You are on page 1of 12

Nucleic Acids Research Advance Access published February 11, 2011

Nucleic Acids Research, 2011, 1–12
doi:10.1093/nar/gkr065

Measurable impact of RNA quality on gene
expression results from quantitative PCR
Joe¨lle Vermeulen1, Katleen De Preter1, Steve Lefever1, Justine Nuytens1,
Fanny De Vloed1, Stefaan Derveaux1,2, Jan Hellemans1,2, Frank Speleman1 and

Downloaded from nar.oxfordjournals.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15, 2011
Jo Vandesompele1,2,*
1
Centre for Medical Genetics, Ghent University Hospital, De Pintelaan 185, B-9000 Ghent, Belgium and
2
Biogazelle, Technologiepark 3, B-9052 Ghent, Belgium

Received August 13, 2010; Revised January 24, 2011; Accepted January 25, 2011

ABSTRACT INTRODUCTION
Compromised RNA quality is suggested to lead to Gene expression quantification plays a central role in a
unreliable results in gene expression studies. wide variety of studies, including biomedical research
Therefore, assessment of RNA integrity and purity with clinical relevance. Among the various methods avail-
is deemed essential prior to including samples in able for gene expression analysis, the reverse transcription
quantitative polymerase chain reaction (RT–qPCR) is the
the analytical pipeline. This may be of particular im-
most rapid, sensitive, accurate and precise method and its
portance when diagnostic, prognostic or thera- use in clinical diagnostic procedures is presently growing
peutic conclusions depend on such analyses. In exponentially (1–5).
this study, the comparative value of six RNA While there is conflicting literature data, it is often sug-
quality parameters was determined using a large gested that RNA integrity and purity are important in
panel of 740 primary tumour samples for which order to obtain reliable results (6–9). RNA degradation
real-time quantitative PCR gene expression results can occur due to inadequate sample handling, prolonged
were available. The tested parameters comprise of storage, suboptimal storage conditions or inter-laboratory
microfluidic capillary electrophoresis based 18S/ shipment of samples (10,11). RNA may be degraded
28S rRNA ratio and RNA Quality Index value, through exposure to heat or UV, or cleavage by RNAse
HPRT1 50 –30 difference in quantification cycle (Cq) enzymes. In addition, the presence of inhibiting compo-
nents such as urea, salts, phenol, heparin or other agents
and HPRT1 30 Cq value based on a 50 /30 ratio
used during sampling or RNA extraction may also com-
mRNA integrity assay, the Cq value of expressed promise with results (12). It would seem, therefore, that a
Alu repeat sequences and a normalization factor rigorous assessment of RNA integrity and purity is essen-
based on the mean expression level of four refer- tial before using RNA samples in downstream applica-
ence genes. Upon establishment of an innovative tions, especially if diagnostic, therapeutic or prognostic
analytical framework to assess impact of RNA conclusions will be drawn. Unfortunately, proper RNA
quality, we observed a measurable impact of RNA quality control is lacking in a substantial number of
quality on the variation of the reference genes, on studies (4). While it is recently listed as a required
the significance of differential expression of prog- element in the Minimum Information for Publication of
nostic marker genes between two cancer patient Quantitative Real-Time PCR Experiments (MIQE) guide-
risk groups, and on risk classification performance lines (13), there remains a great need to explore in detail
the implications of RNA quality on the final results.
using a multigene signature. This study forms the
Various methods have been proposed for the assessment
basis for further rational assessment of reverse of RNA integrity, most often through measurement of
transcription quantitative PCR based results in the size of the ribosomal subunit RNA molecules.
relation to RNA quality. Importantly though, in RT–qPCR analyses, messenger

*To whom correspondence should be addressed. Tel: +32 479 353563; Fax: +32 9 332 6549; Email: joke.vandesompele@ugent.be

The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors.

ß The Author(s) 2011. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/
by-nc/2.5), which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

0.oxfordjournals. (10 ng/ml). nuclease-free water (Sigma) to 0. lacking homology with any other known Alu expression. To assess the purity of the RNA samples. In addition. The assays had cording to three different methods in collaborating an efficiency of 88. A potato nucleic acid sequence of 101 bp. In from 20 ng total RNA from the neuroblastoma tumour samples yielding sufficient cDNA to measure more than principle. This assay aims at on a large panel of RNA samples extracted from neuro. SEM). the higher 1000 target genes (WT-Ovation. NuGEN). these system (Bio-Rad) (Supplementary Data). a DNase treatment was Four reference samples were tested in all runs and used performed using the RQ1 RNase-free DNase according to as inter-run calibrators. with randomly primed whole transcriptome double. software version 3. this the 50 –30 dCq. About 1 ng of each pre-amplification product (15–17). The cycling conditions comprised 10 min PCR-based methods assessing RNA quality might be polymerase activation at 95 C and 40 cycles of 15 s at more relevant given the fact that the targets are also 95 C and of 60 s at 60 C.0. Two qPCR assays target- ing either the 30 end or the 50 start of the low abundant Total RNA was extracted from 6 neuroblastoma cell lines reference gene HPRT1 were developed and validated and 740 fresh frozen neuroblastoma tumour biopsies ac. reverse primer: GCCTCAGCCTCC water and 1 ml of either RNA from the sample to be tested CGAGTAG) and RT–qPCR was performed in a 384-well . In brief. The cycling con- iScript Select reverse transcriptase according to the manu- ditions were comprised of 3 min polymerase activation at facturer’s instructions and subsequently diluted with 95 C and 55 cycles of 15 s at 95 C and 30 s at 60 C.5 ng/ml cDNA (total followed by a dissociation curve analysis from 60 C to RNA equivalents) and stored at 20 C.375 ml forward and (Bio-Rad).4 SEM) and 94. Cq values were compared to the Cq value Downloaded from nar.75 ml 2 SYBR carried out using the iScript Select cDNA Synthesis Kit Green I master mix (Roche).9% (±1. being interrupted if mRNA is Sample preparation fragmented due to degradation.5 ml) (Eurogentec). inhibitory agent (positive control) or Therefore. erence Cq value).75 ml matching and validated in-house (forward primer: CATGGTGAA forward and reverse primer (5 mM). The principle of this qPCR-based assay is based on the fact that anchored oligo-dT primed reverse transcription proceeds from the 30 MATERIAL AND METHODS poly-A tail to the 50 start.85 ml nuclease-free ACCCCGTCTCTA. PCR amplifica. 0. was amplified with specific primers by abundant repeats in the human genome (1 million real-time qPCR using an iQ5 real-time PCR detection copies). pre-amplification to generate single-stranded cDNA Microfluidic capillary electrophoresis. the 30 Cq value in itself was also evaluated as an alternative stranded cDNA synthesis followed by SPIA-based DNA RNA quality parameter. As the 50 Cq was below detection level for linear and isothermal pre-amplification method starts various samples (hence no 50 –30 dCq could be calculated). 2.15 ml fluorescein (1 mM).org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. An RT–qPCR assay was designed SPUD template (5000 molecules/ml).0 laboratories as described in Vermeulen et al. Briefly. 95 C. (14). the more degraded the RNA sample. The HPRT1 30 and 50 Cq values the manufacturer’s instructions (Promega). cultured neuroblastoma cells and subsequently applied along with microfluidic-based capillary electrophoresis The 50 /30 ratio mRNA integrity assay. 2011 In this study. measurement of the integrity of a reference gene mRNA blastoma tumours recently used in a qPCR-based prog. 4-fold dilution Concentration of each RNA sample was measured using series (Supplementary Data). based on 6-point. strument (LC480. it might be more appropriate to directly assess nuclease-free water (negative control to determine the ref- the quality of the mRNA fraction. Bio-Rad) in order to determine a 18S/28S rRNA ratio and an RNA quality SPUD assay. all mRNAs in a given RNA sample (12). qPCR. By random integration during evolution. 10 ng of each total RNA sample was reverse primer (5 mM each). 0. Real-time qPCR was per- the Nanodrop 1000 Spectrophotometer (Thermo formed on oligo-dT primed cDNA in a 384-well plate in- Scientific). (enzyme) inhibitors (9). respectively.6% (±0. using our in silico analysis pipeline (18). 1 ml nuclease-free water and reverse transcribed using anchored oligo-dT primers and 2 ml cDNA (1 ng total RNA equivalents). two qPCR-based assays using minute of the negative control assay and a difference in Cq or amounts of RNA for investigation of mRNA integrity delta-Cq (dCq) > 1 was used as a cut-off designating the or purity were first validated in RNA samples from presence of inhibitors of the SPUD assay.5 ml containing 3. followed by a dissociation run measured using the same technology when doing RT– from 60 C to 95 C for melting curve analysis. Prior to the cDNA synthesis. a index (RQI) (according to the manufacturer’s quantitative PCR-based assay was used to detect PCR instructions). Real-time qPCR amplifications Reverse transcription of RNA and cDNA synthesis was were performed in 7. total RNA isolate was analysed on a High Sensitivity Assessment of RNA purity and integrity Chip (Experion. were determined and the difference in Cq value between A sample pre-amplification method was applied starting both assays was calculated and defined as the 50 –30 dCq. 2 ml amount of mRNA (19). Subsequently. that is considered to be representative of the integrity of nostic multigene signature validation study (14). repeats got embedded in the 30 -UTR of thousands of tion mixtures (15 ml) contained SYBR Green I Master Mix coding genes and can thus be used as a reference for the buffer (7. Alu repeat sequences are the most human sequence. Roche).2 Nucleic Acids Research. 2011 RNA is the target and not the ribosomal RNA transcripts.

lines (13).rdml. The primer sequences and target genes of each erence genes (HMBS. A normalization factor was A multigene expression signature was built using 30 calculated based on the arithmetic mean Cq value of training samples using either the Prediction Analysis of four stably expressed reference genes (HMBS. standard for exchanging quantitative PCR (qPCR) data HPRT1 30 Cq value.(21)]. assay are available in the rdml files together with their The gene-specific variation for each reference gene was corresponding RTPrimerDB ID (18) (http://www calculated as the standard deviation of the log2 trans.oxfordjournals. UBC and Alu) and Multivariate logistic regression analysis was performed validated using our in silico analysis pipeline (18) using SPSS (version 17). Microarrays (PAM) method (22) or the correlation signa- SDHA. tial expression between the two RNA quality groups in function of increasing group size. Data availability Statistical analysis The experimental data are available as Supplementary Assessment of impact of RNA quality on qPCR results Data in RDML (Real-time PCR Data Markup was performed with six individual quality parameters. . In brief. Nucleic Acids Research. one-sided). HPRT1 50 –30 dCq. distribution interval (1.e. sequences (HMBS. com. Alu Cq value and a normalization (25) (http://www. A dilution series consisting of six 10-fold Stage 4 tumours (irrespective of MYCN status) or with serial dilution points. The HR subgroup comprised neuro. subgroup comprised of all other patients. a qPCR Bioconductor MCR estimate (24) and the ROC packages. 2011 RNA sample. one for This agent thus constitutes an useful positive control for each quality group.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. formed expression of the reference gene under study normalized with the exponentiated arithmetic mean Cq value of the three other reference genes. A clear expression variation was calculated) and constitutes a increase in 50 –30 dCq was noticed upon exposure of the measure for the impact of the RNA quality parameter originally high quality RNA samples to an elevated tem- under investigation on the results (reference gene expres. The mean difference The 50 /30 ratio mRNA integrity assay. UBC).05. In principle. Normalization factor.org). 2011 3 plate instrument (LC480. Reference gene Figure 1 shows that addition of heparin results in PCR variation and significance of differential expressions were (enzyme) inhibition leading to higher Cq values compared measured in sample groups with increasing numbers of to the reference Cq value of the negative control either best quality samples or worst quality samples (Figure 1a) and to a difference in Cq or dCq > 1 for starting with 50 samples up to the total number of each dilution point (Figure 1b) (Supplementary Data). molecules.org) according to the MIQE guide- factor based on the arithmetic mean Cq value of four ref. outside this distribution are significantly different (P < 0. perature (Figure 2a) (Supplementary Data). Downloaded from nar. SDHA. Six RNA samples between the best quality and worst quality sample from cultured neuroblastoma cells were artificially groups was calculated as the area between the curves degraded by heat exposure (10 or 20 min at 80 C) and divided by the numbers of group comparisons (i.64*standard deviation). HPRT1. starting from 1 000 000 molecules INSS Stages 2 and 3 tumours with MYCN amplification down to 10 molecules of SPUD template was created and patients younger than 12 months with INSS Stages 2– using 10 ng/ml yeast tRNA as carrier. HPRT1. Roche) using the same condi. HPRT1. the subsequently subjected to 50 /30 ratio mRNA integrity total number of intervals across which the reference gene and microfluidic electrophoresis analyses. SDHA. PCR inhibition of an assay containing 10 000 SPUD ence gene expression variation or significance of differen. Roche) and data quality) and 1 (worst RNA quality) to make them analysed using qbasePLUS 1. The Mann– RESULTS Whitney test was used to measure the significance of dif- ferential expression of a marker gene between the HR and Validation of the methods for RNA purity and integrity non-HR subgroups. Heparin (0. This resulted in two curves.4 [http://www. samples analysed. comparable. RQI. assay was designed for 59 prognostic genes and 5 reference respectively. a structured and universal data including 18S/28S rRNA ratio. UBC) (20). assessment blastoma patients >12 months at diagnosis with INSS SPUD assay. Therefore the six RNA quality (Supplementary Data).rtprimerdb. Language) format. The non-HR was added as inhibitory agent to the individual wells. Real-time qPCR was performed parameter values were first re-scaled between 0 (best RNA in a 384-well plate instrument (LC480.2) was used to train and test the prognostic signature and to Gene expression analysis of prognostic marker genes evaluate its performance by receiver operating character- Gene expression analysis was performed according to a istic (ROC) area under the curve (AUC) analyses using the procedure described elsewhere (14). Total RNA sion variation or significance of differential expression). the more degraded the ture method (23) and tested on the remaining samples.qbaseplus. electropherograms were in accordance with the above A null distribution interval was determined upon 100 described results showing a progressive reduction in size random permutations and calculation of a 90% of the 18S and 28S peaks and an elevation of the base line . the higher the normalization factor is. Values tions as described above.4 U/ml) 4 tumours with MYCN amplification. The R-language for statistical computing (version 2. denoting the difference of the refer.6.

the more degraded the sample. two replicates for each dilution point. 10 min heat exposure. four replicates for each dilution point) to determine the reference Cq value shown as black diamonds. Figure 2. the more degraded the sample. (b) Microfluidic capillary electrophoresis for one representative sample. or with an inhibitory agent (heparin. 20 min heat exposure). 2011 Figure 1. (b) Difference in Cq (dCq) (y-axis) between the Cq values of the control assays performed in the presence of water and the Cq values of the assays containing the inhibitory agent. (a) Dilution series consisting of six 10-fold serial dilution points. The x-axis represents the samples (intact. starting from 1 000 000 molecules down to 10 molecules of SPUD template. and suboptimal RNA quality The SPUD assay was used for the detection of inhibitors contribute to 50 –30 dCq values deviating from zero in the in a large panel of RNA samples extracted from fresh six untreated samples. 2011 Downloaded from nar. qPCR-based SPUD assay for the detection of PCR (enzyme) inhibitors.4 Nucleic Acids Research. Individual wells were supplemented with water (negative control. the higher the difference in Cq (50 –30 dCq). The longer the heat exposure. Assay characteristics. frozen neuroblastoma biopsies in 11 different laboratories . the lower the RNA Quality Index (RQI).org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15.oxfordjournals. (a) qPCR based 50 /30 ratio mRNA integrity assay for 6 RNA samples from cultured neuroblastoma cells artificially degraded by heat exposure (at 80 C). of tumour samples RT and PCR variability. The longer the heat exposure. The dCq values >1 indicate the presence of PCR (enzyme) inhibitors. resulting in a decrease of the RQI as shown in Figure 2b Assessment of RNA purity and integrity in a large series for one representative RNA sample. grey squares) leading to an increase in Cq values.

oxfordjournals.001). we now studied the impact of cantly larger effects for specific combinations (data not RNA quality on risk classification performance using a shown). the negative 10 log of the Downloaded from nar. we also evaluated the impact of has an impact on the remaining variation of normalized RNA quality parameters on the differential expression of reference gene expression.5). normalized ref. Nucleic Acids Research. All parameters were significantly samples for at least one RNA quality parameter (as correlated to each other (Figure 4. the Cq value of ex. the 50 –30 dCq and the normaliza- tion factor had the largest impact on reference gene vari- ation when comparing best quality with worst quality Influence of RNA integrity on multigene level samples. D and C are seem not to be sensitive to RNA quality (as exemplified measured on unamplified RNA. negative predictive value and overall accuracy as non-HR) using the six RNA quality parameters in all ROC–AUC analysis of either a 6-gene or a 59-gene . we wanted to determine if RNA quality risk patients). mean difference in significance is 4. mean cDNA. We used two different classification second step. The same conclusions can be compromised RNA samples is >2-fold in the 50 most obtained. calculating the variation. N and A param- eters seem to result in more genes where better quality Influence of RNA integrity on single gene level RNA leads to more significant results. R. pressed Alu repeat sequences (A) (measure for the significance of differential expression is higher for overall mRNA content) and a normalization factor high-quality samples for at least one RNA quality param- based on the arithmetic mean Cq value of four stably ex. including 18S/28S rRNA clearly show an influence of RNA quality on single gene ratio (S). and Cq (A) and the normalization factor (N) (r2 = 0. we ex. Within each of the sampling. difference in significance is 0. significance quency of the different parameters are displayed in of differential expression seems to be better in low-quality Figure 3 and Table 1. HPRT1. 15 low risk patients versus 15 high samples. P < 0. hence displaying a low variation across all small sample sizes (e. we per- worst quality samples starting with 50 samples up to all formed a Mann–Whitney test for differential expression 615 samples for which expression and clinical data were analysis and plotted the 10 log P-values as boxplots. (Figure 6 and Supplementary Table S2). 8 genes UBC) (Supplementary Table S1). available.1%. For this gene-specific variation of each reference gene (standard analysis. there is an opposite effect.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. all but 8 samples for which clinical and expression data were avail- (1. specificity. (Table 2). R. The significance of differential expression the negative control and thus considered to be free of in. for 19 genes. A and N on pre-amplified for 50 /30 ratio assay for target PLAT in Figure 6b. eter (as exemplified for 50 /30 ratio for target SLC25A5 in pressed reference genes (N) (HMBS. HPRT1 30 Cq value (C). The results determined and compared. The variation in the 50 most quality. The frequency distribution and cumulative fre.95). Not unex. Out of 740 samples.5 versus 1. non-HR group was measured using the same procedure trapolate that the applied extraction methods generally of growing sample subgroups of opposite RNA quality as provide RNA free of PCR inhibitors. SDHA. positive predictive between two risk groups of cancer patients (HR versus value.68). 2011 5 using 5 different protocols. 75% = 3. Based on this. This approach reference genes as shown in Figure 5 for one representative was applied on the representative genes depicted in reference gene (UBC) and RNA quality parameter Figure 6a–c for quality parameter D (cut-off for good (HPRT1 50 –30 dCq). D and C par- ameters seem to result in as many or more genes where Variation of reference genes. we studied the impact of RNA quality on the algorithms PAM (22) and correlation signature (23) and significance of differential expression of marker genes determined the sensitivity. used for assessment of reference gene variation. the genes within such an experimental set-up. Figure 6a. a reduction in sig- ations of 4 reference genes with 6 RNA quality parameters nificance can be appreciated (Figure 6f). Instead of In order to evaluate the RNA integrity of the neuro. gene expression studies are often based on samples. On average. i. In principle. no bias (Figure 6e) and for CAMTA1. multigene signature in the samples for which clinical data were available (30 training samples and 585 test Differential expression of prognostic marker genes. there is quality samples was observed for 18 out of 24 combin.29). HPRT1 50 –30 difference in Cq or differential expression for a substantial number of genes dCq (D). For 27 genes. S. grouped according to the number of samples belonging RNA quality has a clear influence on the noise of the to the 75% percentile of the best quality. for SLC25A5. for 5 genes there are conflicting data. S. the highest correlation was found between Alu Figure 6c. the differential expression is intact samples (3. We also tested the effect of different combinations of After establishment of a clear link between RNA quality the six RNA quality parameters but could not find signifi. mean difference in significance is 8. 2011 blastoma samples. A significant difference in more significant when testing on a higher proportion of reference gene variation between best quality and worst high-quality samples (Figure 6d). erence gene expression levels should be similar in all In practice. RQI (R). while for PLAT.13). we performed 100 000 random sampling of 15 variation) was calculated for sample subgroups with patients from the HR group and 15 patients from the increasing numbers of either best quality samples or non-HR group. of each marker gene (n = 59) between the HR and hibitors affecting the SPUD assay. Here. and single gene results. six different quality parameters were P-value of the Mann–Whitney test was used. confirmed in a separate run) were within 1 cycle of able (n = 615). exemplified for 50 /30 ratio assay for target CAMTA1 in pectedly.54). In the samples) (Table 3). All RNA quality parameters were investigated. Therefore.g. To this purpose.e.

SDHA. N. S.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. RNA Quality Index determined by microfluidic capillary electrophoresis. A. R. HPRT1. .6 Nucleic Acids Research. Frequency distribution (left axis) and cumulative frequency (right axis) of six RNA quality parameters measured in 740 neuroblastoma tumour samples. normalization factor based on the arithmetic mean Cq value of four reference genes (HMBS. 2011 Downloaded from nar.oxfordjournals. HPRT1 30 Cq value. C. UBC). D. 18S/28S rRNA ratio determined by microfluidic capillary electrophoresis. HPRT1 50 –30 dCq (difference in quantification cycle value). 2011 Figure 3. Alu Cq value.

D. P < 0. HPRT1 30 Cq value. A. A. Asterisks indicate correlation between 30 Cq and 50 –30 dCq values was not calculated because 30 Cq is included in the calculation of 50 –30 dCq (not independent parameters).5 (2.8 6 (0. R.5) 2. Alu Multivariate logistic forward conditional regression Cq value. RNA quality of the tumour samples from sur- neuroblastoma tumour samples vivors was comparable with those from non-survivors n = 740 Median (range) Mean Not available (%) (data not shown).0 (0.4 to 45. D. RNA Quality Index determined by microfluidic capillary electrophoresis. normalization factor based on the arithmetic mean Cq value of four reference genes (HMBS.6 (1 to 10) 6.1) signature method appears less sensitive to RNA quality N 28. N. phoresis. S. normalization factor based on the arithmetic mean Cq testing all RNA quality parameters in a model indicated value of four reference genes (HMBS.9 (7. Nucleic Acids Research.2 (27. N. 2011 C 33.8 to 27.9) A 10.oxfordjournals. Missing values for D and C due to severely degraded RNA were replaced by the value from the sample with the worst quality according to the parameter under study (these imputed values were not taken into account for the correlation analysis).2) best quality and worst quality samples. the correlation signature method is used.3) 11. C.2) 1 0 (0) previous paragraph.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. UBC).2) 29. 18S/28S rRNA ratio determined by microfluidic capillary electrophoresis. Frequencies of six RNA quality parameters measured in 740 classifier. HPRT1. HPRT1. The correlation Downloaded from nar. C.8) impact on classification performance when comparing D 2. Alu Cq value. HPRT1 50 –30 dCq. SDHA.1) than the PAM method and the 59-gene classifier is less sensitive to RNA quality than the 6-gene classifier when S.8) 33.3 30 (4.3 to 39. HPRT1 30 Cq value. SDHA.8 53 (7. that the 50 –30 dCq and normalization factor quality Figure 4.5 (24. UBC). HPRT1 50 –30 dCq (difference in quantification cycle value).1 30 (4.001).8 14 (1. Correlation scatterplots between six RNA quality parameters measured on 740 neuroblastoma tumour samples indicate a significant correlation (Pearson.0 to 4. The impact of RNA quality was determined using the same procedure as described in the S 1. R.1 to 10. RNA Quality Index determined by microfluidic capillary electrophoresis. 18S/28S rRNA ratio determined by microfluidic capillary electro. . 2011 7 Table 1. The 50 –30 dCq had the largest overall R 7.

A.81a. a useful quality parameter is needed as well as a statistically significant together with other parameters measurable outcome. we demonstrated a measurable influence of RNA quality on the gene expression results. In contrast to samples (red curves) starting with 50 samples up to all 615 samples (quality here defined by 50 –30 dCq value). Mean difference in normalized reference gene standard variation between best quality and worst quality samples based on six RNA quality parameters in 615 neuroblastoma tumour samples RNA quality parameter HMBS HPRT1 SDHA UBC Mean a S 0.24 to 0. RNA Quality Index determined by microfluidic capillary electro- phoresis.12 0. D.81a 0. DISCUSSION eters to have an impact on outcome prediction using the An often underestimated critical issue underlying the reli- 59-gene PAM signature [odds ratios 4.31 to 0. Undisputedly.65 (95% CI: 1.26).26) (0.62 A 0. not aim to identify the best parameter measuring RNA relation signature (Supplementary Table S3).23) a Significant difference (P < 0. N. 2011 quality of an unprecedentedly large series of 740 clinical ation signature.19– ability of gene expression results is the quality of the RNA 18.33) (0. Variation is measured in increasing sion of prognostic marker genes and on the classification numbers of both best quality samples (blue curve) and worst quality performance using a multigene signature.e. We did significant independent parameters using the 6-gene cor.42a 0. The null distribution was some reports in the literature (8.78a 0. In this study. and the normalization factor was the only RNA samples using six RNA quality parameters.47a 0. Variation of reference gene UBC calculated as the standard deviation of the log transformed gene expression levels after normal. Upon careful interpretation of the impact of RNA quality on the results using the novel methods.24 to 0.72a 0. resulting in successful pre-amplification of partially compromised RNA samples in case of random priming (23). RNA quality on the final results. A substantial impact Table 2.66a 0.62 C 0.66a 0.58a 0. SDHA.27 to 0. A significant—albeit imperfect—positive correlation was found between all RNA quality parameters. R. C.18 0. samples. we Figure 5.41 0.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. each par- ameter appears to have a different appreciation of RNA quality.06 0.59a 0. Therefore. Grey boxes: the value with the largest mean difference (quality parameter with biggest impact on variation) for each reference gene.60a 0. b Mean difference in reference gene UBC variation according to 50 –30 dCq (example shown in Figure 5).92).75a 0. on the significance of differential expres- genes (HMBS.49a 0.32a 0. clearly observed an effect on the variation of reference ization with the arithmetic mean Cq value of three other reference gene expression.28) (0. Alu Cq value. HPRT1 30 Cq value.05) (smaller in permutated samples than in ranked samples based on RNA quality). a possible explan- ation for the lower correlation between these parameters is the use of a different RT priming strategy. 18S/28S rRNA ratio determined by microfluidic capillary electrophoresis.25 0.18. we developed and applied an analytical framework using novel methods for evaluation of RNA quality in relation to qPCR results. c Based on 100 random permutations. respectively]. and 17.19 0.72 90% null distribution intervalc (0.oxfordjournals.12 R 0. . we have examined the using the 6-gene PAM signature and the 59-gene correl- Downloaded from nar. normalization factor based on the arithmetic mean Cq value of three reference genes (i.17–76.86 (95% CI: 4.51a 0.60a 0. Similar results were obtained for the other reference does not completely resolve the effect of compromised genes and RNA quality parameters. This might also explain why some samples classified as bad quality based on RQI or 18S/28S rRNA ratio (measured on total RNA) turn out to be better quality samples based on the other methods (measured on cDNA). HPRT1 50 -30 dCq (difference in quantification cycle value). 2011 parameters were the only significant independent param.56 N 0. HPRT1 50 –30 dCq and HPRT1 30 Cq value were measured on cDNA obtained from anchored oligo-dT priming of original RNA.b 0.37 D 0.72a 0.06 0. While the Alu repeat sequence expression level and the normalization factor based on four reference genes were determined on randomly primed pre-amplified material. Rather. To assess the impact of RNA quality on the These parameters were also found to be independently results.8 Nucleic Acids Research. quality. HPRT1). S. the results obtained determined upon 100 random permutations (grey curves denote 90% from this study indicate that the process of normalization distribution range). excluding reference gene under evaluation).

Nucleic Acids Research. (i) The negative 10log of the P-value is calculated for increasing numbers of both best quality samples (blue curve) and worst quality samples (red curves) starting with 50 samples up to all 615 samples (a–c). The effect of RNA quality on the significance of differential expression of a marker gene between tumours from two risk groups of neuroblastoma patients (HR versus non-HR) (Mann–Whitney test) was investigated using two different approaches.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. 2011 9 Downloaded from nar. 2011 Figure 6.oxfordjournals. (continued) . The null distribution was determined upon 100 random permutations (grey curves denote 90% distribution range).

Indeed.8 0.0927 0. the results seem better when RNA was of lower biopsies.0 11.05 COR 6b PAM 59b Values represent the mean difference in performance between two groups of divergent RNA quality.4 1.6 1. Surprisingly. AUC. 2011 D 6.8 8. SPEC.8 12. while others are not. assay ampli- lower RNA quality generally results in higher Cq values.0 2.7 0. Alu Cq value.3 S 2. representative gene (CAMTA1) for which differential expression is more significant in worst quality samples (c and f). of RNA quality on the standard deviation of each of the standardization and improvement of pre-analytical pro- four reference genes when normalized by the three other cedures for in vitro diagnostics (M.0 7.01 <0. 18S/28S rRNA ratio determined by microfluidic capillary electrophoresis.01 0.3 1.0 4.2 C 0. including qPCR assay and sion stability is influenced by RNA quality and that genes gene expression specific characteristics such as amplicon display varying sensitivity to RNA degradation (7).8 7.5 14. Grey boxes: the value with the largest difference (quality parameter with biggest impact on classification performance) for each performance parameter.01 <0. the 50 –30 dCq and normalization factor quality patients. A.9 4.8 0.1 0. a Worst RNA quality samples show 6.3 9.4 1. Personal reference genes was observed.8 D 0.7 S 9.2 6. UBC).2 7.5 C 5. NVP. positive predictive value.7 3.7 1.6 6.9 5.8 1.5 3.05 <0.5 5.01 0. COR/PAM: correlation signature/prediction analysis of microarray method using 59 or 6 genes. classifier and by the nature of the applied classification ing on the input amount (such as 18S/28S rRNA ratio.3 2. be the least sensitive to RNA quality and an expression RNA quality has also a noticeable influence on the sig. ferential expression between the two risk groups. no clear ation is not simply due to sampling noise that occurs when explanation was found as to why some genes were more the number of input molecules are low.0 1.8 1.6 2.1 4.5 R 8.9a 9.5 A 13. This puzzling observation is in accordance to useful parameters to qualify RNA samples. distance to 30 -end.01 0.0 6.5 0. The correlation signature algorithm seems to RQI and 50 –30 dCq). HPRT1 30 Cq value.7 7. algorithm.0 16. negative predictive value.01 <0.1 6.8 0.3 0. HPRT1 50 -30 dCq (difference in quantification cycle value).5 2. furthermore. transcript length. representative gene (PLAT) for which expression results are not sensitive to RNA quality (b and e).4 1.5 6.1 5.8 1.6 1.6 0. b Comparison of the values of all RNA quality parameters for a specific performance between two classification methods (t-test).7 14. the sensitive to RNA quality than others (data not shown).2 2. R.3 0. overall accuracy as receiver operating characteristic area under the curve analysis.4 1.7 8.3 2. specificity.2 3.2 0. the same observations are made rity is influenced by the number of genes included in the when RNA quality metrics are used that are not depend.4 0.1 5. N.6 2.0 COR 59 versus 0.6 5. Upon extensive correlation analyses our previous report indicating that reference gene expres. RNA Quality Index determined by microfluidic capillary electrophoresis. for a few qPCR expression results obtained on fresh frozen genes.9% lower sensitivity compared to best quality samples.9 12. sensitivity.0762 <0.8 4.3 N 1. This is in accordance with communication). SDHA.4 1.0651 PAM 59b COR 6b COR 59 versus 0. they ‘appear to constitute the most’ quality.9 5. Impact of RNA quality on risk classification performance in 585 neuroblastoma tumour samples SENS SPEC AUC PPV NPV SENS SPEC AUC PPV NPV (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) COR 59 COR 6 S 4. .3 0.8 A 6.54).7 1.6 2. genes between two divergent risk groups of cancer Overall.oxfordjournals.9 4.876 PAM 6 versus <0.4 0.351 <0.5 Downloaded from nar. Representative gene (SLC25A5) for which differential expression is more significant in best quality samples (a and d).3 D 5.8 4.234 <0.9 N 2.8 4.7 1.0 3. C.0 R 13. As such.01 0.2 3.158 0.4 0.9 0.9 0. measured Cq values of the reference genes are found in Our data further show that the performance of gene a range well below values at which sampling noise is expression based classification in function of RNA integ- expected. The advantage findings within the EU FP7 Spedia project on of using a 50 /30 ratio assay or the normalization factor to Figure 6. 2011 Table 3. using different parameters.6 0.1 0.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15.0 D 0.4 N 5.6 15.8 0.7 12. normalization factor based on the arithmetic mean Cq value of four reference genes (HMBS. mean Cq value and magnitude of dif- it is important to note that the observed increase in vari.0 3. D. As length. PPV.4 5.3 2.3 5. S.05 <0. Some genes appear to be sensitive to RNA parameters appear to have the largest influence on the quality.9 7.478 <0. fication efficiency. Each boxplot represents differential expression between HR and non-HR groups according to the fraction of samples with best quality (dCq 50 –30 < 3. SENS. HPRT1.2 R 8.3 R 8.3 A 1.10 Nucleic Acids Research.0 C 13.3 3.4 5.9 0.1 0.0 N 9.6 3.610 PAM 6 versus 0. Kubista.4 5.7 1. signature built with a larger number of genes results in nificance of differential expression of individual marker more robust classification.3 A 2.0 C 14.3 4.5 PAM 59 PAM 6 S 5.1 1.2 10. Continued (ii) The impact of RNA quality parameters on the differential expression of genes between 100 000 random samplings of 15 patients belonging to two risk groups (HR versus non-HR) (Mann–Whitney test) (d–f).05 <0.

and Wittwer.S.S.H. Genet. Belgian Foundation Against Cancer (SCIE2006-25).. Furthermore. The results from this study demonstrate that monitor. 30. the 50 –30 dCq value in itself is not tation and use of the microfluidic capillary electrophoresis only depending on the RNA quality. Exp. further studies are warranted to establish a Children Cancer Fund Ghent. 9. Rev. the Belgian Society of expression study. Prime Minister’s Office. (2005) Impact of RNA quality eter to measure the impact on the gene expression results. Ginzinger. Eveno.J. ing RNA quality is of critical importance to obtain mean. and S. Barlet. and Auffray. Trends traction and storage of high quality RNA. (2000) Absolute quantification of mRNA using random primed reverse transcription could allow samples real-time reverse transcription polymerase chain reaction assays.A. need of proper RNA integrity control and proposes a 7. Van Cauwenberge. Diagn..A. on reference gene expression stability. (2002) Gene quantification using real-time increased if gene expression profiling is performed imme. Detection of rare mRNAs via quantitative RT-PCR. Nonetheless. The number of useable samples can also be 3. the gene expression measurement Scientific Research Flanders (G. 169–193.C.V. (1992) trained to use standard operating procedures for the ex.. 2011 11 assess the integrity of an RNA sample before its use in a SUPPLEMENTARY DATA gene expression study is that it specifically ‘addresses’ the Supplementary Data are available at NAR Online. with some degree of RNA fragmentation to become J.11). This is not the case for other methods such as microfluidic electrophor- esis that predominately inspect the ribosomal RNA profile ACKNOWLEDGEMENTS to infer RNA quality. 18–23. and Bustin. clinical history of patients. (BOF) to J.A.. Of note.g.. RNA quality is unacceptable. 2. 2011 technical assistance.S. The last two factors do Neuroblastoma Group (SIOPEN). 5. Bustin.0198.oxfordjournals. but may also be (Experion).M.P. Endocrinol. European performance (qPCR efficiency). 187–197. the sensitivity to deg. quantitative PCR: an emerging technology hits the mainstream. charge: VLK (Flemisch League Against Cancer). diately after sampling of the tumour (which is clearly a 4...). Biotechniques. sidered. However. This study confirms the user-independent classifiers of microcapillary electrophoresis traces. eligible (27).D. We propose that pilot experiments are Paediatric Haematology and Oncology.D. and J.08). e56. the inability to diagnose or assess prognosis of patients due to inferior Conflict of interest statement.S.S. if the aim is to es. Isabelle De Roose mRNA transcripts that form the actual template in RT– and their team (Bio-Rad) for support with the implemen- qPCR analyses. . 39. a more stringent assessment of RNA quality is probably initiated by the Belgian State. the Ghent method (RT–qPCR versus microarray versus massively University Research Fund [Bijzonder Onderzoeksfonds parallel sequencing). the Foundation Fournier cut-off value for inclusion of a given sample in a gene Majoie pour l’Innovation. Martin. these factors should be con.V. the Gesellschaft fuer not need to be taken into account if the RNA quality Paediatrische Onkologie und Haematologie parameter is used to rank the samples in the same study (GPOH-Germany). the Belgian initiated that include positive and negative control Kid’s Fund (to J. (2005) ingful and reliable gene expression data and to ensure Towards standardization of RNA quality assessment using reproducibility of the results.. Nucleic Acids Res. Undoubtedly...).J. the Institute expression difference. and Vandesompele. Technology in Flanders (to S.J. Pharm. Funding for open access dividual patient compared to drawing statistical conclu... 5254.L. Belgian program of Interuniversity Poles of Attraction. integrity of a messenger RNA molecule.K.C.S. such methods We thank Els De Smet and Nurten Yigit for their excellent provide an indication of total RNA quality but are not Downloaded from nar.A.).O. the for the Promotion of Innovation by Science and intra-group expression variability. the Fund for Scientific is expected that this value will depend on the observed Research Flanders (to K. the Fund for radation of the target..P. Graudens.B.. Therefore.T.V.org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15. The evaluation of qPCR assays targeting the 30 -end and the 50 start of other reference gene is needed in order to confirm our results and establish 50 –30 dCq as a FUNDING valuable RNA quality parameter.R. 6. We also would like to acknowledge necessarily most appropriate to predict the integrity of Arnaud Remy..P. Bustin.. known that storage of RNA samples might lead to deg. Imbeaud. the Foundation pour la recherche samples in order to establish a study-specific cut-off. (2009) Reliability of real-time radation (10. The under the FP6 [project: STREP: EET-pipeline (037260)]. 503–512. We are indebted to all members of the influenced by reverse transcriptase efficiency and assay International Society of Paediatric Oncology. None declared. Perez-Novo.C. (2008) Real-time quantitative PCR – opportunities compromising factor in retrospective studies) as it is and pitfalls. reverse-transcription PCR in clinical diagnostics: gold standard or substandard? Expert Rev. 25. Eur.M. Science Policy Programming. the Clearly. Mol.F. Hematol. (COG-United States) for providing tumour samples and tablish a quality cut-off value. cut-off value will also depend on the purpose of the study.G.S. Boulanger.. It Nuovo-Soldati (to J. the target abundance.. 4. and the Children’s Oncology Group according to RNA quality. 263–264. efforts should be made to overcome this problem and to increase the percentage of eligible cases for gene expression profiling REFERENCES in a clinical setting.D. laboratories should be 1. Nucleic Acids Research. Bachert. Weis.E. and the nature of the samples (e.A. Tan..C. Schroeder. 8. the European Community fresh frozen versus formalin fixed paraffin embedded). framework to assess the value of an RNA quality param. 33. needed when a therapeutic decision is required for an in.E. Mueller.X. 56.). For instance. sions for a group of samples. Claeys. Zaborski.C. Mary Grace Brubacher.L. Mol.]. Speleman. Murphy.F.V.H.

.T..K.R. Lefever. (2009) RDML: (2004) Linear mRNA amplification from as little as 5 ng total structured language and reporting guidelines for real-time RNA for global gene expression analysis. Naranjo. Przybylski.. Hastie.. Herrler.J. does not preclude microarray analysis of small amounts of RNA.E. Genome Biol.G.F.M. and Vandesompele.B. Proc. 663–671.. Diagnosis of multiple cancer types by shrunken centroids of gene (2009) The MIQE guidelines: minimum information for expression.J. Garson..A. Stat. Poustka. quantitative PCR data.J. and Biotechniques. prognostic role of a gene signature from tumorigenic Van Roy. Genome Biol.T. Dafforn. real-time PCR depend on short amplicons and a proper 1973–1981.. . 1198–1201.N. Taylor.. Lefever. De Preter. 13...G. De Paepe. Research0034. Huber. Nolan..A.T. D942–945.S..M. Clin.S. 2. 22.F. Shipley. Tibshirani.D.M.H. linear nucleic acid amplification expression measurements from degraded RNA by quantitative systems for highly multiplexed applications.S.. 2065–2069. Narasimhan.M.J... 27. Vandesompele. 543. (2006) Quantification of and software for management and automated analysis of mRNA using real-time RT-PCR....J. Vermeulen. Stangegaard.V.A.W. 1. Antonov..P... 99. Chem..J. Smith. 611–622.S. 854–857. 2011 epidemiological studies.X. Sherlock. (2005) Novel isothermal. et al. (2009) RNA pre-amplification enables large-scale and quality of resulting cDNA.A. Hands. Pattyn..F... et al. Natl Acad. Notes.F.... (2003) Moderate degradation Research0034.R. and Bustin. Center of Life Science in Freising in nucleic acid preparations. (2007) qBase relative quantification framework 12. Genet.J. Weihenstephan.F.M.S...W.N. (2002) Kubista. Tonini.oxfordjournals. Untergasser. 37. Scaruffi.. RT–qPCR gene-expression studies on limiting sample amounts.B...H.M.S.. R19.S. Hellemans. real-time quantitative PCR data.R. Van Roy.N.. 17. 8. real-time quantitative RT-PCR data by geometric averaging of 11.. 21.U.. Kurn. Med.A.P. Anal. N. Vercruysse.F.G. Liu. Biol.J. Chen.M.F..1– Rammensee. Lancet Oncol. normalization. and Chu. (2009) the real-time qRT-PCR performance. Germany. Goldstein. (2002) Accurate normalization of Downloaded from nar. Mortier. De Paepe..H.. and Stevanovic.M. (2006) RNA integrity and the effect on 18. Nolan. J.T. Protoc.T. 649–657.S.P.J. Lefever. and Speleman.A.. Speleman. 16.K. Appl. De Preter.. Hands. Kopf-Sill... Biochem.N. and Bustin.. Derveaux.....S.M. 27. Chen. Pfaffl.T. 126–139. Stephens...K.. Dalerba. and Bastaki. Technische Universita¨t Mu¨nchen. Bustin.S.W.F.C.G.S.M. Nucleic Acids Res.. and Dufva.J. 2011 8.. 37. Mutat. (2005) 2nd International qPCR Symposium & SPUD: a quantitative PCR assay for the detection of inhibitors Application Workshop. 85. Schoor.J.. Aspects Med.. (2009) Predicting outcomes for children with 24.L.A. Biological sample collection and processing for molecular De Paepe. and Pattyn. Pattyn.. Chen.G.. and Pfaffl. Holland. publication of quantitative real-time PCR experiments. high-dimensional classification tasks.N. 25. Article37. Stenzl.. 1040–1050.R.. multiple internal control genes.D. Vandesompele. Dufva.A. Pirotta. and Clarke. Nat.G. Deng.W. 51.A. Huggett.. Chem... (2003) 20..B.K.E.J.A.S. Res... Mol. Oberli. Purohit. Lato. 3..K.. Lab Invest. BMC Res. Nucleic Acids Res.. and Mansmann... Engl. Mol. Hennenlotter.Y.E. Vandesompele.I..J. and transcription using random pentadecamer primers increases yield Vandesompele. De Smet. (2006) 19.. Weinschenk. Iglehart.L.G. breast-cancer cells. Lewicki.S.. Hellemans.R.. (2006) Reverse Yigit..M.J. 217–234. 1559–1582. Ruschhaupt. 26. De Preter. 1192–1196. Wang.J..T. 55. RTPrimerDB: the portal for real-time PCR primers and probes.... Clin. Vandesompele. Biotechniques.S. Hellemans. Bravo. 308–310.11.A. Hoey. Altermatt.M. Shedden. Geurts. 10. Wang.. Nolan. Gurney. (2007) The 14..J.D..org at Teilbibl Chemie Biologie undGeowissenschaften der Technischen Universitaet on February 15.. Baltzer. Fleige. Poppe.K. 217–226. Speleman. USA.F. Biotechniques.. et al. Vermeulen.. Swerts. (2005) Reliable gene Wang. 40...R..O. Ogunkolade.R.S..P. Heath.. 356. 351. 35.A. Pillarisetty.R.D.12 Nucleic Acids Research. and Jaggi.J. Koritala.R.R.J. Benes. Hellemans. 10. Pattyn.G.A.A..A. Speleman.. Mueller. 23. (2004) neuroblastoma using a multigene-expression signature: a A compendium to ensure computational reproducibility in retrospective SIOPEN/COG/GPOH study. Eskenazi. 9.A.M. and Fleischmann. 15...J. 235. Sci... 3.A. Corvin..P. 37. 6567–6572.T.