Genetic expression profiles and biologic pathway alterations in head and neck squamous cell carcinoma

Abstract

Head and neck squamous cell carcinoma (HNSCC) is associated with considerable mortality and morbidity and is a major public health concern worldwide. To date, > 20 studies incorporating DNA microarray analyses have examined genomewide genetic expression changes associated with the development of HNSCC. The authors identified published reports of genetic expression profiles of HNSCC by Medline database search. They performed a review of the reports to identify genes that have been found repeatedly to exhibit substantially altered expression in HNSCC. Genes with altered expression were subsequently examined in the context of defined biologic systems with the use of GenMapp 2.0 pathway analysis software. Genes most commonly found to exhibit altered expression were those encoding for cytoskeletal and extracellular matrix proteins, inflammatory mediators, proteins involved in epidermal differentiation, and cell adhesion molecules. Results of GenMapp 2.0 analysis suggested global down-regulation of genes that encode for ribosomal proteins and enzymes in the cholesterol biosynthesis pathway; and up-regulation of genes that encode for matrix metalloproteinases and genes that bear on the inflammatory response. The review indicated that there are several genes and pathways that exhibit substantially altered expression in cancerous versus noncancerous states across studies. Further investigation into the genomic, proteomic, and functional consequences of these gene expression alterations may provide insight into the pathophysiology of HNSCC. Cancer 2005. © 2005 American Cancer Society.

Head and neck squamous cell carcinoma (HNSCC) is the 5th most common cancer worldwide, and accounts for approximately 2.8% of all malignancies in the United States.1 In 2004, an estimated 38,530 Americans developed HNSCC and 11,060 died of this disease.2 Considerable advances in the surgical and medical treatments for HNSCC over the past two decades unfortunately have not improved overall disease outcomes.3 Local tumor recurrence affects approximately 60% of patients and metastases develop in 15–25% of patients.4 Fewer than 30% of patients with HNSCC experience ≥ 3 years of disease-free survival, and many suffer significant morbidity associated with impaired speech and swallowing.3

The TNM staging system that is most often used to classify patients with HNSCC is based on the clinical, radiologic, and pathologic examination of tumor specimens.5 This system does not adequately address the molecular heterogeneity of HNSCC tumors, the magnitude and importance of which have become increasingly evident. An ideal staging system would provide perfect prediction of survival. That is, all patients with a particular stage would survive to a particular point in time. However, as with most other solid tumors, the ability of staging to predict prognosis in HNSCC is limited because patients with tumors of the same clinicopathologic stage do not have the same disease progression, response to clinical treatments, rate of disease recurrence, and survival.6 For example, a patient with HNSCC diagnosed with early-stage disease has an approximately 60% chance of survival through 5 years (based on national data).7 Therefore, at diagnosis, patients with early-stage HNSCC cannot be informed with much precision whether they will be alive in 5 years. Similar problems exist for other stages when attempting to make survival predictions. A variety of successively more refined staging systems have been developed and proposed for HNSCC with the belief that the predictive ability of staging can be improved.6, 8–16 However, several recent studies by Groome et al. show that none of the staging systems currently used or proposed for HNSCC can account for even 30% of the variation in survival.16–18 In other words, > 70% of the variation in survival among patients with HNSCC is unaccounted for and remains to be explained. Clearly, the current classification scheme for HNSCC has room for significant improvement.

There are several clinical situations in which molecular characterization of HNSCC could greatly improve treatment choice, potentially improve survival, and minimize morbidity. For example, premalignant lesions of the oral cavity have a small but tangible risk of developing into frank squamous cell carcinoma. The presence of chromosome polysomies and other histopathologic features of these lesions have some predictive power in distinguishing at-risk lesions from innocent ones,19–21 but these features are not universally applicable. If specific gene expression profiles exist that distinguish which premalignant lesions are at high risk for carcinogenic progression versus those that are not, they could be used to improve diagnosis and direct therapies on a more individualized basis.

Molecular characterization of HNSCC could potentially also be applied to patient care in the context of cervical lymph node metastases. Patients with HNSCC may have occult neck metastases, depending on tumor size, location, and regional lymphatic drainage. Whereas tumors of the tongue and tongue base, floor of mouth, and hypopharynx have been associated with high rates of metastasis (up to 75%), tumors of the larynx, lower alveolar ridge, and buccal mucosa have lower risks (< 20%).22–27 Thus, the management of patients without clinically apparent neck lymph node metastasis (the “N0 neck”) ranges from conservative observation to surgery and/or radiotherapy. Patients with N0 tumors in locations associated with high rates of lymph node metastasis have been treated with either prophylactic radiotherapy or a neck dissection. However, currently, there is disagreement on how best to treat patients with small tumors in locations where the rate of cervical occult metastasis is less clear. If gene expression profiles associated with lymph node disease or metastatic potential (and subsequent mortality) existed in these tumors, patients could benefit from more accurate counseling and therapy.

Predicting and preventing tumor recurrence are additional challenges for physicians who treat patients with HNSCC—challenges that potentially could be met by determining the molecular features of neoplastic, preneoplastic, and recurrent head and neck lesions. It is difficult to find reports in the literature that distinguish between persistent, recurrent, or second primary tumors. The standard approach to assessing whether surgery has completely removed the primary invasive cancer — and thus providing the basis for whether subsequent clinical lesions are persistent or recurrent — is the evaluation of surgical margins to determine if they are histologically free of malignancy. If pathologic evaluation indicates that the resection contained histologically normal-appearing mucosa, tumor that presents later in follow-up is believed to be recurrent. However, studies of HNSCC by Brennan et al.28 and Califano et al.29 provide evidence that “clear” surgical margins are frequently abnormal at the molecular level: the margins contain the same p53 mutations or loss of heterozygosity (LOH) patterns found in the primary tumor. Therefore, clinical lesions that previously were classified as “recurrent” cancer might be persistent disease left behind at the time of surgical resection.28, 29 Under the concept of “field cancerization” proposed by Slaughter et al.,30 however, it is also possible that histologically normal margins may not contain cells that are clonal outgrowths of the original tumor, but instead comprise “primed” mucosa arising from a separate clone.31 In this instance, the molecular profile might be more similar to that of a preneoplastic oral lesion than to an invasive HNSCC tumor. By determining the gene expression profiles of normal, preneoplastic, and cancerous epithelium, surgeons may be able to characterize more accurately the malignant or premalignant status of surgical margins, and thus adjust their resection fields accordingly.

Molecular Studies of Head and Neck Squamous Cell Carcinoma

There has been great interest in finding specific genes whose expression patterns would correlate with, and eventually predict, relevant clinical outcomes.32, 33 For example, LOH at various genetic loci is believed to be associated with loss of tumor suppressor genes, as is microsatellite instability owing to nucleotide expansion or deletion. A current model of HNSCC progression associates loss of chromosomal arms 3p, 9p, and 17p with conversion from normal to dysplasia. Subsequent loss of 11q, 13q, and 14q is associated with progression to carcinoma in situ, with loss of 6p, 8p, 8q, and 4q seen in more advanced stages with invasion.29 Although the order in which these losses happen may not be clearly established, frequent LOH on chromosomes 3p and 9p has been reported in HNSCC in several reports.31, 34, 35

Allelic imbalance also results from oncogene amplification. Although loss of 11q13 heterozygosity occurs in approximately 29% of dysplastic head and neck lesions and in 61% of invasive HNSCC,29 amplifications of this chromosome region have been reported in up to 70% of primary HNSCC36 and are correlated with late stage, aneuploidy,37, 38 poor prognosis,39 tumor recurrence, and distant metastasis.40 Amplification of the CCND1 (cyclin D1) gene on chromosome 11q13 is believed to play an important etiologic role in these correlations.37, 41

Ultimately, allelic imbalance by LOH or amplification at one genetic locus has limited ability to predict outcomes for all HNSCC. Analysis of early lesions using a panel of microsatellite markers29, 42 has facilitated this task, but is still not the most streamlined approach for the high-throughput studies needed to correlate allelic imbalance with clinical outcomes.

Differential expression of a few individual genes or proteins in relation to HNSCC also has been studied. These include tumor suppressor genes, such as p53 and p16; oncogenes, such as ras, erb-B, and c-erbB-2; and cell surface proteins or adhesion molecules, such as heat shock protein (HSP70), integrins, and cathepsins. Although p53 mutation or overexpression is the most common event involved in human cell malignant transformation,43 mutations in the p53 gene have been identified in 21–75% of patients with HNSCC.44–48 Overexpression of wildtype p53 also has been detected in HNSCC.46, 49 However, a 5-year follow-up of patients with p53 overexpression in the surgical margins of their HNSCC tumors did not predict further malignant disease.50 Similarly, the presence or absence of p53 protein did not predict the outcome of oral dysplasia.51 Studies of H-ras, K-ras, and N-ras oncogenes showed overexpression of ≥ 1 ras oncogene in a fraction of HNSCC.52, 53 There was no consistent association between ras overexpression and histologic differentiation and TNM classification. Gene amplification and/or rearrangement of the erbB oncogene was observed in 13% of the HNSCC tumor specimens examined.54 Its overexpression was found in 67% of tumor specimens and was associated with poorer outcome (e.g., advanced clinical stage, lymph node disease, and death).54 Expression of c-erbB-2 was detected in > 50% of the 80 tumor specimens studied. This expression increased as the histologic grade increased and correlated with poor disease survival.55

Although all of these studies have contributed greatly to our current understanding of HNSCC, only a few studies have assessed the heterogeneity of marker levels or expression among tumors of different stage. Also, reports showing associations between survival and a particular marker did not control for stage and other known prognostic markers, such as life style factors and comorbidities. Given that these studies have had limited value in explaining the heterogeneous nature of HNSCC tumors and predicting disease outcome, it seems imperative to assess multiple markers and cellular pathways simultaneously, to optimize predictions of tumor progression and clinical outcomes.

DNA Microarrays

DNA microarrays are platforms on which several hundred or thousand oligonucleotides or cDNA of known genes and expressed sequence tags are printed.56, 57 Microarray technology allows simultaneous assessment of many gene transcripts at a time. Although epigenetic changes and translational and posttranslational modification events often influence cellular phenotype, large-scale genetic profiling may still provide a glimpse of cellular changes that occur during preneoplastic or neoplastic tumorigenesis. Genomic-scale expression profiles allow an investigator to evaluate global transcript variability in the context of specific biologic themes and pathways such as proliferation, carcinogen metabolism, apoptosis, and radiosensitivity. To this end, single-gene variability is minimized and expression “signatures” are generated. Tissue samples with similar profiles can be clustered on the basis of gene expression, or conversely, tissue samples believed to be similar can be analyzed for genetic expression variability. In the context of HNSCC, potential themes to be addressed with the help of microarray data include elucidating the progression of disease from dysplasia to cancer to metastasis on the molecular level; determining whether there are specific markers or profiles predictive for cancer or metastasis, and if so, whether such profiles also predict outcomes such as mortality; and identifying biologic pathways necessarily altered in HNSCC tumorigenesis, potentially illuminating novel therapeutic targets.

Investigators have recently begun to apply DNA microarray technology to the study of HNSCC.58–60 Although there are myriad advantages to using this technique, several weaknesses are also important to note. Perhaps the largest drawback of microarray-based examination of HNSCC tumors is the lack of well defined standards for their use, interpretation, and validation. Variability in tissue procurement, tumor cell isolation, RNA extraction, choice of array platform, and data normalization and analysis prevents rigorous comparison of results from one laboratory with those from others.

Surgical tissue collection and processing can significantly vary from site to site. Ideally, tissue specimens should be flash frozen immediately after surgical excision, and subsequently stored at −80 °C until they are processed for RNA extraction. This is primarily because of the significant effects that tissue ischemia can have on genetic expression profiles. These effects manifest in as little as 20 minutes of ischemia at room temperature, and are largely independent of overall RNA quality.61

The contribution of stromal cell “contamination“ to the genetic expression profiles of HNSCC tumor cells is an important consideration in designing DNA microarray studies. Some researchers have addressed this issue by estimating tumor cell content of samples before RNA extraction and microarray analysis. This is usually achieved by histologic inspection of tissue sections directly adjacent to experimental tumor samples, and by setting an arbitrary threshold value of percent tumor cell content each sample must minimally have to be used for subsequent DNA microarray analysis. Although this method is, at first glance, an efficient way to ensure a relatively homogenous population of cells, preliminary work in our laboratory suggests that estimation of tumor-to-stromal cell ratios in tumor samples by histologic evaluation of adjacent tissue sections is imprecise (data not shown). This is mainly due to the nature of HNSCC tumors, which are not simple, homogeneous masses with discrete edges, but rather complex, three-dimensional entities with varying degrees of spicular projections and interlaced stromal components, not to mention the potential clonal heterogeneity among tumor cells themselves. Estimation of percent tumor cell content within one or a few histologic sections of a tissue sample does not ensure the same percent tumor cell content within the entire sample. Nevertheless, estimated tumor-to-stromal cell ratios can be useful in establishing the simple presence or absence of cancer cells, and reiterating the pathologic diagnoses of experimental samples.

A method to isolate relatively homogenous populations of tumor cells from stromal components involves the use of laser capture microdissection (LCM). LCM is a technique by which specific regions (e.g., tumor cells) within a histologic tissue section are microscopically isolated from other tissue components (e.g., stromal cells) by special laser-equipped microscopes.62 Nearly homogenous populations of HNSCC tumor cells are thus able to be isolated for subsequent RNA extraction and microarray work. Truly tumor cell-specific genetic expression profiles can be obtained as a result. Conversely, stromal cell-specific expression profiles also may be generated by this method. The importance of the interaction between cancer cells and their surrounding stromal or nontumor components is well recognized in epithelial tumors and other human malignancies.63–68 Comprehensive genetic profiling of HNSCC, therefore, ought not to evaluate only cancer cell-specific gene expression, but stromal cell expression as well, to provide a more complete understanding of genetic expression alterations in HNSCC tumorigenesis.

Another consideration in microarray experimental planning is the choice of appropriate controls. This is of course dependent on the question being addressed. Investigation into genes differentially expressed between normal and cancerous tissue samples may designate matched or unmatched normal tissue samples as controls, in which a matched normal tissue sample is taken from the same patient from whom the experimental tumor sample was obtained and an unmatched normal tissue sample is obtained from patients without cancer. However, matched “normal” epithelium may confound interpretations of gene expression changes occurring in HNSCC tumorigenesis, given that histologically normal tissue samples may already harbor preneoplastic genetic changes that have yet to manifest phenotypically. Microarray studies investigating HNSCC tumor progression may compare expression profiles of tumor samples of different TNM classification, in which normal control tissue samples may or may not be employed. Ideally, tumor samples of different TNM classification would be otherwise matched for patient age, gender, smoking/drinking history, and other variables to minimize further confounding factors, although logistically this may not be easily accomplished. Fortunately, programs that can incorporate these variables as analytic covariates, such as GenPlus (http://www.enodar.com/technology6.htm), are now available.

In addition to the variability in choosing control tissue samples, the choice of experimental tissue samples is another potential source of study variability. Although HNSCC tumors are histologically similar, arising from epithelial cells within the mucosa lining of the upper aerodigestive tract, there are clear differences in the growth patterns, metastatic rates, and prognoses of tumors within different head and neck subsites. Whether these differences are partially due to differences in genetic expression profiles remains to be established clearly. The impact that anatomic subsite-specific genetic expression differences might have on comparison of HNSCC tumors can be reduced by evaluating separately tumor specimens from different subsites.

Perhaps, the most significant source of heterogeneity among studies of DNA microarray tumor profiling comes from the various methods of data generation and analysis. Microarray data analyses all consist of 2 basic steps: 1) the establishment of a normalized hybridization signal for each transcript on the array and 2) the subsequent statistical determination of which signals differ significantly between different arrays. For both of these distinct steps, there is ongoing debate about which methods are best. The details of data normalization and analysis are beyond the scope of the current review, and the reader is directed to other sources.69–72

The multivariable differences among different DNA microarray-based studies, together with the huge amount of data generated from microarray experiments and the inherent genetic heterogeneity of most solid tumors, create a situation in which a surfeit of data with no meaningful results is one potential outcome. Published data from such studies ought not to be viewed as endpoints of research endeavors, but rather as screening tools for identifying potential genes for validation and further investigation. Nevertheless, the ability to measure the expression of thousands of genes simultaneously gives researchers a powerful method to analyze global genetic events responsible for HNSCC progression.

Common Gene Alterations in Head and Neck Squamous Cell Carcinoma

Our search of the current literature revealed > 20 studies that incorporated DNA microarray analysis in the study of HNSCC (Table 1).73–99 There is considerable heterogeneity among these studies in terms of study design, number of samples, sites and stages of disease, ratio of tumor-to-stromal cells analyzed, choice of microarray platform, and validation of results by other laboratory methodologies. Some studies restricted their selection of tumor specimens to specific tumor sites, such as the oral cavity or hypopharynx, to reduce variability. Some studies employed separation of tumor cells from stromal cells by LCM, whereas others estimated the percentage of tumor cells within tumor samples before RNA extraction and analysis.

Table 1. HNSCC Studies Incorporating DNA Microarray Analysisa

Authors	Tissue samples	LCMb	Array platform	Validation
HNSCC: head and neck squamous cell carcinoma; LCM: laser capture microdissection; OC: oral cavity; LN: lymph node; Y: yes; N: no; NA: not applicable; qRT-PCR: quantitative real-time polymerase chain reaction; OP: oropharynx; L: larynx; HP: hypopharynx; TMA: tissue microarray; IHC: immunohistochemistry; H&N: head and neck; SCC: squamous cell carcinoma. a Summary of head and neck carcinoma genetic profiling literature. Studies were identified by Medline database search, and are listed alphabetically by the first author. The types of samples analyzed, microarray platforms used, array data validation methodologies, and whether laser capture microdissection was employed are listed for each study. b Minimum percentage of tumor cell content of samples in non-LCM studies is shown in parentheses.
Alevizos et al.73	5 primary OC (unknown LN status)5 matched normal OC controls	Y	Affymetrix HuGeneFL (∼7000)	qRT-PCR
Al Moustafa et al.74	4 primary, LN− OC 4 matched normal OC controls Samples made into cell lines	NA	Affymetrix (∼12,530)	RT-PCR Western blot
Belbin et al.75	12 primary, 5 recurrent 12 OC, 2 OP, 2 L, 1 HP 7 LN+, 10 LN− 1 normal human epithelial keratinocyte cell line (Bio Whittaker, NHEK-6168)	N (70%)	Custom cDNA microarray (∼9216)	None
Belbin et al.76	9 primary OC All LN+	N (70%)	Custom cDNA microarray (∼17,840)	TMA IHC
Chin et al.32	7 primary 5 OC, 2 OP 7 matched normal OC controls	N	Microarray Centre, Ontario Cancer Institute Human 19K cDNA microarray (∼13,131)	IHC
Chung et al.78	55 primary HNSCC, 5 recurrent 15 OC, 14 OP, 7 HP, 24 L 26 LN+, 14 LN−, 20 LN unknown Common reference sample consisting of a pool of total RNA derived from a randomly chosen subset of 30 of the HNSCC samples 3 unmatched normal controls	N	Agilent cDNA microarray (∼12,814)	IHC
Cromer et al.79	34 primary HP SCC 8 matched, 2 unmatched normal controls	N (70%)	Affymetrix HG-U95A (∼12,650)	qRT-PCR
El-Naggar et al.80	11 primary OC, 1 OP Tumors divided into 6 “conventional” SCC and 6 “subtypes,” including basaloid, papillary, sarcomatoid, and verrucous carcinomas. LN status not given 12 matched normal controls from farthest margin of resection	N (90%)	Research Genetics cDNA arrays GF200 (∼5184) or GF211 (∼4048)	IHC qRT-PCR
Ginos et al.81	41 HNSCC 18 OC, 4 OP, 1 HP, 1 L, 3 sinus 25 primary, 16 locally recurrent 19 LN+, 21 LN−, 1 LN unknown 13 unmatched normal controls	N (50%)	Affymetrix HG-U133A (∼14,500)	IHC qRT-PCR
Gonazlez et al.82	3 primary OC LN status not given 3 matched normal controls	N	1) Differential display RT-PCR 2) Incyte Genomics UniGEM V cDNA array (∼9350) 3) Research Genetics cDNA array GF204 (∼5184)	IHC RT-PCR
Ha et al.83	7 primary OC, 1 primary L8 H&N dysplastic lesionsLN status not given6 unmatched normal H&N controls from OC, OP, HP, and L5 matched normal controls	N (85%)	Affymetrix HG-U95A.v2 (∼12,650)	qRT-PCR
Hwang et al.84	5 primary OCLN status not given5 matched normal controls	Y	Affymetrix HuGeneFL (∼7000)	qRT-PCR
Irie et al.85	11 primary OC 4 LN+, 7 LN− 11 matched normal controls	Y	Clontech Human Cancer 1.2 Atlas cDNA array (∼1176)	None
Jeon et al.86	25 HNSCC cell lines 1 immortalized oral keratinocyte line Normal human oral keratinocytes control	NA	Incyte Genomics Human GEM2 cDNA array (∼9000)	qRT-PCR
Kuriakose et al.87	22 primary HNSCC 16 OC/OP, 4 L, 1 HP, 1 maxillary sinus 9 LN+, 13 LN− 22 matched normal controls	N	Affymetrix HG-U95Av2 (∼12,650)	qRT-PCR
Leethanakul et al.88	5 primary HNSCC 2 OC, 1 OP, 1 L, 1 tongue hyperplasia LN status not given 5 matched normal controls	Y	Clontech cDNA array (∼588)	None
Leethanakul et al.89	5 primary OC LN status not given 3 cDNA libraries from normal controls	Y	Custom oral cancer-specific cDNA microarray (∼384)	None
Mendez et al.90	19 primary, 7 recurrent OC 8 LN+, 18 LN− 2 premalignant lesions 18 matched and unmatched normal OC controls	N (60%)	Affymetrix Test-1 and HuGeneFL (∼7000 genes)	qRT-PCR
Moriya et al.91	2 human oral SCC cell lines LN status not given normal OC controls	NA	Custom in-house cDNA microarray containing 1423 independent clones from 817 genes from a stomach cancer library	RT-PCR
Nagata et al.92	15 primary OC 5 LN+, 10 LN− control: pooled mRNA from OC of 58 patients without cancer	N	Takara IntelliGene Human Cancer CHIP 2.1 cDNA array (557)	IHC qRT-PCR
O'Donnell et al.93	18 primary OC as an “initial set” 11 LN+, 7 LN− 4 primary OC as a “validation set” 3 LN+, 1 LN−	N	Affymetrix HG-U133A (∼14,500)	IHC qRT-PCR
Roepman et al.94	82 primary OC/OP to find predictor genes 45 LN+, 37 LN− 22 primary OC/OP as a “validation set” 10 LN+, 12 LN−	N (50%)	Qiagen Human Array-Ready Oligo set (version 2.0) (∼21,329)	None
Schmalbach et al.95	20 primary OC/OP 13 LN+, 7 LN− 4 OSSC cell lines 4 unmatched normal OC controls	N (70%)	Affymetrix HG-U95A.v2 (∼10,000)	IHC
Sok et al.96	7 primary, 1 recurrent, 1 secondary 5 OC, 2 HP, 1 L, 1 max sinus 4 LN+, 5 LN− 9 matched controls	N	Affymetrix HG-U95A (∼12,650)	None
Squire et al.97	5 primary tongue 2 tongue SCC cell lines 5 matched normal tongue controls	N	Clonetech Atlas Human cDNA Expression array (588)	None
Villaret et al.98	16 HNSCC from multiple sites 6 primary, 7 recurrent, 3 LN 6 LN+, 7 LN− 22 unmatched controls from multiple different organs (e.g., tonsil, soft palate, esophagus, skin, lung, small intestine, stomach, heart, brain, kidney)	N	Synteni cDNA microarray (∼985)	None
Warner et al.99	6 HNSCC cell lines 20 primary OC 13 LN+, 7 LN−	N (80%)	Microarray Centre, Ontario Cancer Institute Human 19K cDNA microarray (≈13,131)	qRT-PCR

Most of the studies primarily describe global changes in gene transcription that distinguish normal head and neck squamous epithelia from carcinoma.73–75, 77, 79–92, 96–98 A study from our laboratory90 also demonstrated hierarchical clustering of transcriptional profiles that distinguished preneoplastic versus cancerous epithelium. In that study, patients with verrucous leukoplakia and erythroplakia, both premalignant conditions, clustered with a higher degree of relatedness to oral squamous cell carcinoma (SCC) samples than to normal controls.90 This phenomenon was also observed in a study by Ha et al.,83 in which the authors examined 7 cases of HNSCC compared with 8 head and neck dysplastic lesions and 11 normal matched and unmatched controls. They found genes differentially expressed in 1) primary HNSCC versus unmatched normal controls, 2) premalignant lesions versus unmatched normal controls, and 3) primary HNSCC versus premalignant lesions.83 As was the case in our study, Ha et al. found that premalignant and malignant tissue samples showed closer association than normal and premalignant tissue samples, suggesting that changes in genetic expression may occur before the development of malignancy.

Different gene expression profiles distinguishing cervical metastatic disease from nonmetastatic disease were found in a few studies.76, 78, 79, 93–95, 99 Schmalbach et al.95 demonstrated that oral cavity and oropharyngeal SCC tumors with cervical lymph node metastases exhibited a different gene expression profile compared with tumors without associated metastatic disease and with unmatched normal oral mucosa. In a study by Chung et al.,78 tumors of the oropharynx, hypopharynx, and larynx clustered significantly according to metastatic cervical lymph node status. Warner et al. assessed 6 HNSCC cell lines and 20 primary oral cavity SCC tumor specimens and found that patients clustered into 2 groups by binary tree-like structured vector quantization (BTSVQ) analysis,100 with significant differences in cervical lymph node metastasis, as well as in patients' gender and tumor stage.99 Cromer et al.79 evaluated the gene expression profiles of 34 hypopharyngeal tumor specimens and identified a subset of genes that were associated with metastatic potential. O'Donnell et al.93 identified a 116-gene signature set that differentiated primary tumor specimens according to metastatic lymph node status, and showed that tumor specimens from lymph node metastases clustered together with lymph node-positive primaries. Belbin et al.76 evaluated genetic expression profiles of tumor progression in 9 patients with Stage III/IV oral SCC, and identified 140 genes whose expression increased, and 94 genes whose expression decreased, in disease progression from normal to invasive tumor to lymph node metastasis in ≥ 4 of the 9 patients. Finally, a study by Roepman et al.94 established a metastatic predictor set of 102 genes based on expression profiles from 82 HNSCC tumor specimens (45 metastatic and 37 nonmetastatic) of the oral cavity and oropharynx. The performance of this predictor set was dependent on tumor tissue specimen storage times, exhibiting improving performance with shorter storage times. When the predictor set was assessed among expression profiles of 22 independent tumor samples, all stored for < 5 years, lymph node status was correctly predicted in 86.4% of the tissue specimens.94 These findings, taken together, suggest that there might be a metastatic gene expression signature present in some primary tumors that predisposes them to metastasize.

Other data also suggest that distinct histologic subtypes exhibit different gene expression signatures. A small study comparing six HNSCC tumor specimens with conventional histology with six less common histologic subtypes (including basaloid, papillary, sarcomatoid, and verrucous HNSCC) demonstrated that certain genes were differentially expressed in the conventional HNSCC histologic type versus the less common phenotypic variants.80

Some of the reviewed studies correlated expression profiles with clinical outcomes. For example, Chung et al.78 identified tumor profiles that clustered into four groups, which exhibited significantly different rates of disease recurrence-free survival. Ginos et al.81 examined > 50 specimens from multiple sites and identified a set of genes with altered expression that clustered patients according to tumor recurrence. In a recent study of seven tumor specimens and their matched normal controls by Chin et al.,77 the authors found that elevated protein expression of one particular marker, osteonectin, was shown to be a powerful, independent predictor for short disease-free interval and poor overall survival.

A few studies investigated possible mechanisms of altered gene expression observed by DNA microarray analyses. Based on their results of comparative genome hybridization (CGH), Cromer et al.79 localized several genes that were overexpressed in hypopharyngeal SCC to chromosomal regions that were amplified, thereby providing one potential explanation for increased gene transcription. These results were in accordance with those from a study by Squire et al.,97 which combined DNA microarray analysis with CGH and spectral karyotyping to identify changes in gene expression and correlate these changes with chromosomal dosage and structure alterations.

When the results from all studies are evaluated together, we identified 84 genes with common alterations in transcriptional expression across multiple studies (Table 2). It is noteworthy that some of these changes, although identified in multiple reports, also may have been present in samples from studies in which they were not identified, simply due to the absence of probes for these genes in the various array platforms used. Arrays used by the different research groups contained probes for as few as 384 genes to as many as 21,329 genes (Table 1). Given the significant heterogeneity of experimental designs and microarray platforms used, we did not seek to determine which platforms contained probes for which genes, but rather, we listed the genes whose expression levels were reported in multiple studies to be significantly altered in HNSCC, irrespective of whether these changes occur early in the precancerous phase or late in metastases.

Table 2. Common Gene Alterations in Head and Neck Carcinomaa

Gene name		Unigene ID	Symbol	No. of studies
Up-regulated genes
a Genes found in multiple studies of head and neck squamous cell carcinoma tumor profiling are listed alphabetically and by up-regulation versus down-regulation. Unigene cluster ID and gene symbol are given for each gene. Genes with conflicting data from one or more studies are also listed, with the conflicting study or studies indicated. ID: identification; Hs.: xxxxx; kD: kilodalton; ECM: extracellular matrix.
Cadherin 11, type 2, OB-cadherin (osteoblast)		Hs.116471	CDH11	5
Cathepsin L		Hs.418123	CTSL	5
Chemokine (C-X-C motif) ligand 10		Hs.413924	CXCL10	6
Chemokine (C-X-C motif) ligand 1 (melanoma growth-stimulating activity, alpha)		Hs.789	CXCL1	5
Collagen, type I, alpha 1		Hs.172928	COL1A1	5
Collagen, type IV, alpha 1		Hs.17441	COL4A1	5
Collagen, type IV, alpha 2		Hs.508716	COL4A2	5
Collagen, type V, alpha 2		Hs.445827	COL5A2	8
Fibronectin 1		Hs.203717	FN1	11
Integrin, alpha 6		Hs.133397	ITGA6	7
Integrin, beta 4		Hs.370255	ITGB4	5
Interferon, alpha-inducible protein (clone IFI-6-16)		Hs.523847	G1P3	6
Interleukin-6 (interferon, beta 2)		Hs.512234	IL6	6
Interleukin-8		Hs.624	IL8	6
Matrix metalloproteinase 1 (interstitial collagenase)		Hs.83169	MMP1	8
Matrix metalloproteinase 3 (stromelysin 1, progelatinase)		Hs.375129	MMP3	6
Matrix metalloproteinase 10 (stromelysin 2)		Hs. 2258	MMP10	7
Matrix metalloproteinase 12 (macrophage elastase)		Hs.1695	MMP12	5
Microfibrillar-associated protein 2		Hs.389137	MFAP2	5
Parathyroid hormone-like hormone		Hs.89626	PTHLH	5
Periostin, osteoblast-specific factor		Hs.136348	POSTN	9
Profilin 2		Hs.91747	PFN2	5
ras homolog gene family, member C		Hs.502659	RHOC	5
Runt-related transcription factor 1 (acute myeloid leukemia 1; aml1 oncogene)		Hs.149261	RUNX1	5
Secreted phosphoprotein 1 (osteopontin, bone sialoprotein I, early T-lymphocyte activation 1)		Hs.313	SPP1	6
Secreted protein, acidic, cysteine rich (osteonectin)		Hs.111779	SPARC	6
Serine (or cysteine) proteinase inhibitor, clade E, member 1		Hs.414795	SERPINE1	6
Superoxide dismutase 2, mitochondrial		Hs.487046	SOD2	5
Tenascin C (hexabrachion)		Hs.143250	TNC	7
Thrombospondin 2		Hs.371147	THBS2	5
Transforming growth factor, beta-induced, 68 kD		Hs.369397	TGFBI	5
Trophoblast glycoprotein		Hs.82128	TPBG	5
Down-regulated genes
Absent in melanoma 1		Hs.486074	AIM1	6
Actin binding LIM protein 1		Hs.438236	ABLIM1	5
Aldehyde dehydrogenase 9 family, member A1		Hs.2533	ALDH9A1	6
Annexin A1		Hs.494173	ANXA1	7
BENE protein		Hs.185055	BENE	6
Carcinoembryonic antigen-related cell adhesion molecule 1		Hs.512682	CEACAM1	5
Clusterin		Hs.436657	CLU	5
Creatine kinase, mitochondrial 1 (ubiquitous)		Hs.425633	CKMT1	6
Cyclin G2		Hs.13291	CCNG2	5
Cystatin B		Hs.695	CSTB	5
Deoxyribonuclease I-like 3		Hs.476453	DNASE1L3	5
Dual specificity phosphatase 5		Hs.2128	DUSP5	5
Envoplakin		Hs.500635	EVPL	7
Epithelial membrane protein 1		Hs.436298	EMP1	8
ECM protein 1		Hs.81071	ECM1	8
Hydroxyprostaglandin dehydrogenase 15-(NAD)		Hs.77348	HPGD	5
Interleukin-1 receptor antagonist		Hs.81134	IL1RN	10
Keratin 13		Hs.463032	KRT13	11
Keratin 15		Hs.80342	KRT15	7
Keratin 4		Hs.371139	KRT4	11
KIAA0089 protein (GPD1L: glycerol-3-phosphate dehydrogenase 1-like)		Hs.82432	GPD1L	6
Mal, T-cell differentiation protein		Hs.80395	MAL	10
Monoamine oxidase B		Hs.46732	MAOB	5
Monoglyceride lipase		Hs.277035	MGLL	5
Neuromedin U		Hs.418367	NMU	6
Paired-like homeodomain transcription factor 1		Hs.84136	PITX1	5
Periplakin		Hs.192233	PPL	7
Sciellin		Hs.115166	SCEL	5
Serine (or cysteine) proteinase inhibitor, clade B (ovalbumin), member 1		Hs.381167	SERPINB1	6
Serine protease inhibitor, Kazal type 5		Hs.331555	SPINK5	6
Tetranectin (plasminogen binding protein)		Hs.476092	TNA	5
Transglutaminase 3 (E polypeptide, protein-glutamine-gamma-glutamyltransferase)		Hs.2022	TGM3	12
Zinc finger protein 185 (LIM domain)		Hs.16622	ZNF185	6
Genes with conflicting expression data	Unigene ID	Symbol	No. of studies	Conflicting studies
Carcinoembryonic antigen-related cell adhesion molecule 5	Hs.220529	CEACAM5	5 down, 1 up	Belbin et al.75
Collagen, type I, alpha 2	Hs.489142	COL1A2	9 up, 1 down	Moriya et al.91
Cystatin A (stefin A)	Hs.518198	CSTA	6 down, 1 up	El Naggar et al.80
Dual specificity phosphatase 6	Hs.298654	DUSP6	4 up, 1 down	Belbin et al.76
FAT tumor suppressor homolog 1 (Drosophila)	Hs.481371	FAT	6 up, 1 down	Jeon et al.86
Fibroblast activation protein, alpha	Hs.516493	FAP	4 up, 1 down	Mendez et al.90
Integrin, alpha 3 (antigen CD49C, alpha 3 subunit of VLA-3 receptor)	Hs.265829	ITGA3	5 up, 1 down	Al Moustafa et al.74
Integrin, beta 1 (fibronectin receptor, beta polypeptide, antigen CD29 includes MDF2, MSK12)	Hs.429052	ITGB1	4 up, 1 down	Jeon et al.86
Keratin 16 (focal nonepidermolytic palmoplantar keratoderma)	Hs.432448	KRT16	4 up, 2 down	Al Moustafa et al.,74 Jeon et al.86
Keratin 17	Hs.2785	KRT17	5 up, 2 down	Al Moustafa et al.,74 Jeon et al.86
Laminin, alpha 3	Hs.436367	LAMA3	4 up, 2 down	Al Moustafa et al.,74 Jeon et al.86
Laminin, beta 3	Hs.497636	LAMB3	5 up, 1 down	Jeon et al.86
Laminin, gamma 2	Hs.530509	LAMC2	6 up, 1 down	Jeon et al.86
Lumican	Hs.406475	LUM	4 up, 1 down	Moriya et al.91
Plasminogen activator, urokinase	Hs.77274	PLAU	10 up, 1 down	Jeon et al.86
Tumor necrosis factor, alpha-induced protein 3	Hs.211600	TNFAIP3	5 up, 1 down	Al Moustafa et al.74
TYRO3 protein tyrosine kinase	Hs.381282	TYRO3	4 down, 1 up	Leethanakul et al.88
v-fos FBJ murine osteosarcoma viral oncogene homolog	Hs.25647	FOS	3 down, 2 up	Ha et al.,83 Ginos et al.81
v-myc myelocytomatosis viral oncogene homolog (avian)	Hs.202453	MYC	4 up, 1 down	Al Moustafa et al.74

Among the genes that were identified in these studies to be significantly and commonly altered in HNSCC (Table 2), several have been reported to be involved with HNSCC. These include genes encoding for matrix metalloproteinases, integrins, collagens, interleukins 6 and 8, fibronectin, tenascin C, and cathepsin L.101–111 Several genes less well characterized for their role in HNSCC, such as cadherin 11, periostin, osteonectin, and transglutaminase 3, also exhibited altered expression in these studies. When we assessed the pooled results for gene ontologies using the Cancer Genome Anatomy Project (CGAP; http://cgap.nci.nih.gov), we found that genes encoding for extracellular matrix (ECM) and integral membrane proteins, proteins involved in epidermal development and differentiation, and cell adhesion molecules were most frequently identified to have altered expression in HNSCC (Table 3).

Table 3. Ontology of Genes Altered in Head and Neck Carcinomaa

Ontology	No. of genes identified	Genes
a Numbers and symbols of genes identified in Table 2 with commonly identified ontologies, based on the Cancer Genome Anatomy Project (http://cgap.nci.nih.gov).
Extracellular matrix	22	COL1A1, COL1A2, COL4A1, COL4A2, COL5A2, ECM1, FN1, LAMA3, LAMB3, LAMC2, LUM, MFAP2, MMP1, MMP3, MMP10, MMP12, POSTN, SPARC, SPP1, TGFB1, THBS2, TNC
Cell adhesion	20	CDH11, CEACAM1, CEACAM5, FAT, FN1, IL8, ITGA3, ITGA6, ITGB1, ITGB4, LAMA3, LAMB3, LAMC2, POSTN, SPP1, TGFB1, THBS2, TNC, TPBG, TYRO3
Integral to membrane	16	BENE, CDH11, CEACAM1, CEACAM5, EMP1, FAP, FAT, G1P3, ITGA3, ITGA6, ITGB1, ITGB4, MAL, MAOB, TPBG, TYRO3
Epidermal development/differentiation	12	COL1A1, EMP1, EVPL, KRT13, KRT15, KRT16, KRT17, LAMA3, LAMB3, LAMC2, PTHLH, SCEL, TGM3
Signal transduction	11	ANXA1, CXCL1, CXCL10, IL6, IL8, LAMA3, MAL, NMU, PLAU, RHOC, TYRO3
Ca^2+ binding	10	ANXA1, CDH11, DNASE1L3, FAT, MMP1, MMP3, MMP12, SPARC, TGM3, THBS2
Cytoskeleton/cytoskeleton organization	10	ABLIM1, CXCL1, EVPL, KRT4, KRT13, KRT15, KRT16, KRT17, PFN2, PPL
Cell growth/proliferation	9	CXCL1, CXCL10, EMP1, IL6, IL8, KRT16, MYC, PTHLH, TGFB1
Inflammation	9	ANXA1, CXCL1, CXCL10, FOS, IL1RN, IL8, MGLL, SPINK5, SPP1
Zn^2+ binding	9	ABLIM1, MMP1, MMP3, MMP10, MMP12, RUNX1, SCEL, TNFAIP3, ZNF185
Cell motility/chemotaxis	8	ANXA1, CXCL1, CXCL10, IL8, MMP12, PLAU, SPP1, TPBG
Proteolysis	7	CTSL, FAP, MMP1, MMP3, MMP10, MMP12, PLAU
Basement membrane	6	COL4A1, COL4A2, LAMA3, LAMB3, LAMC2, SPARC
DNA binding	6	DNASE1L3, FOS, MYC, PITX1, RUNX1, TNFAIP3
Skeletal development	6	CDH11, COL1A1, COL1A2, PITX1, POSTN, TNA
Transcriptional regulation	6	FOS, MYC, PITX1, RUNX1, SOD2, TNFAIP3
Redox activity	5	ALDH9A1, GPD1L, HPGD, MAOB, SOD2

Although a few studies identified genetic expression differences that grouped cases according to metastatic status,76, 78, 79, 93–95, 99 only 1 gene, transglutaminase 3, was common to ≥ 3 of these studies. This is, perhaps, indicative of the heterogeneity of experimental design and microarray platforms used in these studies, and highlights the importance of independent validation of DNA microarray data by alternative methods.

Considering the heterogeneity of the different studies we reviewed and the large volume of data, some contradictory findings in the pooled gene expression data were not surprising. In several instances, we found that multiple studies reported a significant up-regulation of a particular gene, whereas ≥ 1 study reported a down-regulation of the same gene, and vice versa (Table 2). Sometimes, these contradictory data could be explained by differences in experimental design. For example, studies by Al Moustafa et al.,74 Jeon et al.,86 and Moriya et al.91 examined genetic expression profiles of cell lines derived from HNSCC tumors. These cell lines were relatively devoid of stromal cells, and therefore exhibited different expression profiles compared with whole-tumor specimens containing both epithelial and stromal cells. Collagen type I alpha 2 (COL1A2), for example, is a gene primarily expressed by fibroblasts within the stroma. Nine of the studies we reviewed reported significant up-regulation of COL1A2 in HNSCC tumor specimens compared with normal tissue specimens, but Moriya et al.91 found COL1A2 to be down-regulated in HNSCC cell lines. The absence of stromal cells in HNSCC cell culture offers one possible explanation for these disparate results.

Gene Alterations and Biologic Pathways

We examined reported alterations in global genetic expression across all these studies in the context of specific biologic pathways using GenMapp 2.0 software (http://www.genmapp.org).112 Because standardized, quantitative expression data from all of the studies reviewed were not readily available, we simply assigned a value of +1 to genes reported to be up-regulated and −1 to those reported to be down-regulated in cancer. We found that several major biologic systems or pathways were globally altered in HNSCC, at least in this set of studies. Among these are some previously implicated in HNSCC pathogenesis, such as cell cycle control (Fig. 1A), matrix metalloproteinases, and inflammatory response systems.113–115 It is noteworthy that other biologic systems previously not clearly associated with HNSCC were found to exhibit significant alterations in cancerous versus normal states as well. The most striking example is the down-regulation of expression of genes encoding many of the cytoplasmic ribosomal proteins (Fig. 1B).73, 81, 83, 84, 89, 90, 93, 94 Only a few published studies have evaluated the role of cytoplasmic ribosomal proteins and carcinogenesis. Significant changes in the expression of several ribosomal proteins have been reported to occur in colorectal carcinoma.116 A study from our laboratory described a generalized down-regulation of ribosomal protein genes in cancer based on DNA microarray analyses of oral SCC tumors.90 In that study, 14 of the 75 significantly down-regulated genes found in oral SCC encode for ribosomal proteins.90 A recent paper by Sengpiel et al.117 also reported a significant decrease in mRNA levels of ribosomal protein S19 in HNSCC compared with normal epithelial cells.

Figure 1. (A) Global up-regulation of genes that mediate cell cycle were identified using GenMapp 2.0 software (http://www.genmapp.org). Genes labeled red were found to be up-regulated by one or more studies, and those labeled blue were found to be down-regulated. Three genes, CCND2 (cyclin D2), CDKN1A (cyclin-dependent kinase inhibitor 1A; p21; Cip1), and CDKN1B (cyclin-dependent kinase inhibitor 1B; p27; Kip1), were either up-regulated or down-regulated by multiple studies, as indicated by the blue borders around red labels. (B) Global down-regulation of genes encoding for cytoplasmic ribosomal proteins was identified by GenMapp 2.0 analysis. Six genes, encoding for cytoplasmic ribosomal proteins L4, L21, L37A, S6, S21, and S29, were found to be either up-regulated or down-regulated by multiple studies, as indicated by the red borders around blue labels. Dark bars: up-regulated; light bars: down-regulated; open bars: not found.

Based on studies in zebrafish, there is evidence that some ribosomal proteins may act as tumor suppressors.118 Given that many mammalian tumor suppressors are necessary for normal embryonic development,119 Amsterdam et al.118 assessed several mutant zebrafish lines, all heterozygous for recessive embryonic lethal mutations, to see if these mutations conferred an increased risk for cancer development. Among the 12 mutant zebrafish lines exhibiting moderate to very high frequencies of cancer incidence, 11 were heterozygous for a mutation in a different ribosomal protein gene, suggesting that ribosomal genes may behave as haploinsufficient tumor suppressors in zebrafish. Whether ribosomal proteins act as tumor suppressors in HNSCC tumorigenesis in humans is unknown. Given that a large number of ribosomal protein genes are down-regulated in HNSCC, the significance of such global down-regulation in HNSCC warrants further investigation.

Another pathway we found to be dysregulated was that for mevalonate metabolism. Components of this pathway include products that are important for various cellular functions, including those that affect cell cycle progression, membrane integrity, and protein synthesis.120 In addition to the de novo synthesis of cholesterol, the pathway is responsible for generating other key cellular compounds such as ubiquinone and dolichol, which are essential components of mitochondrial respiration and N-glycosylation of proteins, respectively.120 Posttranslational modifications of several small G proteins, such as Ras and Rho, by isoprenylation also utilize products of the mevalonate pathway.121 We found that many genes of the mevalonate pathway were down-regulated in HNSCC in the reviewed studies, including the gene encoding for 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase, which catalyzes the rate-limiting conversion of HMG-CoA to mevalonate. Transcriptional down-regulation of the mevalonate pathway, if correlated with a decrease in the levels of encoded proteins, would be expected to lead to enhanced apoptosis, potentially via the activity of RhoA guanosine triphosphase (GTPase).122 This proapoptotic phenomenon, if truly present in HNSCC tumors, might represent a protective response to counter the uncontrolled cell proliferation characteristic of malignancy.

Conclusions

We reviewed the head and neck squamous cell carcinoma genetic expression profiling literature and identified 24 studies that incorporated DNA microarray analysis (Table 1). From the pooled gene expression data, we identified a group of genes reported by multiple studies to be significantly up-regulated or down-regulated in HNSCC. When we evaluated ontologies of these genes via the CGAP, we found that genes associated with cell adhesion, ECM organization, epidermal development, and membrane integrity were highly represented. When evaluated in the context of defined biologic systems with the use of GenMapp 2.0 software, results of theses studies showed that many genes encoding for cell cycle regulating proteins, matrix metalloproteinases, inflammatory response mediators, enzymes of the mevalonate pathway, or ribosomal proteins exhibited global, simultaneous up-regulation or down-regulation.

Although these findings may help focus selection of markers for further analysis for their value in the understanding and management of HNSCC, there are limitations to the utility of the list of genes and pathways presented in the current study. The weakness of the current status of HNSCC tumor profiling stems primarily from limited sample sizes and heterogeneity in experimental design and execution. The results presented herein may or may not hold true when results from studies of much larger sample size comprising tumor specimens of more uniform characteristics become available. In addition, validation of genetic expression changes found in these studies by alternative methods was performed in some but not all studies. For our review, we did not have access to each study's raw expression data, and therefore did not take into account how different fold changes in gene expression might impact different clinical outcomes or biologic pathways. Ideally, all published studies containing DNA microarray analyses would have raw expression data available in the public domain, to permit a systematic, comprehensive metaanalysis to be performed.