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
2 views
PDF
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
   www.Bioechniques.com 󰀳󰀸󰀹 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 o diagnostic and therapeutic targets. Microarray expression profiling has provided an exciting new technology or attempting to identify gene-based classifiers that correlate  with breast cancer diagnosis, disease prognosis, or prediction o response to treatment (1–4). A useful technique to confirm or use such classifiers is quantitative reverse transcription–  polymerase chain reaction (RT-qPCR) assays o the selected genes (5–7). Te 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 ollow-up inormation. 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, R-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 or 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 R-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 proiling; R-qPCR; FFPE; normalization; biomarkersSupplementary material or this article is available at www.Bioechniques.com/article/113388. *I.S.-N. and A.G.-P. contributed equally to this work. Recent reports demonstrate the easibility o quantiying gene expression by using RNA isolated rom blocks o or-malin-fixed, paraffin-embedded (FFPE) tumor tissue. Te development o molecular tests or clinical use based on archival materials would be o great utility in the search or and validation o important genes or gene expression pro-files. In this study, we compared the perormance o different normalization strategies in the correlation o quantita-tive data between resh rozen (FF) and FFPE samples and analyzed the parameters that characterize such correlation or each gene. otal RNA extracted rom FFPE samples presented a shif in raw cycle threshold (Cq) values that can be explained by its extensive degradation. Proper normalization can compensate or the effects o RNA degradation in gene expression measurements. We show that correlation between normalized expression values is better or mod-erately to highly expressed genes whose expression varies significantly between samples. Nevertheless, some genes had no correlation. Tese genes should not be included in molecular tests or clinical use based on FFPE samples. Our results could serve as a guide when developing clinical diagnostic tests based on R-qPCR analyses o FFPE tissues in the coming era o treatment decision-making based on gene expression profiling. Reports   www.Bioechniques.com 󰀳󰀹󰀰 Department o Pathology o Hospital Universitario La Paz (Madrid, Spain). issue samples were procured between 1991 and 1998. he histopathological eatures o each sample were reviewed by an experienced breast pathologist to conirm diagnosis and similar tumor content ( ≥ 70%). Approval from the Ethical Committee o Hospital Universitario La Paz (HULP code PI-405) was obtained for the conduct o the study. Isolation of RNA and cDNA synthesis For extraction o RNA rom FFPE tissue, 15 5-µm sections were cut rom each archival block. Paraffin was removed by  xylene extraction ollowed by ethanol  washes. RNA was isolated rom tissue slices using the MasterPure RNA Purifi-cation Kit (Epicenter Biotechnologies, Madison, WI, USA). RNA was isolated rom 10 10-µm sections o FF tissue with RIzol Reagent (Invitrogen, Carsbald, CA, USA) and cleaned up with Qiagen RNeasy spin columns. otal 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 rom 1 µg total RNA according to the High Capacity cDNA Reverse Transcription Kit protocol (Applied Biosystems, Foster City, CA, USA). 󰁑uantitative RT-PCR  R-qPCR amplifications were perormed  with aqMan Gene Expression Assays  products in an ABI PRISM 7900 H Sequence Detection System (Applied Biosystems). Reactions were carried out using aqMan Low Density Arrays (LDAs; 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 o the microfluidic card.  We used two configurations of TLDAs containing reerence and breast cancer  prognosis–related genes. he irst one  was configured to measure the expression o 95 genes in duplicate or two samples  per card. Te second one was configured to analyze expression o 63 genes in triplicate, also or 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 Inormation or Publication o 󰁑uantitative Real-time PCR Experi-ments (MIQE) guidelines (13). Raw Cq data, experimental annotation and sample annotation are available in the RDML data ormat (Supplementary able S2).cDNAs derived rom matched FF and FFPE tumor tissue pairs were run simulta-neously in the same card. Te 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 sofware (Applied Biosystems). Te maximum Cq value was set at 40. Cq values were recorded in Microsot Excel 2003 or subsequent calculations and analyzed using GenEx sofware rom MultiD Analyses (Version 4.3.2;   Göteborg, Sweden) and Prism 4 rom 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.   www.Bioechniques.com 󰀳󰀹󰀲  Vol. 󰀴󰀸 | No. 󰀵 | 󰀲󰀰󰀱󰀰 Reports Normalization factors  We applied two dierent published methods to identiy 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 or all genes in a given set o samples, and determines the optimal number o control genes needed or 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 bioinormatics tool, NormFinder, to find the best endogenous control gene and to determine the best combination o two genes, taking into consideration two subgroups, FF and FFPE samples, in order to find the best endogenous control genes in both materials. he program  provides a stability value for each candidate gene and highlights the best gene with the lowest stability value or one group analysis (NormFinder), as well as the best combination of two genes for a two-gene normalization actor or two subgroups (NormFinder2). Te perormance o the mean Cq per sample was also tested, which has been  previously validated as a new method or miRNA R-qPCR data normalization (17). In addition, we developed a new model or selecting control genes based on housekeeping coefficient o variation (CV)   and Pearson correlation coei- 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 o variation or control gene i  in material   j  , and r  ij   is the Pearson correlation coeicient o expression o gene i  with the mean gene expression per sample in material   j  . Tis equation provides a value  a i , which permits the ranking o 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. Tose genes  with r ≤ 0 should not be considered or NorMean analysis. Using this ranking, we calculated different normalization factors by stepwise inclusion o control genes and geometric averaging o their expression levels. Te 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 shif 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. Ten, 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 o 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 R-qPCR between FF and FFPE tissue pairs. Statistical analyses were performed with the SPSS sofware 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 R-qPCR using LDAs. Te mean raw Cq  was 26.71 and the median raw Cq was 26.91 or 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 rom FFPE material than in RNA prepared from FF samples (9,10,23). Te average expression levels of each tumor are visualized separately for FF and FFPE samples as box  plots in Figure 1. Te raw Cq values are available in Supplementary ables S1 and S2. Capillary electrophoresis analysis showed that RNA extracted rom archival FFPE breast cancer specimens was essentially degraded (Supplementary Figure S1), which is consistent with previous observa- tions (9,10). Tis loss o intact amplicon template explains the five raw Cq shif observed in FFPE specimens with respect to the matched FF samples. Te samples with a greater degree o RNA degradation showed higher median Cq values, confirming the inverse relation between average Cq and the quality o RNA. Comparison of gene expression profiles between FF and FFPE tissues  We compared the R-qPCR raw data o the 30 FF and FFPE sample  pairs to measure the sample correlation coefficients. Correlation was assessed by calculating the Pearson’s correlation coefficient o raw Cqs or 95 genes. A mean correlation coefficient o 0.81 ± 0.073 was ound between the corresponding Cq values in the 30 tumors, ranging rom 0.63 to 0.92. Figure 2 shows the mean raw Cq of each gene in both materials. In spite o the observed Cq shif, 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 o expression of each gene is maintained in both materials; that is, genes with low Cqs (high level o 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. Tis gives rise to a much more relevant question, as it allows the assessment o whether the expression  profile o a given gene is conserved in both materials. In this case, the correlations were lower, with a mean correlation coefficient o 0.33 ± 0.189, ranging rom 0.005 to 0.81. Tese results indicate that RNA degradation affects correlation in gene expression patterns between FF and FFPE materials, despite the act that relative levels o expression are conserved or each gene. Tis could be due to the extensive degradation o RNA obtained rom FFPE. Several studies have described that the observed changes in Cq  values induced by RNA ragmentation can be partially compensated by appropriate normalization (9,10,23). Identification of endogenous control genes  We used two previously described methods o normalization, geNorm (15) and NormFinder (16), to select genes with more stable expression. Given that mean gene expression o each sample reflects the quality o the RNA isolated rom that sample, we also tested the perormance o mean Cq per sample as a normalization actor. Finally, we have developed a new model or selecting control genes.  We propose that the normalization actor should be able to control or the various levels o experimental variability in R-qPCR, mainly the quantity o RNA and differences in enzymatic efficiencies, as well as or differences in the quality o the starting material. As we have seen, the mean expression o each sample reflects its RNA quality. Corre-spondingly, we generated a model that took into account the control gene coefficient o variation as a measure o 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). Te first part o the model should correct or   the differences in RNA quantity and enzyme efficiencies, and the second part should correct or the differences in RNA quality.  We called this model NorMean. Te best endogenous control genes obtained by each o these our methods are shown in able 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 (able 2). Te highest correlations were observed between geNorm and NorMean, while NormFinder2 and NorMean showed the lowest correlations. It was striking to observe that the degree o correlation or each gene in the same material was very high, independent o the normalization method employed. Moreover, we would like to highlight the good perormance o single gene–based normalization actors compared with mutigene normalization methods, although it has been described that multigene normalization methods outperorm single gene–derived normalization methods (15,16,24). Tese 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 or each gene across the 30 matched samples ranged -0.098–0.951 (mean = 0.56). Te respective values for NorMean ranged 0.074–0.936 (mean = 0.57).  We then compared the perormance o each normalization method by contrasting the percentage o genes that show a significant correlation coefficient between FF and FFPE (Figure 3). Raw Cq data showed only 40% o significantly correlated genes between both materials. All normalization actors included in this study increased the percentage o genes significantly correlated. Te best results were obtained with the normalization actor derived rom NorMean, with >80% o 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  www.prior.com • 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   www.BioTechniques.com 󰀳󰀹󰀴 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 LDA; 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 LDAs, 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 R-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). Terefore, 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. Te 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 www.mdbioproducts.com 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