activity. During NSC activation, B
cells produce transit-amplifying C cells, and C cells give rise to
large numbers of PSA-NCAM (polysialylated neural cell adhesion
molecule)-positive neuroblast A cells, which repopulate the ol-
factory bulb (Doetsch, 2003). A contentious issue in the field
lies in the understanding of the ependyma. While several studies
demonstrated that ependymal multi-ciliated cells (E cells)
contain stem cell activities (Johansson et al., 1999; Coskun
et al., 2008; Nakafuku et al., 2008), others suggested that E cells
were structural cells, which did not divide and, therefore, could
not serve as NSCs. Previously, we have demonstrated that
during embryonic cortical development, immunoreactivity for
CD133 labeled almost all cells in the germinal ventricular zone
(VZ) lining the ventricular surface, which were considered
NSCs (Coskun et al., 2008). Postnatally, CD133 still labeled the
layer of cells lining the ventricular surface, which is referred to
as the ependyma. It is likely that embryonic VZ cells turn into
ependymal cells postnatally. However, whether they still main-
tain NSC activity has been debated. Using prominin1 gene-
based lineage-tracing studies, we have shown that CD133
+
cells
give rise to neuroblast A cells in the rostral migratory stream
(RMS) and interneurons in the olfactory bulb, suggesting that
CD133
+
ependymal cells still maintain NSC activity (Coskun
et al., 2008). An alternative interpretation of the study, however,
is that while the prominin1 promoter labeled all CD133-express-
ing cells, not all, but only the mono-ciliated CD133 and GFAP
double-positive cells manifested NSC activities (Codega et al.,
2014; Beckervordersandforth et al., 2010). There is no evidence,
however, to prove or disprove such an interpretation. Clearly, the
CD133 population is heterogeneous in the ependymal and/or
subependymal regions. The lack of comprehensive knowledge
of the molecular makeups/signatures of the ependymal and sub-
ependymal cells hampered the understanding of lineage rela-
tionships among the complex cell populations in those areas.
In this study, we carried out single-cell transcriptome analyses
on adult mouse CD133
+
ependymal and some CD133
sube-
pendymal cells. Using weighted gene coexpression network
analysis (WGCNA) (Zhang and Horvath, 2005; Langfelder and
Horvath, 2008), we uncovered specific gene-regulatory modules
correlated with what appeared to be critically staged cells along
stem cell activation and subsequent neural lineage develop-
ment, including CD133
+
E cells, CD133 and GFAP double-posi-
tive B cells, and DLX2
+
and doublecortin
+
(DCX
+
) neuroblast A
cells (Coskun et al., 2008; Doetsch et al., 1999; Bonaguidi
et al., 2011; Zhao et al., 2008). Moreover, we revealed and vali-
dated a tight link between CD133
+
adult NSC activation and
vasculature development/angiogenesis-related signals.
Single-cell transcriptome analysis indicated that CD133
+
/
GFAP
cells in the ependyma expressed vascular endothelial
growth factor (VEGF) receptors (VEGFRs) but not basic fibro-
blast growth factor (bFGF) receptors (bFGFRs); in contrast,
CD133
+
/GFAP
+
subependymal B cells appeared to express
bFGFRs but not VEGFRs. When bFGF was introduced into the
lateral ventricles, only subependymal cells were mitotically
further activated. On the contrary, when VEGF was added,
mitotic activation of the ependymal cells took place. The epen-
dymal/ventricular surface of the fourth ventricle has not been re-
ported to be neurogenic in vivo in postnatal mice (Martens et al.,
2002). When exposed to VEGF and bFGF, ependymal quiescent
NSCs lining the fourth ventricle were also mitotically activated.
Using ROSA26-tdTomato reporter mice with electroporation of
prominin1 promoter-driven Cre, we found that CD133
+
/GFAP
ependymal cells of the fourth ventricle not only became mitoti-
cally activated but also created a lineage by differentiating into
MAP2
+
neurons and GFAP
+
glia, indicative of endogenous
NSC activity.
We predict that the single-cell transcriptome analysis estab-
lished here would be a powerful tool for identification of molec-
ular signatures of tissue stem cells and cancer stem cells, as
well as revealing molecular features of complex biological sys-
tems, including various neural circuits in the brain.
RESULTS
Quality Control of Single-Cell RNA-Seq
To establish single mouse CD133
+
and CD133
adult NSCs and/
or progenitor cell transcriptome profiles, we used fluorescence-
activated cell sorting (FACS) to isolate CD133
+
and CD133
cells, followed by manual picking of single cells under the micro-
scope (Figure S1). Single cells were subjected to RNA-seq
sample preparation using previously published protocols (Kuri-
moto et al., 2007; Tang et al., 2009, 2010; Ramsko
¨ld et al.,
2012; Xue et al., 2013), with modifications at several steps (for
details, see Supplemental Information). We used single-cell
qRT-PCR to detect a few well-known markers, including
CD133 (prominin1), GFAP, MBP, CNP, DCX, and DLX2, and
based on the presence or absence of their expression to repre-
sent a few different types of cells in the ependymal/subependy-
mal regions (Figure S1). From a total of more than 200 single cells
analyzed by qRT-PCR, we picked 28 cells representing CD133
+
and different CD133
populations for subsequent RNA deep
sequencing using both SOLiD-3 and Illumina HiSeq 2500
sequencing platforms. On average, about 15–20 million reads
mapped to the reference genome (Ensembl genome browser,
Mus_musculus, GRCm38.74) (Table S1) were obtained per sam-
ple, and FPKM (fragments per kilobase of transcript per million
mapped reads) conversion was performed to represent levels
of expression. Acquisition of good-quality single-cell RNA-seq
data is presumably important to the downstream extrapolation
of useful biological information. Therefore, we performed the
following assays to assess the validity/quality of our single-cell
RNA-seq data: (1) sequencing reproducibility controls and (2)
control for batch effect. For sequencing reproducibility controls,
we evaluated the reproducibility of our RNA-seq platform
by independently sequencing quadruplicate cDNA samples
aliquoted from one cDNA library and found Pearson correlation
coefficients of the sequencing results to be 0.996 or higher
based on log-transformed FPKM, indicative of good sequencing
1176 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
reproducibility and that even sampling of the cDNA library for
sequencing could be achieved with our protocol (Figure 1A).
For control for batch effect, to assess sequencing batch effect
of our system, we sequenced a single-cell cDNA library twice,
6 months apart, and found the Pearson correlation coefficient
to be 0.998 of the two batches, demonstrating minimal
sequencing batch effect (Figure 1B). To assess potential tech-
nical and biological variations of single-cell RNA-seq results,
we compared two-cell-stage mouse developing embryos, where
the two sister cells are presumably very ‘‘similar.’’ RNA-seq re-
sults indicated that Pearson correlation coefficients of similar
cells were within the 0.901–0.950 range (Figure 1C). Assuming
these samples are the ‘‘same,’’ the maximal technological
variations of our single-cell RNA-seq system should be less
than or within the range of 0.05 (i.e., 1–0.950 = 0.05) to 0.1 (i.e.,
1–0.90 = 0.1) at most. Since early developing mouse embryos,
such as two-cell-stage embryos, carry relatively large amounts
of RNA (about 0.47 ng per cell) (Olsza
nska and Borgul, 1993),
Figure 1. Quality Controls of Single-Cell
RNA-Seq Analysis
(A) Quadruplicate independent sequencing of the
same cDNA sample demonstrating good repro-
ducibility as shown by the Pearson correlation
coefficient being 0.996–0.997 among all four
sequencing results. P, pooled sample.
(B) The same single-cell cDNA sample (S10) was
subjected to two independent sequencings in two
batches, with the exact sequencing time being
6 months apart. High Pearson correlation coeffi-
cient (0.998) indicates minimum sequencing batch
effect and, therefore, reliable sequencing tech-
nology. TR, technical repetition.
(C) Estimation of technological and biological var-
iations of single-cell sequencing analysis. Two-
cell-stage mouse embryos were sequenced either
as a whole or separately as individual cells. Pear-
son correlation coefficients among the four data-
sets are between 0.901 and 0.95. Assuming these
cells are identical, our technical variation is ranging
between 0.05 and 0.1. E, embryo.
(D) Pearson correlation coefficient of the three
pairs of half-cell RNA-seq.
(E) Saturation curves of RNA-seq data with
FPKM R0.1.
(F) Number of genes detected in pooled samples
and single-cell samples (FPKM R0.1).
See also Figures S1,S2,S3, and S4 and Table S1.
including stored maternal RNA, it is still
unknown whether such an assessment
of technical variations may apply to adult
somatic NSCs, which contain less
amount of RNA (about 10–20 pg per cell)
(Tietjen et al., 2003; Beckervordersand-
forth et al., 2010). Unfortunately, there
are no two somatic cells in the adult sys-
tem known to be truly identical; therefore,
we decided to sequence the two halves of
one NSC using our single-cell RNA-seq
system instead. Presumably, materials
from the two halves of one NSC should give the same
sequencing result. The challenge was that the total amount of
RNA materials in this situation were even less than that in one
single NSC. Moreover, we found that when we split the 5-ml sin-
gle-cell lysate into two halves (2.5 ml), the cell-membranous
debris prevented a homogeneous even split of the material. In
the end, Pearson correlation coefficients of the three pairs of
half-cell RNA-seq were 0.84, 0.82, and 0.80 (Figure 1D). More-
over, we also used the External RNA Control Consortium
(ERCC) spike-in method and found that when about 5 pg (similar
to the total estimated mRNAs in half of a somatic cell) of ERCC
standard poly(A) RNAs were spiked into the half-cell cDNA
RNA-seq system, the Pearson correlation coefficient between
ERCC concentrations and detected ERCC read counts was
0.93 (Figure S2A). All of these observations demonstrated the
validity and reasonable technical variability of our single-cell
RNA-seq system. To examine whether the sequencing depths
of our data were sufficient, we carried out saturation analyses
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. 1177
(legend on next page)
1178 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
and found that with a sequencing depth of 15–20 million total
reads, when we chose an FPKM cutoff at 0.1, all samples
reached saturation in mRNA detection with about 8 million reads
used (Figure 1E). The vast majority of single-cell samples satu-
rated with only 5 million reads used (Figure 1E). If we chose an
FPKM cutoff at 1, the number of reads required for saturation
was even less, about 2–3 million (Figure S2B). Additional quality
control results are shown in the Supplemental Information (Fig-
ures S2C–S2F).
To assess the addition and/or averaging effect when seq-
uencing heterogeneous population samples, from approxi-
mately 200 single-cell cDNA libraries confirmed for CD133
positivity via RT-PCR, we collected and mixed ten CD133
+
and
ten CD133
single-cell libraries, respectively, and sequenced
the mixed/pooled samples (Figure 1F, samples P1, P2, P3, and
P4). The pooling process increased the cDNA complexity of
the samples for subsequent sequencing. Since random sam-
pling and sequencing of one such cDNA mix four times gave
highly correlated sequencing results (Pearson correlation coeffi-
cient being 0.996 among the quadruplicate samples; Figure 1A),
we were confident that sequencing of the pooled sample was
saturated. The number of genes detected in single-cell seq-
uencing samples as well as after the pooling of ten single-cell li-
braries as CD133
+
population and CD133
population samples
using an FPKM cutoff at 0.1 (Figure 1F) demonstrated that
pooled samples contained more expressed genes than any of
the single-cell samples. Obviously, population-based analyses
did not represent any of the single cells and would likely give
false information on gene co-expression. For example, genes
co-detected in the same population-based samples— therefore,
‘‘seemingly co-expressed’’—could, in fact, be expressed in
different cells in that population. On the other hand, if we pooled
sequencing data of all four pooled samples (P1–P4), with an
FPKM cutoff at 0.1, we detected 12,701 genes expressed in 40
CD133
+
and CD133
cells. However, if we pooled sequencing
data of all 28 single cells together (S1–S28), we detected a total
of 13,611 genes expressed in 28 CD133
+
and CD133
cells, sug-
gesting that single-cell detection was more sensitive, as more
genes could be detected. To carry out a more precise com-
parison, we designed an experimental paradigm so that the
exact ten CD133
+
and seven CD133
single cells were both
sequenced as individual cells and as pooled population samples
after mixing individual cDNAs as described earlier. With detec-
tion saturation at FPKM > 0.1, we found that there were more
than 1,000 genes, which were detected when the single-cell
transcriptome was analyzed individually followed by pooling of
the sequencing data but were undetectable in pooled-then-
sequenced samples (Figure S3). This is because pooling the
cDNAs before sequencing would end up mixing/averaging all
cDNA species so that genes that are highly expressed in rare
cells (e.g., one out of ten cells) will be diluted or averaged out
and, therefore, could fall below the established detection
threshold. It is worth noting that, here, we only demonstrated
how mixing and averaging cDNA samples would lead to reduced
sequencing detection sensitivity. The inevitable stochastic loss
of transcripts associated with single-cell RNA-to-cDNA conver-
sion and PCR amplification was not addressed here, which likely
contributed to the fact that our half-cell RNA-seq samples only
correlated with each other by about 80%–84%, not 99.6%.
Nevertheless, with regard to a heterogeneous population, sin-
gle-cell-resolution, RNA-seq data provide more detailed and
abundant information of the molecular properties and cellular
compositions of the population compared to population-based
analyses. Additionally, with this approach, we detected more
than 600 novel multi-exonic long noncoding RNAs (lncRNAs)
(Figure S4) that failed to be detected in previously reported
lncRNA expression profiles using microdissected adult subven-
tricular zone (SVZ) tissues (Ramos et al., 2013). This revealed
the potential discovery power of single-cell RNA-seq. Taken
together, these studies demonstrated that single-cell transcrip-
tome profiling provides new information that could be difficult
to acquire using population-based analyses.
Gene-Network Modules Underlying NSC Activation
Identified by WGCNAs
Unsupervised hierarchical clustering of 28 single-cell samples
and 4 pooled-cell samples (FPKM R0.1) demonstrated that
data from all eight CD133
+
/GFAP
single and two pooled
CD133
+
/GFAP
cell samples (ten cells per sample) clustered
together naturally (Figure 2A, blue frame). In addition, CD133
and DLX2
+
neuronal lineage samples clustered together (Fig-
ure 2A, red frame). The rest of the samples formed a loose cluster
(Figure 2A, yellow frame), within which there were two more
closely correlated sub-clusters (Figure 2A, brown and royal
blue frames). S9 and S10 in the royal blue subgroup expressed
both CD133 and GFAP.
When single-cell RNA-seq data (FPKM R0.1) were subjected
to WGCNAs, multiple gene-network modules were obtained. We
decided to focus initially on modules that were shared by more
than one sample so that shared properties of cell subpopulations
Figure 2. WGCNA Revealed Gene-Network Modules Enriched in CD133
+
E Cells, GFAP
+
/CD133
+
B Cells, DLX2
+
A Cells, and SOX10
+
/OLIG1
+
O Cells
(A) Unsupervised hierarchical clustering of 28 single-cell samples and four pooled-cell datasets with FPKM cutoff of 0.1.
(B) WGCNA dendrogram indicating expression of different gene modules in all 28 sing le-cell samples.
(C) Eigengene expression of four selected modules across all 28 single-cell samples. Color code of the modules is preserved. Expressions of a few known
markers for E, B, O, and A cells were also shown in 28 samples. Based on these two sets of information, it is obvious that the blue module has enriched expression
in 8 CD133 single-positive cells, the brown module is expressed in 3 O cells, the red module is highly expressed in 11 DLX2
+
A cells, and the royal blue module is
expressed in 2 GFAP
+
and CD133
+
B cells.
(D) Expanded view of expression of all genes in each of the four modules across all 28 single-cell samples.
(E) Hub-gene network of the red module (the A module). Size of the dots represents hubness. Red highlights the genes known for neurogenesis.
(F) Hub-gene network of the brown module (the O module). Size of the dots represents hubness. Brown highlights the genes known for oligodendrocyte
differentiation.
See also Figure S5.
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. 1179
could be revealed. Four modules of interest were identified: blue,
red, brown, and royal blue (Figures 2B–2D). Module Eigengene
(i.e., the first principal component of a given module, which
can be considered a representative of the gene expression pro-
files in a module) expression across all single cells is presented in
Figure 2C. Based on limited marker gene expression—PROM1
(CD133) as an E-cell marker, GFAP as a B-cell marker, and
Figure 3. Mitotic Activation of Ependymal
CD133
+
Cells by VEGF
(A) Hub-gene network of the blue module (the
CD133-specific E module). The size of the dots
represents hubness. Blue highlights the genes
being discussed in the text.
(B) A representative sagittal section from the adult
SVZ region stained with CD133 (red) and Flt1
(green).
(C–F) Fluorescent photomicrographs of represen-
tative sagittal sections from the adult SVZ region
stained with CD133 (red) and the proliferation
marker Ki67 (green). The animals were injected
with saline (C), bFGF (D), VEGF (E), or bFGF +
VEGF (F). Arrows and the insets in (E) and (F)
indicate CD133
+
/Ki67
+
ependymal cells from the
same panels.
(G) A representative sagittal section from the SVZ
region of adult mouse that was injected with
bFGF + VEGF followed by 7-day oral BrdU
administration. BrdU
+
nuclei (red) are present
within the SVZ. The arrows and the lower inset
show BrdU
+
nuclei within the ependymal layer
from the same panel. The upper inset is taken from
a different section and shows additional BrdU
+
cells in the ependymal layer adjacent to the lateral
ventricle.
(H) The percentage of Ki67
+
cells in the sub-
ependymal region under saline, bFGF+VEGF,
bFGFVEGF+, and bFGF+VEGF+ conditions.
(I) The percentage of Ki67
+
/CD133
+
ependymal
cells under saline, bFGF+VEGF, bFGFVEGF+,
and bFGF+VEGF+ conditions.
(J) Quantification of BrdU
+
cells in the ependymal
layer under saline, bFGF+VEGF, bFGF-VEGF+,
and bFGF+VEGF+ conditions.
Scale bars, 100 mm in (B)–(G). Error bars indicate
the SD from three different experiments. *p < 0.05;
**p < 0.01. CC, corpus callosum; cp, choroid
plexus; LV, lateral ventricle.
DLX2 and DCX as A-cell markers—we as-
signed CD133
+
/GFAP
cells as E cells,
and the blue module turned out to be an
E-cell-specific module. Similarly, S9 and
S10 appeared to belong to B cells due
to GFAP and prominin1 double positivity,
and the royal blue module seemed to be
B cell specific, although it is based on
only two cells captured and sequenced
in these analyses. While it is possible
that these two B cells only represent a
subset of the B-cell population, this is still
the first exploration of the molecular sig-
natures of some B cells. Hub-gene-network analysis of the mod-
ules revealed a hierarchical organization of highly connected
genes in each module, through which key controlling (hub) genes
in the modular network can be identified (Figures 2E, 2F, and 3A).
The red module contains many neuronal differentiation hub
genes, including, for example, Dlx1,Dlx2, and Sox5, which
resemble A cells (Figure 2E). On the other hand, the brown
1180 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
module hub genes contain Sox10,Olig1,Mog, and Mag, which
are all oligodendrocyte differentiation genes; therefore, this
module is likely oligodendrocyte specific and, hence, referred
to as the ‘‘O-cell’’ module (Figure 2F). Although only 8 E cells,
11 A cells, 3 O cells, and 2 B cells were sequenced, being able
to easily assign different modules to different cells (i.e., E, B, A,
and O cells) (Figures 2D–2F and 3A) and to extrapolate the
hub-gene networks indicated that WGCNA was powerful in
providing useful information related to cell-type specificity.
Although, based on previous knowledge, B and A cells are likely
derived from E cells, we could not yet establish a link between E
cells and O cells. While more sampling of cells will be needed—
specifically with regard to B and O cell populations to assess
whether the identified modules represent the entire B and O
cell populations or only a subset of those cells—the discovery
of these modules would certainly help with better understanding
of the neurogenic zone cells as well as the molecular features of
different cell types along the process of stem cell activation and
differentiation in future studies.
E-Cell Module Is Enriched in Other Stem Cell Markers
and in Immune-Responsive and Angiogenesis-Related
Genes
When the CD133 ependymal-cell-specific blue module was
examined in combination with Gene Ontology (GO) analysis, ma-
jor GO terms associated with the blue (E) module were related to
vasculature development/angiogenesis and injury immune re-
sponses. Some of the big ‘‘hubs’’ in this module were FLT1,
KDR, and TEK, which are VEGF receptors and an angiopoietin
receptor, respectively (Figure 3A). Immunocytochemical ana-
lyses indicated that CD133
+
ependymal cells did express the
VEGF receptor FLT1, suggesting that this module indeed
marked ependymal cells but not other lineage cells such as
endothelial cells (Figure 3B). To our surprise, a hematopoietic
stem cell (HSC) marker CD34 and a mesenchymal stem cell
(MSC) marker ENG (CD105) were also present in this module.
Since CD133 labels the VZ in the developing cortex (Coskun
et al., 2008; Fischer et al., 2011; Uchida et al., 2000), it will be
interesting to know whether these angiogenesis-, hematopoie-
sis-, and mesenchyme-related features are present early on or
are acquired later during postnatal development. CD133
+
epen-
dymal cells also express extracellular matrix proteins such as
fibronectin (FN1) and collagen (COL4A2), which are features of
the mesenchymal lineage. It remains to be determined whether
epithelial-to-mesenchymal transition (EMT) is involved in the
development of adult tissue stem cells.
We tried to address whether the E (blue) module, which was
extrapolated from transcriptomes of eight CD133
+
GFAP
cells,
actually represented features shared by ependymal NSCs. We
analyzed five additional single-cell and two 10-cell pools of
CD133
+
GFAP
ependymal cells via RNA-seq. We then plotted
the expression levels of 30 hub genes of the E (blue) module in
eight initially sequenced single cells, as well as the ‘‘5 + 20’’
additionally sequenced samples, and found that the expression
data clustered considerably well, suggesting that these 33 cells
share common properties regarding E-module hub-gene ex-
pression (Figure S5). Based on these data, it appears that,
from the hub-gene expression perspective, a large number of
CD133
+
GFAP
ependymal cells are similar. Therefore, without
knowing whether all or only a fraction of ependymal CD133
+
cells
are quiescent NSCs, we speculated that the E module probably
represented common features of most, if not all, CD133
+
epen-
dymal cells, including CD133
+
ependymal NSCs.
Based on the transcriptome data indicating the expression of
VEGF receptors (FLT1 and KDR) as hub genes in the CD133
+
ependymal cell-specific gene regulatory network, we confirmed
that FLT1 and CD133 were indeed co-expressed in the ependyma
(Figure 3B), which is a very quiescent zone, because under normal
conditions, Ki67
+
cells in the ependymal layer are extremely rare
(Coskun et al., 2008; Ihrie and Alvarez-Buylla, 2011). Subse-
quently, we explored the function of VEGF signaling in regulating
the proliferation and differentiationof adult ependymal NSCs (Fig-
ures 3D–3J). In order to determine whether ependymal cells were
responsive to mitogen stimulation, we injected bFGF, VEGF, or
bFGF + VEGF into the lateral ventricles of adult mice and exam-
ined the proliferation of CD133
+
cells. Compared to saline injec-
tions, bFGF treatment mainly increased Ki67 labeling in the sube-
pendymal region (i.e., 10.81% ±0.31% versus 15.94% ±0.40%),
with minimal effects on the ependyma (i.e., 0.16% ±0.01% versus
0.22% ±0.04%) (Figures 3C, 3D, 3H, and 3I). In contrast, when
VEGF alone was injected, the ependyma (i.e., 0.16% ±0.01%
versus 0.74% ±0.07%) was mitotically activated, but not so
much the subependyma (i.e., 10.81% ±0.31% versus 11.88% ±
1.04%) (Figures 3C, 3E, 3H, and 3I). When bFGF and VEGF
were applied together, we detected an increased number of
Ki67
+
ependymal cells (i.e., 0.16% ±0.01% versus 0.94% ±
0.11%) and subependymal cells (Figures 3F, 3H, and 3I). Injected
animals were also administered with bromodeoxyuridine (BrdU)
orally for 7 days following surgery. In agreement with the Ki67
data, animals that were injected with bFGF and VEGF together
showed numerous BrdU
+
ependymal cells (Figures 3G and 3J).
These results indicated that while subependymal NSCs (B cells)
were mitotically responsive to bFGF, ependymal CD133
+
quies-
cent cells were more responsive to VEGF. Since the vascular sys-
tem has been postulated to be oneof the major components of the
NSC niche (Ihrie and Alvarez-Buylla, 2011; Mirzadeh et al., 2008)
and also vascular reconstructionis a necessary step via VEGF/an-
giopoietin signaling during injury, the VEGF responsiveness of
CD133
+
cells might serve to coordinate the activationof quiescent
NSCs and angiogenesis processes to facilitate neural repair.
Since quiescent CD133
+
ependymal cells can give rise to acti-
vated NSCs (B cells) (Coskun et al., 2008), CD133
+
/GFAP
+
cells
might represent E-to-B transitioning cells. B cells express VEGF
according to single-cell RNA-seq results, which could potentially
serve as an endogenous mitogen to activate ependymal cells
even though, after nerve injury, VEGF levels in the CNS would
be much higher. Since CD133
+
cells exist throughout the epen-
dyma of the lateral, third, and fourth ventricles (Figure 5A), as
well as the central canal of the spinal cord (Coskun et al., 2008),
but VEGF-expressing B cells (i.e., subependymal NSCs) are
only known to exist in the forebrain, it is possible that the lack of
B-cell-secreted VEGF is part of the reason for CD133
+
stem cells
not to manifest stem cell activity in the CNS regions other than the
forebrain, under normal conditions.
To establish a CD133
+
ependymal cell downstream-lineage-
tracing paradigm, we used ROSA26-tdTomato reporter mice
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. 1181
(Figure 4A). When a plasmid containing CD133 (prominin1) min-
imum promoter-driven Cre recombinase was electroporated into
postnatal mouse lateral ventricle walls, tdTomato-expressing
CD133-lineage cells were clearly labeled (Figures 4B–4I).
In this approach, only CD133
+
cells could acquire Cre recombi-
nase activity, which expels the stop cassette in the ROSA26
locus, allowing tdTomato to be permanently expressed not
only in the CD133
+
cells but also in their downstream lineage.
With this approach, the lineage included CD133
+
ependymal
cells, GFAP
+
astrocytes or B cells, and DCX
+
neuroblasts (Fig-
Figure 4. Labeling of the Progeny of CD133
+
Cells Surrounding the Lateral Ventricles
(A) A diagram illustrating the injection of mP2
plasmid into the lateral ventricle of ROSA26-
tdTomato animals followed by electroporation.
(B–E) Sagittal sections of P14 ROSA26-tdTomato
reporter mice showing tdTomato
+
cells (red)
around the lateral ventricle; the animals were in-
jected with mP2 followed by electroporation at P7.
Arrows in (E) point out migrating tdTomato
+
cells in
the RMS.
(F–I) Fluorescent images of ROSA26-tdTomato
forebrain sections stained with CD133 (F, green),
GFAP (G and H, green), and doublecortin (I, green).
The arrows in (F), (H), and (I) denote tdTomato
+
cells that are immunoreactive for CD133, GFAP,
and doublecortin, respectively.
Scale bars, 250 mm in (B) and (C); 50 mm in (D)–(H);
and 40 mm in (I). CC, corpus callosum, LV, lateral
ventricle, cp, choroid plexus.
ures 4B–4I), consistent with what we pre-
viously reported using ROSA26-LacZ re-
porter mice (Coskun et al., 2008).
Activation of CD133
+
Ependymal
Quiescent NSCs Lining the Fourth
Ventricle
To examine whether VEGF and bFGF
together can elicit activation of CD133
+
ependymal quiescent NSCs in non-
neurogenic brain regions, we injected
bFGF and/or VEGF into the fourth
ventricle and observed mitotic activation
of cells in the ependyma and the immedi-
ately adjacent parenchymal regions
(Figures 5A–5E). To trace the lineage of
activated CD133
+
ependymal cells lining
the fourth ventricle, we electroporated
a plasmid carrying prominin1 (CD133)
mP2 promoter driving Cre recombinase
(Coskun et al., 2008) into the fourth
ventricle of ROSA26-tdTomato reporter
mice, with or without co-injection of
VEGF and bFGF (Figure 5F). We found
that injection of VEGF and bFGF was suf-
ficient to mitotically activate the ependy-
mal CD133
+
cells as indicated by Ki67
immunoreactivity and BrdU incorporation
(Figures 5D and 5E). Following electroporation, tdTomato
+
cells
only appeared in the ependymal layer of the fourth ventricle in
control mice without VEGF or bFGF injections (Figure 5G). Simi-
larly, with bFGF injection alone, only a few tdTomato-labeled
cells were observed (Figure 5H). However, the addition of
bFGF and VEGF together triggered the CD133
+
ependymal
NSC activation process (Figures 5I–5L). tdTomato
+
cells,
indicative of the downstream lineage of CD133
+
ependymal
NSCs lining the fourth ventricle, were found to migrate into the
parenchyma and differentiate into Map2
+
neurons and GFAP
+
1182 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
astrocytes (Figures 5M–5P). These data indicate that at least a
subset, if not all, of the CD133
+
ependymal cells throughout
the CNS are dormant NSCs and, upon VEGF and bFGF stimula-
tion, as in the case of injury, can likely be mitotically activated
and differentiate into downstream neural-lineage cells.
DISCUSSION
In this study, through single-cell transcriptome analyses, we pro-
vide an initial high-resolution picture of the molecular biological
characteristics of the CD133
+
ependymal quiescent NSCs and
Figure 5. Activation of Dormant CD133
+
NSCs in the Fourth Ventricle by Treatment
of FGF and VEGF
(A) A sagittal section of adult fourth ventricular
region stained with CD133 (red).
(B–D) Immunostained sections of the postnatal
fourth ventricular region from control (B, saline)
and experimental (C, bFGF, and D, bFGF + VEGF
administration) groups showing numerous prolif-
erating Ki67
+
cells (green, arrows).
(E) A fluorescent image showing BrdU
+
cells
(green) double-labeled with CD133 (red) lining the
fourth ventricle following the local administration
of bFGF and VEGF.
(F) Diagram illustrating the experimental para-
digm of CD133
+
ependymal cell lineage-tracing
studies. The prominin1 minimum promoter-driven
Cre construct used for electroporation is shown.
(G–L) Sagittal sections of P14 ROSA26-tdTomato
reporter mice showing tdTomato
+
cells (red)
around the fourth ventricle; the animals were in-
jected with mP2 + saline (G), mP2 + bFGF (H), or
mP2 + FGF + VEGF (I–L) followed by electropo-
ration at P7. The inset in (I) shows the higher
magnification of tdTomato
+
cells. (K and L)
Fluorescent images of ROSA26-tdTomato mice
stained with CD133 showing tdTomato
+
(red) and
CD133
+
(green) cells lining the fourth ventricle
following the injection of FGF + VEGF and elec-
troporation of mP2.
(M–P) tdTomato
+
cells (red) in the parenchyma of
ROSA26-tdTomato reporter mice at P21; the ani-
mals were injected with mP2 + FGF + VEGF fol-
lowed by electroporation at P7. The bracket in (M)
indicates a group of cells at the ependymal layer
expressing tdTomato, while arrows point to some
of the apparently migrating tdTomato
+
cells. The
arrows in (N) denote tdTomato
+
cells with bushy
processes resembling astrocytes. (O) and (P)
show tdTomato
+
cells expressing MAP2 or GFAP,
respectively.
Scale bars, 100 mm in (A) and (G)–(J); 40 mmin
(B)–(E); and 75 mm in (M)–(P). 4V, fourth ventricle;
cb, cerebellum; cp, choroid plexus.
other cells in close proximity, some of
which could be the progeny of CD133
+
cells. While single-cell analyses in combi-
nation with WGCNA revealed much infor-
mation about the cells, modules that were
shared by subsets of samples appeared
to be quite effective in revealing common features or molecular
signatures of specific groups of cells (ependymal CD133
+
E
cells, A cells, B cells, etc.). Moreover, hub-gene analyses were
powerful in revealing important regulatory factors for specific
biological features of the subsets of cells. It is rather peculiar
that the E-module hubs are enriched for angiogenesis-related
and immune factors, as well as MSC and HSC markers CD105
and CD34, blurring the boundaries between neural and immune
or epithelial and mesenchymal lineages. Whether this stands for
a transitional state during EMT or whether tissue-specific stem
cells are vague in their germ-layer specificity is an interesting
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. 1183
question and remains to be determined. The hub-gene network
analysis was useful for discovering that VEGF was a critical
trigger to mitotically activate ependymal CD133
+
NSCs in
different parts of the ventricular system in the CNS. Given that
vasculature disruption and reconstruction are closely linked to
CNS injury and repair, our data support the notion that CNS
injury is one of the major stimuli that activates CD133
+
quiescent
NSCs. The fact that we were able to drive CD133
+
ependymal
cells lining the fourth ventricle into neurogenesis further supports
the idea that dormant NSCs exist throughout the CNS, lining the
ventricular surface including the central canal of the spinal cord,
and upon injury, those cells can be activated through exposure
to factors such as VEGF and potentially participate in neural
repair.
Although single-cell RNA-seq is set to be used to study cellular
diversity, it turned out to be particularly useful for the identifica-
tion of gene programs shared by subsets of cells, more so than
differential gene expression analyses using population-based
analyses, because, as discussed before, genes detected to be
highly expressed in population-based samples may not be ex-
pressed by every single cell in that population. Those genes
will surface in population-based differential expression analyses
and end up masking genes that are truly unique to cell subpop-
ulations. Therefore, for the discovery of shared common features
of cell subpopulations, single-cell transcriptome analyses have
proven to be rather effective.
Clearly, when the whole genome-wide transcriptome is
considered, there is substantial heterogeneity in gene expres-
sion patterns of the CD133
+
/GFAP
ependymal cell popula-
tion. Pearson’s correlation coefficients of pairwise samples
were about 0.6–0.7. Therefore, even if these cells are all quies-
cent NSCs, they may reside at different cellular states such as
being quiescent or at differently poised states for activation.
Alternatively, these cells might be innately distinct in that
some of them are structural cells and others are NSCs.
Additional single-cell transcriptome analyses will allow us to
fully unfold the heterogeneity of these cells as well as the
processes underlying quiescent stem cell activation during
injury in other regions of the CNS. The analyses might also
reveal the ongoing homeostatic stem cell activity in the fore-
brain. Moreover, this powerful method could be used in the
near future to successfully solve different biological problems
that are currently masked by cellular heterogeneity, including
the identification of cancer stem cells or molecular character-
istics of individual neurons that are functioning in neural
circuits.
EXPERIMENTAL PROCEDURES
Isolation of Single Cells from Postnatal SVZ
Three-week-old mice were anesthetized and euthanized in accordance with
Tongji University institutional guidelines. The brains were quickly removed
from the skull and put into cold PBS. After several washes with cold PBS at
37C for 30 min, the tissue was dissected under a dissection microscope,
and the walls of the lateral ventricles were obtained and enzymatically di-
gested using Papain (Worthington LS003127) at 37C for 30 min. Dissociated
cells were labeled with phycoerythrin (PE)-conjugated CD133 (eBiosciences
12-1331) for 1 hr. After several PBS washes, labeled cells were sorted using
the Becton Dickinson Fluorescence-Activated Cell Sorter to isolate CD133
+
or CD133
cells, followed by manual picking of single cells under the
microscope.
Fragment Library Construction and Sequencing
After the generation of cDNA fr om a single cell, 500 ng of cDNA was used
for the SOLiD system’s low-input fragment library preparation. Using the
Covaris S2 System (Covaris), cDNA was sheared into 100- to 150-bp short
fragments according to the manufacturer’s instructions, and the ends of the
target DNA were repaired using the end-polishing enzymes 1 and 2 (Epi-
centre ER0720). The sheared DNA was then purified with the SOLiD Library
Column Purific ation Kit (QIA GEN 28706) an d subsequently ligated to SOLiD
P1 and P2 adaptors . The resultin g ligated population was resolved on a 3%
agarose gel, and the 80- to 300-bp fractions were excised. The adaptor-
ligated DNA was subjected to nine cycles of PCR amplification. The ampli-
fied PCR products were purifi ed with the SOLiD Library Column Purifica tion
Kit (QIAGEN 28106) to yield the SOLiD fragment library ready for emulsion
PCR. Emulsion PCR reactions were performed by mixing 500-pg single-cell
libraries with 1.6 billion beads (1-mm diameter) with P1 primers covalently
attached to their surfaces. The 50-base-pair sequences were obtained us-
ing SOLiD sequencing.
WGCNA
A signed network was constructed using any gene that was expressed at an
FPKM value of 0.1 or higher in at least one of the samples. Soft power param-
eter was estimated and used to derive a pairwise distance matrix for selected
genes using the topological overlap measure, and the dynamic hybrid cut
method was used to detect clusters. The node centrality, defined as the
sum of within-cluster connectivity measures, was used to rank genes for
‘‘hub-ness’’ within each cluster. For visual analysis of the constructed
networks by hard thresholding of edge distances, the closest 150 edges
were represented using Cytoscape 3.0.0.
Intraventricular Injections, Electroporation, and BrdU
Administration
After postnatal C57B6 mice (wild-type or ROSA26-tdTomato) were fully
anesthetized by using isoflurane, a Hamilton syringe was lowered 2.5 mm
down to the later al ventricle from 0.3 mm Bregma and 1.2 mm lateral coor-
dinates using a st ereotaxic app aratus. For the fourth-vent ricle injections,
postnatal C57B6 mice (wild-type or ROSA26-tdTomat o), fully anesthetized
using isoflurane, received a Hamilton syring e lowered 2 mm down to t he
fourth ventr icle from 6.0 mm Bregma coordinates at the midline using a ste-
reotaxic apparatus. Subs equently, 2 ml of saline, fibroblast growth factor
(FGF), VEGF, FGF + VEGF, or mP2 was injected into the right lateral ventricle
or the fourth ventricle of postnatal animals. The animals that were injected
with mP2 were subjected to electroporation with the BTX ECM 830 electro-
porator (70 V, 100 ms). The cell proliferation marker BrdU was administered
orally from the day of injection for 7 days. Animals were perfused, and their
brain tissues were processe d as previousl y described (Coskun et al., 2008)
7 or 14 days after intraventricular injections.
Immunostaining
SuperFrost Plus slides (Fisher Scientific) with tissue sections were dried
at room temperature and rinsed a few times with PBS. For the detection
of BrdU-incorporated cells, the sections were immersed in 2 N hydrochloric
acid (HCl) at 50C for 30 min to denature the DNA and then rinsed
twice with 0.1 M borate buffer for 15 min. After several PBS washes, all
slides were imm ersed in blocki ng solution (2 % normal donkey serum
and 0.15% Triton X-100 in PBS) for 1 hr at room temperature. Primary anti-
body incubation was carried out overnight at 4Cwiththefollowinganti-
bodies at specified dilutions in blocking solution: anti-GFAP (Sigma),
1:1,000; anti-BrdU (BD Biosciences), 1:400; anti-CD133 (eBioscience),
1:1,000; anti-Flt-1 (Abcam), 1:500; MAP2 (Sigma), 1:750; doublecortin
(Abcam), 1:500; and Ki67 (Vector), 1:500). The following day, slides were
incubated with the appropriate Cy2- or Cy3-conjugated secondary anti-
bodies (Jacks on Immunoresearch) for 1 hr at room tempera ture at 1:500
dilution. Hoechst staining was used to label the nuclei. The slides were
examined by using Olympus fluorescent microscopes or a confocal system
1184 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
(Zeiss; Axioplan-LSM 510-META). Quantification of Ki67
+
,Ki67
+
/CD133
+
,
and BrdU
+
cells was performed by counting the labeled cells within the
ependyma or subependyma from three separa te experiments. Statistical
analysis was performed using a one-way ANOVA.
ACCESSION NUMBERS
The GEO accession number for the RNA-seq data and processed files re-
ported in this paper is GEO: GSE61288.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
five figures, and one table and can be found with this article online at http://
dx.doi.org/10.1016/j.cell.2015.04.001.
AUTHOR CONTRIBUTIONS
Y.L., Y.E.S., and S.L. designed the study. Y.L., V.C., A.L., J.Y., L.C., W.G., Z.S.,
C. Li, Y.C., D.L., J.W., C. Lin, Z.D., H.Z., J.Z., J.L., H. Liu, J.D.V., S.H., and S.L.
carried out experiments or contributed critical reagents and protocols. J.Y.,
W.G., K.Z., H. Lin, and Y.E.S. analyzed the data and performed statistical an-
alyses. Y.E.S., Y.L., V.C., J.Y., and S.L. wrote the manuscript. All the authors
read and approved the manuscript.
ACKNOWLEDGMENTS
We thank Gun Woo Byeon for his assistance in data analyses. We thank
Xiao Chen from Jiangxi Science & Technology Normal University for making
the graphical abstract and some figures. This work was supported by the
National Key Basic Research Program of China (grants 2011CBA01106,
2011CB965102, 2010CB945202, 2012CB966300, 2010CB945600,
2011CB966204, and 2014CB964602); the National Natural Foundation of
China (grants 91319309, 31271371, 81330030, 31271450, and 31271375);
and grants from the NIH (P01 GM081621-01A1), the Transcriptome and
Epigenetics Core of Center for Study of Opioid Receptors and Drugs of
Abuse (CSORDA center; grant NIH-P50DA005010), and the Intellectual
and Developmental Disabilities Research Center (IDDRC center; grant
NIH-P30HD004612) at the University of California, Los Angeles. The project
is also supported by Yunnan and Shanghai local grants 2012HA013,
2014FC004, 13XD1403600, and ZJ2014-ZD-002.
Received: August 26, 2014
Revised: January 5, 2015
Accepted: March 26, 2015
Published: May 21, 2015
REFERENCES
Beckervordersandforth, R., Tripathi, P., Ninkovic, J., Bayam, E., Lepier, A.,
Stempfhuber, B., Kirchhoff, F., Hirrlinger, J., Haslinger, A., Lie, D.C., et al.
(2010). In vivo fate mapping and expression analysis reveals molecular hall-
marks of prospectively isolated adult neural stem cells. Cell Stem Cell 7,
744–758.
Bonaguidi, M.A., Wheeler, M.A., Shapiro, J.S., Stadel, R.P., Sun, G.J., Ming,
G.L., and Song, H. (2011). In vivo clonal analysis reveals self-renewing and
multipotent adult neural stem cell characteristics. Cell 145, 1142–1155.
Codega, P., Silva-Vargas, V., Paul, A., Maldonado-Soto, A.R., Deleo, A.M.,
Pastrana, E., and Doetsch, F. (2014). Prospective identification and purifica-
tion of quiescent adult neural stem cells from their in vivo niche. Neuron 82,
545–559.
Coskun, V., Wu, H., Blanchi, B., Tsao, S., Kim, K., Zhao, J., Biancotti, J.C., Hut-
nick, L., Krueger, R.C., Jr., Fan, G., et al. (2008). CD133+ neural stem cells in
the ependyma of mammalian postnatal forebrain. Proc. Natl. Acad. Sci. USA
105, 1026–1031.
Doetsch, F. (2003). The glial identity of neural stem cells. Nat. Neurosci. 6,
1127–1134.
Doetsch, F., Caille
´, I., Lim, D.A., Garcı
´a-Verdugo, J.M., and Alvarez-Buylla, A.
(1999). Subventricular zone astrocytes are neural stem cells in the adult
mammalian brain. Cell 97, 703–716.
Fischer, J., Beckervordersandforth, R., Tripathi, P., Steiner-Mezzadri, A., Nin-
kovic, J., and Go
¨tz, M. (2011). Prospective isolation of adult neural stem cells
from the mouse subependymal zone. Nat. Protoc. 6, 1981–1989.
Ihrie, R.A., and Alvarez-Buylla, A. (2011). Lake-front property: a unique
germinal niche by the lateral ventricles of the adult brain. Neuron 70, 674–686.
Johansson, C.B., Momma, S., Clarke, D.L., Risling, M., Lendahl, U., and
Frise
´n, J. (1999). Identification of a neural stem cell in the adult mammalian
central nervous system. Cell 96, 25–34.
Kurimoto, K., Yabuta, Y., Ohinata, Y., and Saitou, M. (2007). Global single-cell
cDNA amplification to provide a template for representative high-density oligo-
nucleotide microarray analysis. Nat. Protoc. 2, 739–752.
Langfelder, P., and Horvath, S. (2008). WGCNA: an R package for weighted
correlation network analysis. BMC Bioinformatics 9, 559.
Li, L., and Clevers, H. (2010). Coexistence of quiescent and active adult stem
cells in mammals. Science 327, 542–545.
Lugert, S., Basak, O., Knuckles, P., Haussler, U., Fabel, K., Go
¨tz, M., Haas,
C.A., Kempermann, G., Taylor, V., and Giachino, C. (2010). Quiescent and
active hippocampal neural stem cells with distinct morphologies respond
selectively to physiological and pathological stimuli and aging. Cell Stem
Cell 6, 445–456.
Martens, D.J., Seaberg, R.M., and van der Kooy, D. (2002). In vivo infusions of
exogenous growth factors into the fourth ventricle of the adult mouse brain in-
crease the proliferation of neural progenitors around the fourth ventricle and
the central canal of the spinal cord. Eur. J. Neurosci. 16, 1045–1057.
Meacham, C.E., and Morrison, S.J. (2013). Tumour heterogeneity and cancer
cell plasticity. Nature 501, 328–337.
Merkle, F.T., Mirzadeh, Z., and Alvarez-Buylla, A. (2007). Mosaic organization
of neural stem cells in the adult brain. Science 317, 381–384.
Ming, G.L., and Song, H. (2011). Adult neurogenesis in the mammalian brain:
significant answers and significant questions. Neuron 70, 687–702.
Mirzadeh, Z., Merkle, F.T., Soriano-Navarro, M., Garcia-Verdugo, J.M., and Al-
varez-Buylla, A. (2008). Neural stem cells confer unique pinwheel architecture
to the ventricular surface in neurogenic regions of the adult brain. Cell Stem
Cell 3, 265–278.
Morrison, S.J., and Spradling, A.C. (2008). Stem cells and niches: mechanisms
that promote stem cell maintenance throughout life. Cell 132, 598–611.
Nakafuku, M., Nagao, M., Grande, A., and Cancelliere, A. (2008). Revisiting
neural stem cell identity. Proc. Natl. Acad. Sci. USA 105, 829–830.
Nolan, D.J., Ginsberg, M., Israely, E., Palikuqi, B., Poulos, M.G., James, D.,
Ding, B.S., Schachterle, W., Liu, Y., Rosenwaks, Z., et al. (2013). Molecular sig-
natures of tissue-specific microvascular endothelial cell heterogeneity in organ
maintenance and regeneration. Dev. Cell 26, 204–219.
Olsza
nska, B., and Borgul, A. (1993). Maternal RNA content in oocytes of
several mammalian and avian species. J. Exp. Zool. 265, 317–320.
Ramos, A.D., Diaz, A., Nellore, A., Delgado, R.N., Park, K.Y., Gonzales-Roybal,
G., Oldham, M.C., Song, J.S., and Lim, D.A. (2013). Integrat ion of genome-
wide approaches identifies lncRNAs of adult neural stem cells and their prog-
eny in vivo. Cell Stem Cell 12, 616–628.
Ramsko
¨ld, D., Luo, S., Wang, Y.C., Li, R., Deng, Q., Faridani, O.R., Daniels,
G.A., Khrebtukova, I., Loring, J.F., Laurent, L.C., et al. (2012). Full-length
mRNA-Seq from single-cell levels of RNA and individual circulating tumor
cells. Nat. Biotechnol. 30, 777–782.
Shalek, A.K., Satija, R., Adiconis, X., Gertner, R.S., Gaublomme, J.T., Ray-
chowdhury, R., Schwartz, S., Yosef, N., Malboeuf, C., Lu, D., et al. (2013).
Single-cell transcriptomics reveals bimodality in expression and splicing in im-
mune cells. Nature 498, 236–240.
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. 1185
Shapiro, E., Biezuner, T., and Linnarsson, S. (2013). Single-cell sequencing-
based technologies will revolutionize whole-organism science. Nat. Rev.
Genet. 14, 618–630.
Snippert, H.J., and Clevers, H. (2011). Tracking adult stem cells. EMBO Rep.
12, 113–122.
Tang, F., Barbacioru, C., Wang, Y., Nordman, E., Lee, C., Xu, N., Wang, X.,
Bodeau, J., Tuch, B.B., Siddiqui, A., et al. (2009). mRNA-Seq whole-transcrip-
tome analysis of a single cell. Nat. Methods 6, 377–382.
Tang, F., Barbacioru, C., Nordman, E., Li, B., Xu, N., Bashkirov, V.I., Lao, K.,
and Surani, M.A. (2010). RNA-Seq analysis to capture the transcriptome land-
scape of a single cell. Nat. Protoc. 5, 516–535.
Tietjen, I., Rihel, J.M., Cao, Y., Koentges, G., Zakhary, L., and Dulac, C. (2003).
Single-cell transcriptional analysis of neuronal progenitors. Neuron 38,
161–175.
Uchida, N., Buck, D.W., He, D., Reitsma, M.J., Masek, M., Phan, T.V., Tsuka-
moto, A.S., Gage, F.H., and Weissman, I.L. (2000). Direct isolation of human
central nervous system stem cells. Proc. Natl. Acad. Sci. USA 97, 14720–
14725.
Wu, J., and Tzanakakis, E.S. (2013). Deconstructing stem cell population het-
erogeneity: single-cell analysis and modeling approaches. Biotechnol. Adv.
31, 1047–1062.
Xue, Z., Huang, K., Cai, C., Cai, L., Jiang, C.Y., Feng, Y., Liu, Z., Zeng, Q.,
Cheng, L., Sun, Y.E., et al. (2013). Genetic programs in human and mouse early
embryos revealed by single-cell RNA sequencing. Nature 500, 593–597.
Yan, L., Yang, M., Guo, H., Yang, L., Wu, J., Li, R., Liu, P., Lian, Y., Zheng, X.,
Yan, J., et al. (2013). Single-cell RNA-Seq profiling of human preimplantation
embryos and embryonic stem cells. Nat. Struct. Mol. Biol. 20, 1131–1139.
Zhang, B., and Horvath, S. (2005). A general framework for weighted gene co-
expression network analysis. Stat. Appl. Genet. Mol. Biol. Published online
August 12, 2005. http://dx.doi.org/10.2202/1544-6115.1128.
Zhao, C., Deng, W., and Gage, F.H. (2008). Mechanisms and functional impli-
cations of adult neurogenesis. Cell 132, 645–660.
1186 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
Supplemental Figures
Figure S1. Structure of the Adult Rodent Forebrain Neurogenic Zone and Pipeline for Isolation of CD133
+
or CD133
Cell in the SVZ, Related
to Figure 1
A diagram of major cell types in the adult mouse subventricular/subependymal zone (SVZ/SEZ). Multi-ciliated ependymal cells (E, blue) line the lateral ventricles
(LV). We have previously demonstrated that a subpopulation of CD133
+
E cells represent an additional -perhaps more quiescent- stem cell population (Coskun
et al., 2008). Neuroblasts (A, red) ensheathed by glial tunnels migrate along the rostral migratory stream (RMS). Glial tunnels are formed by SVZ astrocytes (B,
royal blue), which are regarded also as SEZ neural stem cells (NSCs). Rapidly dividing transit-amplifying cells (C, green) are the transitional cells from B to A cells.
SVZ astrocytes or NSCs are GFAP
+
, neuroblasts are GFAP
-
, Dlx2
+
, Dcx
+
, PSA-NCAM
+
, and C cells are GFAP
-
, Dlx2
+
, PSA-NCAM
-
. A schema for single-cell
isolation by FACS is presented.
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. S1
Figure S2. Quality Control and Additional Data Analyses of RNA-Seq Results, Related to Figure 1
(A) ERCC Spike-in controls. 0.5ul of ERCC RNA mix at 1:400 dilution, when added into 5 ul of 50pg total mouse RNA (approximately the amount of RNA in 5 mouse
cells) in the ‘‘lysis-reverse-transcription’’ system, generated about equal amount of reads (ERCC RNA versus mouse cell poly-A RNA) after sequencing (Left
panel), suggesting ERCC mix and RNAs from about 5 mouse cells contain same amount of poly-A RNA. Similarly, when 0.5 ul of ERCC Spike-In Mix (1:4,000
dilution) was spiked into a half of a single-cell lysis mix prior cDNA synthesis (about 5pg of total RNA in 5ul reaction system), also generated similar amount of
reads. Scatter plot showing the reads counts of ERCC transcripts against the attomoles per transcript spiked in. Axes are plotted on a log2 scale. The high ‘‘r’’
values suggest that our single-cell RNA sequencing system could reliably and quantitatively detect small (single-cell level) amount of poly-A RNAs in the samples.
(B) Saturation curves of RNA sequencing data. Saturation curves of RNA sequencing data of one pooled sample (P2) and two single-cell samples [S10 (CD133
+
/
GFAP
+
) and S1 (CD133
+
/GFAP
-
)] with FPKM cutoff from 0.01 to 1.
(C) Comparison of RNA-Seq and TaqMan real-time PCR measures. Prom1, MBP, CNP and GFAP identified by RNA-Seq analysis were re-evaluated by TaqMan
real-time PCR using the same corresponding cDNA samples respectively. Reasonably good correlations were found between RNA-seq and q-PCR results.
(D) Coverage of sequenced transcripts and gene saturation. Plots of 50to 30coverage of sequenced transcripts. We included published dataset by Ramsko
¨ld
et al., 2012 as reference.
(E) Distribution of abundantly expressed and low-level expressed genes in pooled data sets. Densities (number) of genes expressed at different levels (Log
2
FPKM) were plotted.
(F) Distribution of abundantly expressed and low-level expressed genes in single-cell data sets. Densities (number) of genes expressed at different levels (Log
2
FPKM) were plotted.
S2 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
Figure S3. Single-Cell Transcriptome Analyses Provide More Sensitive Data with High Resolutions, Related to Figure 1
(A) Experimental design of single cell and pooled cell transcriptome analyses.
(B) Gene detection overlap between 28 individually sequenced single cells in combina tion and 4 10 cell-pooled samples.
(C) Gene detection overlap of the exact same 17 cells when cells were sequenced as individual cells (green) or when 10 CD133
+
cells were pooled and 7 CD133
-
cells were pooled and then sequenced as two populations (pink and blue respectively); FPKM cutoff was 0.1, 1, and 10, respectively. Single-cell analyses
detected more genes than pooled samples.
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. S3
Figure S4. Analysis of lncRNA from Mouse Neurogenic Zone Cells by Single-Cell RNA-Seq, Related to Figure 1
(A) Pipeline for identification of novel long non-coding RNAs.
(B) Venn diagram depicting the overlap between our catalog of SVZ-associated lncRNAs and UCSC Genome Browser BED File of Identified lncRNAs.
(C) Novel lncRNAs were categorized based on intersection with protein-coding genes including five categorie s, which are sense, antisense, intronic, bidirectional,
and intergenic.
S4 Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc.
Figure S5. Expression Levels of 30 Hub Genes of the E Module in CD133
+
GFAP
Single-Cell and Pooled Sequenced Samples, Related to
Figure 2
The black dots represent 8 initially sequenced single cells. The stars in blue represent additionally sequenced 5 single cells. The stars in red represent 10-cell
pools of CD133
+
GFAP
-
ependymal cells.
Cell 161, 1175–1186, May 21, 2015 ª2015 Elsevier Inc. S5
