Identification of Novel Candidate Oncogenes and Tumor Suppressors in Malignant Pleural Mesothelioma Using Large-Scale Transcriptional Profiling

Abstract

Malignant pleural mesothelioma (MPM) is a highly lethal, poorly understood neoplasm that is typically associated with asbestos exposure. We performed transcriptional profiling using high-density oligonucleotide microarrays containing ∼22,000 genes to elucidate potential molecular and pathobiological pathways in MPM using discarded human MPM tumor specimens (n = 40), normal lung specimens (n = 4), normal pleura specimens (n = 5), and MPM and SV40-immortalized mesothelial cell lines (n = 5). In global expression analysis using unsupervised clustering techniques, we found two potential subclasses of mesothelioma that correlated loosely with tumor histology. We also identified sets of genes with expression levels that distinguish between multiple tumor subclasses, normal and tumor tissues, and tumors with different morphologies. Microarray gene expression data were confirmed using quantitative reverse transcriptase-polymerase chain reaction and protein analysis for three novel candidate oncogenes (NME2, CRI1, and PDGFC) and one candidate tumor suppressor (GSN). Finally, we used bioinformatics tools (ie, software) to create and explore complex physiological pathways. Combined, all of these data may advance our understanding of mesothelioma tumorigenesis, pathobiology, or both.

Malignant pleural mesothelioma (MPM) is a highly lethal malignancy primarily resulting from previous asbestos exposure.1,2 Approximately 3000 patients are diagnosed with MPM in the US annually and the incidence worldwide is projected to rise substantially in the next 2 decades.3–5 Treatment options are few and most patients die within 2 years of diagnosis. The pathological diagnosis of MPM is often complex and requires immunohistochemical staining with a panel of antibodies. There are three distinct histological subtypes of MPM based on the microscopic appearance of the major malignant elements: epithelial, sarcomatoid, mixed. The majority of MPMs are epithelial (50%). Patients with this subtype are generally thought to experience a somewhat longer survival than do patients with the other subtypes.6,7 However, clinicians have long noticed a certain degree of variability in patient survival and response to therapy that is not explained solely by the stage or histological appearance and may therefore be due to more subtle biological variability among tumors.

The long latency period between asbestos exposure and tumor development implies that multiple, and likely diverse, genetic changes are required for the malignant transformation of mesothelial cells. Several oncogenes and tumor suppressors have been hypothesized to play a role in MPM carcinogenesis.8 These include the tumor suppressors RASSF1A, NF2, and p16 as well as the growth factors/proto-oncogenes/signaling molecules c-fos, c-jun, EGF, VEGF, MetAP2, c-Met, c-myc, and fra-1. Of note, MPM does not appear to involve aberrant expression of many well-studied growth control genes such as H-ras, K-ras, p53, and Rb9–11 although SV40 T antigen has been proposed to inactivate p53 function in some MPM tumors.12

Large-scale transcriptional studies using microarrays have been reported in multiple cancers. To date, studies of this sort reported for MPM have used relatively small numbers of tissues and/or cell lines.13–16 This is likely the consequence of the relatively low incidence of MPM and by extension, the difficulty of acquiring large numbers of MPM pathological specimens. One exception is a recent study by Pass and colleagues.17 These investigators profiled 21 MPM patients and created a 27-gene neural network classifier that could predict clinical outcome with moderate accuracy. Previously, we profiled 31 MPM tumors to develop and validate a six-gene diagnostic test for MPM13 and a four-gene prognostic test for patients undergoing cytoreductive surgery.18 In the current study, we profile 40 MPM tumors, as well as normal lung and pleural tissues (n = 9), and tumor-derived and nontumorigenic cell lines (n = 5) using microarrays containing ∼22,000 genes to validate a previously described MPM diagnostic test13 and better characterize the molecular pathways involved in the pathogenesis of MPM.

Materials and Methods

Tissues and Cell Lines Profiled Using Microarrays

Discarded MPM surgical specimens (n = 40), normal pleura specimens (n = 5), and normal lung specimens (n = 4) were freshly collected (and snap-frozen) from patients who underwent surgery at Boston’s Brigham and Women’s Hospital (BWH) between October 1998 and August 2000. All of these patients underwent extrapleural pneumonectomy with heated intrapleural cis-platin chemotherapy delivered after the specimens were removed. All normal specimens were obtained from patients who were never diagnosed with MPM. Two human MPM cell lines (MS589 and MS428) were kindly provided by Jonathan A. Fletcher, M.D., Department of Pathology, BWH. The JMN1B MPM cell line19,20 has been described previously. The SV40-immortalized, nontumorigenic mesothelial cell line (Met-5A)21 and the MPM cell line MSTO-211H22 were purchased from the American Type Culture Collection (Rockville, MD). Normal tissues were obtained from additional consented patients undergoing treatment for diseases other than MPM. All MPM samples used in these studies contained relatively pure tumor (greater than 50% tumor cells per high-power field examined in a section adjacent to the tissue used). The microscopic slides from the patients’ resection specimens were reviewed by one of the authors (J.G.), and the diagnosis and histological subclassification of MPM confirmed in all cases. Linked clinical and pathological data were obtained for all patients who contributed tumor specimens. Specimens and data were rendered anonymous to protect patient confidentiality. Studies using human tissues were approved by and conducted in accordance with the policies of the Institutional Review Board at BWH.

Microarray Experiments

For preparation of RNA for microarray analysis, JMN1B, MS589, MS428, MSTO-211H, and Met5A cells were grown in RPMI 1640 medium (Invitrogen Life Technologies, Carlsbad, CA) supplemented with 10% calf serum. Sample preparation and hybridization to Affymetrix human U133A oligonucleotide probe arrays was performed essentially as described in the Affymetrix Expression Analysis Technical Manual using total RNA (7 μg) that was prepared from whole tumor blocks and cells in exponential growth using TRIzol reagent (Invitrogen Life Technologies). Hybridization experiments were scanned visually for artifacts and gene expression levels (ie, Affymetrix signal) for each microarray were generated and scaled to a target intensity of 100 using Affymetrix Microarray Suite v.5.0 (Sants Clara, CA).

Gene Selection and Clustering Algorithms

Before unsupervised cluster analysis and principal components analysis,23 the expression level of each of the 22,000 genes on the microarray was stated relative to the median value of that gene in all normal samples, tumor samples, and cell lines analyzed and then log₂-transformed. The 1405 most variable transcripts (SD, more than threefold) were chosen to perform cluster analysis using CLUSTER and TREEVIEW software24 after median centering and normalization of the data. Other clustering algorithms including partitioning around medoids25 and principal components analysis used the S-PLUS statistical package.26 Partitioning around medoids is similar to k-means clustering, but uses medoids instead of centroids and minimizes a sum of dissimilarities instead of a sum of squared Euclidean distances.

Using S-PLUS, we identified potential tumor molecular markers by determining which of the genes on the microarray was expressed at significantly (P < 0.05) different average expression levels between all 40 tumor samples and all (lung and pleura) normal samples using a two-sided Student’s (parametric) t-test for pair-wise comparisons of average gene expression levels. Genes discriminatory between subclasses discovered during cluster analysis and between epithelial and mixed MPM histological subtypes were similarly identified. The SAM algorithm27 was used to estimate the false discovery rate.

Pathway Analysis of Candidate MPM Oncogenes

Pathway analysis was performed using Ingenuity pathways analysis (www.ingenuity.com), a web-delivered application that enables biologists to discover, visualize, and explore potentially relevant pathobiological networks. Genes found to be statistically significantly up-regulated in MPM so identified were used to create a total of 12 pathway networks. Seven of these networks with at least one gene in common were subsequently merged for display. The combined network is displayed graphically as nodes (genes/gene products) and edges (the biological relationships between the nodes). The length of an edge reflects the evidence supporting that node-to-node relationship. Edges supported by more articles from the literature are shorter. For simplicity, genes are referred to by their LocusLink symbol.

Quantitative Reverse Transcriptase (RT)-Polymerase Chain Reaction (PCR)

For preparation of RNA for quantitative RT-PCR analysis, MS589, MS428, MS924, JMN1B, and the normal primary human mesothelial cell line HM328,29 were grown in M199/MCDB105 [1:1 (v/v) medium + 15% newborn calf serum + 0.4 μg/ml hydrocortisone] with or without epidermal growth factor (EGF) (10 ng/ml). This medium is permissive for growth both of normal and neoplastic human mesothelial cells.28,29 Note that HM3 cells and MS924 cells were used only for quantitative RT-PCR analysis and were not analyzed using microarrays. MS924 was kindly provided by Jonathan A. Fletcher, M.D., Department of Pathology, BWH. All cells were maintained in a humidified 5% CO₂ incubator at 37°C.

Quantitative RT-PCR was performed using a SYBR-Green fluorometric-based detection system and reagents purchased from Applied Biosystems (Foster City, CA; www.appliedbiosystems.com, see technical bulletin no. 4310251 for details and protocol). Using TRIzol reagent, we isolated total RNA (2 μg) from frozen tumor specimens, preconfluent, exponentially growing cultures, or mitogen-deprived growth-arrested cultures. The isolated RNA was then reverse-transcribed into complementary DNA (cDNA) with the use of Taq-Man reverse transcription reagents (Applied Biosystems), and quantified using all controls recommended by the manufacturer. Both culture conditions (ie, ±EGF) permit the growth of the mesothelioma lines but the −EGF condition induces a growth-arrested state for HM3 cells, as described previously.28 All RT-PCR primers were used at a final concentration of 800 nmol/L in the reaction mixture. (For simplicity, genes are referred to by their LocusLink symbol.) Primer sequences have been previously published for GAPD,30 CALB2 (calretinin),13 CLDN7 (claudin-7),13 ANXA8 (annexin A8, also known as vascular anticoagulant-β, VAC-β),13 TACSTD1 (tumor-associated calcium signal transducer 1),13 CD200 (also known as MRC OX-2),13 TITF1 (thyroid transcription factor 1).13 Other primers (synthesized by Invitrogen Life Technologies) were as follows (forward and reverse): NME2 (5′-GACCTGAAAGACCGACCATT-3′ and 5′-AATCTGCTGGATTGGTCT-CC-3′), PDGFC (5′-TGTCATGCCACAATTCACAG-3′ and 5′-TGCCATCTCTCTGGTTCAAG-3′), CRI1 (5′-TGCCGGCTACAGAGTATCAG-3′ and 5′-ACTGATCAAACGG-GGT-CTTC-3′), GSN (5′-ACGGCTGAAGGACAAGAA-GA-3′ and 5′-TTCCAACCCAGACAAAGACC-3′). PCR amplification of cDNA was performed using a Stratagene MX 3000P device with appropriate controls. The comparative CT equation (Applied Biosystems) describes the exponential nature of PCR-based amplification and was used exactly as stated to measure the relative expression levels of NME2, PDGFC, and CRI1 relative to those of GAPDH.

Western Blot Analysis

To prepare the cell lysates for Western blot analysis, 98-483, H-2052, JMN1B, MS589, MS428, MSTO-211H, and Met5A cells were grown in RPMI 1640 medium (Invitrogen Life Technologies) supplemented with 10% calf serum. Note that 98-483 and H-2052 cells were used only in Western blot analysis and were not analyzed using microarrays or quantitative RT-PCR. The 98-483 MPM cell line was kindly provided by Jonathan A. Fletcher, M.D. The H-2052 MPM cell line was purchased from American Type Culture Collection. Cell lines subjected to protein analysis were homogenized and lysed at 4°C in a buffer containing 20 mmol/L Tris-HCl (pH = 8), 137 mmol/L NaCl, 2 mmol/L ethylenediamine tetraacetic acid, 10% glycerol, 1% Nonidet P-40, 200 mmol/L sodium orthovanadate, 1 mmol/L dithiothreitol, 1 mmol/L phenyl-methyl sulfonyl fluoride, 2 mg/ml leupeptin, and 5 mg/ml aprotinin. The normal pleura protein preparation was similarly made by pooling the five normal pleura specimens analyzed using microarrays. Protein quantity was estimated using the Bio-Rad protein assay (Bio-Rad, Richmond, CA). For Western blot analysis, aliquots of cell lysates were combined with a protein solubilization stock buffer consisting of 250 mmol/L Tris-HCl (pH 7.4), 2% sodium dodecyl sulfate, 30% glycerol, and 0.01% bromophenol blue (pH 6.8) to a final concentration of 2 μg of protein/μl and denatured by boiling for 5 minutes. Samples containing ∼50 μg of protein per lane were separated using sodium dodecyl sulfate-polyacrylamide gel electrophoresis (12% gel) and transferred onto Hybond ECL nitrocellulose membranes (Amersham, Arlington Heights, IL) according to standard procedures. All subsequent incubations were performed at room temperature unless otherwise stated. Nitrocellulose membranes were incubated with agitation overnight at 4°C in a blocking buffer consisting of 5% nonfat dry milk in phosphate-buffered saline (PBS) (136 mmol/L NaCl, 2.7 mmol/L KCl, 10 mmol/L Na₂PO₄, and 1.76 mmol/L KH₂PO₄, pH 7.2) and 0.05% Tween-20. Antibodies specific to human NME2 (Santa Cruz Biotechnology, Santa Cruz, CA), GSN (Research Diagnostics, Flanders, NJ), CRI1 (AKA EID-1, U.S. Biological, Swampscott, MA), and β-actin (Sigma, St. Louis, MO) were diluted 1:100 in PBS (for NME2), used at 2 μg/ml in PBS (for CRI1), diluted 1:3000 in PBS (for GSN), and diluted 1:500 in PBS (for β-actin) for 1 hour of incubation at room temperature. Horseradish peroxidase-conjugated secondary antibodies (DAKO,Carpinteria, CA) were diluted 1:10,000 in PBS. Membranes were washed after the primary and secondary antibodies in three changes of 0.05% Tween 20 in PBS for 5 minutes each. Chemiluminescent detection was performed using ECL reagents (Amersham) and the size of protein bands estimated by comparison to Full Range Rainbow molecular marker standard (Amersham).

Immunohistochemistry

Indirect immunoperoxidase analysis of NME2 and CRI1 protein was performed on a tissue array consisting of 66 MPM tumors, 2 lung adenocarcinoma tumors, and 4 normal pleura tissues. This array was constructed from the same 40 MPM tumors analyzed using microarrays and an additional 26 formalin-fixed paraffin-embedded MPM tumor tissues obtained from pathological specimens of patients undergoing extrapleural pneumonectomy at BWH. Because the tumors were anonymized after deidentified clinical information was provided, we were not able to directly compare mRNA data from microarrays with protein data from tissue microarrays on a patient-by-patient basis. The microscopic slides from the patients’ resection specimens were reviewed by one of the authors (J.G.), and the diagnosis and histological subclassification of MPM confirmed in all cases. Two to four representative 1-mm cores were used for each tumor. Construction of the array was performed by the Dana Farber/Harvard Cancer Center Tissue Microarray Core Facility, and was approved by the BWH Institutional Review Board. For immunostaining, 6-μm sections were cleared with xylenes and rehydrated through a graded series of alcohols. NME2 and CRI1 detection was performed using an accelerated avidin/biotin peroxidase procedure per the manufacturer’s recommended protocol (Vectastain ABC Elite kit; Vector Laboratories, Burlingame, CA) and primary antibody concentrations and incubation times identical to those used in Western blot analysis. Endogenous peroxidase activity was quenched by incubation in a 0.3% solution of H₂O₂ in methanol for 5 minutes. Blocking of nonspecific activity was accomplished using diluted serum of the secondary antibody species (Vector Laboratories). Primary antibodies were replaced with PBS as a negative control. Sections were developed in diaminobenzidine (DAB kit, Vector Laboratories), lightly counterstained with Gill’s hematoxylin, dehydrated through a graded series of alcohols, and permanently mounted. Images were digitally captured using an Olympus T041 microscope. To quantify protein expression levels, slides were read by a pathologist (J.G.) who was blinded to the gene expression data. Individual tissues were scored as follows: no expression, <25% positive cells, 25 to 50% positive cells, 50 to 75% positive cells, >75% positive cells. The intensity of staining was subjectively scored either as strong or weak. The statistical significance of observed differences was assessed using χ² analysis with P < 0.05.

Results and Discussion

Unsupervised Cluster Analysis

We have previously profiled 31 MPM tumors using microarrays containing ∼12,000 genes and used these data to design a diagnostic test to distinguish MPM from adenocarcinoma of the lung.13 In the current study, we examined global gene expression profiles of 40 MPM tumors, 9 normal tissues (lung and pleura), and five cell lines using microarrays containing ∼22,000 genes. Unlike our initial effort,13 both of these normal tissue types were included as controls because MPM arises from mesothelial cells of the pleura and frequently envelopes lung tissues. Of the five cell lines, four are MPM-derived and 1 (Met-5A) is a nontumorigenic immortalized mesothelial cell line. For simplicity, genes are referred to by their LocusLink symbol. We did not use samples previously profiled13 in cluster analysis in the current study because it would substantially limit the number of genes available for use due to inherent platform differences.

Of the 22,000 genes represented on the microarray, we chose 1405 genes with the most variable expression levels across all samples to perform two-dimensional unsupervised cluster analysis. The dendrogram specifying the arrangement of samples is shown in Figure 1 with linked clinical data available in Supplemental Table 1 at http://ajp.amjpathol.org. Four distinct subclasses can be seen, two of which consist entirely of MPM samples. These were arbitrarily designated C1 and C2 (n = 17 and n = 14, respectively; Figure 1). The other two subclasses consist primarily of normal tissues and cell lines (normal and cell, respectively; Figure 1). C1 and C2 tumors consisted almost entirely of epithelial (88%) and mixed (78%) histological subtypes, respectively, demonstrating that varied morphological appearance may have a basis in distinct gene expression patterns. It has been reported using large sample cohorts that patients with epithelial histology generally experience a significantly longer disease-free survival than patients with nonepithelial histology.5 Nevertheless, we did not find any statistically significant survival differences between C1 and C2 subclasses, possibly because of the relatively small number of specimens examined.

Figure 1.

Defining MPM subclasses using unsupervised hierarchical clustering. Two-dimensional unsupervised hierarchical clustering was performed using the 1405 most variable transcripts across MPM tumor samples (n = 40, black letters), normal lung samples (n = 4, green letters), normal pleura samples (n = 5, red letters), and cell lines (n = 5, blue letters). The dendrogram specifying the arrangement of samples is shown across the top with major nontumor tissue types (ie, normal or cell line) and both major tumor subclasses (ie, C1 and C2) identified using yellow bars found immediately below. Individual gene expression levels (in rows) for each sample (in columns) were normalized and expressed relative to the median value for each gene in all samples according to the scale at the bottom left. Bars to the right of the figure refer to regions of differentially expressed genes shown in greater detail in Figures 2 and 3 (for genes overexpressed in subclasses C1 and C2, respectively) and Supplemental Figures 2 to 6 at http://ajp.amjpathol.org (for numbered bars 1 to 5, respectively).

Unexpectedly, a total of nine tumors were not contained within either C1 or C2, and seven of these samples clustered opposite the normal samples. The cause of this observation was not immediately clear, because all MPM samples contained relatively pure tumor content and the seven samples clustering nearest the normal tissues were >90% tumor. Nor did we observe any noticeable or statistically significant trends in clinical data for these nine samples. Although these nine samples were not tightly clustered on a single branch of the dendrogram, we initially hypothesized that these samples represented a third tumor subclass distinct from C1/C2 tumors.

To test this hypothesis, we first used the expression data for the 1405 highly variable genes from above in other unsupervised hierarchical clustering algorithms and found that the general arrangement of samples was maintained. When we excluded normal tissues and cell lines and repeated the analysis, we found that the resulting dendrogram contained three major subbranches: one contained all C1 tumors, another contained all C2 tumors, and the third contained four of the nine non-C1/C2 samples (tumors 07, 10, 36, and 20), with the remaining five distributed between the C1 subbranch (n = 3) and C2 subbranch (n = 2).

Next we clustered all of the samples using a nonhierarchical algorithm (ie, partitioning around medoids) and the same 1405 genes with six predicted clusters: normal lung, normal pleura, cell lines, C1, C2, and a putative C3 tumor cluster consisting of the nine non-C1/C2 samples from Figure 1. We found that the normal pleura samples, normal lung samples, and cell lines were all assigned to unique clusters in this analysis. Of the three remaining clusters, one contained only C1 tumors, another contained all C2 tumors and four of the nine non-C1/C2 tumors, and the third consisted of five C1 tumors and five of the nine non-C1/C2 tumors

Finally, a three-dimensional graphical display of the results of principal components analysis demonstrates that C1/C2 tumors can be separated easily from normal tissues and the cell lines along the first principal component, while the same cannot be said of the other nine samples (Supplemental Figure 1 at http://ajp.amjpathol.org). These combined results suggest that the nine non-C1/C2 samples do not represent a distinct third subclass of MPM. They do not cluster tightly with the majority of MPM tumors in hierarchical cluster analysis likely because they have distinct gene expression patterns at least with respect to a subset of the 1405 genes used in these analyses. The existence of two distinct tumor classes and the nine that do not always cluster within C1 or C2 is consistent with the great biological variability observed in vivo as manifested in patient clinical behavior and the diverse cytogenetic features of MPM tumors. This may reflect multiple nonoverlapping carcinogenic pathways in MPM especially considering its extraordinary long latency period.

Genes Distinctive of MPM Subclasses

We observed several groups of genes with distinct expression patterns among cell lines and normal and tumor tissues (Figure 1). Genes highly expressed in subclasses C1 and C2 are listed individually and shown in greater detail in Figures 2 and 3, respectively. Other distinctive gene clusters (delineated by bars 1 to 5 in Figure 1) are listed individually and shown in greater detail in Supplemental Figures 2 through 6 at http://ajp.amjpathol.org, respectively. Genes expressed at relatively low levels in normal tissues and most of the nine non-C1 or -C2 tumors are designated by bar 2 in Figure 1 (for details see Supplemental Figure 3 at http://ajp.amjpathol.org). This group of genes includes multiple histone genes, SON, which encodes for a DNA-binding protein with sequence homology to MYC and MOS oncoproteins, and CALU, a gene that has been found to be down-regulated in metastatic head and neck squamous cell carcinoma relative to the primary tumor,31 and associated with prognosis in MPM.17 This provides evidence for the suggestion above that the non-C1 and -C2 tumors are inherently different from the other tumors at least with respect to the expression levels of certain genes. Furthermore, these gene expression patterns are likely responsible for the grouping of non-C1 or -C2 tumors nearer to normal tissues from unsupervised hierarchical cluster analysis. However, the high tumor cell content of these non-C1 and -C2 MPM samples is evident by examining the gene cluster designated by bar 4 in Figure 1 (for details see Supplemental Figure 5 at http://ajp.amjpathol.org) in which genes, such as surfactants and hemoglobin, that are typically expressed at high levels in normal lung and pleura are expressed at much lower levels in the relevant tumor tissues.

Figure 2.

Genes with elevated expression levels in C1 MPM tumors. Genes whose expression levels are elevated in C1 MPM tumors (from Figure 1) are shown in greater detail and annotated with gene symbol (ie, LocusLink identifier) and gene name (ie, Unigene title). Individual sample identifiers have been removed for the sake of clarity. Relative gene expression levels are given by the scale at the bottom left.

Figure 3.

Genes with elevated expression levels in C2 MPM tumors. Genes whose expression levels are elevated in C2 MPM tumors (from Figure 1) are shown in greater detail and annotated with gene symbol (ie, LocusLink identifier) and gene name (ie, Unigene title). Individual sample identifiers have been removed for the sake of clarity. Relative gene expression levels are given by the scale at the bottom left.

Genes overexpressed in C1 (Figure 2) include those with known or inferred function such as proto-oncogenes (eg, MAF, MET), tumor suppressors (eg, GAS1, WT1), and multiple genes associated with cytoskeletal/support (eg, keratins, cadherins, and other proteoglycans), signaling (eg, KDR), apoptosis (eg, MST4), and proliferation (eg, FGF18). A large number of C1-overexpressed genes have not been previously described in MPM and/or were initially characterized in other tissue types. For example, NMU codes for a neuropeptide with potent effects on smooth muscle contraction. Its role, if any, in MPM (or cancer in general) is unknown. Several C1-overexpressed genes are of particular interest. CD47 is a gene that has been implicated in diverse cellular functions such as tumor cell chemotaxis,32 integrin function,33 and apoptosis.34 The c-Met proto-oncogene (MET) has widely documented roles in tumorigenesis. Moreover, high levels of MET mRNA and protein have been previously found in MPM tumors and cell lines35,36 and have been proposed to promote angiogenesis37 and cell motility.38 Recent evidence suggests that the MET pathway can be activated by SV40,39 an interesting observation considering the unclear role of this virus in MPM etiology. Finally, we observed high mRNA levels of the PLA2G2A gene in C1 tumors. This gene is highly expressed in many other neoplasms where it has been proposed to promote carcinogenesis40 or mediate profound anti-tumor effects41 in a tissue/disease-specific manner. These latter observations are consistent with our previous studies showing PLA2G2A to be preferentially expressed in MPM tumors from patients with relatively good prognosis.18

Genes highly expressed in subclass C2 are listed in Figure 3 and consist mainly of extracellular matrix and structural proteins such as collagen, actin, biglycan, and fibronectin. Several of these genes (eg, COMP) are typically expressed in other tissue types and in less differentiated cells. This combined with an overall anabolic expression profile is consistent with the morphological appearance and clinical behavior of mixed histological subtype (ie, C2 subclass) MPM tumors. We also discovered at least two potentially interesting (nonmatrix-related) genes overexpressed in C2: RAB23 and BMP4. RAB proteins are small GTPases that are members of the RAS superfamily and RAB23 encodes a protein that is a negative regulator of the Sonic hedgehog signaling pathway.42 This pathway is important in mammalian embryogenesis and its activation in adults can result in various tumors including a subset of non-small cell lung cancers.43 BMP4 is another Sonic hedgehog pathway regulator that is also highly expressed in C2 tumors, and together these genes may indicate a potential role for this pathway in MPM. A complete listing of the 113 genes with average expression levels that were significantly different between C1 and C2 tumors is found in Supplemental Table 2 at http://ajp.amjpathol.org. Using SAM, we estimated a false-discovery rate of 15% for this gene list.

Identification of MPM Histological Molecular Markers

We identified those genes with expression patterns that were reflective of MPM histological subtype and found a total 54 genes with average expression levels significantly different between MPM histological classes with a false-discovery rate of ∼15% (Supplemental Table 3 at http://ajp.amjpathol.org). Twenty-five and twenty-nine of these genes were expressed at relatively higher levels in epithelial and mixed MPM tumors, respectively. In fact, many of the genes that we found to be distinctive of C1 and C2 subclasses from hierarchical clustering were also differentially expressed in a statistically significant manner between epithelial and nonepithelial tumors. This finding provides additional evidence that the morphological appearance of MPM tumor cells may reflect distinct gene expression patterns.

Identification of MPM Tumor Molecular Markers

In a related analysis, we identified 328 genes that were differentially expressed at significantly higher levels in MPM tumors and 311 genes that were differentially expressed at significantly higher levels in all normal samples with an estimated false-discovery rate of 10% for both gene lists (Supplemental Table 4 at http://ajp.amjpathol.org). Among those genes significantly underexpressed in MPM tumor tissues was gelsolin (GSN, P = 7.4 × 10⁻⁸), a gene known to be down-regulated in multiple neoplasms and capable of suppressing tumorigenicity by inhibiting protein kinase C activation.44 In this study, elevated GSN levels in the lung cancer cell line PC10 resulted in decreased proliferation in vitro and complete inhibition of tumorigenicity in nude mice.

By visually scanning the list of genes up-regulated in MPM, we noticed several genes that had known physiological roles suggestive of a role in carcinogenesis including platelet-derived growth factor C (PDGFC, P = 0.018) CREBBP/EP300 inhibitory protein 1 (CRI1, P = 2.7 × 10⁻⁴), and nucleoside-diphosphate kinase 2 (NME2, P = 1.5 × 10⁻⁹). These genes have well documented roles in growth and/or differentiation in multiple cancers, but have not previously been implicated in MPM pathogenesis.

Platelet-derived growth factor (PDGF) family members have been previously described in MPM,8 and PDGF-C has been recently described as a novel transforming growth factor45 that participates in an autocrine signaling loop.46 Blocking of the functional receptor reduces tumor formation in nude mice.46 The PDGF receptor is a tyrosine kinase that functions in the Akt and MAPK pathways, but is not likely involved with Erk phosphorylation.46 PDGF-C has a binding pattern similar to PDGF (A and B) and, of particular note for MPM, is a more potent mitogen than either in cells of mesenchymal origin.47 PDGF-C is highly expressed specifically in mesenchymal precursors and the myoblasts of smooth and skeletal muscle.48

The NME2 gene is a transcriptional regulator with diverse functions that binds and cleaves DNA via covalent bond formation and catalyzes phosphoryl transfer.49 It is a transcriptional regulator of specific genes (eg, activation of c-MYC) in a cell- and tissue-specific manner, is part of the N-MYC downstream pathway, and experimental evidence suggests that high levels of NME2 are mutagenic.50 NME2 protein has also been implicated in signal transduction pathways51 and in control of cell adhesion and migration.52 Finally, NME2 protein has been shown to bind single-stranded telomeric DNA and the RNA component of telomerase.53 The physiological role of this interaction (if any) has yet to be determined. Recently, however, NME2 immunoreactivity was shown to be positively associated in a statistically significant manner with telomerase activity in hepatocellular carcinoma.54

CRI1 (also known as EID-1) is a CREB-binding protein and potential oncogene that, in functional assays, antagonizes the action of pRb, p300, and CREB-binding protein (CBP) histone acetylase activity.55,56 Although CRI1 is capable of binding and sequestering wild-type unphosphorylated (active) pRb, there is also evidence that CRI1 acts at points downstream of pRb in differentiation and proliferation control pathways.55 These observations are particularly notable in the context of mesothelioma because, although mutational or deletional loss of pRb is rare in most tumor types including MPM,11 many MPMs are found to have deletions, mutations, or promoter methylation of p16^INK4a,57 a regulator of pRb via inhibitory action on cdk4.

Validation of Microarray Data

To initially validate microarray data, we performed quantitative RT-PCR for NME2 and GSN in all samples profiled with microarrays. (To facilitate comparison between platforms, we stated the expression level of both experimental genes relative to those of GAPDH in each sample.) We compared gene expression levels generated from quantitative RT-PCR to those from microarray analysis and found that the relative ranking of all tissues was relatively similar in both data acquisition platforms. For example, of the 10 MPM tumors with the highest and lowest NME2 levels from microarray analysis, a total of 8 and 7 were similarly ranked, respectively, using quantitative RT-PCR. Similar results were obtained for GSN.

To determine whether these genes are useful for study in vitro and additionally validate microarray data, we quantified the mRNA levels of all three genes up-regulated in MPM (PDGFC, CRI1, NME2) in four MPM cell lines and HM3 cells, a normal (nontransformed) mesothelial cell line58 not profiled using microarrays, with and without EGF in the culture medium as a control for the proliferation status of the cells (Figure 4). Both conditions (ie, ±EGF) permitted the growth of the MPM lines, but the −EGF condition induced a growth-arrested state for HM3 cells, as described previously.28 We found that expression levels were typically higher in tumor cell lines relative to growth-arrested (−EGF) HM3 cells. EGF produced a twofold increase in CRI1 mRNA levels for HM3 cells, but no or little increase for the tumor lines, suggesting that the MPM cell lines were no longer dependent on exogenous mitogen for expression of this gene (Figure 4A). Curiously, one tumor cell line (MS428) exhibited a twofold decrease in PDGFC mRNA in the presence of EGF (Figure 4C).

Figure 4.

Validation of select MPM molecular markers. We examined, in normal and tumor cell lines with and without EGF in the culture medium, the mRNA levels of three genes that were found to be significantly differentially expressed between MPM tumors and normal tissues, as described in the text. We performed quantitative RT-PCR using four MPM cell lines (MS589, MS428, MS924, and JMN1B) and one normal mesothelial cell line (HM3). Individual mRNA levels were normalized to GAPDH and expressed relative to those in HM3 cells (−EGF) for CRI1 (A), NME2 (B), and PDGFC (C). Expression levels for tumor markers were typically at least twofold higher in tumor cell lines relative to HM3 (−EGF) cells for all three genes. Black bars, −EGF; white bars, +EGF.

Validation of a Previously Described MPM Diagnostic Test

MPM can be difficult to distinguish from lung adenocarcinoma without a surgical biopsy in patients presenting with unilateral malignant pleural effusion. We have previously described a simple, but highly accurate test based on the ratios of the expression levels of six genes that can overcome this difficulty. This test uses the combined score (ie, geometric mean; see Materials and Methods) of three gene pair ratios to distinguish between lung adenocarcinoma and MPM: calretinin/claudin-7, VAC-β/TACSTD1, MRC-OX/TITF.13 Using quantitative RT-PCR, we obtained the relative expression level of each of these six genes in the 40 MPM tumors profiled in the current study and used these data to calculate the combined score of the three diagnostic ratios. As per the original publication,13 samples with a combined score >1 and <1 were called MPM and lung adenocarcinoma, respectively. We found that this test was 97.5% accurate with a single MPM tumor being misdiagnosed (tumor 36).

Protein Analysis of MPM Candidate Oncogenes and Tumor Suppressors

A subset of MPM candidate oncogenes and tumor suppressors from above were examined for protein expression in MPM cell lines using Western blot analysis (Figure 5) and in MPM tumors using immunohistochemical analysis of an MPM tissue array (Figure 6). CRI1 protein was detected in four of the five MPM cell lines examined but not in normal pleural tissue (Figure 5A). No CRI1 protein was detected in MS589 cells, consistent with mRNA levels from Figure 4A. CRI1 was also faintly detected in the Met-5A cell line. Protein levels of NME2 had approximately similar expression patterns in Met-5A, MPM cell lines, and normal pleura (Figure 5A). Two bands of similar molecular weight were recognized by the anti-NME2 antibody in MS428 cell lysates. The cause of this phenomenon is unknown and occurs in a small number of cell lines (eg, HeLa and K562 cells; personal communication, Santa Cruz Biotechnology). GSN protein was only detected in normal pleural tissue and not in any cell line examined (Figure 5B). These results provide evidence that transcription and translation are coupled for NME2 and CRI1 (compare Figures 4 and 5). Of note, CRI1 and NME2 were both detected in the Met-5A cell line. One explanation for this observation is that MPM cells acquire expression of these (and other) candidate oncogenes in a stepwise progression during tumorigenesis because Met-5A, while nontumorigenic, is a mesothelial cell-derived SV40-immortalized cell line.

Figure 5.

Protein analysis of candidate MPM tumor-related genes in vitro. Western blot analysis was performed for CRI1 and NME2 (A) and GSN (B) in nontumorigenic SV40-immortalized mesothelial cells (Met-5A) and six human MPM cell lines (98-483, MS589, MS428, JMN1B, 211-H, H-2052) as described in the Materials and Methods using β-actin as a loading control and pooled normal pleura as a control tissue.

Figure 6.

Protein analysis of candidate MPM tumor-related genes in human MPM tissues. We performed indirect immunohistochemical analysis of MPM tissue arrays consisting of 66 MPM tumors, 2 lung adenocarcinoma tumors, and 4 normal pleura tissues using antibodies to NME2 (A and B) and CRI1 (C). Expression of NME2 antigen varied widely in intensity in a diverse percentage of tumor cells (see Table 1), while CRI1 antigen was consistently expressed at moderate levels in all tumors examined. Neither protein was detected in normal pleura tissue or stromal elements of tumor tissues. Original magnifications: ×4 (A); ×10 (B and C).

NME2 protein showed variable and heterogeneous expression in the normal and tumor tissues present on the MPM tissue array (Figure 6, A and B, and Table 1). It is important to note that no NME2 protein was detected in any normal pleura tissue or in stromal cells from tumor tissues. NME2 protein was detected in a small number of tumor cells in both lung adenocarcinoma tissues and the MPM sarcomatoid histological subtype. The majority of epithelial (78%, 35 of 45) and mixed (74%, 14 of 19) MPM subtypes had detectable NME2 protein. Nearly half of all epithelial (44%, 20 of 45) and one-third of mixed (37%, 7 of 19) MPM subtypes had NME2 expression in at least 50% of tumor cells. When NME2 protein was detected, it was nearly always present at high levels with 74% (26 of 35) and 86% (12 of 14) of epithelial and mixed subtypes scored as strong expression, respectively. There was no statistically significant correlation between histological subtype and NME2 protein expression with respect to positive and negative cases, strongly and weakly staining cases, or to cases with moderate and extensive staining (ie, <50% and >50% of tumor cells positive, respectively). CRI1 protein expression displayed a more homogenous staining pattern on the MPM tissue array (see representative MPM tissue in Figure 6C). Similar to NME2 expression, CRI1 protein was not detected in any of the normal pleura tissues or stromal elements in tumor tissues, but was detected in nearly all tumor cells from MPM and lung adenocarcinoma tissues.

Table 1.

Scoring of NME2 Protein Expression in MPM Tissue Array

Positive cells	Tissue type
	Ept	Mixed	Sarc	ADCA	Np
0	10	5	1	1	4
<25%	10	6	1	1
25–50%	5	1
50–75%	7	5
>75%	13	2

Column/row intersections represent the distribution of normal and tumor tissues with NME2 protein expression detected in an increasing number of cells. For tumor tissues, only tumor cells were used for quantification and not surrounding stromal cells, which were consis-tently negative for NME2 protein. 

Ept, MPM epithelial subtype; mixed, MPM mixed subtype; Sarc, MPM sarcomatoid subtype; ADCA, lung adenocarcinoma; Np, normal pleura. 

Pathway Analysis of MPM Candidate Oncogenes

We analyzed all 328 genes found to be statistically significantly up-regulated in MPM tumors relative to normal tissues from above to discover novel pathophysiological pathways with potential relevance to MPM. Genes that were unknown (ie, expressed sequence tags) or uncharacterized in the scientific literature were discarded. We discovered a total of 12 networks, 7 of which had at least one gene in common and were merged for display. An enlarged portion of the resulting network is shown in Figure 7 and a high-resolution picture of the entire network can be found in Supplemental Figure 7 at http://ajp.amjpathol.org. The combined network consisted of 210 genes, approximately half of which (46%, 97 of 210) were MPM up-regulated genes. (For simplicity, genes are referred to by their LocusLink symbol in Figure 7 and the text.) Five genes (TP53, SP1, SPARC, FN1, MYCN) were shared by at least two networks. Of these, SP1 and SPARC were not previously described in MPM in the published scientific literature. The SP1 transcription factor has multiple functions and can drive the expression of many cancer-related genes such as VEGF. SPARC (also known as osteonectin, ON) is a matrix-associated protein that can elicit changes in cell shape, inhibit cell-cycle progression, and influence the synthesis of extracellular matrix. Note in Figure 7 that SPARC binds to and modulates vitronectin (VTN), an MPM up-regulated gene that promotes cell adhesion and spreading. Pathway analysis revealed that the SPARC/VTN interaction can also modulate adhesion and spreading of endothelial cells, which may function in angiogenesis in the context of MPM tumorigenesis.

Figure 7.

Pathway analysis of candidate MPM tumor-related genes. Genes found to be statistically significantly up-regulated in MPM were used to create a combined network of physiological pathways as described in the Materials and Methods. A portion of the resulting pathway is shown here. The intensity of the node color (red) indicates the experimentally determined degree of up-regulation of MPM tumor-associated genes expressed as a fold-increase in average expression levels relative to normal tissues. Other node colors are white (genes that are not user specified but are incorporated in networks through relationships with other genes) and yellow (genes that are shared by two or more networks in a merged diagram). Nodes are displayed using various shapes that represent the functional class of the gene product (see legend). Edges are displayed with various labels and shapes that describe the nature of the relationship between the nodes (eg, B for binding, T for transcriptional regulation) and links to experimental results and gene summaries supporting such a relationship. The edge labels and shapes have been removed from the diagram for the sake of clarity but can be found in Supplemental Figure 7 at http://ajp.amjpathol.org. The length of an edge reflects the evidence supporting that node-to-node relationship. Edges supported by more articles from the literature are shorter. For simplicity, genes are referred to by their LocusLink symbol.

We found it interesting that several MYCN-regulated genes are highly expressed in MPM tumors (including NME2 which is also regulated by MYC) even though MYCN is not generally thought to play a role in MPM carcinogenesis (Figure 7). Although MPM tumors and cell lines contained low, but reliably detectable levels of MYCN mRNA by microarray analysis, this gene may be physiologically relevant during MPM tumorigenesis because MYCN protein has been detected in some MPM tumors,59 and not all neuroblastoma tumors or cell lines have MYCN genomic amplification yet still contain detectable mRNA and protein.60 Alternatively, genes regulated by multiple myc family members could be transcriptionally activated primarily by MYC in MPM.

Other genes forming prominent nodes in pathway analysis include well-studied transcription factors regulating proliferation and apoptosis (E2F1, E2F4, FOS, MYC), cell cycle regulators (RB1), and growth factors (VEGF). Several of these networks contain genes that have been previously described in MPM and/or are consistent with current knowledge. For example, the up-regulation of genes transcriptionally regulated by TP53 and RB1 tumor suppressors is consistent with the lack of mutations in these genes in MPM.9–11 Similarly, VEGF and MYC have been previously described as potential proto-oncogenes in MPM.8 Curiously, we found high levels of the FMR1 gene in MPM tumors. Mutations in this gene are responsible for X-linked mental retardation (ie, fragile X syndrome) and the role of this gene (if any) in cancer is unknown. This mRNA binding protein has not previously been described in MPM and forms a node with other genes that have mRNA to which it is capable of binding. These genes (eg, GNAS, SPP1, CDK4, HINT1, COLL11A1, CDH11, CBX3) are themselves up-regulated in MPM. Although the exact functional relevance of these results is unknown at present, at least one of these genes is in a pathway with direct cancer relevance (eg, SPP1, which interacts with the anti-apoptotic protein IGFBP2, also expressed at high mRNA levels in MPM).

In other studies, investigators have found that AP-1 transcription factor family members FOS and FOSL1 (also known as fra-1) were associated with MPM tumorigenesis, consistent with our pathway analysis showing that multiple MPM up-regulated gene products form a node with FOS (Figure 7). In these analyses, FOSL1 mRNA levels were found to be elevated in both rat mesothelioma cell lines and rat mesothelial cells exposed to asbestos, but FOS mRNA levels were up-regulated only in the latter.61–64 This finding prompted these investigators to suggest that FOSL1 may substitute for FOS in AP-1 complexes during human MPM tumorigenesis.62 Consistent with these findings, we have also observed relatively low (or absent) mRNA levels of FOS and substantially higher mRNA levels of FOSL1 in human MPM cell lines in the current study and in those cell lines previously profiled with microarrays in our laboratory.13 However, we observed FOS mRNA levels that are 56-fold higher than those of FOSL1 in the human MPM specimens profiled in the current study, consistent with a similar eightfold difference previously observed.13 In fact, the average expression level of FOS in human MPM tumors in the current study is in the 98th percentile relative to the average expression level of all of the genes represented on the microarray. The high intertumor variability of FOS expression levels (average, 1009; range, 5.7 to 3718) accounts for the fact that it is not statistically significantly elevated compared to normal tissues. Despite the lack of FOS mRNA in rat and human MPM cell lines in our studies and others, the high levels of this gene in human MPM tumors necessitates its continued study in MPM pathogenesis.

Several of the genes highly expressed in MPM relative to normal tissues have been previously shown to have prognostic value in MPM. Pass and colleagues17 profiled 21 MPM patients and created a 27-gene neural network classifier that could predict clinical outcome with high accuracy. Of these 27 genes, 5 were found to be statistically significantly up-regulated in MPM in the current study (CRIP1, IGFBP5, KIAA0275, KIAA1237, MYC). It is noteworthy that NME1 was reported as a prognostic gene in MPM by these authors.17 This gene is highly related to NME2, a tumor marker discovered in the current study. Clearly, this family of genes is important in MPM and deserves additional study.

In this report we describe an extensive transcriptional analysis of MPM tumors leading to the identification of at least two possible subclasses of MPM with different gene expression patterns that correlate loosely with tumor histology. We also identify and validate a subset of MPM tumor-related genes with aberrant expression that may contribute to MPM tumorigenesis. Finally, we performed pathway analysis of MPM candidate oncogenes to identify potential cellular pathways, the dysregulation of which may contribute to and/or reflect MPM tumorigenesis and pathogenesis.