Comparison of gene expression profiling by reverse transcription quantitative PCR between fresh frozen and formalin-fixed, paraffin-embedded breast cancer tissues

Please download to get full document.

View again

of 7
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Document Description
Comparison of gene expression profiling by reverse transcription quantitative PCR between fresh frozen and formalin-fixed, paraffin-embedded breast cancer tissues
Document Share
Document Transcript 󰀳󰀸󰀹 Vol. 󰀴󰀸 | No. 󰀵 | 󰀲󰀰󰀱󰀰 Reports Introduction   󰁑uantification methods for gene expression have become important tools both in the understanding of the molecular events under- lying human breast cancer and in the identi-fication of diagnostic and therapeutic targets. Microarray expression profiling has provided an exciting new technology for attempting to identify gene-based classifiers that correlate  with breast cancer diagnosis, disease prognosis, or prediction of response to treatment (1–4). A useful technique to confirm or use such classifiers is quantitative reverse transcription–  polymerase chain reaction (RT-qPCR) assays of the selected genes (5–7). 󰀀e development of molecular tests for clinical use has been limited by the lack of available clinical samples for validation of candidate biomarkers. Fresh frozen (FF) samples are difficult to collect for large scale studies, complicated to process, and expensive to store. Formalin-fixed, paraffin-embedded samples (FFPE) are stable at room temperature and easily storable, and—most important— they constitute a widely available archive of clinical samples linked to precious clinical and follow-up information. Set against these advantages is the fact that RNA isolated from FFPE is considered a poor material for gene expression analysis, owing to its extensive degradation. While microarray-based studies are highly sensitive to RNA degradation, RT-qPCR appears to be more robust and tolerates partial degradation of RNA (8). Although RNA degradation leads to a loss of amplifiable templates, optimized normal- ization strategies could effectively compensate for this bias (9–11). Normalization is essential to control for experimental errors, such as the inherent variability of RNA, variability of extraction protocols that may co-purify inhib- itors, and different reverse transcription and PCR efficiencies (12). In this study, we analyzed the corre-lation in gene expression measurements by RT-qPCR between breast cancer FF and FFPE tissues, and evaluated the perfor-mance of different normalization methods in compensating for the effect of RNA degra- dation. We also investigated the factors that could influence obtaining reliable results from FFPE samples.  Materials and methods Tissue specimens Matching pairs of FF and FFPE biopsies of breast tumors from 30 patients (60 samples in total) were retrieved from the Comparison of gene expression profiling by reverse transcription quantitative PCR between fresh frozen and formalin-fixed, paraffin-embedded breast cancer tissues Iker Sánchez-Navarro 1 *, Angelo Gámez-Pozo 1 *, Manuel González-Barón 2 , Álvaro Pinto-Marín 2 , David Hardisson 3 , Rocío López 1 , Rosario Madero 4 , Paloma Cejas 1 , Marta Mendiola 1, 3 , Enrique Espinosa 2 , and Juan Ángel Fresno Vara 1 1  Laboratory of Molecular Pathology and Oncology, Research Unit, Hospital Universitario La Paz, Madrid, Spain,  2  Department of Medical Oncology, Hospital Universitario La Paz, Madrid, Spain, 3  Department of Pathology,  Hospital Universitario La Paz, Madrid, Spain, and 4   Biostatistics Unit, Hospital Universitario La Paz, Madrid, Spain  BioTechniques  48:389-397 (May 2010) doi 10.2144/000113388 Keywords: breast cancer; gene expression profiling; RT-qPCR; FFPE; normalization; biomarkersSupplementary material for this article is available at *I.S.-N. and A.G.-P. contributed equally to this work. Recent reports demonstrate the feasibility of quantifying gene expression by using RNA isolated from blocks of for-malin-fixed, paraffin-embedded (FFPE) tumor tissue. 󰀀e development of molecular tests for clinical use based on archival materials would be of great utility in the search for and validation of important genes or gene expression pro-files. In this study, we compared the performance of different normalization strategies in the correlation of quantita-tive data between fresh frozen (FF) and FFPE samples and analyzed the parameters that characterize such correlation for each gene. Total RNA extracted from FFPE samples presented a shi in raw cycle threshold (Cq) values that can be explained by its extensive degradation. Proper normalization can compensate for the effects of RNA degradation in gene expression measurements. We show that correlation between normalized expression values is better for mod-erately to highly expressed genes whose expression varies significantly between samples. Nevertheless, some genes had no correlation. 󰀀ese genes should not be included in molecular tests for clinical use based on FFPE samples. Our results could serve as a guide when developing clinical diagnostic tests based on RT-qPCR analyses of FFPE tissues in the coming era of treatment decision-making based on gene expression profiling. Reports 󰀳󰀹󰀰 Department of Pathology of Hospital Universitario La Paz (Madrid, Spain). Tissue samples were procured between 1991 and 1998. The histopathological features of each sample were reviewed by an experienced breast pathologist to confirm diagnosis and similar tumor content ( ≥ 70%). Approval from the Ethical Committee of Hospital Universitario La Paz (HULP code PI-405) was obtained for the conduct of the study. Isolation of RNA and cDNA synthesis For extraction of RNA from FFPE tissue, 15 5-µm sections were cut from each archival block. Paraffin was removed by  xylene extraction followed by ethanol  washes. RNA was isolated from tissue slices using the MasterPure RNA Purifi- cation Kit (Epicenter Biotechnologies, Madison, WI, USA). RNA was isolated from 10 10-µm sections of FF tissue with TRIzol Reagent (Invitrogen, Carsbald, CA, USA) and cleaned up with Qiagen RNeasy spin columns. Total RNA isolated was quantified and qualitatively assessed using spectrophotometer OD 260   measurements and capillary electropho- resis (Agilent 2100 Bioanalyzer, Agilent, Santa Clara, CA, USA). We normalized to total RNA input; therefore, first-strand cDNA was synthesized from 1 µg total RNA according to the High Capacity cDNA Reverse Transcription Kit protocol (Applied Biosystems, Foster City, CA, USA). 󰁑uantitative RT-PCR  RT-qPCR amplifications were performed  with TaqMan Gene Expression Assays  products in an ABI PRISM 7900 HT Sequence Detection System (Applied Biosystems). Reactions were carried out using TaqMan Low Density Arrays (TLDAs; Applied Biosystems) containing 50 µL TaqMan Universal PCR Master Mix (Applied Biosystems) and 50 µL cDNA template corresponding to 100 ng total RNA per channel of the microfluidic card.  We used two configurations of TLDAs containing reference and breast cancer  prognosis–related genes. The first one  was configured to measure the expression of 95 genes in duplicate for two samples  per card. 󰀀e second one was configured to analyze expression of 63 genes in triplicate, also for two samples per card. For a complete list of the Gene Expression Assays included in each configuration, see Supplementary Table S1. We followed the Minimum Information for Publication of 󰁑uantitative Real-time PCR Experi-ments (MIQE) guidelines (13). Raw Cq data, experimental annotation and sample annotation are available in the RDML data format (Supplementary Table S2).cDNAs derived from matched FF and FFPE tumor tissue pairs were run simulta-neously in the same card. 󰀀e raw average Cq values, defined as the point at which the fluorescence rises above the background fluorescence (14), were obtained using the SDS 2.2 soware (Applied Biosystems). 󰀀e maximum Cq value was set at 40. Cq values were recorded in Microsoft Excel 2003 for subsequent calculations and analyzed using GenEx soware from MultiD Analyses (Version 4.3.2;   Göteborg, Sweden) and Prism 4 from Graphpad (La  Jolla, CA, USA). 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 19 20 21 66 67 68 69 72 73 74 75 76 77 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 19 20 21 66 67 68 69 72 73 74 75 76 77 010203040 FF FFPE       C     q Figure 1. Comparison of expression levels between FF and FFPE archived samples.  Average expres-sion levels of each tumor are visualized separately for the FF and matched FFPE samples as box plots, where the median raw Cqs are represented by lines, the Cq 25th and 75th percentile values by boxes, and the ranges by whiskers. Median Cq values were approximately 5 units greater in RNA derived from FFPE material than in RNA prepared from FF samples. Table 1. Top ranked candidate control genes selected by each of the normalization methods.MethodFFFFPE geNorm 18S, IPO8, POLR2A, HMBS, UBC, PPIAGUSB, IPO8, B2M, POLR2A, UBC, SD  HA NormFinder 18SPPIA NormFinder2 HPRT1, B2M  NorMean IPO8, POLR2A, UBC, SD  HA geNorm provided distinct lists of control genes for FF and FFPE materials, with some shared genes, and NormFinder selected a different best control gene for each material. NormFinder2 and NorMean identi-fied the same set of control genes in both materials. 󰀳󰀹󰀲 Vol. 󰀴󰀸 | No. 󰀵 | 󰀲󰀰󰀱󰀰 Reports Normalization factors  We applied two different published methods to identify genes that were expressed at essentially constant levels in our samples to use as endogenous controls. Vandesompele et al. (15) have written an application called geNorm that automati-cally calculates the gene stability measure for all genes in a given set of samples, and determines the optimal number of control genes needed for normalization. Andersen et al. (16) have described a novel evaluation strategy that considers whether the candidate genes show variation not only across the studied sample set, but also across sample subgroups. We used their bioinformatics tool, NormFinder, to find the best endogenous control gene and to determine the best combination of two genes, taking into consideration two subgroups, FF and FFPE samples, in order to find the best endogenous control genes in both materials. The program  provides a stability value for each candidate gene and highlights the best gene with the lowest stability value for one group analysis (NormFinder), as well as the best combination of two genes for a two-gene normalization factor for two subgroups (NormFinder2). 󰀀e performance of the mean Cq per sample was also tested, which has been  previously validated as a new method for miRNA RT-qPCR data normalization (17). In addition, we developed a new model for selecting control genes based on housekeeping coefficient of variation (CV)   and Pearson correlation coeffi- cient of its expression with the mean gene expression per sample. We called this method NorMean: , ijiij CV ar  = ∑∑ [Eq. 1]  where CV  ij   is the coefficient of variation for control gene i  in material  j  , and r  ij   is the Pearson correlation coefficient of expression of gene i  with the mean gene expression per sample in material  j  . 󰀀is equation provides a value  a i , which permits the ranking of control genes. Control genes with the lowest  a i  values are those  with the most stable expression (low CV) and highest positive correlation with mean gene expression per sample. 󰀀ose genes  with r ≤ 0 should not be considered for NorMean analysis. Using this ranking, we calculated different normalization factors by stepwise inclusion of control genes and geometric averaging of their expression levels. 󰀀e optimal number of control genes for normal- ization was determined by comparing the  percentage of significantly correlated genes between FF and FFPE materials using each normalization factor. Given that there is a considerable shi in mean Cq values between FF and FFPE samples, we preferred to use CV over standard deviation, as previously done by others (18–21), because standard deviation is very sensitive to mean value, whereas CV expresses the standard deviation as a percentage of the sample mean. Data presentation and calculations Gene expression values were calculated based on the modified ∆∆Cq method described before (22). Once the appropriate gene or set of genes to be used as controls were selected, we calculated a normalization factor (NF) using the geometric mean of the genes selected by each method. 󰀀en, we calculated normalized expression values for each gene in each sample, given by the 2 -∆Cq   value, where .   Cq Cq NF  ∆ = − [Eq. 2] Statistical analysis Correlation in the quantitative data between 30 matched FF and FFPE samples was determined using the Pearson correlation coefficient among the different calculated expression values normalized with each of the normalization factors. Statistical signifi-cance was  P   < 0.05. A discriminant analysis  was performed in order to identify the factors that could explain the measured correla-tions in gene expression data generated by RT-qPCR between FF and FFPE tissue pairs. Statistical analyses were performed with the SPSS soware package, version 9 (Chicago, IL, USA). Results and discussion   󰁑uantitative RT-PCR data  We analyzed the expression of 95 genes on 30 FF and FFPE matched breast cancer tissues by RT-qPCR using TLDAs. 󰀀e mean raw Cq  was 26.71 and the median raw Cq was 26.91 for FF samples, and were 32.18 and 31.88, respectively, for FFPE samples. In accordance  with previously published observations, the median Cq values were approximately 5 units     1    8    S    A    C    T    B    A    K    A    P    2    A    L    D    H    4    A    1    A    P    2    B    1    A    U    R    K    A    A    Y    T    L    2    B    2    M    B    A    G    1    B    B    C    3    B    C    L    2    B    I    R    C    5    C    1    6   o   r    f    6    1    C    2    0   o   r    f    4    6    C    9   o   r    f    3    0    C    C    N    B    1    C    C    N    E    2    C    D    6    8    C    D    C    4    2    B    P    A    C    D    C    A    7    C    E    N    P    A    C    H    D    H    C    O    L    4    A    2    C    T    S    L    2    D    C    K    D    T    L    E    C    T    2    E    G    L    N    1    E    R    B    B    2    E    S    M    1    E    S    R    1    E    S    R    2    E    X    T    1    F    B    X    O    3    1    F    G    F    1    8    F    L    T    1    G    A    P    D    H    G    M    P    S    G    N    A    Z    G    P    R    1    2    6    G    P    R    1    8    0    G    R    B    7    G    S    T    M    3    G    U    S    B    H    M    B    S    H    O    X    B    1    3    H    P    R    T    1    H    R    A    S    L    S    I    G    F    B    P    5    I    L    1    7    R    B    I    P    O    8    K    I    A    A    1    4    4    2    K    N    T    C    2    L    G    P    2    M    C    M    6    M    E    L    K    M    K    I    6    7    M    M    P    1    1    M    M    P    9    M    S    4    A    7    M    T    D    H    M    Y    B    L    2    M    Y    O    1    5    B    N    M    U    N    U    S    A    P    1    O    R    C    6    L    O    X    C    T    1    P    E    C    I    P    G    K    1    P    G    R    P    I    T    R    M    1    P    O    L    R    2    A    P    P    I    A    P    Q    L    C    2    P    R    C    1    Q    S    C    N    6    L    1    R    A    B    6    B    R    F    C    4    R    P    L    P    0    R    T    N    4    R    L    1    R    U    N    D    C    1    S    C    U    B    E    2    S    D    H    A    S    E    R    F    1    A    S    L    C    2    A    3    S    T    K    3    2    B    T    B    P    T    F    R    C    T    G    F    B    3    T    S    P    Y    L    5    U    B    C    U    C    H    L    5    W    I    S    P    1    Y    W    H    A    Z    Z    N    F    5    3    3 010203040 Mean Cq FFMean Cq FFPE Gene       C     q Figure 2. Comparison of mean Cq of each gene in FF and FFPE archived samples.  Mean raw Cq of the 95 genes expression values in FF and FFPE samples. The relative level of expression of each gene is maintained in both materials in spite of the Cq shift between FF and FFPE. Table 2. Comparison of normalization methods on separated materials.FFgeNormNorm FinderNorm Finder2NorMeanMean Cq         F        F        P        E geNorm 10.9600.8550.9790.953 NormFinder 0.78910.8410.9270.912 NormFinder2 0.7280.78610.7990.865 NorMean 0.9780.7640.65610.923 Mean Cq 0.9110.8790.8220.8831 Average Pearson correlation coefficients for gene expression normalized data calculated with the distinct normalization factors within the same material.  Vol. 󰀴󰀸 | No. 󰀵 | 󰀲󰀰󰀱󰀰 Reports higher in RNA derived from FFPE material than in RNA prepared from FF samples (9,10,23). 󰀀e average expression levels of each tumor are visualized separately for FF and FFPE samples as box  plots in Figure 1. 󰀀e raw Cq values are available in Supplementary Tables S1 and S2. Capillary electrophoresis analysis showed that RNA extracted from archival FFPE breast cancer specimens was essentially degraded (Supplementary Figure S1), which is consistent with previous observa- tions (9,10). 󰀀is loss of intact amplicon template explains the five raw Cq shi observed in FFPE specimens with respect to the matched FF samples. 󰀀e samples with a greater degree of RNA degradation showed higher median Cq values, confirming the inverse relation between average Cq and the quality of RNA. Comparison of gene expression profiles between FF and FFPE tissues  We compared the RT-qPCR raw data of the 30 FF and FFPE sample  pairs to measure the sample correlation coefficients. Correlation was assessed by calculating the Pearson’s correlation coefficient of raw Cqs for 95 genes. A mean correlation coefficient of 0.81 ± 0.073 was found between the corresponding Cq values in the 30 tumors, ranging from 0.63 to 0.92. Figure 2 shows the mean raw Cq of each gene in both materials. In spite of the observed Cq shi, the correlation in Cq values between matched FF and FFPE tumor sample pairs was very high. In our opinion, however, these high correlation coefficients in raw  pre-normalized data merely indicate that the relative level of expression of each gene is maintained in both materials; that is, genes with low Cqs (high level of expression) in FF samples have also low Cqs in FFPE samples, and vice versa.Similarly, the gene correlation coefficients were determined across the sample series in FF and in FFPE. 󰀀is gives rise to a much more relevant question, as it allows the assessment of whether the expression  profile of a given gene is conserved in both materials. In this case, the correlations were lower, with a mean correlation coefficient of 0.33 ± 0.189, ranging from 0.005 to 0.81. 󰀀ese results indicate that RNA degradation affects correlation in gene expression patterns between FF and FFPE materials, despite the fact that relative levels of expression are conserved for each gene. 󰀀is could be due to the extensive degradation of RNA obtained from FFPE. Several studies have described that the observed changes in Cq  values induced by RNA fragmentation can be partially compensated by appropriate normalization (9,10,23). Identification of endogenous control genes  We used two previously described methods of normalization, geNorm (15) and NormFinder (16), to select genes with more stable expression. Given that mean gene expression of each sample reflects the quality of the RNA isolated from that sample, we also tested the performance of mean Cq per sample as a normalization factor. Finally, we have developed a new model for selecting control genes.  We propose that the normalization factor should be able to control for the various levels of experimental variability in RT-qPCR, mainly the quantity of RNA and differences in enzymatic efficiencies, as well as for differences in the quality of the starting material. As we have seen, the mean expression of each sample reflects its RNA quality. Corre-spondingly, we generated a model that took into account the control gene coefficient of variation as a measure of expression stability, and the Pearson correlation coefficient with the mean gene expression per sample as a measure of correlation with global sample behavior (see “Materials and methods” section). 󰀀e first part of the model should correct for   the differences in RNA quantity and enzyme efficiencies, and the second part should correct for the differences in RNA quality.  We called this model NorMean. 󰀀e best endogenous control genes obtained by each of these four methods are shown in Table 1. Comparison of normalization methods Given that different methods of selection of control genes provided diverse lists of such genes, we evaluated the correlation of expression data using distinct normalization factors (Table 2). 󰀀e highest correlations were observed between geNorm and NorMean, while NormFinder2 and NorMean showed the lowest correlations. It was striking to observe that the degree of correlation for each gene in the same material was very high, independent of the normalization method employed. Moreover, we would like to highlight the good performance of single gene–based normalization factors compared with mutigene normalization methods, although it has been described that multigene normalization methods outperform single gene–derived normalization methods (15,16,24). 󰀀ese results suggest that effective normalization could be achieved by different strategies in one material. Subsequently, we assessed the correlation of normalized gene expression data between FF and FFPE samples. When data were normalized using geNorm, correlation coefficients for each gene across the 30 matched samples ranged -0.098–0.951 (mean = 0.56). 󰀀e respective values for NorMean ranged 0.074–0.936 (mean = 0.57).  We then compared the performance of each normalization method by contrasting the percentage of genes that show a significant correlation coefficient between FF and FFPE (Figure 3). Raw Cq data showed only 40% of significantly correlated genes between both materials. All normalization factors included in this study increased the percentage of genes significantly correlated. 󰀀e best results were obtained with the normalization factor derived from NorMean, with >80% of genes • ProScan TM II S eries stages with IntelligentScanning Technology provide the highest precision in its class F LUORESCENCE  I LLUMINATION M OTORIZED  S TAGES •Lumen 200/200Pro series 200Watt fluorescence illuminationsystems MICROSCOPE A UTOMATION  W  HERE  V  ISION  M EETS  P RECISION Prior Scientific, Inc. 80 Reservoir Park Drive Rockland, MA. 02370 800-877-2234 • Compatible with Prior Scientific’s line of NanoScanZPiezo Z Stages•OEM and custom stage systems are available • Built-in high speed light attenuator and six position high speed filter wheel•Pre-centered bulb with 2,000 hour life•Stabilized DC power supplyeliminatesvariations in light intensity 󰀳󰀹󰀴 showing significant correlation between FF and FFPE materials (Figure 3). To assess the reproducibility of the  proposed model of reference gene selection,  we applied the different normalization methods to a different set of TLDAs containing 50 genes and 13 controls (63-gene TLDA; see “Materials and methods” section and Supplementary Tables S1 and S2). We carried out the same analyses described in the previous paragraph. In this case, raw Cq data showed <30% of significantly corre-lated genes between both materials. Once more, all normalization factors improved the  percentage of genes with significant positive correlation (Figure 3). Despite the fact that in this series of TLDAs, NormFinder2 was the normalization factor that provided the best results (88% of significantly correlated genes), NorMean showed a similar perfor- mance (84% of significantly correlated genes). Moreover, it presented the maximum stability among the different models tested: in both the 95-gene and the 63-gene series of TLDAs, over 80% of genes showed positive correlation between FF and FFPE materials. NorMean provides a method to select control genes that resemble the mean expression value, which has been previously shown to outperform the current normal-ization strategies (17). However, mean expression value normalization is only valid if a large number of genes are profiled, which occurs in initial screening experiments but almost never in subsequent studies on a limited number of genes.  Analysis of assay performance Our results consistently show that the NorMean normalization procedure is suitable to efficiently compensate for the changes in expression levels resulting from RNA degra-dation. Some genes showed a positive corre- lation between FF and FFPE tissues even  without normalization—it is very significant that this group of genes includes estrogen receptor (  ESR1 ) and progesterone receptor (  PGR )—but the great majority showed a statistically significant positive correlation coefficient between FF and FFPE materials only when appropriately normalized (  ERBB2 ,    AURKA  , and  CCNB1 ). Nevertheless, there also exist a number of genes whose expression  profiles cannot be compared between FF and FFPE samples, because their normalized expression values lack correlation (  FLT1 ,   AKAP2 , and  FGF18) . We investigated the factors that could explain the behavior of these poorly correlated genes. It has been described that amplicon length is related to RT-qPCR performance in severely degraded RNA (9,10). It has been also reported that genes with the largest dynamic range in expression present the highest  positive correlation between FF and FFPE (23). 󰀀erefore, we carried out a discriminant analysis in order to define the parameters that could characterize the genes lacking corre- lation between normalized expression values in FF and FFPE samples. We evaluated the influence of amplicon length, CV, mean and median Cq, trimmed mean Cq, minimum Cq, range, interquartilic range, and standard deviation in each material independently. 󰀀e discriminant analysis selected CV of the normalized expression values and minimum raw Cq as the most influential parameters on the correlation of gene expression between FF and FFPE. These results agree with  previously published data. CV is a measure of the variability in expression of a gene and minimum Cq reflects the level of expression of that gene. Both parameters have been identified as the most decisive in the positive correlation of gene expression measure-ments between RT-qPCR and microarrays North America: 888-USMDBIOInternational: +41-44-986-2628 ST2, also know as IL-1 R4 is an nterleukin-1 receptor family gly-coprotein that contributes to Th2 immune responses. ST2 is expressed on the surface of mast cells, activated  Th2 cells, macrophages and cardio myocytes.It binds IL-33, a cytokine that is up regulated by inflamma-tion or mechanical strain in smooth muscle cells, keratinocytes, airway epithelia and cardiac fibroblasts T1/ST2 Products Mouse T1/ST2 ELISA • Mouse T1/ST2 Antibody • Human ST2L Antibody • Human IL-18 R Antibody • Mouse T1/ST2 ELISA ST2 (IL-1 R4) Levels 00.511.522.533.5Sample 1Sample 2Sample 3Sample 4    S   T   2   (  n  g   /  m   L   ) ST2 levels from mouse serum samples from an OVA-induced asthma study  . Figure 3. Comparison of normalization methods between materials.  Percentage of genes significantly cor-related between FF and FFPE archived breast cancer samples using distinct normalization methods.
Similar documents
View more...
Search Related
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks