|
Sign In to gain access to subscriptions and/or personal tools.
|
DNA Hybridization Arrays for Gene Expression Analysis of Human Oral Cancer
R. Todd1,* and
D.T.W. Wong2
1 Laboratory of Oral & Maxillofacial Surgery, Harvard School of Dental Medicine, 188 Longwood Avenue, Boston, MA 02115, and Massachusetts General Hospital, 1 Fruit Street, Boston, MA 02114; and
2 Laboratory of Molecular Pathology, Division of Oral Pathology, Department of Oral Medicine and Diagnostic Sciences, Harvard School of Dental Medicine, Boston, MA 02115;
Correspondence: *corresponding author, 188 Longwood Avenue, Boston, MA 02115, randy_todd{at}hms.harvard.edu
 |
ABSTRACT
|
|---|
DNA hybridization arrays permit global gene expression profiling to be done in a single experiment. The evolution and challenges of DNA hybridization arrays are reflected in the variety of experimental platforms, probe composition, hybridization/signal detection methods, and bioinformatic interpretation. In tumor biology, DNA hybridization arrays are being used for gene/gene pathway discovery, diagnosis, and therapeutic design. Similar applications are advancing our understanding of oral cancer cell biology.
Key Words: microarray • gene expression • oral cancer
|
INTRODUCTION
|
|---|
DNA hybridization arrays permit the global analysis of gene expression in complex biologic systems. To date, this approach has been used to examine internal cellular events across different cell lines, physiologic responses to environmental changes in an intact organism, serial time points to a cell or organism, and gene expression associated with pathologic conditions. DNA hybridization arrays have been quickly adopted in cancer research for better understanding, diagnosis, and treatment of human malignancies. In this article, we will review the predominant methods of DNA hybridization array analysis, discuss the application of global gene expression analysis to the study of human cancers, and highlight head and neck/oral cancer investigations using this technology.
|
DNA HYBRIDIZATION ARRAY ANALYSIS
|
|---|
DNA hybridization arrays permit the simultaneous gene expression analysis of thousands of transcripts. Composed of gene-specific sequences immobilized at known sites on a solid-state matrix, DNA hybridization arrays are, in principle, similar to Northern or Southern blotting analysis. The sequence (or probes) is queried by labeled nucleic acids (target) from biological samples. Relative levels of the reporter-labeled target within the biological samples are detected and analyzed. While the overall strategy is similar, DNA hybridization arrays vary by format, probe composition, hybridization/signal detection methods, and bioinformatic interpretation.
Array Format
DNA hybridization arrays vary according to matrix, probe composition/density, array size, and type of label. Commonly used DNA hybridization array formats are: macroarrays, microarrays, and high-density oligonucleotide arrays. Macroarrays are nylon or nitrocellulose membrane-based arrays that contain cDNAs spotted at low density (typically between 200 and 5000 sequences). Radiolabeling is the most common reporter used in macroarray analysis, though chemiluminescence is also used (Fig. 1A ) (Rajeevan et al., 1999). Microarrays, arguably the most popular DNA hybridization arrays by virtue of their low cost and versatility, are either 2.5 x 7.5-cm glass- or plastic-slide-based (Fig. 1B) (DeRisi et al., 1996). The sequences (clones, PCR products, or oligonucleotides) are spotted at higher density, and a competitive, fluorescent dye scheme is used for detection. High-density oligonucleotide arrays use 25-mers photolithographically generated on a 1 x 1-cm silicon chip (Fig. 1C). Because of limited specificity and binding affinity, multiple oligonucleotides (and mismatch oligonucleotides) are spotted for each sequence. From 40,000 to 60,000 probes are on each array. Like microarrays, a fluorescent reporter is used for detection. Several alternative array formats exist. For example, microelectronic arrays consist of sets of electrodes covered by a thin layer of agarose coupled with an affinity moiety (Fig. 1D) (Edman et al., 1997). Probes are immobilized by means of biotin-avidin. Approximately 1000 genes are examined on each array. Each microelectrode is ~ 80 mm in diameter and is capable of generating an electric field that controls probe deposition and hybridization. Microelectronic arrays, along with several other formats, are one of several emerging DNA hybridization technologies.
View larger version (58K):
[in this window]
[in a new window]
|
Figure 1. Array platforms. (A) Macroarrays use nucleic acid probes deposited on membrane filters. Samples are usually radioactively labeled and hybridized in parallel. Detection is performed either through autoradiography or phosphorimaging. (B) Microarrays are formated on glass or plastic. Fluorescently labeled control and experimental samples are hybridized to the array in a competitive manner. Detection is performed by means of a fluorescent scanner. (C) High-density oligonucleotide arrays use photolithographically synthesized probes on a silicon matrix. Due to the limited length and specificity of the probes, a mismatch pair is added to determine specific hybridization. (D) Microelectronic arrays are an emerging technology that uses an electric field generated by individually controllable electrodes to immobilize probes and to control target hybridization. Washing is accomplished by reversing the electric fields. (Reproduced with permission from Freeman et al., 2000)
|
|
Probes and Target
Advances in probe design and target-labeling strategies have made DNA hybridization arrays possible (Ohyama et al., 2000). Probes, or the DNA sequences applied to the matrix, have been made increasingly available with genome sequencing and the databases that are being generated. Ideally, probes should be species-specific and gene-specific to avoid non-specific hybridization. Therefore, probes are best identified if constructed from regions of low-sequence conservation. Three classes of probes exist (Freeman et al., 2000). cDNA probes, ranging in size from hundreds to thousands of base pairs, are the least specific probes, because they are not designed for regions of low-sequence conservation. While cDNA probes have the greatest degree of non-specific hybridization, they are more flexible for analyzing cross-species target sample. Oligonucleotide probes, typically 25 bases in length, lack sequence specificity, thus requiring multiple oligonucleotide probes per gene sequence (as well as probes incorporating deliberate mismatches to control against mispriming). While high-density oligonucleotide arrays enjoy the highest specificity, they lack cross-species applications and are difficult and costly to manufacture. PCR product probes, ranging from 200 to 500 base pairs in length, share the strengths and weaknesses of both probe designs. Only macroarrays, microarrays, and microelectonic arrays use cDNAs, oligonucleotides, and PCR products as probes.
Target is generated from the biologic samples by means of an amplification step that also incorporates either a chemical or radiochemical reporter. These amplification steps include reverse transcription, amplified antisense RNA, tyramine signal amplification, and PCR (Luo et al., 1999; Freeman et al., 2000; Ohyama et al., 2000). However, signal amplification, while allowing for representation of genes with low copy numbers, must be tightly kinetically regulated to prevent gene representation alteration.
Hybridization and Signal Detection
Even distribution of target and wash solutions over the array is the crucial element of successful hybridization (Freeman et al., 2000). Targets must be evenly distributed over probes to facilitate the accurate representation of gene activity. Wash solutions must also be evenly distributed to remove non-hybridized target and reduce non-specific binding. Typically, macroarrays use traditional membrane methods as in Southern blotting. Array data inconsistency is frequently introduced by variations in these hybridization and washing steps using macroarrays. Microarrays and high-density oligonucleotide arrays use flow cells, a sealed chamber through which solutions can be directed. Microelectronic chips use a directed electric field in solution. Traditional membrane and flow cell hybridization methods have many inherent drawbacks (Freeman et al., 2000). Target-probe pairs have different melting temperatures and hybridization kinetics; therefore, conditions are determined by the probe hybridization mean. In addition, variations in target concentrations lead to variations in hybridization times and intensities. Electronic arrays actively concentrate target over the probes by virtue of DNA's negative charge. Reversal of the charge allows all but hydrogen-bonded target to be washed. Theoretically, this active, rather than passive, hybridization technique generates optimal conditions and reduced experimental times (Heller et al., 2000).
Accurate signal detection is imperative for gene expression studies (Luo et al., 1999). The two major detection methods in array experiments are radioisotopes and fluorescence. Radioisotopic detection is imaged by radiographic film or phosphorimaging. Though radiographic imaging is more accessible, phosphorimaging has a superior dynamic range, requires a shorter exposure time, and utilizes a direct digital output (Freeman et al., 2000). Fluorescence detection allows for competitive hybridization. Experimental and control samples are hybridized with separate fluorescent tags, allowing for detection of both samples in the same experiment by means of a confocal laser scanner. For example, if a control sample is labeled with a green dye and an experimental sample is labeled with a red dye, the relative amounts of each sample that hybridize to the array can determine the relative amounts of each sequence within both samples. Macroarray experiments typically use radioisotopic detection, while microarray and high-density oligonucleotide array experiments typically use fluorescent reporters.
Bioinformatics
Each DNA hybridization array experiment generates thousands of data points, and each study (containing many experiments) can result in millions of data points (Sherlock, 2000). Therefore, the interpretation and verification of array databases present a major challenge. While no universal algorithm exists for array data management, many experiments undergo a process of normalization, unsupervised analysis, and supervised analysis (Young, 2000). Initially, background hybridization is determined, followed by normalization of the signal. Background subtraction is typically performed based on either the area around each spot or spots with the lowest signal intensity (Freeman et al., 2000). To allow for comparison between different arrays/array experiments, normalization is performed by means of a selected housekeeping gene, an equilibration of the sum of intensities for all control and experimental spots, or application of an exogenous synthetic RNA standard.
Requiring no additional data outside the gene expression database itself, unsupervised analysis is commonly used for exploratory tasks, such as an unbiased discovery of gene expression patterns. Data grouped in an unsupervised analytic strategy are termed "clusters" (Tamayo et al., 1999). Several mathematical models exist for the clustering. Cluster algorithms group similar profiles based on a distance metric, usually by the statistical correlation coefficient or Euclidean distance (Freeman et al., 2000). The primary clustering strategies are hierarchical, K-means, and self-organizing maps (SOM) (Golub et al., 1999; Young, 2000). Hierarchical clustering begins by a calculation of the distance between individual datapoints, then a grouping of points by proximity. An organizational "tree" is formed, with gene relatedness reflected by the length of each "branch". While hierarchical clustering is easy to implement, the creation and order of branches can be very arbitrary. K-means clustering (k is the number of expected clusters) organizes gene sequences according to the proximity of the predetermined clusters. SOMs organize the clusters into a "map" based on relatedness (Toronen et al., 1999). Unlike K-means strategies, SOMs are recalculated based on genes within the cluster center, as well as adjacent clusters. To make algorithms run more quickly and analyses easier to understand, dimension reduction is performed to remove uninformative sequences (or those genes whose expression is the same regardless of condition). Dimension reduction in unsupervised analyses is commonly accomplished by either principal component analysis or independent component analysis.
Supervised analysis of DNA hybridization array databases requires that genes or conditions be identified by pre-existing classifications determined outside of the array database generation alone. This classification information "supervises" or drives the analysis of the experimental results. While data grouping in unsupervised strategies is termed "clustering", data groups from supervised analyses are termed "classification". Supervised analysis uses known groupings to create rules for reliably assigning genes to these groups. Therefore, examples for the phenotype or class being explored must have examples for which gene expression has been previously associated. Several mathematical models are used in supervised analysis of array databases, including logistic regression, neural networks, and linear discriminant analysis. Among the goals of dimension reduction in supervised analysis is the selection of relevant gene expression and a reduction in the number of genes in the discriminatory set sufficient for correct classification. Weighted analysis is performed by the removal of a single gene from the discriminatory data set (starting with the lowest weight) until the classification performance of the set drops off significantly (Golub et al., 1999). In short, while unsupervised analyses can find novel profile groupings, they often fail to reproduce groupings that are known from independent experimental approaches. Supervised analysis is a superior means of classifying gene expression; however, the success of supervised learning is highly dependent on the quality of data from previous experiments. Supervised analysis has been used to predict the function of previously undefined genes and categorizes gene expression associated with cell physiology/pathophysiology (Brown et al., 2000).
|
DNA HYBRIDIZATION ARRAY ANALYSIS OF HUMAN CANCERS
|
|---|
While the relationship between gene expression profiles and human cellular behavior is speculative at present, molecular "fingerprinting" is widely believed to have immense clinical promise for disease diagnostics and therapeutics (Young, 2000). In addition to DNA array platforms, global gene expression measurements are performed by several other methods, including library construction/direct sequencing, subtractive hybridization, differential display, and SAGE analysis (Ricci and El-Deiry, 2000). Cancer gene expression profiles have been performed on a variety of experimental systems, including human cell lines, animal models, and human tissues. From the studies published to date, three predominant strategies are emerging for DNA hybridization array-based experiments: specific, phenotype-associated gene expression, tumor classification by global molecular profiling, and therapeutic discovery (Ricci and El-Deiry, 2000).
Experimental Systems
The technical strengths and limitations of experimental systems for DNA array studies are similar to those of other experimental platforms. For example, cell lines are excellent for obtaining "starting material" pure populations and for single-gene manipulation; however, gene expression profiles in culture can have little resemblance to in vivo human gene expression, the gold standard. The use of human tissue, however, has many technical obstacles (Emmert-Buck et al., 2000). To date, the effects of excision and ischemic times on gene expression are not known. Human biopsy specimens "dilute" cancer tissue with normal neighboring tissue and inflammatory infiltrate. While techniques like laser capture microdissection allow for 93-100% purity of cell populations, the efficiency of recovery of a diverse and complex transcriptome has not been established (Fend and Raffeld, 2000). The large amount of target sample needed for array hybridization often requires an amplification step that may distort relative gene levels (Luo et al., 1999; Ohyama et al., 2000). However, as array databases become more common and standardized, assessment of experimental systems will become more reliable and accurate.
Phenotype-associated Gene Expression Profiles
Phenotype-focused DNA hybridization array experiments typically examine tumor-associated molecular events (for example, myc-induced gene expression) or cellular phenotype (for example, tumor invasion or metastasis) (Califano et al., 2000). While a wide range of array platforms is used for these experiments, the experimental systems predominantly used are cell lines and animal models. Performing oligonucleotide array gene expression analysis on fibroblasts ectopically expressing myc, Coller and co-workers identified 27 up-regulated and 9 repressed growth, cell cycle, signaling and adhesion sequences downstream of the proto-oncogene (Coller et al., 2000). Similar experiments have been performed for such genes as E2F1-3 (Muller et al., 2000), laminin-5 (Calaluce et al., 2001), p73a (Vikhanskaya et al., 2001), HPV E6/E7 (Nees et al., 2001), TNF-alpha, and Fas (Manos and Jones, 2001). Gene expression analyses that compare normal with tumor samples, or that use experimental systems mimicking a particular feature of tumorigenesis, are being more frequently performed (Manos and Jones, 2001). Using high-density oligonucleotide arrays, Clark and co-workers identified gene expression associated with metastases by comparing poorly metastatic with highly metastatic melanoma cells (Clark et al., 2000). Overexpression of RhoC or expression of a dominant-negative RhoC construct corresponded with the metastatic phenotype in both in vitro and in vivo assays. Differential gene expression profiles of gastric carcinoma cell lines with and without metastatic potential to the peritoneum or lymph nodes in a nude mouse model were identified by means of oligonucleotide arrays (Hippo et al., 2001). Similarly, a radiosensitivity gene profile was uncovered by cDNA microarrays comparing radiosensitive and radioresistant cervical carcinoma cell lines (Aschary et al., 2000). While sequences in expression profiles might be grouped by function, little is known about the majority of genes identified within each experiment. However, the functional relevance of identified genes is less important in the application of expression profiles for tumor classification.
Cancer Classification Gene Profiles
The majority of DNA array studies with human tissue attempt to classify cancers based on gene expression. Macroarrays, microarrays, and high-density oligonucleotide arrays have been used with various degrees of success. In a landmark paper, Golub and co-workers identified a gene set that can discriminate between forms of leukemia (Golub et al., 1999). Differential gene expression between acute myeloid leukemias and acute lymphoblastic leukemias was identified by means of high-density oligonucleotide arrays. The bioinformatic strategy based on unsupervised and supervised analysis to create a Class Prediction gene set outlines a means of discovering and predicting cancer classes based on a molecular profile. More recently, Khan and co-workers used gene expression profiling (with high-density oligonucleotide arrays) to separate a group of pediatric tumors (Khan et al., 2001). Ewing sarcomas, neuroblastomas, rhabdomyosarcomas, and non-Hodgkin's lymphomas all present as "small, round cell tumors" (He and Friend, 2001). However, treatment and prognosis of each are separate. Data from 88 tumors were analyzed by means of artificial neural networks to recognize and categorize complex gene patterns. Interestingly, genes currently used in tumor diagnosis were shown not to have a tumor-specific pattern. For example, the MIC2 antigen, used to diagnose Ewing sarcoma, also was expressed in several rhabdomyosarcoma samples. Molecular profiling has been performed on a variety of other solid tumors, including cancer of the breast, lung, colon, ovary, bladder, liver, and brain (Ricci and El-Diery, 2000). While total RNA requirements for DNA array hybridization vary according to platform, nucleic acid isolation from pure cell populations in histologically heterogeneous solid tumors remains an important technical challenge (Ricci and El-Diery, 2000). However, strategies are emerging to generate sufficient quantity and quality of starting material from laser-dissected cells for DNA array hybridization (Luo et al., 1999; Ohyama et al., 2000).
DNA Hybridization Arrays and Cancer Therapeutics
DNA array technology is beginning to enter cancer therapeutics research primarily through two strategies: biomarker identification and drug discovery. The previously described classification schemes hold a great deal of promise for improved diagnosis and treatment of cancer. Among the many possibilities are earlier diagnosis/better prediction of "pre-malignant" lesion transformation, identification of malignancy in an equivocal biopsy sample, subclassification of histologically identical phenotypes (presumably by differential responses to therapy), and more sensitive monitoring. In a recent report, Hughes and co-workers explore the application of molecular profiling to drug discovery. In a yeast experimental system, they first explored gene function by mutating the target gene and identifying the resultant change in gene expression (Hughes et al., 2000). Next, the gene profile of cells treated with a topical anesthetic, dyclonine, was obtained. By matching genetic permutation pathways with the pharmacologically induced pathways, they uncovered the mechanism of dyclonine. DNA array analysis will be important in several areas of drug development, including target identification, target validation, efficacy optimization, toxicity reduction, and identification of appropriate clinical trial participants (Young, 2000).
Early studies of drug resistance/sensitivity of human cancers to available chemotherapeutic agents by DNA array analysis have proved promising. By means of a macroarray-based approach, several genes, including IL-6, IL-8, and monocyte chemotactic protein-1, have been demonstrated to be differentially regulated in cancer cells resistant to paclitaxel (Duan et al., 1999). Butte and co-workers developed relevance networks to identify gene expression and chemotherapeutic susceptibility. Using the NCI60 (a set of 60 human cancer cell lines used by the National Cancer Institute Developmental Therapeutics Program), they correlated baseline gene expression with the gene expression following growth inhibition by thousands of anticancer agents (Butte et al., 2000). Through relevance networks, linking specific genes to other genes or phenotypes (like anticancer agents), strong associations were found between genes such as LCP1 and the anticancer agent NSC 624044, a thiazolidine carboxylic acid derivative.
|
DNA HYBRIDIZATION ARRAY ANALYSIS OF HUMAN ORAL CANCERS
|
|---|
Head and neck cancer is the tenth most common cancer in the United States, sixth most common cancer worldwide, and third most common cancer in developing nations (Parkin et al., 1999). Oral squamous cell carcinoma is the most common form of head and neck cancer. Clinical examination, biopsy (and H&E histology), and imaging for oral cancer diagnosis have shown little improvement in sensitivity and specificity. In spite of advances in surgery, radiation, and chemotherapy, five-year cancer survival rates have changed little over the past 40 years and remain among the worst of cancers of all anatomic sites. It is hoped that a better understanding of the molecular determinants of oral carcinogenesis will improve diagnosis, treatment, and monitoring of the disease.
While preliminary molecular progression models exist for oral carcinogenesis, the precise molecular targets and pathways remain unclear (Califano et al., 2000). Chromosomal structural alterations are associated with dysplasia (9p21, 3p21, 17p13), carcinoma in situ (11q13, 13q21, 14q31), and invasive carcinoma (4q26-28, 6p, 8p, 8q) (Califano et al., 1996; Mao, 1997). Recently, there has been an explosion of gene expression associated with oral carcinogenesis; however, few genes have been consistently identified or functionally implicated with the oral cancer phenotype. Certain genes have a stronger association with oral cancer development. Among these are p16ink4a, APC, and p53, with 70%, 50%, and 90% involvement, respectively, in head and neck cancers (Mao, 1997). All three are located in chromosomal regions linked to oral cancer development: 9p21 (p16ink4a), 5q21-22 (APC), and 17p13 (p53). Several molecular events, such as altered expression of DOC-1 and RAR-β, that have no obvious structural chromosomal change, have been associated with oral cancer development (Todd et al., in press). Likely, there are several genes/pathways yet to be identified as having a true causal relationship with oral cancer development. Identification of potentially novel diagnostic and therapeutic targets is now being addressed by DNA arrays.
To identify oral-cancer-associated gene expression, investigators have combined DNA hybridization arrays with other molecular techniques for examining differential gene expression. Chan and co-workers combined cDNA representational difference analysis (RDA) and a macroarray technique to characterize transformation-related genes in oral keratinocyte cell lines (Chang et al., 1998). Three hundred eighty-four differentially expressed gene fragments, detected by RDA, were arrayed on a filter: Sixty-nine genes (from 99 clones) were confirmed by hybridization of the arrays in triplicate. A similar strategy was used by Villeret and co-workers (Villaret et al., 2000). An oral-cancer-specific subtraction library derived from two oral cancer cell lines and six normal oral epithelial cell lines was arrayed on a glass slide. Fluorescent probes from normal and tumor tissues were used to hybridize the arrays. Nine known genes and 4 unknown genes were identified to be overexpressed in human oral cancers. However, neither approach used pure cell populations of normal and malignant oral keratinocytes in vivo.
Leethanakul and co-workers, using cDNA microarrays, have identified several differentiation and growth-related genes in head and neck cancers (Leethanakul et al., 2000). Normal and malignant keratinocytes were obtained from biopsy specimens by means of laser-capture microdissection (LCM) (Fig. 2). Total RNA from 5000 cells was reverse-transcribed and used for the generation of radiolabeled cDNA probes. A commercially available cDNA array, containing 588 cancer-related genes, was hybridized. Several expected and novel gene expression patterns were noted. Cytokeratin expression (2E, 2P, 6A-F, 7, 13-15, 17-19) was shown to be down-regulated fromm two- to 20-fold in malignant keratinocytes. However, cyclin D1, MMP-7,-10,-14, TGF- /β, PDGF, HGF, and VEGF-C were all elevated. Wnt and MAP kinase signal pathway genes, including ERK1, JNK isoforms, p38, and ERK6, were all elevated. Apoptosis inhibitors IAP and Akt 2 were also overexpressed in the malignant keratinocytes. No experiments confirming the gene expression findings were performed.
View larger version (119K):
[in this window]
[in a new window]
|
Figure 2. Laser-capture microdissection. (Panel A) Oral cancers commonly present as white or red-white mucosal lesions. They can be exophytic, endophytic, or a combination of the two. (Panel B) While advanced oral cancers can be several centimeters in size, many oral cancers present as lesions that are millimeters in diameter and are frequently overlooked. (Panel C) Like other cancers, oral cancers have heterogeneous cell populations in addition to the malignant epithelium (ME): connective tissues/fibroblasts (CT), vascular epithelium (VE), and acute/chronic inflammatory infiltrate (INFL). (Panel D) Laser-capture microdissection allows for the isolation and transfer of malignant oral epithelium (inset) for DNA, RNA, and protein studies. (Reproduced with permission from Todd et al., 2001)
|
|
Oral-cancer-related gene expression has also been investigated by high-density oligonucleotide arrays. Compared with other DNA hybridization arrays, high-density oligonucleotide arrays require higher amounts of target sample. Therefore, generation of target sample requires higher amounts of starting material or the addition of an amplification step. Ohyama and co-workers demonstrated the feasibility of generating target sample from LCM-procured tissues suitable for hybridizing high-density oligonucleotide arrays for gene expression profiling (Ohyama et al., 2000). RNA was successfully isolated by LCM from three paired cases of oral cancer and linearly amplified by T7 RNA polymerase. Hybridization of the samples to the HuGenFL GeneChip® probe arrays revealed that from 26.5 to 33.0% of the ~ 7000 represented genes are expressed in each of the six samples. These results demonstrate that LCM-generated tissues can generate sufficient quality cRNA for high-density oligonucleotide microarray analysis, an important step in the determination of comprehensive gene expression profiling by this high-throughput technology.
Using this optimized technology, Alevizos et al. (2001) examined 5 paired cases of oral cancer and analyzed them using three different bioinformatic tools (GeneChip® software, GeneCluster algorithm, and Matlab analysis). Collectively, the bioinformatic tools revealed that about 600 genes are associated with oral cancer. These oral-cancer-associated genes include oncogenes, tumor suppressors, transcription factors, xenobiotic enzymes, metastatic proteins, differentiation markers, and genes that have not been implicated in oral cancer. The database created provides a verifiable global profile of gene expression during oral carcinogenesis, revealing the potential role of known genes as well as genes that have not been previously implicated in oral cancer (Fig. 3).
View larger version (85K):
[in this window]
[in a new window]
|
Figure 3. Computational tools to display gene expression profile data of five-paired cases of oral cancer. (A) Scatter plot identifies outlying genes between normal and malignant oral keratinocytes. (B) Hierarchical cluster analysis done on intensity values standardized by dividing by root mean square. Cosine correlation of similarity coefficient and complete linkage clustering classified the samples as shown. Normal ‘A’; Tumor ‘B’. (C) Self-organizing map (SOM) group expressed genes into co-expressed clusters (GeneCluster). (D) Principal component analysis (PCA) identifies the most significant expression patterns in all the genes examined. (E, F) Screen shoot of GeneSpring displays of selected gene expression (E, β-actin; F, GRO1), comparing normal and tumor oral keratinocyte expression. Note that expression of β-actin is relatively similar between normal and tumor specimens as well as between samples. GRO1, on the other hand, is dramatically overexpressed in malignant oral keratinocytes.
|
|
DNA hybridization arrays have been used as a novel approach to the identification of gene clusters predictive of oral cancer radiation responsiveness (Hanna et al., 2001). For this study, radiation resistance was defined as < 40% decrease in tumor size at the end of 6 weeks of treatment with a total 60-70 Gy. Total RNA was harvested from homogenized radioresistant and radiation-sensitive oral squamous cell carcinomas. Radiolabeled cDNAs were used to hybridize commercially available DNA arrays containing 1187 tumor-related genes. Hierarchical cluster analysis of genes showed a three-fold or greater difference between tumor types. With a predictor set of 60 genes, the radiation responses of two subsequent cases were correctly predicted. While genes previously associated with radiation responsiveness, such as c-jun and XRCC1, were found within this predictor gene set, most of the identified sequences had no previous association. The authors plan to refine this gene expression profile to develop a more accurate means of identifying patients with non-responsive oral cancers.
|
CONCLUSIONS
|
|---|
DNA array technology and bioinformatics are rapidly evolving and becoming better able to address important biologic questions. Identification of molecular pathways responsible for the malignant phenotype will be a major contribution of this experimental approach. Molecular fingerprinting of the oral cancer cell will translate into earlier detection, better classification, more predictable response to treatment, and development of novel interventions.
|
ACKNOWLEDGMENTS
|
|---|
This work was supported by the National Institute of Dental and Craniofacial Research (NIDCR) grants R29 DE11983 (RT), K02 DE00456 (RT), P01 DE12467 (DTWW), and P30 DE11814 (DTWW), the Harvard University William F. Milton Fund (RT), and the Oral and Maxillofacial Surgery Foundation (RT).
Received for publication July 20, 2001.
Revision received November 29, 2001.
Accepted for publication December 12, 2001.
|
REFERENCES
|
|---|
- Alevizos I, Mahdevappa M, Ohyama H, Zhang X, Kohno Y, Colevas D, et al. (2001). Head and neck/oral cancer gene expression profiling assisted by laser capture microdissection and GeneChip analysis. Oncogene20:6196–6204.[CrossRef][Medline]
[Order article via Infotrieve]
- Aschary MP, Jaggernauth W, Gross E, Alfieri A, Klinger HP, Vikram B (2000). Cell lines from the same cervical carcinoma but with different radiosensitivities exhibit different cDNA microarray patterns of gene expression. Cytogenet Cell Genet91:39–43.[CrossRef][Medline]
[Order article via Infotrieve]
- Brown MP, Grundy WN, Lin D, Cristianini N, Sugnet CW, Furey TS, et al. (2000). Knowledge-based analysis of microarray gene expression data by using support vector machines. Proc Natl Acad Sci USA97:262–267.[Abstract/Free Full Text]
- Butte AJ, Tamayo P, Slonim D, Golub TR, Kohane IS (2000). Discovering functional relationships between RNA expression and chemotherapeutic susceptibility using relevance networks. Proc Natl Acad Sci USA97:12182–12186.[Abstract/Free Full Text]
- Calaluce R, Kunkel MW, Watts GS, Schmelz M, Hao J, Barrera J, et al. (2001). Lamin-5-mediated gene expression in human prostate carcinoma cells. Mol Carcinog30:119–129.[CrossRef][Medline]
[Order article via Infotrieve]
- Califano J, van der Riet P, Westra W, Nawroz H, Clayman G, Piantadosi S, et al. (1996). Genetic progression model for head and neck cancer: implications for field cancerization. Cancer Res56:2488–2492.[Abstract/Free Full Text]
- Califano A, Stolovitzky G, Tu Y (2000). Analysis of gene expression microarrays for phenotype classification. Proc Int Conf Intell Syst Mol Biol8:75–85.[Medline]
[Order article via Infotrieve]
- Chang DD, Park NH, Denny CT, Nelson SF, Pe M (1998). Characterization of transformation related genes in oral cancer cells. Oncogene16:1921–1930.[CrossRef][Medline]
[Order article via Infotrieve]
- Clark EA, Golub TR, Lander ES, Hynes RO (2000). Genomic analysis of metastasis reveals an essential role for RhoC. Nature406:532–535.[CrossRef][Medline]
[Order article via Infotrieve]
- Coller HA, Grandori C, Tamayo P, Colbert T, Lander ES, Eisenman RN, et al. (2000). Expression analysis with oligonucleotide microarrays reveals that Myc regulates genes involved in growth, cell cycle, signaling, and adhesion. Proc Natl Acad Sci USA97:3260–3265.[Abstract/Free Full Text]
- DeRisi J, Penland L, Brown PO, Bittner ML, Meltzer PS, Ray M, et al. (1996). Use of a cDNA microarray to analyze gene expression patterns in human cancer. Nat Genet14:457–460.[CrossRef][Medline]
[Order article via Infotrieve]
- Duan Z, Feller AJ, Penson RT, Chabner BA, Seiden MV (1999). Discovery of differentially expressed genes associated with paclitaxel resistance using cDNA array technology: analysis of interleukin (IL) 6, IL-8, and monocyte chemotactic protein 1 in the paclitaxel-resistant phenotype. Clin Cancer Res5:3445–3453.[Abstract/Free Full Text]
- Edman CF, Raymond DE, Wu DJ, Tu E, Sosnowski RG, Butler WF, et al. (1997). Electric field directed nucleic acid hybridization on microchips. Nucleic Acids Res25:4907–4914.[Abstract/Free Full Text]
- Emmert-Buck MR, Strausberg RL, Krizman DB, Bonaldo MF, Bonner RF, Bostwick DG, et al. (2000). Molecular profiling of clinical tissues specimens: feasibility and applications. J Mol Diagn2:60–66.[Free Full Text]
- Fend F, Raffeld M (2000). Laser capture microdissection in pathology. J Clin Pathol53:666–672.[Abstract/Free Full Text]
- Freeman WM, Robertson DJ, Vrana KE (2000). Fundamentals of DNA hybridization arrays for gene expression analysis. BioTechniques29:1042–1055.[Medline]
[Order article via Infotrieve]
- Golub TR, Slonin DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov CP, et al. (1999). Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science286:531–537.[Abstract/Free Full Text]
- Hanna E, Shrieve DC, Ratanatharathorn V, Xia X, Breau R, Suen J, et al. (2001). A novel alternative approach for prediction of radiation response of squamous cell carcinoma of the head and neck. Cancer Res61:2376–2380.[Abstract/Free Full Text]
- He YD, Friend SH (2001). Microarrays—the 21st century divining rod? Nat Med7:658–659.[CrossRef][Medline]
[Order article via Infotrieve]
- Heller MJ, Forster AH, Tu E (2000). Active microelectronic chip devices which utilize controlled electrophoretic fields for multiplex DNA hybridization and other genomic applications. Electrophoresis21:157–164.[CrossRef][Medline]
[Order article via Infotrieve]
- Hippo Y, Yashiro M, Ishii M, Taniguchi H, Tsutsumi S, Hirakawa K, et al. (2001). Differential gene expression profiles of scirrhous gastric cancer cells with high metastatic potential to peritoneum or lymph nodes. Cancer Res61:889–895.[Abstract/Free Full Text]
- Hughes TR, Marton MJ, Jones AR, Roberts CJ, Soughton R, Armour CD, et al. (2000). Functional discovery via a compendium of expression profiles. Cell102:109–126.[CrossRef][Medline]
[Order article via Infotrieve]
- Khan J, Wei JS, Ringner M, Saal LH, Ladanyi M, Westermann F, et al. (2001). Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks. Nat Med7:673–679.[CrossRef][Medline]
[Order article via Infotrieve]
- Leethanakul C, Patel V, Gillespie J, Pallente M, Ensley JF, Koontongkaew S, et al. (2000). Distinct pattern of expression of differentiation and growth-related genes in squamous cell carcinomas of the head and neck revealed by the use of laser capture microdissection and cDNA arrays. Oncogene19:3220–3224.[CrossRef][Medline]
[Order article via Infotrieve]
- Luo L, Salunga RC, Guo H, Bittner A, Joy KC, Galindo JE, et al. (1999). Gene expression profiles of laser-captured adjacent neuronal subtypes. Nat Med5:117–122.[CrossRef][Medline]
[Order article via Infotrieve]
- Manos EJ, Jones DA (2001). Assessment of tumor necrosis factor receptor and Fas signaling pathways by transcriptional profiling. Cancer Res61:433–438.[Abstract/Free Full Text]
- Mao L (1997). Leukoplakia: molecular understanding of pre-malignant lesions and implications for clinical management. Mol Med3:442–448.
- Muller H, Bracken AP, Vernell R, Moroni MC, Christians F, Grassilli E, et al. (2000). E2Fs regulate the expression of genes involved in differentiation, development, proliferation, and apoptosis. Genes Dev15:267–285.
- Nees M, Geoghegan JM, Hyman T, Frank S, Miller L, Woodworth CD (2001). Papillomavirus type 16 oncogenes downregulate expression of interferon-responsive genes and upregulate proliferation-associated and NF-kappa b-responsive genes in cervical keratinocytes. J Virol75:4283–4296.[Abstract/Free Full Text]
- Ohyama H, Zhang X, Kohno Y, Alevizos I, Posner M, Wong DT, et al. (2000). Laser capture microdissection-generated target sample for high-density oligonucleotide array hybridization. Biotechniques29:530–536.[Medline]
[Order article via Infotrieve]
- Parkin DM, Pisani P, Fernlay J (1999). Global cancer statistics. CA Cancer J Clin49:33–64.[Abstract/Free Full Text]
- Rajeevan MS, Dimulescu IM, Unger ER, Vernon SD (1999). Chemiluminescent analysis of gene expression on high-density filter arrays. J Histochem Cytochem47:337–342.[Abstract/Free Full Text]
- Ricci MS, El-Deiry WS (2000). Novel strategies for therapeutic design in molecular oncology using gene expression profiles. Curr Opin Mol Ther2:682–690.[Medline]
[Order article via Infotrieve]
- Sherlock G (2000). Analysis of large-scale gene expression. Curr Opin Immunol12:201–205.[CrossRef][Medline]
[Order article via Infotrieve]
- Tamayo P, Slonim D, Mesirov J, Zhu Q, Kitareewan S, Dmitrovsky E, et al. (1999). Interpreting patterns of gene expression with self-organizing maps: methods and application to hematopoietic differentiation. Proc Natl Acad Sci USA96:2907–2912.[Abstract/Free Full Text]
- Todd R, Gutkind JS, Shillitoe EJ, Wong DT (2001). Solid tumors: microarray analysis of oral cancers. In: Microarrays in cancer research. Wong DT, Todd R, Warrington J, editors. Natick, MA: Eaton Press (in press).
- Toronen P, Kolehmainen M, Wong C, Castren E (1999). Analysis of gene expression data using self-organizing maps. FEBS Lett451:142–146.[CrossRef][Medline]
[Order article via Infotrieve]
- Vikhanskaya F, Marchini S, Marabese M, Galliera E, Broggini M (2001). p73a overexpression is associated with resistance to treatment with DNA-damaging agents in a human ovarian cancer cell line. Cancer Res61:935–938.[Abstract/Free Full Text]
- Villaret DB, Wang T, Dillon D, Xu J, Sivam D, Cheever MA, et al. (2000). Identification of genes overexpressed in head and neck squamous cell carcinoma using a combination of complementary DNA subtraction and microarray analysis. Laryngoscope110:374–381.[CrossRef][Medline]
[Order article via Infotrieve]
- Young RA (2000). Biomedical discovery with DNA arrays. Cell102:9–15.[CrossRef][Medline]
[Order article via Infotrieve]
Journal of Dental Research, Vol. 81, No. 2,
89-97 (2002)
DOI: 10.1177/154405910208100202
 CiteULike  Complore  Connotea  Del.icio.us  Digg  Reddit  Technorati  Twitter What's this?
This article has been cited by other articles:
|
|
|
|
 
B. Dysvik, E. N. Vasstrand, R. Lovlie, O. A-A. Elgindi, K. W. Kross, H. J. Aarstad, A. Chr. Johannessen, I. Jonassen, and S. O. Ibrahim
Gene Expression Profiles of Head and Neck Carcinomas from Sudanese and Norwegian Patients Reveal Common Biological Pathways Regardless of Race and Lifestyle
Clin. Cancer Res.,
February 15, 2006;
12(4):
1109 - 1120.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
TOP
ABSTRACT
INTRODUCTION