Meaningful quantification of mRNA using real-time PCR
Reverse transcription (RT) followed by the fluorescence-based real-time polymerase chain reaction (PCR) is the most sensitive method for the detection of low abundance mRNA and has become widely used for the quantification of steady-state mRNA lev-els. Assays are easy to perform, capable of high throughput, and can combine high sensitivity with reliable specificity. They address many of the difficulties inherent in conventional PCR, but careful experimental design and validation remain essential for accurate quantitative measurements of mRNA levels.
Three characteristics have made real-time techniques the most popular method for characterising or confirming gene expression patterns and comparing mRNA levels in different sample populations: conceptual simplicity, practical ease and high through-put 1. Real-time techniques integrate the amplification and analysis steps of the PCR reaction by monitoring the amount of DNA produced during each PCR cycle. The technology is still relatively novel, with the first practical real-time fluorescence-based quantitative PCR method, the 5’-nuclease assay, developed in 1996 2. Neverthe-less, their sensitivity, specificity and wide dynamic range have made real-time PCR-based assays the gold standard for the detection and quantification of both DNA 3 and RNA 4. There are currently five chemistries that use fluorescent dyes to monitor PCR reactions in real-time during the PCR. The simplest method uses fluorescent dyes, e.g. SYBR Green, that bind specifically to double-stranded-DNA. The 5’-nuclease assay (Taqman), Molecular Beacons, adjacent linear oligoprobes and Scorpions rely on the hybridisation of fluorescence-labelled oligonucleotides to the correct amplicon.
The problem is that, while quantification per se is simple, the interpretation and re-porting of that quantitative data is not. In principle there are two ways to quantitate mRNA levels: either relative to some internal control or absolute per cell number, to-tal RNA or unit mass of tissue. Relative quantification is usually described as ade-quate for most purposes, with the extra conditions and treatments required by absolute quantification requiring restricting its usefulness 5. In practice, I believe that this is not so.
Relative quantification determines the changes in steady-state mRNA levels of a gene across multiple samples and expresses it relative to the levels of a co-amplified inter-nal control mRNA. Therefore, relative quantification does not require standards with known concentrations and the reference can be any nucleic acid, as long as its concen-tration and length of amplicon are known. During the RT-PCR assay, target Ct are compared directly to reference Ct and results are expressed as ratios of the target-specific signal to the internal reference. This produces a corrected relative value for the target-specific mRNA product that can be compared between samples for an esti-mate of the relative expression of target mRNA in those samples, e.g. c-myc mRNA levels relative to GAPDH mRNA levels are x times more, or less in colorectal tumour cells compared with paired normal colonic biopsies. The crucial flaw with this ap-proach is that the most common reference mRNAs are transcribed from reference genes, whose expression may be regulated and whose levels usually vary significantly with treatment or between individuals (see below).
However, relative quantification can generate useful and biologically relevant infor-mation when used appropriately. For example, an experiment designed to compare the degree of activated T-cell infiltration in colorectal cancers might usefully measure In-terleukin-2 receptor (IL-2R) mRNA levels relative to those of a T-cell-specific marker (CD3 or CD8) 6.
This is based on the use of an external standard dilution series with predetermined known concentrations and results are expressed as actual numbers of mRNA or DNA molecules. Its most obvious application is in quantifying tumour cells or infectious particles like viruses or bacteria in body fluids, but it in the absence of suitable reference genes that can act as normalisers, it is also usefully applied to quantitate changes in mRNA levels. The accuracy of absolute quantification depends entirely on the accuracy of the standards. In general, standard curves are highly reproducible and allow the generation of specific and reproducible results. Nevertheless, it is difficult to calibrate these standards so that they permit universal, absolute quantification and re-sults may not be comparable with those obtained using different probe/primer sets for the same markers, and most certainly will be different from results obtained using dif-ferent techniques 7. Furthermore, external standards cannot detect or compensate for inhibitors that may be present in the samples. For this it is necessary to spike the sam-ple with an internal control, ideally a synthetic amplicon, that either can be co-amplified with the target using a different fluorophore to detect its amplification product or is amplified in a separate reaction.
Standard curves can be constructed using from in vitro T7-transcribed sense RNA transcripts or single-stranded sense-strand oligodeoxynucleotides. The standard curve is generated by performing serial dilutions of the standard and assaying each dilution together with positive and negative control reactions. To maximise accuracy, the dilu-tions are made over the range of copy numbers that include the amount of target mRNA expected in the experimental RNA samples. The Ct value is inversely propor-tional to the log of the initial copy number 8. Therefore, a standard curve is generated by plotting the Ct values against the logarithm of the initial copy numbers. The copy numbers of experimental RNAs can be calculated after real-time amplification from the linear regression of that standard curve, with the y-intercept giving the sensitivity the slope the amplification efficiency.
Biologically meaningful reporting of target mRNA copy numbers requires accurate and relevant normalisation to some standard. Clearly, the quality of quantitative data cannot be better than the quality of the normaliser and any variation in the normaliser will obscure real changes and produce artifactual changes. RT-PCR-specific errors in the quantification of mRNA transcripts are easily compounded by any variation in the amount of starting material between samples. This is especially relevant when the samples have been obtained from different individuals or when comparing samples from different tissues, and will result in the misinterpretation of the expression pro-files of the target genes. Consequently, the question of appropriate standardisation arises 9 and constitutes one of the most critical aspects of experimental design.
The accepted method for minimising these errors and correcting for sample-to-sample variation is to amplify, simultaneously with the target, a cellular RNA specified by a housekeeping gene, such as glyceraldehyde-3-phosphate dehydrogenase (GAPDH), that serves as an internal reference against which other RNA values can be normalised 10. However, it is now clear that there is no single RNA with a constant expression level among different tissues of an organism 9;11, and that their use as internal calibra-tors is inappropriate 12. Ribosomal RNA (rRNA), which makes up the bulk of a total RNA sample, has been proposed as an alternative normaliser 13;14, and its levels re-main fairly constant during serum-stimulation of tissue culture cells 15;16. However, reservations remain concerning its expression levels, transcription by a different RNA polymerase and possible imbalances in rRNA and mRNA fractions between different samples 17.
For some applications, e.g. when preparing RNA from tissue culture cells, from nu-cleated cells or from microdissected cells, it is straightforward to normalise mRNA levels to copies per cell number. Alternatively, normalisation to total cellular RNA appears to be relatively reliable 4 since there is little variability in total cell RNA con-tent within the same tissue type between individuals 12. However, there are obvious limitations to this approach as total RNA levels may be increased in highly proliferat-ing cells and this will affect the accuracy of any comparison of absolute copy numbers between normal and tumour cells. Nevertheless, combining absolute quantification using a target-specific standard curve with reporting of copy numbers relative to total RNA is as close as it is possible to achieve precise and biologically meaningful quan-tification and allow straightforward comparison between different laboratories.
Quantification and quality assessment of RNA
Normalisation to total RNA content requires accurate quantification of the RNA and the most commonly used method, absorbance measurement at OD260 in a spectrophotometer, may not be sufficiently accurate for this purpose. In addition, it can waste significant amounts of precious RNA. The RiboGreen RNA quantification assay re-lies on a proprietary dye that exhibits significant fluorescence enhancement on bind-ing to nucleic acids and can be detected in a spectrofluorometer, fluorescence mi-croplate reader or filter-based fluorometer 18. However, it does not provide qualitative information. The Agilent Bioanalyser/Biorad Experion systems are based on microfluidics tech-nology, and a LabChip cassette consists of a series of interconnected microchannels and reservoirs embedded in a palm-sized chip architecture. Migrations through the channels are monitored and controlled using the benchtop Agilent 2100 Bioanalyser or Biorad Experion instruments. The software calcu-lates the ratio of 28S:18S rRNA in the sample to provide a simultaneous qualitative assessment for each sample. Results can be viewed as gel-like images, electro-pherogrammes or in tabular formats. However, these methods report the quality of the rRNA, rather than the mRNA. Hence we advocate the use of a 3':5' assay to measure the integrity of the mRNA.
An important consideration when using total RNA for normalisation is the lack of in-ternal control for RT or PCR inhibitors. All quantitative methods assume that the RNA targets are reverse transcribed and subsequently amplified with similar effi-ciency. The risk with normalisation against external standards is that a proportion of the samples might contain some inhibitor that significantly reduces the efficiency of the RT-PCR reaction, resulting in inaccurate quantification. Therefore, it is necessary to develop universal internal standards that can be added to the RNA preparation to monitor the efficiency of reverse transcription reaction.
Optimisation and consistency are as critical for obtaining reproducible results using real-time RT-PCR as they are for conventional methods. However, real-time RT-PCR assays are significantly less variable than any conventional RT-PCR protocol which is subject to significant error 19. Where measured, the coefficient of variation for Ct data has been shown to be very low at less than 2% for the Taqman 2;20 and as low as 0.4% for the Lightcycler 21, which is significantly better than the 14% reported for conven-tional RT-PCR 22. Reproducibility is influenced by parameters such as distribution statistics (Poisson’s law) 23, and Ct data are less reproducible when working with very low copy numbers due to the stochastic effects in the quantification of small numbers of target molecules 24. Particle distribution statistics predict that it will require a much higher number of replicates to differentiate 5 from 10 copies of RNA than for the dif-ferentiation of 5,000 from 10,000 copies. Of course, this emphasises the importance of repetitive testing in clinical samples and one of the strengths of these assays is the ease with which it is possible to determine multiple Ct values for every sample, which encourages replicate determinations of the same sample and permits the application of statistical analyses to the quantification procedure.
Normal biopsies contain a range of different cell types, a problem exacerbated in het-erogeneous tumour samples that also include normal and inflammatory cells as well as diversely evolved cell populations. In addition, normal cells adjacent to a tumour may be phenotypically normal, but genotypically abnormal or exhibit altered gene ex-pression profiles due to their proximity to the tumour 25. Hence expression profiling of such biopsies provides a composite of the whole population, and cannot identify ex-pression limited to subpopulations or sections of the biopsy. This may result in the masking or loss of the expression profile of a specific cell type or it may be ascribed to and dismissed as illegitimate transcription 26 because of the bulk of the surrounding cells. Nevertheless, most in vivo RNA extractions and subsequent analyses are carried out from such biopsies with little regard for the different cell types contained within that sample. The critical importance of microdissection for maximising the accuracy of quantitative gene profiling of individual cells or cell populations is demonstrated by the significant differences that have been detected in the gene expression profiles of microdissected and bulk tissue samples 27;28.
Laser capture microdissection (LCM) 29 is the most powerful technique for extracting pure subpopulations of cells from heterogeneous in vivo cell samples for detailed mo-lecular analysis 30. Quantitative isolation of RNA from such small samples is possible 31;32, and mRNA can be isolated 33 and expression levels accurate and reproducible quantified from archival paraffin-embedded tissue specimens 34-36, even after immu-nohistochemical staining 37. Therefore, it may now be possible to use a single sample for immunocytochemistry, in situ hybridisation as well as quantitative RT-PCR or mi-croarray analysis.
The RT step is the least monitored, yet crucially important part of any protocol design to generate sensitive and accurate quantification of steady-state mRNA levels. First, the amount of cDNA produced by the reverse transcriptase must accurately represent RNA input amounts. Therefore, the dynamic range, sensitivity and specificity of the enzyme are prime consideration for a successful RT-PCR assay. Second, RT reactions are usually carried out between 40oC and 50oC and at these low temperatures there can be problems with the relative nonspecificity of the RT reaction 38. This results in non-specific priming by both forward and reverse primers and is a particular problem with very low concentrations of starting template. Here such non-specific side reac-tions can outcompete the desired reaction and, if the genuine target concentration is sufficiently low, results in the complete inhibition of specific product amplification. Third, G/C- or secondary structure-rich mRNA templates 39 also pose problems for reverse transcriptases transcribing at 40-50oC. This is because such templates can cause the enzyme to stop, dissociate from the RNA template, or skip over looped-out regions of RNA.
The comparative easy with which real-time RT-PCR assays can generate quantitative data has created the impression that this technique is simple, speedy, sensitive and specific and that data can be subjected to objective statistical analysis. In reality, sig-nificant doubts remain about the relevance and comparability of real-time RT-PCR data and statistical analyses of the numerical data may obscure and allowing misinter-pretation of the actual results. The widespread use of housekeeping genes to report relative changes of mRNA levels has made it impossible to compare results from dif-ferent laboratories, and must be discouraged. Instead, normalisation procedures be-tween samples must be properly validated to achieve biologically relevant interpreta-tion of data. Today’s main challenge is to develop experimental protocols and designs that are rigorously controlled and allow meaningful global comparisons.
1. Orlando, C., Pinzani, P., and Pazzagli, M., Developments in quantitative PCR, Clin. Chem. Lab Med. 36, 255, 1998
2. Heid, C. A. et al., Real time quantitative PCR, Genome Res. 6, 986, 1996
3. Mackay, I. M., Arden, K. E., and Nitsche, A., Real-time PCR in virology, Nu-cleic Acids Res. 30, 1292, 2002
4. Bustin, S. A., Absolute quantification of mRNA using real-time reverse tran-scription polymerase chain reaction assays, J Mol. Endocrinol. 25, 169, 2000
5. Freeman, W. M., Walker, S. J., and Vrana, K. E., Quantitative RT-PCR, pit-falls and potential, Biotechniques 26, 112, 1999
6. Bustin, S. A. et al., Expression of HLA class II in colorectal cancer, evidence for enhanced immunogenicity of microsatellite-instability-positive tumours, Tumour Biol. 22, 294, 2001
7. Schuurman, R. et al., Multicenter comparison of three commercial methods for quantification of human immunodeficiency virus type 1 RNA in plasma, J Clin. Microbiol. 34, 3016, 1996
8. Higuchi, R. et al., Kinetic PCR analysis, real-time monitoring of DNA ampli-fication reactions, Biotechnology (NY) 11, 1026, 1993
9. Thellin, O. et al., Housekeeping genes as internal standards, use and limits, J Biotechnol. 75, 291, 1999
10. Karge, W. H., Schaefer, E. J., and Ordovas, J. M., Quantification of mRNA by polymerase chain reaction (PCR) using an internal standard and a nonradioac-tive detection method, Methods Mol. Biol. 110, 43, 1998
11. Haberhausen, G. et al., Comparative study of different standardization con-cepts in quantitative competitive reverse transcription-PCR assays, J. Clin. Microbiol. 36, 628, 1998
12. Bustin, S. A., Quantification of mRNA using real-time RT-PCR, trends and problems, J Mol Endocrinol 28, (in press), 2002
13. Bhatia, P. et al., Comparison of glyceraldehyde-3-phosphate dehydrogenase and 28S- ribosomal RNA gene expression as RNA loading controls for north-ern blot analysis of cell lines of varying malignant potential, Anal. Biochem. 216, 223, 1994
14. Zhong, H., Simons, J. W., Direct comparison of GAPDH, beta-actin, cyclo-philin, and 28S rRNA as internal standards for quantifying RNA levels under hypoxia, Biochem. Biophys. Res. Commun. 259, 523, 1999
15. Schmittgen, T. D., Zakrajsek, B. A., Effect of experimental treatment on housekeeping gene expression, validation by real-time, quantitative RT-PCR, J. Biochem. Biophys. Methods 46, 69, 2000
16. Goidin, D. et al., Ribosomal 18S RNA Prevails over Glyceraldehyde-3-Phosphate Dehydrogenase and beta-Actin Genes as Internal Standard for Quantitative Comparison of mRNA Levels in Invasive and Noninvasive Hu-man Melanoma Cell Subpopulations, Anal. Biochem. 295, 17, 2001
17. Solanas, M., Moral, R., and Escrich, E., Unsuitability of using ribosomal RNA as loading control for Northern blot analyses related to the imbalance between messenger and ribosomal RNA content in rat mammary tumors, Anal. Bio-chem. 288, 99, 2001
18. Jones, L. J. et al., RNA quantitation by fluorescence-based solution assay, Ri-boGreen reagent characterization, Anal. Biochem. 265, 368, 1998
19. Souaze, F. et al., Quantitative RT-PCR, limits and accuracy, Biotechniques 21, 280, 1996
20. Gerard, C. J. et al., Improved quantitation of minimal residual disease in mul-tiple myeloma using real-time polymerase chain reaction and plasmid-DNA complementarity determining region III standards, Cancer Research 58, 3957, 1998
21. Wittwer, C. T. et al., The LightCycler, a microvolume multisample fluorime-ter with rapid temperature control, Biotechniques 22, 176, 1997
22. Zhang, J. et al., Two variants of quantitative reverse transcriptase PCR used to show differential expression of alpha-, beta- and gamma-fibrinogen genes in rat liver lobes, Biochem. J. 321, 769, 1997
23. de Vries, T. J. et al., Reproducibility of detection of tyrosinase and MART-1 transcripts in the peripheral blood of melanoma patients, a quality control study using real-time quantitative RT-PCR, Br. J. Cancer 80, 883, 1999
24. Peccoud, J., Jacob, C., Theoretical uncertainty of measurements using quanti-tative polymerase chain reaction, Biophys. J. 71, 101, 1996
25. Deng, G. et al., Loss of heterozygosity in normal tissue adjacent to breast car-cinomas, Science 274, 2057, 1996
26. Chelly, J. et al., Illegitimate transcription, transcription of any gene in any cell type, Proc. Natl. Acad. Sci. U. S. A. 86, 2617, 1989
27. Fink, L. et al., cDNA array hybridization after laser-assisted microdissection from nonneoplastic tissue, Am. J. Pathol. 160, 81, 2002
28. Sugiyama, Y. et al., Microdissection is essential for gene expression profiling of clinically resected cancer tissues, Am. J. Clin. Pathol. 117, 109, 2002
29. Emmert-Buck, M. R. et al., Laser capture microdissection, Science 274, 998, 1996
30. Walch, A. et al., Tissue microdissection techniques in quantitative genome and gene expression analyses, Histochem. Cell Biol. 115, 269, 2001
31. Bohle, R. M. et al., Cell type-specific mRNA quantitation in non-neoplastic tissues after laser-assisted cell picking, Pathobiology 68, 191, 2000
32. Dolter, K. E., Braman, J. C., Small-sample total RNA purification, laser cap-ture microdissection and cultured cell applications, Biotechniques 30, 1358, 2001
33. Bock, O., Kreipe, H., and Lehmann, U., One-step extraction of rna from ar-chival biopsies, Anal. Biochem. 295, 116, 2001
34. Godfrey, T. E. et al., Quantitative mRNA expression analysis from formalin-fixed, paraffin- embedded tissues using 5' nuclease quantitative reverse tran-scription- polymerase chain reaction, J. Mol. Diagn. 2, 84, 2000
35. Specht, K. et al., Quantitative gene expression analysis in microdissected ar-chival formalin-fixed and paraffin-embedded tumor tissue, Am. J. Pathol. 158, 419, 2001
36. Cohen, C. D. et al., Laser microdissection and gene expression analysis on formaldehyde-fixed archival tissue, Kidney Int. 61, 125, 2002
37. Fink, L. et al., Immunostaining and laser-assisted cell picking for mRNA analysis, Lab. Invest. 80, 327, 2000
38. Raja, S. et al., Increased sensitivity of one-tube, quantitative RT-PCR, Bio-techniques 29, 702, 2000
39. Malboeuf, C. M. et al., Thermal effects on reverse transcription, improvement of accuracy and processivity in cDNA synthesis, Biotechniques 30, 1074, 2001