Abstract
Background: In this work, gene expression profile was examined in 19 cases of oral cancer (OC) obtained from patients from Sweden (n=8) and UK (n=11) and the findings were tested for correlation to patient's clinicopathological data. Materials and Methods: Following total RNA extraction, cDNA synthesis, labeling with fluorescent dyes and hybridization to the 21 k human oligonucleotide microarrays, slides were scanned and images were subjected to Genepix and J-Express analysis. Results for selected genes were validated by quantitative reverse transcriptase polymerase chain reaction (Q-RT-PCR). Results: 42 genes were identified as being differentially expressed. These included 39 genes of known functions (such as fatty acid synthase (FASN), 5′ nucleotidase, ecto (NT5E), high mobility group AT-hook (HMGA1), and v-fos FBJ murine osteosarcoma viral oncogene homolog (FOS)) and 3 novel genes; 26 (67%) of the 39 genes with known functions were previously reported in oral/head and neck tumors examined from other populations. Hierarchical clustering of the samples using the 42 genes demonstrated that samples mainly clustered in the same population. Conclusion: These results illustrate that microarrays can be used to identify distinct patterns of gene expression in different populations, but with no direct association to clinicopathological parameters. The fact that 67% of the 39 genes with known functions found in this work were previously reported in oral/head and neck tumors from other populations provides clear evidence that development of these tumors follows the same biological pathways irrespective of the source of the samples used.
Oral squamous cell carcinoma (OSCC) constitutes a serious public health problem worldwide, with 400,000 new cases being reported annually (1), of which about two-thirds occur in developing countries. In Sweden, and in the United Kingdom, the relative frequency of OSCCs is about 1.4% (2) and 3% (3), respectively. Incidence of OSCC in Sweden has been relatively stable for the last 40 years. In the UK, however, there has been an alarming increase in the number of young patients diagnosed with this cancer since the 1970s (1). It is not clear why the OSCC frequency is increasing in these patients. The risk factors appear to be the same (4), but the survival among younger patients is better than that of older patients (5).
Cigarette smoking and high alcohol consumption are the main causative factors of OSCCs in the Western World (1). Socioeconomic status also seems to be an important confounding factor, as lifestyles are often influenced by individual social status (6). In Sweden, the use of oral snuff (snus) is very popular, but the association between use of snus and development of OSCC is debatable. Several cohort and case–control studies have examined the relationship between development of OSCC and snus use, with contradictory outcomes (7-11). Although there is agreement that the health risks related to use of snus are not fully understood, it has been reported that snus increases the risk of pancreatic and gastric cancer (12-14). In addition, an increased risk for OSCC development among snus users compared to non users was demonstrated in a recent study (8).
South Asian immigrants living in the UK have been shown to have a high risk for developing OSCC, perhaps due to the sustained habit of betel quid chewing after immigration (15).
Improvements in surgery, radiotherapy and chemotherapy have not affected the five-year survival rate of OSCC (35-50%) significantly over the last four decades. A better understanding of the molecular basis for development and progression of OSCC might improve our knowledge related to tumor classification, prediction of disease outcome and choice of treatment. Molecular therapy is an evolving field targeting specific molecules affecting cancer cells (16-18). The fact that OSCC infrequently affects patients without a history of alcohol and tobacco use also raises the need for in-depth molecular studies to search for genetic alterations that might explain the interindividual differences related to carcinogenic susceptibility.
Microarrays have now become a commonly used technique in cancer-related studies, allowing global gene expression profiling in tumor cells, and generating large amounts of useful information. Gene expression profiling in OSCC is increasing (19), with valuable information related to the molecular mechanisms possibly involved in OSCC development. This knowledge may be extremely valuable in clinical applications such as early diagnosis and therapeutic planning.
Our aim in this study was to examine the gene expression profile in OSCCs from two European populations, and to search for possible differences both within and between the samples examined, and to test for correlation of any differences with the clinicopathological data.
Materials and Methods
Patients. Samples of OSCCs were acquired from consecutive patients with previously untreated OSCCs from a South London population in the UK, and from Sweden. A total of 19 samples, 11 from the UK (average age 62.2, range 41-91, SD±17.96 years) and 8 from Sweden (average age 67.8, range 56-84, SD±9.90 years) were obtained. Tissue samples were stored in the tissue storage and RNA stabilization solution RNAlater™ (Ambion, Inc., Woodlands, TX, USA), and were transported to the Department of Biomedicine at the University of Bergen where they were stored at −20°C until RNA purification and microarray experiments.
All tumors were staged following the 1987 UICC staging system (20), and had their histopathological diagnosis confirmed by two of the Authors (SW/SOI) using either fresh frozen/or 10% formalin-fixed, paraffin-embedded tissue sections stained with hematoxylin and eosin (H&E). The tumors were histologically graded into high, moderate or poorly differentiated carcinomas. To exclude gene expression alterations because of stromal cell contamination, by analysis of the corresponding H&E-stained sections it was confirmed pathologically that each tumor specimen contained ≥70% tumor tissue. For all the patients, data on clinicopathological parameters were available (Table I).
Tissue samples and RNA extraction. Total RNA was extracted from all the samples using TRIzol® reagent (Gibco BRL, Carlsbad, CA, USA)/RNeasy Fibrous Tissue Kit (Qiagen Inc., Valencia, CA, USA) following the manufacturer's instructions. Quality and quantity of the extracted RNA were determined spectrophotometrically with a Beckman DU®530 Life Science Spectrophotometer (Beckman Coulter, Inc., Fullerton, CA, USA) and by an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA).
cDNA synthesis, hybridization and scanning. Synthesis and labeling of the cDNA was carried out using Fairplay Microarray Labeling Kit (Stratagene, La Jolla, CA, USA), following the manufacturer's instructions. Synthesized cDNA was labelled with Cy™3 (normal cDNA) and Cy™5 (tumor cDNA) monoreactive dyes (GE Healthcare Life Sciences, Little Chalfont, Buckinghamshire, UK), and samples were hybridized to the 21 k human oligonucleotide microarrays, containing 21,521 oligonucleotides (AROS Human oligo v2.0 set from OPERON; Operon Biotechnologies Inc., Huntsville, AL, USA), representing 21,521 human genes (UniGene clusters), printed on Corning Ultra GAPS slides at the Norwegian Microarray Consortium (www.mikromatrise.no). Labeled cDNA was hybridized on a Ventana Discovery® XT System (Ventana Medical Systems Inc., Tucson, AZ, USA) according to the manufacturer's protocols. Slides were scanned by Agilent DNA Microarray Scanner BA (Agilent Technologies) and the microarray data was stored as TIFF format images. The images were further analyzed with GenePix Pro v5.0 (Molecular Devices Corp., Sunnyvale, CA, USA) where bad spots, and spots not detected were flagged, and the final results containing all statistical values were stored as a GPR-file.
Statistical analysis. The 632/352 median intensity values obtained from the GenePix Pro were processed and merged into a gene expression matrix using J-Express Pro software package (version 2.7; www.molmine.com). A pre-processing of each array was performed by removing controls and spikes, as well as flagged and empty spots. Multiple spots for the same gene were combined and represented by the median intensity value. The global lowess normalization method was applied to each array individually. Genes with values missing in more than 40% of the patients were removed. Remaining missing values were imputed with KNN input (K=10), the data were scale normalized [interarray normalization as described by Yang et al. (21)], and finally genes showing little variation across the samples (SD less than 0.5) were removed. This resulted in log-ratios for 2,439 genes across the 19 samples studied.
Hierarchical clustering with distance matrix, based on Pearson correlation and average-linkage (WPGMA), was performed on the 2,439 genes to cluster patients with similar gene expression profiles. Significance analysis of microarrays (SAM) was performed with the 2,439 genes, using two classes (UK, Sweden; un-paired, with 1,000 permutations) to find genes differentially expressed between the Swedish and UK samples. The percentage of genes falsely identified as being differentially expressed, i.e. the false discovery rate (FDR), was set to less than 4%, resulting in a final total of 42 genes. The function of these 42 genes was identified using the Oligo Microarray Database from Operon (www.operon.com), searching the Human Genome Oligo Set V2, the Cancer Genome Anatomy Project (CGAP) database (http://cgap.nci.nih.gov) using the Gene Finder tool, and the GeneCards database (http://www.genecards.org/).
The J-Express Pro Gene Ontology (GO) tool was used to search for expression changes related to biological functions. A GO directed acyclic graph (DAG) mapped a selection of the top 100 SAM genes, and the genes in the GO terms was done to the number of the genes in a GO DAG based on the 2,439 genes left from the pre-processing step. The Fisher-Irwin exact test was used to calculate a p-value for all GO terms, using a p-value cut-off at 0.05. GO terms with less than 3 genes and an enrichment score (ratio of the relative frequency of genes from a GO term in the selected set to the relative frequency of genes from the same term in the full set) below 2 were removed.
Quantitative real-time RT-PCR. To validate gene expression profile for selected candidate genes, quantitative RT-PCR was performed for three genes: high mobility group AT-hook (HMGA1), nuclear pore complex interacting protein (NPIP) and poly(A) binding protein, cytoplasmic 1 (PABPC1). HMGA1 was chosen for its possible dual role in cancer. We selected PABPC1 based on its function and a high average fold-change value, while NPIP was chosen mainly because of its high fold-change value. Aliquots of the same RNA (1-2 μg) used for the microarray hybridization experiments were also used for synthesis of the cDNA, performed with High Capacity cDNA Reverse Transcription kit (Applied Biosystems), following the manufacturer's instructions. Real-time PCR was performed with probes for each gene (listed in Table II) using an ABI 7900 HT (Applied Biosystems) and 384-well optical plates (Applied Biosystems). Each reaction contained 1 μl cDNA, 5 μl 2X TaqMan Universal Mastermix (Applied Biosystems), 0.5 μl Taqman assay-on-demand (AOD) probe and water to a final volume of 10 μl, and was run in triplicate. Cycling parameters were 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. Serial diluted standards were run on the same plate and the relative standard curve method was used to calculate the gene expression as described elsewhere (22). β-Actin was used as endogenous normalization control to adjust for unequal amounts of RNA. For statistical analysis, we used GraphPad prism software (GraphPad Software Inc., La Jolla, CA, USA), and the Mann-Whitney U-test.
Results
Nineteen patients were included in this study, 11 from the UK and 8 from Sweden (Table I). All 19 tumors, except two from UK that were diagnosed with a verrucous carcinoma, were classified as SCC. There were 7 (36.8%) females and 12 (63.2%) males. A total of 12 (63%) out of 19 tumors were classified as tumors of stage 3 or 4, and 7 (37%) as tumors of stage 1 or 2. Among the 19 cases, there was one (5.3%) ex-smoker, 3 (15.8%) non-smokers and 13 (68.4%) smokers. Data on smoking habits was missing for two patients. Two of the Swedish patients were snus users, one was a non-user; data for the rest of the Swedish cases were not available. One UK patient used smokeless tobacco. Four patients were heavy alcohol drinkers, four were regular drinkers, four moderate and three were non-drinkers; data were missing for four of the Swedish cases.
Gene expression profile. Global gene expression profiles of the 19 tumors were obtained following hybridizing of the cDNA and human universal reference cDNA to the 21 k human oligonucleotide microarrays. Following grouping of the tumors by country, SAM analysis was performed, and 42 genes were found as being significantly differentially expressed between the samples from the two populations (Table III). Of the 42 genes found, 39 (93%) were of known function (including ribosomal protein S6 (RPS6), apolipoprotein L, 3 (APOL3), chitinase 3-like 1 (cartilage glycoprotein-39) (CHI3L1), PABPC1, NT5E, FASN and FOS, among others), while 3 (7%) were novel genes of unknown function. The biological processes of the known genes were related to signal transduction, cell–cell signaling, metastasis, cytoskeleton and transcription. Of the 39 known genes, 26 (67%) have been reported as being up- or down-regulated in oral/head and neck cancer, according to the Cancer Genome Anatomy Project (CGAP, http://cgap.nci.nih.gov/). These genes are listed in Table IV.
Based on the 2,439 genes found, we performed hierarchical clustering of the samples from the two countries and the results showed a tendency for most of the patients to group together based on country of origin (Figure 1). All subgroups included tumors of different stages. One case from the UK (UK10) and one case from Sweden (SW7) did not group with any of the other cases. Interestingly, UK10 is the only case among the UK patients with a history of smokeless tobacco use (in addition to smoking), and SW 7 is the only case with a distant metastasis. However, due to small sample size, no certain conclusions can be drawn from these observations.
We also performed a clustering after performing the SAM analysis, and the results also showed a tendency for the samples to group together based on country of origin (Figure 2), except for two Swedes (SW1 and SW9) that clustered with the cases from the UK. Although a distinct separation of the two populations was expected based on the SAM analysis, as this analysis searches for genes differentially expressed between the two populations, this was not the case in this study. This might suggest a similarity in the nature of these cases that might be related to similar biological pathways involved in development of the OSCCs from these two populations.
GO mapping was performed to search for alterations in gene expression related to gene function, but only one GO term fulfilled our criteria, that of isomerase activity.
Quantitative RT-PCR. Results of the quantitative RT-PCR confirmed the differential expression status of HMGA1, NPIP and PABPC1 (Figure 3) examined. P-values for all three genes was statistically significant (0.055 and lower).
Discussion
In this study, we examined the global gene expression profile of 19 cases of OSCCs from UK and Sweden, and found 42 genes to be significantly differentially expressed between the samples from the two nations. Among the genes found, 39 (including NT5E, APOL3, FASN, HMGA1, FOS, stanniocalcin (STC2), and protein arginine methyltranferase 1 (PRMT1) among others) were of known functions, encoding for proteins involved in tumor-promoting processes such as cell growth, metastasis, signal transduction and cell–cell signaling. Of these, 26 (67%) were previously reported as being differentially expressed in oral/head and neck cancer. Of interest, we found 3 novel genes of unknown function to be differentially expressed in the cases examined. Findings of the same 26 genes being previously reported in oral/head and neck tumors examined from other populations illustrates that the development of these tumors follows similar biological pathways irrespective of the source of the samples studied. Our results correlate well with other findings from Western countries (19, 23) and also with our previous studies that compared different populations (24, 25).
One of the genes verified by RT-PCR was HMGA1, high mobility group AT-hook 1. This gene encodes a protein with a role in tumor progression and metastasis. HMGA1 was selected for validation because of its function as a DNA-binding protein, with a role in transcriptional regulation of a number of genes (26, 27). It has a low expression level in most adult tissues, but is overexpressed in different malignancies (26, 27). Interestingly, HMGA1 has been suggested as a prognostic marker, showing increased expression not only in cancerous tissue, but also in premalignant lesions (28). In this work, PABPC1, poly A binding protein, cytoplasmic 1, has also been verified by RT-PCR. This protein comprises a part of a complex which binds to mRNA, initiates translation and causes mRNA circularization by binding both to the 5′-end and the 3′-poly(A)-tail of mRNA (29). In esophageal cancer, down-regulation of PABPC1 has been reported to be related to locally invasive tumors and poor prognosis (30). PABPC1 complex also includes proteins encoded by the genes EIF4E (cap-binding protein) and EIF4G1 (eukaryotic translation initiation factor 4 gamma, 1) (29). In our study, EIF4G1 was found to be differentially expressed between the samples from Sweden and UK. This gene, along with eukaryotic translation initiation factor 4B (EIF4B), also found to be differentially expressed in our study, is associated with tumor suppressor protein p53-induced inhibition of protein synthesis (31). EIF4B has been found as being downregulated in nasopharyngeal cancer, indicating a possible relation to p53 function in this cancer (32). We also validated expression of a third gene, NPIP, by RT-PCR. This gene, named nuclear pore complex interacting protein, has not been largely studied, and has no known relation to cancer. Although Hornan et al. (33) found NPIP to be overexpressed in the macula (a region of the human retina), the function of this gene remains unclear.
The gene NT5E, 5′-nucleotidase, ecto (also named CD73), catalyzes the dephosphorylation of ribo- and deoxyribo-nucleotide 5′-monophosphates, preferably adenosine 5′-monophosphate, a molecule suspected to have tumorigenic effects (34, 35). An up-regulation of NT5E has been found in several human malignancies (34, 35), including head and neck cancer (36). APOL3, apolipoprotein 3, a signal transducer and a member of the APOL gene family, was reported as being up-regulated in an earlier study that we performed with OSCCs from Sri Lanka (25). Horrevoets et al. (37) showed that APOL3 is activated by tumor necrosis factor alpha (TNF-α), a cytokine with both antitumor and tumor promoting properties (38). TNF-α has the ability to activate pathways involved in inflammation and proliferation, as well as apoptotic pathways, and is considered a potential therapeutic target (38). FASN, fatty acid synthase, is the main catalytic enzyme in de novo fatty acid biosynthesis (39). Over-expression of FASN has been reported in many human epithelial tumors, and the enzyme is considered as a possible therapeutic target (39). Up-regulation of FASN was shown in papillary thyroid carcinoma (40), in combination with activation of PI3K/AKT signaling, which is important for cancer development and progression.
Elevated levels of CHI3L1, cartilage glycoprotein-39, has been found in head and neck cancer and other malignancies, and is a marker of poor prognosis (41). The role of its gene in tumor development is poorly understood. Elevated levels of its protein expression has been found in non-malignant diseases (42). EPHX1 encodes the protein microsomal epoxide hydrolase, an enzyme involved in the metabolism of alkene and arene oxides, and in some cases it may activate and detoxify carcinogenic compounds (43, 44). It has been suggested that specific genotypes of EPHX1 represent an elevated risk of OSCC (43, 44). HBA1 (hemoglobin, alpha 1) (25, 45), and RPS6 (ribosomal protein S6) (46) have both been reported as being differentially expressed in head and neck cancer, but their functions in tumorigenesis need to be examined more closely. PRMT1, protein arginin methyltransferase 1, has also been reported in our earlier work (25). This gene is a histone-modifying enzyme, responsible for arginine methylation of histones, which affects chromatin structure and promote tumorigenesis (47).
FOS, v-fos FBJ murine osteosarcoma viral oncogene homolog, is also known as c-FOS or the transcription complex AP-1. FOS is an oncogene involved in numbers of tumor-promoting processes including cell proliferation and metastasis. Like HMGA1, FOS may have a dual role, though acting both as an oncogene and a tumor suppressor gene, with a possible pro-apoptotic function (48). In OSCCs, FOS appears to be involved in tumorigenesis, but its specific role is not yet known (49). Our findings of differential expression of FOS in the cases examined are supported by others (19).
In the current work, the genes described were expressed differentially in the tumors examined from Sweden compared to UK tumors. The global gene expression profiles found were more similar between the individuals from the same population. This may indicate that the mechanisms for OSCC development involve the same biological pathways in samples from different populations, although the sample size of our study is too small to draw any certain conclusions. However, similar results were previously reported by the same group when examining large samples from Norway and the Sudan (24). These findings indicate that further studies of the functions of these genes might reveal insights into their role in tumor development. We found two genes that are of special interest, HMGA1 and FOS, both suggested to have tumor-promoting as well as tumor-suppressing functions. Herein lies a possible explanation for the differential expression found in different tumors, and further studies of the function of these two genes in OSCC development is therefore necessary. When examining tobacco habits, most of the patients studied were smokers, though former use of smokeless tobacco by the Swedes cannot be discounted. Tumors were localized to the tongue, predominantly in the cases from UK. Other tumor sites (gingival, floor of mouth) were found in both sample groups. As tongue cancer is the most common site for intraoral cancer in the Western world, and the incidence of tongue cancer in UK and Sweden is fairly equal (50), the dominance of the tongue tumors among the UK patients probably is coincidental.
To conclude, this work identified 42 genes as being differentially expressed between the samples from the two countries. These genes included 39 known genes, of which 26 were previously reported in oral/head and neck cancer examined from other populations. The study illustrated that cDNA microarray can be used to identify distinct patterns of gene expression from different populations. The findings provide further evidence and support our previous findings indicating that development of OSCCs follows the same biological pathways irrespective of the source of the samples and the genes to be selected for analysis. Application of the cDNA microarray technology represents an advance in identification of genes associated with oral cancer, and may help in development of international detection and therapeutic strategies for this disease. A closer look at the functions of the genes described above would possibly reveal more information about the tumorigenesis, in particular the two genes with dual functions, HMGA1 and FOS. New findings could perhaps reveal if these genes are possible biomarkers for OSCC in the UK and Sweden.
Acknowledgements
The skilled technical assistance of Inger Ottesen and Dipak Sapkota is highly appreciated. We would like to thank Bjarte Dysvik and Inge Jonassen at the Computational Biology Unit, HIB, Bergen, Anne-Kristin Stavrum, Harald Breilid and Rita Holdhus at the Norwegian Microarray Consortium for their assistance.
Footnotes
- Received December 7, 2009.
- Revision received March 23, 2010.
- Accepted March 23, 2010.
- Copyright© 2010 International Institute of Anticancer Research (Dr. John G. Delinassios), All rights reserved