Relative expression software tool (REST

RT-PCR使用

© 2002 Oxford University PressNucleic Acids Research, 2002, Vol. 30, No. 9 e36

Relative expression software tool (REST©) for group-wise comparison and statistical analysis of relative expression results in real-time PCR

Michael W. Pfaffl*, Graham W. Horgan1 and Leo Dempfle2

Institute of Physiology, FML-Weihenstephan, Center of Life and Food Sciences, Technical University of Munich,

Germany, 1Biomathematics and Statistics Scotland, Rowett Research Institute, Bucksburn, Aberdeen AB21 9SB, UK and 2Department of Animal Science, Center of Life and Food Sciences, Technical University of Munich, Germany

Received January 21, 2002; Revised and Accepted March 4, 2002

ABSTRACT

Real-time reverse transcription followed by polymerasechain reaction (RT–PCR) is the most suitablemethod for the detection and quantification of mRNA. Itoffers high sensitivity, good reproducibility and a widequantification range. Today, relative expression isincreasingly used, where the expression of a targetgene is standardised by a non-regulated referencegene. Several mathematical algorithms have beendeveloped to compute an expression ratio, based onreal-time PCR efficiency and the crossing point devi-ation of an unknown sample versus a control. But allpublished equations and available models for thecalculation of relative expression ratio allow only forthe determination of a single transcription differencebetween one control and one sample. Therefore anew software tool was established, named REST©(relative expression software tool), which comparestwo groups, with up to 16 data points in a sampleand 16 in a control group, for reference and up tofour target genes. The mathematical model used isbased on the PCR efficiencies and the mean crossingpoint deviation between the sample and control group.Subsequently, the expression ratio results of the fourinvestigated transcripts are tested for significanceby a randomisation test. Herein, development andapplication of REST© is explained and the useful-ness of relative expression in real-time PCR usingREST© is discussed. The latest software version ofREST© and examples for the correct use can bedownloaded at http://www.wzw.tum.de/gene-quantifi-cation/.

INTRODUCTION

Reverse transcription (RT) followed by polymerase chain reaction(PCR) is a powerful tool for the detection and quantification ofmRNA. Nowadays real-time RT–PCR is widely and increasingly

used, because of its high sensitivity, good reproducibility andwide quantification range (1,2). It is the most sensitive methodfor the detection and quantification of gene expression levels,in particular for low abundance mRNA (1,2), in tissues withlow concentrations of mRNA (e.g. bone marrow, fatty tissues),from limited tissue samples (e.g. biopsies, single cells) (3,4)and to elucidate small changes in mRNA expression levels(1,2,5). However, it is a very complex technique with varioussubstantial problems associated with its true sensitivity,reproducibility and specificity and, as a fully quantitativemethodology, it suffers from the problems inherent in real-timeRT–PCR. Generally, two quantification strategies can beperformed: an absolute and a relative quantification. In absolutequantification the absolute mRNA copy number per vial orcapillary is determined by comparison with appropriateexternal calibration curves (2). An absolute quantificationmakes it easier to compare expression data between differentdays and laboratories, because the calibration curve is a non-changing solid and reliable basis. The relative expression isbased on the expression ratio of a target gene versus areference gene and is adequate for most purposes to investigatephysiological changes in gene expression levels. Trends can bebetter explained by relative quantification, but the results arestrongly dependent on the reference gene and the normalisationprocedure used. Some mathematical models have already beendeveloped to calculate the relative expression ratios of singlesamples (6–8; http://wendang.chazidian.com/pebiodocs/04303859.pdf), with or without efficiency correction.Equation 1 shows the most convenient mathematical model,which includes an efficiency correction for real-time PCRefficiency of the individual transcripts (6).

Ratio = (Etarget)?CPtarget(control – sample)/(Eref)?CPref(control – sample)

1

The relative expression ratio of a target gene is computed,based on its real-time PCR efficiencies (E) and the crossingpoint (CP) difference (?) of an unknown sample versus acontrol (?CPcontrol – sample). In mathematical models the targetgene expression is normalised by a non-regulated referencegene expression, e.g. derived from housekeeping genes,glyceraldehyde-3-phosphate dehydrogenase (GAPDH), albumin,actins, tubulins, cyclophilin, 18S ribosomal RNA (rRNA) or

*To whom correspondence should be addressed at present address: Institute of Physiology, FML-Weihenstephan, Life Science Zentrum Weihenstephan, Technische Universitaet Muenchen, Weihenstephaner Berg 3, D-85350 Freising–Weihenstephan, Germany. Tel: +49 8161 71 3511; Fax: +49 8161 71 4204; Email: pfaffl@wzw.tum.de

RT-PCR使用

e36 Nucleic Acids Research, 2002, Vol. 30, No. 9

28S rRNA (9–11). But all published equations and availablemodels for the calculation of relative expression ratios allowfor the determination of only a single transcription differencebetween one control and one sample (n = 1), e.g. given in anDNA array experiment, and not for a group-wise comparisonfor more samples (n > 2), given in an experimental trial.

Therefore, a new software tool was established, namedREST© (relative expression software tool), which comparestwo groups, with up to 16 data points in the sample groupversus 16 data points in the control group, and tests the groupdifferences for significance with a newly developed random-isation test. Nevertheless, the successful application of real-timeRT–PCR and REST© depends on a clear understanding of thepractical problems. Therefore, a clear experimental design,application and validation of the applied real-time RT–PCRremains essential for accurate and fully quantitative measure-ment of mRNA transcripts. This paper explains the develop-ment of REST© application, discusses the technical aspectsinvolved in an experimental trial and illustrates the usefulnessof relative expression in real-time RT–PCR using REST©.MATERIALS AND METHODS

Animal experiment, total RNA extraction and reverse transcription

Total RNA extraction was performed from rat liver asdescribed previously (12). Adult rats were either fed withphysiological zinc concentrations (control group, 58 p.p.m. Zn,n = 7) or suffered 22–29 days under zinc depletion (samplegroup, 2 p.p.m. Zn, n = 6) (W.Windisch, manuscriptsubmitted). Isolated total RNA integrity was electrophoreticallyverified by ethidium bromide staining and by an averageoptical density (OD) OD260/OD280 nm absorption ratio of 1.97(range 1.78–2.09). Either 330, 1000 or 3000 ng total RNA wasreverse transcribed with 100 U Superscript II Plus RNaseH– reverse trancriptase (Gibco Life Technologies, Gaithersburg,MD) in a volume of 40 µl, using 100 µM random hexamerprimers (Pharmacia Biotech, Uppsala, Sweden) according tothe manufacturers’ instructions. Therefore, concentrations of8.25, 25 or 75 ng cDNA (=reverse transcribed total RNA)perµl were achieved.Optimisation of RT–PCR

Highly purified salt-free primer for target gene metallothionein(MT) (forward primer, CTC CTG CAA GAA GAG CTG CT;reverse primer, TCA GGC GCA GCA GCT GCA CTT) andfor reference gene GAPDH (forward primer, GTC TTC ACTACC ATG GAG AAG G; reverse primer, TCA TGG ATGACC TTG GCC AG) were generated commercially (MWGBiotech, Ebersberg, Germany). The MT primer set is able toamplify the transcripts of MT isoform 1 and MT isoform2mRNA. Conditions for real-time PCRs were optimised in agradient cycler (Mastercycler Gradient, Eppendorf, Germany)with regard to Taq DNA polymerase (Roche MolecularBiochemicals, Basel, Switzerland), forward and reverseprimers, MgCl2 concentrations (Roche Molecular Biochemicals)and various annealing temperatures (54–66°C). RT–PCRamplification products were separated on a 4% high resolutionNuSieve agarose (FMC Bio Products, Rockland, ME) gel electro-phoresis and analysed with the Image Master system (Pharmacia

PAGE 2 OF 10

Biotech). Optimised conditions were transferred to thefollowing LightCycler real-time PCR protocol.LightCycler real-time PCR

For determination of test and software variations all applica-tions of different total cDNA input were performed in triplets(MT 1–3 and GAPDH 1–3). Real-time PCR mastermix wasprepared as follows (to the indicated end-concentration): 6.4 µlwater, 1.2 µl MgClµl reverse primer (0.4 2 (4 mM), 0.2 µl forward primer (0.4 µM),0.2 µM) and 1 µl LightCycler–Fast StartDNA Master SYBR Green I (Roche Molecular Biochemicals).Nine microlitres of master-mix was filled in the glass capillariesand a 1 µl volume of cDNA (either 8.25, 25 or 75 ng) wasadded as PCR template. Capillaries were closed, centrifugedand placed into a cycling rotor. A four-step experimental runprotocol was used: (i) denaturation program (10 min at 95°C);(ii) amplification and quantification program repeated 40 times(15 s at 95°C; 10 s at 60°C for MT or 10 s at 58°C for GAPDH;20 s at 72°C; 5 s at 86°C for MT or 5 s at 84°C for GAPDHwith a single fluorescence measurement); (iii) melting curveprogram (60–99°C with a heating rate of 0.1°C per s and acontinuous fluorescence measurement); (iv) cooling programdown to 40°C. To improve SYBR Green I quantification a hightemperature fluorescence measurement point at the end of thefourth segment was performed (13). It melts the unspecificPCR products below the chosen temperature, e.g. primerdimers, eliminates the non-specific fluorescence signal andensures accurate quantification of the desired GAPDH and MTreal-time RT–PCR product, respectively. For the describedmathematical model it is necessary to determine the CPs foreach transcript. The CP is defined as the point at which thefluorescence rises appreciably above the backgroundfluorescence. In this study the Second Derivate MaximumMethod was performed for CP determination, using LightCyclerSoftware 3.5 (Roche Molecular Biochemicals).Statistics

For statistical evaluations of the determined CP variations andcalculated relative expression variations (Tables 1–3), datawere analysed for significant differences by ANOVA usingapproximate tests (Sigma Stat for Windows Software®,Version 2.0; Jandel Corporation).

DEVELOPMENT OF REST©

Our goal was the development of a software tool that allowsfor a relative quantification between groups, and a subsequenttest for significance of the derived results with a suitable statisticalmodel. Further, the software must be able to run on a widelyavailable platform, which can be used worldwide on differentcomputer systems. For that reason it was programmed to run inMicrosoft Excel® (Microsoft Corporation). In what follows,the four pages of REST© and the statistical model, a Pair WiseFixed Reallocation Randomisation Test© are described indetail.

Page 1—Introduction

On the introduction page the basic settings are made for theREST© application (Fig. 1). Up to four genes and one referencegene can be labelled. Different background colours in thespreadsheets and the print command are shown and described.

RT-PCR使用

PAGE 3 OF 10

Nucleic Acids Research, 2002, Vol. 30, No. 9 e36

Figure 1. Page 1—Introduction.

Pink cells indicate cells for data input, blue cells indicate dataoutput, grey cells are used for calculation purposes and outputof the CP variation, the red box will start the RandomisationTest itself and the printer icon indicates ‘print this page’. Further,the relative expression equation is given with direct links to thedata input section on page CP input + randomisation test.Page 2—PCR efficiency

The PCR efficiency calculation is facultative and notobligatory for the user (Fig. 2). To generate the data basis forthe determination of PCR efficiency of each transcript, it isrecommended to use various dilutions in triplets of a pool of allavailable cDNAs. This ensures the best estimation of the PCRefficiency. If the user wants to determine the real-time PCRefficiencies, an import via copy and paste of cDNA startingconcentrations in dilution row and the corresponding CPvalues measured by the real-time PCR machine is possible.Depending on the real-time PCR platform used, CP values canbe determined either by the Threshold Cycles = Fit PointMethod (all platforms) or Second Derivate Maximum Method(only LightCycler). Up to three CPs can be inserted in the table(run 1–3) per cDNA starting concentration and REST© deter-mines the slope with a logarithmic algorithm, as publishedearlier (1,6,14), as well as an indication of the linearity of thislogarithmic alignment using Pearson’s correlation coefficient.The real-time PCR efficiencies were calculated from the slope,according to the established equation E = 10[–1/slope] (1,14). E isin the range from 1 (minimum value) to 2 (theoreticalmaximum and optimum). If no real-time PCR efficiencies arecalculated here, REST© assumes an optimal efficiency ofE=2.0 on the following pages and further procedures.

Page 3—CP input + randomisation test

On the top the calculated PCR efficiencies or alternativelyE=2.0 are shown and will be the basis for the calculation andrandomisation test (Fig. 3). Up to 16 CP data per group(control or sample group) can be inserted for the referencegene and up to four target genes (input section of page 3 is notshown). On clicking the red box, the Randomisation Testapplication window will appear. Here the range of the data setmust be defined, for the control group and sample group, bytouching the last cell containing the last CP data point (on thebottom right of the pink input window). Further, the number ofrandomisations can be chosen and the randomisation test willbe started on clicking OK. It is recommended that at least 2000randomisations be performed (see next section statisticalmodel).

The numeric results of the randomisation test are given in theRandomisation Data Output box: the concerned Genes, the CPmean of control group (Control Means), the CP mean ofsample group (Sample Means), the Expression Ratios normalisedby the reference gene, the corresponding p-Values, theExpression Ratios-nn not normalised by the reference gene,the corresponding p-Values-nn and the number of Random-isations performed. To simplify matters for the user, additionalanswer sentences were created according to the calculatedresults. They are divided into the Randomisation Test Results(normalised by reference gene expression) and RandomisationTest Results (not normalised by reference gene expression).The sentences tell the user if the sample group in comparisonwith the control group is up- or down-regulated and illustratesthe factor of regulation and if this up- or down-regulation issignificantly different or not. For up-regulation, the factor of

regulation is equal to the given value in the Randomisation

RT-PCR使用

e36 Nucleic Acids Research, 2002, Vol. 30, No. 9

PAGE 4 OF 10

Figure 2. Page 2—PCR efficiency.

Data Output box. In the case of down-regulation, the regulationfactor is illustrated as a reciprocal value (1/expression ratio or1/expression ratio-nn, respectively).Page 4—Ratio + variation output

The mean CP of the genes, the CP variations and the coefficient ofvariation (CV) are calculated and shown to illustrate thereproducibility and variation of the investigated group datasubsets (Fig. 4).

Statistical model: Pair Wise Fixed Reallocation Randomisation Test©

Differences in expression between control and treated sampleswere assessed in group means (Fig. 1) for statistical signifi-cance by randomisation tests (15,16; http://www.bioss.ac.uk/smart/unix/mrandt/slides/frames.htm). Permutation or random-isation tests are a useful alternative to more standard para-metric tests for analysing experimental data. They have theadvantage of making no distributional assumptions about thedata, while remaining as powerful as more standard tests (16).The rationale for the randomisation test is that standard para-metric tests (such as analysis of variance or t-tests) depend onassumptions, such as normality of distributions, whose validityis doubtful. In our case, where the quantities of interest arederived from ratios and variances can be high, normal distributionswould not be expected, and it is unclear how a parametric testcould best be constructed. A randomisation test avoids makingany assumptions about distributions, and is instead based onone we know to be true: that treatments were randomly allocated.The test is conducted as follows.

A statistical test is based on the probability of an effect aslarge as that observed occurring under the null hypothesis of notreatment effect. If this hypothesis is true, the values in onetreatment group were just as likely to have occurred in theother group. The randomisation test repeatedly and randomlyreallocates the observed values to the two groups, and notes theapparent effect (expression ratio in our case) each time. Theproportion of these effects which are as great as that actuallyobserved in the experiment gives us the P-value of the test.They calculate P-values by obtaining the proportion ofrandom allocations of the mean observed data to the controland treated sample groups that would give greater indicationsof a treatment effect than that observed. If this is small, thenthere is evidence that the observed treatment effect is notsimply the result of random allocation. Thus, the test makes noassumptions concerning the distribution of measured geneexpression in any hypothesised population—it assumes onlythe random allocation of treatment. In practice, it is impracticalto examine all possible allocations of data to treatment groups,and a random sample is drawn. If 2000 or more samples aretaken, a good estimate of the P-value (SE < 0.005 at P = 0.05)is obtained. In the applied Pair Wise Fixed ReallocationRandomisation Test© for each sample, the CP values for refer-ence and target genes are jointly reallocated to control andsample groups (=pair wise fixed reallocation), and the expressionratios are calculated on the basis of the mean values as

described above. They are deemed to give greater indications

RT-PCR使用

PAGE 5 OF 10

Nucleic Acids Research, 2002, Vol. 30, No. 9 e36

Figure 3. Page 3—CP input + randomisation test.

of a treatment effect than that actually observed if |log R| > |log R0|where R0 is the true expression ratio and R the result of reallo-cation. In the Pair Wise Fixed Reallocation RandomisationTest© a two-sided test was performed. The randomisation testswere carried out using a Microsoft Excel® macro (MicrosoftCorporation) attached to a purpose-built spreadsheet andrunning in the background of REST©.RESULTS

Confirmation of primer specificity

Specificity of RT–PCR products was documented with highresolution gel electrophoresis and resulted in a single productwith the desired length (MT, 106 bp; GAPDH, 197 bp). Inaddition, a LightCycler melting curve analysis was performedwhich resulted in single product-specific melting temperatures:87.4°C (GAPDH) and 89.7°C (MT). No primer primer–dimerformations were generated during the applied 40 real-timePCR amplification cycles.

Real-time PCR amplification efficiencies and variationReal-time PCR efficiencies were calculated from the slopesgiven in LightCycler software (Roche Molecular BiochemicalsLightCycler Software®, Version 3.5). The corresponding real-time PCR efficiency (E) of one cycle in the exponential phase

was calculated according to the equation: E = 10[–1/slope], asdescribed earlier (1,6,14). Investigated transcripts showed real-time PCR efficiency rates for MT (EMT = 1.67) and GAPDH(EGAPDH = 1.88) in the investigated range from 120 pg to 75 ngcDNA input, repeated six times, with high linearity [Pearsoncorrelation coefficient (r) > 0.989].

To mimic different reverse transcription efficiencies and toconfirm precision and reproducibility of real-time PCR, as wellas for REST©, three replicates of real-time RT–PCR at each ofvarious cDNA input concentrations (three times more andthree times less concentrated) were performed and real-timeRT–PCR and REST© variations (CV) were determined. Asshown in Table 1, variations of investigated transcripts arebased on the CP variation and remained stable between 2.43and 10.03% for MT and 1.59 and 12.89% for GAPDH; thelatter showing a dependence on the cDNA input in real-timePCR. CP itself decreased with increasing cDNA input in bothfactors and groups.

Variation and reproducibility of REST©

On the basis of the previously published mathematical model(6), REST© calculates the relative expression ratios on thebasis of group means for target gene MT versus reference geneGAPDH and tests the group ratio results for significance.Normalised and not-normalised expression results were

compared.