Partitioning of 17 840 complementary DNA human complementary DNA clones into 9 distinct categories based on expression data for 9 patients with head and neck squamous cell carcinoma (HNSCC). Categories 1 and 9 are highlighted by thick boxes.
Representative immunostaining patterns of moesin in normal epithelium of the tongue (A), corresponding squamous carcinoma of the tongue (B), and head and neck squamous cell carcinoma metastatic to the lymph node (C) (hematoxylin-eosin, original magnification ×400).
Belbin TJ, Singh B, Smith RV, Socci ND, Wreesmann VB, Sanchez-Carbayo M, Masterson J, Patel S, Cordon-Cardo C, Prystowsky MB, Childs G. Molecular Profiling of Tumor Progression in Head and Neck Cancer. Arch Otolaryngol Head Neck Surg. 2005;131(1):10-18. doi:10.1001/archotol.131.1.10
Copyright 2005 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2005
To assess gene expression changes associated with tumor progression in patients with squamous cell carcinoma of the oral cavity.
A microarray containing 17 840 complementary DNA clones was used to measure gene expression changes associated with tumor progression in 9 patients with squamous cell carcinoma of the oral cavity. Samples were taken for analysis from the primary tumor, nodal metastasis, and “normal” mucosa from the patients’ oral cavity.
Tertiary care facility.
Nine patients with stage III or stage IV untreated oral cavity squamous cell carcinoma.
Our analysis to categorize genes based on their expression patterns has identified 140 genes that consistently increased in expression during progression from normal tissue to invasive tumor and subsequently to metastatic node (in at least 4 of the 9 cases studied). A similar list of 94 genes has been identified that decreased in expression during tumor progression and metastasis. We validated this gene discovery approach by selecting moesin (a member of the ezrin/radixin/moesin [ERM] family of cytoskeletal proteins) and one of the genes that consistently increased in expression during tumor progression for subsequent immunohistochemical analysis using a head and neck squamous cell carcinoma tissue array.
A distinct pattern of gene expression, with progressive up- or down-regulation of expression, is found during the progression from histologically normal tissue to primary carcinoma and to nodal metastasis.
Head and neck squamous cell carcinoma (HNSCC) represents the fifth most common malignancy worldwide, representing a major international health problem. The 5-year survival rate for this disease has improved only marginally over the past decade; as a result, over 30 100 cases and 7800 deaths occur annually in the United States.1 There are few molecular markers that can be reliably used in early detection or as indicators of prognosis. Several karyotypic or gross chromosomal aberrations have been investigated as markers of disease progression and/or outcome, including gains of 3q, 8q, 9q, 20q, 7p, 11q13, and 5p and losses of 3p, 9p, 21q, 5q, 13q, 18q, and 8p.2 In recent years, studies involving microarrays have identified specific genes whose expression has changed in HNSCC samples compared with normal tissue. Villaret et al,3 using arrays containing 985 genes, examined 16 HNSCC cases and identified 9 overexpressed genes. Using laser capture microdissection of HNSCC cells to measure the expression of 588 known cancer-related genes, Leethanakul et al4 demonstrated increased expression of genes related to the WNT and NOTCH signaling pathways, as well as a decrease in expression of differentiation markers such as cytokeratins. Another expression-profiling study using laser capture microdissection of normal and malignant oral epithelium identified about 600 differentially expressed genes with oligonucleotide arrays.5 Microarray studies of HNSCC have also focused on molecular classification of HNSCC. Hanna et al6 were able to identify 60 tumor-related genes from a complementary DNA (cDNA) microarray containing 1187 genes that could successfully predict the radiation response of tumor samples. Our preliminary study of molecular classification of HNSCC is the first use of microarray gene expression data to predict outcome in this disease.7
We report herein the use of microarrays containing 17 840 cDNA clones to monitor gene expression changes associated with tumor progression in 9 patients with HNSCC. Our analysis to categorize genes based on their expression patterns has identified 140 genes that consistently increase in expression during progression from normal tissue to invasive tumor and subsequently to metastatic node (in at least 4 of the 9 cases studied). A similar list of 94 genes has been identified that decrease in expression during tumor progression and metastasis.
We obtained institutional review board approval and informed patient consent from each patient before obtaining invasive tumor tissue. Matched normal surgical margin and metastatic lymph node tissue were also obtained at the time of surgery for HNSCC at Memorial Sloan-Kettering Cancer Center. These patients had stage III or stage IV untreated oral cavity squamous cell carcinoma (SCC) with nodal involvement. Only patients undergoing surgical treatment with curative intent, with a history of smoking and no prior nonsurgical treatment for head and neck cancer, were included in the study. To rule out gene expression alterations due to stromal cell contamination, we confirmed that each tumor specimen used in our study contained greater than 70% cancer cells by analysis of corresponding hematoxylin-eosin –stained sections. Tissue samples were flash-frozen in liquid nitrogen and stored at –80°C prior to RNA extraction. Tissues were homogenized in TRIzol reagent (Invitrogen/Life Technologies, Carlsbad, Calif) using a Brinkmann PT 10/35 Tissue Homogenizer (Brinkmann Instruments Inc, Westbury, NY) and total RNA extracted using TRIzol reagent following the protocol of the manufacturer. Total RNA was subsequently cleaned by column purification using an RNeasy Mini Total RNA purification kit (Qiagen, Valencia, Calif).
For microarray construction, a set of 17 840 sequence-verified human IMAGE (Integrated Molecular Analysis of Gene Expression) cDNA clones representing both known genes and expressed sequence tags were spotted onto polylysine-coated microscope slides using a custom robot designed and built at the Albert Einstein College of Medicine, Bronx, NY.8 Prior to hybridization, slides were preprocessed as described previously (http://microarray1k.aecom.yu.edu). One round of linear amplification of tissue messenger RNA was carried out using a modified T7 linear amplification protocol.9 Fluorescent labeling of probes was carried out by reverse transcription of amplified RNA with Superscript II reverse transcriptase (Invitrogen/Life Technologies) in a reaction containing 2.5mM deoxyadenosine triphosphate, deoxycytidine triphosphate, deoxyguanosine triphosphate, 1mM deoxythymidine triphosphate, 10 U of RNase Out (Invitrogen/Life Technologies), 4 μL of Cy-labeled deoxyuridine triphosphate (Amersham-Pharmacia, Piscataway, NJ) in a final volume of 40 μL. Reactions were carried out at 42°C for 2 hours. Hybridization of fluorescently labeled probes to cDNA arrays was carried out overnight at 50°C in a buffer containing 30% formamide, 3× saline sodium citrate, 0.75% sodium dodecyl sulfate (SDS), and 100 ng of human Cot-1 DNA. Following hybridization, slides are briefly washed with a solution of 1× SSC, 0.1% SDS, then washed for 20 minutes at room temperature in 0.2× saline sodium citrate, 0.1% SDS and 20 minutes at room temperature in 0.1× saline sodium citrate (without SDS). Slides are immediately dried as before and scanned using the GenePix 4000A microarray scanner (Axon Instruments, Foster City, Calif). Red (Cy5) and green (Cy3) signal intensities for each element on the array were calculated using GenePix Pro 3.0 software. This software gives an integrated intensity per spot for each channel in addition to an integrated background count. In all subsequent analysis, we used the mean background subtracted intensity for the 2 channels. For each spot, we calculated the mean intensity over the spot in the 2 channels and from this subtracted the median of the background intensity.
For each patient in our study, we compared gene expression patterns between histologically normal adjacent mucosa (Cy3-labeled sample) and invasive HNSCC tumor (Cy5-labeled sample). In a second series of array experiments, we compared invasive HNSCC tumor (Cy3-labeled sample) to metastatic lymph node (Cy5-labeled sample) from the same patients. The ratio of the fluorescence intensities of the 2 dyes therefore represented a measure of differential gene expression between the 2 tissue samples of interest.
During analysis, several inherent features of the expression data were considered. First, noise levels in microarray experiments were often intensity dependent with noise increasing as the average fluorescence intensity (A) decreased. Therefore, a given fluorescence ratio was less significant at lower than at higher intensities. To account for this effect, we computed an intensity- dependent noise factor. We used this factor to compute a weighted score for each log ratio in a manner similar to a statistical z score. A second problem encountered was that a given gene sometimes had a large expression ratio in 1 or 2 samples and changed only modestly (or not at all) in the remaining ones. To account for this, we transformed the weighted log ratio values as follows:
where p[x] is the probability distribution function for the normal (gaussian) distribution. This calculation bound the ratio measurement to lie between –1.0 and 1.0, thereby minimizing the scores of genes having a huge differential ratio in 1 or 2 samples and only modest changes in the remaining ones. We used this “bounded score” to determine genes that were either over or underexpressed (bounded score >0.5 or <–0.5, respectively) in each microarray experiment.
We partitioned our entire gene set according to their patterns of expression across the 9 patients (Figure 1). To do this, we evaluated the expression pattern for each gene and each patient individually. For a given patient, the initial categorization placed a gene in 1 of 3 categories, depending on whether the gene was overexpressed (bounded score >0.5), underexpressed (bounded score <–0.5), or unchanged in invasive HNSCC compared with normal adjacent mucosa. Having established the initial category, a second categorization placed a gene in 1 of 3 categories depending on whether this gene was overexpressed (bounded score >0.5), underexpressed (bounded score <–0.5), or unchanged when comparing lymph node metastasis with invasive HNSCC from the same patient. As a result, the gene was placed into 1 of 9 categories depending on its overall expression pattern in the 2 microarray experiments. This process was repeated for each patient individually until the gene was finally assigned to the category in which it most often appeared. This entire process was repeated for every gene represented on the array until all were assigned to 1 of 9 categories.
Using the algorithm described in the previous section, a gene that increased in expression in tumor tissue relative to normal tissue, and then increased again in metastatic node tissue relative to the original invasive tumor, was placed in category 1. Genes belonging to category 2 increased in expression in tumor tissue relative to corresponding normal mucosa, but did not increase further in proceeding to metastatic disease. Conversely, a gene that consecutively decreased in expression was placed in category 9. Genes belonging to category 8 decreased in expression in tumor tissue relative to corresponding normal mucosa, but did not decrease further in proceeding to metastatic disease. Categories 4 and 6 represented genes that did not alter expression when comparing normal mucosa with primary tumor tissue, but subsequently increased (category 4) or decreased (category 6) when comparing primary tumor tissue with metastatic lymph nodes. Genes that were unchanged, or for which data were discarded owing to low signal, were placed in category 5 by default.
Three different HNSCC tissue arrays were constructed for this study at Memorial Sloan-Kettering Cancer Center, New York, NY. Five-micrometer sections of normal and tumor tissues were embedded in paraffin and stained with hematoxylin-eosin to identify viable, morphologically representative areas of the specimen from which needle core samples could be taken. From each specimen, triplicate tissue cores with diameters of 0.6 mm were punched and arrayed onto a recipient paraffin block using a precision instrument (Beecher Instruments, Silver Spring, Md).10 Five-micrometer sections of these tissue array blocks were cut and placed onto microscope slides and used for immunohistochemical analysis. A total of 23 normal tongue epithelium, 30 dysplastic epithelial lesions from patients with HNSCC, and 42 HNSCC and 7 lymph node metastases were analyzed.
Standard immunoperoxidase procedures were used for immunohistochemical analysis of moesin expression. We used mouse monoclonal clone 38/87 at 1:50 (4 μg/mL) with microwave pretreatment of the slides (Neomarkers, Fremont, Calif); staining conditions were optimized on sections from formalin-fixed, paraffin-embedded tissue controls as specified by the manufacturer. Antibody reactivity was detected by using diaminobenzidine as chromogen, and sections were counterstained with hematoxylin. The primary antibody was omitted for negative controls. All specimens (n = 102) used for analysis of the association between moesin expression and progression of HNSCC were assessed using the nonparametric Wilcoxon–Mann-Whitney and Kruskall-Wallis tests.11 The consensus value of the 3 representative cores from each tumor sample arrayed was used for statistical analyses. Expression values were displayed as median and 25% to 75% interquartile ranges, and mean values were accompanied by 95% confidence intervals and range.
The focus of our analysis was to identify genes that were consistently underexpressed or overexpressed during HNSCC tumor progression. We were especially interested in identifying genes that showed a consistent pattern of expression across each patient’s 3 tissue samples. For example, a gene that consistently decreased its expression in experiments comparing “normal” adjacent mucosa with invasive tumor tissue and decreased expression again when comparing invasive tumor with metastatic lymph node tissue may represent a tumor suppressor gene or a potential target for therapeutic intervention. By combining the 2 data sets for each patient in our study, we were able to partition all genes into 9 categories based on their patterns of gene expression during tumor progression. Initial partitioning of these data was based simply on the category in which a given gene most often appeared. We initially identified 347 genes appearing in category 1 and 303 genes in category 9 (Figure 1). When these 2 data sets were reduced by insisting that a given gene appear in a single category at least 4 times, we identified 140 genes that consistently increased in expression during tumor progression. Among the category 9 data set, 94 genes consistently decreased in expression (at least 4 of 9 patients). Complete lists of genes belonging to categories 1 and 9, as well as associated expression data, are available from our laboratory Web site (http://microarray.info). Partial lists of genes included in these 2 data sets are listed in Table 1 and Table 2.
Categories 1 and 9 include genes originating from diverse functional categories and many already associated with neoplastic disease states. In category 1, we observed an increase in the expression of the ECM glycoprotein tenascin (TN) during tumor progression. Strong expression of TN in HNSCC was confirmed previously.12 The increase in the ECM protein osteopontin confirmed detection of the protein in premalignant and malignant lesions arising from oral epithelium.13 Although its function in carcinogenesis remains unclear, it had been demonstrated that tumor-derived osteopontin was able to inhibit macrophage function and enhance survival of SCC metastases.14 Also increased in expression during tumor progression was the gene coding for arachidonate 5-lipoxygenase, a key enzyme implicated in the production of eicosanoids, which are thought to play a role in tumor promotion, progression, and metastasis. Expression of lipoxygenase had been shown to be up-regulated in cancer of the colon, prostrate, and bladder.15- 17 Lipoxygenase inhibitors have also been demonstrated to have antiproliferative effects in several of these diseases.16,17 Although the mechanism is not well understood, it had been shown that induction of apoptosis in prostate cancer cells by selenium could be reversed by the metabolites of this enzyme, suggesting that 5-lipoxygenase may be a molecular target of selenium’s anticancer action.18 Another category 1 gene was Stanniocalcin-1 (STC1), a polypeptide hormone believed to be a regulator of mineral homeostasis. Enhanced expression of STC1 had been observed in tissues of hepatocellular carcinoma and colorectal cancer compared with corresponding cancer-free tissue, suggesting that it might be a useful molecular marker for detection of tumor cells.19 Expression of STC1 had also been observed in the hypoxic regions of human breast and colon cancer cells, an observation of interest in that it was known that the presence of hypoxic regions within solid tumors was associated with a more aggressive tumor phenotype and poorer prognosis.20 We also identified a subunit (gamma 2) of laminin 5, one of the principal components of anchoring filaments and keratinocyte adhesion. Interestingly, this specific subunit of laminin 5 was previously reported in the cytoplasm of budding carcinoma cells at the invasive front of oral SCC.21 Among other category 1 genes identified was moesin, a member of the ERM family of proteins that function as membrane-cytoskeletal linkers and play a role in the regulation of cell adhesion and cortical morphogenesis.22 Moesin had not been previously identified in HNSCC. However, immunohistochemical analysis of skin tumors revealed that the staining pattern for moesin varied among different types of skin tumors, with invasive SCC showing an intense and heterologous staining of the cytoplasm and cell membrane.23 Overexpression of another family member, ezrin, had been identified as being associated with metastasis in a murine model of osteosarcoma.24 High ezrin expression had also been shown to be involved in the process of invasion by endometrial cancer cells.25
A variety of genes decreased in expression during tumor progression. For example, it had been previously reported that the degree of keratinization appears to be a valuable parameter in identifying patients with oral SCC at risk for developing regional metastases to the neck.26 We observed a loss of keratins 4, 13, and 15 during tumor progression. Loss of keratin 4 and keratin 13 had been previously observed in tongue SCC, with the loss of keratin 13 being related to the invasive and metastatic ability of the cancer cell line.27,28 Most interesting is the loss of expression of occludin, one of the major structure-determining transmembrane proteins located at cellular tight junctions. Using an in vitro invasion assay with oral SCC cell lines, Shibata et al29 showed that inhibition of invasion by malotilate involved development of cell-cell adhesions and a dose-dependent increase in zonula occludin. The loss of occludin had also been observed in poorly differentiated gastrointestinal adenocarcinomas.30 Among other genes identified in category 9 (decreasing in expression) were 4 potential tumor suppressor genes: BRUSH-1, MXI1, oxidative stress-response (OSR1) gene, and occludin. BRUSH-1 was part of a multiprotein complex linking signal transduction to actin, thereby regulating cellular shape and motility. Furthermore, loss of BRUSH-1 expression in both breast cancer cell lines and invasive tumors identified it as a potential tumor suppressor.31MXI1 belongs to the Mad family of transcription factor proteins, which function as potent antagonists of Myc oncoproteins.32 It had been shown that MXI1 can suppress prostate tumor cell proliferation in vitro.33
We examined the expression of moesin in tissue arrays containing normal epithelium, as well as corresponding SCC of the tongue and lymph nodes. Whereas normal squamous epithelium of the tongue showed negative or positive basal layer staining for moesin, advanced SCC showed a strong positive staining in the membrane and cytoplasm of most tumor cells (Figure 2). These results agreed with the moesin staining patterns observed in invasive SCC of the skin.13 Interestingly, tumor cells invading lymph nodes also displayed a positive moesin staining. Both membrane and cytoplasmic expression of moesin increased when comparing normal epithelium (median percentage of positive cells, 10.0%) with dysplastic epithelium (median percentage of positive cells, 20.0%) and with tumor samples (median percentage of positive cells, 81.5%) (Table 3). Each of these comparisons reached a statistically significant association (Mann-Whitney test, P<.001). However, moesin expression was not increased significantly when comparing tumor samples and lymph node metastases (median percentage of positive cells, 90.0%; Mann-Whitney test, P = .82). All lymph nodes invaded by tumor cells displaying positive moesin staining showed overexpression of this protein in the HNSCC tumor. Overall, moesin expression was found to be significantly associated with HNSCC progression (Kruskall-Wallis test, P<.001). The prognostic utility of moesin expression had previously been reported in bladder tumors, where the membranous expression of the protein was found to be associated with the overall survival of patients with bladder cancer.34
With the use of microarrays containing 17 840 cDNA clones, we have observed that progression of HNSCC to metastatic disease is accompanied by a vast number of changes in gene expression. It is important to note that owing to the relatively small population of patients studied, it was not possible to assess the statistical significance (or lack thereof) of these gene changes. Furthermore, the fact that all patients involved in our study presented with late-stage disease made it impossible to assess any correlation to characteristics such as tumor stage. However, we have been able to identify many important gene changes that occur in this disease. Perhaps most interesting is the identification of genes, including the ERM protein moesin and 4 potential tumor suppressors that show a consistent pattern of expression during head and neck tumor progression. We are now analyzing the levels and cellular distribution of these and other proteins using tissue arrays containing HNSCC samples in the hopes of identifying potential prognostic markers for this disease.
Correspondence: Thomas J. Belbin, PhD, Department of Pathology, Albert Einstein College of Medicine, 1300 Morris Park Ave, Bronx, NY 10461 (email@example.com).
Submitted for Publication: April 28, 2004; accepted September 16, 2004.
Acknowledgment: We thank Thomas Harris, PhD, for critical review and helpful discussions during the preparation of the manuscript. We also thank the assistance of Aldo Massimi and the Albert Einstein College of Medicine Microarray Facility.