Review Article
Single-cell gene expression proﬁling using reverse transcription quantitative
real-time PCR
Anders Ståhlberg a,b,*, Martin Bengtsson b,c,1
a
Lundberg Laboratory for Cancer, Department of Pathology, Sahlgrenska Academy at University of Gothenburg, Gula Straket 8, 413 45 Gothenburg, Sweden
b
TATAA Biocenter, Odinsgatan 28, 411 03 Gothenburg, Sweden
c
Oxford Centre for Diabetes, Endocrinology & Metabolism, University of Oxford, Churchill Hospital, Oxford OX3 7LJ, UK
a r t i c l e i n f o
Article history:
Accepted 7 January 2010
Available online 11 January 2010
Keywords:
Cell heterogeneity
Cell lysis
Gene expression
qPCR
Real-time PCR
Reverse transcription
RT-qPCR
Single cell
Single-cell biology
Single-cell gene expression
a b s t r a c t
Even in an apparently homogeneous population of cells there are considerable differences between individual
cells. A response to a stimulus of a cell population or tissue may be consistent and gradual while
the single-cell response might be binary and apparently irregular. The origin of this variability may be
preprogrammed or stochastic and a study of this phenomenon will require quantitative measurements
of individual cells. Here, we describe a method to collect dispersed single cells either by glass capillaries
or ﬂow cytometry, followed by quantitative mRNA proﬁling using reverse transcription and real-time
PCR. We present a single cell lysis protocol and optimized priming conditions for reverse transcription.
The large cell-to-cell variability in single-cell gene expression measurements excludes it from standard
data analysis. Correlation studies can be used to ﬁnd common regulatory elements that are indistinguishable
at the population level. Single-cell gene expression proﬁling has the potential to become common
practice in many laboratories and a powerful research tool for deeper understanding of molecular
mechanisms.
Ó 2010 Elsevier Inc. All rights reserved.
1. Introduction
Cells have a remarkable ability to cooperate and jointly construct
complex structures such as tissues, organs and whole organisms.
These constructions are normally accurately tuned and
respond to stimuli with high precision. During development, cells
differentiate to specialized cell types, each with particular functions
in the environment they reside. In many aspects, individual
cells exhibit a high degree of variability and responses to identical
stimuli may be very different even in a seemingly homogeneous
population [1–6].
Gene expression proﬁling is a pivotal research tool in molecular
biology. By default, measurements are made on large pools of cells
as this will increase reliability of the recordings. For tissues, different
cell types are mixed uncontrollably and the measured gene
expression proﬁle has unknown contribution from different cell
types. In addition, cell population measurement will not reveal
how a particular transcript is distributed among the cells
(Fig. 1A–B). Bulk measurements easily miss potentially important
gene correlations (Fig. 1C–D) where single cell analysis would indicate
coupled transcriptional regulations, which might be controlled
by the same molecular mechanism (Fig. 1C–D) [7].
Observed heterogeneity may indicate the presence of specialized
cell types or originate in the random nature of the transcription
machinery [1–6].
A typical single cell contains $1 pg mRNA, which is equivalent
to a few hundred thousand molecules transcribed from about ten
thousand genes [8]. The high sensitivity of reverse transcription
quantitative real-time PCR (RT-qPCR) makes it possible to detect
even a single molecule. RT-qPCR is also characterized by high
reproducibility and wide dynamic range [9–11]. These properties
make RT-qPCR suitable for single-cell gene expression proﬁling.
Even if single-cell gene expression proﬁling using RT-qPCR has
been successfully applied to several different applications [6,12–
14], it has still not become common practice for laboratories. In
this paper, we describe the workﬂow of single cell RT-qPCR including:
cell collection, cell lysis, RT, qPCR and data analysis.
2. Description of method
Single cell RT-qPCR constitutes several sequential steps, outlined
in Fig. 2. Our intention with this paper is to present the most
common experimental approaches with suitable references and in
1046-2023/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.ymeth.2010.01.002
* Corresponding author. Address: Lundberg Laboratory for Cancer, Department of
Pathology, Sahlgrenska Academy at University of Gothenburg, Gula Straket 8, 413
45 Gothenburg, Sweden. Fax: +46 31828733.
E-mail address: anders.stahlberg@neuro.gu.se (A. Ståhlberg).
1
Present address: Nuevolution A/S, Ronnegade 8, 2100 Copenhagen, Denmark.
Methods 50 (2010) 282–288
Contents lists available at ScienceDirect
Methods
journal homepage: www.elsevier.com/locate/ymeth
detail describe the most appropriate experimental setup for analysis
of cells in suspension. For successful single-cell gene expression
proﬁling good laboratory practice is essential. All RNA work requires
completely RNase free conditions. Furthermore, PCR contamination
must be avoided, since even negative (zero values)
are used in data analysis. Physical separation, i.e. different rooms,
of pre-PCR, PCR and post-PCR is recommended.
3. Single cell collection and lysis
A majority of single-cell studies published deploy one of three
methods to collect cells: Flow cytometry, glass capillaries, and laser
capture. This review will describe the two former while laser
capture and laser microdissection are described in detail elsewhere
[15–17]. For some applications, the origin and surroundings of a
cell is vital information and this excludes ﬂow cytometry as collection
method, leaving laser capture as best option. Where high
numbers of cells are needed, and where collections of intact cells
are of importance, ﬂow cytometry is recommended. Glass capillaries,
as we use them in this article, will also collect intact cells.
Single cell suspensions are prepared from tissue using mechanical
separation, enzymatic treatment, and/or non-physiological
buffers. The yield of functionally viable, dissociated cells will be
dependent on several parameters such as concentrations, incubation
times and temperatures. An informative website for different
cell dissociation approaches is: www.tissuedissociation.com. As
the expression of some genes may be altered by the cell treatment
it is recommended that the expression of the genes of interest is
quantiﬁed also in an untreated sample. For example, one part of
the biological sample is saved for total RNA isolation and the
remaining is used for cell dissociation and collection.
3.1. Single cell sorting using ﬂow cytometry
Several ﬂow cytometry instruments, such as FACSDiva and
FACSVantage (both BD Biosciences, San Jose, CA, USA) can be used
to sort out individual cells [18]. PCR plates (96-well) with lysis buffer
should be prepared in advance. In addition to standard ﬂow
cytometry calibration, the instrument needs to be carefully calibrated
to deposit single cells in the center of each collection tube.
This can easily be tested by sorting $50 beads/cells on the plastic
Fig. 1. Cell heterogeneity and correlated transcript levels. Single cell measurements can distinguish the cases in (A) and (C) from (B) and (D), respectively, while cell
population measurements cannot.
Fig. 2. Overview of single-cell gene expression proﬁling using RT-qPCR.
A. Ståhlberg, M. Bengtsson / Methods 50 (2010) 282–288 283
ﬁlm covering the 96-well PCR plate. It can also be worthwhile to
check the calibration every second plate, because the sorting arm
may be displaced over time. A small volume of lysis buffer will increase
the risk of a cell sticking to the wall of the tube, while too
large volume will likely interfere with downstream reactions. We
have found that $5 ll lysis buffer is a suitable volume to work with
using ﬂow cytometry.
For practical reason, a signiﬁcant number of cells are needed for
calibration, and thousands of cells are wasted by the ﬂow cytometry
instrument. For most cell types it is recommended to keep the
cells on ice before sorting. If speciﬁc antibodies are used to sort out
subpopulations, cells need to be kept in medium that do not interfere
with the ﬂuorophores used (PBS with $2.5% serum is usually
compatible with both the ﬂuorescence-activated cell sorting procedure
and keeping the cells viable). If no ﬂuorophores are used, cells
should be kept on cell type-speciﬁc medium to avoid altered gene
expression. Viable cells can be enriched by ﬂow cytometry by adding
7-aminoactinomycin D (Sigma–Aldrich) before sorting. As a positive
control 1–2 wells may be used to collect $30 cells. This
number of cells will level out most single cell variability, and if
these controls result in negative expression data, reliable single cell
data for that particular gene will most likely be hard to obtain.
After ﬁnished sorting, each 96-well PCR plate should be sealed
and put on dry ice and stored at À80 °C until RT. As a control for
contamination, some liquid without cells can be collected from
the ﬂow cytometry instrument to eliminate the chance that
mRNAs from lysed cells have contaminated the collected cells.
3.2. Single cell collection using glass capillaries
This collection method requires a phase contrast microscope,
preferably inverted, mounted with micromanipulators [6,19].
Shortly prior cell collection, the dissociated cells should be re-plated
on Petri dishes, allowing cell adhesion. Usually, several dishes
can be prepared and maintained in optimal conditions until single
cell collection. Depending on the use of ﬂuorophores, dissociated
cells should be kept on cell speciﬁc medium or medium which
do not interfere with the ﬂuorophores used. Cell collection is simpliﬁed
if the cells have attached slightly to the dish and are physically
separated from each other. Cells may be washed before
collection to remove dead cells and loose debris that could contaminate
the pipette.
Borosilicate glass capillaries (Hilgenberg, GmbH) with outer
diameter of 1.5–1.6 mm and wall thickness of 0.16 mm can be
pulled to pipettes using a patch-clamp pipette puller (Heka PIP5).
The diameter of the tip should be $10 lm, but the width is adjusted
to cell size. This is substantially wider than standard
patch-clamp pipettes and large enough to allow passage of an intact
cell ensuring collection of all mRNA. An alternative method
is to penetrate the cell membrane with a patch-clamp pipette
(diam. $0.5 lm) and only collect the cytoplasm [20,21]. This has
the advantage of allowing mRNA harvest from intact tissue, and
in conjunction with patch-clamp recording. However, it is technically
very challenging which limits the number of samples that
can be collected within reasonable time limits. In addition, normalization
of expression levels poses a particular difﬁculty when an
unknown fraction of the cell’s mRNA is harvested.
The glass pipette is mounted on a hydraulic micromanipulator
(Narishige or equivalent) on the inverted microscope and connected
by a ﬂexible plastic tubing to the mouth of the researcher. If the pipette
size is right it is readily feasible to control the pressure inside the
pipette and collect individual cells with minimum volume of extracellular
solution ((0.1 ll). Pipettes should be emptied in standard
PCR tubes containing 2 ll lysis solution. This small volume allows
high concentration of detergents needed for cell lysis and RNase
inhibition. To simplify the process of emptying the glass capillary,
a stand with tube holder and an adjustable capillary mount is recommended
(Fig. 3). Careful breaking of the glass tip to the bottom of the
tube facilitates complete emptying of the capillary. We have not detected
any inhibition of RT or difference in transcript number for
cells with or without small traces of borosilicate glass (data not
shown). PCR tubes containing single cells should be stored on ice
during collection and kept in À80 °C until RT. As a control for contamination
and inhibition of downstream reactions, $1 ll of the
buffer or medium surrounding the cells should be collected and analyzed
together with the single-cell samples.
3.3. Cell lysis
The amount of protein, salts and debris in the single-cell samples
is very small, allowing us to omit mRNA puriﬁcation. We have
Fig. 3. Glass capillary emptying. The tip of the glass capillary contains a single cell that is emptied into the PCR tube containing 2 ll lysis buffer. To speed up and simplify the
procedure, our local workshop constructed a stand for the tube and a moveable capillary holder allowing precise emptying.
284 A. Ståhlberg, M. Bengtsson / Methods 50 (2010) 282–288
pooled up to ten cells, each collected individually, and not observed
any inhibitor effects on reverse transcription [19]. However,
for efﬁcient and complete RT we need to ascertain that the cell is
lysed and that all available mRNA is accessible to the enzyme. Cell
lysis can be performed with heat, mechanic forces, osmotic pressure,
enzymatic treatment and detergents. In addition to efﬁcient
cell lysis, RNase activity needs to be blocked. RNase activity can
be inhibited by RNase inhibitors, such as RNaseOUT (20 U, Invitrogen),
and chaotropic salts. For mammalian cells, especially mouse
b-cells, we have found that the use of guanidine thiocyanate is efﬁcient
to lyse cells. In addition, low concentrations ($50 mM) of
guanidine thiocyanate is compatible with RT, for some genes it
may even enhance the RT efﬁciency and act as an RNase inhibitor
[19]. The maximum concentration of guanidine thiocyanate that
may be used in the lysis buffer depends on the dilution factor between
lysis and reverse transcription. For mouse b-cells, 2 ll lysis
buffer containing 250 mM guanidine thiocyanate is sufﬁcient for
efﬁcient lysis and also compatible with 10 ll RT reactions. RNA/
DNA carrier may be added to the lysis buffer, but we have not
found any positive effects from the use of carrier in our experimental
setup. We have also evaluated heat incubation during lysis
(80 °C, 5 min) but it does not signiﬁcantly affect mRNA recovery
(data not shown). Commercial single cell lysis buffers, such as
CellsDirect (Invitrogen) and CelluLyser (TATAA Biocenter), may
also be used without additional extraction or puriﬁcation.
3.4. RNA spike
As a control for the entire workﬂow from single cell collection
to data analysis, an RNA spike may be added in the lysis buffer.
Fig. 4. RT priming. Identical amounts of puriﬁed total RNA was used as starting material and four genes were measured (Ins1, Ins2, Rps29 and Hprt). (A) Determination of
optimal RT primer concentration using oligo(dT), random hexamers and gene speciﬁc primer. The gene speciﬁc primer was identical to reverse qPCR primer. Agarose gel
electrophoresis analysis for Ins2 revealed that increasing concentration of gene speciﬁc primer resulted in formation of erroneous PCR products. Melting curve analysis could
not distinguish between correct and erroneous PCR products. Unspeciﬁc PCR products were observed for Ins2, Hprt and Rps29 using gene-speciﬁc primers. (B) Relative RT
reaction yields are shown for various primer combinations. 2.0 lM oligo(dT), 2.0 lM random hexamers and 0.25 lM GSP was used for all RT primer combinations. Agarose gel
inspection showed formation of erroneous PCR products for Hprt and Ins2 using gene speciﬁc primer. The Ins1 assay was speciﬁc for all conditions while Rps29 occasionally
formed erroneous PCR products using gene speciﬁc primer. Any combination of priming methods was invariably superior or equal to the single best priming method used.
Relative cDNA yield was arbitrarily set to a value of one for the gene with lowest expression in the group of genes with the lowest RT primer concentration tested in (A) and in
left group (Olig) in (B) [22]. Values are means ± SEM for 3 separate experiments. Abbreviations used: Olig, oligo(dT); Rand, random hexamer; GSP, gene speciﬁc primer.
A. Ståhlberg, M. Bengtsson / Methods 50 (2010) 282–288 285
However, the use of RNA spike will not reveal any information
about cell quality, lysis efﬁciency, mRNA accessibility and mRNA
quality. Setting up single-cell gene expression proﬁling for the ﬁrst
time or changing the experimental setup, RNA spikes may be useful
to include as a control for RNA stability, RT and qPCR. From our
experience, data from the RNA spike always results in almost identical
Cq-values if all single cells are run in parallel. RNA spikes can
either be commercial or generated by in vitro transcription [19].
The sequence of the RNA spike should be unique compared to
the transcriptome of the single cells analyzed. Alternatively, the
RNA spike should be used in high excess compared to the endogenous
expression.
3.5. Reverse transcription
High RT efﬁciency is needed to ensure that low transcript numbers
can be detected and quantiﬁed by qPCR. There are several different
reverse transcriptases and priming methods available. We
have previously shown that the variability in RT efﬁciency is highly
gene dependent [22,23]. Therefore, we prefer to use a blend of oligo(dT)
and random hexamers to prime the reverse transcription
reaction (Fig. 4). Gene-speciﬁc primers can bind unspeciﬁcally to
the mRNA [22] leading to cDNA similar to that of random hexamer
priming, but with the gene speciﬁc primer in the 30
end. If the gene
speciﬁc primer is identical to the reverse PCR primer, the speciﬁcity
will be reduced in the PCR and the number of products unrelated
to the intended target will increase. Use of gene speciﬁc
primer identical to the reverse primer should therefore be carefully
optimized before use or avoided completely. It may also be worthwhile
to test different reverse transcriptases, as the efﬁciency varies
greatly and gene-dependently [23].
For many genes we have seen that the use of SuperScript III
(Invitrogen) gives sufﬁcient cDNA yield. We frequently use the following
reverse transcription protocol: A mix of a single cell in 2 ll
lysis buffer (containing 50 mM guanidine thiocyanate, Sigma–Aldrich),
0.5 mM dNTP (Sigma–Aldrich), 2.0 lM oligo (dT15, Euroﬁns
MWG Operon), 2.0 lM random hexamers (Euroﬁns MWG Operon)
is incubated at 65 °C for 5 min in a total volume 6.5 ll. dNTP, oligo(dT15)
and random hexamers can be added directly to the frozen
cells without thawing the samples. Beware of batch differences of
oligo(dT) and random hexamers; check new batches before using
them. Then cool the samples on ice or cooling block to <25 °C.
Add 50 mM Tris–HCl (pH 8.3), 75 mM KCl, 3 mM MgCl2, 5 mM
dithiothreitol, 20 U RNaseOUT and 50 U SuperScript III (all Invitrogen)
to total volume of 10 ll. The ﬁnal concentration of guanidine
thiocyanate in RT should be 650 mM. The following temperature
proﬁles may be used for RT: 25 °C for 5 min, 50 °C for 60 min,
55 °C for 15 min and 70 °C for 15 min. Only minor differences have
been observed using other incubation times and temperatures as
long the reverse transcriptases have not been inactivated by heat.
To increase the amount of mRNA, pre-ampliﬁcation methods may
be applied [24,25].
The quantitative contribution from genomic DNA to the expression
signal is generally insigniﬁcant, as it would only increase the
copy number by two for most genes. Nevertheless, PCR primers
should whenever possible be designed to cross introns. DNase
treatment is another option, but difﬁcult to evaluate in single-cell
samples. For +/À measurements however, both these precautions
are required.
3.6. Quantitative real-time PCR
The number of genes that can be analyzed from a single cell is
limited by the number of transcripts of the studied genes. We have
shown that >$20 target molecules per PCR are needed for accurate
quantiﬁcation [19]. Furthermore, we have shown that highest
reproducibility for measurements of limited samples is achieved
when the dilution between RT and qPCR is kept at a minimum,
even at the expense of qPCR-replicates [19]. For example, the accuracy
in qPCR will be higher if all cDNA is quantiﬁed in a single qPCR
reaction instead of splitting the cDNA sample in two and analyze
the duplicate. However, loading too much of the RT reaction may
inhibit the qPCR, in part due to inhibitory effects of reverse transcriptase
enzyme on PCR [19,21,26–28]. The inhibition depends on
the amount of reverse transcriptase and Taq polymerase used in
the two reactions. Using SuperScript III (Invitrogen) and JumpStart
(Sigma–Aldrich) the ratio should be <2 (U/U). This ratio may be different
using other enzymes. Practically, about 5–10 genes can successfully
be analyzed from the same cell, a higher number would
require pre-ampliﬁcation of cDNA.
Pre-ampliﬁcation of single-cell cDNA allows virtually limitless
number of measured genes, and has been used on single cells in
concert with microarrays [29]. In a pioneering work by Brady
and Iscove [30], it was shown that polyA-cDNA can be globally
ampliﬁed, followed by gene-speciﬁc PCR on the resulting pool.
Alternatively, a multiplex PCR on the single-cell cDNA, using low
concentrations of all gene-speciﬁc primers, results in a population
of PCR products that are quantiﬁed by qPCR. This way, up to 40
[31] and 100s (TaqManÒ
PreAmp Master Mix, p/n 4391128, Applied
Biosystems) of transcripts have been analyzed in a single cell.
However, pre-ampliﬁcation introduces one extra source of potential
bias which needs to be veriﬁed. For more information on
pre-ampliﬁcation methods, the reader is referred elsewhere
[24,25].
Normal qPCR guidelines are used to design working assays for
single-cell gene expression proﬁling [9,32]. Assays should be gene
speciﬁc, have high sensitivity, dynamic range and reproducibility
[9,32]. For most real-time PCR instruments, typical quantiﬁcation
cycle (Cq) values are between 30 and 40 analyzing single cells.
Many single cells will and should not result in any detectable
PCR product (no Cq-value). Using melting curve analysis, non-speciﬁc
PCR products may be identiﬁed. To eliminate detection and
quantiﬁcation bias, assays should not give rise to any primer-dimers,
at least not before Cq = 40. From our experience is the assay
speciﬁcity, i.e. lack of erroneous PCR products such as primer-dimers,
one of the most important parameter when optimizing single
cell assays. If non-speciﬁc PCR fragments are formed in
addition to the correct PCR product, data should only be used qualitatively.
Probe-based qPCR assays can be used but they will not reveal
important information about interfering PCR products that are
formed. Another alternative is to use one-step RT-qPCR assays.
Running RT and qPCR in the same reaction tube eliminates one
handling step and consequently reduces the risk of contamination.
The main drawback with one-step RT-qPCR is that RT and qPCR
most often have different optimal reaction conditions, reducing
the sensitivity, efﬁciency and reproducibility of respective reactions.
All assays should be validated by gel electrophoresis, even
if probes are used.
The following qPCR protocol has been proven to work well:
10 ll reactions containing 10 mM Tris (pH 8.3), 50 mM KCl,
3 mM MgCl2, 0.3 mM dNTP, 1 U JumpStart Taq polymerase (all Sigma–Aldrich),
0.5 Â SYBR Green I (Invitrogen), 200–400 nM of each
primer (Euroﬁns MWG Operon) and 1–4 ll cDNA (cDNA should be
diluted to avoid qPCR inhibition from the RT reaction). Premade
qPCR mixes can also be used. However, assay performance usually
differs somewhat between qPCR mixes and should not be changed
between runs for samples to be compared. Primers should be designed
according to standard procedure [9,11]. Primer3 (http://
www.bioinformatics.nl/cgi-bin/primer3plus/primer3plus.cgi.) is
one freely available primer design software that may be used.
For samples with low RNA quality short amplicon lengths are recommended
[33,34]. The following temperature proﬁle may be ap-
286 A. Ståhlberg, M. Bengtsson / Methods 50 (2010) 282–288
plied for most qPCR instruments: 95 °C for 3 min followed by 50
cycles of ampliﬁcation (95 °C for 20 s, 60 °C for 20 s, and 72 °C
for 20 s). The predenaturation step is to activate the antibody-inactivated
Taq DNA polymerase and the cycling program is for primers
designed with melting temperatures of 60 °C using Primer3 and
amplicon lengths <250 bp.
3.7. Data analysis
Single-cell gene expression data may either be determined as
relative quantities or as absolute mRNA/cDNA numbers. The efﬁciency
of reverse transcription is gene dependent and <100%
[22,23]. Hence, the number of cDNA molecules is a lower-limit
estimate of the number of mRNA molecules present in the cell.
The RT efﬁciency can be estimated experimentally by using known
amounts of synthetic RNA mimicking the properties of the native
mRNA. However, producing full length mRNAs is challenging and
it is common to use a truncated RNA molecule, which may not
be reverse transcribed with the same efﬁciency as the native
mRNA. Alternatively, the RT-qPCR variability can be used to determine
the reverse transcription efﬁciency using mathematical modeling
[19]. The number of cDNA molecules can be determined
using standards with known concentrations as described [9–11].
If double-stranded DNA standards are used, one cycle should be
subtracted from each acquired Cq-value to compensate for the fact
that cDNA is single stranded [22].
Fig. 5A shows representative Cq-values of two genes with medium
(Chgb) and high (Ins2) expression in 37 individual mouse bcells
[19]. Fig. 5B shows the distribution of Ins2-expressing cells
in linear-scale and Fig. 5C shows the distribution in log10-scale. It
is a highly skewed distribution where $40% of all transcripts originate
from only three cells, resulting in a better ﬁt for lognormal
distribution than for normal distribution (p = 0.63 and p < 0.01,
respectively, with a Shapiro–Wilk normality test where a high value
corresponds to a good ﬁt). In lognormal distributions, the typical
cell is best characterized by the geometric mean. Extrapolation
of cell population data to single cell data by dividing the total
expression by the number of cells, i.e. arithmetic mean, will overestimate
the transcript number for a typical single cell in a
population.
Fig. 6 shows how the mRNA copy number of three hypothetical
genes may vary over time. Genes 1 and 3 have correlated expression,
while gene 2 is uncorrelated with genes 1 and 3. Gene expression
in mammalian cells occurs in bursts [4,5,35]. The total amount
of a speciﬁc mRNA at a given time point in a cell is determined by
the frequency and duration of the bursts and the mRNA degradation.
The observed distributions with lognormal features are consistent
with the burst theory [5]. The nature and consequences of
the stochastic behavior of gene expression have recently gained
new insights but still this is a relatively new ﬁeld of research [4].
Maybe the most critical step in single-cell gene expression proﬁling
using RT-qPCR is to ensure that expression data between single
cells are comparable. How can we compensate for differences
in accessible mRNA content among individual cells, and how do
we separate between experimental and biological variability in
the mRNA pool of individual cells? Today, single cell data are reported
as absolute mRNA/cDNA copies or as relative quantities
without any normalization except to the number of cells (one).
The use of RNA spike will only compensate for technical errors
downstream of cell lysis. The expression of all genes changes over
time, consequently the use of reference genes is not valid to normalize
single cell data (Fig. 6). In addition, most actively expressed
genes will at some time point be expressed below the detection
limit of the RT-qPCR assay due to the stochastic behavior of gene
expression. A strong correlation between the geometric mean
and the frequency of cells with observed transcripts of a speciﬁc
gene is usually observed. For example, single b-cells only occasionally
lack Ins2 (high expressed) expression, while Chgb (medium exFig.
5. Single cell gene expression data. (A) Cq-values of Chgb and Ins2 from 37
single cells. Cells not expressing, respectively, gene was arbitrarily given a value of
40. (B) Distribution of Ins2 expression shown in linear-scale. Cq-values of Ins2 were
converted to number of cDNA molecules using standard curves of known
concentrations. (C) Distribution of Ins2 expression shown in log10-scale. The
geometric and arithmetic mean are shown in both linear- and log10-scale. Data have
been published before [19].
A. Ståhlberg, M. Bengtsson / Methods 50 (2010) 282–288 287
pressed) is only detectable in about 50% of the cells. (Fig. 5A) [19].
Another confounding effect when analyzing single cell data is that
compared cells usually are not exactly in the same phase in the cell
cycle. In many experimentally setups, cell cycle synchronization is
impossible or even unwanted. Exactly how the cell cycle affects
gene expression is not known. One interesting approach to test
for normalizing single cell data would be to inject a speciﬁc mRNA
molecule with known concentration into the cell using glass capillaries,
followed by standard cell collection and measurement.
The fact that a minority of cells constitute the major part of the
total number of transcripts for a given gene in a cell population results
in that observed correlations at cell population level actually
may occur in different cells (Fig. 1C–D). In a previous study [6], we
found that Ins1, Ins2 and Actb were all up-regulated in b-cells at cell
population level when the glucose concentration was increased.
Interestingly, single-cell correlations between the genes revealed
that Ins1 and Ins2 expression increased in the same cells (high correlation),
while Actb was up-regulated in other cells; i.e. Actb
showed no single-cell correlation with Ins1/Ins2. The positive correlation
between Ins1 and Ins2 and that they share almost identical
promoter regions [36] suggest that they have synchronized transcriptional
bursts (Fig. 6) [3,4,7]. Consequently, correlation at single
cell level can be used as a tool to identify genes that truly
share regulatory mechanisms, not just being affected by the same
stimulus.
4. Concluding remarks
Single-cell gene expression proﬁling is a powerful tool to better
understand molecular mechanisms. However, single-cell gene
expression proﬁling is rarely applied, in part due to a lack of understanding
of single-cell biology. The stochasticity of gene expression
makes it difﬁcult to separate technical variation from biologically
relevant cell-to-cell variation. For example, some cells in a larger
population are expected to be expressed below the detection limit,
even for well-optimized RT-qPCR assays. Many laboratories have
access to equipment for single cell collection and real-time PCR
instruments. Due to the very few transcripts being measured, each
step in the process from cell preparation to data analysis need to be
carefully optimized and obviously follow good laboratory practice,
but each step is otherwise identical as for cell population measurement.
When single cell experiments are applied for the ﬁrst time, it
may be worthwhile to analyze a few highly expressed genes to get
started, even if the genes are not of biological interest. One major
advantage using RT-qPCR for single-cell gene expression proﬁling
is that RT-qPCR for cell population measurement is standard procedure
for many laboratories, which limits the cost, time and effort to
perform the experiments. Analyzing individual cells opens up new
doors for molecular biologists and the research ﬁeld of single-cell
biology is growing rapidly.
Acknowledgments
This work was partly supported from the Swedish Society for
Medical Research, Socialstyrelsens Foundation, Sigurd and Elsa
Goljes Minne, Foundation Blanceﬂor Boncompagni-Ludovisi née
Bildt and Wilhelm and Martina Lundgren’s Research Foundation.
References
[1] M.B. Elowitz, A.J. Levine, E.D. Siggia, P.S. Swain, Science 297 (2002) 1183–1186.
[2] W.J. Blake, M. Kaern, C.R. Cantor, J.J. Collins, Nature 422 (2003) 633–637.
[3] J.M. Raser, E.K. O’Shea, Science 309 (2005) 2010–2013.
[4] A. Raj, A. van Oudenaarden, Cell 135 (2008) 216–226.
[5] A. Raj, C.S. Peskin, D. Tranchina, D.Y. Vargas, S. Tyagi, PLoS Biol. 4 (2006) e309.
[6] M. Bengtsson, A. Ståhlberg, P. Rorsman, M. Kubista, Genome Res. 15 (2005)
1388–1392.
[7] D. Volfson, J. Marciniak, W.J. Blake, N. Ostroff, L.S. Tsimring, J. Hasty, Nature
439 (2006) 861–864.
[8] E.S. Kawasaki, Ann. NY Acad. Sci. (2004) 92–100.
[9] T. Nolan, R.E. Hands, S.A. Bustin, Nat. Protoc. 1 (2006) 1559–1582.
[10] M.W. Pfafﬂ, Nucleic Acids Res. 29 (2001) 2002–2007.
[11] S.A. Bustin, J. Mol. Endocrinol. 29 (2002) 23–29.
[12] L. Warren, D. Bryder, I.L. Weissman, S.R. Quake, Proc. Natl. Acad. Sci. USA 103
(2006) 17807–17812.
[13] J. Gründemann, F. Schlaudraff, O. Haeckel, B. Liss, Nucleic Acids Res. 36 (2008)
e38.
[14] A. Diercks, H. Kostner, A. Ozinsky, PLoS One 4 (2009) e6326.
[15] V. Espina, J.D. Wulfkuhle, V.S. Calvert, A. van Meter, W. Zhou, G. Coukos, D.H.
Geho, E.F. Petricoin III, L.A. Liotta, Nat. Protoc. 1 (2006) 586–603.
[16] K.M. Keays, G.P. Owens, A.M. Ritchie, D.H. Gilden, M.P. Burgoon, J. Immunol.
Methods 302 (2005) 90–98.
[17] G.I. Murray, Acta Histochem. 109 (2007) 171–176.
[18] A. Ståhlberg, M. Bengtsson, M. Hemberg, H. Semb, Clin. Chem. 55 (2009) 2162–
2170.
[19] M. Bengtsson, M. Hemberg, P. Rorsman, A. Ståhlberg, BMC Mol. Biol. 9 (2008)
63.
[20] B. Liss, O. Franz, S. Sewing, R. Bruns, H. Neuhoff, J. Roeper, EMBO J. 20 (2001)
5715–5724.
[21] B. Liss, Nucleic Acids Res. 30 (2002) e89.
[22] A. Ståhlberg, J. Håkansson, X. Xian, H. Semb, M. Kubista, Clin. Chem. 50 (2004)
509–515.
[23] A. Ståhlberg, M. Kubista, M.W. Pfafﬂ, Clin. Chem. 50 (2004) 1678–1680.
[24] A. Dafforn, P. Chen, G. Deng, M. Herrler, D. Iglehart, S. Koritala, S. Lato, S.
Pillarisetty, R. Purohit, M. Wang, S. Wang, N. Kurn, Biotechnology 37 (2004) 3–
7.
[25] E. Wang, L.D. Miller, G.A. Ohnmacht, E.T. Liu, F.M. Marincola, Nat. Biotech. 18
(2000) 457–459.
[26] L.N. Sellner, R.J. Coelen, J.S. Mackenzie, Nucleic Acids Res. 20 (1992) 1487–
1490.
[27] D.P. Chandler, C.A. Wagnon, H. Bolton Jr., Appl. Environ. Microbiol. 64 (1998)
669–677.
[28] O. Suslov, D.A. Steindler, Nucleic Acids Res. 33 (2005) e181.
[29] N.N. Iscove, M. Barbara, M. Gu, M. Gibson, C. Modi, N. Winegarden, Nat.
Biotech. 20 (2002) 940–943.
[30] G. Brady, N.N. Iscove, Methods Enzymol. 225 (1993) 611–623.
[31] A. Peixoto, M. Monteiro, B. Rocha, H. Veiga-Fernandes, Genome Res. 14 (2004)
1938–1947.
[32] S.A. Bustin, V. Benes, J.A. Garson, J. Hellemans, J. Huggett, M. Kubista, R.
Mueller, T. Nolan, M.W. Pfafﬂ, G.L. Shipley, J. Vandesompele, C.T. Wittwer, Clin.
Chem. 55 (2009) 611–622.
[33] S. Fleige, M.W. Pfafﬂ, Mol. Aspects Med. 27 (2006) 26–39.
[34] S. Fleige, V. Walf, S. Huch, C. Prgomet, J. Sehm, M.W. Pfafﬂ, Biotechnol. Lett. 28
(2006) 601–613.
[35] J.R. Chubb, T. Trcek, S.M. Shenoy, R.H. Singer, Curr. Biol. 16 (2006) 1018–1025.
[36] S.J. Giddings, L.R. Carnaghi, J. Biol. Chem. 263 (1988) 3845–3848.
Fig. 6. mRNA expression over time. The transcription process is known to occur in
bursts. The total amount of a speciﬁc mRNA is determined by the frequency and
duration of transcriptional bursts and mRNA degradation. The green (Gene 1) and
black (Gene 3) curves illustrate correlated mRNAs, while the red curve (Gene 3)
represents an uncorrelated mRNA (for illustration only).
288 A. Ståhlberg, M. Bengtsson / Methods 50 (2010) 282–288