How to do successful gene expression analysis using real-time PCR
Stefaan Derveaux, Jo Vandesompele *, Jan Hellemans
Center for Medical Genetics, Ghent University Hospital, Ghent, Belgium
Biogazelle, Ghent, Belgium
a r t i c l e i n f o
Article history:
Accepted 25 November 2009
Available online 5 December 2009
Keywords:
RT-qPCR
Quality control
Experiment design
Assay design
Reference gene validation
a b s t r a c t
Reverse transcription quantitative PCR (RT-qPCR) is considered today as the gold standard for accurate,
sensitive and fast measurement of gene expression. Unfortunately, what many users fail to appreciate
is that numerous critical issues in the workflow need to be addressed before biologically meaningful
and trustworthy conclusions can be drawn. Here, we review the entire workflow from the planning
and preparation phase, over the actual real-time PCR cycling experiments to data-analysis and reporting
steps. This process can be captured with the appropriate acronym PCR: plan/prepare, cycle and report.
The key message is that quality assurance and quality control are essential throughout the entire RT-qPCR
workflow; from living cells, over extraction of nucleic acids, storage, various enzymatic steps such as
DNase treatment, reverse transcription and PCR amplification, to data-analysis and finally reporting.
Ó 2009 Elsevier Inc. All rights reserved.
1. Chance favors the prepared mind
Reverse transcription quantitative PCR (RT-qPCR) distinguishes
itself from other methods available for gene expression in terms of
accuracy, sensitivity, and fast results. Because of this, the technology
has established itself as the golden standard for medium
throughput gene expression analysis. Due to its apparent simplicity,
also inexperienced users can rapidly produce results; however,
care should be taken when performing RT-qPCR as numerous critical
quality issues may arise throughout the entire workflow influencing
the accuracy of the results and the reliability of the
conclusions. Intensive quality control is an important and necessary
part that can be captured with the appropriate acronym
PCR: plan/prepare, cycle and report (Fig. 1). In this first section,
we will illustrate four important preparative steps in the RT-qPCR
workflow, prior to starting the actual real-time PCR quantifications.
Spending careful attention to experiment design, sample
and assay quality control, and selection of proper reference genes
for normalization will significantly increase the chance of successful
results; preparation is everything.
1.1. Experiment design
One of the most neglected points in setting up an RT-qPCR
study is experimental design. Nevertheless, proper set-up of the
experiment saves time, cuts down on reagent cost and increases
the accuracy and precision of the results. Experiment design involves
3 important aspects. First, power analysis should enable
careful assessment of the number of biological samples needed
to draw meaningful and statistically significant results. Often,
researchers are confronted with too few samples to demonstrate
significance for the observed differential expression in their experiment.
Second, the proper run layout strategy should be selected.
Two different experimental set-ups can be followed in an RT-qPCR
study [1]. According to the preferred sample maximization method,
as many samples as possible are analyzed in the same run. This
means that different genes should be analyzed in different runs if
not enough free wells are available to analyze the different genes
in the same run. In contrast, the gene maximization set-up analyzes
multiple genes in the same run, and spreads samples across
runs if required. The latter approach is often used in commercial
kits or in prospective studies. It is important to realize that in a relative
quantification study, the experimenter is usually interested
in comparing the expression level of a particular gene between different
samples. Therefore, the sample maximization method is
highly recommended because it does not suffer from (often underestimated)
technical, run-to-run variation between the samples.
However, irrespective of the set-up, inter-run calibration is required
to correct for possible run-to-run variation whenever not
all samples are or can be analyzed in the same run. For this purpose,
the experimenter needs to analyze so-called inter-run calibrators
(IRC); these are identical samples that are tested in both
runs. By measuring the difference in quantification cycle or normalized
relative quantity between the IRCs in both runs, it is possible
to calculate a correction or calibration factor to remove the
run-to-run difference, and proceed as if all samples were analyzed
in the same run. The qBase quantification model incorporates state
of the art inter-run calibration schemes [1].
1046-2023/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.ymeth.2009.11.001
* Corresponding author. Address: Center for Medical Genetics Ghent, Ghent
University Hospital, De Pintelaan 185, B-9000 Ghent, Belgium. Fax: +32 9 3326549.
E-mail address: Joke.Vandesompele@Ugent.be (J. Vandesompele).
Methods 50 (2010) 227–230
Contents lists available at ScienceDirect
Methods
journal homepage: www.elsevier.com/locate/ymeth
Apart from the theoretical considerations of gene versus sample
maximization there are also practical reasons to think about run
layout. As the cost of microtiter plates is negligible compared to
that of samples and reagents, there is no need to force reactions
in a single plate until it is completely full. Rather, it may be more
appropriate to leave some wells empty in order to have a more logical
or practical pipetting scheme. For example, in a 96-well plate 9
genes can be measured in 10 samples (in this example no replicated
reactions); however, analyzing only 8 genes allows putting
1 sample or gene in a single column or row, respectively. This
not only facilitates run annotation, but also allows easy pipetting
with a multi-channel pipette or dispensing robot and prevents
mistakes during reaction set-up. In line with this, some experiments
are better served with a 384-well instrument for its lower
consumable cost, faster results and improved data quality as technical
run-to-run variation can be more easily avoided.
1.2. Sample extraction and quality assurance
The need for high quality nucleic acids has been recognized for
many years in the microarray field; the cost of such experiments
pushed people to assess the purity and integrity of the input RNA
molecules. In contrast, RT-qPCR is relatively cheap and – more
worrying – almost always result in beautiful sigmoidal amplification
curves, even if the template is highly degraded as is the case
for e.g. formalin fixed paraffin embedded tissues. An uninformed
user is thus easily mislead, and might not appreciate the impact
of RNA quality on the results. A well known phrase in computer
science ‘garbage in, garbage out’ also applies to real-time PCR. As
RT-qPCR performance is affected by the RNA integrity, Fleige and
Pfaffl recommend an RNA quality score (RIN or RQI) higher than
five as good total RNA quality and higher than eight as perfect total
RNA for downstream applications [2]. A study on the impact of
RNA quality on the expression stability of reference genes indicated
that it is inappropriate to compare degraded and intact samples,
necessitating sample quality control prior to RT-qPCR
measurements [3]. Our own recent data indicate that RNA quality
has a profound impact on the results, in terms of the significance of
differential expression, variability of reference genes and classification
performance of a multi-gene signature (Vermeulen et al., submitted
for publication). In addition or as an alternative to the use of
capillary gel electrophoresis methods that assess the integrity of
the ribosomal RNA molecules (as discussed in [2,3]), PCR based
tests are also frequently used to determine mRNA integrity. In
one such a test, the ratio between the 5’ and 3’ end of a universally
expressed gene is measured upon anchored oligo-dT cDNA synthesis,
reflecting integrity of that particular poly-adenylated transcript
[4]. Finally, another PCR based assay is often used in clinical diagnostics
to determine sample purity. By comparing the Cq value of a
known concentration of a spiked DNA or RNA molecule in both a
negative water control and in the sample of unknown quality,
enzymatic inhibition can be determined [5].
Apart from purity and integrity, there is a third important
requirement related to the input RNA material; it should be free
of contaminating DNA. While it is sometimes possible to design
an assay that spans a large intron such that residual DNA is not
co-amplified, this strategy is not recommended. First, it makes
the design process more cumbersome, and secondly, for up to
20% of the human genes, it will not work as these genes are either
single exon genes or have one or more processed pseudogenes
(retropseudogene or intronless copy) in the genome. The most efficient
strategy in our hands is to do a proper DNase treatment, followed
by a careful check for absence of DNA through qPCR analysis
of a DNA target on the crude RNA [6]. When no signal is observed,
the RNA was free of DNA.
A last point of attention with respect to RNA samples is the increased
use of sample pre-amplification protocols. Indeed, many
biological specimens are valuable and often the amount of RNA extracted
is limiting large-scale gene-expression studies. RNA preamplification
methods address this issue by producing micrograms
of cDNA starting from a few nanograms of total RNA. Prior to using
one of the different amplification protocols that are available, there
is only one important criterion that has to be evaluated: fold
changes in expression between two samples should be preserved
before and after amplification. As different targets might be preamplified
with a different efficiency, it is irrelevant to compare
genes before and after pre-amplification. As such, transcript variant
analysis is only possible before pre-amplification.
1.3. Assay design and quality control
Another important preparative step is the design and empirical
validation of a real-time PCR assay. Assays are available from multiple
sources. You can design by yourself using free or commercially
available design software, or you can buy assays from specialized
vendors that provide assays of the shelf or design on demand. A third
alternative is to look for published and experimentally validated assays
in public databases such as rtprimerdb.org [7]. The latest version
contains almost 8000 assays from mainly five organisms
(human, mouse, rat, rice and Arabidopsis thaliana). Irrespective the
Fig. 1. Overview of all quality control tools throughout the entire qPCR workflow.
228 S. Derveaux et al. / Methods 50 (2010) 227–230
source of the primer and probe sequences, it is of great importance to
validate the assay, both in silico and empirically.
People typically perform specificity analysis of an RT-qPCR primer
pair by doing a BLAST or BiSearch query [8]. However, there is
more quality control required than only specificity assessment.
One should also inspect the presence of SNPs in the primer annealing
regions (which may hamper efficient annealing of the primer or
prevent amplification of the variant allele at all), and model the
secondary structure of the amplicon (using e.g. UNAFold software).
It is well known that hairpins overlapping primer annealing sites
significantly hamper efficient annealing and negatively impact
the PCR efficiency [9]. All these tools are integrated in an
automated in silico assay evaluation pipeline available from rtprimerdb.org
(http://www.rtprimerdb.org).
After in silico assay evaluation, the primers need to be empirically
validated by doing an actual RT-qPCR experiment and
inspecting the length using gel electrophoresis (once), and inspecting
the melt curve when using SYBR Green I. In addition, a standard
curve needs to be run in order to estimate the PCR efficiency.
Important to realize is that the more dilution points and the wider
the range (dilution factor), the more precise PCR efficiency can be
determined [1]. Best practice is to use a mixture of representative
samples as input material for the dilution series.
An important aspect of assay design is to select the correct target
sequence, which in many cases is not easy to do, especially for
the large proportion of genes that have alternatively spliced isoforms.
If no prior knowledge is available on the function of these
transcript variants, one may decide to design an assay for a part
of the transcript that is expressed in most or all of the variants.
Dedicated splice variant quantification requires a specific work
flow and its own controls. A straightforward and reliable strategy
was published by Vandenbroucke and colleagues [10]. The key
message of that study is that so-called absolute or equimolar (same
molarity for the various splice variant targets) standard curves are
required for accurate assessment of splice variants, and that the
primers should be the driving force behind the specificity of the
detection (instead of relying on a probe). Absolute or equimolar
standard curves are curves in which the number of molecules is
known; an efficient way to produce them is to purify a PCR product
or synthesize a single stranded long oligonucleotide template, both
diluted in carrier DNA (10–100 ng of E. coli or yeast tRNA).
1.4. Reference gene validation
A final step in the preparation phase is the selection of proper
reference genes for data normalization. Important to realize is that
any gene expression quantification result is finally composed of
two sources of variation; on the one side there is inherent technical
or experimentally induced variation and on the other side, there is
the true biological variation that underlies the phenomenon under
investigation. The very purpose of any normalization strategy is to
remove the technical variation as much as possible, ending up with
the true biological changes. At the 3rd London qPCR Symposium
organized by Professor Stephen Bustin (April 2005), normalization
against 3 or more validated reference genes was considered as the
most appropriate and universally applicable method. While many
algorithms have been reported to date to evaluate candidate reference
genes in terms of expression stability (or suitability as normalizing
gene) [11], the geNorm method [12] was the first
(http://medgen.ugent.be/genorm) and has established itself as
the de facto standard with more than 2000 citations (Google Scholar,
September 2009). A typical pilot experiment for evaluation of
reference gene expression stability measures around 10 candidate
reference genes in 10 representative samples from each tissue or
sample group. The genes are selected such that they represent different
biological pathways and expression abundance levels. The
relative gene expression values are subsequently imported in the
geNorm program and ranked according to their expression stability.
In a subsequent analysis, the software is able to indicate how
many reference genes are optimally required to remove most of
the technical variation (which depends on the expression stability
of the tested genes and on the heterogeneity of the samples under
investigation). Typically, between 3 and 5 genes are required for
accurate normalization. It is clear that any report documenting
small expression changes without validating reference genes is
unacceptable as it has been shown that the use of a single non-validated
reference gene results in a significant bias (ranging from
more than 3-fold in 25% of the results up to 6-fold in 10% of the results)
[12].
Besides selection of stably expressed reference genes in a pilot
experiment with representative samples, it remains important to
assess their expression stability in the final experiment. In the qba-
sePLUS
software, expression stability calculations are automatically
performed if more than one reference gene is used for normalization.
This approach provides the required reference gene quality
control in each experiment.
2. Action and reaction
There are two basic guidelines in PCR set-up to assure successful
reactions: maximize precision through the proper use of calibrated
pipettes or automated liquid handling systems and minimize contamination
by using filter tips and gloves and by keeping amplified
products away from the PCR set-up area. RT and PCR replicates can
be included to evaluate variation in reverse transcription efficiency
and qPCR precision while no-template controls allow verification of
the absence of contamination in your qPCR reactions.
Before setting up the qPCR reaction one needs to decide on how to
monitor the increase in PCR products as the reaction proceeds. One
can choose between intercalating dyes such as SYBR green I or one of
the many probe technologies that are available. Both detection
methods can deliver excellent results and each has its advantages
and disadvantages: probes enable multiplex quantification whereas
intercalating dyes are the most cost-effective choice, especially
when a limited number of samples are quantified.
A last consideration with significant consequences is the volume
of the PCR reaction. In 96- or 384-well plates, PCR reaction
volumes can be easily scaled down to 10 ll and 5 ll, respectively.
This will not only save on reagents cost, but also allows the use of
less material from your precious samples (as only the final concentration
of the template impacts the Cq value). Below the proposed
reaction volumes, one is likely to encounter evaporation problems
and loss of precision. High-throughput systems (now available
from Roche, Biotrove and Fluidigm) enable fast screening of thousands
of reactions in small (sub)microliter volumes. Of note, there
can be a reduction in sensitivity due to the limited amount of sample
that can be added to the reaction. A sample pre-amplification
step might solve this issue.
3. Trust, but verify
3.1. Relative quantification
The polymerase chain reaction is an exponential process whereby
the specifically amplified product ideally doubles each cycle. As
such, the measured Cq value (standard name for threshold cycle or
crossing point value according to the RDML guideless, http://
www.rdml.org [13]) is a logarithmic value that needs to be converted
into a linear relative quantity using the following exponential
function: RQunkn = 2^(Cqcal À Cqunkn) (Cqcal is the Cq value of
the calibrator sample (or reference sample, e.g. untreated control),
S. Derveaux et al. / Methods 50 (2010) 227–230 229
Cqunkn the Cq value of the unknown sample, RQunkn the quantity of
the unknown sample relative to the calibrator, and 2 the base of
the exponential function, denoting 100% PCR efficiency). Pfaffl
was the first to realize that the use of a gene specific PCR efficiency
(when measured accurately) improves the accuracy of the results
[14]. However, in his quantification model, only one reference gene
can be inserted in the equation for normalization. As already indicated,
normalization is a crucial step in the calculation workflow in
which sample related technical variation is cancelled out. We and
others have shown that the use of multiple stably expressed reference
genes results in more accurate data and statistically more
significant results, and allows reliable quantification of small
expression differences [15,12]. The use of efficiency corrected
multiple reference genes is enabled in the universally applicable
qBase quantification model [1], forming the basis of the qbasePLUS
software (http://www.qbaseplus.com) that also comes in a free
version. Such a quantification model not only provides the required
flexibility (in terms of selection of one or more reference genes,
either or not with correction for gene specific PCR efficiency) but
also employs state of the art error propagation rules (providing
confidence measures to the final quantification result) and interrun
calibration schemes (see also higher). The latter makes it possible
to perform large multi-plate experiments in which many
more samples are studied than actually fit in one physical plate
or PCR run. This feature not only enables large multi-centric studies
but also prospective evaluation of patients in the clinic [16].
3.2. Biostatistical analysis
It is beyond the scope of this article to review all statistical tests to
determine significance of a difference in gene expression between 2
or more groups, to identify a diagnostic or prognostic RNA marker
with high confidence, to find correlations between gene expression
patterns or samples, or to identify relevant pathways or new sample
subgroups. However, a few important messages can be conveyed.
First, it is good practice to log transform the final gene expression
results (i.e. the normalized relative quantities), in order to make
the data distribution more symmetrical (as gene expression data is
often log normally distributed [17]). Together with the Central Limit
Theorem, this allows the use of parametric statistical tests and calculations
that rely on a distribution that resembles a normal distribution
(e.g. classic t-test, confidence intervals, Analysis of Variance)
[18]. Secondly, independent biological replicates are required to
draw meaningful and reliable conclusions. The minimum number
depends on the statistical test and on the power one wants to
achieve (e.g. for confidence interval analysis, at least three replicates
are needed, for a non-parametric paired test (Wilcoxon signed-rank
test), at least six pairs are needed). It must be clear that statistics on
repeated measurements (e.g. PCR replicates) are absolutely nonsense,
as only technical variation is measured. Third, the statistical
test should be selected prior to doing the actual experiment, whereby
the choice is based on the question that needs to be addressed, the
number of data points, and the distribution of the data. If in doubt, a
(bio)statistician should be consulted. Fourth, if the data speak for
themselves, don’t interrupt. This means that sometimes, the difference
is so striking or obvious that there is no need to perform a
statistical test.
3.3. Reporting guidelines – the tower of Babel
Having generated high quality data, the next step is to communicate
results to colleagues and collaborators and possibly to
submit for publication. One major limitation today is that the
real-time PCR instruments’ software and third party data analysis
software speak a different language and as such create data files
that are difficult if not impossible to understand by people not
having the same software. To address this issue, the international
Real-time PCR Data Markup Language (RDML) consortium was
founded (http://www.rdml.org) [13] with one major goal, i.e. the
development of an XML-based Real-Time PCR Data Markup Language
(RDML) standard. RDML enables straightforward exchange of qPCR
data and related information between qPCR instruments and third
party data analysis software, between colleagues and collaborators
and between experimenters and journals or public repositories.
RDML has recently become part of the Minimum Information
for Publication of Quantitative Real-Time PCR Experiments (MIQE)
guidelines and accompanying checklist to guarantee inclusion of
key data information when reporting experimental results [19]
(http://www.rdml.org/miqe/). MIQE describes the minimum information
necessary for evaluation of RT-qPCR experiments under the
form of a checklist to accompany the initial submission of a manuscript
to the publisher. By providing all relevant experimental
conditions and assay characteristics, reviewers can assess the
validity of the protocols used. Full disclosure of all reagents, sequences,
and analysis methods is necessary to enable other investigators
to reproduce results. Following these guidelines will
encourage better experimental practice, allowing more reliable
and unequivocal interpretation of quantitative PCR results.
4. Conclusions
Real-time quantitative RT-PCR is a wonderful method for fast,
accurate, sensitive and cost-effective gene expression analysis.
However, the simplicity of the technology itself makes it vulnerable
for abuse in experiments in which the operator does not perform
the required quality control throughout the entire
procedure. In this review, we outlined the different steps in the
work flow and indicated point by point where and how critical issues
can be resolved. Following the advice in this paper, any user
should be able to do (more) successful gene expression profiling
using the RT-qPCR technology.
References
[1] J. Hellemans, G. Mortier, P.A. De, F. Speleman, J. Vandesompele, Genome Biol. 8
(2007) R19.
[2] S. Fleige, M.W. Pfaffl, Mol. Aspects Med. 27 (2006) 126–139.
[3] C.A. Perez-Novo, C. Claeys, F. Speleman, C.P. Van, C. Bachert, J. Vandesompele,
Biotechniques 39 (2005) (pp. 52, 54, 56).
[4] T. Nolan, R.E. Hands, S.A. Bustin, Nat. Protoc. 1 (2006) 1559–1582.
[5] T. Nolan, R.E. Hands, W. Ogunkolade, S.A. Bustin, Anal. Biochem. 351 (2006)
308–310.
[6] J. Vandesompele, P.A. De, F. Speleman, Anal. Biochem. 303 (2002) 95–98.
[7] S. Lefever, J. Vandesompele, F. Speleman, F. Pattyn, Nucleic Acids Res. 37 (2009)
D942–D945.
[8] T. Aranyi, G.E. Tusnady, Methods Mol. Biol. 402 (2007) 385–402.
[9] J. Hoebeeck, L.R. van der, B. Poppe, S.E. De, N. Yigit, K. Claes, R. Zewald, G.J. de
Jong, P.A. De, F. Speleman, J. Lab. Invest. 85 (2005) 24–33.
[10] I.I. Vandenbroucke, J. Vandesompele, A.D. Paepe, L. Messiaen, Nucleic Acids
Res. 29 (2001) E68.
[11] J. Vandesompele, M. Kubista, M.W. Pfaffl, in: J. Logan, K. Edwards, N. Saunders
(Eds.), Real-Time PCR Current Technology and Applications, Caister Academic
Press, Norfolk, 2009.
[12] J. Vandesompele, P.K. De, F. Pattyn, B. Poppe, R.N. Van, P.A. De, F. Speleman,
Genome Biol. 3 (2002).
[13] S. Lefever, J. Hellemans, F. Pattyn, D.R. Przybylski, C. Taylor, R. Geurts, A.
Untergasser, J. Vandesompele, Nucleic Acids Res. 37 (2009) 2065–2069.
[14] M.W. Pfaffl, Nucleic Acids Res. 29 (2001) e45.
[15] J. Hellemans, O. Preobrazhenska, A. Willaert, P. Debeer, P.C. Verdonk, T. Costa,
K. Janssens, B. Menten, R.N. Van, S.J. Vermeulen, R. Savarirayan, H.W. Van, F.
Vanhoenacker, D. Huylebroeck, P.A. De, J.M. Naeyaert, J. Vandesompele, F.
Speleman, K. Verschueren, P.J. Coucke, G.R. Mortier, Nat. Genet. 36 (2004)
1213–1218.
[16] J. Vermeulen, F. Pattyn, P.K. De, L. Vercruysse, S. Derveaux, P. Mestdagh, S.
Lefever, J. Hellemans, F. Speleman, J. Vandesompele, Nucleic Acids Res. (2009).
[17] M. Bengtsson, A. Stahlberg, P. Rorsman, M. Kubista, Genome Res. 15 (2005)
1388–1392.
[18] H. Motulski, Intuitive Biostatistics, Oxford University Press, 1995.
[19] 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.
230 S. Derveaux et al. / Methods 50 (2010) 227–230