- Open Access
Development of a microarray platform for FFPET profiling: application to the classification of human tumors
© Duenwald et al; licensee BioMed Central Ltd. 2009
Received: 15 January 2009
Accepted: 28 July 2009
Published: 28 July 2009
mRNA profiling has become an important tool for developing and validating prognostic assays predictive of disease treatment response and outcome. Archives of annotated formalin-fixed paraffin-embedded tissues (FFPET) are available as a potential source for retrospective studies. Methods are needed to profile these FFPET samples that are linked to clinical outcomes to generate hypotheses that could lead to classifiers for clinical applications.
We developed a two-color microarray-based profiling platform by optimizing target amplification, experimental design, quality control, and microarray content and applied it to the profiling of FFPET samples. We profiled a set of 50 fresh frozen (FF) breast cancer samples and assigned class labels according to the signature and method by van 't Veer et al  and then profiled 50 matched FFPET samples to test how well the FFPET data predicted the class labels. We also compared the sorting power of classifiers derived from FFPET sample data with classifiers derived from data from matched FF samples.
When a classifier developed with matched FF samples was applied to FFPET data to assign samples to either "good" or "poor" outcome class labels, the classifier was able to assign the FFPET samples to the correct class label with an average error rate = 12% to 16%, respectively, with an Odds Ratio = 36.4 to 60.4, respectively. A classifier derived from FFPET data was able to predict the class label in FFPET samples (leave-one-out cross validation) with an error rate of ~14% (p-value = 3.7 × 10-7). When applied to the matched FF samples, the FFPET-derived classifier was able to assign FF samples to the correct class labels with 96% accuracy. The single misclassification was attributed to poor sample quality, as measured by qPCR on total RNA, which emphasizes the need for sample quality control before profiling.
We have optimized a platform for expression analyses and have shown that our profiling platform is able to accurately sort FFPET samples into class labels derived from FF classifiers. Furthermore, using this platform, a classifier derived from FFPET samples can reliably provide the same sorting power as a classifier derived from matched FF samples. We anticipate that these techniques could be used to generate hypotheses from archives of FFPET samples, and thus may lead to prognostic and predictive classifiers that could be used, for example, to segregate patients for clinical trial enrollment or to guide patient treatment.
While genome-wide mRNA profiling with microarrays has been widely used with fresh frozen (FF) total RNA, few discovery platforms have reliably been applied to formalin-fixed paraffin-embedded tissue (FFPET) samples. Some approaches that assay fewer transcripts (e.g., DASL [Illumina, San Diego, CA] or HTG [Tucson, AZ]) are promising, but do not allow for unbiased discovery of diagnostic signatures, which requires a genome-wide profiling method [2, 3]. For example, the DASL assay has been modified to accommodate several thousand genes and was used to derive an expression signature correlated with survival in hepatocellular carcinoma patients . While such sub-genomic platforms may be useful when the target genes are known, applications such as the discovery of biomarkers and the development of de novo classifiers specifically benefit from a more comprehensive genomic profile. Standard extraction and amplification microarray protocols (e.g., the Arcturus Paradise Reagent System  and NuGEN's Ovation FFPE method ) and array platforms (Affymetrix Human X3P arrays [7, 8]) have been adapted to handle FFPET samples, but typically generate detectable present call rates only on the order of 30% [7–9], and if the block is more than 5 years old, present call rates can drop below 20% . While successful whole-genome profiling from FFPE has recently been shown with one color Affymetrix arrays , similar approaches for two-color arrays have not yet been developed.
The importance of expression-based classification of human tumors to predict treatment response or disease outcome is highlighted by recent publications [9, 12]. van't Veer et al  were one of the first to apply microarray methods for profiling a group of young, lymph node negative patients with primary invasive breast carcinoma with known treatment outcome. Using the initial profiling data as training and test sets [1, 13], a 70-gene prognostic signature was identified to predict disease progression for early-stage estrogen receptor-positive and negative tumor patients. The test was subsequently developed into the first DNA microarray-based expression in vitro diagnostic test (MammaPrint™) cleared by the FDA for actionable decisions in the risk-management of disease. However, these landmark analyses and the subsequent tests have two disadvantages – they require as much as 5 μg total RNA and they require FF tumor samples .
In the clinical diagnosis of patients with cancer, it is routine to obtain a FFPET sample, but generally rare to obtain a FF sample. Consequently, the requirement of FF samples has limited expression profiling to patients treated at specialized research centers. The ability to use FFPET samples would make this technology available for virtually all cancer patients both in the context of retrospective analyses of banked samples and clinical trials seeking to identify molecular tumor characteristics associated with patient outcomes to treatment. Being able to do such analyses from FFPET samples would simplify sample biopsy collection requirements and enable retrospective studies to develop and test hypotheses for prognostic classifiers for other cancers. As additional tests are developed and as molecular profiling methods mature, health-care providers will come to rely more on such classifiers in the risk-management of disease.
In this report, our primary focus was to develop sample processing and classification methods with archived FFPET samples for hypothesis generation. To this end, we optimized a microarray platform and applied it to the profiling of FFPET samples. We demonstrated that FFPET samples can be accurately assigned to class labels using a classifier developed from fresh frozen samples, and we show that a classifier derived from FFPET samples that performs well in classifying FF samples can be developed.
Human tissue samples and reagents
Matched pairs of Fresh Frozen and Formalin-Fixed Paraffin-Embedded breast cancer samples were obtained from Genomics Collaborative (Bioserve, Beltsville, MD) and Cytomyx (Lexington, MA). Colon carcinoma tissues were also from Cytomyx. RNA extraction reagents for FFPET samples were obtained from Epicentre Biotechnologies (Madison, WI). Jurkat total RNA and amplification reagents used in this study were from Ambion (Austin, TX). Matched FF and FFPET liver and muscle RNAs were obtained from MPI Research (Mattawan, MI). The Universal Human Reference (UHR) total RNA was obtained from Stratagene (La Jolla, CA). Cy-dye reagents were from GE Health sciences (Piscataway, NJ). Quantitative PCR reagents were purchased from Applied Biosystems (now Life Technologies, Foster City, CA). Microarrays were designed at Rosetta and manufactured by Agilent Technologies (Santa Clara, CA).
RNA extraction from matched FF and FFPET samples
Total RNA from FF samples was extracted by the vendor (Bioserve, Beltsville, MD) immediately prior to shipment. For FFPET samples, the extraction protocol is adopted from MasterPure RNA Purification kit (Epicentre Biotechnologies, Madison, WI). Briefly, three 10 μm sections were subjected to paraffin solubilization with xylene. Tissue was pelleted from solution by centrifugation and residual xylene was removed by two ethanol rinses. The tissue pellets were then air dried and digested overnight in a lysis buffer with Proteinase K. Digested protein and other cellular components were removed by ammonium acetate precipitation and centrifugation followed by a DNase I treatment of the resulting supernatant. A second ammonium acetate precipitation to remove any residual protein was then performed prior to ethanol precipitation and nuclease-free water rehydration of the purified total RNA.
Quantitative PCR for Total RNA and cRNA QC
List of qPCR primers
Ribosomal protein L13a
Glyceraldehyde-3 Phosphate Dehydrogenase
Amyloid beta precursor protein binding protein 1
RNA target preparation (cRNA amplification)
Target cRNA labelling and hybridization
Five micrograms of cRNA from each experimental sample were dried down and labelled with Cy3 dyes (Invitrogen), and co-purified with the same mass of cRNA from the UHR pool labelled with Cy5. Hybridizations were done in fluor-reversed pairs as described . Labeling, microarray hybridizations, scanning of the slides, and image processing were described in Marton et al .
Since we are mostly interested in ratio profiles between FFPET samples, we developed methods to recover such information by re-ratioing the ratio profiles derived from each array experiment between the FFPET samples and the corresponding hybridizations with UHR as the reference. Re-ratio of the ratio profiles effectively cancels out the UHR profile while accounting for dye labelling biases and leaving only the FFPET profiles of interest. In other words, given a typical configuration in which the common reference (C) exists in two array-based ratio profiles, if hybridization 1 consists of the ratio C versus A (A/C), and hybridization 2 consists of the ratio C versus B (B/C), re-ratioing of A/C and B/C creates a new ratio experiment B versus A (A/B).
Clustering, statistical analysis and classification methods
Gene clustering was performed independently by using an agglomerative hierarchical algorithm . Pairwise similarity metrics among genes are calculated on the basis of expression ratio measurements across all experimental samples. A detailed procedure and calculation was described in van 't Veer et al . We defined the Odds Ratio (OR) for this study as the ratio of odds of developing distant metastases within 5 years for a patient in the referenced study with a tumor characterized by a poor prognosis signature to the odds of developing metastases without this signature (2 × 2 Table). The development of the FFPET-based classifier using supervised classification was performed with following three steps: 1) selection of discriminating candidate genes by their correlation with the category; 2) determination of the optimal set of reporter genes using a leaving-one-out cross validation procedure; 3) prediction of class label (or prognosis or diagnostic) based on the gene expression of the optimal set of reporter genes. The significance of the classifier was estimated by the OR and the 95% confidence interval. The p-value associated with the OR was calculated by Fisher's exact test. More detailed of statistical analysis and classification were reported by van 't Veer et al .
Novel DNA microarray probe design
Probe content for FFPET 44 k v1.0 development array
# of Probes
Current default OSD
Current default OSD with increased weighting towards 3' end
Current default OSD weighted towards 100 nt from 3' end
Tile at 30 nt up to 400 nt from 3' end
Tiled house keeping probes to 1000 nt from 3' end
For each probe, the potential to cross-hybridize was calculated by comparing the probe sequence against the human transcriptome using sequence similarity and the 'BLAST' algorithm. The binding affinities between the probe and the transcript sequences identified by BLAST were computed in terms of dG (delta G). The cross-hybridization score reflects the smallest difference between the dG value of self binding and the largest dG value of the probe binding to any other molecular species. However, despite making the cross hybridization filters more lenient, we noticed that the cross hybridization can only be minimized since certain probes still contain relatively higher GC content compared to the majority of probes on the array in order to meet the 3' distance requirement. Selection of the genes for the HumFFPET 44 k v2.0 array was kept as close as possible to those on the current default Human 44 k v1.1 array. The human 44 k v1.1 and HumFFPET v2.0 array share 20,327 probes in common while the HumFFPET v2.0 array has 19,231 unique probes that are specifically designed for FFPET samples. Documentation for the HumFFPET 44 k v2.0 array will be available in the Gene Expression Omnibus (GEO) website in support of this publication, and the HumFFPET 44 k v2.0 array pattern will be publicly available through the Agilent eArray ordering system.
Results and Discussion
Formalin-preserved samples present multiple challenges to whole-genome RNA profiling methodologies. To overcome the extensive degradation, contaminants from the formalin treatment and limited RNA mass availability, we sought to improve 1) robustness of the amplification protocol, 2) quality control assessment of FFPET samples, 3) microarray performance through probe selection and 4) experimental design by validating the use of UHR as the reference channel in two-color hybridization experiments.
Improvement of the robustness of the RT-IVT amplification protocol
We started with a commercially available two-round RT-IVT protocol for the target cRNA preparation for non-FFPET samples (Additional file 1) but observed it was not sufficiently robust in our hands to be adopted as a high-throughput, automated protocol. We reasoned it was essential that the protocol have properties such as a linear input-dose response curve and be robust to mass imbalances and that these properties may be dependent on sample mass input. We determined the optimal mass input amounts in both the first and second round RT reactions that provided both a linear dose-response and insensitivity to mass input imbalances. The relationship between the input amount and yield for both FF (Jurkat cell line) and FFPET samples (colon carcinoma) is shown in Figure 1a. The cRNA yields for FF and FFPET are directly proportional to the corresponding total RNA input with a good linear dose-dependence curve; we note that the cRNA yield from the FF total RNA is significantly higher (Figure 1a). It is our experience that a linear relationship between total RNA input and cRNA output is necessary to ensure representative amplification and to avoid artifacts or false positive gene expression signatures on different microarray platforms. Figure 1b shows the cRNA yield distribution of a set of 20 breast carcinoma FFPET sample. The input for the first round amplification was 100 ng total RNA, which yielded 1–2 μg cRNA; the input for the second round was 500 ng, which yielded 60–115 μg cRNA. In the first round amplification, the input total RNA only contains a small fraction of amplifiable RNA (mRNA), usually less than 1% for FFPET samples. However, the input of 500 ng cRNA for the second round is derived from oligo dT-based amplification from the first round, and all cRNA molecules should contain the specific polyT tails for further amplification. Essentially, all cRNA molecules with polyT tails in the second round are expected to be amplifiable following the conversion into T7-promoter-containing double strand cDNA. Thus, we attribute the lower yields in the first round to the lower amount of amplifiable mRNA relative to the second round amplification. This is consistent with the prediction that cRNA input used in the second round contains a greater percentage of amplifiable RNA than the total RNA used in the first round amplification.
Robustness to sample mass imbalances
The rationale for using Jurkat RNA is to model the tolerance of the platform to sample mass variations in terms of mRNA content when the expression data of experimental samples are presented as ratios. Each microarray platform has different degrees of tolerance to sample mass imbalance, which can result from either operational variations during the sample preparation process, or from the variation of amplifiable mRNA content in the samples. For example, different FFPET samples contain very different amounts of amplifiable mRNA even when measured total RNA concentrations are the same (data not shown). We used a mass imbalance of intact Jurkat RNA to assess the potential impact of degradation and chemical modifications that would vary across FFPET samples. While evaluating imbalanced FFPET samples directly was more appealing, we reasoned that the primary impact from the degradation and chemical modifications of the fixation and embedding procedures would be loss of amplifiable mRNA, which the Jurkat RNA experiment adequately approximates.
To quantify the degree to which the amplification steps are susceptible to mass imbalance, we performed titrations of input mass for the first round amplification with 25 to 200 ng of Jurkat total RNA sample to model the impact of mass imbalance on microarrray data quality. Following the first round amplification using different mass inputs, a fixed amount (500 ng) of cRNA derived from each input mass titration in the first round amplification was used for the second round amplification. The resulting cRNA from the second round amplification was labelled and hybridized in fluor-reversed pair that was formed to reflect the initial first round mass imbalance between the reference input of 100 ng and the other mass inputs which were originally titrated in the first round amplification. The same experiments were done to titrate input mass for the second round amplification by holding the mass of the first round constant at 100 ng and varying the second round from 250 to 2,000 ng, with the input of 500 ng used as the baseline to form the reference for the different fluor-reversed pairs.
In these mass imbalance experiments, hybridizations of fluor-reversed pairs that are formed between different mass input are still defined as 'same-vs-same' hybridizations since the exactly same mRNA-containing total RNA are used in the amplification whether the input varied in the first or the second round. If there is no amplification bias resulting from the initial mass inputs, either for the first round or the second round, then these same-vs-same hybridizations should have shown no signatures of differential expression beyond background level. Same-vs-same hybridization data are presented as a heat map in Figure 1c, which shows the second round amplification is less susceptible to spurious signatures resulting from mass imbalance than the first round. Fewer than 500 differential signature genes were detected in the hybridization with up to 2-fold mass imbalance for the first round titration experiment and below 200 for the second round, which corresponds to ~1% and 0.5%, respectively, of false positive rate for the microarray hybridization with 44,000 gene probes. These false positive rates are within the background level normally predicted and defined as the pass and fail metrics for different microarray platforms (Affymetrix and Agilent). Based on the mass imbalance experiment and our operational requirement of the high-throughput amplification procedures, the optimal sample mass inputs for the first and second round amplification were determined to be 100 and 500 ng respectively. The selected inputs have a high tolerance to unintended mass imbalances either up or down from the intended input, which serves as a buffer against the potential sample quality variation inherent in FFPET sample sets and the operational variation during the amplification process.
Use of quantitative PCR for assessing the relative quality of FFPET total RNA and amplified cRNA
The extent and nature of RNA degradation in FFPET blocks depends on FFPET preparation method, length of storage and storage conditions [9, 17, 18]. While the RNA extraction procedure can be optimized to increase RNA yield (data not shown), most FFPET blocks were not prepared with RNA quality preservation in mind. Even if sufficient amounts of total RNA can be recovered from poorly prepared and stored FFPET blocks, one cannot be certain the microarray experimental results derived therefrom will yield biologically meaningful data, even after applying bioinformatic and data processing approaches to compensate for the effects of degradation. For FF samples, the quality the total RNA can be assessed by the integrity of the 18S to 28S ribosomal RNAs (measured by abundance ratio) or the RNA Integrity Number (RIN) . It is not feasible to employ similar methods for assessing the quality of FFPET-derived total RNA samples. Thus, there is a need to measure the relative total RNA quality of FFPET samples and to determine whether data are impacted by poor total RNA quality or by the procedures used to store or process the samples. In particular, we sought to determine whether the quality of the microarray hybridization data for a given sample could be predicted from the integrity of the starting total RNA or the amplified cRNA prior to microarray hybridization.
First, we noted that mean log ratio of some samples displayed a dependence on the distance of the probe to the 3' end of the message. We reasoned that a non-zero slope of this plot indicates a bias in the data quality (i.e., a data artifact), and that the 3' slope of mean log ratio could be used as a key quality metric for microarray hybridization. In fact, the 3' slope metric is analogous to the RNA Degradation metric  on Affymetrix microarrays, which was originally developed to measure bias in array data due to degradation of total RNA. In an ideal experiment, number of signatures should not correlate to 3' slope. From a plot of the slope vs Ct, we determined the region of no correlation to Ct is around a slope of 0.15; therefore, we use 0.15 as the threshold for 3' bias. We suspect the 3' slope metric is a measure of variation in cRNA length introduced by amplification bias or RNA quality. The variation in cRNA length is likely to be induced during first strand cDNA synthesis and likely can be applied to any mRNA amplification method utilizing reverse transcription.
In summary, the hybridization data of a sample of ideal quality should show no correlation between the Ct count of the total RNA or the amplified cRNA and the detected expression pattern, either positive or negative. Since the measured Ct count from the total RNA and amplified cRNA correlate with the quality of microarray hybridization (Figure 2), the qPCR method can be utilized as a cost- and time-effective manner to assess the relative quality of total RNA and cRNA for FFPET microarray assays. It should be noted that measured Ct count in terms of sample quality will be contextual not only on the basal expression of the transcripts in the sample type of interest, but also within the experimental FFPET sample set. Our approach of estimating FFPET sample quality using a qPCR method and the method's effectiveness at predicting array performance (at either total RNA and cRNA levels) improves our confidence in the data quality of the FFPET profile.
Development and optimization of an FFPET array
Using the HumFFPET 44 k 2.0 array with a set of FF and FFPET samples, we found that the measured expression ratio correlation between FF and FFPET samples is also improved on the HumFFPET 44 k 2.0 array (r = 0.80) over the Human 44 k array (r = 0.70, Additional file 2), which compared favorably to correlation observed by others . While the performance of FFPET RNA profiling would not be expected to match that from FF samples, the correlation between matched FFPET and FF samples can be used as a way to optimize an FFPET profiling platform (including microarray and quantitative PCR). Our results suggest that the optimization of probe design on HumFFPET 44 k 2.0 array could improve overall performance of the platform for the profiling of FFPET samples.
Determination of the Optimal Reference Sample for FFPET Profiling
An experiment on the Agilent two-color microarray platform requires the pairing of differentially-labelled experimental and reference samples. Frequently, samples are hybridized to two separate slides in fluor-reversed pairs. The competitive hybridization of differentially labelled samples in the two-color array platform has the potential for increased consistency and reduced susceptibility to protocol variation . The fluor-reversed hybridization is performed to minimize dye bias when two samples are labelled with different dyes. Initially, we focused on self-reference pools because previous studies  clearly showed that for optimal microarray performance, the reference pool should be as similar to the experimental samples as possible (i.e., a self-reference pool). However, such self-reference pools present several challenges, including that they require twice as much sample and that poor quality samples in the pool can skew the expression data and create data artifacts.
Is FFPET RNA Profiling of Sufficient Quality to Classify Samples?
A number of papers have recently been published utilizing microarray technology to identify prognostic and diagnostic biomarkers as a tool to predict treatment response or disease outcome [12, 23]. Such patient segmentation efforts have been successful using microarray data of FF samples. Ideally, one would design a clinical trial to address a specific hypothesis, then collect samples and analyze outcome data for tumor signatures, with validation performed in an independent cohort. While prognostic classifiers of FFPET samples have been reported [4, 10, 11, 23, 24], our aim was to determine whether the improvements we describe enabled the generation of hypotheses and classifiers that could be used in subsequent clinical trials. Therefore, we addressed two questions: 1) how will data from FFPET samples perform in predicting class labels derived from FF samples; and 2) how does the predictive power of a FFPET-derived classifier compare to that of a classifier derived from FF samples?
Performance of FFPET samples in FF-derived classifiers
To answer the first question, we chose a well established biomarker, the breast cancer prognostic signature identified by van 't Veer et al. . In the van 't Veer's publication, the authors developed a "good prognosis" template using the average expression of the 70 prognostic genes of patients with a good outcome, and then calculated a correlation coefficient to each patient's expression profile to derive a prognostic score. For the present study, we obtained 50 matched pairs (FF and FFPET) of breast cancer samples and calculated the prognostic scores using the methods and templates (both "good template" and "poor template") of van 't Veer et al . Class labels ("good prognosis" and "poor prognosis") were assigned to each of the 50 patients using the FF samples; we then tested how well the FFPET data could match the FF sample class labels (using the same method and templates). Since no clinical or outcome data were available for these samples, the present study does not allow for a direct validation of the prognostic signatures (that is, this was not an objective for this study).
Performance of classifiers derived from FFPET samples
Comments on the use of FFPET samples for personalized medicine
Enabling personalized medicine in the near future will rely to a large extent on extracting data from well-annotated archived samples for which the three-to-five year outcome of the subject is known. We consider this approach as a bridging strategy for patient stratification and enrollment, during which time the hypotheses are tested and confirmed. Part of this strategy includes integrating FDA guidelines relating to retrospective sample and data mining; key factors in this regard are (1) storage conditions; (2) samples are representative of intended use; (3) samples meet inclusion/exclusion criteria; and (4) performance is comparable to that expected from a prospective study . The FDA has expressed reluctance to accept retrospective data as support for a label that will be used to make clinical decisions because the protocol was not an integral part of development and may not meet the agency's scientific standards for the assay . Therefore, while we demonstrated the ability to develop hypothesis-generating classifiers from FFPET samples based on concordance with classifiers from FF samples, we anticipate significant work remains to validate the results.
We demonstrated the ability to derive gene expression-based classifiers from FFPET data that sort patient samples into class labels that recapitulates the sorting of FF samples. The method involved the development of an optimized microarray platform with two-round RT-IVT amplification using 100 ng of total RNA input for the target preparation, a quantitative PCR method for assessing the relative quality of FFPET samples and a custom microarray with content and probe features specifically optimized for FFPET profiling. We found this microarray platform reliably and reproducibly measured differential gene expression of FFPET samples with a good correlation to corresponding FF samples. FFPET samples were correctly assigned to class labels developed from FF-derived classifiers. Although we cannot directly attribute the success of the classifiers to the optimizations performed on the profiling platform, further study with FFPET samples also showed our platform is of sufficient quality to enable hypothesis generation that could be validated with FF or FFPET samples in controlled, well-designed clinical trials.
We wish to acknowledge the contributions of the Gene Expression Laboratory staff who assisted with various aspects of this work. We thank Chris Raymond for cloning and developing an assay for APPBP1. The support of Mark Parrish, Chris Wright, Joel Klappenbach and Deborah Kessler is gratefully acknowledged.
- van't Veer LJ, Dai H, Vijver van de MJ, He YD, Hart AA, Mao M, Peterse HL, Kooy van der K, Marton MJ, Witteveen AT, Schreiber GJ, Kerkhoven RM, Roberts C, Linsley PS, Bernards R, Friend SH: Gene expression profiling predicts clinical outcome of breast cancer. Nature. 2002, 415: 530-6. 10.1038/415530a.View ArticleGoogle Scholar
- Ravo M, Mutarelli M, Ferraro L, Grober OM, Paris O, Tarallo R, Vigilante A, Cimino D, De Bortoli M, Nola E, Cicatiello L, Weisz A: Quantitative expression profiling of highly degraded RNA from formalin-fixed, paraffin-embedded breast tumor biopsies by oligonucleotidenucleotide microarrays. Lab Invest. 2008, 88: 430-40. 10.1038/labinvest.2008.11.View ArticlePubMedGoogle Scholar
- Rimsza LM, Leblanc ML, Unger JM, Miller TP, Grogan TM, Persky DO, Martel RR, Sabalos CM, Seligmann B, Braziel RM, Campo E, Rosenwald A, Connors JM, Sehn LH, Johnson N, Gascoyne RD: Gene expression predicts overall survival in paraffin-embedded tissues of diffuse large B-cell lymphoma treated with R-CHOP. Blood. 2008, 112: 3425-33. 10.1182/blood-2008-02-137372.PubMed CentralView ArticlePubMedGoogle Scholar
- Hoshida Y, Villanueva A, Kobayashi M, Peix J, Chiang DY, Camargo A, Gupta S, Moore J, Wrobel MJ, Lerner J, Reich M, Chan JA, Glickman JN, Ikeda K, Hashimoto M, Watanabe G, Daidone MG, Roayaie S, Schwartz M, Thung S, Salvesen HB, Gabriel S, Mazzaferro V, Bruix J, Friedman SL, Kumada H, Llovet JM, Golub TR: Gene expression in fixed tissues and outcome in hepatocellular carcinoma. N Engl J Med. 2008, 359: 1995-2004. 10.1056/NEJMoa0804525.PubMed CentralView ArticlePubMedGoogle Scholar
- Furusato B, Shaheduzzaman S, Petrovics G, Dobi A, Seifert M, Ravindranath L, Nau ME, Werner T, Vahey M, McLeod DG, Srivastava S, Sesterhenn IA: Transcriptome analyses of benign and malignant prostate epithelial cells in formalin-fixed paraffin-embedded whole-mounted radical prostatectomy specimens. Prostate Cancer Prostatic Dis. 2008, 11: 194-7. 10.1038/sj.pcan.4501007.View ArticlePubMedGoogle Scholar
- Scicchitano MS, Dalmas DA, Mertiaux MA, Anderson SM, Turner LR, Thomas RA, Mirable R, Boyce RW: Preliminary comparison of quantity, quality and microarray performance of RNA extracted from formalin-fixed, paraffin-embedded and unfixed frozen tissue samples. J Histochem Cytochem. 2006, 54: 1229-10.1369/jhc.6A6999.2006.View ArticlePubMedGoogle Scholar
- Frank M, Döring C, Metzler D, Eckerle S, Hansmann M-L: Global gene expression profiling of formalin-fixed paraffin-embedded tumor samples: a comparison to snap-frozen material using oligonucleotide microarrays. Virchows Archiv. 2007, 450: 699-711. 10.1007/s00428-007-0412-9.View ArticlePubMedGoogle Scholar
- Lee S, Medina D, Tsimelzon A, Mohsin SK, Mao S, Wu Y, Allred DC: Alterations of gene expression in the development of early hyperplastic precursors of breast cancer. Am J Pathol. 2007, 171: 252-62. 10.2353/ajpath.2007.061010.PubMed CentralView ArticlePubMedGoogle Scholar
- Farragher SM, Tanney A, Kennedy RD, Paul Harkin D: RNA expression analysis from formalin fixed paraffin embedded tissues. Histochem Cell Biol. 2008, 130: 435-45. 10.1007/s00418-008-0479-7.View ArticlePubMedGoogle Scholar
- Paik S, Kim CY, Song YK, Kim WS: Technology insight: Application of molecular techniques to formalin-fixed paraffin-embedded tissues from breast cancer. Nat Clin Pract Oncol. 2005, 2: 246-54. 10.1038/ncponc0171.View ArticlePubMedGoogle Scholar
- Linton KM, Hey Y, Saunders E, Jeziorska M, Denton J, Wilson CL, Swindell R, Dibben S, Miller CJ, Pepper SD, Radford JA, Freemont AJ: Acquisition of biologically relevant gene expression data by Affymetrix microarray analysis of archival formalin-fixed paraffin-embedded tumours. Br J Cancer. 2008, 98: 1403-14. 10.1038/sj.bjc.6604316.PubMed CentralView ArticlePubMedGoogle Scholar
- Glas AM, Floore A, Delahaye LJ, Witteveen AT, Pover RC, Bakx N, Lahti-Domenici JS, Bruinsma TJ, Warmoes MO, Bernards R, Wessels LF, Van't Veer LJ: Converting a breast cancer microarray signature into a high-throughput diagnostic test. BMC Genomics. 2006, 7: 278-81. 10.1186/1471-2164-7-278.PubMed CentralView ArticlePubMedGoogle Scholar
- Vijver van de MJ, He YD, van't Veer LJ, Dai H, Hart AA, Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton MJ, Parrish M, Atsma D, Witteveen A, Glas A, Delahaye L, Velde van der T, Bartelink H, Rodenhuis S, Rutgers ET, Friend SH, Bernards R: A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med. 2002, 347: 1999-2009. 10.1056/NEJMoa021967.View ArticlePubMedGoogle Scholar
- van't Veer LJ, Bernards R: Enabling personalized cancer medicine through analysis of gene-expression patterns. Nature. 2008, 452: 564-70. 10.1038/nature06915.View ArticlePubMedGoogle Scholar
- Marton MJ, DeRisi JL, Bennett HA, Iyer VR, Meyer MR, Roberts CJ, Stoughton R, Burchard J, Slade D, Dai H, Bassett DE, Hartwell LH, Brown PO, Friend SH: Drug target validation and identification of secondary drug target effects using DNA microarrays. Nat Med. 1998, 4: 1293-301. 10.1038/3282.View ArticlePubMedGoogle Scholar
- Eisen MB, Spellman PT, Brown PO, Botstein D: Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA. 1998, 95: 14863-8. 10.1073/pnas.95.25.14863.PubMed CentralView ArticlePubMedGoogle Scholar
- Cronin M, Pho M, Dutta D, Stephans JC, Shak S, Kiefer MC, Esteban JM, Baker JB: Measurement of gene expression in archival paraffin-embedded tissues. AJP. 2004, 164: 35-42.PubMed CentralPubMedGoogle Scholar
- Penland SK, Keku TO, Torrice C, He X, Krishnamurthy J, Hoadley KA, Woosley JT, Thomas NE, Perou CM, Sandler RS, Sharpless NE: RNA expression analysis of formalin-fixed paraffin-embedded tumors. Lab Invest. 2007, 87: 383-91.PubMedGoogle Scholar
- 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: 3-10.1186/1471-2199-7-3.PubMed CentralView ArticlePubMedGoogle Scholar
- Gautier L, Cope L, Bolstad BM, Irizarry RA: Affy – Analysis of Affymetrix GeneChip data at the probe level. Bioinformatics. 2004, 20: 307-315. 10.1093/bioinformatics/btg405.View ArticlePubMedGoogle Scholar
- He YD, Dai H, Schadt EE, Cavet G, Edwards SW, Stepaniants SB, Duenwald S, Kleinhanz R, Jones AR, Shoemaker DD, Stoughton RB: Microarray standard data set and figures of merit for comparing data processing methods and experiment designs. Bioinformatics. 2003, 19: 956-65. 10.1093/bioinformatics/btg126.View ArticlePubMedGoogle Scholar
- MAQC Consortium: The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nat Biotechnol. 2006, 24: 1151-61. 10.1038/nbt1239.PubMed CentralView ArticleGoogle Scholar
- Bibikova M, Chudin E, Arsanjani A, Zhou L, Garcia EW, Modder J, Kostelec M, Barker D, Downs T, Fan JB, Wang-Rodriguez J: Expression signatures that correlated with Gleason score and relapse in prostate cancer. Genomics. 2007, 89: 666-72. 10.1016/j.ygeno.2007.02.005.View ArticlePubMedGoogle Scholar
- Coudry RA, Meireles SI, Stoyanova R, Cooper HS, Carpino A, Wang X, Engstrom PF, Clapper ML: Successful application of microarray technology to microdissected formalin-fixed, paraffin-embedded tissue. J Mol Diagn. 2007, 9: 70-9. 10.2353/jmoldx.2007.060004.PubMed CentralView ArticlePubMedGoogle Scholar
- Philip R, Chan MM, Gutman SI: FDA Perspectives on Validating Proteomic Biomarkers for In Vitro Diagnostics Use. American Drug Discovery. 2008, 3: 30-33.Google Scholar
- Usdin Steve: Proof is in the pudding. Biocentury. 2008, 16: A1-21.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.