| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Clinical Trials |
1 Thoracic Oncology, Karmanos Cancer Institute, 2 Department of Computer Science, and 3 Center for Molecular Medicine and Genetics, Wayne State University, Detroit, Michigan; 4 Department of Pathology, Harper University Hospital, Detroit, Michigan; 5 Department of Pathology, Cardinal Bernardin Cancer Center, Loyola University, Maywood, Illinois; and 6 Division of Thoracic Surgery, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts
ABSTRACT
Purpose: Clinical outcomes for malignant pleural mesothelioma (MPM) patients having surgery are imprecisely predicted by histopathology and intraoperative staging. We hypothesized that gene expression profiles could predict time to progression and survival in surgically cytoreduced pleural mesothelioma of all stages.
Experimental Design: Gene expression analyses from 21 MPM patients having cytoreductions and identical postoperative adjuvant therapy were performed using the U95 Affymetrix gene chip. Using both dChip and SAM, neural networks constructed a common 27 gene classifier, which was associated with either the high-risk and low-risk group of patients. Data were validated using real-time PCR and immunohistochemical staining. The 27 gene classifier was also used for validation in a separate set of 17 MPM patients from another institution.
Results: The groups predicted by the gene classifier recapitulated the actual time to progression and survival of the test set with 95.2% accuracy using 10-fold cross-validation. Clinical outcomes were independent of histology, and heterogeneity of progression and survival in early stage patients was defined by the classifier. The gene classifier had a 76% accuracy in the separate validation set of MPMs.
Conclusions: These data suggest that pretherapy gene expression analysis of mesothelioma biopsies may predict which patients may benefit from a surgical approach.
Introduction
Malignant pleural mesothelioma (MPM) is an orphan disease of which the management has defied curative options (1) . The incidence of MPM worldwide is increasing, and the estimated number of deaths over the next 30 years from MPM in Western Europe alone could approach 250,000 (2) . Our understanding of the pathophysiology of the disease, however, has improved with the advent of molecular investigations of the downstream effects of asbestos (3) , as well as the realization that SV40 is linked with the disease in >60% of the cases (4) . Studies of the molecular genetics of the disease have revealed the importance of signal transduction pathways including the epidermal growth factor receptor (5) and activator protein (6) , as well as asbestos-SV40 interactions, which could explain such phenomena as increased angiogenesis (7) , telomerase (8 , 9) , MET overexpression (10 , 11) , abrogation of p53-induced apoptosis (12) , and abnormalities of immune surveillance (13) among other findings.
There have been few global investigations using gene array technologies of the multiple genetic events in MPM due to the lack of uniformly treated patients with carefully documented survival and recurrence data, and the absence of large archives of banked frozen tissue with adequately preserved RNA. In addition to the gene-based classification (14) , the ability to prognosticate on the basis of gene expression data from biopsy harvests could have important therapeutic implications for this group of unfortunate patients. Our hypothesis is that gene arrays performed on operative specimens from uniformly treated mesothelioma patients could predict survival and time to progression, and could cluster groups of patients as good or poor risk. Our test set population included stages I-III MPM patients resected by one surgeon. We analyzed our gene expression data using two programs, significance analysis microarrays (SAM) and dCHIP, to see whether concordant genes from both analyses could correctly cluster the patients into good or poor risk cohorts.
Materials and Methods
Patient Population.
Tumor and normal tissues from 21 patients with MPM treated at the National Cancer Institute and the Karmanos Cancer Institute between August 1993 and May 2001 were used as the test set for the studies described below. Consent was received and the project approved by the local Institutional Review Boards. All of the patients had cytoreductive surgery, mediastinal lymph node dissection with staging by the International Mesothelioma Interest Group staging system (15)
, followed by postoperative cisplatinum-based chemotherapy. All of the patients were followed from the surgery until death or through February 2003 with computerized tomography of the chest every 3 months. Interval change, which suggested progression of the MPM, was documented whenever possible by histological confirmation of disease. Tumors were classified as having epithelial, sarcomatoid, or mixed histology MPM by two of the authors (F. L. and M. C.). All of the tumor specimens contained a tumor cellularity of >85%.
Gene Expression Profiling.
RNA isolation and cRNA synthesis were performed as described (16)
. Each probe array was scanned twice (Hewlett-Packard GeneArray Scanner), the images overlayed, and the average intensities of each probe cell compiled. dChip and SAM were used for expression profiling.
Real-Time Quantitative Reverse Transcription-PCR.
Real-time quantitative reverse transcription-PCR was used to confirm differential gene expression profile of Plasmolipin, Calumenin, and IGFBP5. cDNAs were synthesized from 3 µg total RNA using an oligodeoxythymidylic acid primer and the Superscript II reverse transcription kit (Life Technologies, Inc.). The primers used were: Plasmolipin sense, 5'-ATGTTCGTCGCTGTCTTCCT-3'; antisense, 5'-ATCATCACCAAACACGCAAA-3'; Calumenin sense, 5'-CGCTGGATTTACGAGGATGT-3'; antisense 5'-GCAGGAAAGCTGTGAACTCC-3'; IGFBP5 sense 5'-GAGCTGAAGGCTGAAGCAGT-3', antisense 5'-GAATCCTTTGCGGTCACAAT-3'.
The amplified PCR products were 248 bp Plasmolipin fragment, a 235 bp Calumenin fragment, and a 237 bp IGFBP5 fragment. Reactions containing 1 µg cDNA, SYBR green sequence detection reagent, and sense and antisense primers were assayed on an ABI 7700 sequence detection system (Applied Biosystems). The PCR conditions were one cycle at 50°C for 2 min, one cycle at 95°C for 10 min, and 40 cycles of 95°C for 15 s, 60°C for 1 min. The accumulation of PCR product was measured in real time as the increase in SYBR green fluorescence, and the data were analyzed using the Sequence Detector program v1.6.3 (Applied Biosystems). Standard curves relating initial template copy number to fluorescence and amplification cycle were generated using the amplified PCR product as a template and were used to calculate gene concentration (normalized to ß-actin values) in nM/µg cDNA.
Immunohistochemical Staining.
Deparaffinized sections were stained manually, using the avidin-biotin complex technique. In brief, 4 µm sections were collected on gelatin-coated slides, deparaffinized, and rehydrated via passage through graded alcohol solutions, according to standard histological techniques. Heat-induced epitope retrieval was obtained by heating slides for 20' in a citrate buffer (2 mM citric acid/10 mM Na citrate; pH 6) in a steam bath, followed by cooling down for 20' at room temperature. Slides were presaturated with 0.05% BSA in PBS for 10', quenched in 2% H2O2 for 10', and preincubated with 0.5% nonimmune goat serum (Sigma) for 10' at room temperature. The sections were then incubated with the primary antibodies. Antibodies used, dilutions, and incubation conditions were monoclonal Nm 23 antibody (Santa Cruz Biotechnology-H1; sc 465) 1:20 dilution, overnight incubation at 4°C; monoclonal Integrin 
6 (Santa Cruz Biotechnology BQ; sc-13542) 1:10 dilution, overnight incubation at 4°C; and polyclonal FGF-7 (Santa Cruz Biotechnology-C19; sc-1365), 1:40 dilution, 2-h incubation at room temperature. After three washes in BSA/PBS (5' each) localization of the primary antibody and color development were obtained by incubation with biotinylated antimouse (monoclonal antibodies) or antigoat (FGF-7) secondary antibodies (Vector), followed by incubation with streptavidin-coupled horseradish peroxidase, using an avidin-biotin detection kit, with AEC as the chromogen (Vector Labs, Burlingame, CA). Sections were lightly counterstained with hematoxylin (Sigma) and coverslipped. Positive controls were included with each run and included colon cancer for Nm23, normal skin for Integrin 6, and ductal carcinoma of breast for FGF-7. Negative controls were included with each run by omitting the primary antibody. Score was separately assessed for intensity and extent of positivity. Intensity was scored on a scale of 1 (lightest) to 3 (strongest). The extent of tumor cells positivity was scored as negative, no stain; focal, 
10% positive tumor cells; moderate, 
10% positive cells 50%; and diffuse, 50% positive tumor cells. Positive stain occurred in the cytoplasm.
Data Analysis and Neural Network Methods.
T tests were used to identify differences in mean gene expression levels between comparison groups. We constructed of the classifiers using a multilayer feedforward backpropagation neural network. A three layer feed forward artificial neural network was trained based on the expression value of differentially regulated genes, and the accuracy of the neural network was confirmed with 10-fold cross-validation. The 10-fold cross-validation involves constructing 10 different classifiers in 10 training runs. Each training run divides the available data into a training set, including 90% of the data and a validation set including 10% of the data. The classifier is constructed based on the training set and tested on the validation set. Each training run leaves out different patterns such as at the end of the entire process, each available patient will have been left out at least once. The performance reported represents an average of these 10 training runs. The process ensures that the resulting classifiers are always tested on data that has never been used during the training, thus testing its generalization abilities. We calculated the intersection of the genes selected by the dChip and Sam, and constructed a similar neural network classifier based on only those genes that were selected by both methods. To inspect visually the expression profiles of the genes, a hierarchical clustering was constructed for each group of genes using average linkage and standardized Euclidean distance.
The performance of each classifier was assessed in two ways. Firstly, each classifier was tested and assessed using 10-fold cross-validation, as described above. Secondly, we used the trained classifier on a previously unseen data set of 17 mesothelioma patients including 8 long-term survivors and 9 short-term survivors. These data were obtained from the Division of Thoracic Surgery of the Brigham and Women’s Hospital (16) . Kaplan Meier survival plots and log rank tests were used to assess differences between poor and good risk groups using MedCalc Software (Mariakerke, Belgium).
Onto-Express (OE) Analysis.
Using OE (17)
, we performed a functional analysis to identify the main biological processes and pathways involving these prognostic genes. OE is a tool designed to mine the available functional annotation data and find relevant biological processes. The input to OE is the list of GenBank accession numbers, Affymetrix probe IDs, or UniGene cluster IDs. OE constructs a functional profile for each of the Gene Ontology categories (18)
, including cellular component, biological process, and molecular function, as well as biochemical function and cellular role, as defined by Proteome. As biological processes can be regulated within a local chromosomal region (e.g., imprinting), an additional profile is constructed for the chromosome location. Statistical significance values are calculated for each category using either a hypergeometric or binomial distribution depending on the number of genes on the array used.
Results
Hierarchical Profile Clustering Yields Two Mesothelioma Subsets.
Using the U95 Affymetrix oligonucleotide arrays, we generated profiles for 21 MPMs, which had been defined as "short-term" and "long-term" groups based on their actual survival from operation as either < or >12 months (Table 1)
. Median survival time for the short-term group was 3.6 months (range, 2–10 months) compared with 58 months (range, 25–102 months) for the long-term group (P < 0.0001). The time from surgery to radiographic progression of disease was a median of 1.8 months (range, 1–8 months) and 14.5 months (range, 6–95 months) for the short- and long-term groups, respectively (P < 0.0001).
|
. These 95 genes were used to construct an artificial neural network with a 95–14-2 architecture. This network achieved an accuracy of 90.5% measured by 10-fold cross-validation. The sensitivity and specificity of this artificial neural network classifier were 90% and 90.9%, respectively.
|
Common Genes between dChip and SAM.
To define which important genes were common to dChip and SAM, an unpaired two group comparison was performed on the dChip normalized data. Interestingly, the intersection of the 95 genes selected by dChip with the 100 genes selected by SAM yielded only 27 common genes. This suggested that: (a) the data were rather noisy and the subset of good predictors might be <95 genes, and (b) at least one but possibly both methods picked up a number of genes that were less than ideal predictors. To investigate these hypotheses, we built a classifier based on the genes common between the two sets (i.e., 95 dChip genes and 100 SAM genes as opposed to the most significant 27 in each analysis). The clustering dendogram for these common 27 genes is seen in Fig. 1B
. This classifier yielded an accuracy of 95.2% on the validation dataset with 10-fold cross-validation. The confusion matrix is shown in Table 2
.
|
, there was excellent correlation between the array expression data and the real-time PCR, with the best correlation seen for IGFBP5 (0.99); those for plasmolippin and calumenin were 0.86. Immunohistochemical analysis of integrin -6, fibroblast growth factor, and IGFBP5 was performed for patients 4 and 39, each representing the two clusters to determine whether RNA expression was reflected appropriately by the corresponding proteins in tumors (Fig. 3)
. When one compares the expression of the proteins by immunohistochemistry with the clustergram (Fig. 1B)
, it is seen that patient 4 with a more favorable prognosis had elevated levels of integrin and FGF7, and both of these proteins were scored as having a stronger and more focal intensity with more cells staining in patient 4 than patient 39. Staining for NM23A revealed approximately equal intensity and quantity of cellular staining.
|
|
. These biological processes include cell proliferation (P = 0.016), lipid metabolism (P = 0.001), positive regulation of cell proliferation (P = 0.025), cell-cell signaling (P = 0.095), regulation of transcription from Pol II promoter (P = 0.071), pathogenesis (P = 0.010), cell cycle arrest (P = 0.003), negative regulation of cell proliferation (P = 0.034), inflammatory responses (P = 0.022), apoptosis induction by extracellular signals (P = 0.001), DNA damage response, activation of p53 (P = 0.001), and RNA processing (P = 0.006). The P was calculated using a binomial distribution (21)
.
|
. No observed propensity for segregation of epithelial mesotheliomas into the good-risk group or other histologies into the poor-risk group was observed. Time-to-progression curves for the gene set also mirrored the test set actual clinical data (Fig. 4, C and D)
.
|
. Prognostic heterogeneity, however, was detected within stages when survivals and progression were examined based on the gene classifier. When surgically staged I and II patients were grouped together, their survival was significantly different depending on which cluster the 27 gene classifier assigned them to. As seen in Fig. 5
C, the patients in either stage 1 or 2 who were assigned to the poor-risk cluster were found to have significantly shorter survival than those assigned to the other cluster (P = 0.0042), and the time to progression of the stage I and II patients (Fig. 5D)
was 1.8 months for those assigned to the poor-risk cluster and 15 months for those in the good-risk cluster (P = 0.0042). These differences were only suggestive for the stage III mesothelioma patients with regard to survival (Fig. 5, E and F)
. There was, however, a statistically longer time to progression for those stage III mesotheliomas who were clustered as good-risk (17 month median) compared with the poor-risk cluster (2.6 month median; P = 0.0397).
|
, and based on our 27 important genes, we applied hierarchical clustering on the patients from B&W dataset. The resulting dendogram revealed two groups (Fig. 6A)
from the B&W data, and the predicted survival of the B&W patients (Fig. 6C)
was similar to their actual (Fig. 6B)
survival (median survival, 5.5 months versus 5 months) for the poor-risk groups. Moreover, the survival differences between the gene-predicted 5 good-risk patients in the B&W and the 12 poor-risk patients approached significance (P = 0.0511).
|
Of the 27 genes found to be important in this investigation, 18 have been thoroughly classified in the literature, and few have been associated with MPM. Clusterin SP-40, acidic protein rich in leucine, and cysteine-rich protein were overexpressed in the vast majority of the MPMs, and, hypothetically, could be part of some pathway common to mesothelial carcinogenesis. Selenium binding protein was consistently overexpressed in good-risk patients, and was the only gene common to both the test set and validation set. SIVA or CD27-binding protein is part of an apoptotic pathway induced by CD27 antigen, a member of the tumor necrosis factor receptor superfamily. CD27 regulates the death and differentiation of T and B cells, and provides signals needed for the correct activation of specific T cells. Whether SIVA is important in MPMs for immune surveillance, and as such, could contribute to a longer survival, is unknown. Short chain dehydrogenase/reductase (retSDR1) reduces all-trans-retinal during bleached visual pigment regeneration (22) . In the absence of retSDR1, and under low concentrations of circulating retinol, the lack of production of vitamin A active metabolites could contribute to cancer development and progression. Retinoic acid decreases the synthesis of fibronectin and laminin, and migration of MPM implying that retinoids and their binding may decrease MPM local invasion and tumor progression (23) . BTG 2 is a member of a group of structurally related antiproliferative proteins, which mediate a common signal transduction growth arrest and differentiation pathway. This pathway interacts with N-methyl transferase, the chief enzyme for post-translational modifications of proteins by protein methylation (24) . BTG influences substrates of this methylation pathway including heterogeneous nuclear ribonucleoproteins (hnRNP), which was generally lower in expression in the good-risk group. hnRNP H1 influences pre-mRNA processing, and its role in cancer is unclear; however, hnRNP H1 has been identified as a protein that binds to the negative regulator spicing element of the Rous Sarcoma Virus (25 , 26) , and also binds a G-rich element downstream of the core SV40 late polyadenylation signal and stimulates 3' processing (27) . These data suggest that hnRNP H1 could modulate late viral element activity in two known carcinogenic viruses, of which one (SV40) has now been implicated in the pathogenesis of MPM (4) .
Calumenin, a calcium binding protein localized in the endoplasmic reticulum involved in protein folding and sorting, was uniformly underexpressed in the good-risk group. Calumenin has been found to bind serum amyloid P (28) and is hypothesized to participate in the immunological defense system. Calumenin has not specifically been associated with MPM before, but serum amyloid A and P have been noted to be connected with MPM (29, 30, 31) . Nm23a, variably expressed in the good-risk group, was uniformly overexpressed in the poor-risk group. This transcription factor associated with highly metastatic cells (32) plays a role in myc expression. It has been shown that myc can up-regulate nm23-H1 and nm23-H2 expression (33) , and data are variable regarding the prognostic significance of nm23 in human neoplasms (34, 35, 36, 37, 38, 39) .
Our data analysis is novel compared with other reports in the literature. Differences in array platforms, data analysis packages, or modifications in the filtering of the genes may result in differences in classifier genes among reports examining the same disease. We have demonstrated this phenomenon in our own work by comparing the top significant genes of interest from dChip and SAM, and found that there was not uniform agreement between the two programs on these genes, and a 10-fold cross-validation testing yielded differing results for the two programs. We feel that by selecting only those genes that were common to the two analyses, the resulting classifiers are more robust. Moreover, neither the top 27 genes from dChip or SAM could segregate survival differences among the B&W MPMs; however, the common set was 76% accurate in this independent set of specimens. It is likely that the patient population from the National Cancer Institute/Karmanos experience is different from that of the Boston experience. Our group included patients who were in International Mesothelioma Interest Group stage III, usually on the basis of metastases to the mediastinal lymph nodes at the time of surgery. The Boston MPMs were all stage I or II by the B&W classification, and in this classification, stage II patients may have intraparenchymal lymph node involvement, which would be stage III in the International Mesothelioma Interest Group classification. Moreover, the Boston MPM expression array data would be based on tumors that at the time of surgery did not have disease in the mediastinal lymph nodes (B&W classification stage III). Criteria selection for gene expression differed from the validation set in that the B&W group stipulated minimal expression of 500 and at least 2-fold differences. Given the huge differences in clinical presentation of these subsets, it is actually gratifying that a 76% validation of our gene set could be achieved in the Boston patients.
We have described a set of 27 genes that segregates good-risk and poor-risk surgically treated MPM patients with validation in an independent set of MPM patients. The classifier-predicted good-risk group had significantly longer survival and significantly longer time to progression than the poor-risk group. Moreover, these classifications were independent of the "conventional wisdom" that epithelial mesotheliomas have a better prognosis than biphasic and sarcomatoid histotypes. The data also reflect staging heterogeneity in relation to stage-related prognosis. Good-risk MPMs who were pathological stage I and II had strikingly different survival and time to progression compared with stage I and II patients assigned to the poor-risk cohort. These data reflect the clinical scenario that certain patients with favorable prognostic demographics clinically (i.e., epithelial histology and early stage) nevertheless may recur quickly after operation despite maximal cytoreduction. Such data, after validation in larger numbers of patients, could be important to guide clinical decisions regarding appropriate therapy before committing the MPM patient to surgery.
ACKNOWLEDGMENTS
The superb technical assistance of Valerie Murphy, Julia Sluchak-Carlsen, and Apurvi Desai was crucial for the successful completion of these series of studies.
FOOTNOTES
Grant support: Early Detection Research Network, National Cancer Institute/NIH U01 CA 084986, by Merit Review Funding from the Department of Veterans Affairs, and by a grant from the Mesothelioma Applied Research Foundation (R. B.).
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.
Requests for reprints: Harvey I. Pass, Harper University Hospital, 3990 John R, Suite 2102, Detroit, MI 48201. Phone: (313) 745-8746; Fax: (313) 993-0572; E-mail: hpass{at}dmc.org
Received 4/21/03; revised 10/12/03; accepted 10/14/03.
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  D. Nguyen, S.-J. Lee, E. Libby, and C. Verschraegen Rate of Thromboembolic Events in Mesothelioma Ann. Thorac. Surg., March 1, 2008; 85(3): 1032 - 1038. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. N. Hayes, S. Monti, G. Parmigiani, C. B. Gilks, K. Naoki, A. Bhattacharjee, M. A. Socinski, C. Perou, and M. Meyerson Gene Expression Profiling Reveals Reproducible Human Lung Adenocarcinoma Subtypes in Multiple Independent Patient Cohorts J. Clin. Oncol., November 1, 2006; 24(31): 5079 - 5090. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R Hassan, R Alexander, K Antman, P Boffetta, A Churg, D Coit, P Hausner, R Kennedy, H Kindler, M Metintas, et al. Current treatment options and biology of peritoneal mesothelioma: meeting summary of the first NIH peritoneal mesothelioma conference Ann. Onc., November 1, 2006; 17(11): 1615 - 1619. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Lopez-Rios, S. Chuai, R. Flores, S. Shimizu, T. Ohno, K. Wakahara, P. B. Illei, S. Hussain, L. Krug, M. F. Zakowski, et al. Global Gene Expression Profiling of Pleural Mesotheliomas: Overexpression of Aurora Kinases and P16/CDKN2A Deletion as Prognostic Factors and Critical Evaluation of Microarray-Based Prognostic Prediction. Cancer Res., March 15, 2006; 66(6): 2970 - 2979. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. I. Pass, D. Lott, F. Lonardo, M. Harbut, Z. Liu, N. Tang, M. Carbone, C. Webb, and A. Wali Asbestos exposure, pleural mesothelioma, and serum osteopontin levels. N. Engl. J. Med., October 13, 2005; 353(15): 1564 - 1573. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. H. Ahmed, R. Lutjens, L. D. van der Stap, D. Lekic, V. Romano-Spica, M. Morales, G. F. Koob, V. Repunte-Canonigo, and P. P. Sanna Gene expression evidence for remodeling of lateral hypothalamic circuitry in cocaine addiction PNAS, August 9, 2005; 102(32): 11533 - 11538. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. J. Gordon, G. N. Rockwell, P. A. Godfrey, R. V. Jensen, J. N. Glickman, B. Y. Yeap, W. G. Richards, D. J. Sugarbaker, and R. Bueno Validation of Genomics-Based Prognostic Tests in Malignant Pleural Mesothelioma Clin. Cancer Res., June 15, 2005; 11(12): 4406 - 4414. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. J. Gordon, G. N. Rockwell, R. V. Jensen, J. G. Rheinwald, J. N. Glickman, J. P. Aronson, B. J. Pottorf, M. D. Nitz, W. G. Richards, D. J. Sugarbaker, et al. Identification of Novel Candidate Oncogenes and Tumor Suppressors in Malignant Pleural Mesothelioma Using Large-Scale Transcriptional Profiling Am. J. Pathol., June 1, 2005; 166(6): 1827 - 1840. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Cesario, A. Catassi, L. Festi, A. Imperatori, A. Pericelli, D. Galetta, S. Margaritora, V. Porziella, V. Cardaci, P. Granone, et al. Farnesyltransferase Inhibitors and Human Malignant Pleural Mesothelioma: A First-Step Comparative Translational Study Clin. Cancer Res., March 1, 2005; 11(5): 2026 - 2037. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. Kettunen, A.G. Nicholson, B. Nagy, H. Wikman, J.K. Seppanen, T. Stjernvall, T. Ollikainen, V. Kinnula, S. Nordling, J. Hollmen, et al. L1CAM, INP10, P-cadherin, tPA and ITGB4 over-expression in malignant pleural mesotheliomas revealed by combined use of cDNA and tissue microarray Carcinogenesis, January 1, 2005; 26(1): 17 - 25. [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 |
