- Email: [email protected]
Influence of RNA Labeling on Expression Profiling of MicroRNAs
The Journal of Molecular Diagnostics, Vol. 14, No. 1, January 2012 Copyright © 2012 American Society for Investigative Pathology and the Association for Molecular Pathology. Published by Elsevier Inc. All rights reserved. DOI: 10.1016/j.jmoldx.2011.08.005
Technical Advance Influence of RNA Labeling on Expression Profiling of MicroRNAs
John S. Kaddis,* Daniel H. Wai,* Jessica Bowers,† Nicole Hartmann,‡ Lukas Baeriswyl,‡ Sheetal Bajaj,* Michael J. Anderson,* Robert C. Getts,† and Timothy J. Triche*§ From the Departments of Pathology and Laboratory Medicine,* Children’s Hospital Los Angeles Saban Research Institute and Keck School of Medicine, University of Southern California, Los Angeles, California; Genisphere LLC,† Hatfield, Pennsylvania; the Novartis Institutes for BioMedical Research,‡ Novartis, Basel, Switzerland; and the USC/CHLA Genome Core Laboratory,§ Children Hospital Los Angeles, Los Angeles, California
Although a number of technical parameters are now being examined to optimize microRNA profiling experiments, it is unknown whether reagent or component changes to the labeling step affect starting RNA requirements or microarray performance. Human brain/lung samples were each labeled in duplicate, at 1.0, 0.5, 0.2, and 0.1 g of total RNA, by means of two kits that use the same labeling procedure but differ in the reagent composition used to label microRNAs. Statistical measures of reliability and validity were used to evaluate microarray data. Cross-platform confirmation was accomplished using TaqMan microRNA assays. Synthetic microRNA spike-in experiments were also performed to establish the microarray signal dynamic range using the ligation-modified kit. Technical replicate correlations of signal intensity values were high using both kits, but improved with the ligation-modified assay. The drop in detection call sensitivity and miRNA gene list correlations, when using reduced amounts of standard-labeled RNA, was considerably improved with the ligation-modified kit. Microarray signal dynamic range was found to be linear across three orders of magnitude from 4.88 to 5000 attomoles. Thus, optimization of the microRNA labeling reagent can result in at least a 10-fold decrease in microarray total RNA requirements with little compromise to data quality. Clinical investigations bottlenecked by the amount of starting material may use a ligation mix modification strategy to reduce total
12
RNA requirements. ( J Mol Diagn 2012, 14:12–21; DOI: 10.1016/j.jmoldx.2011.08.005)
MicroRNA (miRNA) expression profiling platforms have burgeoned in the last decade1– 4 and promise to aid in the diagnosis, prognosis, and therapeutic treatment of human diseases such as cancer,5 cardiovascular disease,6 and diabetes.7 In response, intra-, inter-, and cross-platform comparisons have recently emerged that examine the use of microarray technology alone8,9 or in contrast to other profiling methods.10 –14 Such studies have reported on the reliability and validity of widely used miRNA discovery tools and to highlight the advantages and limitations of these platforms. However, there also remains a need to evaluate preexperimental factors, such as sample handling, processing, storage, and nucleic acid quality, that may influence miRNA expression profiling experiments.15 Whereas three studies have demonstrated that miRNA detection was possible in formalin-fixed paraffin-embedded (FFPE) samples of up to 1016 or 1217 years old using extraction kits from different manufacturers,18 stability and comparability of miRNA expression signal using paired freshfrozen (FF) versus FFPE samples was reported in human and murine specimens using different tissues ranging in age from newly acquired19 to up to 3 years of age.17,20 In contrast, when detecting the expression of longer RNA transcripts, Abdueva et al21 found differences in paired FF versus FFPE samples of up to 7.5 years of age, but Supported in part by the National Cancer Institute, Strategic Partnering to Evaluate Cancer Signatures (U01CA-114757, to T.J.T.). Accepted for publication August 22, 2011. Disclosures: R.C.G. and J.B. are employed by Genisphere LLC and provided the microarrays and corresponding reagents for the RNA titration experiments. N.H. and L.B. are employed by Novartis and provided the microarrays and corresponding reagents for the synthetic spike-in experiments. Supplemental material for this article can be found at http://jmd. amjapthol.org or at doi: 10.1016/j.jmoldx.2011.08.005. Address reprint requests to Timothy J. Triche, M.D., Ph.D., Department of Pathology, Children’s Hospital Los Angeles, 4650 Sunset Blvd., Los Angeles, CA 90027. E-mail: [email protected].
RNA Labeling For MicroRNA Profiling 13 JMD January 2012, Vol. 14, No. 1
concluded that useable profiles were obtainable. A number of studies examined RNA extraction methods for miRNA, with one that compared three isolation kits and found general comparability among them,22 two that used five kits and identified a preferred approach,23,24 and one that evaluated eight kits and found R2 correlations exceeding 0.857 in all comparisons made.25 Conflicting reports have surfaced on the length of time RNA can be stored after extraction for miRNA profiling, with one study describing degradation after 3 days26 and another demonstrating stability of up to 10 months,27 although cells from different species were used in each study. Likewise, the quality of RNA28 has also been called into question, with some showing that medium- to high-integrity samples are required for reproducible miRNA profiling using bovine29 and murine30 samples, and another reporting negligible effects on highly degraded human samples.31 Consideration of these experimental factors is crucial to ensuring valid and reliable miRNA expression profiling results within and across laboratories. With the exception of next-generation sequencing (NGS), all miRNA expression profiling approaches use a labeling method to tag the target molecules of interest. In the case of transcripts greater than 200 nucleotides in length, the method used to amplify and/or label the molecules of interest before detection has been shown to be important. Studies examining linear,32–36 exponential,37 and linear versus exponential38,39 amplification methods revealed that although total RNA requirements can be reduced significantly by using amplified RNA, quality metrics such as comparability, sensitivity, and accuracy were, in some cases, compromised when compared with those using nonamplified RNA. However, these measures can be improved when robust data methods were used.40 Next, evaluation of direct, indirect, and other cDNA labeling approaches have shown that the method chosen can reduce total RNA requirements, but that gains in signal intensity values were also achieved alongside, in some cases, a loss in labeling performance.41– 46 Similar observations have been made when comparing cRNA labeling methods.47–50 Even when performing replicate experiments within51 or across52 laboratories, RNA labeling has been shown to contribute to the variability seen in gene expression outcome. Although the body of work examining target labeling is extensive, it is unclear whether and how these findings can be applied to low abundance short transcripts ⬍200 nucleotides. This is especially true in the case of miRNAs, when considering that they represent only 0.5% to 9.2% of total RNA in human and 0.1% to 1.3% in mouse and rat tissue samples,53 but have also been shown to be present, at some level, in a variety of human bodily fluids.54 Because many studies are limited in the amount of total RNA available, such as those using tissues derived from needle biopsies or laser capture microdissection, more sensitive target labeling procedures are critical. Accordingly, a growing number of amplification and labeling strategies have been developed to aid in the detection of these low-abundance RNAs.55 Although there are at least two studies that have examined the effects of amplifica-
tion on miRNA expression,10,56 there are currently no studies that compare amplification or labeling effects when direct labeling of mature miRNA is used. We therefore conducted a study to directly compare the Genisphere FlashTag biotin-HSR (biotin-HSR) labeling kit to the previously recommended FlashTag biotin-only (biotin-only) assay (Genisphere, Hatfield, PA) for the Affymetrix GeneChip miRNA Arrays. The biotin-HSR kit is a modified version of the biotin-only assay. More specifically, a smaller more efficient labeling molecule, having a higher number of biotins per unit mass of DNA, has been used in the biotin-HSR kit compared to the DNA dendrimer of the original FlashTag biotin kit. A total of 32 chips were used for this evaluation, followed by crossplatform validation using TaqMan miRNA assays. In addition, the linear dynamic range of the GeneChip was assessed using synthetic miRNA spiked-in at different amounts, labeled with the biotin-HSR kit, on 12 chips. A total of 44 microarray chips were used in this study.
Materials and Methods Human Normal Tissue RNA FirstChoice Total Brain (Lot#0906005) and Lung (Lot#0904002) RNA samples were obtained from Ambion, Inc., certified to contain small RNAs, and quantified at 1 mg/ml (Applied Biosystems/Ambion, Foster City, CA).
Labeling of RNA and Microarray Processing Brain and lung total RNA samples were each labeled using both the FlashTag biotin-only and FlashTag biotin-HSR RNA labeling kits (Genisphere, Hatfield, PA) for the Affymetrix GeneChip miRNA array. Each microarray contains sequences for 6703 miRNA probes, 499 small nucleolar RNA (snoRNA), including 274 C/D and 127 H/ACA box RNA, 22 small cajal body–specific RNA (scaRNA), 10 5.8S ribosomal RNA, 162 Affymetrix controls, and 22 oligonucleotide spike-in controls. Probes were developed by Affymetrix using the Sanger miRNA database v11 and contained additional snoRNAs and scaRNAs from the Ensembl database and snoRNABase. RNA labeling was performed by the kit manufacturer and efficiency assessed using an enzymelinked oligosorbent assay. Hybridization, washing, and scanning of slides were performed according to the Affymetrix and Genisphere protocols. Scanning was performed using the miRNA-1_0 library file from Affymetrix for the Flashtag biotin-only and miRNA-1_0_2Xgain for the FlashTag biotin-HSR labeled samples. Refer to NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) study GSE30045 to access Affymetrix CEL file data.
RNA Titration Experiments Total RNA was prepared at recommended and reduced input amounts from two different tissues to examine the performance of the Affymetrix GeneChip miRNA array and to evaluate the significance of using a new RNA
14 Kaddis et al JMD January 2012, Vol. 14, No. 1
labeling kit on the quality of data generated. Brain samples were prepared containing 0.1 g, 0.2 g, 0.5 g, and 1.0 g of total RNA. Two aliquots from each sample were taken and labeled with either the biotin-only or biotin-HSR kit. This process was carried out in duplicate for each tissue. A total of 32 microarray chips were used for this study.
Synthetic miRNA Spike-In Experiments The mirVana miRNA Reference Panel v9.1 (Lot#072307) was obtained from Ambion, Inc. and contains an equimolar pool of 470 human, 224 murine, and 42 rat synthetic miRNA oligonucleotides according to the Sanger miRBase sequence database, release 9.1 (Applied Biosystems/Ambion, Foster City, CA). Twofold serial dilution samples were then prepared that contained from 5000 to 4.88 attomoles of miRNA (12 total, including a negative control). Synthetic miRNAs were labeled using the FlashTag biotin-HSR RNA labeling kits. Hybridization, washing, and scanning of slides were performed by Novartis according to Affymetrix and Genisphere protocols. Scanning was performed using the miRNA-1_0_2Xgain library file from Affymetrix.
Array Quality, Data Processing, and Detection Calls Raw intensity signal values of Affymetrix spike-in controls indicated that array hybridization was successful (ie, bioB⬍bioC⬍bioD⬍Cre). Likewise, raw intensity values from an additional five Genisphere oligonucleotide spikeins control probes were all found to be ⬎1000, used to demonstrate poly(A) tailing and ligation (3 RNA oligos), ligation (1 poly(A) RNA), and ligation and lack of RNases in the RNA sample (1 poly(dA) DNA; data not shown). Because labeling was an experimental factor being tested, before analysis, raw signal intensity values were Robust Multichip Average (RMA) background corrected, quantile normalized, median polish summarized, and log2 transformed separately for FlashTag biotin-only and FlashTag biotin-HSR chip data.57– 61 Processing was performed using Partek Genomics Suite, version 6.5, build 6.10.0412 copyright 2010 (Partek, St. Louis, MO). Present/absent detection calls for each probe, on every array, were generated using the Affymetrix detection algorithm and statistical significance testing implemented through the miRNA QCTool, version 1.0.33.0. Details on how the algorithm determines these calls are described in Appendix A of the Affymetrix miRNA QC Tool User’s Guide 6 that, along with the software utility, are freely available on the manufacturer’s website.
Reverse Transcription and Real-Time PCR Total RNA was converted into cDNA and quantified using individual Taqman assay kits for each miRNA of interest (Applied Biosystems, Foster City, CA). Real-time PCR followed the reverse transcription (RT) reaction, which was modified to allow for the simultaneous synthesis of
cDNA by combining RT primers into pools A and B. The multiplex RT protocol is a modification of the original kit instructions and made available through the manufacturer by request. RNU6B, SNORD44, and SNORD48 were selected as candidate artificial normalization controls and previously shown to be stable in a panel of 38 normal human tissues, including Brain and Lung.62 Identification of hsa-let-7d and hsa-miR-151-5p as candidate endogenous human miRNA normalization controls, specific to the present experiment, was performed, using microarray data, in the following way: i) A signal detection call of “present” in both Brain and Lung tissue across all input RNA titrations and replicates using both the old and new labeling methods (79 of 847 miRNA probesets remained); ii) the absence of a statistically significant fold change difference between brain and lung tissue (20 of 79 remained); iii) removal of half the probesets with the lowest mean signal intensity values and those previously shown63 as inadequate (eight of 20 remained); and iv) use of NormFinder64 to rank (data not shown) and select the most stable miRNAs (two of eight remained). Subsequent analysis of all five candidate control Ct values by geNorm65 identified the combination of no more than two miRNAs, ie, hsa-miR-151-5p/SNORD44 and hsa-let-7d/SNORD44, as the most stably expressed housekeeping miRNAs in pools A and B, respectively (data not shown). The expression of miRNAs, relative to selected controls, was determined using the 2(–⌬⌬Ct) method,66 and as also described online in an Applied Biosystems document titled “User Bulletin #2: ABI Prism 7700 Sequence Detection System.”
Data Analysis The mean or median was used as a measure of central tendency for all continuous numerical variables, depending on whether the values followed a normal Gaussian probability distribution. Student’s t-test was used when performing two-group comparisons using parametric data. An F-test for equality of variance was performed before significance testing. A Wilcoxon rank-sum test was used when performing two-group comparisons using nonparametric data. For descriptive statistics, significance was indicated if P was ⬍0.05. Coefficient of variation (CV) for replicate chips was calculated as the SD of the processed signal intensity value over the mean for miRNAs called as present. The Pearson product-moment correlation coefficient (r) was used to measure the linear relationship of signal intensity values between replicate chips. Spearman’s rank correlation coefficient () was used to evaluate intra- and cross-platform miRNA gene list comparisons. Unlike r, measures the relative rank of a miRNA gene in list A against its relative rank in list B, thereby minimizing the effects of outlier data. For both r and , the null hypothesis of no relationship was indicated if P was ⬎0.05. Statistical significance for differential miRNA gene expression was determined using a false discovery rate (FDR) P value of ⱕ0.05, unless otherwise noted. Data analysis was performed using either Partek
RNA Labeling For MicroRNA Profiling 15 JMD January 2012, Vol. 14, No. 1
Genomics Suite or SAS software, version 9.1.3 SP 4 (SAS Institute, Cary, NC).
Results Reliability (RNA Titrations) Qualitative and quantitative measures of random error within and between labeling kits were examined at all input amounts of total RNA from both brain and lung tissue. The number of present and absent detection calls was first calculated for all replicate chips (see Supplemental Figure S1 at http://jmd.amjpathol.org). Percent concordant (present and absent) calls for all probes on the array using total RNA labeled with the biotin-only kit was found to range from a grand mean of 95.0% for brain to 95.8% for lung. Those values increased to 96.6% and 97.4%, respectively, when the RNA was labeled using the biotin-HSR kit, although this was not a statistically significant finding (P ⬎ 0.05 for both comparisons made). On restricted examination of only the 847 human miRNA gene probes, a statistically significant increase in the number of detection calls was seen when concurrently examining the differences in labeling kits alongside the amount of input RNA used for microarray hybridization (Figure 1). For brain samples, as the amount of biotin-only labeled RNA is reduced from 1.0 g to 0.1 g, the mean percentage of detected overlapping probes was decreased by 31%. However, this was statistically significantly improved to a mean value of only 7% when using biotin-HSR labeled RNA (P ⫽ 0.0490). Likewise, for lung samples, the loss was statistically significantly improved from a mean value of 25% to 5% (P ⫽ 0.04). Irrespective of detection call, the reproducibility of processed signal intensity values between duplicate chips was next examined (Table 1). A statistically significant improvement in the mean Pearson’s correlation coefficient value was observed for duplicate chips labeled using the biotin-HSR kit. This finding was true for RNA isolated from either brain or lung tissue (P ⬍ 0.0001 for both comparisons made). Individual correlations are provided in a 16 ⫻ 16 correlation matrix table for both brain and lung samples (see Supplemental Tables S1 and S2 at http://jmd.amjpathol.org). Finally, the CVs of the processed signal intensity values were calculated for all study samples (Figure 2). There was a statistically significant reduction in the grand median CV value for brain samples labeled with the biotin-HSR versus biotin-only kit (2.59% to 1.25%; Wilcoxon rank-sum P ⬍ 0.05). However, a statistically significant reduction in the grand median CV value for lung samples was not observed (1.96% to 1.16%; Wilcoxon rank-sum P ⬎ 0.05).
Validity (RNA Titrations and RT Real-Time PCR) Systematic error was evaluated by examining the interplatform agreement and cross-platform concordance of brain to lung expression ratios for selected human miRNAs. First, using differential miRNA gene expression analysis, the numerical rank of all probes from the 1.0 g
Figure 1. Comparison of detected miRNA genes. Probes for all 847 human miRNAs were examined in (A) brain and (B) lung samples. The number of present calls is indicated by the y axis. Comparisons on the x axis were done to determine the number of overlapping present calls on chips containing 1.0 g of input RNA vs. 0.5, 0.2, and 0.1 g of starting material. One vs. 1.0 g indicates the maximum possible overlap. Evaluations using the biotin-only kit are represented by filled circles, biotin-HSR by filled squares, and biotinonly versus biotin-HSR by open triangles. Statistical significance of *P ⱕ 0.05.
biotin-only labeled RNA list was compared to the order from lists using reduced amounts of input RNA (Table 2). This process was repeated following sequential application of filters, defined in Table 2, to the 1.0 g biotin-only miRNA gene list to selectively reduce the number of
Table 1. Correlation of Replicate Chip Signal Intensity Values Pearson product-moment correlation (r) Labeling kit Tissue Brain Lung
Biotin-only global Mean (⫾1 SD)
Biotin-HSR global Mean (⫾1 SD)
P value
0.955 (0.023) 0.976 (0.011)
0.978 (0.010) 0.985 (0.006)
⬍0.0001 ⬍0.0001
The processed signal intensity value for each probe on a chip was correlated with the corresponding probe on a replicate array (7819 probes total). Individual correlations between replicate arrays also were all found to be statistically significant (P ⬍ 0.0001; see Supplemental Tables S1 and S2 at http://jmd.amjpathol.org). Global mean calculations did not include correlations of a chip with itself, ie, where r ⫽ 1.
16 Kaddis et al JMD January 2012, Vol. 14, No. 1
Figure 2. Coefficient of variation within and between RNA labeling kits. The percent CV was averaged for replicate chips at each input RNA amount (0.1–1.0 g) within and between labeling kits for both brain and lung samples. The interquartile range of the processed signal intensity value is represented by shaded boxes, median value by a black line within a box, the 5th and 95th percentiles by the whiskers outside of each box and the number of detected miRNA genes by filled boxes. CVs were calculated only for miRNA genes called as present.
Figure 3. Evaluation of statistically significant, differentially expressed miRNA genes using Biotin-HSR labeled RNA. Genes were included in this analysis using only human miRNA probes detected as present in at least brain or lung tissue, with absolute fold change ⱖ1.5, and an FDR p ⬍0.05. A total of 419, 376, 218, and 184 differentially expressed miRNA genes were identified when using 1.0 g, 0.5 g, 0.2 g, and 0.1 g of input RNA, respectively.
candidate probes evaluated. Regardless of miRNA gene filters used, as the amount of biotin-only labeled RNA is reduced, correlations with the 1.0 g biotin-only miRNA gene list also falls. However, when using reduced amounts of biotin-HSR labeled RNA, gene list correlations remain steady. Depending on the miRNA gene list filter used, there is a 60% to 93% reduction in the variation of correlation values when using reduced amounts of biotin-HSR versus biotin-only labeled RNA. Second, differentially expressed miRNA gene lists were also generated for biotin-HSR labeled samples at each input RNA amount and evaluated for overlap in statistically significant transcripts (Figure 3). Of the 1779 human probes, a total of 474 differentially expressed miRNA genes were identified. As the amount of input RNA increases, the percentage of miRNA genes unique to each list also increases from 4.3% for
Table 2.
0.1 g and 4.1% for 0.2 g to 8.8% for 0.5 g and 16.9% for 1.0 g. The percentage of miRNA genes that were detected in more than 1 list, but not by all, is 22.8% for 0.1 g, 34.4% for 0.2 g, 55.6% for 0.5 g, and 68.0% for 1.0 g. There were 134 miRNA genes robustly detected by all lists with a minimum absolute fold change of at least 10 or more for 86 genes, and two to ⬍10 for the remaining 48 genes. Next, we took the 272 miRNA genes, from Table 2, and compared it with the 474 identified miRNA genes from Figure 3, to determine the gene list effects of using a new labeling kit (see Supplemental Table S3 at http://jmd. amjpathol.org). Of the 272 miRNA genes identified using 1.0 g of biotin-only labeled RNA, 136 of them were also
miRNA Gene List Agreement within and Between RNA Labeling Kits
Gene list filters applied* (number of probes) Biotin-only kit 0.5 g 0.2 g 0.1 g Range (Max–Min) Biotin-HSR kit 0.5 g 0.2 g 0.1 g Range (Max–Min) % Reduction in range using Biotin-HSR kit
0 (n ⫽ 7819) 0.765 0.693 0.588 0.177 0.791 0.766 0.721 0.070 60
1 (n ⫽ 1779)
2 (n ⫽ 601)
3 (n ⫽ 422)
Spearman Rank Correlation Coefficient 0.734 0.961 0.972 0.662 0.915 0.938 0.558 0.830 0.879 0.176 0.131 0.093 0.762 0.733 0.692 0.070 60
0.962 0.957 0.937 0.025 81
0.968 0.970 0.960 0.010 89
4 (n ⫽ 272) 0.978 0.949 0.903 0.075 0.973 0.977 0.972 0.005 93
The brain versus lung differentially expressed miRNA gene list, derived from1.0 g of biotin-only labeled RNA, was used in all pairwise correlations made with lists generated using reduced amounts of starting material. The use of at least 1.0 g of biotin-only labeled total RNA is required by the manufacturer and is thus defined as the standard. *Filters to the 1.0 g biotin-only miRNA gene list included the following: 0 ⫽ none, ie, all probes on the array; 1 ⫽ human probes only; 2 ⫽ filter 1 ⫹ probe must be present in at least one tissue; 3 ⫽ filter 2 ⫹ absolute fold change ⱖ1.5; and 4 ⫽ filter 3 ⫹ false discovery rate P ⬍ 0.05. All correlations were statistically significant (P ⬍ 0.0001; data not shown).
RNA Labeling For MicroRNA Profiling 17 JMD January 2012, Vol. 14, No. 1
Figure 4. Correlation of microarray data using TaqMan MicroRNA assays. Selection of human miRNAs based on statistically significant differential expression 1) in the brain (one high, two lows) and lung (one high, one low), and 2) found exclusively using either Biotin-HSR labeled RNA (one high, one low) or Biotin-Only labeled RNA (two lows). Two additional miRNAs that were not statistically significant by either labeling method (one high, one low), and one low detected using Biotin-only labeled RNA, were found to be undetectable by TaqMan Assay and therefore excluded. Eight miRNAs were used to generate these graphs. Correlations were made using both Biotin-Only (left column) and Biotin-HSR (right column) labeled RNA. Fold change values and names of each miRNA can be found in Supplemental Table S3 (available at http://jmd.amjpathol.org). Linear regression line determined using least-squares method.
18 Kaddis et al JMD January 2012, Vol. 14, No. 1
found to be statistically significant in one or more, but not all, biotin-HSR miRNA gene lists; 122 were common to all biotin-HSR miRNA gene lists; and 14 were unique to the biotin-only miRNA gene list. To further investigate the differences in these miRNA gene lists, a cross-platform analysis was undertaken comparing microarray results to TaqMan MiRNA assays for 11 selected human miRNAs (Figure 4). Brain-to-lung expression ratios derived from microarrays using the biotin-HSR kit better correlated with TaqMan assay results than did data from chips using biotin-only labeled RNA. Pearson’s correlation coefficient improved from 0.915 to 0.986 for 0.1 g, 0.964 to 0.995 for 0.2 g, 0.975 to 0.994 for 0.5 g, and 0.994 to 0.996 for 1.0 g samples. The use of expression ratios to correlate microarray with TaqMan data has been previously demonstrated.67
Linear Dynamic Range (Synthetic miRNA Spike-In Experiments) To ensure that the use of the biotin-HSR labeling kit did not adversely affect the linearity of the signal intensity values, the concentration range in which valid brain and lung sample measurements can be made was examined by using spike-in synthetic miRNAs at increasing equimolar amounts (Figure 5). The change in amount of spike-in used on the array varied by three orders of magnitude from a low of 4.88 to a high of 5000 attomoles (0-attomole sample used as control). The change in signal intensity values across this range was found to be linear for human (r ⫽ 0.991), mouse (r ⫽ 0.989), and rat (r ⫽ 0.987) synthetic spike-in miRNAs. Signal intensity values from brain and lung samples of the same probes used to detect the synthetic human miRNAs were also plotted on the same graph and shown to fall within the linear dynamic range tested.
plemental Tables S1 and S2 at http://jmd.amjpathol.org). These values were consistent with other studies that examined miRNA replicate chip signal intensity correlations.9,10,13 Second, improvements in the grand median CV values between labeling kits for each tissue was also observed (Figure 2). Although the reduction in CVs was statistically significant only for brain samples, median CV values never exceeded 4%, regardless of the amount of input RNA or labeling kit used. This is in stark contrast to data from Sato et al,9 who found median CVs across five different miRNA microarrays, using two different tissues, to range from 15% to 100%. Likewise, detection calls were also improved by labeling of RNA with the biotinHSR kit (see Supplemental Figure S1 at http://jmd. amjpathol.org). The fact that these improvements were not statistically significant suggest that both kits performed equally well, with values ranging from 93.3%⫺98.4% concordant. This data were generally superior to the numbers reported by Sato et al,9 as well as the percentages in the MicroArray Quality Control project,67 although that study examined protein coding gene expression profiling. Taken together, this data shows that both the biotin-only and biotin-HSR labeling approaches are reliable, and that changes to the labeling reagent in the biotin-HSR improves reproducibility of the data. This is the first study to show that changes to the miRNA labeling reagent allows a reduction in the amount of starting material used. First, it was shown that the number of present calls for human miRNA genes was reduced by only a mean value of 7% and 5% for brain and lung samples, respectively, when the amount of biotin-HSR labeled RNA used was decreased by 10-fold from 1.0 g to 0.1 g (Figure 1). However, the decline was increased to a mean value of 31% and 25% respectively, when using the biotin-only labeled RNA. This large detection loss seen using biotin-only labeling was similar
Discussion Previous studies in the mRNA literature have shown that when laboratories are left to use their own labeling protocols, the comparability of microarray data within and across centers is fair to low,68 but can be improved when standardizing technical, experimental, and analytical factors.67 This is the first study to examine the issue in small noncoding RNAs. Here, we analyzed miRNA expression in human brain and lung samples and evaluated the linear dynamic range of the Affymetrix miRNA GeneChip. Our objective was to determine the differences, using two RNA labeling kits, in the reliability and validity of miRNA expression profiling data using recommended and reduced amount of input RNA. With the exception of the ligation mix, each of the two kits used contained identical reagents. Several improvements to the quality of data were seen when different RNA labeling kits were used. First, a statistically significant increase in the mean correlation coefficient for technical replicates in the brain and lung were seen when comparing samples labeled using the biotin-HSR versus biotin-only kits (Table 1; see also Sup-
Figure 5. Dynamic range of the miRNA GeneChip using biotin-HSR labeling kit. An equimolar pool of synthetic human, mouse, and rat miRNA was labeled using the Biotin-HSR kit and hybridized in amounts ranging from 4.88 to 5000 attomoles (0 datapoint not shown; 12 arrays total). Probes for 465 of 470 human, 223 of 224 mouse, and 42 of 42 rat synthetic miRNAs were detected in all hybridizations (rat and mouse data not shown). Global mean (⫾ SEM) represented by X and error bars, respectively. Linear regression line in black. Intensity values from the same human probes detected in all brain or lung hybridizations (0.1 to 1.0 g) were plotted as black squares or white triangles, respectively. Brain and lung data plotted as grand mean (⫾ grand SEM).
RNA Labeling For MicroRNA Profiling 19 JMD January 2012, Vol. 14, No. 1
to that observed by Lynch et al., who reduced input RNA requirements by 2.5-, 5-, and 25-fold and found a 14%, 14%, and 33% decrease, respectively, in expression detection sensitivity.61 These data show that detection calls using as little as 0.1 g of biotin-HSR labeled RNA were comparable to those generated when using 1.0 g of biotin-only labeled RNA. When considering that as much as 5 g of total RNA were required to produce reliable signatures,69 a reduction in the amount of input RNA is of particular relevance, especially in cases in which nonrenewable clinical specimens are used for miRNA microarray profiling. Next, differentially expressed miRNA gene lists, from reduced input RNA amounts, were correlated with the one derived using 1.0 g total RNA labeled with the biotin-only kit, which was the standard recommended amount by Affymetrix/Genisphere. We found that, in contrast to the miRNA gene list correlations using reduced amounts of biotin-only labeled RNA, variation in miRNA gene rankings were minimized when using reduced amounts of biotin-HSR labeled RNA, suggesting that the use of less starting material does not affect the order of identified genes on a list (Table 2). Finally, when directly comparing filtered miRNA gene lists, generated using different amounts of biotin-HSR labeled RNA (Figure 3), at least 80% of the genes from each list were differentially co-detected in one or more samples of varying input RNA amounts, indicating a high degree of overlap regardless of the amount of starting material used. A cross-platform validation was undertaken to ensure the integrity of the microarray data. Although next generation sequencing technologies could have been used,11 we choose quantitative real-time PCR because it is widely accepted as a gold standard for cross-platform comparison1 and currently offers a time and cost savings over other methods. Eleven human miRNAs were selected based on statistically significant high/low differential expression values in both brain and lung samples, as well as a small subset with no difference (Figure 4). Much like the reports from some9,22,67 but not all11 investigators, we found good correlation between TaqMan and microarray data. Furthermore, the correlation values remained ⬎0.98 for all comparisons made using microarray values from any of the RNA dilutions labeled using the biotin-HSR. In contrast, although still high, the correlations fell to a minimal value of 0.915 when evaluating biotin-only labeled RNA samples. Subsequently, our spike-in experiments showed that the dynamic range of the microarray remained linear for equimolar increases in synthetic miRNAs that varied by three orders of magnitude. Similar results were reported by Wang and company, who demonstrated linearity over two orders of magnitude using the Agilent microarray platform.70 These experiments were performed to ensure that the signal intensity values in our experiments did not exceed the linearity limits of the microarray when using the new RNA labeling kit. Finally, after the conclusion of these experiments and analysis, the GeneChip miRNA 2.0 array was released by Affymetrix to expand on coverage of 131 organisms (vs. 71 on the original array) using Sanger miRNA database
v15 (vs. v11) and to add 2202 probesets unique to premiRNA hairpins (vs. 0). Although it was not tested, as that was not the primary objective of this study, the biotin-HSR kit is not limited to the original chip type, as it is now the recommended strategy to label RNA for the 2.0 array. It should also be noted that, currently, this kit has been optimized only for short RNAs such as miRNAs, and is not developed for longer RNA transcripts. Results from this study highlight RNA labeling as an important factor that can be used as a means of reducing input RNA requirements and improving data quality for miRNA microarray expression profiling studies. Labeling reagent modifications can result in time and cost savings, by eliminating or minimizing the need for preamplification strategies, and enables the profiling of clinical samples with limited quantities of total RNA. The biotin-HSR labeling kit is reliable, is valid, benchmarks well with the currently accepted gold-standard quantitative real-time PCR, and is an improvement over the previous generation of Genisphere miRNA labeling reagents.
Acknowledgments Precision Biomarker Resources performed all of the hybridization, washing, and scanning steps for the RNA titration experiments. The authors thank James Cravens and Stacy Berger (City of Hope) for their input in data review and presentation, and Frank Staedtler (Novartis Institute of Biomedical Research), for his reading and critical review of the manuscript.
References 1. Benes V, Castoldi M: Expression profiling of microRNA using realtime quantitative PCR, how to use it and what is available. Methods 2010, 50:244 –249 2. Creighton CJ, Reid JG, Gunaratne PH: Expression profiling of microRNAs by deep sequencing. Brief Bioinformatics 2009, 10:490 – 497 3. Liu CG, Spizzo R, Calin GA, Croce CM: Expression profiling of microRNA using oligo DNA arrays. Methods 2008, 44:22–30 4. William K, Jian-Jun Z, Lili H, Jin QC: Strategies for profiling MicroRNA expression. J Cell Physiol 2009, 218:22–25 5. Long Q, Johnson BA, Osunkoya AO, Lai YH, Zhou W, Abramovitz M, Xia M, Bouzyk MB, Nam RK, Sugar L, Stanimirovic A, Williams DJ, Leyland-Jones BR, Seth AK, Petros JA, Moreno CS: Protein-coding and microRNA biomarkers of recurrence of prostate cancer following radical prostatectomy. Am J Pathol 2011, 179:46 –54 6. Ono K, Kuwabara Y, Han J: MicroRNAs and cardiovascular diseases. FEBS J 2011, 278:1619 –1633 7. Caporali A, Meloni M, Vollenkle C, Bonci D, Sala-Newby GB, Addis R, Spinetti G, Losa S, Masson R, Baker AH, Agami R, le Sage C, Condorelli G, Madeddu P, Martelli F, Emanueli C: Deregulation of microRNA-503 contributes to diabetes mellitus-induced impairment of endothelial function and reparative angiogenesis after limb ischemia. Circulation 2011, 123:282–291 8. Sah S, McCall MN, Eveleigh D, Wilson M, Irizarry RA: Performance evaluation of commercial miRNA expression array platforms. BMC Res Notes 2010, 3:80 9. Sato F, Tsuchiya S, Terasawa K, Tsujimoto G: Intra-platform repeatability and inter-platform comparability of microRNA microarray technology. PLoS One 2009, 4:e5540 10. Chen Y, Gelfond J, McManus L, Shireman P: Reproducibility of quantitative RT-PCR array in miRNA expression profiling and comparison with microarray analysis. BMC Genomics 2009, 10:407 11. Git A, Dvinge H, Salmon-Divon M, Osborne M, Kutter C, Hadfield J, Bertone P, Caldas C: Systematic comparison of microarray profiling,
20 Kaddis et al JMD January 2012, Vol. 14, No. 1
12.
13.
14.
15.
16.
17.
18. 19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
real-time PCR, and next-generation sequencing technologies for measuring differential microRNA expression. RNA 2010, 16:991– 1006 Pradervand S, Weber J, Lemoine F, Consales F, Paillusson A, Dupasquier M, Thomas J, Richter H, Kaessmann H, Beaudoing E, Hagenbuchle O, Harshman K: Concordance among digital gene expression, microarrays, and qPCR when measuring differential expression of microRNAs. Biotechniques 2010, 48:219 –222 Willenbrock H, Salomon J, Sokilde R, Barken KB, Hansen TN, Nielsen FC, Moller S, Litman T: Quantitative miRNA expression analysis: comparing microarrays with next-generation sequencing. RNA 2009, 15:2028 –2034 Yauk CL, Rowan-Carroll A, Stead JD, Williams A: Cross-platform analysis of global microRNA expression technologies. BMC Genomics 2010, 11:330 Nelson PT, Wang WX, Wilfred BR, Tang G: Technical variables in high-throughput miRNA expression profiling: much work remains to be done. Biochim Biophys Acta 2008, 1779:758 –765 Xi Y, Nakajima G, Gavin E, Morris CG, Kudo K, Hayashi K, Ju J: Systematic analysis of microRNA expression of RNA extracted from fresh frozen and formalin-fixed paraffin-embedded samples. RNA 2007, 13:1668 –1674 Szafranska AE, Davison TS, Shingara J, Doleshal M, Riggenbach JA, Morrison CD, Jewell S, Labourier E: Accurate molecular characterization of formalin-fixed. paraffin-embedded tissues by microRNA expression profiling. J Mol Diagn 2008, 10:415– 423 Hoiby PE, Watzele M: miRNA extraction and profiling on FFPE samples. Biochemica 2008, 3:23–25 Li J, Smyth P, Flavin R, Cahill S, Denning K, Aherne S, Guenther SM, O’Leary JJ, Sheils O: Comparison of miRNA expression patterns using total RNA extracted from matched samples of formalin-fixed paraffin-embedded (FFPE) cells and snap frozen cells. BMC Biotechnol 2007, 7:36 Zhang X, Chen J, Radcliffe T, Lebrun DP, Tron VA, Feilotter H: An array-based analysis of microRNA expression comparing matched frozen and formalin-fixed paraffin-embedded human tissue samples. J Mol Diagn 2008, 10:513–519 Abdueva D, Wing M, Schaub B, Triche T, Davicioni E: Quantitative expression profiling in formalin-fixed paraffin-embedded samples by affymetrix microarrays. J Mol Diagn 2010, 12:409 – 417 Ach RA, Wang H, Curry B: Measuring microRNAs: comparisons of microarray and quantitative PCR measurements, and of different total RNA prep methods. BMC Biotechnol 2008, 8:69 Doleshal M, Magotra AA, Choudhury B, Cannon BD, Labourier E, Szafranska AE: Evaluation and validation of total RNA extraction methods for microRNA expression analyses in formalin-fixed, paraffin-embedded tissues. J Mol Diagn 2008, 10:203–211 Hammerle-Fickinger A, Riedmaier I, Becker C, Meyer HH, Pfaffl MW, Ulbrich SE: Validation of extraction methods for total RNA and miRNA from bovine blood prior to quantitative gene expression analyses. Biotechnol Lett 2010, 32:35– 44 Wang WX, Wilfred BR, Baldwin DA, Isett RB, Ren N, Stromberg A, Nelson PT: Focus on RNA isolation: obtaining RNA for microRNA (miRNA) expression profiling analyses of neural tissue. Biochim Biophys Acta 2008, 1779:749 –757 Bravo V, Rosero S, Ricordi C, Pastori RL: Instability of miRNA and cDNAs derivatives in RNA preparations. Biochem Biophys Res Commun 2007, 353:1052–1055 Mraz M, Malinova K, Mayer J, Pospisilova S: MicroRNA isolation and stability in stored RNA samples. Biochem Biophys Res Commun 2009, 390:1– 4 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 Biol 2006, 7:3 Becker C, Hammerle-Fickinger A, Riedmaier I, Pfaffl MW: mRNA and microRNA quality control for RT-qPCR analysis. Methods 2010, 50: 237–243 Ibberson D, Benes V, Muckenthaler MU, Castoldi M: RNA degradation compromises the reliability of microRNA expression profiling. BMC Biotechnol 2009, 9:102 Jung M, Schaefer A, Steiner I, Kempkensteffen C, Stephan C, Erbersdobler A, Jung K: Robust microRNA stability in degraded RNA
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
preparations from human tissue and cell samples. Clin Chem 2010, 56:998 –1006 Zhu B, Xu F, Baba Y: An evaluation of linear RNA amplification in cDNA microarray gene expression analysis. Mol Genet Metab 2006, 87:71–79 Viale A, Li J, Tiesman J, Hester S, Massimi A, Griffin C, Grills G, Khitrov G, Lilley K, Knudtson K, Ward B, Kornacker K, Chu CY, Auer H, Brooks AI: Big results from small samples: evaluation of amplification protocols for gene expression profiling. J Biomol Techn 2007, 18:150 –161 Singh R, Maganti RJ, Jabba SV, Wang M, Deng G, Heath JD, Kurn N, Wangemann P: Microarray-based comparison of three amplification methods for nanogram amounts of total RNA. Am J Physiol Cell Physiol 2005, 288:C1179 –C1189 Patel OV, Suchyta SP, Sipkovsky SS, Yao J, Ireland JJ, Coussens PM, Smith GW: Validation and application of a high fidelity mRNA linear amplification procedure for profiling gene expression. Vet Immunol Immunopathol 2005, 105:331–342 Francois P, Garzoni C, Bento M, Schrenzel J: Comparison of amplification methods for transcriptomic analyses of low abundance prokaryotic RNA sources. J Microbiol Methods 2007, 68:385–391 Nagy ZB, Kelemen JZ, Feher LZ, Zvara A, Juhasz K, Puskas LG: Real-time polymerase chain reaction-based exponential sample amplification for microarray gene expression profiling. Anal Biochem 2005, 337:76 – 83 Puskas LG, Zvara A, Hackler L, Jr., Van Hummelen P: RNA amplification results in reproducible microarray data with slight ratio bias. Biotechniques 2002, 32:1330 –1334, 1336, 1338, 1340 Bak M, Conley L, Hedegaard J, Larsen LA, Sorensen P, Bendixen C, Tommerup N: Evaluation of two methods for generating cRNA for microarray experiments from nanogram amounts of total RNA. Anal Biochem 2006, 358:111–119 Cope L, Hartman SM, Gohlmann HW, Tiesman JP, Irizarry RA: Analysis of Affymetrix GeneChip data using amplified RNA. Biotechniques 2006, 40:165–166, 168, 170 Yu J, Othman MI, Farjo R, Zareparsi S, MacNee SP, Yoshida S, Swaroop A: Evaluation and optimization of procedures for target labeling and hybridization of cDNA microarrays. Mol Vis 2002, 8:130 – 137 Richter A, Schwager C, Hentze S, Ansorge W, Hentze MW, Muckenthaler M: Comparison of fluorescent tag DNA labeling methods used for expression analysis by DNA microarrays. Biotechniques 2002, 33:620 – 628, 630 Ouellet M, Adams PD, Keasling JD, Mukhopadhyay A: A rapid and inexpensive labeling method for microarray gene expression analysis. BMC Biotechnol 2009, 9:97 Manduchi E, Scearce LM, Brestelli JE, Grant GR, Kaestner KH, Stoeckert CJ, Jr.: Comparison of different labeling methods for twochannel high-density microarray experiments. Physiol Genomics 2002, 10:169 –179 Kimura N, Tamura TA, Murakami M: Evaluation of the performance of two carbodiimide-based cyanine dyes for detecting changes in mRNA expression with DNA microarrays. Biotechniques 2005, 38: 797– 806 Badiee A, Eiken HG, Steen VM, Lovlie R: Evaluation of five different cDNA labeling methods for microarrays using spike controls. BMC Biotechnol 2003, 3:23 t Hoen PA, de Kort F, van Ommen GJ, den Dunnen JT: Fluorescent labelling of cRNA for microarray applications. Nucleic Acids Res 2003, 31:e20 Sawada A, Mizufune S, Kaji N, Tokeshi M, Baba Y: Evaluation of amplified cRNA targets for oligonucleotide microarrays. Anal Bioanal Chem 2007, 387:2645–2654 Ma C, Lyons-Weiler M, Liang W, LaFramboise W, Gilbertson JR, Becich MJ, Monzon FA: In vitro transcription amplification and labeling methods contribute to the variability of gene expression profiling with DNA microarrays. J Mol Diagn 2006, 8:183–192 Lonergan W, Whistler T, Vernon SD: Comparison of target labeling methods for use with Affymetrix GeneChips. BMC Biotechnol 2007, 7:24 Zakharkin SO, Kim K, Mehta T, Chen L, Barnes S, Scheirer KE, Parrish RS, Allison DB, Page GP: Sources of variation in Affymetrix microarray experiments. BMC Bioinformat 2005, 6:214
RNA Labeling For MicroRNA Profiling 21 JMD January 2012, Vol. 14, No. 1
52. Ach RA, Floore A, Curry B, Lazar V, Glas AM, Pover R, Tsalenko A, Ripoche H, Cardoso F, d’Assignies MS, Bruhn L, Van’t Veer LJ: Robust interlaboratory reproducibility of a gene expression signature measurement consistent with the needs of a new generation of diagnostic tools. BMC Genomics 2007, 8:148 53. Tissot C: Application Note: Analysis of miRNA content in total RNA preparations using the Agilent 2100 bioanalyzer. Agilent Technologies, Santa Clara, CA. Publication number 5989-7870EN 2008, 1– 8 54. Weber JA, Baxter DH, Zhang S, Huang DY, How Huang K, Jen Lee M, Galas DJ, Wang K: The MicroRNA spectrum in 12 body fluids. Clin Chem 2010, 56:1733–1741 55. Mora JR, Getts RC: High-sensitivity detection methods for low-abundance RNA species: applications for functional genomics research. Expert Rev Mol Diagn 2007, 7:775–785 56. Saba R, Booth S: Target labelling for the detection and profiling of microRNAs expressed in CNS tissue using microarrays. BMC Biotechnol 2006, 6:47 57. Wu Z, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F: A model-based background adjustment for oligonucleotide expression arrays. J Am Stat Assoc 2004, 99:909 –917 58. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 2003, 4:249 –264 59. Rao Y, Lee Y, Jarjoura D, Ruppert AS, Liu CG, Hsu JC, Hagan JP: A comparison of normalization techniques for microRNA microarray data. Stat Appl Genet Mol Biol 2008, 7:Article22 60. Pradervand S, Weber J, Thomas J, Bueno M, Wirapati P, Lefort K, Dotto GP, Harshman K: Impact of normalization on miRNA microarray expression profiling. RNA 2009, 15:493–501 61. Lynch A, Hadfield J, Dunning M, Osborne M, Thorne N, Tavare S: The cost of reducing starting RNA quantity for Illumina BeadArrays: a bead-level dilution experiment. BMC Genomics 2010, 11:540 62. Application Note: TaqMan MicroRNA assays: Endogenous controls for real-time quantitation of miRNA using TaqMan MicroRNA assays. Applied Biosystems, Foster City, CA. publication 127AP11-01 2007, 1– 8 63. Peltier HJ, Latham GJ: Normalization of microRNA expression levels in quantitative RT-PCR assays: identification of suitable reference
64.
65.
66.
67.
68.
69.
70.
RNA targets in normal and cancerous human solid tissues. RNA 2008, 14:844 – 852 Andersen CL, Jensen JL, Orntoft TF: Normalization of real-time quantitative reverse transcription-PCR data: a model-based variance estimation approach to identify genes suited for normalization, applied to bladder and colon cancer data sets. Cancer Res 2004, 64:5245– 5250 Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, Speleman F: Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol 2002, 3:(7)research0034.1-research0034.11 Livak KJ, Schmittgen TD: Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) method. Methods 2001, 25:402– 408 Shi L, Reid LH, Jones WD, Shippy R, Warrington JA, Baker SC, Collins PJ, de Longueville F, Kawasaki ES, Lee KY, Luo Y, Sun YA, Willey JC, Setterquist RA, Fischer GM, Tong W, Dragan YP, Dix DJ, Frueh FW, Goodsaid FM, Herman D, Jensen RV, Johnson CD, Lobenhofer EK, Puri RK, Schrf U, Thierry-Mieg J, Wang C, Wilson M, Wolber PK, Zhang L, Amur S, Bao W, Barbacioru CC, Lucas AB, Bertholet V, et al: The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nature Biotechnol 2006, 24:1151–1161 Bammler T, Beyer RP, Bhattacharya S, Boorman GA, Boyles A, Bradford BU, Bumgarner RE, Bushel PR, Chaturvedi K, Choi D, Cunningham ML, Deng S, Dressman HK, Fannin RD, Farin FM, Freedman JH, Fry RC, Harper A, Humble MC, Hurban P, Kavanagh TJ, Kaufmann WK, Kerr KF, Jing L, Lapidus JA, Lasarev MR, Li J, Li YJ, Lobenhofer EK, Lu X, Malek RL, Milton S, Nagalla SR, O’Malley JP, Palmer VS, Pattee P, et al: Standardizing global gene expression analysis between laboratories and across platforms. Nature Methods 2005, 2:351–356 Calin GA, Liu C-G, Sevignani C, Ferracin M, Felli N, Dumitru CD, Shimizu M, Cimmino A, Zupo S, Dono M, Dell’Aquila ML, Alder H, Rassenti L, Kipps TJ, Bullrich F, Negrini M, Croce CM: MicroRNA profiling reveals distinct signatures in B cell chronic lymphocytic leukemias. Proc Natl Acad Sci U S Am 2004, 101:11755–11760 Wang H, Ach RA, Curry B: Direct and sensitive miRNA profiling from low-input total RNA. RNA 2007, 13:151–159
Technical Advance Influence of RNA Labeling on Expression Profiling of MicroRNAs
John S. Kaddis,* Daniel H. Wai,* Jessica Bowers,† Nicole Hartmann,‡ Lukas Baeriswyl,‡ Sheetal Bajaj,* Michael J. Anderson,* Robert C. Getts,† and Timothy J. Triche*§ From the Departments of Pathology and Laboratory Medicine,* Children’s Hospital Los Angeles Saban Research Institute and Keck School of Medicine, University of Southern California, Los Angeles, California; Genisphere LLC,† Hatfield, Pennsylvania; the Novartis Institutes for BioMedical Research,‡ Novartis, Basel, Switzerland; and the USC/CHLA Genome Core Laboratory,§ Children Hospital Los Angeles, Los Angeles, California
Although a number of technical parameters are now being examined to optimize microRNA profiling experiments, it is unknown whether reagent or component changes to the labeling step affect starting RNA requirements or microarray performance. Human brain/lung samples were each labeled in duplicate, at 1.0, 0.5, 0.2, and 0.1 g of total RNA, by means of two kits that use the same labeling procedure but differ in the reagent composition used to label microRNAs. Statistical measures of reliability and validity were used to evaluate microarray data. Cross-platform confirmation was accomplished using TaqMan microRNA assays. Synthetic microRNA spike-in experiments were also performed to establish the microarray signal dynamic range using the ligation-modified kit. Technical replicate correlations of signal intensity values were high using both kits, but improved with the ligation-modified assay. The drop in detection call sensitivity and miRNA gene list correlations, when using reduced amounts of standard-labeled RNA, was considerably improved with the ligation-modified kit. Microarray signal dynamic range was found to be linear across three orders of magnitude from 4.88 to 5000 attomoles. Thus, optimization of the microRNA labeling reagent can result in at least a 10-fold decrease in microarray total RNA requirements with little compromise to data quality. Clinical investigations bottlenecked by the amount of starting material may use a ligation mix modification strategy to reduce total
12
RNA requirements. ( J Mol Diagn 2012, 14:12–21; DOI: 10.1016/j.jmoldx.2011.08.005)
MicroRNA (miRNA) expression profiling platforms have burgeoned in the last decade1– 4 and promise to aid in the diagnosis, prognosis, and therapeutic treatment of human diseases such as cancer,5 cardiovascular disease,6 and diabetes.7 In response, intra-, inter-, and cross-platform comparisons have recently emerged that examine the use of microarray technology alone8,9 or in contrast to other profiling methods.10 –14 Such studies have reported on the reliability and validity of widely used miRNA discovery tools and to highlight the advantages and limitations of these platforms. However, there also remains a need to evaluate preexperimental factors, such as sample handling, processing, storage, and nucleic acid quality, that may influence miRNA expression profiling experiments.15 Whereas three studies have demonstrated that miRNA detection was possible in formalin-fixed paraffin-embedded (FFPE) samples of up to 1016 or 1217 years old using extraction kits from different manufacturers,18 stability and comparability of miRNA expression signal using paired freshfrozen (FF) versus FFPE samples was reported in human and murine specimens using different tissues ranging in age from newly acquired19 to up to 3 years of age.17,20 In contrast, when detecting the expression of longer RNA transcripts, Abdueva et al21 found differences in paired FF versus FFPE samples of up to 7.5 years of age, but Supported in part by the National Cancer Institute, Strategic Partnering to Evaluate Cancer Signatures (U01CA-114757, to T.J.T.). Accepted for publication August 22, 2011. Disclosures: R.C.G. and J.B. are employed by Genisphere LLC and provided the microarrays and corresponding reagents for the RNA titration experiments. N.H. and L.B. are employed by Novartis and provided the microarrays and corresponding reagents for the synthetic spike-in experiments. Supplemental material for this article can be found at http://jmd. amjapthol.org or at doi: 10.1016/j.jmoldx.2011.08.005. Address reprint requests to Timothy J. Triche, M.D., Ph.D., Department of Pathology, Children’s Hospital Los Angeles, 4650 Sunset Blvd., Los Angeles, CA 90027. E-mail: [email protected].
RNA Labeling For MicroRNA Profiling 13 JMD January 2012, Vol. 14, No. 1
concluded that useable profiles were obtainable. A number of studies examined RNA extraction methods for miRNA, with one that compared three isolation kits and found general comparability among them,22 two that used five kits and identified a preferred approach,23,24 and one that evaluated eight kits and found R2 correlations exceeding 0.857 in all comparisons made.25 Conflicting reports have surfaced on the length of time RNA can be stored after extraction for miRNA profiling, with one study describing degradation after 3 days26 and another demonstrating stability of up to 10 months,27 although cells from different species were used in each study. Likewise, the quality of RNA28 has also been called into question, with some showing that medium- to high-integrity samples are required for reproducible miRNA profiling using bovine29 and murine30 samples, and another reporting negligible effects on highly degraded human samples.31 Consideration of these experimental factors is crucial to ensuring valid and reliable miRNA expression profiling results within and across laboratories. With the exception of next-generation sequencing (NGS), all miRNA expression profiling approaches use a labeling method to tag the target molecules of interest. In the case of transcripts greater than 200 nucleotides in length, the method used to amplify and/or label the molecules of interest before detection has been shown to be important. Studies examining linear,32–36 exponential,37 and linear versus exponential38,39 amplification methods revealed that although total RNA requirements can be reduced significantly by using amplified RNA, quality metrics such as comparability, sensitivity, and accuracy were, in some cases, compromised when compared with those using nonamplified RNA. However, these measures can be improved when robust data methods were used.40 Next, evaluation of direct, indirect, and other cDNA labeling approaches have shown that the method chosen can reduce total RNA requirements, but that gains in signal intensity values were also achieved alongside, in some cases, a loss in labeling performance.41– 46 Similar observations have been made when comparing cRNA labeling methods.47–50 Even when performing replicate experiments within51 or across52 laboratories, RNA labeling has been shown to contribute to the variability seen in gene expression outcome. Although the body of work examining target labeling is extensive, it is unclear whether and how these findings can be applied to low abundance short transcripts ⬍200 nucleotides. This is especially true in the case of miRNAs, when considering that they represent only 0.5% to 9.2% of total RNA in human and 0.1% to 1.3% in mouse and rat tissue samples,53 but have also been shown to be present, at some level, in a variety of human bodily fluids.54 Because many studies are limited in the amount of total RNA available, such as those using tissues derived from needle biopsies or laser capture microdissection, more sensitive target labeling procedures are critical. Accordingly, a growing number of amplification and labeling strategies have been developed to aid in the detection of these low-abundance RNAs.55 Although there are at least two studies that have examined the effects of amplifica-
tion on miRNA expression,10,56 there are currently no studies that compare amplification or labeling effects when direct labeling of mature miRNA is used. We therefore conducted a study to directly compare the Genisphere FlashTag biotin-HSR (biotin-HSR) labeling kit to the previously recommended FlashTag biotin-only (biotin-only) assay (Genisphere, Hatfield, PA) for the Affymetrix GeneChip miRNA Arrays. The biotin-HSR kit is a modified version of the biotin-only assay. More specifically, a smaller more efficient labeling molecule, having a higher number of biotins per unit mass of DNA, has been used in the biotin-HSR kit compared to the DNA dendrimer of the original FlashTag biotin kit. A total of 32 chips were used for this evaluation, followed by crossplatform validation using TaqMan miRNA assays. In addition, the linear dynamic range of the GeneChip was assessed using synthetic miRNA spiked-in at different amounts, labeled with the biotin-HSR kit, on 12 chips. A total of 44 microarray chips were used in this study.
Materials and Methods Human Normal Tissue RNA FirstChoice Total Brain (Lot#0906005) and Lung (Lot#0904002) RNA samples were obtained from Ambion, Inc., certified to contain small RNAs, and quantified at 1 mg/ml (Applied Biosystems/Ambion, Foster City, CA).
Labeling of RNA and Microarray Processing Brain and lung total RNA samples were each labeled using both the FlashTag biotin-only and FlashTag biotin-HSR RNA labeling kits (Genisphere, Hatfield, PA) for the Affymetrix GeneChip miRNA array. Each microarray contains sequences for 6703 miRNA probes, 499 small nucleolar RNA (snoRNA), including 274 C/D and 127 H/ACA box RNA, 22 small cajal body–specific RNA (scaRNA), 10 5.8S ribosomal RNA, 162 Affymetrix controls, and 22 oligonucleotide spike-in controls. Probes were developed by Affymetrix using the Sanger miRNA database v11 and contained additional snoRNAs and scaRNAs from the Ensembl database and snoRNABase. RNA labeling was performed by the kit manufacturer and efficiency assessed using an enzymelinked oligosorbent assay. Hybridization, washing, and scanning of slides were performed according to the Affymetrix and Genisphere protocols. Scanning was performed using the miRNA-1_0 library file from Affymetrix for the Flashtag biotin-only and miRNA-1_0_2Xgain for the FlashTag biotin-HSR labeled samples. Refer to NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) study GSE30045 to access Affymetrix CEL file data.
RNA Titration Experiments Total RNA was prepared at recommended and reduced input amounts from two different tissues to examine the performance of the Affymetrix GeneChip miRNA array and to evaluate the significance of using a new RNA
14 Kaddis et al JMD January 2012, Vol. 14, No. 1
labeling kit on the quality of data generated. Brain samples were prepared containing 0.1 g, 0.2 g, 0.5 g, and 1.0 g of total RNA. Two aliquots from each sample were taken and labeled with either the biotin-only or biotin-HSR kit. This process was carried out in duplicate for each tissue. A total of 32 microarray chips were used for this study.
Synthetic miRNA Spike-In Experiments The mirVana miRNA Reference Panel v9.1 (Lot#072307) was obtained from Ambion, Inc. and contains an equimolar pool of 470 human, 224 murine, and 42 rat synthetic miRNA oligonucleotides according to the Sanger miRBase sequence database, release 9.1 (Applied Biosystems/Ambion, Foster City, CA). Twofold serial dilution samples were then prepared that contained from 5000 to 4.88 attomoles of miRNA (12 total, including a negative control). Synthetic miRNAs were labeled using the FlashTag biotin-HSR RNA labeling kits. Hybridization, washing, and scanning of slides were performed by Novartis according to Affymetrix and Genisphere protocols. Scanning was performed using the miRNA-1_0_2Xgain library file from Affymetrix.
Array Quality, Data Processing, and Detection Calls Raw intensity signal values of Affymetrix spike-in controls indicated that array hybridization was successful (ie, bioB⬍bioC⬍bioD⬍Cre). Likewise, raw intensity values from an additional five Genisphere oligonucleotide spikeins control probes were all found to be ⬎1000, used to demonstrate poly(A) tailing and ligation (3 RNA oligos), ligation (1 poly(A) RNA), and ligation and lack of RNases in the RNA sample (1 poly(dA) DNA; data not shown). Because labeling was an experimental factor being tested, before analysis, raw signal intensity values were Robust Multichip Average (RMA) background corrected, quantile normalized, median polish summarized, and log2 transformed separately for FlashTag biotin-only and FlashTag biotin-HSR chip data.57– 61 Processing was performed using Partek Genomics Suite, version 6.5, build 6.10.0412 copyright 2010 (Partek, St. Louis, MO). Present/absent detection calls for each probe, on every array, were generated using the Affymetrix detection algorithm and statistical significance testing implemented through the miRNA QCTool, version 1.0.33.0. Details on how the algorithm determines these calls are described in Appendix A of the Affymetrix miRNA QC Tool User’s Guide 6 that, along with the software utility, are freely available on the manufacturer’s website.
Reverse Transcription and Real-Time PCR Total RNA was converted into cDNA and quantified using individual Taqman assay kits for each miRNA of interest (Applied Biosystems, Foster City, CA). Real-time PCR followed the reverse transcription (RT) reaction, which was modified to allow for the simultaneous synthesis of
cDNA by combining RT primers into pools A and B. The multiplex RT protocol is a modification of the original kit instructions and made available through the manufacturer by request. RNU6B, SNORD44, and SNORD48 were selected as candidate artificial normalization controls and previously shown to be stable in a panel of 38 normal human tissues, including Brain and Lung.62 Identification of hsa-let-7d and hsa-miR-151-5p as candidate endogenous human miRNA normalization controls, specific to the present experiment, was performed, using microarray data, in the following way: i) A signal detection call of “present” in both Brain and Lung tissue across all input RNA titrations and replicates using both the old and new labeling methods (79 of 847 miRNA probesets remained); ii) the absence of a statistically significant fold change difference between brain and lung tissue (20 of 79 remained); iii) removal of half the probesets with the lowest mean signal intensity values and those previously shown63 as inadequate (eight of 20 remained); and iv) use of NormFinder64 to rank (data not shown) and select the most stable miRNAs (two of eight remained). Subsequent analysis of all five candidate control Ct values by geNorm65 identified the combination of no more than two miRNAs, ie, hsa-miR-151-5p/SNORD44 and hsa-let-7d/SNORD44, as the most stably expressed housekeeping miRNAs in pools A and B, respectively (data not shown). The expression of miRNAs, relative to selected controls, was determined using the 2(–⌬⌬Ct) method,66 and as also described online in an Applied Biosystems document titled “User Bulletin #2: ABI Prism 7700 Sequence Detection System.”
Data Analysis The mean or median was used as a measure of central tendency for all continuous numerical variables, depending on whether the values followed a normal Gaussian probability distribution. Student’s t-test was used when performing two-group comparisons using parametric data. An F-test for equality of variance was performed before significance testing. A Wilcoxon rank-sum test was used when performing two-group comparisons using nonparametric data. For descriptive statistics, significance was indicated if P was ⬍0.05. Coefficient of variation (CV) for replicate chips was calculated as the SD of the processed signal intensity value over the mean for miRNAs called as present. The Pearson product-moment correlation coefficient (r) was used to measure the linear relationship of signal intensity values between replicate chips. Spearman’s rank correlation coefficient () was used to evaluate intra- and cross-platform miRNA gene list comparisons. Unlike r, measures the relative rank of a miRNA gene in list A against its relative rank in list B, thereby minimizing the effects of outlier data. For both r and , the null hypothesis of no relationship was indicated if P was ⬎0.05. Statistical significance for differential miRNA gene expression was determined using a false discovery rate (FDR) P value of ⱕ0.05, unless otherwise noted. Data analysis was performed using either Partek
RNA Labeling For MicroRNA Profiling 15 JMD January 2012, Vol. 14, No. 1
Genomics Suite or SAS software, version 9.1.3 SP 4 (SAS Institute, Cary, NC).
Results Reliability (RNA Titrations) Qualitative and quantitative measures of random error within and between labeling kits were examined at all input amounts of total RNA from both brain and lung tissue. The number of present and absent detection calls was first calculated for all replicate chips (see Supplemental Figure S1 at http://jmd.amjpathol.org). Percent concordant (present and absent) calls for all probes on the array using total RNA labeled with the biotin-only kit was found to range from a grand mean of 95.0% for brain to 95.8% for lung. Those values increased to 96.6% and 97.4%, respectively, when the RNA was labeled using the biotin-HSR kit, although this was not a statistically significant finding (P ⬎ 0.05 for both comparisons made). On restricted examination of only the 847 human miRNA gene probes, a statistically significant increase in the number of detection calls was seen when concurrently examining the differences in labeling kits alongside the amount of input RNA used for microarray hybridization (Figure 1). For brain samples, as the amount of biotin-only labeled RNA is reduced from 1.0 g to 0.1 g, the mean percentage of detected overlapping probes was decreased by 31%. However, this was statistically significantly improved to a mean value of only 7% when using biotin-HSR labeled RNA (P ⫽ 0.0490). Likewise, for lung samples, the loss was statistically significantly improved from a mean value of 25% to 5% (P ⫽ 0.04). Irrespective of detection call, the reproducibility of processed signal intensity values between duplicate chips was next examined (Table 1). A statistically significant improvement in the mean Pearson’s correlation coefficient value was observed for duplicate chips labeled using the biotin-HSR kit. This finding was true for RNA isolated from either brain or lung tissue (P ⬍ 0.0001 for both comparisons made). Individual correlations are provided in a 16 ⫻ 16 correlation matrix table for both brain and lung samples (see Supplemental Tables S1 and S2 at http://jmd.amjpathol.org). Finally, the CVs of the processed signal intensity values were calculated for all study samples (Figure 2). There was a statistically significant reduction in the grand median CV value for brain samples labeled with the biotin-HSR versus biotin-only kit (2.59% to 1.25%; Wilcoxon rank-sum P ⬍ 0.05). However, a statistically significant reduction in the grand median CV value for lung samples was not observed (1.96% to 1.16%; Wilcoxon rank-sum P ⬎ 0.05).
Validity (RNA Titrations and RT Real-Time PCR) Systematic error was evaluated by examining the interplatform agreement and cross-platform concordance of brain to lung expression ratios for selected human miRNAs. First, using differential miRNA gene expression analysis, the numerical rank of all probes from the 1.0 g
Figure 1. Comparison of detected miRNA genes. Probes for all 847 human miRNAs were examined in (A) brain and (B) lung samples. The number of present calls is indicated by the y axis. Comparisons on the x axis were done to determine the number of overlapping present calls on chips containing 1.0 g of input RNA vs. 0.5, 0.2, and 0.1 g of starting material. One vs. 1.0 g indicates the maximum possible overlap. Evaluations using the biotin-only kit are represented by filled circles, biotin-HSR by filled squares, and biotinonly versus biotin-HSR by open triangles. Statistical significance of *P ⱕ 0.05.
biotin-only labeled RNA list was compared to the order from lists using reduced amounts of input RNA (Table 2). This process was repeated following sequential application of filters, defined in Table 2, to the 1.0 g biotin-only miRNA gene list to selectively reduce the number of
Table 1. Correlation of Replicate Chip Signal Intensity Values Pearson product-moment correlation (r) Labeling kit Tissue Brain Lung
Biotin-only global Mean (⫾1 SD)
Biotin-HSR global Mean (⫾1 SD)
P value
0.955 (0.023) 0.976 (0.011)
0.978 (0.010) 0.985 (0.006)
⬍0.0001 ⬍0.0001
The processed signal intensity value for each probe on a chip was correlated with the corresponding probe on a replicate array (7819 probes total). Individual correlations between replicate arrays also were all found to be statistically significant (P ⬍ 0.0001; see Supplemental Tables S1 and S2 at http://jmd.amjpathol.org). Global mean calculations did not include correlations of a chip with itself, ie, where r ⫽ 1.
16 Kaddis et al JMD January 2012, Vol. 14, No. 1
Figure 2. Coefficient of variation within and between RNA labeling kits. The percent CV was averaged for replicate chips at each input RNA amount (0.1–1.0 g) within and between labeling kits for both brain and lung samples. The interquartile range of the processed signal intensity value is represented by shaded boxes, median value by a black line within a box, the 5th and 95th percentiles by the whiskers outside of each box and the number of detected miRNA genes by filled boxes. CVs were calculated only for miRNA genes called as present.
Figure 3. Evaluation of statistically significant, differentially expressed miRNA genes using Biotin-HSR labeled RNA. Genes were included in this analysis using only human miRNA probes detected as present in at least brain or lung tissue, with absolute fold change ⱖ1.5, and an FDR p ⬍0.05. A total of 419, 376, 218, and 184 differentially expressed miRNA genes were identified when using 1.0 g, 0.5 g, 0.2 g, and 0.1 g of input RNA, respectively.
candidate probes evaluated. Regardless of miRNA gene filters used, as the amount of biotin-only labeled RNA is reduced, correlations with the 1.0 g biotin-only miRNA gene list also falls. However, when using reduced amounts of biotin-HSR labeled RNA, gene list correlations remain steady. Depending on the miRNA gene list filter used, there is a 60% to 93% reduction in the variation of correlation values when using reduced amounts of biotin-HSR versus biotin-only labeled RNA. Second, differentially expressed miRNA gene lists were also generated for biotin-HSR labeled samples at each input RNA amount and evaluated for overlap in statistically significant transcripts (Figure 3). Of the 1779 human probes, a total of 474 differentially expressed miRNA genes were identified. As the amount of input RNA increases, the percentage of miRNA genes unique to each list also increases from 4.3% for
Table 2.
0.1 g and 4.1% for 0.2 g to 8.8% for 0.5 g and 16.9% for 1.0 g. The percentage of miRNA genes that were detected in more than 1 list, but not by all, is 22.8% for 0.1 g, 34.4% for 0.2 g, 55.6% for 0.5 g, and 68.0% for 1.0 g. There were 134 miRNA genes robustly detected by all lists with a minimum absolute fold change of at least 10 or more for 86 genes, and two to ⬍10 for the remaining 48 genes. Next, we took the 272 miRNA genes, from Table 2, and compared it with the 474 identified miRNA genes from Figure 3, to determine the gene list effects of using a new labeling kit (see Supplemental Table S3 at http://jmd. amjpathol.org). Of the 272 miRNA genes identified using 1.0 g of biotin-only labeled RNA, 136 of them were also
miRNA Gene List Agreement within and Between RNA Labeling Kits
Gene list filters applied* (number of probes) Biotin-only kit 0.5 g 0.2 g 0.1 g Range (Max–Min) Biotin-HSR kit 0.5 g 0.2 g 0.1 g Range (Max–Min) % Reduction in range using Biotin-HSR kit
0 (n ⫽ 7819) 0.765 0.693 0.588 0.177 0.791 0.766 0.721 0.070 60
1 (n ⫽ 1779)
2 (n ⫽ 601)
3 (n ⫽ 422)
Spearman Rank Correlation Coefficient 0.734 0.961 0.972 0.662 0.915 0.938 0.558 0.830 0.879 0.176 0.131 0.093 0.762 0.733 0.692 0.070 60
0.962 0.957 0.937 0.025 81
0.968 0.970 0.960 0.010 89
4 (n ⫽ 272) 0.978 0.949 0.903 0.075 0.973 0.977 0.972 0.005 93
The brain versus lung differentially expressed miRNA gene list, derived from1.0 g of biotin-only labeled RNA, was used in all pairwise correlations made with lists generated using reduced amounts of starting material. The use of at least 1.0 g of biotin-only labeled total RNA is required by the manufacturer and is thus defined as the standard. *Filters to the 1.0 g biotin-only miRNA gene list included the following: 0 ⫽ none, ie, all probes on the array; 1 ⫽ human probes only; 2 ⫽ filter 1 ⫹ probe must be present in at least one tissue; 3 ⫽ filter 2 ⫹ absolute fold change ⱖ1.5; and 4 ⫽ filter 3 ⫹ false discovery rate P ⬍ 0.05. All correlations were statistically significant (P ⬍ 0.0001; data not shown).
RNA Labeling For MicroRNA Profiling 17 JMD January 2012, Vol. 14, No. 1
Figure 4. Correlation of microarray data using TaqMan MicroRNA assays. Selection of human miRNAs based on statistically significant differential expression 1) in the brain (one high, two lows) and lung (one high, one low), and 2) found exclusively using either Biotin-HSR labeled RNA (one high, one low) or Biotin-Only labeled RNA (two lows). Two additional miRNAs that were not statistically significant by either labeling method (one high, one low), and one low detected using Biotin-only labeled RNA, were found to be undetectable by TaqMan Assay and therefore excluded. Eight miRNAs were used to generate these graphs. Correlations were made using both Biotin-Only (left column) and Biotin-HSR (right column) labeled RNA. Fold change values and names of each miRNA can be found in Supplemental Table S3 (available at http://jmd.amjpathol.org). Linear regression line determined using least-squares method.
18 Kaddis et al JMD January 2012, Vol. 14, No. 1
found to be statistically significant in one or more, but not all, biotin-HSR miRNA gene lists; 122 were common to all biotin-HSR miRNA gene lists; and 14 were unique to the biotin-only miRNA gene list. To further investigate the differences in these miRNA gene lists, a cross-platform analysis was undertaken comparing microarray results to TaqMan MiRNA assays for 11 selected human miRNAs (Figure 4). Brain-to-lung expression ratios derived from microarrays using the biotin-HSR kit better correlated with TaqMan assay results than did data from chips using biotin-only labeled RNA. Pearson’s correlation coefficient improved from 0.915 to 0.986 for 0.1 g, 0.964 to 0.995 for 0.2 g, 0.975 to 0.994 for 0.5 g, and 0.994 to 0.996 for 1.0 g samples. The use of expression ratios to correlate microarray with TaqMan data has been previously demonstrated.67
Linear Dynamic Range (Synthetic miRNA Spike-In Experiments) To ensure that the use of the biotin-HSR labeling kit did not adversely affect the linearity of the signal intensity values, the concentration range in which valid brain and lung sample measurements can be made was examined by using spike-in synthetic miRNAs at increasing equimolar amounts (Figure 5). The change in amount of spike-in used on the array varied by three orders of magnitude from a low of 4.88 to a high of 5000 attomoles (0-attomole sample used as control). The change in signal intensity values across this range was found to be linear for human (r ⫽ 0.991), mouse (r ⫽ 0.989), and rat (r ⫽ 0.987) synthetic spike-in miRNAs. Signal intensity values from brain and lung samples of the same probes used to detect the synthetic human miRNAs were also plotted on the same graph and shown to fall within the linear dynamic range tested.
plemental Tables S1 and S2 at http://jmd.amjpathol.org). These values were consistent with other studies that examined miRNA replicate chip signal intensity correlations.9,10,13 Second, improvements in the grand median CV values between labeling kits for each tissue was also observed (Figure 2). Although the reduction in CVs was statistically significant only for brain samples, median CV values never exceeded 4%, regardless of the amount of input RNA or labeling kit used. This is in stark contrast to data from Sato et al,9 who found median CVs across five different miRNA microarrays, using two different tissues, to range from 15% to 100%. Likewise, detection calls were also improved by labeling of RNA with the biotinHSR kit (see Supplemental Figure S1 at http://jmd. amjpathol.org). The fact that these improvements were not statistically significant suggest that both kits performed equally well, with values ranging from 93.3%⫺98.4% concordant. This data were generally superior to the numbers reported by Sato et al,9 as well as the percentages in the MicroArray Quality Control project,67 although that study examined protein coding gene expression profiling. Taken together, this data shows that both the biotin-only and biotin-HSR labeling approaches are reliable, and that changes to the labeling reagent in the biotin-HSR improves reproducibility of the data. This is the first study to show that changes to the miRNA labeling reagent allows a reduction in the amount of starting material used. First, it was shown that the number of present calls for human miRNA genes was reduced by only a mean value of 7% and 5% for brain and lung samples, respectively, when the amount of biotin-HSR labeled RNA used was decreased by 10-fold from 1.0 g to 0.1 g (Figure 1). However, the decline was increased to a mean value of 31% and 25% respectively, when using the biotin-only labeled RNA. This large detection loss seen using biotin-only labeling was similar
Discussion Previous studies in the mRNA literature have shown that when laboratories are left to use their own labeling protocols, the comparability of microarray data within and across centers is fair to low,68 but can be improved when standardizing technical, experimental, and analytical factors.67 This is the first study to examine the issue in small noncoding RNAs. Here, we analyzed miRNA expression in human brain and lung samples and evaluated the linear dynamic range of the Affymetrix miRNA GeneChip. Our objective was to determine the differences, using two RNA labeling kits, in the reliability and validity of miRNA expression profiling data using recommended and reduced amount of input RNA. With the exception of the ligation mix, each of the two kits used contained identical reagents. Several improvements to the quality of data were seen when different RNA labeling kits were used. First, a statistically significant increase in the mean correlation coefficient for technical replicates in the brain and lung were seen when comparing samples labeled using the biotin-HSR versus biotin-only kits (Table 1; see also Sup-
Figure 5. Dynamic range of the miRNA GeneChip using biotin-HSR labeling kit. An equimolar pool of synthetic human, mouse, and rat miRNA was labeled using the Biotin-HSR kit and hybridized in amounts ranging from 4.88 to 5000 attomoles (0 datapoint not shown; 12 arrays total). Probes for 465 of 470 human, 223 of 224 mouse, and 42 of 42 rat synthetic miRNAs were detected in all hybridizations (rat and mouse data not shown). Global mean (⫾ SEM) represented by X and error bars, respectively. Linear regression line in black. Intensity values from the same human probes detected in all brain or lung hybridizations (0.1 to 1.0 g) were plotted as black squares or white triangles, respectively. Brain and lung data plotted as grand mean (⫾ grand SEM).
RNA Labeling For MicroRNA Profiling 19 JMD January 2012, Vol. 14, No. 1
to that observed by Lynch et al., who reduced input RNA requirements by 2.5-, 5-, and 25-fold and found a 14%, 14%, and 33% decrease, respectively, in expression detection sensitivity.61 These data show that detection calls using as little as 0.1 g of biotin-HSR labeled RNA were comparable to those generated when using 1.0 g of biotin-only labeled RNA. When considering that as much as 5 g of total RNA were required to produce reliable signatures,69 a reduction in the amount of input RNA is of particular relevance, especially in cases in which nonrenewable clinical specimens are used for miRNA microarray profiling. Next, differentially expressed miRNA gene lists, from reduced input RNA amounts, were correlated with the one derived using 1.0 g total RNA labeled with the biotin-only kit, which was the standard recommended amount by Affymetrix/Genisphere. We found that, in contrast to the miRNA gene list correlations using reduced amounts of biotin-only labeled RNA, variation in miRNA gene rankings were minimized when using reduced amounts of biotin-HSR labeled RNA, suggesting that the use of less starting material does not affect the order of identified genes on a list (Table 2). Finally, when directly comparing filtered miRNA gene lists, generated using different amounts of biotin-HSR labeled RNA (Figure 3), at least 80% of the genes from each list were differentially co-detected in one or more samples of varying input RNA amounts, indicating a high degree of overlap regardless of the amount of starting material used. A cross-platform validation was undertaken to ensure the integrity of the microarray data. Although next generation sequencing technologies could have been used,11 we choose quantitative real-time PCR because it is widely accepted as a gold standard for cross-platform comparison1 and currently offers a time and cost savings over other methods. Eleven human miRNAs were selected based on statistically significant high/low differential expression values in both brain and lung samples, as well as a small subset with no difference (Figure 4). Much like the reports from some9,22,67 but not all11 investigators, we found good correlation between TaqMan and microarray data. Furthermore, the correlation values remained ⬎0.98 for all comparisons made using microarray values from any of the RNA dilutions labeled using the biotin-HSR. In contrast, although still high, the correlations fell to a minimal value of 0.915 when evaluating biotin-only labeled RNA samples. Subsequently, our spike-in experiments showed that the dynamic range of the microarray remained linear for equimolar increases in synthetic miRNAs that varied by three orders of magnitude. Similar results were reported by Wang and company, who demonstrated linearity over two orders of magnitude using the Agilent microarray platform.70 These experiments were performed to ensure that the signal intensity values in our experiments did not exceed the linearity limits of the microarray when using the new RNA labeling kit. Finally, after the conclusion of these experiments and analysis, the GeneChip miRNA 2.0 array was released by Affymetrix to expand on coverage of 131 organisms (vs. 71 on the original array) using Sanger miRNA database
v15 (vs. v11) and to add 2202 probesets unique to premiRNA hairpins (vs. 0). Although it was not tested, as that was not the primary objective of this study, the biotin-HSR kit is not limited to the original chip type, as it is now the recommended strategy to label RNA for the 2.0 array. It should also be noted that, currently, this kit has been optimized only for short RNAs such as miRNAs, and is not developed for longer RNA transcripts. Results from this study highlight RNA labeling as an important factor that can be used as a means of reducing input RNA requirements and improving data quality for miRNA microarray expression profiling studies. Labeling reagent modifications can result in time and cost savings, by eliminating or minimizing the need for preamplification strategies, and enables the profiling of clinical samples with limited quantities of total RNA. The biotin-HSR labeling kit is reliable, is valid, benchmarks well with the currently accepted gold-standard quantitative real-time PCR, and is an improvement over the previous generation of Genisphere miRNA labeling reagents.
Acknowledgments Precision Biomarker Resources performed all of the hybridization, washing, and scanning steps for the RNA titration experiments. The authors thank James Cravens and Stacy Berger (City of Hope) for their input in data review and presentation, and Frank Staedtler (Novartis Institute of Biomedical Research), for his reading and critical review of the manuscript.
References 1. Benes V, Castoldi M: Expression profiling of microRNA using realtime quantitative PCR, how to use it and what is available. Methods 2010, 50:244 –249 2. Creighton CJ, Reid JG, Gunaratne PH: Expression profiling of microRNAs by deep sequencing. Brief Bioinformatics 2009, 10:490 – 497 3. Liu CG, Spizzo R, Calin GA, Croce CM: Expression profiling of microRNA using oligo DNA arrays. Methods 2008, 44:22–30 4. William K, Jian-Jun Z, Lili H, Jin QC: Strategies for profiling MicroRNA expression. J Cell Physiol 2009, 218:22–25 5. Long Q, Johnson BA, Osunkoya AO, Lai YH, Zhou W, Abramovitz M, Xia M, Bouzyk MB, Nam RK, Sugar L, Stanimirovic A, Williams DJ, Leyland-Jones BR, Seth AK, Petros JA, Moreno CS: Protein-coding and microRNA biomarkers of recurrence of prostate cancer following radical prostatectomy. Am J Pathol 2011, 179:46 –54 6. Ono K, Kuwabara Y, Han J: MicroRNAs and cardiovascular diseases. FEBS J 2011, 278:1619 –1633 7. Caporali A, Meloni M, Vollenkle C, Bonci D, Sala-Newby GB, Addis R, Spinetti G, Losa S, Masson R, Baker AH, Agami R, le Sage C, Condorelli G, Madeddu P, Martelli F, Emanueli C: Deregulation of microRNA-503 contributes to diabetes mellitus-induced impairment of endothelial function and reparative angiogenesis after limb ischemia. Circulation 2011, 123:282–291 8. Sah S, McCall MN, Eveleigh D, Wilson M, Irizarry RA: Performance evaluation of commercial miRNA expression array platforms. BMC Res Notes 2010, 3:80 9. Sato F, Tsuchiya S, Terasawa K, Tsujimoto G: Intra-platform repeatability and inter-platform comparability of microRNA microarray technology. PLoS One 2009, 4:e5540 10. Chen Y, Gelfond J, McManus L, Shireman P: Reproducibility of quantitative RT-PCR array in miRNA expression profiling and comparison with microarray analysis. BMC Genomics 2009, 10:407 11. Git A, Dvinge H, Salmon-Divon M, Osborne M, Kutter C, Hadfield J, Bertone P, Caldas C: Systematic comparison of microarray profiling,
20 Kaddis et al JMD January 2012, Vol. 14, No. 1
12.
13.
14.
15.
16.
17.
18. 19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
real-time PCR, and next-generation sequencing technologies for measuring differential microRNA expression. RNA 2010, 16:991– 1006 Pradervand S, Weber J, Lemoine F, Consales F, Paillusson A, Dupasquier M, Thomas J, Richter H, Kaessmann H, Beaudoing E, Hagenbuchle O, Harshman K: Concordance among digital gene expression, microarrays, and qPCR when measuring differential expression of microRNAs. Biotechniques 2010, 48:219 –222 Willenbrock H, Salomon J, Sokilde R, Barken KB, Hansen TN, Nielsen FC, Moller S, Litman T: Quantitative miRNA expression analysis: comparing microarrays with next-generation sequencing. RNA 2009, 15:2028 –2034 Yauk CL, Rowan-Carroll A, Stead JD, Williams A: Cross-platform analysis of global microRNA expression technologies. BMC Genomics 2010, 11:330 Nelson PT, Wang WX, Wilfred BR, Tang G: Technical variables in high-throughput miRNA expression profiling: much work remains to be done. Biochim Biophys Acta 2008, 1779:758 –765 Xi Y, Nakajima G, Gavin E, Morris CG, Kudo K, Hayashi K, Ju J: Systematic analysis of microRNA expression of RNA extracted from fresh frozen and formalin-fixed paraffin-embedded samples. RNA 2007, 13:1668 –1674 Szafranska AE, Davison TS, Shingara J, Doleshal M, Riggenbach JA, Morrison CD, Jewell S, Labourier E: Accurate molecular characterization of formalin-fixed. paraffin-embedded tissues by microRNA expression profiling. J Mol Diagn 2008, 10:415– 423 Hoiby PE, Watzele M: miRNA extraction and profiling on FFPE samples. Biochemica 2008, 3:23–25 Li J, Smyth P, Flavin R, Cahill S, Denning K, Aherne S, Guenther SM, O’Leary JJ, Sheils O: Comparison of miRNA expression patterns using total RNA extracted from matched samples of formalin-fixed paraffin-embedded (FFPE) cells and snap frozen cells. BMC Biotechnol 2007, 7:36 Zhang X, Chen J, Radcliffe T, Lebrun DP, Tron VA, Feilotter H: An array-based analysis of microRNA expression comparing matched frozen and formalin-fixed paraffin-embedded human tissue samples. J Mol Diagn 2008, 10:513–519 Abdueva D, Wing M, Schaub B, Triche T, Davicioni E: Quantitative expression profiling in formalin-fixed paraffin-embedded samples by affymetrix microarrays. J Mol Diagn 2010, 12:409 – 417 Ach RA, Wang H, Curry B: Measuring microRNAs: comparisons of microarray and quantitative PCR measurements, and of different total RNA prep methods. BMC Biotechnol 2008, 8:69 Doleshal M, Magotra AA, Choudhury B, Cannon BD, Labourier E, Szafranska AE: Evaluation and validation of total RNA extraction methods for microRNA expression analyses in formalin-fixed, paraffin-embedded tissues. J Mol Diagn 2008, 10:203–211 Hammerle-Fickinger A, Riedmaier I, Becker C, Meyer HH, Pfaffl MW, Ulbrich SE: Validation of extraction methods for total RNA and miRNA from bovine blood prior to quantitative gene expression analyses. Biotechnol Lett 2010, 32:35– 44 Wang WX, Wilfred BR, Baldwin DA, Isett RB, Ren N, Stromberg A, Nelson PT: Focus on RNA isolation: obtaining RNA for microRNA (miRNA) expression profiling analyses of neural tissue. Biochim Biophys Acta 2008, 1779:749 –757 Bravo V, Rosero S, Ricordi C, Pastori RL: Instability of miRNA and cDNAs derivatives in RNA preparations. Biochem Biophys Res Commun 2007, 353:1052–1055 Mraz M, Malinova K, Mayer J, Pospisilova S: MicroRNA isolation and stability in stored RNA samples. Biochem Biophys Res Commun 2009, 390:1– 4 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 Biol 2006, 7:3 Becker C, Hammerle-Fickinger A, Riedmaier I, Pfaffl MW: mRNA and microRNA quality control for RT-qPCR analysis. Methods 2010, 50: 237–243 Ibberson D, Benes V, Muckenthaler MU, Castoldi M: RNA degradation compromises the reliability of microRNA expression profiling. BMC Biotechnol 2009, 9:102 Jung M, Schaefer A, Steiner I, Kempkensteffen C, Stephan C, Erbersdobler A, Jung K: Robust microRNA stability in degraded RNA
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
preparations from human tissue and cell samples. Clin Chem 2010, 56:998 –1006 Zhu B, Xu F, Baba Y: An evaluation of linear RNA amplification in cDNA microarray gene expression analysis. Mol Genet Metab 2006, 87:71–79 Viale A, Li J, Tiesman J, Hester S, Massimi A, Griffin C, Grills G, Khitrov G, Lilley K, Knudtson K, Ward B, Kornacker K, Chu CY, Auer H, Brooks AI: Big results from small samples: evaluation of amplification protocols for gene expression profiling. J Biomol Techn 2007, 18:150 –161 Singh R, Maganti RJ, Jabba SV, Wang M, Deng G, Heath JD, Kurn N, Wangemann P: Microarray-based comparison of three amplification methods for nanogram amounts of total RNA. Am J Physiol Cell Physiol 2005, 288:C1179 –C1189 Patel OV, Suchyta SP, Sipkovsky SS, Yao J, Ireland JJ, Coussens PM, Smith GW: Validation and application of a high fidelity mRNA linear amplification procedure for profiling gene expression. Vet Immunol Immunopathol 2005, 105:331–342 Francois P, Garzoni C, Bento M, Schrenzel J: Comparison of amplification methods for transcriptomic analyses of low abundance prokaryotic RNA sources. J Microbiol Methods 2007, 68:385–391 Nagy ZB, Kelemen JZ, Feher LZ, Zvara A, Juhasz K, Puskas LG: Real-time polymerase chain reaction-based exponential sample amplification for microarray gene expression profiling. Anal Biochem 2005, 337:76 – 83 Puskas LG, Zvara A, Hackler L, Jr., Van Hummelen P: RNA amplification results in reproducible microarray data with slight ratio bias. Biotechniques 2002, 32:1330 –1334, 1336, 1338, 1340 Bak M, Conley L, Hedegaard J, Larsen LA, Sorensen P, Bendixen C, Tommerup N: Evaluation of two methods for generating cRNA for microarray experiments from nanogram amounts of total RNA. Anal Biochem 2006, 358:111–119 Cope L, Hartman SM, Gohlmann HW, Tiesman JP, Irizarry RA: Analysis of Affymetrix GeneChip data using amplified RNA. Biotechniques 2006, 40:165–166, 168, 170 Yu J, Othman MI, Farjo R, Zareparsi S, MacNee SP, Yoshida S, Swaroop A: Evaluation and optimization of procedures for target labeling and hybridization of cDNA microarrays. Mol Vis 2002, 8:130 – 137 Richter A, Schwager C, Hentze S, Ansorge W, Hentze MW, Muckenthaler M: Comparison of fluorescent tag DNA labeling methods used for expression analysis by DNA microarrays. Biotechniques 2002, 33:620 – 628, 630 Ouellet M, Adams PD, Keasling JD, Mukhopadhyay A: A rapid and inexpensive labeling method for microarray gene expression analysis. BMC Biotechnol 2009, 9:97 Manduchi E, Scearce LM, Brestelli JE, Grant GR, Kaestner KH, Stoeckert CJ, Jr.: Comparison of different labeling methods for twochannel high-density microarray experiments. Physiol Genomics 2002, 10:169 –179 Kimura N, Tamura TA, Murakami M: Evaluation of the performance of two carbodiimide-based cyanine dyes for detecting changes in mRNA expression with DNA microarrays. Biotechniques 2005, 38: 797– 806 Badiee A, Eiken HG, Steen VM, Lovlie R: Evaluation of five different cDNA labeling methods for microarrays using spike controls. BMC Biotechnol 2003, 3:23 t Hoen PA, de Kort F, van Ommen GJ, den Dunnen JT: Fluorescent labelling of cRNA for microarray applications. Nucleic Acids Res 2003, 31:e20 Sawada A, Mizufune S, Kaji N, Tokeshi M, Baba Y: Evaluation of amplified cRNA targets for oligonucleotide microarrays. Anal Bioanal Chem 2007, 387:2645–2654 Ma C, Lyons-Weiler M, Liang W, LaFramboise W, Gilbertson JR, Becich MJ, Monzon FA: In vitro transcription amplification and labeling methods contribute to the variability of gene expression profiling with DNA microarrays. J Mol Diagn 2006, 8:183–192 Lonergan W, Whistler T, Vernon SD: Comparison of target labeling methods for use with Affymetrix GeneChips. BMC Biotechnol 2007, 7:24 Zakharkin SO, Kim K, Mehta T, Chen L, Barnes S, Scheirer KE, Parrish RS, Allison DB, Page GP: Sources of variation in Affymetrix microarray experiments. BMC Bioinformat 2005, 6:214
RNA Labeling For MicroRNA Profiling 21 JMD January 2012, Vol. 14, No. 1
52. Ach RA, Floore A, Curry B, Lazar V, Glas AM, Pover R, Tsalenko A, Ripoche H, Cardoso F, d’Assignies MS, Bruhn L, Van’t Veer LJ: Robust interlaboratory reproducibility of a gene expression signature measurement consistent with the needs of a new generation of diagnostic tools. BMC Genomics 2007, 8:148 53. Tissot C: Application Note: Analysis of miRNA content in total RNA preparations using the Agilent 2100 bioanalyzer. Agilent Technologies, Santa Clara, CA. Publication number 5989-7870EN 2008, 1– 8 54. Weber JA, Baxter DH, Zhang S, Huang DY, How Huang K, Jen Lee M, Galas DJ, Wang K: The MicroRNA spectrum in 12 body fluids. Clin Chem 2010, 56:1733–1741 55. Mora JR, Getts RC: High-sensitivity detection methods for low-abundance RNA species: applications for functional genomics research. Expert Rev Mol Diagn 2007, 7:775–785 56. Saba R, Booth S: Target labelling for the detection and profiling of microRNAs expressed in CNS tissue using microarrays. BMC Biotechnol 2006, 6:47 57. Wu Z, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F: A model-based background adjustment for oligonucleotide expression arrays. J Am Stat Assoc 2004, 99:909 –917 58. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 2003, 4:249 –264 59. Rao Y, Lee Y, Jarjoura D, Ruppert AS, Liu CG, Hsu JC, Hagan JP: A comparison of normalization techniques for microRNA microarray data. Stat Appl Genet Mol Biol 2008, 7:Article22 60. Pradervand S, Weber J, Thomas J, Bueno M, Wirapati P, Lefort K, Dotto GP, Harshman K: Impact of normalization on miRNA microarray expression profiling. RNA 2009, 15:493–501 61. Lynch A, Hadfield J, Dunning M, Osborne M, Thorne N, Tavare S: The cost of reducing starting RNA quantity for Illumina BeadArrays: a bead-level dilution experiment. BMC Genomics 2010, 11:540 62. Application Note: TaqMan MicroRNA assays: Endogenous controls for real-time quantitation of miRNA using TaqMan MicroRNA assays. Applied Biosystems, Foster City, CA. publication 127AP11-01 2007, 1– 8 63. Peltier HJ, Latham GJ: Normalization of microRNA expression levels in quantitative RT-PCR assays: identification of suitable reference
64.
65.
66.
67.
68.
69.
70.
RNA targets in normal and cancerous human solid tissues. RNA 2008, 14:844 – 852 Andersen CL, Jensen JL, Orntoft TF: Normalization of real-time quantitative reverse transcription-PCR data: a model-based variance estimation approach to identify genes suited for normalization, applied to bladder and colon cancer data sets. Cancer Res 2004, 64:5245– 5250 Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, Speleman F: Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol 2002, 3:(7)research0034.1-research0034.11 Livak KJ, Schmittgen TD: Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) method. Methods 2001, 25:402– 408 Shi L, Reid LH, Jones WD, Shippy R, Warrington JA, Baker SC, Collins PJ, de Longueville F, Kawasaki ES, Lee KY, Luo Y, Sun YA, Willey JC, Setterquist RA, Fischer GM, Tong W, Dragan YP, Dix DJ, Frueh FW, Goodsaid FM, Herman D, Jensen RV, Johnson CD, Lobenhofer EK, Puri RK, Schrf U, Thierry-Mieg J, Wang C, Wilson M, Wolber PK, Zhang L, Amur S, Bao W, Barbacioru CC, Lucas AB, Bertholet V, et al: The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nature Biotechnol 2006, 24:1151–1161 Bammler T, Beyer RP, Bhattacharya S, Boorman GA, Boyles A, Bradford BU, Bumgarner RE, Bushel PR, Chaturvedi K, Choi D, Cunningham ML, Deng S, Dressman HK, Fannin RD, Farin FM, Freedman JH, Fry RC, Harper A, Humble MC, Hurban P, Kavanagh TJ, Kaufmann WK, Kerr KF, Jing L, Lapidus JA, Lasarev MR, Li J, Li YJ, Lobenhofer EK, Lu X, Malek RL, Milton S, Nagalla SR, O’Malley JP, Palmer VS, Pattee P, et al: Standardizing global gene expression analysis between laboratories and across platforms. Nature Methods 2005, 2:351–356 Calin GA, Liu C-G, Sevignani C, Ferracin M, Felli N, Dumitru CD, Shimizu M, Cimmino A, Zupo S, Dono M, Dell’Aquila ML, Alder H, Rassenti L, Kipps TJ, Bullrich F, Negrini M, Croce CM: MicroRNA profiling reveals distinct signatures in B cell chronic lymphocytic leukemias. Proc Natl Acad Sci U S Am 2004, 101:11755–11760 Wang H, Ach RA, Curry B: Direct and sensitive miRNA profiling from low-input total RNA. RNA 2007, 13:151–159