You are on page 1of 6

Genes and Immunity (2005) 6, 279–284

& 2005 Nature Publishing Group All rights reserved 1466-4879/05 $30.00
www.nature.com/gene

REVIEW
Real-time RT-PCR normalisation; strategies
and considerations
J Huggett1,2, K Dheda1,2,3, S Bustin4 and A Zumla1,2
1
Centre for Infectious Diseases and International Health, University College London, London, UK; 2Royal Free Medical School, London,
UK; 3Department of Thoracic and HIV Medicine, Royal Free Hospital, London, UK; 4Centre for Academic Surgery, Barts and
the London, Queen Mary’s School of Medicine and Dentistry, London, UK

Real-time RT-PCR has become a common technique, no longer limited to specialist core facilities. It is in many cases the only
method for measuring mRNA levels of vivo low copy number targets of interest for which alternative assays either do not exist
or lack the required sensitivity. Benefits of this procedure over conventional methods for measuring RNA include its sensitivity,
large dynamic range, the potential for high throughout as well as accurate quantification. To achieve this, however, appropriate
normalisation strategies are required to control for experimental error introduced during the multistage process required to
extract and process the RNA. There are many strategies that can be chosen; these include normalisation to sample size, total
RNA and the popular practice of measuring an internal reference or housekeeping gene. However, these methods are
frequently applied without appropriate validation. In this review we discuss the relative merits of different normalisation
strategies and suggest a method of validation that will enable the measurement of biologically meaningful results.
Genes and Immunity (2005) 6, 279–284. doi:10.1038/sj.gene.6364190
Published online 7 April 2005

Keywords: qPCR; real-time RT-PCR; housekeeping genes; normalisation; reference gene

Introduction Consequently, it is important that an accurate method
of normalisation is chosen to control for this error.
Real-time reverse transcription PCR (real-time RT-PCR) Unfortunately, normalisation remains one of real-time
is an established technique for quantifying mRNA in RT-PCRs most difficult problems.5
biological samples. Benefits of this procedure over Several strategies have been proposed for normalising
conventional methods for measuring RNA include its real-time RT-PCR data. These range from ensuring that a
sensitivity, large dynamic range, and the potential for similar sample size is chosen to using an internal
high throughout as well as accurate quantification. Its housekeeping or reference gene (Table 1). These ap-
enhanced specificity is particularly useful for immuno- proaches are not mutually exclusive and can be
logical research, which frequently involves analysis of incorporated into a protocol at many stages (Figure 1).
proteins derived from different splice variants of the Here we discuss the respective advantages and dis-
original transcript.1 Furthermore, many of the key advantages of each technique.
proteins (eg cytokines and transcription factors) are
found in such low abundance that real-time RT-PCR
quantification of their mRNAs represents the only Normalisation; sample size
technique sensitive enough to measure reliably their Ensuring a similar sample size is obtained, by sampling
expression in vivo.2,3 similar tissue volume or weight, is the first stage of
Although real-time RT-PCR is widely used to quanti- reducing experimental error. This may appear to be
tate biologically relevant changes in mRNA levels, there straightforward, but experimental sample groups of
remain a number of problems associated with its use. similar size are often not representative. It can be
These include the inherent variability of RNA, variability difficult to ensure that different samples contain the
of extraction protocols that may copurify inhibitors, and same cellular material. A good example is blood, which
different reverse transcription and PCR efficiencies.4 is relatively easy to sample and ensure similar volumes
are compared. However, this can be misleading, as is
illustrated when sampling a similar volume of blood
Correspondence: Dr J Huggett, Centre for Infectious Diseases and from HIV þ ve patients (Figure 2). Patients with HIV that
International Health, University College London, Windeyer Institute of
Medical Research, 46 Cleveland St, London W1T 4JF, UK.
have less advanced immunosupression (CD4 counts
E-mail: j.huggett@ucl.ac.uk X200 cells/ml) will yield a higher amount RNA than
Received 1 November 2004; revised 20 December 2004; accepted 21 patients with CD4 counts p200 cells/ml. This is simply
December 2004; published online 7 April 2005 because there are fewer cells per millilitre of blood in the

integrity (depending on Assumes no variation in rRNA/ technique used) mRNA ratio Genomic DNA Give an idea of the cellular May vary in copy number per Rarely used. input RNA for the reverse transcription reaction should particularly when cultured as a monolayer. Cells can be be similar. Sample Total RNA extracted 150 Extract RNA Ensure similar sample size p = 0. not sufficient on its own. Also measured by Resolution of assay is defined work well with rRNA as no real time RT-PCR by the error of the reference polyA tail is present.4. cell. While Frequently overlooked is the concomitant need for ensuring a similar sample size is important it clearly is interpreting the quality of the RNA (Figure 3).6 If the two HIV also primarily measures ribosomal RNA (rRNA). which should be considered for a good normalisation strategy. Real-time RT-PCR normalisation J Huggett et al 280 Table 1 Comparison of the actual amount of RNA used in different reverse transcription reactions with the respective amount of HuPO cDNA measured by real time RT-PCR Normalisation strategy Pros Cons Note Similar sample size/tissue Relatively easy Sample size/tissue volume may Simple first step to reduce volume be difficult to estimate and/or experimental error may not be biological representative Total RNA Ensures similar reverse Does not control for error Requires a good method of transcriptase input. RNA. Difficult to extract with optically or by real time PCR RNA Reference genes ribosomal Internal control that is subject to Must be validated using the Oligo dt priming of RNA for RNAs (rRNA) the same conditions as the RNA same experimental samples. Also Resolution of assay is defined primed with oligo dt for reverse measured by real time RT-PCR by the error of the reference transcription gene Alien molecules Internal control that is subject to Must be identified and cloned Requires more characterisation most of the conditions as the or synthesised. which Genes and Immunity . but not all.035 RNA RNA yield ng / µl 100 Generate cDNA Ensure similar RNA concentration cDNA Measure cDNA by 50 Real time PCR Measure internal reference Result Figure 1 Processes required to generate a real time RT-PCR result. discussed in Figure 2. of mRNAs RNAs (mRNA) the same conditions as the same experimental samples. Normal- ising a sample against total RNA has the drawback of not Normalisation. It would be misleading to directly compare Dheda et al. Total RNA extracted using PAXtubes (Qiagen) as for latter group. it can also be difficult to estimate sample size (cell number) because groups. is not extracted from commercially the biological variability of a the within the cells reference gene There is good correlation between the RNA concentration used and the real time PCR estimation of the different amounts of HuPO cDNA (using omniscript reverse transcriptase). and is likely to confuse the experimental findings. Usually in gene high abundance Reference genes messenger Internal control that is subject to Must be validated using the Most. 0 < 200 > 200 Black arrows indicate points.5 these groups based on sample volume alone. Is without of interest. Normalising to total RNA RNA prior to reverse transcription. are to be assessed then cells will often clump up or have different morphologies. If RNA quality is not good then the measurement can be effected (Figure 4 and Bustin and Nolan4). There are several methods for quantifying treated chemically or enzymatically to assist counting. Can be measured sample size. CD4 count (cells/ml) Figure 2 Total RNA yields from two different groups of HIV patients. When dealing with in vitro cell culture. reverse transcription will not of interest. Unlike the RNA and to be made available mRNA of interest. May introduced at the reverse assessing quality and quantity provide information on the transcription or PCR stages. this will undoubtedly effect gene expression (molecular probes) and the LabChip (Agilent 2100). arguably the most accurate being ribogreen however. contain polyA tails and can be mRNA of interest. RNA quantification controlling for variation inherent in the reverse tran- It is essential to quantitate accurately and quality assess scription7 or PCR reactions.

it is assumed In 1985. glyceraldehyde-3-phosphate dehydrogenase (GAPDH). because this can vary with reverse transcrip- tase type7 it must be validated. tumor cells often extensively and are unsuitable for normalisation pur- have a variable haplotype and actively replicating poses due to large measurement error.5 available. so contain more sets of genetic information inappropriate as a reference when comparing different when compared to nonproliferating cells. However. However. The procedure is simplified as both the gene of interest and the reference gene are measured using real-time RT-PCR.12 In 1984. Real-time RT-PCR normalisation J Huggett et al 281 a b dures are usually not designed to purify DNA. reference genes Normalising to a reference gene is a simple and popular method for internally controlling for error in real-time RT-PCR. so the extraction rate may vary between different samples. the advent of 2 real-time PCR placed the emphasis on quantitative 1 change. this variability is likely to be similar in the study and Genes and Immunity . however. More recently there have been a number of reports that fold and so unless very fine measurements are required demonstrate that the classic reference genes can vary would not cause a problem. As early as 1975. Reference genes can also control for different input RNA amounts used in the reverse transcription step (Figure 5). in mouse tissue by Northern blot. hypoxanthine-guanine phosphoribosyl transferase (HPRT) and 18S ribosomal RNA. which have been collectively called housekeeping genes and benefits from the fact that all the steps required to obtain the final PCR measurement are controlled for. 18S rRNA was reported to increase in expression with cytomegalovirus infection.10 This appears to be an ideal method as it mention of a validation process.11 Another major problem the overall study findings will not be affected because with using this strategy is that RNA extraction proce. Their use was accep- 9 8 table for these non/semiquantitative techniques where a 7 qualitative change was being measured.15 Despite these and other observations there are count- Normalisation.5. Figure 3 Example of good quality RNA assessed by (a) agarose gel What is even more surprising is that these ‘classic’ (5 ml RNA extract). (b and c) by the Agilent Bioanalyser (1 ml RNA reference genes were demonstrated to be regulated over extract) using the RNA LabChip. they contained very different amounts of mRNA. with DNA yields often being low. and should have resulted in a re-evaluation of 18S 28S 0 -1 the use of these reference genes. The notion that these does not require reverse transcription for detection by RNAs require validation is also not new. Total RNA extracted using a decade before the real-time RT-PCR was made PAXtubes (Qiagen) as for Dheda et al. This was not done and 19 24 29 34 39 44 49 54 59 64 69 studies continued to use arbitrarily chosen ‘classic’ Time (seconds) reference genes for this purpose. Rnase protection assays and conventional RT-PCR assays. The most commonly used reference genes include b-actin. As the majority of transcription occurred at a similar rate in different rat procedures measure protein-coding messenger RNAs tissues. In 1989.9 and in 1987 b-actin mRNA was reported to be differen- tially expressed in different leukaemia patient tumour samples. However. genomic DNA less examples of published work that have used a Targeting genomic DNA has also been suggested for particular reference gene for normalisation without any normalisation. This strategy targets RNAs encoded by genes. an assumption that has been reported can be elevated in certain parts of the central nervous system14 wrong. Normalisation.8. They are 3 historical carryovers and were used as references for c10 many years in Northern blots. HPRT was reported to be constitutively that rRNA : mRNA ratio does not change between expressed at low levels in most human tissues but was groups. Piechaczyk et al13 reported that while GAPDH makes up B80% of the fraction. Cells that are proliferating are replicating and Dautry16 reported that probes to b-actin were their DNA. This was Fluorescence 6 because these RNAs are expressed at relatively high 5 levels in all cells and made ideal positive controls if the 4 3 gene of interest was switched off. (mRNA) (comprising 2–5% of the fraction).17–23 Some have bacteria can contain up to 8  more copies of certain argued that the above-mentioned factors are academic as loci than nonreplicating cells. eukaryotic organisms. this difference will usually be p2. Barbu real-time PCR.

Even if the chosen omniscript reverse transcriptase). (c) Different results obtained when 15 ng of good and poor quality RNA are used for reverse transcription are illustrated. The different reference genes are then measured by real-time RT-PCR 1 and variation in the cycle threshold (Ct) or crossing point (Cp) assessed. unless the variability is measure a 2 log change in a gene of interest. Total RNA extracted using PAXtubes (Qiagen) as for Dheda et al. that is.genomebiology.24 However.26 BestKeeper and has serious implications for studies that have used also selects the least variable gene using the geometric unvalidated reference genes.5 2.28 freely available on request. There is good gene can provide the resolution of the assay in question. Norm-Finder. a reference gene RNA that has an error of 1 log may not be ideal.5 to reference gene variability assessment. Gnorm allows the most appropriate reference Of particular note is the findings of Bas et al17 and gene to be chosen by using the geometric mean of the Tricarico et al.gene- Authors must be able to demonstrate that the reference quantification. Real-time RT-PCR normalisation J Huggett et al 282 a 12 b 11 10 9 Fluorescence 8 7 6 5 4 3 2 1 0 -1 19 24 29 34 39 44 49 54 59 64 69 Time (seconds) c RNA quality Duplicate cDNA measurement Duplicate cDNA measurement of HuPO (~copies/reaction) of IL-4 (~copies/reaction) Good 43050 60 43730 50 Poor 248480 410 168600 590 Figure 4 Example of poor quality RNA assessed by the Agilent Bioanalyser using the RNA LabChip (a. copy number. It is no longer acceptable to mean but uses raw data27 instead of data converted to choose blindly any reference gene for normalisation. the measured reference gene variability represents the 0 cumulative error of the entire process. gene is variable it may not matter as long as intergroup difference being measured is greater than the reference gene variation. freely available (http://www. Figure 5 Comparison of the actual amount of RNA used in different reverse transcription reactions with the respective amount Once this variation is defined the chosen reference of HuPO cDNA measured by real time RT-PCR. correlation between the RNA concentration used and the real time Choosing the accepted level of variability will depend on PCR estimation of the different amounts of HuPO cDNA (using the degree of resolution required. The strategy we previously Log relative expression 2 presented5 uses total RNA to normalise the sample prior 1.5 error and does not take into account the RT step.26 This software is reference gene is chosen it can result in altered findings. A third program gene of choice is suitable for the experiment in question. not only Genes and Immunity . This is particularly worrying ples are published by Vandesompele et al. the innate 250 ng 100 ng 50 ng 10 ng variation of the reference gene under investigation and the experimental error associated with the technique. defined this argument simply does not hold.zip. but is sufficient to control groups. b). As RNA normalisation can incorporate 0.5 log actual concentration Considerations when using a reference gene log estimated concentration Reference gene validation exercises are also subject to the problem of normalisation. While There are a number of programs based on the excel genes like GAPDH have been found to be appropriate platform that allow the assessment of multiple reference for certain experimental situations25 they are often not. it is also available at http://www.com/ This occurs when the reference gene is regulated by the 2002/3/7/research/0034/) and the underlying princi- experimental conditions. that is. genes.21 who demonstrated that if the wrong expression of the candidate cDNA.de/BestKeeper-1. (HuPO and IL-4 reactions are performed as described5).

64: 356–360. 50: experimental error that affects the RNA of interest. Andersson M. J Immunol 2004. 172: 6938–6943. furthermore. cule may also not be feasible for small laboratories 10 Talaat AM. Biotechniques 2004. Brinster RL. the rat. real-time RT-PCR data. not mutually exclusive and we recommend attempting 16 Barbu V. 4 Bustin SA. Increased expression of assay remains dependent on the variability of the chosen mRNA encoding interleukin (IL)-4 and its splice variant IL- reference genes. J Bacteriol 2001. that the common criticisms. measurements are to be made. Clin Chem 2004. Johnston wanting to perform limited amounts of real-time RT. but at this lymphocytes. Cohen CD. Huggett JF. Microbiology 2004. 150 (Part 6): 1609–1627.29 but until they are made available stimulates host cell RNA synthesis. Expression of human HPRT in the central nervous system of Conclusion transgenic mice. Differential pattern of oncogene and beta-actin expression in leukaemic cells from normalising real-time RT-PCR data. Defining this is essential as it can generate false results as discussed above. Aseffa A et al. Abebe M. Blanchard JM. Kubista M. multiple genes are chosen the resolution of the particular 3 Fletcher HA. extracted and similar quantity used for the reverse 17 Bas A. glyceraldehyde-3-phosphate-dehydrogenase for normaliza- Spiking the sample with a defined amount of an tion in real-time quantitative reverse transcriptase-polymerase unique or artificial RNA may arguably be the most chain reaction analysis of gene expression in human T accurate method for internal normalisation. ensuring good quality RNA is rRNA oligonucleotide probe. Howard ST. Furukawa T. incorporated at the extraction stage at a defined 7 Stahlberg A. 4delta2 in cells from contacts of Mycobacterium tuberculosis. 18 Bemeur C. Nielsen AK.4 This 5 Dheda K. Nucleic Acids Res 1993. it will not suffer from the inherent 8 Spanakis E. include sets of artificial molecules that could be used 11 Rocha EP. Plotkin SA. 61: 1105–1112. Nucleic Acids Res 1984. Caskey CT. J Mol species or generated synthetically. The stages required to generate the alien mole. Furthermore. Comparison of reverse concentration. Kidney Int 2003. Henger A. This is a robust method for providing accurate 1 Atamas SP. Marty L et al. discussed in this article. real-time RT-PCR consumables would Mycobacterium tuberculosis. interest or reference gene RNAs. How. commercially and more extensive research into this 13 Piechaczyk M. Ste-Marie L. 357: 211–214. Pfaffl M. Nature 1985. Desjardins P et al. Immunology 2004. Lyons R. Researchers should ensure expression analysis in microdissected human renal biopsies. Scand J Immunol 2004. Healthy individuals that possible to measure multiple reference genes due to control a latent infection with Mycobacterium tuberculosis limited sample availability and cost. Nucleic Acids Res 2002. Br J Haematol 1987. 1678–1680. unlike the RNA of 9 Hansen MC. even if express high levels of Th1 cytokines and the IL-4 antagonist IL-4delta2. Hammer K. Forsberg G. for different organisms. However. Nucleic Acids Res 1989. Nolan T. These strategies are AML patients. 112: 669–673. Rook G. Post-transcrip- method is performed. This ‘artificial’ RNA molecule 6 Bustin SA. they stitutive transcripts as controls. J Virol 1975. are not extracted from within the cells. Schlondorff D. Northern blot normalization with a 28S to match sample size. as these RNAs are spiked into the extractant. 12: 6951–6963. Neurosci Lett 2004. Brennand J. Molin S. but they must be used with caution. Life Sci 1997. SA. 29: 23–39. 30: e104. Owiafe P. in the absence of in vitro stimulation. Hammarstrom ML. 65: 83–86. Johnson MA. 15: An alternative internal control strategy uses a molecule 155–166. Validation of housekeeping genes for normalizing RNA represents a potentially excellent method for normalising expression in real-time PCR. 14 Stout JT. Bustin SA. Jeffries D et al. 59: 566–573. Hale W. The key to 19 Schmid H. Decreased beta-actin time reference genes represent a strategy that is simple to mRNA expression in hyperglycemic focal cerebral ischemia in use and can control for every stage of the real-time PCR. Zumla A. Chen HY. This might present a Changes in rRNA levels during stress invalidates results from problem in certain situations (eg different histological mRNA blotting: fluorescence in situ rRNA hybridization samples when studying fibrosis) so would also require permits renormalization for estimation of cellular mRNA validation. ase gene expression in rat tissues. that is artificially incorporated into the sample. beta-actin and (reference gene or alien molecule) also be measured. the extent by which they are effected by the experimental conditions. 21: 3809–3819. Generation of such standards 12 Tanaka S. Ideally. Hammarstrom S. Human cytomegalovirus has been proposed. Faldt R. Quantification of mRNA using real-time reverse can be cloned and in vitro transcribed from another transcription PCR (RT-PCR): trends and problems. 37: 112–119. levels. radiographic data on gene expression using common con- ever. Kilstrup M. Vandesompele et al also advocate the use of multiple References reference genes rather than relying on a single RNA transcript. 15: 297–304. it will be subjected to almost all the transcriptases in gene expression analysis. Problems related to the interpretation of auto- biological fluctuations that effect reference genes. artificial molecule transcription polymerase chain reaction. transcription reaction and finally an internal control Utility of the housekeeping genes 18S rRNA. Validation of endogenous controls for gene to demonstrate that it is valid. Pitfalls of quantitative real-time reverse- Normalisation. The replication-related organization of bacterial genomes. 183: 4747–4751. that is. groups. Irrgang S. As it can then be Endocrinol 2002. Genes and Immunity . J Biomol Tech 2004. 317: 250–252. There are a number of methods that can be used for 15 Blomberg J. Alternative splice variants of cytokines: making a normalisation and is consequently favourable if fine list. Real-time RT-PCR normalisation J Huggett et al 283 measures the variation but also ranks the potential have been addressed before presenting real-time RT-PCR reference genes by how much they differ between study data. Genomic DNA standards for gene expression profiling in PCR. it is not always 2 Demissie A. artificial RNA normalisation will tional regulation of glyceraldehyde-3-phosphate-dehydrogen- remain an unvalidated theoretical ideal. Garner H. 17: 7115. normalisation when using this technology is to be able Kretzler M. Dautry F.

64: 5245–5250. geometric averaging of multiple internal control genes. Am J Respir Crit Care sets. Zumla A. The use of housekeeping genes tion of stable housekeeping genes. Clin Chem 2004. Tichopad A. 270: 41–49. perfused rat liver. differentially regulated (HKG) as an internal control for the detection of gene target genes and sample integrity: BestKeeper—Excel-based expression by quantitative real-time RT-PCR. 30: 1108–1114. 292: 264–266. Pattyn F et al. Real-time RT-PCR normalisation J Huggett et al 284 20 Schulz WA. Sistare F. 29 Cronin M. Increase of beta-actin mRNA upon hypotonic perfusion of cardiac myocytes in vitro. Hallbrucker C. Bianchi S et al. Anal Biochem 2002. Ghosh K. Huggett JF. 509–515. Normalization of real- tissue-specificity in the expression of cytochrome P450 3A time quantitative reverse transcription-PCR data: a model- mRNA. 49: 211–216. expression. 169: 1167–1168. Wolbold R et al. De Preter K. Orntoft TF. 25 Winer J. FEBS Lett 1991. Jung CK. Drug Metab Dispos 2002. Sies H. 26 Vandesompele J. 23 Koch I. Kim LU. Interindividual variability and 28 Andersen CL. Cancer Res 2004. Development and O’Connell C. Folia Biol (Praha) tool using pair-wise correlations. Anal Biochem 1999. Weil R. applied to bladder and colon cancer data respiratory syncytial virus bronchiolitis. Med 2004. Neuvians TP. Quackenbush J. Quantitative real-time normalization of real-time quantitative RT-PCR data by reverse transcription polymerase chain reaction: normaliza. Jensen JL. Accurate 21 Tricarico C. for human tissue biopsies. Type 2 cytokines in for normalization. 3: RESEARCH0034. Biotechnol Lett 2004. Haussinger polymerase chain reaction for monitoring gene expression in D. 27 Pfaffl MW. based variance estimation approach to identify genes suited 24 Dheda K. Vilker V. 309: 293–300. Pinzani P. Universal RNA reference materials for gene validation of real-time quantitative reverse transcriptase. Prgomet C. Eickelmann P. Haskovec C. Williams PM. 26: 2003. Shackel I. Determina- 22 Ullmannova V. 50: 1464–1471. tion to rRNA or single housekeeping genes is inappropriate Genome Biol 2002. Genes and Immunity .