| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Imaging, Diagnosis, Prognosis |
1 Cancer Diagnosis Program, 2 Biometric Research Branch, and 3 Center for Bioinformatics, National Cancer Institute, Bethesda, Maryland; Departments of 4 Surgery, 5 Pediatrics, and 6 Pathology, University of Michigan Medical School and 7 Department of Biostatistics, University of Michigan, Ann Arbor, Michigan; 8 Department of Medical Oncology, Dana-Farber Cancer Institute and Department of Pathology, Harvard Medical School, Boston, Massachusetts; 9 Department of Surgery, H. Lee Moffitt Cancer Center and Research Institute, University of South Florida, Tampa, Florida; Departments of 10 Pathology and 11 Molecular Biology, Memorial Sloan-Kettering Cancer Center, New York, New York; 12 Hamon Center for Therapeutic Oncology Research, University of Texas Southwestern Medical Center, Dallas, Texas; and 13 Whitehead Institute-Massachusetts Institute of Technology Center for Genome Research, Cambridge, Massachusetts
Requests for reprints: Kevin K. Dobbin, Division of Cancer Treatment and Diagnosis, National Cancer Institute, Rockville, MD 20852. Phone: 301-451-6244; E-mail: dobbinke{at}mail.nih.gov.
 |
ABSTRACT |
|---|
 Top ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Key Words: oligonucleotide microarrays • reproducibility • interlaboratory comparison • gene expression
|
INTRODUCTION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
A study is being planned to confirm previously reported associations of gene expression signatures with patient outcome in stage I lung adenocarcinomas (4–6). Four laboratories will participate in the confirmatory study where 600 tumors will be analyzed with the goal of combining data from all sites. The main goal of this preliminary laboratory comparability study was to determine whether differences between the laboratories in tissue processing, RNA extraction, generation of labeled target, hybridization, and scanning result in comparable gene expression measurements from the same samples. If the data from the four laboratories were not comparable, then an alternative experimental design would be required for the confirmation study. The preliminary study was designed to identify sources of variation in gene expression measurements from frozen tissues, cell line samples, and purified RNA samples analyzed with Affymetrix Human Genome U133A arrays.
We are aware of no published studies that evaluate the interlaboratory comparability for microarray data on human tumor specimens. Piper et al. (12) assessed interlaboratory comparability of microarray measurements of yeast cell cultures grown in various conditions. There have additionally been some small, unpublished studies on limited numbers of extracted RNA samples. Because tissue handling and RNA extraction are considered major sources of variability in the assay for tumor tissue specimens, these previous studies are insufficient for our needs.
To assess the results of the interlaboratory comparability, we needed some estimate of the intralaboratory comparability to use as a baseline. Previous studies of intralaboratory comparability did not use the standardized protocols developed for this experiment or involve all aspects of the tumor assay process, including tissue processing (selecting and physically cutting the tissue to be assayed from the frozen tumor block, etc.), extraction of RNA from the tissue specimen, preparation of labeled cRNA target (reverse-transcription, labeling, fragmentation, etc.), and array hybridization, washing, and scanning. To address all of these issues, we designed an experiment to determine the intralaboratory and interlaboratory comparability of DNA microarrays at four laboratories using a standardized protocol.
|
MATERIALS AND METHODS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
Description of Tumor Tissue Samples, Cell Line Samples, and RNA Samples
Frozen tissues were collected at the Department of Pathology of the University of Michigan Health System from surplus tissue from surgical specimens. Selected tissues were large and grossly and histologically homogeneous to reduce variability due to tumor heterogeneity between replicates. As the goal of our study was to examine variability within replicates from a tissue sample and not between different tissue samples, it was decided that the tissues could be of varying histogenesis and/or differentiations. Thus, 12 tissues, predominantly tumors, of diverse histogenesis and differentiation were selected, although 2 lung carcinomas were specifically included. Six frozen OCT blocks were prepared from each of the 12 tissue samples. Tissues were coded A-L and de-identified from the patient after the diagnosis was recorded. RNA quality was roughly assessed for each of the 12 tissues by gel electrophoresis. The OCT blocks were distributed to the four laboratories on dry ice, and each laboratory prepared a frozen section and selected the area to be used for RNA extraction. The tissues selected included two primary lung squamous cell carcinomas; adrenal cortical adenoma; primary gastric adenocarcinoma; normal liver; recurrent renal cell carcinoma, chromophobe cell type; primary malignant gastrointestinal stromal tumor; uterine leiomyoma; primary ovarian papillary serous adenocarcinoma; metastatic renal cell carcinoma, clear cell type; primary large cell lymphoma; and metastatic melanoma to lymph node.
RNA samples extracted from lung adenocarcinoma tumors were obtained from Ardais Corp. (Lexington, MA). The tumors were from a variety of patients with various stages of disease.
Cell line samples were obtained from the University of Texas Southwestern Medical Center at Dallas. The lung adenocarcinoma tumor cell lines used were NCI-H2009, NCI-H1437, NCI-H2087, NCI-H2347, and HCC78. All but HCC78 are deposited in the American Type Culture Collection (Manassas, VA; http://www.atcc.org). Cells were grown to 
90% confluence, trypsinized, washed once in PBS, and pelleted. They were then snap frozen in dry ice/ethanol bath and maintained at –80°C until shipping to the various sites.
Further details on the samples used can be found in the Supplementary Material.
Specimen Preparation and Laboratory Procedures
Tumors were sectioned using a cryostat, and several 5 to 8 µm cryostat sections were quickly obtained and then stained with hematoxylin to select the regions for use in RNA isolation. Typically, these specimens were large enough to obtain all of the material for RNA from a single region (e.g., 5-10 mm3). Care was taken to avoid warming the specimen no more than to –18°C for the shortest time possible. The selected regions were at least 60% tumor cells (tumor cellularity), and tumors having mixed histology (i.e., adenosquamous) were not used. The tumor portions chosen for RNA isolation were obtained by cutting out that region of the tumor using a razor blade or scalpel cooled using dry ice. The material was then placed in labeled tubes and maintained at –80°C before RNA isolation.
RNA Isolation. RNA was isolated using 1 mL volume of Trizol (Invitrogen, Inc., Carlsbad, CA) reagent per sample using the protocol provided by the manufacturer. After RNA precipitation and 70% ethanol wash, the pellet was resuspended in RNase-free water and further purified using the RNeasy columns (Qiagen, Inc., Valencia, CA) as described by the manufacturer. RNA was eluted from the columns with RNase-free water. RNA concentration was determined by spectrometry, one 5 µg aliquot of the sample was used for reverse transcription, and another 1 µg portion was used for assessing RNA quality using the Agilent bioanalyzer.
cRNA Synthesis and Hybridization: First-Strand Synthesis. Briefly, 5 µg of total RNA were resuspended in Ambion (Austin, TX) DEPC-treated water and quantified by A260, and the quality was determined using both the A260 ratio and Agilent bioanalyzer. RNA was converted into double-stranded cDNA by reverse transcription using a cDNA synthesis kit (Invitrogen). The oligo(dT)24 primer [Affymetrix, T7-oligo(dT) Promoter Primer kit] containing a T7 RNA polymerase promoter located 5' to the poly(T) was used. The temperature of incubation was 42°C for 1 hour in a PCR machine. Following incubation, the mixture was quick spun in a centrifuge and placed on ice, and cold premixed second-strand reagents were added.
cRNA Synthesis and Hybridization: Second-Strand Synthesis. The Invitrogen cDNA synthesis kit was used for second-strand synthesis. The incubations were done at 16°C for 2 hours in a PCR machine, and the reaction was stopped using 10 µL of 0.5 mol/L EDTA (Sigma Chemical Co., St. Louis MO). The reactions were immediately treated using a Sample Cleanup module (Affymetrix) and resuspended in 22 µL DEPC water. Labeled cRNA was generated from the cDNA sample by an in vitro transcription reaction that was supplemented with biotin-11-CTP and biotin-16-UTP (Enzo, Farmingdale, NY, via Affymetrix Enzo BioArray High-Yield Transcript Labeling kits) for 6 hours at 37°C in a PCR machine with no shaking. All of the labeled cRNA were purified using Affymetrix GeneChip Sample Cleanup module. The cRNA was quantified (using A260), and 15 µg were fragmented in a total volume of 40 µL fragmentation buffer at 94°C for 35 minutes using a PCR machine. The fragmentation buffer [200 mmol/L Tris-acetate (pH 8.1), 500 mmol/L KOAc, 150 mmol/L MgOAc] was provided in the Affymetrix GeneChip Sample Cleanup module.
Preparation of the Hybridization Cocktail. Fragmented cRNA (15 µg) was used to prepare 300 µL hybridization cocktail (100 mmol/L MES, 1 mol/L NaCl, 20 mmol/L EDTA, 0.01% Tween 20) containing 0.1 mg/mL (3 mL/300 mL) of herring sperm DNA (Promega, Madison, WI, 10 mg/mL) and 500 µg/mL acetylated bovine serum albumin (3 mL/300 mL, Invitrogen, 50 mg/mL). EDTA was obtained from Sigma Chemical, Tween 20 from Pierce Chemical (Rockford, IL), and DEPC-treated water from Ambion. Control cRNA used for comparison of hybridization efficiency between arrays and to standardize the quantitation of measured transcript levels is included as component of Eukaryotic Hybridization Control kit (Affymetrix, 20x) and uses 15 mL/300 mL hybridization cocktail. Before hybridization, the cocktails were heated to 94°C for 5 minutes, equilibrated at 45°C for 5 minutes, and clarified by centrifugation (16,000 x g) at room temperature for 5 minutes. Aliquots of each sample (10 µg fragmented cRNA in 200 µL hybridization cocktail) were prehybridized to U133A arrays at 45°C for 60 minutes and then hybridized for 16 to 18 hours in a rotisserie oven at 60 x g. The arrays were then washed using nonstringent wash buffer (6x saline-sodium phosphate-EDTA) at 25°C followed by stringent wash buffer [100 mmol/L MES (pH 6.7), 0.1 mol/L NaCl, 0.01% Tween 20] at 50°C. After staining with streptavidin-phycoerythrin (Molecular Probes, Eugene, OR), the arrays were washed again with 6x saline-sodium phosphate-EDTA and incubated with biotinylated anti-streptavidin IgG followed by a second staining with streptavidin-phycoerythrin and a third washing with 6x saline-sodium phosphate-EDTA. The arrays were scanned using the GeneArray scanner (Affymetrix). Data analysis was done using Affymetrix GeneChip 5.0 software. Features on the oligonucleotide arrays were carefully reviewed to confirm expression levels and exclude hybridization or washing artifacts.
Statistical Analysis
All data are publicly available for download at http://gedp.nci.nih.gov/ (experiment IDs 615-618). Affymetrix MAS 5.0 gene summaries were obtained for each array under the default parameter settings. Arrays were normalized as described by Wright et al. (13); genes with >50% Affymetrix "present" calls across arrays were identified, and for each array, the expression levels were multiplied by a constant to make the median of the identified genes 500; signal values below 25 were truncated to 25, and the base two logarithm of the normalized intensities served as the signal. No filtering of genes was necessary because the truncation eliminated missing data and large negative log-intensity values. Comparability of two microarray measurements on the same sample was assessed by the Pearson correlation coefficient and the root mean square deviation (14), defined as 
, where G is the number of genes on the array and xi and yi are the normalized log intensity readings for gene i on two different arrays. Pearson correlation was used because it is a common and intuitive measure in microarray studies. This was supplemented by examining the root mean square deviation because it captures aspects of comparability missed by correlation (i.e., systematically higher or lower expression values across genes). Comparability of repeated measurements on the same gene for pairs of replicated samples within a single laboratory, and for pairs of replicated samples between two laboratories, was assessed by intraclass correlation (ICC; refs. 15, 16). The ICC is calculated by fitting an ANOVA model for each gene: log2 (Ysle) = Ss + Ll + Esle, where Y is the normalized intensity, S is the sample effect, L is the laboratory effect, and E is random error. ANOVA models were fit to the frozen tumor specimens and cell lines separately. ANOVA is generally robust, and the assumptions of normal and homogeneous error variances seemed adequate for the vast majority of genes based on 
2 goodness of fit tests and Fligner test for homogeneity of variances. Fitting the model results in estimates of the variance contributed by the different components, 
s2, l2, and e2, for variation attributable to sample, laboratory, and measurement error, respectively. These were estimated from mean squares with adjustment for the partial replication and missing data (see Supplementary Material for details). The estimate of the ICC between laboratories is then 
and within each laboratory is 
. In an ideal experiment, variation attributable to laboratory, l2, and measurement error, s2, would both be near 0, resulting in within-laboratory and between-laboratory ICC values close to 1. Values of the ICC close to 1 indicate good comparability, and values close to 0 indicate poor comparability. Within-laboratory ICC serves as the baseline against which between-laboratory ICC will be assessed. If the variation attributable to laboratory, l2, is large, then the between-laboratory ICC will be small compared with the within-laboratory ICC. If the laboratories are comparable, then the between-laboratory ICC and the within-laboratory ICC will be similar. Separate within-laboratory and between-laboratory ICCs are calculated for each gene.
As a further test of the comparability of the data across sites, we ran hierarchical agglomerative cluster analysis using average linkage separately on the tumor tissue samples, the cell line pellets, and the purified RNA samples. The distance metric (1 – correlation) was used, although Euclidean distance yielded similar results.
|
RESULTS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
High within-laboratory reproducibility of the Stratagene reference RNA samples was expected, but the high within-laboratory reproducibility of both the cell line pellets and the frozen tissue sections is somewhat surprising. The single outlier on the within-laboratory correlations for the tumors (with correlation 0.86) and the three outlier correlations for the between-laboratory tumor correlations (all with correlation around 0.86) all involve a single tumor sample; this particular tumor sample also had a poor Affymetrix RawQ score of 3.97, suggesting quality problems associated with this single microarray. The three outlier between-laboratory cell line correlations for the cell line samples were associated with different sample pairs.
Figure 2 shows box plots of the between-laboratory correlations of repeated measurements on the same samples. The median correlations were 0.93, 0.94, 0.94, and 0.96 for the tumor tissue, cell line pellets, purified RNA, and Stratagene reference RNA based on 66, 57, 30, and 8 pairs of replicated sample measurements, respectively. A similar pattern held for the root mean square deviations (see Supplementary Material). The relationships among the correlations are similar to the within-laboratory results, although the correlations are slightly decreased.
In comparing the between-laboratory with the within-laboratory comparability for the tumors, cell lines, and Stratagene reference RNA, in each case, a decline of 0.02 in median correlation is associated with moving from within-laboratory to between-laboratory comparability. On the other hand, the within-laboratory variations are clearly different for the different sample types, with the tumor tissue most heterogeneous. This suggests that the between-laboratory differences are primarily associated with preparation of labeled cRNA target (reverse-transcription, labeling, fragmentation, etc.) and array hybridization, washing, and scanning and that relatively little additional between-laboratory variability is contributed by tissue handling and RNA extraction.
Hierarchical Clustering Reflects the Biological Variation within the Different Types of Samples
To determine the ability of between-laboratory studies to accurately discriminate among cancer samples, we did hierarchical clustering on the three types of samples: the samples for which purified tumor RNA samples were provided to each laboratory (Fig. 3A), those for which cell line pellets were provided (Fig. 3B), and those for which tumor tissue samples were provided (Fig. 3C). In each of these cases, the hierarchical cluster analysis accurately grouped the related samples together despite the many variables that could separate these samples. It is also interesting to note in Fig. 3C that the primary tumor specimens cluster close together, with the metastatic (l) and normal liver (c) samples distant from the primary tumor group; in interpreting Fig. 3C, recall that the distance between two clusters is represented by the vertical height of the connection between them, not the horizontal distance between them, so that, for instance, the metastatic and normal liver clusters on the right side are far distant from each other.
|
To get a more realistic picture, note that in Fig. 2 the correlations between repeated measures on the same samples decreases by a median of 0.01 when moving from either the cell line or purified RNA to the frozen tissue specimens. This suggests estimating the correlations between biologically different frozen lung adenocarcinoma tissue samples by adding 0.01 to the correlations between the RNA samples and to the correlations between the cell lines. The resulting estimated first quartile, median, and third quartile correlations between different lung adenocarcinoma tissue samples are 0.86, 0.87, and 0.88, respectively. Comparing this with the observed correlations between the repeated observations on the tumor tissues, which had first quartile, median, and third quartile of 0.92, 0.93, and 0.95, respectively, suggests that a more homogeneous set of lung adenocarcinoma tissue samples would still cluster almost entirely by tumor across sites.
Comparability of Repeated Measurements on the Same Gene across Samples
For each gene, we used ANOVA to separate the sources of variation into three types: biological variation due to differences in gene expression among the different biological specimens (the 12 different tissue specimens); laboratory variation resulting from laboratory-to-laboratory variability; and error variation resulting from measurement error, which will be present even when the same sample is measured multiple times in the same laboratory.
Results for the frozen tumor tissues are shown in Fig. 4. In general, the variability attributable to the laboratories was the smallest source of variation followed by variation due to measurement error and finally by biological variation as the largest. This suggests that the different laboratories are contributing some extra variation into the measurements but that variation tends to be relatively small compared with the other sources of variability already present in this type of data obtained from primary tumors. Similar results held for the cell line samples and are given in the Supplementary Material.
|
|
|
|
DISCUSSION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Within-laboratory comparability of measurements was highest for the Stratagene RNA sample, lower for the cell line samples, and lowest for the tumor tissue sections. The additional within-laboratory variability associated with RNA extraction and tissue handling seemed to be relatively minor. Comparability between laboratories generally seemed to be lower than within laboratories for all three types of samples, but the loss in comparability seemed to be fairly minor and chiefly associated with sample labeling, hybridization, and scanning as opposed to tissue handling or RNA extraction. Microarray data between laboratories seems comparable under the protocols used. For future studies, it seems that standardizing equipment, protocols, and reagents associated with preparation of labeled cRNA target (reverse-transcription, labeling, fragmentation, etc.) and array hybridization, washing, and scanning across sites may be important for ensuring comparability.
Much effort was made to standardize the laboratory protocols across the laboratories and to ensure that the Affymetrix scanners, reagents, etc., were as consistent as possible at the sites. The laboratories involved in this study all have extensive experience with Affymetrix gene chips. Three are medium-sized core facilities and one is a large, high-throughput core facility. Thus, it does not seem that these results can necessarily be generalized to less controlled situations or used to justify combining publicly available data from previous studies for analysis. In addition, it is important to bear in mind that some preliminary studies have suggested that comparability across platforms or even across probe set summaries within the same platform may be poor (17), although others have had some success combining data across platforms (18).
It is perhaps surprising how small a role biological variation within the tumors seems to have played, because there was high correlation between all six tumor sections across the tumors (with the exception of the failed arrays of the problematic melanoma). Because the tumors were selected based on apparent homogeneity, these tumors may be particularly homogeneous in gene expression. Nevertheless, the hierarchical clustering analysis was striking because the tumor samples consistently clustered together despite the use of independent tumor sections, the selection of tumor regions at different institutions, and the independent RNA extraction, target labeling, and array hybridization and scanning results. This argues for the reproducibility and potential utility of microarray analysis.
If gene expression signatures are to be clinically useful in the future, they must be converted into standardized assays that are reproducible. How this transition step will occur is not yet clear. It may involve taking the results of microarray studies and developing assays based on different technologies (such as quantitative reverse transcription-PCR). However, the results of this study indicate that it may also be possible to create a standardized assay based on microarray technology that is reproducible enough for clinical use.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Received 6/ 8/04; revised 9/13/04; accepted 10/21/04.
|
REFERENCES |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  C. I. Dumur, M. Lyons-Weiler, C. Sciulli, C. T. Garrett, I. Schrijver, T. K. Holley, J. Rodriguez-Paris, J. R. Pollack, J. L. Zehnder, M. Price, et al. Interlaboratory Performance of a Microarray-Based Gene Expression Test to Determine Tissue of Origin in Poorly Differentiated and Undifferentiated Cancers J. Mol. Diagn., January 1, 2008; 10(1): 67 - 77. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Jayachandran, J.-i. Sazaki, M. Nishizaki, K. Xu, L. Girard, J. D. Minna, J. A. Roth, and L. Ji Fragile Histidine Triad Mediated Tumor Suppression of Lung Cancer by Targeting Multiple Components of the Ras/Rho GTPase Molecular Switch Cancer Res., November 1, 2007; 67(21): 10379 - 10388. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. P. A. Ioannidis Is Molecular Profiling Ready for Use in Clinical Decision Making? Oncologist, March 1, 2007; 12(3): 301 - 311. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. K. Thomas, B. Weir, and M. Meyerson Genomic approaches to lung cancer. Clin. Cancer Res., July 15, 2006; 12(14): 4384s - 4391s. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. S. Verducci, V. F. Melfi, S. Lin, Z. Wang, S. Roy, and C. K. Sen Microarray analysis of gene expression: considerations in data mining and statistical treatment Physiol Genomics, May 16, 2006; 25(3): 355 - 363. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Ma, M. Lyons-Weiler, W. Liang, W. LaFramboise, J. R. Gilbertson, M. J. Becich, and F. A. Monzon In Vitro Transcription Amplification and Labeling Methods Contribute to the Variability of Gene Expression Profiling with DNA Microarrays J. Mol. Diagn., May 1, 2006; 8(2): 183 - 192. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J S Reis-Filho, C Westbury, and J-Y Pierga The impact of expression profiling on prognostic and predictive testing in breast cancer. J. Clin. Pathol., March 1, 2006; 59(3): 225 - 231. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. L. Thompson, B. A. Rosenzweig, P. S. Pine, J. Retief, Y. Turpaz, C. A. Afshari, H. K. Hamadeh, M. A. Damore, M. Boedigheimer, E. Blomme, et al. Use of a mixed tissue RNA design for performance assessments on multiple microarray formats Nucleic Acids Res., December 23, 2005; 33(22): e187 - e187. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. L. Newkirk, J. H.M. Knoll, and P. K. Rogan Distortion of quantitative genomic and expression hybridization by Cot-1 DNA: mitigation of this effect Nucleic Acids Res., December 14, 2005; 33(22): e191 - e191. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Simon Roadmap for Developing and Validating Therapeutically Relevant Genomic Classifiers J. Clin. Oncol., October 10, 2005; 23(29): 7332 - 7341. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Espinosa, J.A. F. Vara, A. Redondo, J.J. Sanchez, D. Hardisson, P. Zamora, F. G. Pastrana, P. Cejas, B. Martinez, A. Suarez, et al. Breast Cancer Prognosis Determined by Gene Expression Profiling: A Quantitative Reverse Transcriptase Polymerase Chain Reaction Study J. Clin. Oncol., October 10, 2005; 23(29): 7278 - 7285. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Bullinger and P. J.M. Valk Gene Expression Profiling in Acute Myeloid Leukemia J. Clin. Oncol., September 10, 2005; 23(26): 6296 - 6305. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Meyerson and D. Carbone Genomic and Proteomic Profiling of Lung Cancers: Lung Cancer Classification in the Age of Targeted Therapy J. Clin. Oncol., May 10, 2005; 23(14): 3219 - 3226. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
