BMC Biotechnology Biomed Central

BMC Biotechnology BioMed Central

Research article Open Access Characterization of the effect of sample quality on high density oligonucleotide microarray data using progressively degraded rat liver RNA Karol L Thompson*1, P Scott Pine1, Barry A Rosenzweig1, Yaron Turpaz2 and Jacques Retief2,3

Address: 1Center for Drug Evaluation and Research, US Food and Drug Administration, Silver Spring, MD, USA, 2Affymetrix Inc., Santa Clara, CA, USA and 3Current- Illumina Inc., San Diego, CA, USA Email: Karol L Thompson* - [email protected]; P Scott Pine - [email protected]; Barry A Rosenzweig - [email protected]; Yaron Turpaz - [email protected]; Jacques Retief - [email protected] * Corresponding author

Published: 13 September 2007 Received: 8 May 2007 Accepted: 13 September 2007 BMC Biotechnology 2007, 7:57 doi:10.1186/1472-6750-7-57 This article is available from: http://www.biomedcentral.com/1472-6750/7/57 © 2007 Thompson et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract Background: The interpretability of microarray data can be affected by sample quality. To systematically explore how RNA quality affects microarray assay performance, a set of rat liver RNA samples with a progressive change in RNA integrity was generated by thawing frozen tissue or by ex vivo incubation of fresh tissue over a time course. Results: Incubation of tissue at 37°C for several hours had little effect on RNA integrity, but did induce changes in the transcript levels of stress response genes and immune cell markers. In contrast, thawing of tissue led to a rapid loss of RNA integrity. Probe sets identified as most sensitive to RNA degradation tended to be located more than 1000 nucleotides upstream of their transcription termini, similar to the positioning of control probe sets used to assess sample quality on Affymetrix GeneChip® arrays. Samples with RNA integrity numbers less than or equal to 7 showed a significant increase in false positives relative to undegraded liver RNA and a reduction in the detection of true positives among probe sets most sensitive to sample integrity for in silico modeled changes of 1.5-, 2-, and 4-fold. Conclusion: Although moderate levels of RNA degradation are tolerated by microarrays with 3'- biased probe selection designs, in this study we identify a threshold beyond which decreased specificity and sensitivity can be observed that closely correlates with average target length. These results highlight the value of annotating microarray data with metrics that capture important aspects of sample quality.

Background fore important to understand how RNA quality affects the It is recommended that the highest quality RNA be used interpretation of the results and also how reliable current for genomic analyses. However, in some cases, such as quality measures are at indicating RNA quality issues. It human autopsy samples or paraffin embedded tissues, has been reported that gene expression profiling on high quality RNA samples may not be available. It is there- Affymetrix GeneChip arrays is relatively tolerant to mod-

Page 1 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

erate RNA degradation and to the 5'-truncation that Results occurs during the two successive rounds of in vitro tran- Effect of sample handling on RNA integrity scription needed to detect small sample quantities [1-3]. The methods used to harvest and preserve source tissue for Some samples fall within a "grey zone" of sample quality, gene expression analyses can impact the quality of iso- where there is some loss of RNA integrity but the samples lated mRNA and the reliability of microarray data gener- still pass RNA quality thresholds. It is unknown how dif- ated from this source. We investigated the relative impact ferences in RNA integrity within the "grey zone" affect the of several different tissue handling conditions on RNA data interpretation. More information is needed to help integrity. These conditions were designed to model the guide the generation of best practice recommendations effect of time between necropsy or sacrifice and sample for sample handling and the evaluation of the quality of processing (incubation at room temperature or 37°C) or genomic studies submitted to public databases to fulfill between removal from storage and sample processing journal requirements and to regulatory agencies. (time of thaw of frozen sample). Fresh liver tissue was incubated up to 6 hr at room temperature without a meas- The recommended method for preparing target from RNA urable effect on RNA integrity, as measured by RIN (Figure for hybridization to Affymetrix microarrays is based on 1). RNA in fresh liver tissue proved to be remarkably sta- the Eberwine procedure [4]. The sample labeling and ble. RNA degradation was only observed after fresh liver amplification method starts with cDNA synthesis from tissue was incubated at 37°C for 120 min or more and the polyadenylation (polyA) site followed by the genera- poor quality RNA (RIN ≤ 7) appeared after 3.5 hours of tion of cRNA from the sense strand of the cDNA via an incubation at 37°C. RNA degradation was much more incorporated T7 primer sequence. Because this process rapid in frozen tissue. Poor quality RNA (RIN ≤ 7) was iso- generates labeled target with a 3' bias, Affymetrix Gene- lated from frozen tissue thawed for 15–30 min at room Chip Rat Expression Set 230 (RAE230A) arrays are temperature. designed to contain probes that reside within the 600 nucleotides (nt) most proximal to the 3' end of each tran- Sample characterization by RNA quality metrics script [5]. Where alternative polyA sites are identified For each of the sample handling conditions that induced within 600 nt of each other, the probe selection region is RNA degradation (37°C incubation or freeze/thaw (F/T)), based on the most upstream site. The housekeeping genes sets of progressively degraded RNA were generated in beta-actin (Actb) and glyceraldehyde-3-phosphate dehy- independent experiments, with each experiment using a drogenase (Gapdh) serve as internal controls of RNA quality and the target preparation process. Probe sets have been designed to hybridize to the 5', middle (M), or 3'- regions of these control transcripts. High signal ratios of the 3' probe set to the 5' probe set are indicative of either RNA degradation or target synthesis problems. It has been recommended that samples should have a 3'/5' signal ratio for Gapdh of no more than 3 [6].

Various methods for measuring sample quality pre- and post-hybridization have been proposed [7-10]. In this study the degree of RNA degradation was standardized by use of an Agilent 2100 Bioanalyzer to assign an RNA integrity number (RIN) to each sample. The RIN software algorithm classifies the integrity of eukaryotic total RNA on a scale of 1 to 10 (most to least degraded) based on the most informative features of an electropherogram of the TimehandlingFigure course 1 conditions of RNA degradation induced by different tissue 18s and 28s rRNA peaks [11]. Time course of RNA degradation induced by different tissue handling conditions. RNA was prepared from liver sections incubated at room temperature (RT) (circles), In this study a set of rat liver samples with a progressive 37°C (triangles), or frozen and thawed at room temperature loss in RNA quality was generated. This dataset was used (F/T) (squares). The dotted line represents the linear trend- to characterize individual probe set sensitivity to RNA line for the RT incubated sample set. Dashed or solid lines degradation and to evaluate the effect of RNA integrity on connect the mean values at hourly or semi-hourly time the sensitivity and specificity of microarray data generated points for the 37°C or F/T handling condition sets, respec- on Affymetrix GeneChip arrays. tively. Symbol shading indicates replicate experiments conducted on different days from independent sources of tissue.

Page 2 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

FigureElectrophoretic 2 tracings of RNA progressively degraded by different handling methods Electrophoretic tracings of RNA progressively degraded by different handling methods. Rescaled tracings were overlaid from three RNA samples from independent experiments with similar RIN values. Tracings for RNA from samples degraded by freeze/thaw (A, C, E, G) or by 37°C incubation (B, D, F, H) and with values of RIN 9.5 (A and B), RIN 8 (C and D), RIN 7 (E and F), or RIN 6 (G and H) are shown.

Page 3 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

Table 1: RNA metrics associated with RIN value and RNA degradation method. Average values and standard deviations are reported for metrics associated with samples in each RIN class (n = 3).

Handling RIN class RIN 28s/18s rRNA cRNA yield cRNA length 3'/5' GAPDH 3'/5' β-actin condition

F/T 9.5 9.5 ± 0.1 1.43 ± 0.15 86 ± 8 1923 ± 157 1.17 ± 0.06 1.70 ± 0.35 8 8.1 ± 0.2 1.00 ± 0.00a 63 ± 4b 1398 ± 116a 1.77 ± 0.21 1.93 ± 0.06 7 7.0 ± 0.1 0.70 ± 0.10a 44 ± 12a 1136 ± 207a 3.05 ± 0.95b 3.62 ± 1.18b 6 6.2 ± 0.1 0.50 ± 0.10a 38 ± 3a 878 ± 62a 5.50 ± 0.8a 5.27 ± 0.86a

37°C 9.5 9.5 ± 0.2 1.35 ± 0.13 74 ± 16 2294 ± 236 1.01 ± 0.02 1.34 ± 0.02 9 9.1 ± 0.1 1.33 ± 0.06 83 ± 7 2110 ± 322 1.24 ± 0.20 1.49 ± 0.26 8 8.2 ± 0.1 1.33 ± 0.06 61 ± 19 1834 ± 293 1.29 ± 0.18 1.60 ± 0.04 7 7.0 ± 0.2 1.07 ± 0.15b 42 ± 6 1210 ± 158a 2.00 ± 0.21a 2.12 ± 0.20a 6 6.1 ± 0.5 0.90 ± 0.10a 39 ± 18a 1244 ± 223a 1.96 ± 0.41a 2.29 ± 0.40a

aP < 0.01 compared to RIN 9.5 set bP < 0.05 compared to RIN 9.5 set single liver lobe from a different individual animal as 27 samples with RIN ≥ 6. All samples had percent present source tissue. Each of these sets contains a minimum of 12 calls within 10% of the mean value (49%). The scale fac- RNA samples across three or more replicate experiments tors (SF) for all but one hybridization were within 2 SD of with RIN values ranging from 9.5 to 5. Electrophoretic the mean (SF 1–3) and all were within 3 SD of the mean. tracings of RNA with RIN measurements of 9.5, 8, 7, or 6 All samples except one RIN 7 sample and 3 RIN 6 samples are shown in Figure 2 for three independent samples in in the F/T set had 3'/5' Gapdh ratios below the recom- each sample handling set. The tracings are highly similar mended threshold of 3. The effect on 3'/5' Gapdh and 3'/ between samples with the same RIN, but show subtle dif- 5' Actb ratios corresponded well with average cRNA tran- ferences between the two handling methods for a given script length (Table 1). The 3' boundaries of the target RIN value. Within each handling condition, 3 samples sequences (TargetSeq) for the 5' Gapdh and Actb probe sets with similar RIN values were grouped as replicate samples are located about 883 and 855 nt, respectively, upstream in all additional analyses on the effects of RNA degrada- of the 3' end of their corresponding RefSeqs and span tion. about 1150 nt in length [see Additional file 1]. Targets that are less than 878 nt in length on average (i.e. the F/T Next, the relationship was examined between RIN and 3 RIN 6 samples) would be expected to exhibit significantly other RNA quality metrics (28s/18s rRNA ratio, cRNA reduced hybridization to these probe sets. yield, and cRNA length) that are generated before or during microarray sample preparation (Table 1). Of these 3 The quality metrics discussed so far are summary metrics metrics, RIN value was most highly correlated with aver- that provide an assessment or surrogate measure of the age cRNA length across a set of 29 samples generated from overall integrity of the sample. Individual probe set sig- both handling methods (r = 0.86). RIN values also corre- nals may vary in their sensitivity to RNA degradation. To lated fairly well with cRNA yield and with 28s/18s ribos- visualize the effect of handling condition and degree of omal RNA ratios (r = 0.84 and r = 0.82, respectively). RNA degradation on individual gene expression profiles, Some differences in the correspondence between RIN we limited the comparison set to the genes that were most value and the other 3 RNA metrics were observed between affected by sample incubation (347 probe sets that were the two different handling methods. In general, there was changed by 2-fold or greater in at least 30% of all non- a stronger decreasing trend in metric value as a function of control F/T or 37°C samples compared to zero time con- RIN for RNA degraded by F/T than for RNA degraded by trols). The relationship between the log2 ratio data for the 37°C incubation. filtered sets of noncontrol samples from the F/T and 37°C incubations was displayed by plotting the heatmap and Microarray quality metrics and signal changes induced by dendrograms resulting from average linkage hierarchical RNA degradation clustering (Figure 3). The samples clustered primarily by To systematically assess the effect of degree of RNA integ- handling condition and then by degree of degradation. In rity on microarray data, RNA samples in each RIN group general, four different patterns of probe set responses are were analyzed on Affymetrix GeneChip Rat Expression visualized in the heat map. The majority of probe sets 230A arrays for both handling conditions. Most of the showed a decrease in signal induced by degradation that global microarray quality metrics that are summarized in was independent of handling condition and observable in Affymetrix report files were within normal ranges for the even moderately degraded (RIN 8) samples. A second,

Page 4 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

tionsFigureHierarchical 3 clustering of significantly changed genes from samples progressively degraded by different tissue handling condi- Hierarchical clustering of significantly changed genes from samples progressively degraded by different tissue handling conditions. Sample labels are concatenated from handling condition, time of incubation, RIN, and study code. Log2 ratios calculated for each sample relative to the average of three zero-time controls are mapped on a green (-1) to red (1) color scale. smaller cluster exhibited an increase in signal induced by responses (Zpf36, Slc25a25). Over expression of Zfp36, degradation by either method. The expression levels of a Btg2, c-Jun, and Egr-1 has also been reported to occur in third subset of genes were selectively altered by ex vivo surgically extirpated prostate tissue after 1 hr of warm incubation at 37°C [see Additional File 2]. The 10 genes ischemia [13]. Dusp1 and Egr-1 are also 2 of 14 gene tran- in this cluster are primarily involved in cellular defense scripts that increased in peripheral blood mononuclear responses like the mitogen activated protein kinase cells prepared by Ficoll-Hypaque density centrifugation at (MAPK) pathway (Dusp1, Hspa1a, c-Jun), immune 21°C compared to 8°C [14]. Independent confirmation response (Cxcl1), response to hypoxia (Egr-1) [12], cell of an increase in Egr1 mRNA levels (5-fold after 1 and 3 hr growth regulation (Btg2, Myd116, Bhlhb2) or other stress incubation at 37°C) was conducted using qRT-PCR (data

Page 5 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

not shown). A fourth cluster of 25 probe sets were selec- RefSeq), 5'-5' distance (distance from 5' end of the RefSeq tively decreased in signal after 2 hr ex vivo incubation at to the 5' end of the TargetSeq), and average RIN 9.5 signal. 37°C. More than 75% of these probe sets hybridize to A mean and standard deviation for each distance metric transcripts that are either highly expressed in immune was calculated within DEC or INV groups (Table 2). Indi- cells or involved in immune function [see Additional file vidual measurements for each probe set in the DEC and 2]. Ex vivo incubation may have caused a selective loss in INV groups are tabulated in Additional file 3. the presence, function or integrity of immune cells in the liver samples. The correlation between each distance metric and probe set sensitivity to RNA integrity was examined in an Characterization of probe set sensitivity to RNA unpaired t-test comparison of DEC and INV probe set degradation metrics. Probe sets that decreased in signal as a function of The correlation between RNA degradation by F/T and decreasing RNA integrity tended to be located signifi- average cRNA length observed in Table 1 suggests that rel- cantly farther from the 3' end of their target transcript ative probe set position on a target reference sequence sequences than INV probe sets (P < 0.0001) (Table 2). may be a determinant of sensitivity to degradation. To While average TargetSeq length did not significantly differ examine this further, probe sets were first identified that between DEC and INV groups, DEC probes tended to map showed a statistically significant difference in signal level to longer RefSeq transcripts and therefore had lower Tar- between RIN groups and a progressively increasing or getSeq Length/RefSeq Length ratios. DEC probe sets also decreasing trend in average signal between RIN 9.5, 8, 7, tended to be lower in signal in undegraded (RIN 9.5) sam- and 6 sets generated by F/T (INC or DEC, respectively). ples than INV probe sets. Only probe sets that mapped to a single reference sequence (RefSeq) transcript containing a terminal polyA Unlike INV probe sets, the 5'-3' distances of DEC probe sequence ≥ 10 nt were selected in order to accurately sets were bimodally distributed with maxima near 650 measure probe set location relative to the reverse tran- and 1600 nt (Figure 5). The DEC probe sets were divided scription initiation site. Using these criteria, 89 DEC and into two groups with 5'-3' distances either less than or 12 INC probe sets were identified. 61 probe sets were also greater than 1000 nt and analyzed further. The majority identified that were relatively invariant in signal as a func- (66/89) of DEC probe sets including tion of RIN (INV). AFFX_Rat_GAPDH_5_at and AFFX_Rat_beta-actin_5_at had 5'-3' distances > 1000 nt. All metrics that were signif- For probe sets classified as INV or DEC, seven measure- icantly different for the DEC group as a whole were also ments were made to characterize the location and length significantly changed for this subset. The 23 DEC probe of each probe set target sequence on its corresponding ref- sets with 5'-3' distances < 1000 nt (which includes erence transcript sequence (Figure 4). The INC set was not AFFX_Rat_beta-actin_M_at) were significantly different further characterized because of the small sample size. The from INV probe sets in TargetSeq length, TargetSeq metrics were RefSeq length, TargetSeq length, TargetSeq Length/RefSeq Length, and average RIN 9.5 signal (Table length/RefSeq length, 5'-3' distance (distance from 5' end 2). of the TargetSeq to the 3' end of the RefSeq), 3'-3' distance (distance from 3' end of the TargetSeq to the 3' end of the

Table 2: Distance metrics associated with probe set sensitivity to F/T RNA degradation. Probe sets were classified as invariant (INV) to degradation or as decreasing in signal (DEC) in response to degradation. DEC probe sets were further divided into two classes based on probe set location relative to the corresponding reference sequence termini.

Distance metric INV DEC DEC (5'-3' < 1000) DEC (5'-3' > 1000)

5'-3' distance 638 ± 386 1381 ± 665a 646 ± 171 1637 ± 576b 3'-3' distance 211 ± 388 973 ± 653a 289 ± 228 1212 ± 579b 5'-5' distance 1072 ± 783 904 ± 764 1250 ± 874 784 ± 689 RefSeq length 1711 ± 845 2285 ± 901a 1896 ± 830 2420 ± 891b TargetSeq length 479 ± 72 451 ± 109 406 ± 111c 467 ± 105 TargetSeq/RefSeq 0.37 ± 0.21 0.22 ± 0.10a 0.25 ± 0.14c 0.21 ± 0.08b a c b Avg log2 RIN 9.5 signal 11.5 ± 1.3 10.0 ± 1.7 10.0 ± 1.7 10.1 ± 1.8

Count61892366

aP < 0.0001 in unpaired two-tailed t-test comparisons of INV and DEC sets bP < 0.001 in a Tukey's post-test comparison of a one-way ANOVA cP < 0.01 in a Tukey's post-test comparison of a one-way ANOVA

Page 6 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

parisons of RIN 9.5 samples with RIN 8 or 7 samples yielded 16 or 255 false positives, respectively, using signals derived with either MAS5 or PLIER. A high number of statistically significant changes in signal level were observed for RIN 9.5 and RIN 6 sample comparisons (9243 using PLIER and 4203 using MAS5).

The effect of RIN level on sensitivity was assessed using receiver operating characteristic (ROC) plots that measure diagnostic accuracy. ROC plots are generated by plotting sensitivity (true positive fraction) versus 1-specificity ProbeFigure set 4 distance metrics (false positive fraction) along a continuum of decision Probe set distance metrics. Illustration of distance met- thresholds (P-value cutoffs). Known gene expression rics for a hypothetical probe set (vertical open bars) in relation to the 5'- and 3'-termini of the corresponding RefSeq changes were modeled in silico using a mixed tissue para- (solid horizontal bar). digm designed to measure microarray performance [15]. Two mixtures composed of different proportions of rat testis, brain, liver, and kidney RNA are the components of Effect of level of RNA integrity on microarray performance a reference material that has signal ratios of 1:4, 2:1, 3:2, Sensitivity (the rate of detection of true positives among and 1:1 in tissue-selective probe sets. This mixed tissue all positives) and specificity (the rate of detection of true RNA design can be effectively modeled in silico from array negatives among all negatives) are important perform- data for each tissue RNA component in the mixture. ance objectives for microarray experiments. In toxicoge- Microarray signal data from rat liver RNA with different nomic experiments that are designed to measure the effect levels of RNA integrity (RIN 9.5, 8, 7, or 6) generated by of time and dose level of treatment on gene expression, F/T, was combined in silico with signal data from rat brain, misleading results can be generated by confounding vari- testis, and kidney RNA. Assay sensitivity for detecting a ables such as RNA degradation, tissue sectioning, diurnal true positive fold change of 1.5-, 2-, or 4-fold at a fixed effects, etc. The effect of RNA degradation on assay specif- false positive rate of 10% was calculated for each RNA icity was measured by comparing control liver samples in quality level from data modeled with proportional differ- which RIN level was the independent variable. Statistical ences in liver-selective (LS) signal and a 1:1 ratio of kid- Analysis of Microarrays (SAM) was applied at a median ney-selective (KS) signal. These calculations were made false discovery rate (FDR) of 0.1 in two-sample compari- using the entire set of 292 LS probe sets unselected for sen- sons of undegraded control liver RNA (RIN 9.5) with liver sitivity to RNA degradation (LS_ALL) or subsets of probe samples of decreasing RIN value generated by F/T. Com- sets that either significantly decrease (LS_DEC) or are invariant in signal (LS_INV) as a function of RNA quality as true positives and 188 KS probe sets as true negatives (Table 3).

Assay sensitivity was markedly decreased by the use of LS_DEC or LS_ALL probe sets as analytes for detecting modeled changes of 1.5-, 2-, and 4-fold as a function of RIN level. The effect was most pronounced for LS_DEC probes, for 1.5-fold change detection, and for RNA quality of RIN ≤ 7. ROC plots that used LS_INV probe sets as analytes showed no change in sensitivity as a function of either RIN level or fold change detection.

Discussion In this study, it was observed that time after thawing had a greater effect on RNA integrity than time of incubation HistogramresponsesFigure 5 toof RNA5'-3' distancesdegradation for probe sets with different of liver tissue after surgical removal at either room temper- Histogram of 5'-3' distances for probe sets with dif- ature or 37°C (Figure 1). Freezing disrupts tissue struc- ferent responses to RNA degradation. 5'-3' distances ture, rendering the tissue highly sensitive to RNA for probe sets that are INV (grey) or DEC (black) are plotted degradation. In contrast, autolysis of fresh liver tissue as a fraction of the total number in each set. appeared to be a much slower process. To minimize the potential impact of RNA degradation on microarray data,

Page 7 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

Table 3: Effect of RIN and probe sensitivity to RNA degradation on assay sensitivity. The percent sensitivity for a false positive rate of 10% [and 95% Confidence Interval] is reported for liver selective probe sets that significantly decrease in response to RNA degradation (LS_DEC), all liver selective probe sets (LS_ALL), and liver selective probe sets that are invariant to RNA degradation (LS_INV).

Fold change RIN group LS_DEC LS_ALL LS_INV

1.5 9.5 97.2 [85.8, 99.7] 92.9 [88.2, 96.1] 100 [94.0, 100] 8 92.4 [78.7, 98.1] 92.0 [87.5, 95.2] 100 [93.6, 100] 7 72.7 [55.2, 85.9] 88.7 [82.0, 93.4] 100 [90.3, 100] 6 68.4 [51, 82.5] 88.3 [82.9, 92.4] 100 [84.5, 100] 2 9.5 99.2 [90.7, 100] 96.3 [93.3, 98.1] 100a 8 94.3 [82.6, 98.7] 94.7 [91.6, 96.8] 99.9 [80.3, 100] 7 84.2 [68.7, 93.6] 93.8 [89.4, 96.7] 100 [39.5, 100] 6 83.6 [68.3, 93.1] 93.3 [89.3, 96.0] 100 [79.7, 100] 4 9.5 99.0 [89.3, 100] 96.9 [94.5, 98.4] 100a 8 96.4 [85.8, 99.4] 95.7 [93.1, 97.5] 100a 7 87.8 [74.1, 95.4] 95.3b 100a 6 87.6 [74.2, 95.2] 94.6 [91.6, 96.7] 100a

aNo CI (for perfect discrimination, data cannot be fitted to a ROC curve) bValue interpolated from empirical data using Weibull cumulative discrimination function resected tissue should be sectioned and either flash frozen sets that are sensitive to sample integrity but not expressed or immersed in a tissue stabilization solution such as in rat liver. RNALater. Archived frozen tissue should be quickly dis- rupted and homogenized in denaturing solutions after A subtle but reproducible difference in the relationship removal from storage. Homogenization can be performed between RIN value and qualitative (electrophoretic trac- more rapidly if tissue is cut into smaller sections prior to ings) or other quantitative RNA metrics (28s/18s rRNA freezing. ratio, cRNA yield, and cRNA length) was observed between samples generated by different tissue handling Although ex vivo incubation of tissue for several hours had methods (Table 1 and Figure 2). The method of degrading little effect on RNA integrity, it did induce changes in the RNA may release or activate ribonucleases with different expression of ischemia-induced and early immediate specificities or differentially affect ribonuclease access to genes, as has been reported by others [13,16,17]. Many substrate. For example, ribonucleases 1 and 4 have differ- inflammatory response gene transcripts are inherently ent pH optima and substrate preferences for poly(C) and unstable as a mechanism to control cellular response to poly(U) [21]. Freeze/thaw may be the more relevant certain stimuli [18]. The increases in signal observed with method of inadvertent sample degradation associated 37°C incubation could result from de novo transcription with toxicogenomic studies. The correlations observed or stabilization of labile mRNAs through, for example, the here between RIN levels generated by F/T of liver tissue activation of MAPK or other signaling pathways [19]. and other sample quality metrics may not necessarily be Incubation of liver sections at 37°C also induced a selec- applicable for other mechanisms of RNA degradation tive decrease in a set of genes associated with immune (e.g., the introduction of exogenous RNase during han- function. Delay in sample processing has been observed dling) or for other tissues. For example, a threshold RIN to cause a decrease in the levels of selective transcripts in value of 7.8 has been recently proposed for optimal RNA blood cells [16]. Alternatively, this result could have reliability for analysis of human tumor samples on arisen from selective loss of an immune cell population in Affymetrix GeneChip arrays, where reliability was defined liver through diffusion or autolysis. as a 3'/5' Gapdh ratio threshold ≤ 1.25 [10]. In our study, only samples with RIN ≥ 9 had 3'/5' Gapdh ratios ≤ 1.25. A majority (≥ 75%) of the probe set signals identified as most sensitive to sample incubation at 37°C or after thaw- More interlaboratory studies are needed to evaluate the ing (see Figure 3) are expressed in whole tissue RNA prep- reproducibility of RIN and its correlation to performance arations from brain, kidney, and heart in addition to liver on multiple array formats before a RIN threshold can be (data not shown). Although differences in the kinetics of recommended as a component of best practices for micro- postmortem RNA degradation have been observed array data generation. The advantage to RIN as a metric is between tissue types [20], it is anticipated that signals for that it is an automated measurement made prior to per- most of these 347 probe sets would also be sensitive to forming expensive in vitro transcription (IVT) assays and sample integrity in these other tissues. Similarly designed array hybridizations. Although average cRNA length cor- studies using other tissues may identify additional probe related well with microarray sample quality in this study,

Page 8 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

this value is currently not an automated measurement fold was reduced for RNA samples with RIN values ≤ 7. that is calculated by Agilent Bioanalyzer software and The effect was greatest for probe sets most sensitive to needs to be estimated from electropherograms by the end sample integrity and was less pronounced for probe sets user. unselected for an effect of RNA degradation on signal level. The primary effect of RNA degradation on samples analyzed on microarrays was a decrease in the average length Conclusion of products that are reverse transcribed and amplified In this paper, we examined the effect of sample integrity using T7 polymerase. The multiple rounds of in vitro tran- on microarray performance through the use of samples scription that are used to generate samples from small with a progressive decrease in RNA quality that was amounts of RNA from biopsies or laser-captured micro- indexed using a sensitive automated metric of RNA degra- dissections also induce a decrease in transcript cRNA yield dation (RIN). We identified a RIN threshold beyond and length [1]. Amplifying target from small samples was which we observed a decrease in assay specificity and sen- associated with a loss of signal for gene transcripts with sitivity. The effect on assay performance could be linked to high GC content and with a greater number and length of a decrease in hybridization of target to probe sets that map predicted hairpin formations [22]. We did not observed more than 600 nt upstream of the transcription termini any difference in GC content between probe sets that were on their corresponding reference sequences. most and least sensitive to RNA degradation generated by F/T in rat liver (data not shown). Methods Animal studies A relatively small percentage (~4%) of probe sets that are Male Sprague Dawley rats were received from Harlan Lab- on RAE230A arrays and expressed in liver were found to oratories (Frederick, MD) at 6 weeks of age and accli- be similar in sensitivity to cRNA target length as the 5'- mated for 6 days. The rats received certified rodent diet probe sets for Gapdh or Actb. Of the probe sets with this #5002C (Purina Mills Inc.) ad lib and drinking water puri- sensitivity that also had verifiable transcription termini, fied by reverse osmosis. The animals were on a 12 hr light/ most (~75%) were located more than 600 nt upstream of dark cycle and euthanasia was performed within 4 to 6 hr the 3'-end of their target sequences. For the remaining after the start of the light cycle. Animal care and proce- DEC probe sets that were located within the designed dures were approved by the Institutional Animal Care and probe selection region, no other measure of probe set Use Committee at the US FDA. length or location was identified that was significantly different from INV probe sets and could explain the Sample generation enhanced sensitivity to RNA degradation. After euthanasia by carbon dioxide inhalation, whole livers were removed and placed in Petri dishes containing In comparisons of probe set level signal data from unde- sterile phosphate buffered saline (PBS). Livers were briefly graded (RIN 9.5) RNA with RNA of progressively decreas- rinsed with PBS to remove blood. Liver sections were pre- ing RNA integrity (RIN 8, 7, or 6), a substantial increase in pared by removing a 2 cm square section from the left lat- the rate of detection of false positives was observed when eral lobe and further sectioning it into 12–16 equal pieces. RIN values are ≤ 7. Comparisons of samples with different Each time course study used a single liver lobe from a RIN levels could occur in toxicogenomic studies where unique animal. For room temperature incubations, the treatment conditions have induced a degree of damage sections were placed in RNAlater (Ambion, Austin, TX) at and vehicle-treated control tissue is unaffected. Similar the end of each incubation period. Incubations at 37°C effects are possible in comparisons of results between sin- were performed in a water bath and terminated by addi- gle and multiple rounds of amplification. In one study, tion of RNAlater. After the samples were incubated in protocol method (one-cycle or two-cycle) was shown to RNAlater overnight, RNA was isolated using Qiagen Midi have a bigger effect on signal variance than tissue type kits (Qiagen, Valencia, CA). To generate samples in which (breast vs. cervix) [3] RNA was degraded after tissue thawing, liver sections were snap frozen in a dry ice/ethanol slurry and stored at - In our analysis of the effect of RNA integrity on assay sen- 70°C. Random tissue sections were thawed at room tem- sitivity, probe set level signals from both "control" and perature for various intervals. At the end of each incuba- "treated" samples were modeled from RNA with the same tion period, samples were homogenized in Qiagen RLT RIN. This design interrogates a decrease in sensitivity in buffer and processed following the Qiagen Midi kit proto- studies where RNA integrity is similar for all samples but col. RNA and cRNA yields were quantitated on a Nano- at lower than optimal levels because of tissue handling or Drop ND-1000 spectrophotometer (NanoDrop RNA isolation method. The accuracy of true positive Technologies, Wilmington, DE). All samples had 260/280 detection of in silico modeled changes of 1.5-, 2-, and 4- ratios ≥ 2.0. RNA integrity was characterized by measuring

Page 9 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

the 28s/18s rRNA ratio and RIN on an Agilent 2100 Bio- Probe set characterization analyzer (Santa Clara, CA). RIN values for the zero-time Probe sets that were significantly increased (INC), point samples were ≥ 9. decreased (DEC), or unchanged (INV) by degradation were identified using the F/T RIN group replicates. Probe cRNA was synthesized from 5 µg total RNA using Affyme- sets were identified as differentially expressed among the trix standard protocols for cDNA synthesis and an IVT four RIN classes (9.5, 8, 7, and 6) by applying a multivar- labeling kit from Enzo Life Sciences (Farmingdale, NY) for iate permutation test (SAM) to provide a median false dis- synthesis of biotin-labeled cRNA. Average cRNA lengths covery rate of 10% using BRB-Array Tools Version 3.5.0 were calculated from standard curves generated from run- developed by Dr. Richard Simon and Amy Peng Lam. ning the RNA 6000 Ladder (Ambion) on an Agilent 2100 From this list of 574 probe sets, subsets of INC and DEC Bioanalyzer and enabling smear analysis in the Agilent probe sets were identified that demonstrated a consist- 2100 Expert software. A region was defined for each smear ently increasing or decreasing monotonic trend in average and the region start size was manually aligned with the signal between RIN 9.5, 8, 7 and 6 groups. From the probe vertical point of symmetry for each electropherogram. The sets that were not identified as significantly changed region start sizes were automatically extrapolated from the between RIN classes by SAM, a subset of INV probe sets standard curve data obtained from the 25 nt marker peak, were identified with signal values that did not signifi- and the first 5 RNA ladder fragment peaks in the RNA cantly change as a function of RIN (signal coefficients of 6000 Nano Assay. The region end size and other values of variation (CV) < 0.01 across all 12 F/T samples) and did the smear analysis table were not used to determine the not show a monotonic trend in average signal as a func- median cRNA length. tion of RIN.

To calculate statistically significant changes in RNA met- Probe set location relative to the 5' and 3' ends of tran- rics between samples with RIN 9.5 and samples with RIN script reference sequences was calculated for DEC and INV < 9.5, one-way ANOVA with a Dunnett's post test was per- probe sets that could be mapped to a single mRNA refer- formed using GraphPad Prism version 3.00 for Windows ence sequence (RefSeq) containing a terminal polyA (GraphPad Software, San Diego, CA). sequence ≥ 10 nt. Mapping was defined as a 100% sequence match between the corresponding RefSeq and Microarray experiments the first (No. 1) and last (No. 11) perfect match (PM) Thirtytwo RNA samples from 7 independent time course probes in the 11 probe series that comprises each probe studies of RNA degradation by F/T or 37°C incubation set. The minimum contiguous sequence of a RefSeq that is were labeled as described above and run on Affymetrix targeted by all 11 probe pairs in a probe set is defined as RAE230A arrays. 15 µg of fragmented cRNA was hybrid- the target sequence (TargetSeq). ized per array. Probe set signals were calculated using the Affymetrix MAS5 algorithm from files scaled to a target Three distance metrics (3'-3' distance, 5'-3' distance, and signal value of 500. The microarray data is available in EBI 5'-5' distance) were determined for each probe set relative ArrayExpress under the accession number E-MEXP-1069 to its mapped position on the corresponding RefSeq. Ref- [23]. Seq lengths used for distance metrics excluded the length of the polyA sequence. The statistical significance of dis- Cluster analysis tance metric data between INV and DEC groups was cal- Log2 signal ratios were calculated for probe sets in each culated in unpaired two-tailed t-test comparisons using non-control sample relative to the average zero-time con- GraphPad Prism version 3.00 for Windows (GraphPad trol signal for each handling condition. Probe sets were Software, San Diego, CA). The statistical significance of identified that were either present in all control samples distance metric data between INV, DEC (5'-3' distance < (n = 8) and had log2 ratios ≤ -1 in at least 50% of non-con- 1000 nt), and DEC (5'-3' distance > 1000 nt) groups was trol 37°C or F/T samples or were present in all non-con- calculated by applying a Tukey's post-test comparison of a trol samples within a handling condition set and had log2 one-way ANOVA using GraphPad software. ratios ≥ 1 in at least 30% of non-control 37°C or F/T samples. A total of 347 probe sets met these criteria within F/ Effect of RNA degradation on false positive and false T or 37°C datasets. Hierarchical clustering was performed negative rates on this subset of probe sets for 28 non-control samples The effect of RNA degradation on the generation of false using Spotfire DecisionSite Functional Genomics software positives was analyzed by applying SAM at a median FDR (Spotfire, Inc., Somerville, MA), Pearson correlation for of 0.1 in two-sample comparisons of control liver RNA the similarity measure, and the unweighted average clus- that differed in RIN value using BRB-ArrayTools v3.5.0. tering method. Gene Ontology classification was per- Signals from the 3 F/T RIN 9.5 samples were compared to formed using DAVID [24]. F/T RIN 8, 7, or 6 sample sets. For this analysis, signals

Page 10 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

were calculated using MAS5 or PLIER in ArrayAssist v4.0 sequence; ROC, receiver operating characteristic; LS, liver- (Stratagene, La Jolla, CA) for each comparison group of 6. selective; KS, kidney-selective

Probe sets with tissue-selective expression were identified Competing interests from body map data generated on RAE230A arrays by YT holds stock in Affymetrix Inc. KLT, BAR, and PSP applying a tissue-selective index cutoff of 5, as previously receive collaborative research support from Affymetrix described [15]. Using this criterion, 292 probe sets with Inc. liver-selective signals (LS_ALL) and 188 probe sets with kidney-selective (KS) signals were identified. A subset of Authors' contributions 30 LS probe sets (LS_DEC) was identified that were signif- JR and KLT designed the study. BAR planned and con- icantly changed by RNA degradation by applying SAM at ducted the laboratory experiments. KLT, PSP, and YT ana- a median FDR of 0.2 and that had decreasing monotonic lyzed the data. The manuscript was drafted by KLT and trends in average log2 signal between the 4 F/T RIN groups edited by JDR. All authors approved the final manuscript. (9.5 to 6). From the remaining LS probe sets that did not show a statistically significant difference in signal level Additional material between RIN groups using SAM, a set of 33 invariant probe sets (LS_INV) was identified that had log2 signals with CV ≤ 0.01 and a lack of monotonic trend in average Additional file 1 Distance metrics for 3 probe sets that are sample quality controls on signal across the 4 RIN groups ordered by decreasing RNA Affymetrix RAE230A arrays. This table contains distance metrics and quality. average signal values for the endogenous control probe sets AFFX_Rat_GAPDH_5_at, AFFX_Rat_beta-actin_5_at, and To model the effect of liver degradation on the diagnostic AFFX_Rat_beta-actin_M_. accuracy of detecting 1.5-fold changes in expression, the Click here for file signal intensity from each F/T sample was used as the liver [http://www.biomedcentral.com/content/supplementary/1472- 6750-7-57-S1.pdf] component to derive in silico signal intensities for the tissue-selective probe sets used as analytes in each of the two Additional file 2 mixtures (Mix1 and Mix2) that comprise a mixed tissue Lists of probe sets that were selectively altered by ex vivo incubation at RNA reference material using the formulas in [15]. "Bio- 37°C. This file provides the probe set identifiers, gene symbols, UniGene logical" replicates of the modeled probe set signals in identifiers, and gene names for probe sets increased or decreased by 37°C Mix1 and Mix2 were calculated for RIN group replicates but not F/T incubation. through the use of a different, randomly assigned inde- Click here for file [http://www.biomedcentral.com/content/supplementary/1472- pendent preparation of pooled brain, kidney, and testis 6750-7-57-S2.pdf] RNA as the complex background (Batches 1–3 in [15]) for each replicate. To model the effect on diagnostic accuracy Additional file 3 of 2-fold and 4-fold changes, the proportion of liver RNA Distance metrics for individual probe sets. This file contains probe set signal in the mixtures was interchanged with the brain identifiers, gene names, RefSeq identifiers, polyA lengths, TargetSeq and testis RNA signal proportions, respectively. For each lengths, RefSeq lengths, TargetSeq/RefSeq lengths, 3'-3' distance, 5'-3' set of 3 RIN group replicates, a two sample t-test compar- distance, 5'-5' distance, and average log2 RIN 9.5 signal for probe sets ison of modeled Mix1 and Mix2 log signal values was classified by their sensitivity to RNA degradation (DEC or INV). 2 Click here for file performed to calculate a P value for each analyte. Subsets [http://www.biomedcentral.com/content/supplementary/1472- of LS_DEC, LS_INV, or LS_ALL analytes were used as true 6750-7-57-S3.pdf] positives and the set of KS analytes were used as true negatives in ROC plots. The web-based program JROCFIT [25] was used to calculate the sensitivity at a 10% false positive rate from ROC plots of fitted data using the fre- Acknowledgements quency of positives and negatives found in each of 36 The study was supported by US FDA intramural funding (KLT, PSP, BAR) exponentially spaced P value bins from 1 to 10-7 (format and through the contribution of microarrays by Affymetrix Inc. 3). Where the data could not be fitted to a curve using JROCFIT, the sensitivity was interpolated from empirical References 1. Luzzi V, Mahadevappa M, Raja R, Warrington JA, Watson MA: Accu- data using the Weibull cumulative distribution function rate and reproducible gene expression profiles from laser in Microsoft Excel. capture microdissection, transcript amplification, and high density oligonucleotide microarray analysis. J Mol Diagn 2003, 5:9-14. Abbreviations 2. Schoor O, Weinschenk T, Hennenlotter J, Corvin S, Stenzl A, Ram- RIN, RNA integrity number; F/T, freeze/thaw; nt, nucle- mensee HG, Stevanovic S: Moderate degradation does not preclude microarray analysis of small amounts of RNA. otide; TargetSeq, target sequence; RefSeq, reference Biotechniques 2003, 35:1192-201.

Page 11 of 12 (page number not for citation purposes) BMC Biotechnology 2007, 7:57 http://www.biomedcentral.com/1472-6750/7/57

3. Cope L, Hartman SM, Gohlmann HWH, Tiesman JP, Irizarry RA: 21. Zhao W, Kote-Jarai Z, van Santen Y, Hofsteenge J, Beintema JJ: Ribo- Analysis of Affymetrix GeneChip data using amplified RNA. nucleases from rat and bovine liver: purification, specificity Biotechniques 2006, 35:165-170. and structural characterization. Biochim Biophys Acta 1998, 4. Van Gelder RNV, von Zastrow ME, Yool A, Dement WC, Barchas JD, 1384:55-65. Eberwine JH: Amplified RNA synthesized from limited quanti- 22. van Haaften RI, Schroen B, Janssen BJ, van Erk A, Debets JJ, Smeets ties of heterogeneous cDNA. Proc Natl Acad Sci USA 1990, HJ, Smits JF, van den Wijngaard A, Pinto YM, Evelo CT: Biologically 87:1663-7. relevant effects of mRNA amplification on gene expression 5. Affymetrix Technical Note: Array Design and Performance profiles. BMC Bioinformatics 2006, 7:200. of the GeneChip Rat Expression Set 230 [http://www.affyme 23. European Bioinformatics Institute ArrayExpress data repos- trix.com/support/technical/technotes/rat230_technote.pdf] itory [http://www.ebi.ac.uk/arrayexpress/] 6. Hoffman EP, Awad T, Palma J, Webster T, Hubbell E, Warrington JA, 24. Dennis Jr G, Sherman BT, Hosack DA, Yang J, Baseler MW, Lane HC, Spira A, Wright G, Buckley J, Triche T, Davis R, Tibshirani R, Xiao W, Lempicki RA: DAVID: Database for Annotation, Visualization, Jones W, Tompkins R, West M: Expression profiling – best prac- and Integrated Discovery. Genome Biol 2003, 4:P3. tices for data generation and interpretation in clinical trials. 25. Eng J: ROC analysis: web-based calculator for ROC curves. Nat Rev Genetics 2004, 5:229-237. [http://www.jrocfit.org]. Baltimore: Johns Hopkins University 7. Dumur CI, Nasim S, Best AM, Archer KJ, Ladd AC, Mas VR, Wilkin- [updated 2006 May 17; cited 2006 September 7]. son DS, Garrett CT, Ferreira-Gonzalez A: Evaluation of quality- control criteria for microarray gene expression analysis. Clin Chem 2004, 50:1994-2002. 8. Archer KJ, Dumur CI, Joel SE, Ramakrishnan V: Assessing quality of hybridized RNA in Affymetrix GeneChip experiments using mixed-effects models. Biostatistics 2006, 7:198-212. 9. Jones L, Goldstein DR, Hughes G, Strand AD, Collin F, Dunnett SB, Kooperberg C, Aragaki A, Olson JM, Augood SJ, Faull RL, Luthi- Carter R, Moskvina V, Hodges AK: Assessment of the relationship between pre-chip and post-chip quality measures for Affymetrix GeneChip expression data. BMC Bioinformatics 2006, 7:211. 10. Copois V, Bibeau F, Bascoul-Mollevi C, Salvetat N, Chalbos P, Bareil C, Candeil L, Fraslon C, Conseiller E, Granci V, Maziere P, Kramar A, Ychou M, Pau B, Martineau P, Molina F, Del Rio M: Impact of RNA degradation on gene expression profiles: Assessment of different methods to reliably determine RNA quality. J Biotech- nol 2007, 127:549-559. 11. Schroeder A, Mueller O, Stocker S, Salowsky R, Leiber M, Gassmann M, Lightfoot S, Menzel W, Granzow M, Ragg T: The RIN: an RNA integrity number for assigning integrity values to RNA measurements. BMC Mol Bio 2006, 7:3. 12. Yan SF, Fujita T, Lu J, Okada K, Shan Zou Y, Mackman N, Pinsky DJ, Stern DM: Egr-1, a master switch coordinating upregulation of divergent gene families underlying ischemic stress. Nat Med 2000, 6:1355-1361. 13. Dash A, Maine IP, Varambally S, Shen R, Chinnaiyan AM, Rubin MA: Changes in differential gene expression because of warm ischemia time of radical prostatectomy specimens. Am J Pathol 2002, 161:1743-1748. 14. Debey S, Schoenbeck U, Hellmich M, Gathof BS, Pillai R, Zander T, Schultze JL: Comparison of different isolation techniques prior gene expression profiling of blood derived cells: impact on physiological responses, on overall expression and the role of different cell types. Pharmacogenomics J 2004, 4:193-207. 15. Thompson KL, Rosenzweig BA, Pine PS, Retief J, Turpaz Y, Afshari CA, Hamadeh HK, Damore MA, Boedigheimer M, Blomme E, Ciurlionis R, Waring JF, Fuscoe JC, Paules R, Tucker CJ, Fare T, Coffey EM, He Y, Collins PJ, Jarnagin K, Fujimoto S, Ganter B, Kiser G, Kay- sser-Kranich T, Sina J, Sistare FD: Use of a mixed tissue RNA design for performance assessments on multiple microarray formats. Nucleic Acids Res 2005, 33:e187. 16. Baechler EC, Batliwalla FM, Karypis G, Gaffney PM, Moser K, Ort- mann WA, Espe KJ, Balasubramanian S, Hughes KM, Chan JP, Begov- ich A, Chang SY, Gregersen PK, Behrens TW: Expression levels for Publish with BioMed Central and every many genes in human peripheral blood cells are highly sensitive to ex vivo incubation. Genes Immun 2004, 5:347-353. scientist can read your work free of charge 17. Miyatake Y, Ikeda H, Michimata R, Koizumi S, Ishizu A, Nishimura N, "BioMed Central will be the most significant development for Yoshiki T: Differential modulation of gene expression among disseminating the results of biomedical research in our lifetime." rat tissues with warm ischemia. Exp Mol Pathol 2004, 77:222-230. Sir Paul Nurse, Cancer Research UK 18. Chen CY, Shyu AB: AU-rich elements: characterization and Your research papers will be: importance in mRNA degradation. Trends Biochem Sci 1995, 20:465-70. available free of charge to the entire biomedical community 19. Wilusz CJ, Wormington M, Peltz SW: The cap-to-tail guide to peer reviewed and published immediately upon acceptance mRNA turnover. Nat Rev Mol Cell Biol 2001, 2:237-246. 20. Inoue H, Kimura A, Tuji T: Degradation profile of mRNA in a cited in PubMed and archived on PubMed Central dead rat body: basic semiquantification study. Forensic Sci Int yours — you keep the copyright 2002, 130:127-132. Submit your manuscript here: BioMedcentral http://www.biomedcentral.com/info/publishing_adv.asp

Page 12 of 12 (page number not for citation purposes) Ann Hematol (2009) 88:1177–1183 DOI 10.1007/s00277-009-0751-5

ORIGINAL ARTICLE

Prediction of qualitative outcome of oligonucleotide microarray hybridization by measurement of RNA integrity using the 2100 Bioanalyzer™ capillary electrophoresis system

Philipp Kiewe & Saskia Gueller & Martina Komor & Andrea Stroux & Eckhard Thiel & Wolf-Karsten Hofmann

Received: 6 April 2009 /Accepted: 29 April 2009 /Published online: 8 May 2009 # Springer-Verlag 2009

Abstract RNA quality is critical to achieve valid results determined optimal cut points for RIN at 7.15 and 8.05, in microarray experiments and to save resources. The respectively. Using the suggested cut points, RIN can RNA integrity number (RIN) can be measured with support the final decision whether a certain RNA sample minimal sample consumption by microfluidics-based is appropriate for successful microarray hybridization. capillary electrophoresis. To determine whether RIN can predict the qualitative outcome of microarray hybridiza- Keywords RNA quality . RIN . Capillary electrophoresis . tion, we measured RIN in total RNA samples from 484 Bioanalyzer . Oligonucleotide microarray. Present calls different experiments by the 2100 Bioanalyzer system and correlated with the percentage of present calls (%pc) of downstream oligonucleotide microarrays. The correla- Introduction tion coefficient for RNA and %pc in all 408 samples for which the bioanalyzer algorithm was able to produce an RNA oligonucleotide microarray platforms are increasingly RIN was 0.475 (p<0.05), ranging from 0.039 to 0.673 for used to create gene expression profiles of tissues involved different tissue- and assay-type subgroups. Multivariate in various medical conditions, particularly in hematological analysis found RIN to be the best predictor of microarray and oncological diseases. The derived data help to quality as assessed by %pc, outperforming the 28S to understand the biology, facilitate diagnosis, or predict 18S ratio. For a %pc threshold of 25% and 35%, we treatment response and prognosis of the disease studied. The quality of RNA recovery and sample processing is of utmost importance to achieve valid results and to save P. Kiewe (*) : E. Thiel : W.-K. Hofmann Department of Hematology, Oncology, and Transfusion Medicine, precious resources, particularly when limited amounts of Charité-University Hospital Benjamin Franklin, RNA are available. RNA preparations can be contaminated Hindenburgdamm 30, by DNA or protein, and they are constantly compromised 12203 Berlin, Germany by degradation. While moderate RNA degradation may still e-mail: [email protected] yield acceptable microarray results, extensively degraded S. Gueller : M. Komor samples should be excluded from analysis [1]. Conven- Department of Hematology, Oncology, Rheumatology, tional methods to assure RNA integrity include gel and Infectiology, University Hospital Frankfurt/Main, electrophoresis under denaturing conditions with determi- Theodor-Stern-Kai 7, 60596 Frankfurt am Main, Germany nation of the 28S and 18S ribosomal RNA band ratio and UV spectrometry determining the ratio of absorbance at A. Stroux 260 and 280 nm (optimal ratio 1.8–2.1). Major drawbacks Institute for Biostatistics and Clinical Epidemiology, of these methods are either large quantities of RNA Charité-University Hospital Benjamin Franklin, Hindenburgdamm 30, required for analysis, dependence on electrophoresis con- 12203 Berlin, Germany ditions, or the inability to detect DNA contamination. 1178 Ann Hematol (2009) 88:1177–1183

The Agilent 2100 Bioanalyzer™ (Agilent Technologies, 5. The scale factor is a global normalization constant Santa Clara, CA, USA) is a microfluidics-based platform based on the trimmed mean of probe set signals or based on capillary electrophoresis that can be used to average differences and is inversely related to array quantify as well as to quickly and reliably assess quality of brightness. RNA samples with minimal sample consumption [2]. In addition to the calculation of a 28S to 18S rRNA ratio, the Except for Bio B, all quality metrics are influenced by system software includes an algorithm to calculate an RNA cumulative errors. integrity number (RIN), a value between 1 and 10 in one- Most quality control measures are not entirely inde- decimal steps. Using RIN, sample integrity is determined pendent of each other. Percentage of present calls and 3′ by the entire electrophoretic trace of the RNA sample to 5′ ratio most significantly correlate with RNA sample instead of the 28S to 18S rRNA ratio alone. RIN can be quality. used as a standardized measure to correlate with results The aim of the present study was to see if RNA from subsequent microarray experiments and determine integrity measured by RIN can predict microarray thresholds for meaningful results [3]. quality. Using a large database of oligonucleotide micro- Quality of oligonucleotide microarray results can be array experiments including samples with different tissue estimated by several metrics provided by an Affymetrix origins as well as cell cultures, RIN was retrospectively software report file: correlated with the percentage of present calls in downstream microarrays. Other factors potentially corre- 1. Percentage of present calls (%pc), an array level lating with the percentage of present calls like 28S to summary of the results of a statistical function designed 18S ratio, RNA concentration, assay, and tissue type to predict the presence or absence of each gene were included in a multivariate linear regression model. transcript [4]. It can be used as a quality metric that is Cut point analysis for RIN predicting %pc below certain sensitive to any error source from RNA sampling to thresholds was performed to support decisions in future scanning and data extraction and is therefore cumula- experiments of whether a sample is suitable for success- tively influenced by all stages in the microarray ful microarray hybridization. process. Furthermore, different array classes, brightness, background measures, and detection algorithms greatly influence this quality metric by up to 40% [5]. Material and methods Different values have reportedly been used as a threshold for poor-quality assays. Finkelstein et al. Samples used 25% as a threshold for outliers, a value which is also recommended by the Tumor Analysis Best Four-hundred and eighty-four RNA preparations were Practices Working Group [6], whereas Weis et al. [7] electrophoretically analyzed using the Agilent 2100 Bio- found a value below 35% to correlate with poor-quality analyzer™ system. In 408 of these samples (84%), an RIN assays in their experiments. could be calculated using the Agilent 2100 Bioanalyzer 2. The average background is calculated from the 2% of Expert Software™; in 76 samples, the algorithm failed to probes with the weakest signal. It is an estimate of produce an RIN. This failure to produce a valid RIN was general nonspecific binding based on low-intensity mostly due to a missing or displaced lower marker, features across an array. extremely small amounts of RNA (<10 ng/ml), or peak 3. Bio B is a probe set designed to measure prelabeled shifts on the time axis despite optically immaculate bacterial nucleotides. It is the signal from internal electropherograms. prelabeled standards and measures the efficacy of The samples originated from various experiments hybridization, washing, and scanning. Bio B is free of performed in our microarray laboratory from 2002 to RNA, amplification, and labeling effects. 2005. RNA samples were prepared from cell lines, human 4. The 3′ to 5′ ratio of a housekeeping gene (e.g., mononuclear bone marrow cells, human CD34-selected glyceraldehyde-3-phosphate dehydrogenase (GAPDH) cells, murine hematopoietic cells, and mouse tissue or beta-actin) is a ratio of probe sets designed to detect (Table 1). RNA was extracted by standardized protocols the 3′ and 5′ regions of the messenger RNA transcript using either TRIzol™ Reagent (Invitrogen, Carlsbad, of a certain housekeeping gene and is reputed to detect CA,USA)ortheRNeasy™ kit (Qiagen, Valencia, CA, RNA degradation. This ratio is thought to indicate USA) according to the manufacturers’ guidelines. RNA RNA quality as well as the bias inherent in the concentration was measured with a NanoDrop™ spec- Affymetrix labeling assay and should be below a value trophotometer (Thermo Fisher Scientific, Wilmington, of 3 [8]. DE, USA). Ann Hematol (2009) 88:1177–1183 1179

Table 1 Sample origin and processing Statistics Sample type N (%) N with valid RIN (%) Correlations between RIN and %pc were calculated for all Nanogram Standard samples and separately for distinct subgroups: assay type (nanogram and standard assay), microarray type Total 484 (100) 408 (100) (HG-U133A, Hu6800 and MG_U74Av2), and tissue 233 (57) 175 (43) origin (cell culture, human mononuclear bone marrow Cell line 195 (40) 153 (38) cells, human CD34-selected cells, murine hematopoietic 42 (10) 111 (27) cells, and mouse tissue). Tissue origin was clustered Human bone marrow 129 (27) 114 (28) into cell line samples and samples of other origin for 94 (23) 20 (5) multivariate analysis. Human CD34 selected 118 (24) 104 (26) For all nonparametric correlations, Spearman’srank 97 (24) 7 (2) correlation coefficient was calculated. All given p values Mouse tissue 26 (5) 24 (6) correspond to two-sided t tests. p values <0.05 were 0 24 (6) considered significant. All factors significantly correlating Mouse hematopoietic cells 16 (3) 13 (3) with %pc in bivariate analysis were included in a forward 0 13 (3) and backward stepwise linear regression model. Cut points for RIN were determined by receiver Total numbers of RNA samples listed for different tissue origins and processing by standard assay or nanogram-scale assay operating characteristic (ROC) curves at a prespecified RIN RNA integrity number sensitivity of 0.8 with specificity determined as 1-false- positive rate on the horizontal axis at the curve intersection. Diagnostic accuracy was determined by measurement of the Oligonucleotide microarrays “area under the curve.” It is used as a measure to indicate how well the statistical test separates poor-RNA-quality All RNA samples regardless of bioanalyzer output were samples from good-RNA-quality samples, with an area of further processed and hybridized with microarrays specif- 0.9 to 1 representing an excellent test and an area of 0.8 to ic for the analyzed type of RNA (Affymetrix, Santa Clara, 0.9 representing a good test. CA, USA) according to the manufacturer’s guidelines. Commercially available statistical software was used Due to the limited RNA content in certain sample (SPSS for Windows, release 15.0). preparations (e.g., CD34-selected cells), a double in vitro transcription technique (nanogram-scale assay) was used in more than half of the experiments (n=233, see Table 1). Results To assay 50-ng total RNA, the standard Affymetrix target amplification protocol was modified by using the first- Calculation and distribution of RIN round complementary RNA (cRNA) product to generate double-stranded complementary DNA that was then used A valid RIN could be calculated in 408 samples. In 76 forasecondroundofinvitrotranscription for synthesis of samples, the algorithm was not able to produce an RIN. biotinylated cRNA [9]. Distribution of missing RINs to different tissues reflects Most experiments using human tissue (n=428) were their proportion within the whole sample set, but missing performed with the HG-U133A array (Affymetrix) and a RINs are overrepresented in the standard assay group smaller amount (n=7) with the Hu6800 array (Affymetrix), (61%). and experiments with mouse tissue (n=49) were performed Mean RIN for all experiments is 8.1 (range, 1.1–10.0). with the MG_U74Av2 mouse array (Affymetrix). For nanogram-scale assays, the mean RIN is 7.1 (range, After hybridization, the microarray was washed and 1.1–10) and for standard assays 9.6 (range, 2.7–10.0), stained using an Affymetrix fluidics station and was p<0.001. Mean RIN for cell line samples is 9.5 (range, 5.7– scanned with an argon-ion confocal laser with 488-nm 10.0) and for samples of other origin 7.3 (range, 1.1–10.0), emission and detection at 570 nm. Fluorescence intensity p<0.001. was normalized to the average fluorescence for the entire microarray. Distribution of %pc GeneChip image analysis was performed using the Microarray Analysis Suites 4.0.6 and 5.0 (Affymetrix) Mean percentage of present calls is 40% (range, 5–57%). including the array quality assessed by the percentage of Likewise, mean %pc for samples with missing RIN is 40% present calls. (range, 8–52%). For nanogram-scale assays, the mean %pc 1180 Ann Hematol (2009) 88:1177–1183

Fig. 1 Scatterplot diagram of all samples (n=408) showing correlation between RIN (x-axis) and %pc (y-axis). Overall Spearman’s rank Fig. 2 Scatterplot diagram including only quality data from human correlation coefficient is 0.475 (p<1×10−6). Note: RIN, RNA integrity mononuclear bone marrow cells processed by nanogram-scale assay number; %pc, percentage of present calls (n=94) showing correlation between RIN (x-axis) and %pc (y-axis). Overall Spearman’s rank correlation coefficient is 0.673 (<1×10−6). Note: RIN, RNA integrity number; %pc, percentage of present calls is 37% (range, 5–55%) and for standard assays 43% (range, 7–57%), p<0.001. Mean %pc for cell line samples is 44% yielded the highest correlation coefficients (Fig. 2), whereas (range, 21–54%) and for samples of other origin 38% only marginal correlations were seen in cell line prepara- (range, 5–57%), p<0.001. tions or CD34-selected cells. Calculations for murine samples are limited by very small numbers and have only Correlation between RIN and %pc been stated for completeness.

For the entire sample set, a statistically significant correla- Correlation between 28S to 18S ratio and %pc tion coefficient of 0.475 (p<1×10−6) was calculated (Fig. 1). Different coefficients were calculated for sample Data on 28S to 18S ratio were available for 453 samples. In subgroups (Table 2). The highest correlations between RIN 379 samples, both 28S to 18S ratio and RIN were available. and %pc were found in the nanogram-scale assay group, Correlation coefficient for 28S to 18S ratio and RIN is whereas samples processed with standard assays showed no 0.544. Correlation coefficient for 28S to 18S and %pc is significant correlation at all. Human bone marrow samples 0.258 (p<0.001).

Table 2 Correlation between RIN and %pc Group Number Mean RIN Mean %pc rp

All samples 408 8.1 40.0 0.475 <1×10-6 Nanogram-scale assay 233 7.1 37.1 0.541 <1×10-6 Standard assay 175 9.5 43.7 0.069 0.363 Cell lines 153 9.5 43.7 0.283 4×10-4 Nanogram-scale assay 42 8.7 40.8 0.291 0.061 Standard assay 111 9.7 44.8 0.075 0.436 All other tissues 255 7.3 37.7 0.578 <1×10-6 -6 Mean RIN and %pc values, Human bone marrow 114 5.9 30.2 0.606 <1×10 Spearman's rank correlation Nanogram-scale assay 94 5.3 29.3 0.673 <1×10-6 coefficients (r), and Standard assay 20 8.8 34.7 0.039 0.87 corresponding p values (significant values in bold script) are Human CD34 selected cells 104 8.1 43.6 0.232 0.018 given for all samples and Mouse tissue 24 9.0 45.5 −0.158 0.462 subgroups defined by sample Murine hematopoietic cells 13 9.4 42.7 0.085 0.781 origins and assay type Ann Hematol (2009) 88:1177–1183 1181

Poor-quality microarrays (%pc<25% and %pc<35%)

Of 44 experiments with a present call metric (%pc) below 25%, RIN could be calculated in 39 (10% of all samples). Median RIN was 2.9 (range, 1.8–10). Of 109 experiments with a %pc below 35%, RIN could be calculated in 93 (23% of all samples). Median RIN was 5.6 (range, 1.2–10). In ROC analysis including all 408 valid samples, a RIN of less or equal 7.15 predicted a %pc below 25% with the prespecified sensitivity of 0.8, a sensitivity of 0.7, and a diagnostic accuracy of 0.8 (Fig. 3). A RIN of less or equal 8.05 predicted a %pc below 35% with the prespecified sensitivity of 0.8, a sensitivity of 0.73, and a diagnostic accuracy of 0.84 (Fig. 4).

Discussion Fig. 3 Receiver operating curve plotting sensitivity (y-axis) and 1− specificity (x-axis) for a cutoff percentage of present calls of 0.25. A The presented systematic analysis of RNA quality and RIN of less or equal 7.15 predicted a %pc below 25% with the corresponding microarray quality metrics demonstrates a prespecified sensitivity of 0.8, a specificity of 0.7, and a diagnostic linear correlation between RNA integrity and the percent- accuracy of 0.8 age of present calls, an important quality metric in microarray experiments using an Affymetrix platform with Influence of RNA quantity on RIN or %pc “perfect match” and “mismatch” hybridization probes. With an overall coefficient of 0.475, the observed correlation The mean RNA concentration of all samples is 1,229 ng/μl seems only moderate; however, in multivariate analysis, (SD 1,804). Mean RNA concentration for nanogram-scale RIN was the most powerful predictor of microarray quality, assays is 445 ng/µl (SD 857) and for standard assays particularly in comparison with the 28S to 18S ratio, a 2,147 ng/μl (SD 2,158), p<0.001. For all samples, the correlation coefficient of RNA concentration with RIN is 0.503 (p<0.001), and the correlation of RNA concentration with %pc is 0.316 (p< 0.001). The inclusion of RNA concentration as control variable in a partial correlation analysis with RIN and %pc for all samples yielded a coefficient of 0.54 indicating an even better prediction with the consideration of RNA concentration.

Linear regression model

Several factors which significantly correlated with %pc in bivariate analysis including RNA concentration, 28S to 18S ratio, tissue type (cell line or mixed tissue), assay type (standard or nanogram-scale assay), and RIN were included into a multiple linear regression model with forward and backward selection. 28S to 18S ratio and RIN remained the only predictive factors with significant prediction of %pc. Standardized Fig. 4 Receiver operating curve plotting sensitivity (y-axis) and 1− coefficient (beta) in stepwise analysis was larger for RIN specificity (x-axis) for a cutoff percentage of present calls of 0.35. A −13 RIN of less or equal 8.05 predicted a %pc below 35% with the (0.378, p=9×10 ) than for 28S to 18S ratio (0.141, prespecified sensitivity of 0.8, a specificity of 0.73, and a diagnostic p=0.006). accuracy of 0.84 1182 Ann Hematol (2009) 88:1177–1183 commonly used RNA quality metric. This is in line with a Although RNA quantity, tissue origin, and assay type study by Copois et al. [10] who compared different may interact with RIN and %pc, they are not independent methods of RNA assessment—degradometer software predictors of microarray quality as seen in the linear [11], 28S to 18S ratio, and an in-house quality scale—with regression model. RNA quantity has been shown to further array quality assessed by determination of the 3′ to 5′ ratio increase the correlation of RIN and %pc in partial of GAPDH and a clustering analysis of full array correlation analysis and should therefore rather be consid- expression (“dispersion tree”). A similar conclusion was ered as an additional factor in quality prediction, however, drawn from a small study comparing RNA quality of 24 of marginal importance. frozen breast cancer samples assessed by RIN, visual Despite a substantial correlation, the dispersion of inspection of the capillary electrophoretic trace, and the coordinates on the correlation curve seems rather wide. 28S to 18S ratio [12]. Jahn et al. assessed bacterial RNA This is certainly due to a large potential for errors occurring quality by RIN and found it to be critical for obtaining at various stages in the experiment from RNA level to meaningful gene expression data. In their study, RIN values microarray data analysis. RNA contamination and degra- below 7.0 resulted in high variation and loss of statistical dation can occur at any of the steps following electropho- significance when gene expression was analyzed by retic analysis; other reactions like biotinylation or quantitative real-time polymerase chain reaction [13]. fragmentation may confound analytic quality and last but Despite the large number of samples included in our not least array hybridization, scanning, and microarray model, analysis was confounded to a certain degree by a manufacture are potential error sources for impaired array large proportion of cell line experiments yielding large quality reflected by the percentage of present calls [16]. RNA quantities with a high degree of purity. In fact, for the Thus, the assessment of sample RNA quality merely 153 cell line experiments with valid RIN, mean RNA provides a snapshot at the beginning of the process, quantity was 2,129 ng/μl, and mean RIN was 9.5 (54% of whereas the percentage of present calls is a metric of array samples with RIN 10) compared with a mean RNA quality incorporating the whole procedure of gene expres- concentration of 1,214 ng/μl and mean RIN of 8.1 (30% sion analysis. of samples with RIN 10) in the whole sample set. Likewise, Therefore, our aim was not to provide an exact cell line experiments yielded a higher %pc (mean 43.7) numeric prediction of the percentage of present calls, compared with samples of other origin (mean 37.7). which is hardly needed in microarray analysis. The larger Moreover, the different assay types are difficult to compare benefit of quality prediction by RIN is the determination due to larger amounts of RNA used in standard assays of a cut point. Samples with RIN values underneath that compared with nanogram-scale assays. It has previously been cut point are not expected to yield meaningful gene array shown that in small samples the amount of RNA entered into results and, in practice, are not to be further processed. the experiment correlates with the percentage of present calls This helps to save valuable resources and improve the and other quality metrics [14]. This was reflected by a higher overall validity of results. In the literature, microarrays RINandagreater%pcinsamples processed with standard with a percentage of present calls below 25% or 35% are assays. Another parameter potentially influencing microarray usually regarded as poor-quality arrays. Using both quality is the cRNA yield further down the line of sample measures as a threshold, we found optimal cut points processing. However, in their extensive analysis of interlabor- for RIN underneath 7.15 and 8.05, respectively. With an atory reproducibility of microarray experiments, Kohlmann acceptable accuracy of 0.8 and 0.84, 11% and 30% of et al. found no obvious correlation between the cRNA yield samples, respectively, would be left below the cut point and microarray quality, concluding that multiple variables and should be exempted from further analysis. would have to be factored into a conclusion on whether a We have estimated that, depending on the number of sample is suitable for microarray hybridization [15]. samples processed on one bioanalyzer chip, RIN determi- Due to a more even distribution of values on the RIN nation by capillary electrophoresis would be cost-effective scale, we found samples derived from heterogeneous tissue even if only 0.3% to 4% of samples were sorted out before with a higher degree of contamination to yield higher hybridization onto microarrays. correlation coefficients compared with samples derived To conclude, we propose that RIN may be routinely used from cell lines. For example, considering only the subgroup for quality prediction in microarray experiments on an of samples derived from human mononuclear bone marrow Affymetrix platform utilizing “perfect match” and “mis- cells and analyzed with the nanogram-scale assay, we found match” hybridization probes. The correlation with the a good correlation coefficient of 0.673. This indicates that percentage of present calls is superior to that seen with the use of RIN to sort out poor-quality samples may be the 28S to 18S ratio. Depending on the threshold for %pc, more valuable in those samples of heterogeneous origin samples with an RIN below 7.15 or 8.05 may be reliably with a greater potential for contamination. excluded from further microarray hybridization. Ann Hematol (2009) 88:1177–1183 1183

References 9. Hofmann WK, de Vos S, Komor M, Hoelzer D, Wachsman W, Koeffler HP (2002) Characterization of gene expression of CD34 + cells from normal and myelodysplastic bone marrow. Blood 1. Schoor O, Weinschenk T, Hennenlotter J, Corvin S, Stenzl A, 100:3553–3560. doi:10.1182/blood.V100.10.3553 Rammensee HG, Stevanović S (2003) Moderate degradation does 10. Copois V, Bibeau F, Bascoul-Mollevi C, Salvetat N, Chalbos P, not preclude microarray analysis of small amounts of RNA. Bareil C, Candeil L, Fraslon C, Conseiller E, Granci V, Mazière P, Biotechniques 35(6):1192–1201 Kramar A, Ychou M, Pau B, Martineau P, Molina F, Del Rio M 2. Mueller O, Hahnenberger K, Dittmann M, Yee H, Dubrow R, (2007) Impact of RNA degradation on gene expression profiles: Nagle R, Ilsley D (2000) A microfluidic system for high-speed assessment of different methods to reliably determine RNA reproducible DNA sizing and quantitation. Electrophoresis quality. J Biotechnol 127:549–559. doi:10.1016/j.jbiotec. 21:128–134. doi:10.1002/(SICI)1522-2683(20000101)21:1<128:: 2006.07.032 AID-ELPS128>3.0.CO;2-M 11. Auer H, Lyianarachchi S, Newsom D, Klisovic MI, Marcucci G, 3. Schroeder A, Mueller O, Stocker S, Salowsky R, Leiber M, Kornacker K (2003) Chipping away at the chip bias: RNA Gassmann M, Lightfoot S, Menzel W, Granzow M, Ragg T degradation in microarray analysis. Nat Genet 35:292–293. (2006) The RIN: an RNA integrity number for assigning integrity doi:10.1038/ng1203-292 values to RNA measurements. BMC Mol Biol 7:3. doi:10.1186/ 12. Strand C, Enell J, Hedenfalk I, Fernö M (2007) RNA quality in 1471-2199-7-3 frozen breast cancer samples and the influence on gene expression 4. Liu WM, Mei R, Di X, Ryder TB, Hubbell E, Dee S, Webster TA, analysis—a comparison of three evaluation methods using micro- Harrington CA, Ho MH, Baid J, Smeekens SP (2002) Analysis of capillary electrophoresis traces. BMC Mol Biol 8:38. doi:10.1186/ high density expression microarrays with signed-rank call algo- 1471-2199-8-38 rithms. Bioinformatics 18:1593–1599. doi:10.1093/bioinformat 13. Jahn CE, Charkowski AO, Willis DK (2008) Evaluation of ics/18.12.1593 isolation methods and RNA integrity for bacterial RNA quantita- 5. Finkelstein DB (2005) Trends in the quality of data from 5168 tion. J Microbiol Methods 75:318–324. doi:10.1016/j.mimet. oligonucleotide microarrays from a single facility. J Biomol Tech 2008.07.004 16:143–153 14. McClintick JN, Jerome RE, Nicholson CR, Crabb DW, Edenberg 6. Tumor Analysis Best Practices Working Group (2004) Expression HJ (2003) Reproducibility of oligonucleotide arrays using small profiling—best practices for data generation and interpretation in samples. BMC Genomics 4:4. doi:10.1186/1471-2164-4-4 clinical trials. Nat Rev Genet 5:229–237. doi:10.1038/nrg1297 15. Kohlmann A, Haschke-Becher E, Wimmer B, Huber- 7. Weis S, Llenos IC, Dulay JR, Elashoff M, Martínez-Murillo F, Wechselberger A, Meyer-Monard S, Huxol H, Siegler U, Rossier Miller CL (2007) Quality control for microarray analysis of M, Matthes T, Rebsamen M, Chiappe A, Diemand A, Rauhut S, human brain samples: the impact of postmortem factors, RNA Johnson A, Liu WM, Williams PM, Wieczorek L, Haferlach T characteristics, and histopathology. J Neurosci Methods 165:198– (2008) Intraplatform reproducibility and technical precision of 209. doi:10.1016/j.jneumeth.2007.06.001 gene expression profiling in 4 laboratories investigating 160 8. Staal FJ, Cario G, Cazzaniga G, Haferlach T, Heuser M, Hofmann leukemia samples: the DACH study. Clin Chem 54:1705–1715. WK, Mills K, Schrappe M, Stanulla M, Wingen LU, van Dongen doi:10.1373/clinchem.2008.108506 JJ, Schlegelberger B (2006) Consensus guidelines for microarray 16. Schuchhardt J, Beule D, Malik A, Wolski E, Eickhoff H, gene expression analyses in leukemia from three European Lehrach H, Herzel H (2000) Normalization strategies for leukemia networks. Leukemia 20:1385–1392. doi:10.1038/sj. cDNA microarrays. Nucleic Acids Res 28:E47. doi:10.1093/ leu.2404274 nar/28.10.e47 Molecular Aspects of Medicine 27 (2006) 126–139 www.elsevier.com/locate/mam

Review RNA integrity and the eﬀect on the real-time qRT-PCR performance

Simone Fleige a, Michael W. Pfaﬄ a,b,*

a Physiology Weihenstephan, Center of Life and Food Sciences (ZIEL), Technical University of Munich, 85350 Freising, Germany b TATAA Biocenter Germany, Freising-Weihenstephan, Germany

Abstract

The assessment of RNA integrity is a critical first step in obtaining meaningful gene expression data. Working with low-quality RNA may strongly compromise the experimental results of downstream applications which are often labour-intensive, time-consuming, and highly expensive. Using intact RNA is a key element for the successful application of modern molecular biological methods, like qRT-PCR or micro-array analysis. To verify RNA quality now- adays commercially available automated capillary-electrophoresis systems are available which are on the way to become the standard in RNA quality assessment. Profiles generated yield information on RNA concentration, allow a visual inspection of RNA integrity, and generate approximated ratios between the mass of ribosomal sub-units. In this review, the importance of RNA quality for the qRT-PCR was analyzed by determining the RNA quality of different bovine tissues and cell culture. Independent analysis systems are described and compared (OD measurement, NanoDrop, Bioanalyzer 2100 and Experion). Advantage and disadvantages of RNA quantity and quality assessment are shown in performed applications of various tissues and cell cultures. Further the comparison and correlation between the total RNA integrity on PCR performance as well as on PCR efficiency is described. On the basis of the derived results we can argue that qRT-PCR performance is affected by the RNA integrity and PCR efficiency in general is not affected by the RNA integrity. We can recommend a RIN higher than five as

* Corresponding author. Present address: Physiology Weihenstephan, Center of Life and Food Sciences (ZIEL), Technical University of Munich, 85350 Freising, Germany. Tel.: +49 8161 71 3511; fax: +49 8161 71 4204. E-mail address: michael.pfaﬄ@wzw.tum.de (M.W. Pfaﬄ).

0098-2997/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.mam.2005.12.003 S. Fleige, M.W. Pfaﬄ / Molecular Aspects of Medicine 27 (2006) 126–139 127 good total RNA quality and higher than eight as perfect total RNA for downstream application. 2005 Elsevier Ltd. All rights reserved.

Keywords: RNA quality; RNA quantity; qRT-PCR; 2100 Bioanalyzer; Experion; Lab-on-chip

Contents

1. The particular importance of RNA quality ...... 127 1.1. Introduction ...... 127 1.2. RNA extraction ...... 128 1.3. RNA quantity and quality assessment ...... 129 2. Integrity of RNA and its effect on real-time qRT-PCR ...... 133 2.1. General aspects...... 133 2.2. Effect on the mRNA quantification ...... 133 2.3. Association between RNA quality and length of the amplified product . . . 136 3. Conclusion ...... 137 References ...... 137

1. The particular importance of RNA quality

1.1. Introduction

The accuracy of gene expression evaluation is recognised to be inﬂuenced by the quantity and quality of starting RNA. Purity and integrity of RNA are critical elements for the overall success of RNA-based analyses. Starting with low quality RNA may strongly compromise the results of downstream applications which are often labour-intensive, time-consuming and highly expensive (Raeymarkers, 1993; Imbeaud et al., 2005). It is preferable to use high-quality intact RNA as a starting point in molecular biological as well as in diagnostic applications. Especially in quantitative RT-PCR, micro-arrays, ribonuclease-protection-assay, in situ hybridization, northern blot analysis, RNA mapping, in vitro translation, cDNA library construction and any kind of array applications the integrity of the used total RNA should be checked. Especially in clinical application with unique or limited tissue material, e.g. obtained after surgery, a reliable RNA quality analysis is necessary (Bustin and Nolan, 2004b). As a consequence, several steps during tissue handling have to be carefully controlled in order to preserve the quality and integrity of the RNA material. It is well known that RNA is sensitive to degradation by postmortem processes and inadequate sample handling or storage (Perez-Novo et al., 2005). Thus the competency to quickly assess RNA quality using minor amounts has become increasingly important as the following measures of mRNA transcripts have become more expensive and more comprehensive. 128 S. Fleige, M.W. Pfaﬄ / Molecular Aspects of Medicine 27 (2006) 126–139

1.2. RNA extraction

The quality and quality of purified RNA is variable and after the extraction during long storage rather unstable (Bustin et al., 2005). Especially long mRNA fragments up to 10 kb are very sensitive to degradation. This can happen through cleavage of RNAses introduced by handling with RNA samples. The most obvious problem concerns the degradation of the RNA and this is best addressed by insisting that every RNA preparation is rigorously assessed for quality and quantity. The extraction and purification procedure of total RNA must fulfill the following criteria (Bustin and Nolan, 2004b; Pfaffl, 2005a):

• free of protein (absorbance 260 nm/280 nm); • free of genomic DNA; • should be undegraded (28S:18S ratio should be roughly between 1.8 and 2.0, with low amount of short fragments); • free of enzymatic inhibitors for RT and PCR reaction, which is strongly dependent on the puriﬁcation and clean-up methods; • free of any substances which complex essential reaction co-factors, like Mg2+ or Mn2+; • free of nucleases for extended storage;

There are a substantial quantity of problems that affect reproducibility, and hence the relevance of results. The source of RNA, sampling techniques (biopsy material, single cell sampling, laser micro-dissection) as well as RNA isolation techniques (either total RNA or poly-adenylated RNA isolation techniques) often vary significantly between processing laboratories (Bustin and Nolan, 2004b; Pfaffl, 2004). The RNA quality can be different between two extraction methods, e.g. performed by hand or by an automatic extraction system. The isolated total cellular RNA with the liquid extraction, e.g. Trizol (Roche Diagnostics, Germany) or TriFast (peqlab, Germany), has different RNA quality, whereas only the type of homogenization is changed (Fleige and Pfaffl, 2006). Due to its inherent susceptibility to ubiquitous RNases and its chemical instability, RNA is readily endangered by base- or enzyme-catalyzed hydrolysis. Researchers must take into account a variety of factors, which influence their ability to obtain high-quality RNA that is free of contamination such as RNases, proteins and genomic DNA. These factors include yield variations, processing requirements, and sample availability of different cells or tissues. The best RNA yield is obtained from tissue that has been diced into small fragments with a scalpel prior to being frozen by submerging in liquid nitrogen. The samples must be homogenized using a bead mill or a mechanical homogenizer (Bustin and Nolan, 2004b). Further problem may arise in the case of research on human or animal tissue sampling techniques and the time dependency until the tissue is stored safely in RNase inhibitors or RNA-later (Ambion, USA). It is often very challenging to decrease this sampling time to a minimum within the framework of clinical routine procedures, or in animal experiments during a slaughtering process. The RNA quality may also be S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 129 impaired in samples stored for a long time or under sub-optimal conditions (Schoor et al., 2003).

1.3. RNA quantity and quality assessment

Conventional methods are often not sensitive enough, not specific for single- stranded RNA, and disposed to interferences from contaminants present in the sample (Imbeaud et al., 2005). The assessment of RNA integrity can do by various methods: the classical gel OD measurement, modern OD measurement via Nano- Drop, old fashioned denaturating agarose gel-electrophoresis or with high innovative lab-on-chip technologies like Bioanalyzer 2100 (Agilent Technologies, USA) and Experion (Bio-Rad Laboratories, USA). Quantity and quality assessment using a UV/VIS spectrophotometer should be performed at multiple wave lengths at 240 nm (background absorption and possible contaminations), 260 nm (specific for nucleic acids), 280 nm (specific for proteins), and 320 nm (background absorption and possible contaminations). On basis of the OD 260 the quantity, and by the ratio of the optical density (OD) of OD 260/280 the quality, OD 260/240 or OD 260/320 the purity and the extraction performance can be verified. An OD 260/280 ratio greater than 1.8 is usually considered an acceptable indicator of good RNA quality (Sambrook et al., 1989; Manchester, 1996). By the presence of genomic DNA the OD 260 measurement can compromised and leading to over-estimation of the actual and real RNA concentration. Further the used buffer and high salt concentrations will interfere with the result of the optical measurement and therefore the calculated RNA concentrations might be over- or under-estimated (own unpublished results). The accuracy of the OD 260/A280 method has been questioned, with a value of 1.8 corresponding to only 40% RNA, with the remainder accounted for by protein (Bustin and Nolan, 2004b). More modern spectrometric methods, like the NanoDrop (ND-3300, NanoDrop Technologies, USA) in combination with RNA RiboGreen dye (Molecular Probes, Invitrogen, USA) can be used as an UV/VIS spectrophotometer for ultra sensitive quantification of RNA. A major advantage of the system is the very low sample consumption of 1–2 ll, which is especially important when using precious materials like human biopsy or laser dissected samples. Since the sample is not contained in a secondary vessel, the sample directly wets the system optics, reducing the variations and contaminations resulting from changing or repositioning the cuvettes. Further the ND-3300 measure a spectra of your sample covering 400–750 nm, giving you more information about the RNA integrity and other chemical contamination or the extracted RNA (ND-3300 user manual V2.5, NanoDrop Technologies, USA). An additional check involves gel electrophoresis with RNA either stained with SYBR Green dye (Molecular Probes) or the less sensitive ethidium bromide (Bustin and Nolan, 2004b). But the assessment of RNA integrity by inspection of the 18S and 28S ribosomal RNA bands using denaturating gel electrophoresis is a cumber- some, low-throughput method and requires significant amounts of precious RNA (Bustin and Nolan, 2004a). Using the RiboGreen (Molecular Probes) reagent, the detection as little as 1 ng RNA/ml is possible, and can be measured reproducible. 130 S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139

In contrast to UV absorbance measurements at 260 nm, where proteins and free ribonucleotides in the mixture interfere with accurate quantitation, the RiboGreen reagent only measures polymeric nucleic acids (Jones et al., 1998; LePecq and Pao- letti, 1966; Karsten and Wollenberger, 1977). Today high innovative lab-on-chip technologies like micro-fluidic capillary electrophoresis were used to do RNA quality and quantity assessments. Certainly, in terms of routinely analyzing large numbers of RNA preparations, it is by far the most convenient and objective way of assessing the quality of RNA. This method has become widely used, particularly in the gene expression profiling platforms (Bus- tin, 2002; Lightfood, 2002; Mueller et al., 2000). The Agilent 2100 Bioanalyzer (Agilent Technologies, USA) and the Expe- rion (Bio-Rad Laboratories, USA) provide a framework for the standardization of RNA quality control. Therefore RNA samples are electrophoretical separated on a micro-fabricated chip and subsequently detected via laser induced fluorescence detection. It requires only a very small amount of RNA sample down to 200 pg total RNA. The use of a RNA ladder as a mass and size standard during electrophoresis allows the estimation of the RNA band sizes. Integrity of the RNA may be assessed by visualization of the 18S and 28S ribosomal RNA bands. An elevated threshold baseline and a decreased 28S:18S ratio, both are indicative of degradation (Mueller et al., 2004). The intact RNA preparation (Fig. 1) shows high 18S and 28S rRNA peaks as well as a small amount of 5S RNA. Degradation of the RNA sample

Fig. 1. Chromatograms of micro-capillary electrophoresis from RNA samples showing different degrees of degradation. A typical electropherogram of high-quality RNA (solid black line, RIN = 7.5) include a clearly visible 28/18S rRNA peak ratio and a small 5S RNA. Partially degraded sample (thin grey line; RIN = 4.5) was indicated by a shift in the electropherogram to shorter fragment sizes and produce a decrease in fluorescence signal as dye intercalation sites are destroyed. S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 131 produces a shift in the RNA size distribution toward smaller fragments and a decrease in fluorescence signal as dye intercalation sites are destroyed. The 28S/ 18S ratios are automatically generated by the both software applications in Experion and Bioanalyzer 2100. The RNA measurement using the lab-on-chip technology appears stable and relatively uninfluenced by contamination. RNA from tissue samples are typically classified according to the observation that the 28S rRNA peak area should be approximately twice the quantity of that of the 18S in total RNA samples for the mRNA quality to be acceptable (Sambrook and Russel, 2001). In general a 2.0 ribosomal ratio is regarded as perfect (Sambrook and Russel, 2001; Mueller et al., 2004). But in practice this value hardly is obtained. The 28S/18S ratio may reflect unspecific damage to the RNA, including sample mishandling, postmortem degradation, massive apoptosis or necrosis, but it can reflect specific regulatory processes or external factors within the living cells. As it is apparent from a review of the literature, the standard 28S/18S rRNA ratio of a 2.0 is difficult to meet, especially for RNA derived from clinical samples, and it now appears that the relationship between the rRNA electropherogram profile and mRNA integrity is up to now unclear (Monstein et al., 1995). Furthermore, the generated ribosomal ratios are dependent on the used capillary-electrophoresis. In an intern study comparing Bioanalyzer 2100 (Agilent Technologies) with Experion (Bio-Rad) both capillary-electrophoreses systems showed differences in the generated ratio value, sensitivity, variation, and reproducibility (data not shown). Nevertheless, both platform showed more or less the same results. However, it is unable to locate the original data for this commonly accepted premise. Based on structural differences alone, it might be expected that the in situ stability of mRNA differs from rRNA. Certainly, RNases will eventually result in the loss of both components, although there are other factors under which in situ rRNA will be completely degraded but mRNA remains intact (Mayne et al., 1999). Santiago et al. (1986) described that the mRNA integrity correspond more closely to the 28S than to the 18S integrity. This would mean that with increased length, there is a greater statistical chance of cleavage. Contrary to this assumption, Miller et al. (2004) expected that the 18S integrity correlated better than 28S with the mRNA, as the length of 18S is more closely aligned with that of the average mRNA. From our findings we can confirm the mRNA quality is more related to the 28S rRNA, which is often much faster degraded than the 18S. In a time dependent total RNA degradation via UV light the 28S rRNA disappeared very quickly (data not shown). Therefore the 28S/18S ratio has to be assessed for every single experiment and this is regarded as inadequate for the assessment of the quality (Marx, 2004). Altogether, it appears that the total RNA with lower rRNA ratios is not necessarily of poor quality especially if no degradation products can observe in the electrophoretic trace (Imbeaud et al., 2005). A new tool for RNA quality assessment is the RNA Integrity Number (RIN, developed by Agilent Technologies) for the lab-on-chip capillary gel-electrophoresis used in the Bioanalyzer 2100 (Mueller et al., 2004). This tool is based on a neuronal network which determines the RIN number from the shape of the curve in the 132 S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 electropherogram (Fig. 1). The software and the algorithm allows the classification of total RNA on a numbering system from 1 to 10, with 1 being the most degraded profile and 10 being the most intact. In this way, interpretation of an electropherogram is facilitated, comparison of samples is enabled and repeatability of experiments. The verification of the RNA integrity before use in different applications permits to compare experiments and classify the significance of results (Mueller et al., 2004; Imbeaud et al., 2005). The dependence of the RNA integrity on various calf tissue samples, white blood cells and four cell lines was determined (Fleige and Pfaffl, 2006). As shown (Table 1) for solid tissues the average RIN is between 6 and 8. Tissues or organs with high content of connecting tissue, e.g. in the gastrointestinal tract like rumen, omasum and jejunum, underlie high RNA degradation through the sampling and extraction procedure and show great RIN variations. The reason for this variability might be the solid and tough structure of the tissues, e.g. connecting or fatty tissue, the RNase enzymatic activity and problems during tissue sampling and storage. Furthermore tissues from the gastrointestinal tract have been washed in physical saline solution to get rid of any disturbing gut substances. Thus the physiological constitution of the tissue, the time and management of tissue sampling has a bearing on the degradation level of RNA. In contrary single cells like white blood cells (WBC) or cell derived from cell lines have higher RIN. Cell sampling and RNA extraction is much faster and easier, because cells are better accessible and were not kept in any sub- optimal conditions. Therefore RNA integrity based on the RIN classification is

Table 1 Average RNA integrity numbers (RIN) of various bovine tissues and cell lines analyzes with the bioanalyzer 2100 (Agilent Technologies) Tissue Quality metrics Mean Std. dev. n Liver 6.49 ±0.86 28 Heart 6.03 ±1.19 23 Spleen 7.28 ±0.60 17 Lung 6.55 ±0.67 22 Rumen 4.70 ±2.81 23 Reticulum 5.47 ±1.29 21 Omasum 6.64 ±1.87 18 Abomasum 7.30 ±0.86 17 Ileum 7.35 ±1.53 17 Jejunum 4.56 ±2.13 20 Colon 7.52 ±0.62 19 Caecum 7.28 ±0.86 16 Lymph node 6.93 ±0.65 26 Kidney cell 8.87 ±0.32 3 Corpus luteum 9.62 ±0.32 5 Granulosa cell 8.38 ±0.41 5 Oviduct 9.40 ±0.29 5 WBC 9.36 ±0.13 5 S. Fleige, M.W. Pfaﬄ / Molecular Aspects of Medicine 27 (2006) 126–139 133 much better and lay around RIN 9. The importance of isolation technique for a good RNA quality is shown in detail in Fleige and Pfaﬄ (2006).

2. Integrity of RNA and its eﬀect on real-time qRT-PCR

2.1. General aspects

For a sensitive and reliable quantitative measurement of low abundant mRNA gene expression real-time quantitative reverse-transcription polymerase-chain-reaction (qRT-PCR) reaction is the method of choice. qRT-PCR shows high sensitivity, good reproducibility and a wide quantification range (Bar et al., 2003; Wang and Brown, 1999). For successful qRT-PCR and micro-array experiments it is important to use intact RNA. It is not known how this image is influenced by sample preparation factors which such as RNA quality, cDNA synthesis and labeling efficiency. Therefore the determination of RNA quality is a critical first step in obtaining meaningful data of gene expression. Many factors present in samples as well as exogenous contaminants have been shown to inhibit the RT as well as the PCR. Some of them derive from the extracted tissue, others stem from inefficient or messy lab management. For example, the presence of haemoglobin, fat, glycogen, cell constituents, Ca2+, high genomic DNA concentration, and DNA binding proteins are important factors (Wilson, 1997; Rossen et al., 1992). Exogenous contaminants such as glove powder and phenolic com- pounds from the extraction process or the plastic ware can have an inhibitory effect. Also unknown tissue-specific factors can influence amplification kinetics but this affect can be ameliorated, in part, by appropriate primer selection (Wilson, 1997; Rossen et al., 1992; Tichopad et al., 2004). There nevertheless, little is known about the possibility of obtaining reasonable qRT-PCR data from RNA samples with impaired quality. Expression differences for some genes can independently confirmed by real-time qRT-PCR. Gene Expression profiles obtained from partially degraded RNA samples with still visible ribosomal bands exhibit a high degree of similarity compared to intact samples and that RNA samples of sub-optimal quality. This might therefore still lead to meaningful results if used carefully (Schoor et al., 2003).

2.2. Eﬀect on the mRNA quantiﬁcation

In view of the observed difference in gene expression stability between intact and degraded RNA samples from the same tissue and the higher gene-specific variation in degraded samples, we propose performing RNA quality control prior to downstream quantification assays (Bustin and Nolan, 2004a). Especially if one aims to accurately quantify small expression differences (Perez-Novo et al., 2005). With that prospect in mind, and with the aim of anticipating future standards by pre-normative research, it is connotatively too identified and analyzed the influence of degraded RNA on the performance on qRT-PCR. In a study from Fleige and Pfaffl (2006) the purity and integrity of RNA samples was assessed, derived from different bovine 134 S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 tissues and cell lines, using the Bioanalyzer 2100 (Agilent Technologies). To test the influence of the RNA integrity (numbered according to the RIN classification), the intact transcriptome of one distinct bovine tissue was degraded factitiously by enzymatic digest or with ultraviolet light. This leads to enzymatic cutoffs or breaks in the native RNA strand resulting in fragments of different lengths. A gradient with several steps of intact RNA (RIN 8–10) down to degraded RNA (RIN = 1–3) was investigated. The effect of RIN on qRT-PCR performance was investigated by correlating the RIN values with the crossing points (CP) of the PCR runs. The expression levels of four genes were assessed, all of different abundance levels ranging from high abundant 18S and 28S rRNA, intermediate abundant b-actin, down to very low expressed IL-1b mRNA samples. The importance of using high-quality RNA is demonstrated by the results shown in Fig. 2. A high-quality RNA (high RIN) determined a lower CP than by a less-quality (lower RIN). High significant relation between RIN and CP (p < 0.01 for the trend) could be shown for all examined genes (n = 4) and tissues (n = 14). With increasing RNA quality the variability of the qRT-PCR result was decreased (Huch et al., 2005). It is well known, that normalization by an internal reference gene reduce or even diminish tissue derived effects on qRT-PCR (Wittwer et al., 1997). Reliability of any relative RT-PCR experiment can be improved by including an invariant endogenous control in the assay to correct for sample-to-sample variations in RT-PCR efficiency and errors in sample quantification. So called relative quantification determines the changes in steady-state mRNA levels of a gene across multiple samples and gives a result relative to the levels of an internal control RNA (Pfaffl, 2001). For many

Fig. 2. Influence of RNA integrity (RIN) on the Crossing Point (CP): Amplification curves from three HKG (18S, 28S, b-Actin) and IL-1b with different quality of employed RNA from corpus luteum. An increase of RNA degradation correlates significantly to the amplified product, such that the CP is decrease with increasing RNA integrity number (RIN). Quantitative analyses use the threshold cycle number (Ct), at which the signal is detected above the background and is in the exponential phase. S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 135 experiments this method is most adequate for investigating physiological changes in gene expression levels. It is based on the expression levels of a target gene versus an internal reference gene, often non-regulated housekeeping gene are prominent candi- dates. To get rid of the RIN dependency the CP data were normalized by an inter- nally expressed reference gene (Fleige and Pfaffl, 2006), according to the DCP method described earlier (Livak and Schmittgen, 2001). The normalized results (Fig. 3), expressed as RIN compared to DCP values showed minor influence of RNA quality on the expression results, and the significant effects could be reduced to a minimum. Sometimes, even intact RNA does not guarantee good results because RNA sample may contain inhibitors that can reduce reaction efficiency (Bustin and Nolan, 2004a; Wong and Medrano, 2005). These factors include length of the amplicon, secondary structure and primer quality. The shapes of amplification curves differ in the steepness of any fluorescence increase and in the absolute fluorescence levels at plateau depending on background fluorescence levels. Therefore PCR efficiency has a major impact on the fluorescence history and the accuracy of the calculated expression result and in critically influenced by PCR reaction components. Efficiency evaluation is an essential marker in real-time gene quantification procedure (Ticho- pad et al., 2003, 2004). The effect of RIN on PCR efficiency was investigated similarly to the above mentioned tissues and various RNA qualities. The efficiency of all investigated genes was not affected by the RNA quality, independent of gene or tissue. A causally determined correlation between the RIN and the CP is shown in Fig. 4 (Fleige and Pfaffl, 2006).

Fig. 3. Influence of RNA integrity (RIN) on the delta CP. The results (CP) from Fig. 2 are normalized with b-Actin. The significant effect of RNA integrity could reduce to a minimum. 136 S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139

Fig. 4. Influence of RNA integrity (RIN) on PCR Efficiency: The Efficiency was generated by Rotor-Gene 3000 software (Corbett-Research). Only four tissues (lymphnode, corpus luteum, caecum, abomasums) were graph, additional results show the same trend.

2.3. Association between RNA quality and length of the ampliﬁed product

The PCR efficiency is also influenced by various factors, among other things by the annealing temperature, the primer length or by the length of the amplified product. And because of exponential amplification of the initial information, any extant error is amplified, too (Tichopad et al., 2002). The new question is, if the PCR efficiency during real-time qRT-PCR is influenced by the RNA quality or not? There- fore again, a gradient with several steps of intact RNA down to degraded RNA were examined with different primer sets, amplifying qRT-PCR products of various lengths. Primer sets for varying lengths of product (50–950 bases) were used to amplify the sequence of b-actin in different tissues and RNA integrity levels. The correlation between RNA integrity and CP were examined. The results of the correlation between the RIN and CP fulfilled the expectations. It is clearly visible that the crossing point is shifted towards lower cycle numbers using intact total RNA or higher RIN. With increasing length of the amplified product, the importance of RNA quality rises. Regarding the results concerning the correlation between the RIN and the CP values, there were some differences in the tested tissues. In some tissues a correlation between the RIN and the crossing point was visible for shorter products and in WBC and corpus luteum this correlation was visible as well for longer products. In general we can point out, that amplification of long product over 400 bp is strongly dependent on a good RNA quality, which should show at least a RIN of 5. Shorter qRT-PCR products, mostly used with the length of 70–250 bp, are more or less ‘‘independent’’ of the RNA quality. Viewing the correlation between the RIN S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 137 and the efficiency of PCR, it is noticeable that the efficiency does not vary within one amplicon length, despite some exceptions. No correlation between the RIN and PCR efficiency (ranging between 1.6 and 1.7) was given (Pfaffl, 2005b). Other studies showed as well an inhibitory effect of poor RNA quality on real- time PCR results. Degraded or impure RNA can limit the efficiency of the RT reaction and reduce yield. RNA should either be prepared from fresh tissue, or from tissue treated with an RNA stabilization solution such as RNA later (Labourier, 2003, 2004). The importance of using full length RNA for reverse transcription depends on the application. As a result, some degradation of the RNA can be tolerated. If it is not possible to use completely intact RNA, a design of primers to anneal an internal region of the gene of interest is useful. Note that for truly quantitative RT-PCR, partially degraded RNA may not give an accurate representation of gene expression (Wang, 2005).

3. Conclusion

In conclusion, while all efforts should be made to obtain high-quality RNA samples that reflect the natural state most reliably, moderately degraded samples with a degradation signature may still lead to a reasonable qRT-PCR expression profile. The normalized expression differences measured with the real-time RT-qPCR are similar to those obtained from high-quality samples. Only the non-normalized values show a correlation between RNA integrity and CP. This findings show the importance of the normalization. The reliability of any relative RT-PCR experiment can be improved by including an invariant endogenous control in the assay to correct for sample-to-sample variations in RT-PCR efficiency and errors in sample quantification. Furthermore, RNA samples of optimal quality can serve as a template for all product lengths whereas for degraded RNA primer pairs for shorter amplicon are more suitable. To be on the safe side with primer pairs it would be helpful to prove the RNA quality before starting the run. Up to now it is still questionable if we can use the 28S/18S ratio or the RIN, both based on the quantity and quality check of the ribosomal sub-units, to make a def- inite statement on the mRNA quality which is our target in qRT-PCR. We are look- ing forward for sensitive methods, comparable to an intelligent algorithm, which prove the real mRNA integrity to have a reliable answer on mRNA quantity and quality.

References

Agilent Technologies, 2004. DNA, RNA, protein and cell analysis. Agilent Application Note, Publication Number 5989-1197EN. Bar, T., Sta˚hlberg, A., Muszta, A., Kubista, M., 2003. Kinetic outlier detection (KOD) in real-time PCR. Nucleic Acids Research 31 (17), e105. Bustin, S.A., 2002. Quantiﬁcation of mRNA using real-time reverse transcription PCR (RT-PCR): trends and problems. Journal of Molecular Endocrinology 29, 23–39. 138 S. Fleige, M.W. Pfaﬄ / Molecular Aspects of Medicine 27 (2006) 126–139

Bustin, S.A., Nolan, T., 2004a. Pitfalls of quantitative real-time reverse-transcription polymerase chain reaction. Journal of Biomolecular Techniques 15, 155–166. Bustin, S.A., Nolan, T., 2004b. Template handling, preparation, and quantification. In: Bustin, S.A. (Ed.), The Real-Time PCR Encyclopaedia A–Z of Quantitative PCR. Published by International University Line, La Jolla, CA, pp. 87–120. Bustin, S.A., Benes, V., Nolan, T., Pfaffl, M.W., 2005. Quantitative real-time RT-PCR—a perspective. Journal of Molecular Endocrinology 34, 597–601. Fleige, S., Pfaffl, M.W., 2006. Impact of RNA integrity on the quantitative real-time RT-PCR performance and relative mRNA quantification models. Submitted to RNA (January 2006). Huch, S., Fleige, S., Pfaffl, M.W., 2005. Determination of RNA quality and its influence on the quantitative real-time RT-PCR. Diploma thesis, Department of Physiology, TUM-Weihenstephan. Imbeaud, S., Graudens, E., Boulanger, V., Barlet, X., Zaborski, P., Eveno, E., Mueller, O., Schroeder, A., Auffray, C., 2005. Toward standardization of RNA quality assessment using user-independent classifiers of microcapillary electrophoresis traces. Nucleic Acids Research 33 (6), e56. Jones, L.J., Yue, S.T., Cheung, C.Y., Singer, V.L., 1998. RNA quantification by fluorescence-based solution assay: RiboGreen reagent characterization. Analytical Biochemistry 265, 368–374. Karsten, U., Wollenberger, A., 1977. Improvements in the ethidium bromide method for direct fluorometric estimation of DNA and RNA in cell and tissue homogenates. Analytical Biochemistry 77, 464–470. Labourier, E., 2003. A better way to isolate RNA and protein from the same sample. Ambion TechNotes 10 (3), 16–17. Labourier, E., 2004. RNAlater for RNA and Protein Stabilization. Ambion TechNotes 11 (3), 7. LePecq, J.B., Paoletti, C., 1966. A new fluorometric method for RNA and DNA determination. Analytical Biochemistry 17, 100–107. Lightfood, S., 2002. Quantitation comparison of total RNA using the Agilent 2100 bioanalyzer, ribogreen analysis, and UV spectrometry. Agilent Application Note, Publication number 5988-7650EN. Livak, K.J., Schmittgen, T.D., 2001. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods 25, 402–408. Manchester, K.L., 1996. Use of UV methods for measurement of protein and nucleic acid concentrations. Biotechniques 20, 968–970. Marx, V., 2004. RNA Quality: Defining the Good, the Bad, the Ugly. Cover Story, Genomics and Proteomics. Mayne, M., Shepel, P.N., Geiger, J.D., 1999. Recovery of height integrity mRNA from brains of rats killed by high-energy focused microwave irradiation. Brain Research Protocols 4, 295–302. Miller, C., Diglisic, S., Leister, F., Webster, M., Yolken, R.H., 2004. Evaluating RNA status for RT-PCR in extracts of postmortem human brain tissue. Biotechniques 36, 6238–6633. Monstein, H.J., Nylander, A.G., Chen, D., 1995. RNA extraction from gastrointestinal tract and pancreas by a modified Chomczynski and Sacci method. Biotechniques 19, 340–344. Mueller, O., Hahnenberger, K., Dittmann, M., Yee, H., Dubrow, R., Nagle, R., Ilsley, D., 2000. A microfluidic system for high-speed reproducible DNA sizing and quantitation. Electrophoresis 21, 128–134. Mueller, O., Lightfoot, S., Schroeder, A., 2004. RNA integrity number (RIN)—standardization of RNA quality control. Agilent Application Note, Publication 5989-1165EN, pp. 1–8. Perez-Novo, C.A., Claeys, C., Speleman, F., Cauwenberge, P.V., Bachert, C., Vandesompele, J., 2005. Impact of RNA quality on reference gene expression stability. Biotechniques 39, 52–56. Pfaffl, M.W., 2001. A new mathematical model for relative quantification in real-time RT-PCR. Nucleic Acids Research 29 (9), e45. Pfaffl, M.W., 2004. Quantification Strategies in real-time PCR. In: Bustin, S.A. (Ed.), The Real-Time PCR Encyclopedia A–Z of Quantitative PCR. Published by International University Line, La Jolla, CA, pp. 87–120. Pfaffl, M.W., 2005a. Nucleic acids: mRNA identification and quantification. In: Worsfold, P.J., Gallagher, P.K. (Eds.), Nucleic Acids, Encyclopedia of Analytical Science, second ed. Academic Press, ISBN 0-12- 764100-9, pp. 417–426. S. Fleige, M.W. Pfaffl / Molecular Aspects of Medicine 27 (2006) 126–139 139

Pfaffl, M.W., 2005b. qPCR 2005: 2nd International qPCR Symposium. The Whole Story of Quantitative PCR—from Tissue Preparation to Bioinformatics. Proceedings, ISBN 3000166874, p. 11. Raeymarkers, L., 1993. Quantitative PCR: theoretical consideration with practical implication. Analytical Biochemistry 214, 582–585. Rossen, L., Norskov, P.l., Holmstrom, K., Rasmussen, F.O., 1992. Inhibition of PCR by components of food sample, microbial diagnostic assay and DNA-extraction solutions. International Journal of Food Microbiology 17, 37–45. Sambrook, J., Fritsch, E.F., Maniatis, T., 1989. Molecular Cloning: A Laboratory Manual, second ed. Cold Spring Harbor Laboratory Press, NY. Sambrook, J., Russel, D.W., 2001. Molecular Cloning: A Laboratory Manual. CSH Laboratory Press, Cold Spring Harbor, NY. Santiago, T.C., Purvis, I.J., Bettany, A.J., Brown, A.J.P., 1986. The relationship between mRNA stability and length in Saccharomyces cerevisiae. Nucleic Acids Research 12, 8347–8360. Schoor, O., Weinschenk, T., Hennenlotter, J., Corvin, S., Stenzl, A., Rammensee, H.-G., Stevanovic, S., 2003. Moderate degradation does not preclude microarray analysis of small amounts of RNA. BioTechniques 35, 1192–1201. Tichopad, A., Dzidic, A., Pfaffl, M.W., 2002. Improving quantitative real-time RT-PCR reproducibility by boosting primer-linked amplification efficiency. Biotechnology Letters 24, 2053–2056. Tichopad, A., Dilger, M., Schwarz, G., Pfaffl, M.W., 2003. Standardized determination of real-time PCR efficiency from a single reaction setup. Nucleic Acids Research 31 (20), e122. Tichopad, A., Didier, A., Pfaffl, M.W., 2004. Inhibition of real-time RT-PCR quantification due to tissue- specific contaminants. Molecular and Cellular Probes 18, 45–50. Wang, T., Brown, M.J, 1999. mRNA quantification by real time TaqMan polymerase chain reaction: validation and comparison with RNase protection [In Process Citation]. Analytical Biochemistry 269, 198–201. Wang, E., 2005. RNA amplification for successful gene profiling analysis. Journal of Translational Medicine 3, 28. Wilson, I.G., 1997. Inhibition and facilitation of nucleic acid amplification. Applied and Environmental Microbiology 63, 3741–3751. Wittwer, C.T., Herrmann, M.G., Moss, A.A., Rasmussen, R.P., 1997. Continuous fluorescence monitoring of rapid cycle DNA amplification. Biotechniques 22 (130-131), 134–138. Wong, M.L., Medrano, J.F., 2005. Real-time PCR for mRNA quantitation. BioTechniques 39, 75–85.