A number of genetic and epigenetic changes underlying the development of nasopharyngeal carcinomas have recently been identified. However, there is still limited information on the nature of the genes and gene products whose aberrant expression and activity promote the malignant conversion of nasopharyngeal epithelium. Here, we have performed a genome-wide transcriptome analysis by probing cDNA microarrays with fluorescent-labeled amplified RNA derived from laser capture microdissected cells procured from normal nasopharyngeal epithelium and areas of metaplasia-dysplasia and carcinoma from EBV-associated nasopharyngeal carcinomas. This approach enabled the identification of genes differentially expressed in each cell population, as well as numerous genes whose expression can help explain the aggressive clinical nature of this tumor type. For example, genes indicating cell cycle aberrations (cyclin D2, cyclin B1, activator of S-phase kinase, and the cell cycle checkpoint kinase, CHK1) and invasive-metastatic potential (matrix metalloproteinase 11, v-Ral, and integrin β4) were highly expressed in tumor cells. In contrast, genes underexpressed in tumors included genes involved in apoptosis (B-cell CLL/lymphoma 6, secretory leukocyte protease inhibitor, and calpastatin), cell structure (keratin 7 and carcinoembryonic antigen-related cell adhesion molecule 6), and putative tumor suppressor genes (H-Ras-like suppressor 3, retinoic acid receptor responder 1, and growth arrested specific 8) among others. Gene expression patterns also suggested alterations in the Wnt/β-catenin and transforming growth factor β pathways in nasopharyngeal carcinoma. Thus, expression profiles indicate that aberrant expression of growth, survival, and invasion-promoting genes may contribute to the molecular pathogenesis of nasopharyngeal carcinoma. Ultimately, this approach may facilitate the identification of clinical useful markers of disease progression and novel potential therapeutic targets for nasopharyngeal carcinoma.
Nasopharyngeal carcinoma poses a major public health challenge in Southeast Asia, and at least 2000 new cases of nasopharyngeal carcinoma are diagnosed each year in the United States (1 , 2) . Although nasopharyngeal carcinoma is classified as a subtype of head and neck squamous cell carcinoma, its unique epidemiology, clinical characteristics, etiology, and histopathology warrant separate efforts for the study of its underlying molecular mechanisms of carcinogenesis (1) . For example, nasopharyngeal carcinoma patients tend to present at a more advanced stage of disease because the primary anatomical site of tumor growth is located in a silent area, and they exhibit higher metastatic potential when compared with other head and neck squamous cell carcinoma (3, 4, 5) . In addition, a strong association of EBV and nasopharyngeal carcinoma has been widely accepted (1 , 6) . Indeed, nasopharyngeal carcinoma can be EBV or non-EBV associated, being the latter with a higher prevalence in the endemic regions, as reflected by the detection of EBV gene products in the vast majority of the nasopharyngeal carcinoma patients (7) . The persistence of the viral genome in tumor cells is usually associated with nonkeratinized squamous cell carcinoma or undifferentiated carcinoma (8 , 9) . Extensive investigations on the contribution of EBV to carcinogenesis have identified at least one viral gene product, the latent membrane protein 1, which acts as a transforming protein in many cellular contexts, and thus represents a candidate molecule for the initiation and/or maintenance of nasopharyngeal carcinoma (6 , 10 , 11) .
A number of studies have recently identified multiple genetic and epigenetic alterations that occur in nasopharyngeal carcinoma. For example, extensive and high-resolution allelotyping studies have revealed potential tumor suppressor gene loci in nasopharyngeal carcinoma on chromosomes 3p, 9p, 9q, 11q, 12q, 13q, 14q, and 16q (12 , 13) . Candidate approach studies have also identified aberrant hypermethylation of genes residing in these loci such as RASSF1A, RARβ2, and p16INK4A (14, 15, 16) . However, for most of these loss of heterozygosity (LOH) loci, the presence of candidate tumor suppressor genes has not been thoroughly investigated. These genetic approaches can eventually lead to the identification of critical genes that are important for nasopharyngeal carcinoma development and progression. Complementary efforts to characterize the gene expression pattern during the malignant conversion of nasopharyngeal carcinoma and, in particular of those genes involved in key intra- and extracellular molecular pathways, could ultimately lead to the identification of functional interactions among genes and their protein products, which in turn may explain the unique biological features and aggressive nature of this particular cancer type. The comprehensive analysis of the gene expression profile of nasopharyngeal carcinoma may help identify aberrantly expressed or mutated genes, which may thus represent novel molecular targets for therapeutic intervention in this disease.
Relatively limited information is currently available on gene expression in nasopharyngeal carcinoma, which has been generated using bulk nasopharyngeal carcinoma tissues or nasopharyngeal carcinoma-derived cells in culture and using low-density cDNA arrays (17 , 18) . Recent technological progress using laser capture microdissection (LCM) has now made it possible to enrich populations of cells from heterogeneous tissues (19, 20, 21) , which is particularly relevant to nasopharyngeal carcinoma as primary tumors often include numerous infiltrating inflammatory cells, nonneoplastic nasopharyngeal epithelium, and stroma. In this regard, the LCM platform enables the isolation of homogeneous cell populations for further downstream applications. In this study, we have focused on EBV-associated nasopharyngeal carcinoma, given its higher prevalence and thus clinical relevance, and used LCM to procure cells from tumors or adjacent nonneoplastic tissues from nasopharyngeal carcinomas for transcriptional profiling by high-density cDNA microarray. This comparative analysis between normal nasopharyngeal epithelium, metaplastic-dysplastic epithelium, and tumor tissue revealed numerous differentially expressed genes. Validity of the microarray results was confirmed by quantitative reverse transcription-PCR and immunohistochemical evaluation of gene products whose levels were assessed to be differentially expressed in normal and tumor cells. These efforts may ultimately provide a better understanding of the molecular mechanisms underlying nasopharyngeal carcinoma carcinogenesis, as well as help identify suitable targets for the development of novel treatment strategies for this aggressive cancer type.
MATERIALS AND METHODS
Informed consent was obtained from each patient before collecting pretreatment specimens from nasopharyngeal carcinoma patients at the Chulalongkorn University Hospital. Each biopsy sample was divided into two sections, one that was submitted for routine histological diagnosis for nasopharyngeal carcinoma, and the remaining section was flash frozen and stored in liquid nitrogen until ready for microdissection. We selected eight clinical samples based on adequacy of the tissue, the presence of nonneoplastic epithelium, and the appropriate status of tissue preservation. Clinical data, including diagnosis, staging, and cell populations isolated from each sample, are included in Table 1⇓ . As patient specimens were too small to allow further sectioning, for immunohistochemical staining a set of eight archival paraffin-embedded nasopharyngeal carcinoma samples unrelated to the frozen samples were selected.
Confirmation the Presence of EBV.
All specimens were confirmed for EBV positivity by either in situ hybridization of EBV-encoded small nonpolyadenylated RNA (EBER) or by PCR of the EBV BamHI-W region from genomic DNA. EBER in situ hybridization was performed by using fluorescein-conjugated EBER PNA probe (Dako, Glostrup, Denmark) according to the manufacturer’s instruction. Primers for PCR of the BamHI-W region were described elsewhere (22) .
Isolation of tumor and nasopharyngeal epithelial cells by LCM was performed as described previously (21) . Briefly, 6–8-μm cryostat sections stained with H&E were prepared for LCM. The Arcturus Pixell II apparatus was used to harvest cells of interest. Target cell number for each sample was ∼2500–5000 cells.
Nucleic Acid Isolation and mRNA Amplification.
For PCR of genomic DNA, DNA was extracted from samples by incubating overnight at 55°C with lysis buffer containing 10 mm Tris-HCl, 1 mm EDTA, 1% Tween 20, and 0.1 mg/ml proteinase K (pH 8.0). Total RNA was extracted using a modified version of the Stratagene RNA microisolation kit (Stratagene, La Jolla, CA), as described previously (21) . Two rounds of mRNA amplification were performed for each sample by using the MessageAmp kit for T7-based amplification (Ambion, Austin, TX), according to the manufacturer’s protocol (23) . Universal Human Reference RNA (Stratagene) was used as a source of common reference RNA and was also subjected to mRNA amplification in parallel to each nasopharyngeal carcinoma sample.
cDNA Microarrays and Analysis.
Human GEM2 cDNA clones were obtained from Incyte Genomics, Inc. (Palo Alto, CA), and these were arrayed onto precoated glass slides at the Advanced Technology Center (National Cancer Institute, Gaithersburg, MD). Each array consisted of 9128 cDNA clones, which included 7102 known genes, 1179 expressed sequence tag cluster, and 122 Incyte-expressed sequenced tags clones. Preparation of fluorescent-conjugated cDNA targets were performed by an indirect labeling approach using amplified RNA as template. Briefly, 1 μg of amplified RNA was used for each reverse transcription with random hexamer primers and a mixture of nucleotides at a final concentration of 0.5 mm dATP, dCTP, and dGTP, 0.1 mm dTTP, and 0.15 mm 5-(3-aminoallyl)-2′ deoxyuridine-5′-triphosphate (Sigma, St. Louis, MO). cDNA targets were then coupled with fluorescent dye, Cy3, or Cy5 (Amersham, Piscataway, NJ) under 0.1 m NaHCO3 for 1 h in the dark. The probes were appropriately mixed, purified, and concentrated before hybridization to arrays at 42°C overnight. After washing, the arrays were scanned with GenePix scanner (Axon Instruments, Union City, CA). Most samples were repeated at least twice using independent cDNA hybridization in which tumor and noncancerous tissues were labeled with green (Cy3) and reference RNA-labeled with red (Cy5).
Statistical analysis of the microarray data were performed using the BRB ArrayTools software developed by Dr. Richard Simon and Amy Peng from Biometric Research Branch of the National Cancer Institute.7 We used both supervised and unsupervised hierarchical clustering analyses with median center correlation and average linkage. We also used a supervised Class Comparison Tool based on univariate F tests to identify differentially expressed genes between predefine cell populations according to their histology. A stringent criteria was used to define the statistical significance of each observed change in gene expression, using F statistics (P < 0.001), the significance of which was confirmed by 2000 random permutations (24) .
Quantitative Reverse Transcription-PCR.
RNA from the second round of amplification was used to generate cDNA. Briefly, 1 μg of aRNA was used as starting material, to which we added 1 μl (3 μg/μl) of random primer and diethyl pyrocarbonate-treated water to total volume 25 μl, then heated the mixture at 65°C for 5 min and chilled on ice. The other components were added as follows: 10 μl of 5× first strand buffer; 5 μl of 0.1 m DTT; 1 μl of 25 mm deoxynucleoside triphosphates; 1 μl of RNase inhibitor; and 6 μl of diethyl pyrocarbonate-treated water. The samples were incubated at 42°C for 2 min. Then 1 μl of Superscript II (40 units/μl; Invitrogen, Inc., Carlsbad, CA) was added, and the samples were incubated at 42°C for 50 min. The reaction was inactivated at 70°C for 5 min, and 1 μl (2 units/μl) of RNAase H was added to degrade the RNA strand by incubating at 37°C for 20 min. Real-time PCR using the ABI prism 7700 sequencer detector system and Qiagen’s Quantitect SYBR Green PCR kit (Qiagen, Valencia, CA) was performed following the manufacturer’s protocol. In brief, the reaction mixture (50 μl of total volume) contained 500 ng of cDNA, gene-specific forward and reverse primers at 1 μm final concentration, and 25 μl of 2× Quantitect SYBR Green PCR Master mix. The real-time cycler conditions were as follows: PCR initial activation step at 95°C for 10 min; 40 cycles each of melting at 95°C for 15 s; and annealing/extension at 60°C for 1 min. A negative control without template was run in parallel to assess the overall specificity of the reaction. The PCR products were analyzed on a 1.2% agarose gel to confirm the size of the amplified product. The comparative threshold cycle (CT) (ΔΔCT) method, which compares differences in CT values of common reference RNA and nasopharyngeal carcinomas or normal tissues, were used to achieve the relative fold changes in gene expression between normal and tumor. The experiments were repeated in triplicate and the mean fold changes and SE between normal and nasopharyngeal carcinoma are reported.
Gene-specific primers were generated using the Primer 3.0 program (provided by Whitehead/Massachusetts Institute of Technology Center for Genome Research, Cambridge, MA). The primer sequences were as follows: CD83, forward 5′-AGAAGGGGCAAAATGGTTCT-3′, reverse 5′-CACCTGTATGTCCCCGAGTT-3′; MMP11, forward 5′-TAGGTGCCTGCATCTGTCTG-3′, reverse 5′-AGAATACCCCTCCCCATTTG-3′; CCND2, forward 5′-AGGCAGCTGACTATATCA-3′, reverse 5′-CTGCTGGCCAACTCTTAC-3′; SLPI, forward 5′-GGGAAGTGCCCAGTGACTTA-3′, reverse 5′-GGCAGGAATCAAGCTTTCAC-3′; PIGR, forward 5′-AGCCTCTTCGATCACTCAGG-3′, reverse 5′-TGGACTGGAGCAGGAAGTCT-3′; and KRT7, forward 5′-CAGGATGTGGTGGAGGACTT-3′, reverse 5′-TTGCTCATGTAGGCAGCATC-3′. In parallel, a human 18 s rRNA forward 5′-CGCCGCTAGAGGTGAAATTC-3′, reverse 5′-TTGGCAAATGCTTTCGCTC-3′ was used as an endogenous control for normalization.
Paraffin-embedded tissue sections were deparaffinized in xylene and hydrated through graded alcohols and distilled water. Sections were subjected to antigen unmasking with AUF solution (Vector Laboratories, Burlingame, CA) for cyclin D2, antigen retrieval with 0.25% trypsin (Invitrogen, Inc.) for carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6), or their combination for keratin 7 (KRT7). Slides were incubated in 3% hydrogen peroxide in ethanol to quench the endogenous peroxidase. The sections were incubated in blocking solution, 2% BSA. We used biotinylated secondary antibody (Vector Laboratories) and Vectastain kit (Vector Laboratories). The slides were developed in 3,3-diaminobenzidine (Sigma). The primary antibodies used in this study were KRT7 Ab-1, CEA/CD66e Ab-2, and MMP11 (Neomarker, Fremont, CA); and cyclin D2 (M-20) (Santa Cruz Biotechnology, Santa Cruz, CA). Methyl green [2% in acetate buffer (pH 4.8)] was used as nuclear counterstain.
We first confirmed the homogeneity of tumor origin of the sample sets by confirming the presence of EBV genome. All tumor specimens were positive for EBV as judged by either EBER in situ hybridization (Fig. 1)⇓ or genomic DNA PCR for EBV BamHI-W region (data not shown). EBER was found in the nuclei of tumor cells but absent in the normal nasopharyngeal epithelial cells (Fig. 1, A and B)⇓ . In the adjacent metaplastic and dysplastic nasopharyngeal epithelium, EBER was mainly localized to the basal and suprabasal layers (Fig. 1C)⇓ , which is consistent with previous studies supporting an involvement of EBV in nasopharyngeal carcinoma carcinogenesis (25) .
According to their cytologic features, we isolated three populations of cells from each nasopharyngeal biopsy, normal pseudostratified columnar ciliated respiratory epithelium (N), squamous metaplastic or dysplastic epithelium (R), and carcinoma (T). Within these clinical specimens, we characterized normal epithelium (4) , metasplasia or dysplasia (4) , and carcinoma (Ref. 6 ; Table 1⇓ ). Each of these cell populations were isolated by LCM and subsequently subjected to total RNA isolation. To overcome the problem of the limited amount of mRNA isolated from LCM procured samples (∼50–100 ng of total RNA), we chose to increase target mRNA by linear amplification with a T7 polymerase-based in vitro transcription system. This method has been shown to exhibit high fidelity and reproducibility and to yield RNA that is suitable for microarray analysis (26, 27, 28) . A universal human reference RNA was selected as the common reference for the microarray study because this frequently used RNA reference may afford the possibility of future comparisons with other data sets. To avoid bias in hybridization to microarrays, the reference RNA was amplified in parallel to those generated from nasopharyngeal carcinoma samples and used as a reference for each nasopharyngeal carcinoma-derived sample for their hybridization to cDNA microarray chips. All arrays included for additional data analyses had a mean target signal intensity of at least 4-fold over the background intensity, and each array was normalized by subtracting the median log-ratio from log-ratio of each spot. The initial analysis of the fluorescence intensity data by global hierarchical clustering of all of the tissue samples showed limited correlation with their histological characteristics. Thus, we performed additional analyses by a supervised class comparison based on the predefined histological characteristics (24) to detect genes differentially expressed according to each of the observed histological groups. By this approach, we identified 650 genes that were differentially expressed among the three groups (P < 0.001). Reanalysis of the data by a supervised hierarchical clustering of samples based on these 650 genes revealed three clusters that matched well with each cell population (Fig. 2B)⇓ . Among these genes, clustering analyses revealed four major clusters: underrepresented in normal; highly represented in normal; underrepresented in tumor; and highly represented in tumor.
We next performed pairwise comparisons among the three cell populations using a supervised class comparison approach. In normal versus tumor comparison, 477 genes were differentially expressed, of which, 402 and 83 genes were highly expressed in normal and tumor, respectively (Fig. 3)⇓ . When metaplastic-dysplastic tissues were compared with tumor cells, 169 genes were differentially expressed, of which, 121 and 48 genes were highly expressed in metaplastic-dysplastic cells and tumor, respectively (Fig. 3)⇓ . By comparing RNA extracted from normal and metaplastic-dysplastic cells, 658 genes were found to be differentially expressed, of which, 457 and 201 genes were highly expressed in normal and metaplastic-dysplastic tissues, respectively (Fig. 3)⇓ . Thus, the transcriptional profiles of metaplasia-dysplasia and nasopharyngeal carcinoma were both quite distinct from those of adjacent normal nasopharyngeal epithelium.
Although the complete list of genes differentially expressed is provided (see supplemental information), we focused our attention on those genes of known function from normal and tumor cells whose aberrant (increased or decreased) expression may play critical roles in determining the malignant phenotypes of nasopharyngeal carcinoma (Tables 2⇓ and 3⇓ ⇓ ⇓ ). Highly represented genes in the carcinoma group included numerous genes involved in cell cycle regulation and DNA replication, including Check 1 Kinase checkpoint homologue (CHK1), cyclin D2 (CCND2), cyclin B1 (CCNB1), likely ortholog of mouse gene rich cluster, C8 gene (GRCC8), and centromere protein F (CENPF). Other set of differentially expressed genes function in cell adhesion and thus may potentially confer the invasive-metastatic property of tumor cells such as integrin β4 (ITGB4), matrix metalloproteinase 11 (MMP11), and syndecan 2 (SDC2). Interestingly, genes involved in Ras/Ral signaling pathways, RalA, as well as in the Wnt/βcatenin pathways, frizzle 7 (FZD7) and claudin1 (CLDN1), were also up-regulated in the tumor compartment.
Among the genes underrepresented in the tumor cell population, there were a number of potential tumor suppressor genes such as epidermal growth factor receptor pathway substrate 8-related protein 1, retinoic receptor responder 1 (RARRES1), H-Ras-like suppressor 3 (HRASLS3), mutated in colorectal cancers (MCCs), and LOH 11 chromosomal region 2 gene A (LOH11CR2A). Down-regulated apoptosis-related genes included in this group were B-cell CLL/lymphoma 6, calpastatin, calpain 3, caspase10, and serine leukocyte protease inhibitor (SLPI). Genes related to the Notch pathway numb homologue (NUMB) and transducin-like enhancer of split 2 (TLE2) were also underrepresented in tumors. Structural genes in this group included KRT7, CEACAM6, mucin 1, and mucin 5b. Within the down-regulated gene group, this analysis identified polymeric immunoglobulin receptor (pIgR) and EBNA1 binding protein 2, which may play role in the EBV pathogenesis in nasopharyngeal carcinoma.
To validate the results of microarray studies, we first performed quantitative reverse transcription-PCR analysis of six selected genes. Consistent with the microarray data, the expression levels of CD83, MMP11, and cyclin D2 were higher in tumor cells, whereas the expression of pIgR, SLPI, and KRT7 was higher in normal tissues and thus underexpressed in tumor cells (Fig. 4A)⇓ . To study further the relationship between RNA and protein levels, we selected a representative set of proteins whose mRNAs were shown to be over- or underexpressed in the normal versus tumor comparison in microarray experiments based on the availability of their corresponding antibodies. A total of eight paraffin blocks from nasopharyngeal carcinoma patients was selected based on the presence of neoplastic and nonneoplastic epithelium within the same section. Tissue sections were subjected to immunostaining with antisera to KRT7, CEACAM6, MMP11, and cyclin D2. All eight samples showed strong cytoplasmic staining of KRT7 and CEACAM6 in histologically normal or metaplastic nasopharyngeal epithelium, which was in striking contrast to its undetectable expression in cancer cells (Fig. 4⇓ , compare left three panels). All nasopharyngeal carcinoma samples were strongly immunoreactive for anti-cyclin D2 and anti-MMP11 (Fig. 4, B and C⇓ , last two panels), in contrast to nonneoplastic tissues. For cyclin D2, the majority of the cancer cells showed strongly positive nuclear staining, whereas a somehow weaker staining limited to a few cells within the basal and suprabasal epithelial layers was observed in normal and metaplastic epithelium. All clinical samples showed stronger and more extensive cyclin D2 immunoreactivity in tumor cells. For MMP11, a stronger expression in tumor cells was found in five of eight tissues, whereas the rest exhibited only barely detectable expression in both nasopharyngeal carcinoma and noncancerous nasopharyngeal epithelium. Although these observations need to be extended to a large sample collection to establish their clinical relevance, their strict correlation with the gene expression profiles, as revealed by microarray analysis, support the validity of the combined use of LCM and cDNA microarrays as an experimental approach for the molecular profiling of nasopharyngeal carcinoma.
In this study, we provide a genome-wide analysis of the global pattern of gene expression in nasopharyngeal carcinoma by using a combination of laser capture isolation of defined cell populations, linear mRNA amplification, and high-density cDNA microarrays. In addition to normal and tumor cells, we were able to procure cells from adjacent metaplastic and dysplastic lesions and thus analyze gene expression patterns covering the spectrum of nasopharyngeal carcinoma carcinogenic progression. Genes that are differentially expressed in each cell phenotype were identified using bioinformatic tools. Overall, the number of genes differentially expressed between normal nasopharyngeal epithelium and tumor cells or metaplasia-dysplasia was 477 and 658 genes, respectively. In contrast, only 169 genes were identified as being differentially expressed between metaplasia-dysplasia and the tumor cell compartment. Taken together, these findings demonstrated a pattern of transcriptional alterations in nasopharyngeal carcinoma carcinogenesis, as well as revealed the existence of numerous genes whose expression levels is up- or down-regulated in this cancer type.
Previous studies addressing the molecular profile of nasopharyngeal carcinoma have been performed using low-density cDNA arrays that included a limited number of genes. For example, in one study that used pooled nasopharyngeal carcinoma and normal tissues the authors described up-regulation of cell cycle regulatory genes such as cyclin B1, cyclin D2, and cyclin A, which is consistent with our present results (17) . Other genes classified as differentially expressed, however, were not significantly different in our experimental analyses. As whole tissues were used in this prior study (17) , we cannot exclude the possibility that some of the detected changes in gene expression were associated with stromal or infiltrating inflammatory cells rather than with the tumor compartment. Another study using similar low-density cDNA arrays compared nasopharyngeal carcinoma cell lines with nasopharyngeal epithelial explants, which resulted in the identification of a set of differentially expressed genes poorly related to those described in our current study (18) . These discrepancies may arise from the different source of RNA used because cultured cells may have adapted to in vitro growth conditions and thus express genes that may be quite different from those normally transcribed in the tumor microenvironment. Instead, to isolate these potentially confounding variables, in the current study we chose to use recently developed techniques that enable the isolation of pure cell populations, which are optimized for the subsequent RNA extraction, linear mRNA amplification, and hybridization to high-density cDNA microarrays, thus providing a suitable platform to perform detailed gene expression analysis as part of an effort aimed to understand nasopharyngeal carcinoma pathogenesis.
Clinically, nasopharyngeal carcinoma has been recognized as an aggressive head and neck tumor by the fact that the majority of patients surrender to metastatic disease (1 , 3) . Indeed, the molecular profiling of nasopharyngeal carcinoma now revealed the presence of molecules consistent with a highly proliferative and invasive behavior. For example, CENPF (mitosin), whose expression is usually low in the G0 phase of the cell cycle, was found to be highly expressed in tumors, which is consistent with the active cycling status of tumor cells (29) . Moreover, aberrant cell cycle regulation throughout key checkpoints was strikingly portrayed. For example, cyclin D2, which activates the G1-phase cyclin-dependent kinases 4 and 6 (30) , the cell cycle G1-S and G2-M checkpoint kinase Chk1 (31) , and discs-large homologue 7, which has been proposed to play a role in checkpoint control and/or DNA repair based on its pattern of expression and homology with the Drosophila tumor suppressor gene dlg1 (32) , were highly represented in tumor mRNAs when compared with normal cells. In addition, favoring deregulated mitotic entry, nasopharyngeal carcinoma cells displayed up-regulation of GRCC8, which is also known as trigger of mitotic entry 1 or Tome-1 (33) cytosolic protein that promotes the activation of cyclin-dependent kinase 1/cyclin B and consequently mitotic entry. These tumor cells also displayed increased expression of cyclin B1 itself, in conjunction with the down-regulation of CDC16, a protein involved in the anaphase promoting complex (34) .
On the other hand, we have observed an up-regulation of stromelysin-3/MMP11 (35) in tumors. This protease has been described to be increased in a variety of cancer types, including breast (36) , lung (37) , and colon (38) , particularly in the nonneoplastic fibroblasts surrounding the malignant epithelial cells. In contrast, expression of MMP11 in nasopharyngeal carcinoma seems to be restricted to the malignant epithelium, as also described in oral squamous carcinomas (39) . This suggests that MMP11 represents a specific marker for carcinomas derived from the head and neck mucosa, a possibility that may warrant further exploration. Nasopharyngeal carcinoma cells also overexpress RalA (40) and integrin β4 (41 , 42) , combined with down-regulation of the proapoptotic caspase 10 (43 , 44) , the anti-inflammatory secretory leukocyte protease inhibitor (45) , and B-cell CLL/lymphoma 6, which can become oncogenic either by its persistent expression or upon accidental down-regulation, as in certain types of human B-non-Hodgkin’s lymphomas (46) . These genes have been shown to contribute to the oncogenic and metastatic potential of many cancers, which may fit well with the aggressive clinical behavior of nasopharyngeal carcinoma.
Histologically, EBV-associated nasopharyngeal carcinomas often present as poorly differentiated or undifferentiated carcinomas. In agreement, our results revealed down-regulation of genes for structural proteins that are associated with cellular differentiation such as KRT7 (47 , 48) , mucin1 (49) , and mucin 5b (50) . These molecules are usually expressed variable levels in normal epithelium and adenocarcinomas, but their expression in squamous cell carcinoma arising from various organs, including lung and head and neck cancer, is often restricted only to more differentiated areas within keratinized foci (50) . Down-regulation of a cell adhesion molecule gene, CEACAM6 (51) , which has been shown to be deregulated and to affect the differentiation of colon carcinomas, was also detected in nasopharyngeal carcinoma. Although these changes may reflect the undifferentiated state of nasopharyngeal carcinomas, they may also be linked to a common molecular alteration underlying these cancer lesions. As such, they may result from the observed reduction in the expression of the transcription factor Kruppel-like factor 4 (KLF4) (52) , which plays a critical role in skin barrier development in late-stage keratinocyte differentiation, or by the decreased expression of genes involved in cell fate determination related to Notch pathway such as numb (53) and TLE2 (54) , which were under expressed in tumor cells. Overall, our results indicate a remarkable perturbation of normal differentiation programs during nasopharyngeal carcinoma carcinogenesis. Whether these alterations in differentiation programs are caused by EBV and its encoded genes, as well as whether aberrant cell differentiation plays a causative role in nasopharyngeal carcinoma or are a direct consequence of the unregulated cell growth that characterizes this aggressive tumor, is at the present unknown and warrants additional investigation.
Of interest, our study revealed the existence of clear differences in the expression of genes involved in signal transduction in nasopharyngeal carcinoma. For example, although the Ras-liked GTPase, RalA, was assessed to be up-regulated, RalB in contrast was down-regulated in nasopharyngeal carcinoma. Although both RalA and RalB are downstream effectors of ras signaling pathway (55) , a distinct function for RalA and RalB has been described. For instance, RalA is involved in anchorage-independent cell proliferation, whereas RalB functions in the survival of tumor cells (56) . Another gene related to ras pathway, Rab21 (57) , was found up-regulated in nasopharyngeal carcinoma, whereas Ras-related associated with diabetes (Rad) (58) was down-regulated in nasopharyngeal carcinoma. Frizzle 7 (59) , which encodes a seven-transmembrane receptor in the Wnt pathway, was found up-regulated in nasopharyngeal carcinoma. Moreover, claudin-1 (60) , a tight junction gene whose expression is positively regulated by the presence of β-catenin, was up-regulated in the tumor cell population. In agreement with these findings, axin 2 (61) , an inhibitory molecule of the Wnt pathway, was underrepresented in nasopharyngeal carcinoma. These results are strongly suggestive of a dysregulated hyperactivity of the Wnt pathway in nasopharyngeal carcinoma. In contrast, we found down-regulation of genes such as transforming growth factor β receptor type III (62) and inhibin βB (63) , which have been shown to participate in cell growth inhibition as part of the transforming growth factor β signaling system. These results are consistent with the deregulation of major signaling pathways in nasopharyngeal carcinoma, which includes up-regulation of growth promoting pathway and decrease expression of biochemical routes involved in limiting cell proliferation.
From our array analysis, we detected lower expression of multiple putative tumor suppressor genes in nasopharyngeal carcinoma. Both HRASLS3 and LOH11CR2A genes reside on chromosome 11q, which shows high frequency of LOH by allelotyping studies (12 , 13) . HRASLS3 is a class II tumor suppressor gene in that its expression is repressed in the absence of gene deletion or mutation (64) and was initially identified by its ability to counteract the transforming activity of ras (65 , 66) . Another class II tumor suppressor gene, RARRES1, located on chromosome 3q, was found to be significantly underexpressed in tumors. RARRES1 has been shown to reduce tumorigenicity of prostate cancer cells, and its decrease expression correlates with promoter hypermethylation in lung, head and neck, and bladder cancer (67) . Other candidate tumor suppressor proteins for chromosome 5q and 16q included MCC (68) and growth arrest-specific 8 (69) , respectively. Collectively, these molecules may represent novel candidate tumor suppressor genes whose reduced expression may enable the unrestricted growth of nasopharyngeal carcinoma.
In summary, our study provides the global expression profile of nasopharyngeal carcinoma, thus generating a valuable platform for the full characterization of this malignancy at the molecular level. Indeed, the combination of LCM and high-density microarray analysis revealed the aberrant expression of key molecules involved in cell cycle regulation, signal transduction, differentiation, and survival, as well as those involved in tissue invasion and metastatic spread. The emerging molecular anatomy of nasopharyngeal carcinoma may now facilitate the identification of proteins whose expression or activity may promote uncontrolled growth in this malignancy, as well as those molecules that correlate with clinical staging, prognosis, or response to treatment, thus representing novel potential diagnostic tools and therapeutic targets for nasopharyngeal carcinoma.
We thank the staff of Department of Otolaryngology and the Radiation Oncology Unit, Department of Radiology, Chulalongkorn University Hospital, for recruiting nasopharyngeal carcinoma patients. We also thank Ju-Seog Lee, National Cancer Institute, NIH, for technical advice and Wichai Pornthanakasem and Sairoong Sakdikul for technical assistance.
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: J. Silvio Gutkind, Oral and Pharyngeal Cancer Branch, National Institute of Dental and Craniofacial Research, NIH, 30 Convent Drive, Building 30, Room 211, Bethesda, MD 20892-4330. Phone: (301) 496-6259; Fax: (301)402-0823; E-mail:
↵7 Internet address: http://linus.nci.nih.gov/BRB-ArrayTools.html.
- Received December 17, 2003.
- Revision received April 23, 2004.
- Accepted May 4, 2004.