Human Cancer Biology

Identification of a Gene Signature for Rapid Screening of Oral Squamous Cell Carcinoma Amy F. Ziober,1Kirtesh R. Patel,1Faizan Alawi,3 Phyllis Gimotty,4 Randall S. Weber,6 Michael M. Feldman,2 Ara A. Chalian,1Gregory S. Weinstein,1Jennifer Hunt,5 and Barry L. Ziober1

Abstract Purpose: Oral cancer is a major health problem worldwide and in the U.S. The 5-year survival rate for oral cancer has not improved significantly over the past 20 years and remains at f50%. Patients diagnosed at an early stage of the disease typically have an 80% chance for cure and functional outcome, however, most patients are identified when the cancer is advanced. Thus, a convenient and an accurate way to detect oral cancer early will decrease patient morbidity and mortality. The ability to noninvasively monitor oral cancer onset, progression, and treatment outcomes requires two prerequisites: identification of specific biomarkers for oral cancers as well as noninvasive access to and monitoring of these biomarkers that could be conducted at the point of care (i.e., practitioner’s or dentist’s office) by minimally trained personnel. Experimental Design: Here, we show that DNA microarray gene expression profiling of matched tumor and normal specimens can identify distinct anatomic site expression patterns and a highly significant gene signature distinguishing normal from oral squamous cell carcinoma (OSCC) tissue. Results: Using a supervised learning algorithm, we generated a 25-gene signature for OSCC that can classify normal and OSCC specimens. This 25-gene molecular predictor was 96% accurate on cross-validation, averaging 87% accuracy using three independent validation test sets and failing to predict non ^ oral tumors. Conclusion: Identification and validation of this tissue-specific 25-gene molecular predictor in this report is our first step towards developing a new, noninvasive, microfluidic-based diagnostic technology for mass screening, diagnosis, and treatment of pre-OSCC and OSCC.

Head and neck cancers are the sixth most common cancer cell carcinoma (OSCC) remains at f50% (2–4). In addition, worldwide and are associated with low survival and high aggressive treatment of OSCC cancer is controversial because it morbidity (1). Cancers of the oral cavity account for 40% of can lead to severe disfigurement and morbidity (5). As a result, head and neck cancers and include squamous cell carcinomas many patients with OSCC cancers are either overtreated or of the tongue, floor of the mouth, buccal mucosa, lips, hard undertreated, with significant personal and socioeconomic and soft palate, and gingiva (2, 3). Despite therapeutic and effects. diagnostic advances, the 5-year survival rate for oral squamous One of the fundamental factors accounting for the poor outcome of patients with OSCC is that a great proportion of oral cancers are diagnosed at advanced stages, and therefore, treated late. Early detection of oral cancer lesions will greatly 1 Authors’Affiliations: Departments of Otorhinolaryngology-Head and Neck improve morbidity associated with late disease treatment and Surgery, 2Pathology and Laboratory Medicine, 3Department of Pathology, School 4 overall patient survival. For example, early detection could lead of Dental Medicine, and Department of Biostatistics and Epidemiology, Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania Health System, to frequent patient monitoring, dietary changes, counseling on Philadelphia, Pennsylvania; 5Department of Pathology, University of Pittsburgh and cessation of smoking and drinking, preventative drug School of Medicine, Pittsburgh, Pennsylvania; and 6Department of Head and Neck administration, and/or lesion removal. As such, early diagnosis Surgery, University of Texas M.D. Anderson Cancer Center, Houston, Texas and treatment of OSCC has been shown to lead to mean Received 3/6/06; revised 5/21/06; accepted 5/25/06. Grant support: NIH grants DE15856-01and DE015626-01(B.L. Ziober). survival of >80% and a good life quality after treatment (6). The costs of publication of this article were defrayed in part by the payment of page However, no methodology exists that could early, accurately, charges. This article must therefore be hereby marked advertisement in accordance and easily mass screen for oral cancer lesions. with 18 U.S.C. Section 1734 solely to indicate this fact. Currently, clinical examination and histopathologic studies Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). are the standard diagnostic methods used to ascertain whether Requests for reprints: Barry L. Ziober, Molecular Tumor Biology Laboratory, biopsied material are precancerous or cancerous lesions (7). Department of Otorhinolaryngology-Head and Neck Surgery, University of Biopsies are invasive procedures typically involving surgical Pennsylvania Health System, 5 Silverstein/Ravdin, 3400 Spruce Street, techniques. Furthermore, biopsies are limited when it comes to Philadelphia, PA 19104. Phone: 215-898-0075; E-mail: bziober@ mail. med.upenn.edu. lesion size. For example, small lesions may not provide enough F 2006 American Association for Cancer Research. material for accurate diagnosis, whereas biopsies taken from doi:10.1158/1078-0432.CCR-06-0535 large lesions may not accurately reflect every histopathologic

Clin Cancer Res 2006;12(20) October 15, 2006 5960 www.aacrjournals.org Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of OSCC aspect of the lesion. Finally, the biopsy, as a diagnostic tool, has cells will require easy access to the site in which these cancers limited sensitivity. Thus, additional methodologies would be typically arise, as well as a readily available source of cells. Oral helpful in screening for premalignant and malignant oral cavity saliva meets both of these criteria. A fundamental cancer lesions. However, to noninvasively detect oral cancer requirement to identify cancer lesions early, accurately, and

Fig. 1. Gene expression profiles of 13 normal oral mucosal and 13 patient-paired OSCC specimens shows distinct separation. A total 2,207 genes were identified after the microarray data was normalized, filtered for only the genes present, and analyzed using ANOVA with Benjamini-Hochberg multiple testing correction factor at P V 0.05. A, to visualize the data, samples and genes were analyzed by unsupervised hierarchical clustering using Cluster and graphically represented inTreeview (17). Gene signatures of interest are highlighted by dark lines.Tree headings for tumor and normal specimens are labeled. B, two major classes of samples were identified using the 2,207 genes identified in (A) representing an exact separation of the tumor and normal specimens.Tree headings for these two groups are labeled. Anatomic sites in the oral cavity in which samples were surgically removed are labeled as well as the stage and tissue number. Samples are subclustered into tongue and non-tongue classes under each of the tumor and normal headings.

www.aacrjournals.org 5961 Clin Cancer Res 2006;12(20) October 15, 2006 Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Human Cancer Biology

Fig. 2. Unsupervised classification analysis of patient-matched normal oral mucosal and OSCC samples. A, principal components analysis was done on tongue and non-tongue oral cavity OSCC and normal samples. Perspective image with the tongue and non-tongue samples as principal components and axes.Tongue (n), and non-tongue samples ( ), clustering of similar samples (encircled points). easily from the cellular population in saliva requires the fluidic ‘‘lab-on-a-chip’’ system for rapid (real-time), point-of- identification of unique biomarkers. care detection and diagnosis of oral cancer. Typically, genetic changes in cancer cells lead to altered gene expression patterns that can be identified long before the cancer phenotype has manifested. When compared with normal Materials and Methods mucosa, those changes that occur in the cancer cell can be used as biomarkers. Attempts to find biomarkers that identify Patient samples and characteristics. Consent was obtained for all premalignant OSCC and cancerous lesions have resulted in patients in this study in accordance with guidelines set forth by the several candidate genes associated with OSCC tumor progres- Institutional Review Board at the University of Pennsylvania and the University of Pittsburgh. All matched patient normal and tumor sion including p53, cyclin D1, and epidermal growth factor samples and unmatched normal and tumor samples were obtained receptor (8, 9). However, to date, no single gene has shown from surgical resection specimens from patients undergoing surgery for sufficient diagnostic utility in OSCC. Thus, as in many other OSCC using standardized procedures. The Penn data set was isolated cancers, clinical diagnosis will require considering the com- and microarrayed at the University of Pennsylvania, whereas the RO bined influence of many genes. Not surprisingly, the expression data set was kindly provided by Dr. Ruth Muschel (Children’s Hospital patterns of many genes have shown dramatic correlations with of Philadelphia, Philadelphia, PA) and has been previously reported tumor behavior and patient outcome. Indeed, microarray (12). The OSCC data set and other human data sets used in this study analyses of several tumor types has shown that global were downloaded from Geo DataSets (National Center for Biotechnol- expression profiling can distinguish tumor from normal cells, ogy Information). In this study, cancers of the oral cavity included as well as the class and subtype of cancer, far superior to current squamous cell carcinomas of the tongue, floor of the mouth, buccal mucosa, lips, hard and soft palate, and gingiva (2, 3). After resection, histopathologic diagnostic systems (10–15). Recent indepen- matched normal and tumor were fresh-frozen in liquid nitrogen. dent studies (15–19) carried out by various research groups Samples were banked at 80jC for storage until later use. All normal indicate that OSCC cells have a unique gene transcription samples were obtained at the greatest distance from the tumor, typically profile, which differs from that of normal cells. Interestingly, 2 to 3 cm, in which no gross appearance of tumor, leukoplakia, or none of these studies have tested the gene profiles for their erythroplakia could be detected. Touch preparation analysis was used ability to identify or predict OSCC. to confirm that each normal-appearing mucosa did not contain tumor. To date, no accurate, cost-efficient, and reproducible method All sections were evaluated cytologically and diagnosis was confirmed. exists that enables the mass screening of patients for OSCC. All tissue sections were fixed and stained with H&E and evaluated by two Recent literature strongly supports the notion that microarray pathologists (J. Hunt and M.M. Feldman) and histologic analyses were analysis of human cancers far exceeds conventional criteria with done to ensure that each tumor specimen was pure for microarray regards to diagnoses. Thus, the goal of this study was to identify analysis, containing >80% tumor tissue, and that each normal section did not contain dysplasia or carcinoma. Those samples that did not a gene expression signature for OSCC. By analyzing microarray meet these criteria were rejected for this study. Archival material of data from patient-matched normal and OSCC tissue and by normal sections were derived from patients who either had surgical validation at the RNA and protein level, and using several tooth extractions or had non–epithelial-related pathology. independent validation gene data sets, we identified a tissue- Extraction of total RNA and microarray analysis. For RNA isolation, specific gene expression signature that can predict OSCC. This each tissue specimen was placed in a liquid nitrogen–chilled mortar OSCC signature is our first step towards developing a micro- and the tissue ground to a fine powder. The liquid nitrogen was

Clin Cancer Res 2006;12(20) October 15, 2006 5962 www.aacrjournals.org Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of OSCC

Fig. 2 Continued. B, expression patterns up-regulated and associated with tongue samples are labeled. Regions identified in Fig. 1as site-specific are enlarged. C, expression patterns up-regulated and associated with non-tongue samples are labeled. Regions identified in Fig. 1as site-specific are enlarged. Headings for tumor and normal specimens are labeled.

www.aacrjournals.org 5963 Clin Cancer Res 2006;12(20) October 15, 2006 Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Human Cancer Biology

Table 1. Differentially expressed genes in normal mucosa and OSCC

Fold* P c Gene name Up-regulated 63.68217 1.21E-07 inhibin, hA (activin A, activin AB a-polypeptide) 71.4553 2.01E-06 MMP-1 (interstitial collagenase) 25.8173 2.01E-06 chemokine (C-X-C motif) ligand 13 (B cell chemoattractant) 9.258157 1.99E-05 MMP-9 (gelatinase B, 92 kDa gelatinase, 92 kDa type IV collagenase) 5.970897 5.93E-05 601658812R1 NIH_MGC_69 Homo sapiens cDNA clone IMAGE:3886131 3¶, mRNA sequence 8.185247 5.93E-05 collagen, type V, a2 9.032335 5.93E-05 H. sapiens NKG5 gene, complete cds 2.404988 6.64E-05 myosin X 10.58925 6.64E-05 collagen, type V, a1 9.743879 7.51E-05 collagen, type V, a2 4.964245 7.70E-05 lysyl oxidase-like 2 14.72698 7.88E-05 parathyroid hormone – like hormone 8.741428 7.88E-05 collagen, type III, a1 (Ehlers-Danlos syndrome type IV, autosomal dominant) 15.03796 8.29E-05 MMP-11 (stromelysin 3) 5.234371 0.000113 snail homologue 2 (Drosophila) 7.123956 0.000174 similar to rat FE65 protein, GenBank accession number X60469 9.496229 0.000201 collagen, type V, a1 9.258569 0.000218 collagen, type I, a2 1.992358 0.000218 epithelial protein lost in neoplasm h 2.240645 0.000219 myosin IB 10.50398 0.000246 collagen, type I, a1 16.73496 0.000246 osteoblast-specific factor 2 (fasciclin I-like) 5.122788 0.000256 collagen, type I, a2 47.11779 0.000256 collagen, type XI, a1 8.796429 0.000261 transforming growth factor, h-induced, 68 kDa 9.410731 0.000261 collagen, type I, a1 2.658922 0.000261 acid phosphatase 5, tartrate resistant 6.703311 0.000261 plasminogen activator, urokinase 4.470378 0.000261 granulysin 21.31044 0.000261 parathyroid hormone – like hormone 3.059548 0.000261 myosin IB 7.238651 0.000261 collagen, type V, a1 4.245012 0.000264 collagen, type IV, a1 4.509005 0.000293 secreted protein, acidic, cysteine-rich (osteonectin) 5.859527 0.000303 KARP-1-binding protein 3.776092 0.000306 melanoma-associated gene 2.849625 0.000313 trophoblast glycoprotein 7.018807 0.000314 triple functional domain (PTPRF interacting) 10.51419 0.000334 fibroblast activation protein, a 7.29039 0.00035 laminin, g2 2.339839 0.000372 ubiquitin-conjugating enzyme E2C 3.226771 0.000421 AL514445 H. sapiens neuroblastoma, H. sapiens cDNA clone CL0BB010ZF08 3¶, mRNA sequence 2.778452 0.000431 RecQ protein-like (DNA helicase Q1-like) 11.6708 0.000431 IFN, a-inducible protein (clone IFI-15K) 3.029643 0.000431 likely orthologue of mouse Shc SH2-domain binding protein 1 8.878598 0.00051 fibronectin 1 7.076595 0.000534 fibronectin 1 30.60455 0.000534 H. sapiens type X collagen gene 3.374103 0.000537 exostoses (multiple) 1

(Continued on the following page)

evaporated, and the tissue was homogenized in Trizol (Invitrogen, rates the T7 promoter. Second-strand cDNA synthesis was followed by Carlsbad, CA). Total RNA was isolated using the Trizol method and in vitro transcription for linear amplification of each transcript and dissolved in RNase-free water. To remove contaminates, the RNA was incorporation of biotinylated CTP and UTP (Enzo RNA Labeling Kit, purified using RNeasy spin columns (Qiagen, Inc., Valencia, CA). Each Affymetrix). The cRNA products were fragmented to 200 nucleotides or specimen typically yielded 50 Ag of total RNA. less, heated at 99jC for 5 minutes and hybridized for 16 hours at 45jC Total RNA samples were submitted to the University of Pennsylvania to U133A GeneChips microarrays (Affymetrix). The microarrays were Microarray Facility for microarray analysis using Affymetrix U133A washed at low (6 saline-sodium phosphate-EDTA) and high (100 chips. Samples were run on an Agilent Bioanalyzer to confirm integrity mmol/L MES, 0.1 mol/L NaCl) stringency and stained with streptavi- and concentration. For target preparation and hybridization, all din-phycoerythrin. Fluorescence was amplified by adding biotinylated protocols were conducted as described in the Affymetrix GeneChip antistreptavidin and an additional aliquot of streptavidin-phycoery- Expression Analysis technical manual. Briefly, 5 to 8 Ag of total RNA thrin stain. A confocal scanner was used to collect fluorescence signal at were converted to first-strand cDNA using Superscript II reverse 3-Am resolution after excitation at 570 nm. The average signal from two transcriptase (Invitrogen) primed by a poly(T) oligomer that incorpo- sequential scans was calculated for each microarray feature.

Clin Cancer Res 2006;12(20) October 15, 2006 5964 www.aacrjournals.org Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of OSCC

Table 1. Differentially expressed genes in normal mucosa and OSCC (Cont’d)

Fold* P c Gene name

6.493025 0.000555 fibronectin 1 4.030122 0.00051 tumor necrosis factor receptor superfamily, member 12A 5.858879 0.000522 predicted protein of HQ3121; H. sapiens clone FLC1492 PRO3121 mRNA, complete cds 6.161157 0.000525 laminin, a3 3.422555 0.000525 TPX2, microtubule-associated protein homologue (Xenopus laevis) 4.454709 0.000525 human lysyl oxidase (LOX) gene, exon 7 39.05315 0.000534 MMP-13 (collagenase 3) 7.076595 0.000534 fibronectin 1 30.60455 0.000534 H. sapiens type X collagen gene 3.374103 0.000537 exostoses (multiple) 1 6.493025 0.000555 fibronectin 1 3.977667 0.000574 triple functional domain (PTPRF interacting) 1.946353 0.000604 glutathione S-transferase N1 3.090632 0.000604 caldesmon 1 1.743859 0.000604 U2-associated SR140 protein 2.595988 0.000604 thymosin, h10 3.296296 0.000659 collagen, type IV, a2 16.18147 0.000715 MMP-12 (macrophage elastase) 4.554788 0.00072 collagen, type VI, a3 2.972363 0.00072 secernin 1 6.32457 0.00072 apolipoprotein L1 1.890579 0.00072 methyl-CpG binding domain protein 4 7.61961 0.00072 fibronectin 1 6.494057 0.000785 chemokine (C-X-C motif) ligand 9 2.451763 0.00082 FAT tumor suppressor homologue 1 (Drosophila) 2.732824 0.00082 myristoylated alanine-rich protein kinase C substrate 9.185588 0.00082 serine (or cysteine) proteinase inhibitor, clade E(nexin, plasminogen activator inhibitor type 1), member 1 12.76825 0.00082 serine (or cysteine) proteinase inhibitor, clade E(nexin, plasminogen activator inhibitor type 1), member 1 9.70648 0.00082 microfibrillar-associated protein 2 8.731814 0.00082 parathyroid hormone-like hormone 5.08355 0.00082 602035015F1 NCI_CGAP_Brn64 H. sapiens cDNA clone IMAGE:4183107 5¶, mRNA sequence 2.451526 0.000861 proteasome (prosome, macropain) subunit, h type, 2 3.163611 0.000861 chromosome 5 open reading frame 13 4.858088 0.000861 chromosome 10 open reading frame 3 15.55873 0.000914 a disintegrin and metalloproteinase domain 12 (meltrin a) 2.277927 0.000914 BUB1 budding uninhibited by benzimidazoles 1 homologue (yeast) 2.804374 0.000925 ribonucleotide reductase M2 polypeptide 2.798002 0.000942 chloride intracellular channel 4 2.252996 0.000942 high-mobility group box 3 3.431822 0.000942 IFN, a-inducible protein (clone IFI-6-16) 8.40618 0.000942 signal transducer and activator of transcription 1, 91 kDa 6.538403 0.000948 MMP-3 (stromelysin 1, progelatinase) Down-regulated 0.039489 4.12E-08 TU3A protein 0.004179 0.00035 cysteine-rich secretory protein 3 0.111661 0.000751 monoamine oxidase B 0.539417 0.000751 elongation factor RNA polymerase II-like 3 0.247429 0.000862

*Fold difference in expression between OSCC and normal tissue obtained using SAM. cAdjusted P value using ANOVA and Benjamini-Hochberg false discovery value of P = 0.001.

Analysis of microarray data. Initial data analysis was done using of probe fluorescence (corrected for nonspecific signal by subtracting Affymetrix Microarray Suite 5.0 to quantitate expression levels for the mismatch probe value) was calculated using the one-step Tukey’s targeted genes; default values provided by Affymetrix were applied to all biweight estimate. This signal value, a relative measure of the expression analysis variables. Border pixels were removed, and the average intensity level, was computed for each assayed gene. Global scaling was applied of pixels within the 75th percentile was computed for each probe. The to allow comparison of gene signals across multiple microarrays: after average of the lowest 2% of probe intensities occurring in each of 16 exclusion of the highest and lowest 2%, the average total chip signal was microarray sectors was set as background and subtracted from all features calculated and used to determine what scaling factor was required to in that sector. Probe pairs were scored positive or negative for the adjust the chip average to an arbitrary target of 150. All signal values from detection of the targeted sequence by comparing signals from the perfect one microarray were then multiplied by the appropriate scaling factor. match and mismatch probe features. The number of probe pairs meeting For statistical analysis, all data was normalized for comparison across the default discrimination threshold (s = 0.015) is used to assign a call of arrays using GeneSpring default normalization settings: data transfor- absent, present, or marginal for each assayed gene, and a P value is mation was set at measurements <0.01 to 0.01 (per chip, normalized to calculated to reflect confidence in the detection call. A weighted mean the 50th percentile; and per gene, normalized to the median). Genes

www.aacrjournals.org 5965 Clin Cancer Res 2006;12(20) October 15, 2006 Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Human Cancer Biology differently expressed between matched patient normal and tumor samples were obtained using GeneSpring and statistical analysis of microarrays (SAM; ref. 16). Briefly, after normalization, all gene expression data were filtered for those genes that were present in the matched patient normal and tumor samples. The 15,311 genes satisfying this filter were further analyzed for differences between normal and tumor samples using ANOVA with Benjamini-Hochberg multiple testing correction factor at P V 0.001. SAM was then used to determine fold expression levels. Unsupervised clustering was done with Cluster (17) at the settings described using Pearson’s correlation distance metric and complete linkage clustering followed by visualiza- tion in Treeview (17). Principal components analysis was used to compare site-specific expression in the oral cavity and was done using GeneSpring (principal components analysis on conditions). Immunohistochemistry. Immunohistochemistry was done to vali- date the differential expression of selected genes in tissue sections and to localize the tissue expression of the genes. Sections were incubated at 70jC for 20 minutes, deparaffinized in xylene (20 minutes at room temperature), and then rehydrated through a series of graded ethanol solutions (20 minutes at room temperature) followed by water (10 minutes at room temperature). Endogenous peroxide was quenched through treatment with hydrogen peroxide solution (15 minutes at room temperature). To enhance antigen exposure, specimens for matrix metalloproteinase-1 (MMP-1) and laminin-5 g2 chain (Ln-5g2) were incubated in 1% sodium citrate solution at 100jC for 10 minutes and then cooled to room temperature. The detection of MMP-3 required no antigen retrieval methods. Nonspecific binding sites were blocked with horse serum (Vector Laboratories, Burlingame, CA) for 30 minutes. The slides were then incubated overnight at 4jC with monoclonal mouse antibodies directed against the Ln-5g2 chain (D4B5; Chemicon, Inc., Billerica, MA), MMP-1 (36665; R&D, Minneapolis, MN), or MMP-3 (SL1 IID4; Chemicon). Biotinylated secondary antibody (anti-mouse, 30 minutes at room temperature; Vector Laboratories), a biotin-avidin complex (30 minutes at room temperature; Vector Laboratories), and a chromogenic substrate (3,3¶-diaminobenzidine, 10 minutes at room temperature; Vector Laboratories) were then applied. Sections were counterstained with hematoxylin for 5 minutes. Quantitative real-time PCR analysis. Changes in mRNA levels were compared by quantitative real-time PCR analysis, using the Bio-Rad (Hercules, CA) MyiQ single-color real-time PCR detection system. All gene-specific primers used for real-time PCR were purchased from Qiagen. Five micrograms of total RNA from normal and tumor specimens were converted to cDNA using Superscript II (Invitrogen) according to the manufacturer’s specifications. PCR reaction mixtures consisted of 2 AL of Faststart DNA Master SYBR Green I mixture [containing TaqDNA polymerase, reaction buffer, deoxynucleotide triphosphate mix (with dUTP instead of dTTP), SYBR Green I dye, and 10 mmol/L of MgCl2], 0.5 Amol/L of each target primer stock, 2 or 4 mmol/L MgCl2 (Ln-5g2 chain, MMP-1, and MMP-3) in a final reaction volume of 20 AL. h-Guswasusedastheinternalcontrolfor Fig. 3. Validation analyses of gene expression profiling. A, quantitative real-time normalization and Universal Human RNA (Stratagene, La Jolla, CA) PCR of matched normal and tumor specimens (n = 5 pairs) for MMP-1 (top), was used as the standard reference (18). Cumulative fluorescence was Ln-5g2(middle),andMMP-3(bottom). All expression values are normalized to expression of internal control gene h-Gus. All real-time reactions were done in measured at the end of the extension phase of each cycle. Product-specific triplicate; bars, SD. Only a total of seven of the paired samples were analyzed by amplification was confirmed by a melting curve analysis and agarose gel real-time PCR due lack of a sufficient sample. electrophoresis analysis. Quantification was done at the log-linear phase of the reaction and cycle numbers obtained at this point were plotted against a standard curve prepared with serially diluted samples. patient-matched tumor and normal samples were used as the training Class prediction. The support vector machine (SVM; GeneSpring) set and cross-validated. The Penn, RO, OSCC, and various tumor data algorithm was used to identify a gene predictor set from the patient- sets were used as the test sets. SVM was set to use the Golub method. In matched normal and tumor microarray data. SVM uses recognition and the Golub method, each gene is tested for its ability to discriminate regression estimates to identify class prediction gene sets using a between the classes using a signal-to-noise score, which is given by: training set of microarray data. The SVM algorithm attempts to find a l l 1 2 : hyperplane that provides separation between the different input data r þ r classes such that there is a maximal distance between the hyperplane 1 2 and the nearest point on any one of the input classes. Furthermore, Where li and ri,i= 1, 2, are the mean and SD of the expression using the training data set, SVM performs a 10-fold cross-validation to values over the samples in class i. Genes with the highest scores are kept select a set of predictor genes which leads to the smallest error rate. The for subsequent calculations. The Golub method of gene selection

Clin Cancer Res 2006;12(20) October 15, 2006 5966 www.aacrjournals.org Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of OSCC

Fig. 3 Continued. B, immunohistochemical staining analysis of MMP-1, Ln-5g2 chain, and MMP-3 in OSCC and normal oral mucosa. In OSCC (b, d ,andf) staining of MMP-1 (b), Ln-5g2chain(D), and MMP-3 (f) was detected around and within the OSCC tumor islands. Staining of normal oral mucosa (a, c,ande) could not detect the expression of MMP-1 (A)or MMP-3 (e).The Ln-5g2chain(c) was correctly detected in the basement membrane of the oral mucosa ("). Magnification, 400. A total of seven different OSCC tumor and normal sections were stained by immunohistochemistry.

calculates the difference in means between the training and test sets with Benjamini-Hochberg multiple testing correction factor at divided by the sum of the SDs to identify the best set of predictors. The P V 0.05 (20). This yielded a highly discriminating set of 2,207 number of genes used for predictors was set at 25 and the 2,207 genes genes. To visualize the gene expression data, hierarchical identified by filtering for only genes present, and analyzed by ANOVA clustering was done independently of normal and tumor P V with Benjamini-Hochberg multiple testing correction factor at 0.05 samples (Fig. 1A). The samples evenly partitioned into two were used as the gene pool. Once this set of 25 predictor genes was major groups corresponding to normal and OSCC samples. identified, it was applied to each test set to test the classification accuracy. All SVM calculations were done using kernel function set to More genes were differentially regulated in the tumor samples polynomial dot product (order 1) and diagonal scaling factor set to 0. than the normal specimens in this 2,207–gene set. Within each Prediction analysis of microarrays (19) and k-nearest neighbors tumor and normal cluster were two major classes. This confirmed SVM results (ref. 9; data not shown). separation was most apparent in the clustering of the normal samples (Fig. 1A and B). Six of the seven (>85%) normal samples that were derived from tongue tissue were in one class. In Results contrast, >80% (five of six) of those samples obtained from sites other than the tongue, including the mandible, floor of Patient-paired normal and tumor specimens. To identify the mouth, gum, and buccal mucosa, made up the other class differently expressed genes in oral mucosa and OSCC, we (Fig. 1B). A similar clustering of the OSCC tongue specimens and obtained matched patient OSCC and normal specimens from non-tongue OSCC specimens was also present in the tumor 13 patients undergoing surgical resection for OSCC at the samples (Fig. 1B). Principal components analysis highlighted University of Pennsylvania and at the University of Pittsburgh. the separation of the tongue and non-tongue specimens in the The clinical characteristics of the 13 paired tumor/normal normal and tumor groups (Fig. 2A). The separation of site- patients are shown in Supplemental Table S1. This group is specific gene expression was readily apparent in the normal representative of the general population of patients with OSCC specimens but was slightly less distinct in the OSCC samples having a median age of 59 in which a greater percentage (54%) (Fig. 2A). Additionally, within the OSCC tumor samples was a were male. All OSCC specimens were located in the oral cavity subcluster of those primary OSCC tumors that were associated and included squamous cell carcinomas of the tongue, floor with nodal disease (Fig. 1).7 Although expression profiling of the mouth, and buccal mucosa (Supplemental Table S1). Of allowed the distinct separation of tongue versus non-tongue the patients reporting, >90% smoked tobacco and/or drank samples, several samples clustered into groups that were alcohol (data not shown). We chose to compare patient-paired inconsistent with their histology. However, these results do normal and tumor specimens in order to provide the most indicate that there are distinct expression patterns within the oral statistically representative database for distinguishing gene cavity that can identify normal and tumor tissue sites of origin. expression difference between tumor and normal samples. All Closer inspection of the site-specific gene expression patterns RNA were isolated at the University of Pennsylvania. identified in Fig. 1 showed distinct gene expression profiles for Distinct expression profiles define tumor and normal samples. normal tongue tissue compared with normal non-tongue sites Initially, to identify differently expressed genes between oral mucosa and OSCC, the microarray data was normalized, filtered for only the genes present, and analyzed using ANOVA 7 Ziober et al., manuscript in preparation.

www.aacrjournals.org 5967 Clin Cancer Res 2006;12(20) October 15, 2006 Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Human Cancer Biology

(Fig. 2B and C). In particular, several enzymes associated with expressed across all samples by microarray. Quantitative real- biochemical processes were up-regulated in the tongue includ- time PCR results were done using the 2 CDDT method (18). As ing cytochrome family members, aldehyde dehydrogenase, shown in Fig. 3, the expression levels of MMP-1, Ln-5g2 chain, monoglyceride lipase, transglutaminase, sulfotransferase, and MMP-3 were all elevated in the tumor specimens as arachidonate 12-lipoxgenase, glutathione S-transferase, and compared with the adjacent normal tissue. The fold expression others. In addition, several signaling transduction molecules, of MMP-1, Ln-5g2 chain, and MMP-3 in tumor versus normal including RAB proteins and Rho-GTPase activating proteins, tissue was in agreement with that determined by microarray and the e-erb-b2 receptor were elevated in the tongue analysis (Supplemental Table S2). Overall, these results specimens (Fig. 2B). confirm, at the RNA level, the differential gene expression Although the non-tongue samples did express some bio- signature that distinguishes OSCC tumors and normal mucosa. chemical enzymes, the number was far less than that for the This sampling of genes was next examined at the protein level tongue samples (Fig. 2C). Unlike the tongue specimens, the in paraffin-embedded archival samples, both OSCC and non-tongue samples had several types of receptors up-regulated normal, by immunohistochemistry. All OSCC sections dis- including growth factor receptors, prostaglandin receptors, and played elevated expression of MMP-1, MMP-3, and Ln-5g2 G protein–coupled receptors. Likewise, several growth factors within the tumor islands (Fig. 3). In contrast, with the were also elevated, for example, the epidermal growth factor, exception of the Ln-5g2, these proteins could not be detected fibroblast growth factor-2, and WNT inhibitory factor. Finally, in the normal sections. Ln-5g2 was detected, as expected, in the non-tongue tissues showed expression for several transcription basement membrane of normal oral epithelium (ref. 21; factors, including ets, zinc finger, AP-2 and p300/CBP, and Fig. 3B). Thus, these results show that we have validated the signaling molecules like H-Ras suppressor and Grb-2 like OSCC gene signature at the RNA and protein levels. proteins. Together, these results indicate that there are unique OSCC gene signature predicts tumor and normal samples. To gene expression patterns between tongue and non-tongue sites identify an OSCC tumor predictor, the patient-matched OSCC in the oral cavity. tumor and normal microarray data was next analyzed by the Gene expression signature for OSCC. To more thoroughly class prediction algorithm SVMs (a supervised machine-learning characterize and identify the most significantly differentially technique; ref. 22). To identify the best gene predictors, the expressed genes in OSCC and normal mucosa, we used a patient-matched OSCC tumor/normal data was trained and combination of ANOVA with the Benjamini-Hochberg multi- cross-validated using the Golub gene selection method (9), with ple testing correction factor with P V 0.001 and SAM to analyze the number of possible gene predictors set at 25 genes. The cross- the patient-matched tumor/normal samples (16, 20). This validation of the matched patient tumor and normal sample resulted in a list of 92 genes that are highly significantly data show that this 25-gene set predictor could classify tumor differentially expressed in OSCC and normal tissue (Table 1). and normal samples with a 100% accuracy (Supplemental Table As shown in Table 1, this list is comprised of a majority of genes S3). Interestingly, a majority of the margins in Supplemental (95%) that are up-regulated in OSCC as compared with normal Table S3, 25 of 26 (96%) were z0.5 and are considered as (similar to that presented in Fig. 1). This list contains genes confident classifications (23, 24). However, the margin for expressed from 2-fold to >70-fold in the OSCC with P values sample 04-0123 was <0.05 and is considered an unreliable that ranged from 1 10 7 to 0.001 (Table 1). Likewise, the prediction. Thus, this reduced the accuracy to 96%. We were able genes which were down-regulated in OSCC ranged from 2- to to retain this 96% accuracy prediction rate using as few as 10 33-fold with P values of 4 10 8 to 7.5 10 4. genes; however, the 25 predictors yielded the best performance Although several genes in Table 1 are relatively unknown, many when applied to other data sets below (data not shown). have been implicated in OSCC development and progression. To test the predictive strength of the 25-gene predictor These include molecules associated with the extracellular matrix, identified by SVM and cross-validation, we tested it on three matrix proteolysis, cell to cell adhesion, migration, and other independent oral cancer data sets. The Penn data set was processes. For example, several collagen chains, two laminin-5 comprised of two normal specimens and 13 OSCC specimens, chains, six different MMPs (MMP-1, -3, -9, -11, -12, and -13), the RO data set (which has been previously reported) consisted and plasminogen activator of urokinase were all up-regulated in of 18 OSCC tumors and the OSCC data set consisted of 5 OSCC. In addition, genes regulating cell to cell adhesion and tumor and 4 normal samples (from GSE1722; Geo DataSet, motility including snail homologue 2, myosin, meltrin a,and National Center for Biotechnology Information; ref. 12). Using lysyl oxidase–like 2 were identified as being up-regulated in SVM and the Golub method of gene selection, all (100%) OSCC. The functions of those genes down-regulated in OSCC specimens of the Penn data set were correctly classified are presently unknown. However, several of the up-regulated (Table 2). However, the margins of two predictions were genes have previously been reported as markers of or being considered unreliable, resulting in an accuracy of 87%. The involved in OSCC tumor development and progression. 25-gene set predictor was able to accurately classify 86% of the Validation of the OSCC gene signature. To confirm the RO OSCC data (one incorrect prediction and two unreliable; findings from the microarray analysis, real-time PCR was done Supplemental Table S4). The OSCC predictor correctly classified using primers specific for a sampling of genes that were at the all nine samples as tumor or normal in the OSCC data set. top, middle, and bottom of the genes listed in Table 1. This However, one prediction was considered unreliable, giving an included MMP-1, Ln-5g2, and MMP-3. The amplification accuracy of 89% (Supplemental Table S5). Similar results for the efficiencies and expression values of the primer/probe sets at above analyses were obtained using k-nearest neighbor and various dilutions were compared with the amplification prediction analysis of microarrays (data not shown). Thus, efficiencies of the internal control gene h-Gus. We selected testing a total of 44 OSCC and normal specimens, the 25-gene h-Gus as the internal control because it was uniformly predictor was able to correctly classify 42 samples producing an

Clin Cancer Res 2006;12(20) October 15, 2006 5968 www.aacrjournals.org Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of OSCC average accuracy rate of >87%. As a follow-up to this initial previously proposed, the development of OSCC involves stromal work, we are beginning prospective studies to test the OSCC and immune-regulatory components. Thus, many of the gene signature(s) ability to predict using a larger sample predictor genes belong to these categories. population and improve the positive identification accuracy rate. Finally, using existing Affymetrix U133A chip–derived data Discussion sets (National Center for Biotechnology Information Geo DataSets) from 20 other human tumors including breast, renal In the present study, we did expression profiling on patient- clear cell tumor, acute myeloid leukemia, lymphoblastic matched normal mucosa and OSCC tumors to identify a gene leukemia, and Barrett’s-associated adenocarcinomas resulted expression signature that was capable of predicting OSCC tissue in accuracies of only 25% (Supplemental Table S7); thus, from normal. The use of patient-paired normal and tumor illustrating the tissue-specificity of the 25-gene predictor. specimens provided the most representative database statistically Many of the genes that make up the 25-gene predictor have for distinguishing gene expression difference between tumor and previously been implicated in and described for OSCC (Table 3). normal samples. By microarray analysis, we identified a highly However, several predictor genes have not been directly significant set of differentially expressed genes between normal associated with OSCC tumorigenesis and will therefore provide and OSCC tissue. Furthermore, when used in the supervised starting points for further investigations. Several of the genes machine algorithm SVM, we were able to classify three inde- identified in the predictor were also present in the set of highly pendent test data sets with accuracies ranging from 87% to significant genes expressed between normal and tumor 100%. This is the first report to date that has used patient tumor/ (Table 1). The predictor set of genes were comprised of several normal samples to identify a gene signature for the prediction epithelial marker genes with categories of potential interest of OSCC. Finally, this report satisfies our first requirement in including genes encoding extracellular matrix components, genes developing a microfluidic lab-on-a-chip system for rapid (real- involved in cell adhesion, including fasciclin; genes involved in time), point-of-care screening, detection and diagnosis of oral cell to cell integrity, for example lysyl oxidase–like 2 and snail- cancer; identification of a gene signature that predicts OSCC. homologue 2; genes encoding hydrolyzing activities, including We have used several approaches for the analysis of gene proteins involved in the degradation of the extracellular matrix expression data with regards to clinicopathologic variables. Our such as MMP-1, MMP-9, MMP-11; and urokinase and cytokines initial approach, gene selection using an ANOVA test with such as inhibin hA and parathyroid hormone–like hormone. As Benjamini-Hochberg multiple testing correction, was used to examine similarities and differences among the paired tumor/ normal samples in their patterns of gene expression. In Table 2. Prediction of Penn data set of OSCC and agreement with our results, several previous studies using normal specimens (87% accurate; 15 correct unmatched and matched OSCC tumor specimens and normal predictions; 2 unreliable) samples showed a similar distinct clustering of normal and tumor samples (13–15). As with these previous studies, we did c b x Sample* True value Prediction t margin n margin not find any hierarchical clustering of the samples as related to 03-0116 n n 0.699 0.699 stage of disease or tumor grade. The most differently expressed 03-0117 t t 0.21 0.21 genes were up-regulated in OSCC samples, whereas those in the 19965T t t 1.285 1.285 normal tissue were down-regulated. More interestingly, and in 20063T t t 1.207 1.207 20282T t t 0.9 0.9 contrast with previous studies, both tumor samples and, to a 20402T t t 1.065 1.065 greater extent, the normal samples were clustered according 21600T t t 1.435 1.435 to their anatomic sites in the oral cavity. Tissue from the tongue 21732T t t 1.091 1.091 comprised one major cluster whereas those from other sites 21984T t t 1.331 1.331 including buccal mucosa, mandible epithelial, gum tissue, and 22341T t t 1.51 1.51 23908T t t 0.11 0.11 floor of the mouth populated the other cluster. 23943 n n 1.16 1.16 Using a highly significant statistical and data-filtering ap- 23943T t t 1.324 1.324 proach, we identified 92 genes differentially expressed between 27904T t t 1.591 1.591 OSCC and normal mucosa (P < 0.001). Furthermore, the strong R1000T t t 1.56 1.56 correlation of real-time PCR with the array data for gene expression and the validation using immunohistochemistry * Sample number. strongly indicate that the 92 genes in this list are representatively cTrue value of the class of each sample, as either tumor (t) or normal (n). This value is compared with the value in the prediction expressed genes in OSCC. Thus, these studies indicate that many column to validate the training set. of the genes identified by microarray analysis are highly relevant bThe predicted class. Prediction was done using SVM, Golub to OSCC development and/or progression. For example, work method for selecting predictor genes selection (9), and a 25-gene by us and others have shown that urokinase, MMP-1, MMP3, predictor. xMargin shows the distance (in arbitrary units) to the hyperplane MMP-11, MMP-13, and laminin-5 (mainly the g2 and a3 for each of the classes, tumor (t) and normal (n). Positive scores chains) are up-regulated and play a significant role in OSCC are assigned to one class, and negative scores are assigned to the tumor development and progression (25–28). The MMPs are other class. The scores are then reported as the margins. This instrumental in the degradation of the extracellular environment corresponds to the distance from the sample to the separating allowing OSCC tumor growth and invasion, whereas laminin-5 decision boundary. The larger the margin, the farther away a score is from the boundary, and the more confident the classification is. plays a fundamental role in tumor growth and migration/ Predictions are considered unreliable when margins are <0.5. invasion (25). In addition, STAT-1 was identified as being up-regulated and is believed to play an important role in the

www.aacrjournals.org 5969 Clin Cancer Res 2006;12(20) October 15, 2006 Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Human Cancer Biology

Table 3. OSCC tumor and normal gene predictors

Affymetrix no. Predictive strength* Gene name 209074_s_at 2.845 TU3A protein 210511_s_at 2.375 inhibin, hA (activin A, activin AB a-polypeptide) 204475_at 2.334 MMP-1 (interstitial collagenase) 203936_s_at 2.072 MMP-9 (gelatinase B, 92 kDa gelatinase, 92 kDa type IV collagenase) 205242_at 2.026 chemokine (C-X-C motif) ligand 13 (B-cell chemoattractant) 203878_s_at 1.824 MMP-11 (stromelysin 3) 221729_at 1.688 collagen, type V, a2 215077_at 1.669 collagen, type III, a1 (Ehlers-Danlos syndrome type IV, autosomal dominant) 212473_s_at 1.669 601658812R1 NIH_MGC_69 H. sapiens cDNA clone IMAGE:3886131 3¶, mRNA sequence 202311_s_at 1.662 collagen, type I, a1 37145_at 1.635 H. sapiens NKG5 gene, complete cds 205479_s_at 1.633 plasminogen activator, urokinase 212488_at 1.627 collagen, type V, a1 201976_s_at 1.584 myosin X 202998_s_at 1.575 lysyl oxidase-like 2 203325_s_at 1.575 collagen, type V, a1 221730_at 1.571 collagen, type V, a2 206300_s_at 1.57 parathyroid hormone-like hormone 211980_at 1.549 collagen, type IV, a1 37892_at 1.527 collagen, type XI, a1 212364_at 1.515 myosin IB 210809_s_at 1.512 osteoblast-specific factor 2 (fasciclin I-like) 221898_at 1.503 602035015F1 NCI_CGAP_Brn64 H. sapiens cDNA clone IMAGE:4183107 5¶, mRNA sequence 213139_at 1.495 snail homologue 2 (Drosophila) 213419_at 1.485 similar to rat FE65 protein, GenBank accession number X60469

*The 25-gene predictor was determined using SVM and the Golub method for gene selection. A higher value defines a better predictor.

tobacco-induced pathogenesis of oral cancer (29). As in several but does not agree with their findings on invasive disease.8 recent studies, we found both snail homologue and lysyl oxidase These differences are more likely due to the dissimilar micro- to be expressed in OSCC. These two genes alter cell to cell array chips used in the Mendez et al. study as the one presented adhesion and are suspected of being important in OSCC tumor here. However, these results do illustrate that distinct genes are invasion (30). It is interesting that many of these genes, normally expressed in OSCC compared with normal samples. The OSCC expressed during wound healing, especially those associated prediction signature from this study will provide the emphasis with proteolysis and motility, are not properly regulated and to continue work with our collaborators from the University of seem to become constitutively overexpressed in OSCC. Pennsylvania School of Engineering and Applied Science in the To date, >20 studies incorporating microarray analysis have design, development and testing of a disposable handheld reported on the genetic changes associated with OSCC. microfluidic device for the clinical assessment of OSCC (32). Unfortunately, many of these studies have used a variety of To test for clinical applicability, we assessed whether different gene expression arrays and platforms, and thus, it is difficult to data sets of genes and tissues could be predicted by the OSCC make a direct comparison with our results (14, 15). In addition, gene expression signature. We selected SVM, a supervised none of these previous studies have tested their OSCC gene machine-learning technique for our prediction studies (22). signature for its ability to predict OSCC using an independent SVM was chosen because it has been used in several microarray validation data set as shown here. As expected, there is a unified studies with success and seems to be superior to similar algo- consensus that distinct gene expression patterns exist when rithms such as k-nearest neighbor and prediction analysis of normal and primary OSCC tumors are compared. For example, microarray (9, 23, 33, 34). Cross-validation of the training set, Ginos et al., who compared OSCC samples to unmatched which consisted of the paired tumor/normal samples from two normal subjects using Affymetrix U133A chips identified institutions, resulted in an accuracy rate of 96% using at least 25 several genes that overlapped with those found in this study genes per class. The most appropriate test of predictive accuracy (13). In addition, Mendez et al. used microarrays to analyze is to validate the predictor on an independent set of samples. both OSCC and normal specimens (31). As with our work here, Therefore, to provide a superior means of testing the predictor, we they concluded that oral carcinomas are distinguishable from did not split the data set and use leave-one-out validation. Instead, normal oral tissue using oligonucleotide arrays that contained we validated and obtained accuracy rates on the 25-gene predictor probes representing only 7,000 full-length human genes. using three independent validation sets: an independent OSCC Interestingly, these authors indicated that there was expression and normal sample set from the University of Pennsylvania, a profile heterogeneity among tumors of a particular histopath- previously published OSCC sample set, and one obtained from ologic grade and stage, and that no statistically significant the National Center for Biotechnology Information Gene Dataset differences in gene expression were found between early-stage disease and late-stage disease (31). Our data agrees with their assessment that there was no correlation with grade and stage 8 Manuscript in preparation.

Clin Cancer Res 2006;12(20) October 15, 2006 5970 www.aacrjournals.org Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of OSCC

(12). The 25-gene predictor had an overall accuracy ranging from chemokine ligand 13, inhibin, and myosins X and IB. Thus, we 86% to 89% for these two validation sets. How these numbers have identified a highly significant set of genes that are compare with the early clinical diagnosis accuracy of OSCC is expressed in OSCC which will provide opportunities for further currently under investigation. It is difficult to determine why some investigations into OSCC development and progression. samples were incorrectly classified. This may be the result of other These results represent the initial stage of bringing a gene tissue components within the sample (i.e., bone) or because some signature into the clinic for patient screening. To date, no samples were mistakenly labeled. accurate, cost-efficient, and reproducible method exists that Finally, we tested the OSCC 25-gene predictor’s ability to enables mass screening of patients for OSCC. Our results show classify non-oral cavity tumor and normal samples. The OSCC that such a method is possible using the OSCC gene signature predictor displayed poor classifications, with accuracies of only identified here. This gene signature, after further testing, could 25% (75% of the samples were predicted incorrectly), when be useful in identifying the site of tumor origin and/or microarray data sets (obtained from National Center for identification of neoplastic lesions before any gross appearance Biotechnology Information GEO DataSets) were derived from of tumor. As a follow-up to this initial work, we have begun non–OSCC human cancers. This indicated that the oral cancer prospective studies using this OSCC gene signature. In addition, gene predictor set was tissue-specific. In addition, several of we are beginning the development of a new, noninvasive, these genes present in Table 2 are those differentially expressed microfluidic-based diagnostic device using this OSCC gene in OSCC tumor and normal mucosa components. Interestingly, signature to distinguish oral cancer cells from normal mucosa. A several of the predictor genes have not been directly associated multidisciplinary group of scientists from the University of with OSCC tumor development or progression, and thus, Pennsylvania School of Medicine and the School of Engineering provide areas for further investigations. Categories of potential and Applied Science is planning to develop a device that will be interest include genes (putatively) encoding extracellular matrix used clinically for the early mass screening of individuals for the components, in particular, several collagen chains; genes presence of OSCC. It is anticipated that this microfluidic lab-on- involved in matrix degradation, including urokinase and three a-chip will be sensitive enough to identify a suspect lesion members of the matrix metalloproteinase family; genes before it is detectable during a routine clinical exam. Finally, regulating cell to cell and cell to extracellular matrix adhesion determining this gene predictor’s ability to clinically diagnose including snail homologue, lysyl oxidase, and fasciclin-like; oral dysplasias and other potential neoplastic oral lesions is and genes involved in cell growth and migration, for example, currently under way as well.

References 1. Shingaki S, Nomura T, Takada M, et al. Impact of 13. Ginos MA, Page GP, Michalowicz BS, et al. Identifi- to microarray data analysis. Boston (MA): Kluwer lymph node metastasis on the pattern of failure and cation of a gene expression signature associated with Academic Publishers; 2003. p. 166 ^ 85. survival in oral carcinomas. Am J Surg 2003;185: recurrent disease in squamous cell carcinoma of the 25. Yuen HW, Ziober AF, Gopal P, et al. Suppression of 278 ^ 84. head and neck. Cancer Res 2004;64:55 ^ 63. laminin-5 expression leads to increased motility, tu- 2. Funk GF, Karnell LH, Robinson RA, et al. Presenta- 14. Somoza-Martin JM, Garcia-Garcia A, Barros- morigenicity, and invasion. Exp Cell Res 2005;309: tion, treatment, and outcome of oral cavity cancer: a Angueira F, et al. Gene expression profile in oral squa- 19 8 ^ 10. National Cancer Data Base Report. Head Neck 2002; mous cell carcinoma: a pilot study. J Oral Maxillofac 26. Ziober BL, Turner MA, Palefsky JM, Banda MJ, 24:165 ^ 80. Surg 2005;63:786 ^ 92. Kramer RH. Type I collagen degradation by invasive 3. Weinberg MA, Estefan DJ. Assessing oral malignan- 15. BelbinTJ, Singh B, Smith RV, et al. Molecular profil- oral squamous cell carcinoma. Oral Oncol 2000;36: cies. Am Fam Physician 2002;65:1379 ^ 84. ing of tumor progression in head and neck cancer. 365^72. 4. Okamoto M, Nishimine M, Kishi M, et al. Prediction of Arch Otolaryngol Head Neck Surg 2005;131:10^ 8. 27. Moilanen M, Sorsa T, Stenman M, et al. Tumor- delayed neck metastasis in patients with stage I/II 16. TusherVG,Tibshirani R, Chu G. Significance analysis associated trypsinogen-2 (trypsinogen-2) activates squamous cell carcinoma of the tongue. J Oral Pathol of microarrays applied to the ionizing radiation procollagenases (MMP-1, -8, -13) and stromelysin-1 Med 2002;31:227 ^ 33. response. Proc Natl Acad Sci US A 2001;98:5116^ 21. (MMP-3) and degrades type I collagen. Biochemistry 5. Ensley JF, Gutkind JS, Jacobs JR, Lippman SM, 17. Eisen MB, Spellman PT,Brown PO, Botstein D.Clus- 2003;42:5414^ 20. editors. Head and neck cancer: emerging perspec- ter analysis and display of genome-wide expression 28. Arora S, Kaur J, Sharma C, et al. Ets-1, and vascular tives. Academic Press; 2003. patterns. Proc Natl Acad Sci US A1998;95:14863 ^ 8. endothelial growth factor expression in oral precan- 6. Epstein JB, Zhang L, Rosin M. Advances in the diag- 18. Livak KJ, Schmittgen TD. Analysis of relative gene cerous and cancerous lesions: correlation with micro- nosis of oral premalignant and malignant lesions. expression data using real-time quantitative PCR and vessel density, progression, and prognosis. Clin J Can Dent Assoc 2000;68:617^21. the 2(DDC(T)). Methods 2001;25:402^ 18. Cancer Res 2005;11:2272 ^ 84. 7. Sobin LH, Wittekind CH. Head and neck tumors. In: 19. TakahashiM,RhodesDR,FurgeKA,etal.Geneex- 29.Tsai SY, HsiehTC, Ardelt B, Darzynkiewicz Z,Wu JM. h Sorbin, Witttekind, eds. TNMclassification of malig- pression profiling of clear cell renal cell carcinoma: Combined effects of onconase and IFN- on prolifer- nant tumors. 5th ed. Berlin (Germany): Springer-Ver- gene identification and prognostic classification. Proc ation, macromolecular syntheses and expression of lag; 1997. p. 17^ 32. Natl Acad Sci U S A 2001;98:9754 ^ 9. STAT-1 in JCA-1 cancer cells. Int J Oncol 2002;20: 8. Greenman J, Homer JJ, Stafford ND. Markers in can- 20. Benjamini Y, Hochberg Y. Controlling the false dis- 891^6. 30. Peinado H, Del Carmen Iglesias-de la Cruz M, cer of the larynx and pharynx. Clin Otolaryngol Allied covery rate: a practical and powerful approach to mul- Olmeda D, et al. A molecular role for lysyl oxidase-like Sci 2000;25:9 ^ 18. tiple testing. J R Stat Soc Ser B 1995;57:289 ^ 300. 2 enzyme in Snail regulation and tumor progression. 9. Vielba R, Bilbao J, Ispizua A, et al. p53 and cyclin D1 21. ZioberAF, Falls EM, Ziober BL.The extracellular ma- EMBO J 2005;24:3446 ^ 58. as prognostic factors in squamous cell carcinoma of trix in oral squamous cell carcinoma: friend or foe? 31. Mendez E, Cheng C, Farwell DG, et al. Transcription- the larynx. Laryngoscope 2003;113:167 ^ 72. Head Neck 2006;28:740^9. al expression profiles of oral squamous cell carcino- 10. Golub TR, Slonim DK, Tamayo P, et al. Molecular 22. FureyTS, Cristianini N, Duffy N, et al. Support vec- mas. Cancer 2002;95:1482^ 94. classification of cancer: class discovery and class pre- tor machine classification and validation of cancer tis- 32.Wang J, Chen Z, Corstjens PL, Mauk MG, Bau HH. diction by gene expression monitoring. Science 1999; sue samples using microarray expression data. A disposable microfluidic cassette for DNA amplifica- 286:531 ^7. Bioinformatics 2000;16:906^14. tion and detection. Lab Chip 2006;6:46 ^ 3. 11. Singh D, Febbo PG, Ross K, et al. Gene expression 23. Tothill RW, Kowalczyk A, Rischin D, et al. An ex- 33. Brown MP, Grundy WN, Lin D, et al. Knowledge- correlates of clinical prostate cancer behavior. Cancer pression-based site of origin diagnostic method based analysis of microarray gene expression data by Cell 2002;1:203 ^ 9. designed for clinical application to cancer of unknown using support vector machines. Proc Natl Acad Sci 12. O’Donnell RK, Kupferman M,Wei SJ, et al. Gene ex- origin. Cancer Res 2005;65:4031^40. U S A 2000;97:262 ^ 7. pression signature predicts lymphatic metastasis in 24. Mukherjee S. Chapter 9. Classifying microarray data 34. BarrierA, Lemoine A, Boelle PY, et al. Colon cancer squamous cell carcinoma of the oral cavity. Oncogene using support vector machines. In: Berrar DP, prognosis prediction by gene expression profiling. On- 2005;24:1244 ^ 51. DubitzkyW, Granzow M, editors. A practical approach cogene 2005;24:6155^ 64.

www.aacrjournals.org 5971 Clin Cancer Res 2006;12(20) October 15, 2006 Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Identification of a Gene Signature for Rapid Screening of Oral Squamous Cell Carcinoma

Amy F. Ziober, Kirtesh R. Patel, Faizan Alawi, et al.

Clin Cancer Res 2006;12:5960-5971.

Updated version Access the most recent version of this article at: http://clincancerres.aacrjournals.org/content/12/20/5960

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2006/10/19/12.20.5960.DC1

Cited articles This article cites 31 articles, 9 of which you can access for free at: http://clincancerres.aacrjournals.org/content/12/20/5960.full#ref-list-1

Citing articles This article has been cited by 9 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/12/20/5960.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/12/20/5960. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research.