Enhanced Interferon Signaling Pathway in Oral Cancer Revealed by Quantitative Proteome Analysis of Microdissected Specimens Using 16O/18O Labeling and Integrated Two-dimensional LC-ESI-MALDI Tandem MS*

Oral squamous cell carcinoma (OSCC) remains one of the most common cancers worldwide, and the mortality rate of this disease has increased in recent years. No molecular markers are available to assist with the early detection and therapeutic evaluation of OSCC; thus, identification of differentially expressed proteins may assist with the detection of potential disease markers and shed light on the molecular mechanisms of OSCC pathogenesis. We performed a multidimensional 16O/18O proteomics analysis using an integrated ESI-ion trap and MALDI-TOF/TOF MS system and a computational data analysis pipeline to identify proteins that are differentially expressed in microdissected OSCC tumor cells relative to adjacent non-tumor epithelia. We identified 1233 unique proteins in microdissected oral squamous epithelia obtained from three pairs of OSCC specimens with a false discovery rate of <3%. Among these, 977 proteins were quantified between tumor and non-tumor cells. Our data revealed 80 dysregulated proteins (53 up-regulated and 27 down-regulated) when a 2.5-fold change was used as the threshold. Immunohistochemical staining and Western blot analyses were performed to confirm the overexpression of 12 up-regulated proteins in OSCC tissues. When the biological roles of 80 differentially expressed proteins were assessed via MetaCore™ analysis, the interferon (IFN) signaling pathway emerged as one of the most significantly altered pathways in OSCC. As many as 20% (10 of 53) of the up-regulated proteins belonged to the IFN-stimulated gene (ISG) family, including ubiquitin cross-reactive protein (UCRP)/ISG15. Using head-and-neck cancer tissue microarrays, we determined that UCRP is overexpressed in the majority of cheek and tongue cancers and in several cases of larynx cancer. In addition, we found that IFN-β stimulates UCRP expression in oral cancer cells and enhances their motility in vitro. Our findings shed new light on OSCC pathogenesis and provide a basis for the future development of novel biomarkers.


microarrays, we determined that UCRP is overexpressed in the majority of cheek and tongue cancers and in several cases of larynx cancer. In addition, we found that IFN-␤ stimulates UCRP expression in oral cancer cells and enhances their motility in vitro. Our findings shed new light on OSCC pathogenesis and provide a basis for the future development of novel biomarkers. Molecular & Cellular Proteomics 8:1453-1474, 2009.
Oral cancer is one of the most common cancers worldwide. In Taiwan, it remains the sixth most prevalent cancer overall and the fourth most common cancer to afflict males. Over the past 2 decades, the overall incidence and morbidity rates of patients with oral cancer have increased continuously. Epidemiological studies show that ϳ50 -70% of patients who undergo surgery for oral cancer die within 5 years (1)(2)(3)(4)(5)(6). This poor prognosis predominantly reflects late stage presentation, secondary cancer occurrence, local recurrence, and metastasis (7) as well as the lack of suitable markers for cancer detection. Therefore, there is an urgent need to identify proteins that are dysregulated in patients with oral cancer. Such proteins would serve as a valuable resource to find markers for the early diagnosis and disease monitoring of patients with oral cancer.
Oral cancer, a subtype of head-and-neck squamous cell carcinoma (HNSCC), 1 can form at various locations within the oral cavity, including the lips, tongue, buccal surfaces, gingiva, palate, floor of mouth, and oropharynx. Tongue and buccal cancers are the most common and most serious types of oral squamous cell carcinoma (OSCC) especially in southeast Asia (2,8). Alcohol abuse, smoking, and betel nut chewing are the main risk factors for OSCC. Genome-wide approaches have revealed many epigenetic and genetic alterations in patients with OSCC, including several biochemical pathways (9 -11). However, these studies have provided little information regarding alterations in the protein profiles of patients with OSCC. Recently state-of-the-art proteomics technologies have revealed alterations in protein abundance, posttranslational modification and turnover, and spatial and temporal distribution within tumor specimens. Using proteomics approaches, aberrantly expressed proteins have been identified in body fluids (12)(13)(14), frozen or paraffin-embedded tissues (15)(16)(17)(18), and cultured cell lines (19 -22). The fold changes in protein expression in samples from healthy and cancerous states as well as the roles of each protein in disease progression must be determined to identify potential candidates for biomarkers and therapeutic targets.
Blood samples are often used in clinical studies because they are less invasive and more convenient than other types of bodily samples and can be analyzed using automatic and high throughput techniques. Unfortunately the extremely dynamic range of protein concentrations in serum and plasma impedes the direct discovery of potential biomarkers (23,24). Proteins can be released into the blood from diseased tissues during cell death or via secretory pathways. To counteract this problem, serum and plasma biomarkers are sometimes identified by analyzing differential protein expression in tumors and adjacent normal tissues (25).
Like many other types of solid tumors, OSCCs often contain heterogeneous cell populations. Laser capture microdissection (LCM) is a common technique used to dissect a particular tumor cell type from heterogeneous cell populations, thereby reducing the tissue complexity and facilitating the discovery of tumor-associated molecules in small samples (9, 26 -28). Several laboratories have studied differential protein expression in microdissected tissue specimens from patients with head-and-neck cancer in efforts to discover novel tumor markers (15, 17, 29 -31). However, the semiquantitative approaches used in these studies may have limited the number of potential markers identified as well as the reliability of the protein quantification. To minimize technical variations and improve the reliability of protein quantification, a variety of sophisticated stable isotope labeling techniques have been developed for MS-based proteomics analysis, including chemical (32,33), metabolic (34,35), and enzymatic (36 -38) labeling techniques. Improvements in the quality and accuracy of quantitative proteomics analysis via such stable isotope labeling strategies have facilitated the discovery of potential tumor markers in malignancies such as OSCC/HNSCC (16,39,40).
Here we describe a strategy consisting of LCM, 18 O labeling, two-dimensional (2D) LC separation and an integrated ESI-MS/MS and MALDI-TOF/TOF MS (ESI-MALDI tandem MS) system. This strategy was used to identify differentially expressed proteins in OSCC cells microdissected from oral cancer tissue biopsies. A computational data analysis pipeline was also developed to calculate the relative abundances of 16 O-and 18 O-labeled peptides (similar to that described in a previous report (26)) and to assist with multidimensional protein identification and quantification. Using three pairs of OSCC specimens, we identified 1233 unique proteins with a false discovery rate less than 3%. Of these, we quantified 977 non-redundant proteins in which 80 proteins displayed Ն2.5fold changes in expression in microdissected tumor cells versus non-tumor cells. We validated these results in 12 selected targets via immunohistochemical staining and Western blot analysis of OSCC tissues. Our findings reveal that the interferon (IFN) signaling pathway is significantly altered in OSCC lesions.

Clinical Specimens
Three pairs of specimens of surgically resected primary OSCC lesions and adjacent non-tumorous tissues were obtained from three male patients for use in LCM. The specimens were immediately embedded in O.C.T. (Optimal Cutting Temperature) compound (Tissue-Tekா O.C.T., Sakura Finetek) and stored at Ϫ70°C until use. Before conducting the LCM experiments, the tissue sections were stained with hematoxylin/eosin and evaluated by a pathologist. All of the tissue samples were collected from patients who had signed informed consent forms prior to participation in the study, which was approved by the Institutional Review Board of Chang Gung Memorial Hospital at Lin-Ko, Taiwan. Clinicopathological data from paired specimens used in LCM, immunohistochemical (IHC) staining, and Western blot analyses are summarized in supplemental Table S1. Head-and-neck tumor tissue microarrays (BC34011, head-and-neck squamous cell carcinoma tissue arrays) containing 60 head-and-neck squamous cell carcinoma tissues and three normal gingival tissues were obtained from US Biomax, Inc. (Rockville, MD). kindly provided by Dr. Kuo-Wei Chang (School of Dentistry, National Yang-Ming University, Taiwan). The SCC4 tongue squamous cell carcinoma line was derived from a 55-year-old male (ATCC number CRL-1624), and the OEC-M1 oral epidermal carcinoma cell line was derived from the gingiva of a Chinese patient (42). The OC3 cells were cultured in a medium composed of DMEM (Invitrogen) containing 10% fetal calf serum and Keratinocyte-SFM (Invitrogen) (at a 1:2 ratio), and the SCC4 and OEC-M1 cells were grown in RPMI 1640 medium containing 10% fetal bovine serum, 25 mM HEPES, and antibiotics at 37°C in 5% CO 2 .

LCM and Protein Extraction for LC-MS/MS Analysis
In preparation for LCM, 8-m cryosections were mounted onto membrane slides. The slides were fixed with 70% ethanol for 30 s, washed with 25% ethanol for 45 s, placed in Mayer's hematoxylin solution for 30 s, rinsed with 75% ethanol, dehydrated once in 95% ethanol, cleared twice in 100% xylene for 30 s each, and thoroughly air-dried. Laser capture microdissection was performed using the Veritas Laser Capture Microdissection and Laser Cutting Systems (Arcturus, Mountain View, CA). Briefly the tissue surrounding the selected area was cut using a UV laser, and the internal areas were irradiated via soft IR laser pulses to dissociate the cut sections from the membrane slides. Several selected areas were then adhered to a CapSure LCM Cap (Arcturus) and immediately transferred to a 0.5-ml microcentrifuge tube for protein extraction. All captured cells were dissolved in 50 -100 l of lysis buffer (7 M urea, 2 M thiourea, 1% Triton X-100, and 50 mM Tris-HCl, pH 8.0) by vortexing at room temperature for 30 min, briefly sonicating the samples in an ice bath, and centrifuging at 20,000 ϫ g for 10 min to remove insoluble debris. The concentrations of the protein extracts were measured via a modified Bradford assay (Bio-Rad), and the proteins were further examined by SDS-PAGE and silver staining as described previously (43,44). We extracted ϳ60 -80 g of protein from 1.0 ϫ 10 6 -1.5 ϫ 10 6 microdissected cells.

Postdigestion 18 O Labeling
The extracted proteins were diluted 6-fold in 100 mM ammonium bicarbonate and digested twice with trypsin (1:50, w/w) at room temperature for 12 h each. Detergent and salts were removed via sequential strong cation exchange (Luna SCX, 5 m, Phenomenex) and C 18 reverse phase (LiChroprep RP18, 5-20 m, Merck) microtip columns. After drying the peptides under a vacuum, the samples were added to 16 18 O labeling was performed overnight at 37°C. Trypsin was inactivated by incubation in a boiling water bath for 15 min and acidification with formic acid to a final concentration of 3% (similar to that described in a previous report (26)).

2D LC Separation
Equally mixed 16 O-and 18 O-labeled peptides (derived from 20 g of proteins/sample) were injected into a BioBasic SCX column (5 m, 2.1 ϫ 150 mm, ThermoElectron) on an HPLC system (Waters Breeze HPLC instrument) and eluted on a 60-min ammonium chloride gradient in the presence of 25% acetonitrile, pH 3.0 (adjusted using formic acid). The effluents were pooled into 20 fractions, dried, and redissolved in 3% acetonitrile containing 0.01% TFA. Each fraction was loaded into a NanoEase C 18 trapping column (5 m, 0.18 ϫ 23.5 mm, Symmetry300 TM ) and separated on a 60-min acetonitrile gradient (ranging from 3 to 40%) on a capillary RP18 column (3.5 m, 0.15 ϫ 150 mm, Symmetry300, Waters). In preparation for integrated ESI and MALDI MS analysis, the effluent was split with a Nano Y-connector (Upchurch Scientific, Oak Harbor, WA) that diverted the flow by a ratio of 1:3 to the ESI source and to a 384-well target plate attached to a ProBot spotting robot (LC Packings/Dionex) with a 10-s collection time. The samples were mixed with ␣-cyano-4-hydroxycinnamic acid matrix (2 mg/ml in 80% ACN and 0.1% TFA) containing 3 fmol of internal standards in the 384-well target plate as previously described (45). The samples were then analyzed using a MALDI-TOF/TOF (Ultraflex TOF/TOF, Bruker Daltonics, Bremen, Germany) MS system under the management of FlexControl (version 2.2) and WarpLC (version 1.0) software (Bruker Daltonics).

LC-ESI-MALDI Tandem MS Analysis
ESI-IT data acquisition was performed using the Esquire3000plus (Bruker Daltonics) with EsquireControl 5.2 software. Peptide fragment spectra were acquired from one MS scan followed by six MS/MS scans of the most abundant parent ions. Each precursor ion was analyzed twice and then excluded in the following minute. Automatic and intelligent MALDI-TOF/TOF data acquisition was performed using WarpLC software (Bruker Daltonics) with an LC-ESI-MALDI work flow, which performs a low redundancy parent ion selection by excluding peptides already identified by ESI-MS/MS analysis. Compounds spanning more than 60% of the MALDI-TOF MS spectra in a 384-well plate were considered background signals and were excluded from the parent precursor list. In each spectrum, eight of the most abundant peaks with signal-to-noise ratios higher than 30 were selected as parent precursors and were used in tandem MS assays in laser-induced fragmentation technique (LIFT TM ) mode with Flex-Control 2.2 software.

MS Data Processing and Database Search
The emerging spectra identified via ESI-MS/MS and MALDI-TOF/ TOF MS were analyzed using DataAnalysis 3.4 and FlexAnalysis 2.4 peak picking software (Bruker Daltonics), respectively, and used in searches of the Swiss-Prot_51.6 database (selected for Homo sapiens, 15,720 entries) assuming trypsin as the digestion enzyme. The MASCOT search engine (version 2.2.03, Matrix Science, London, UK) was used with one missing cleavage site; MS tolerance values of 2.5 and 0.8 Da for ESI-IT and WARP-ESI-MALDI data sets, respectively; MS/MS tolerance values of 0.6 Da for both data sets; and variable modifications of peptide including methionine oxidation and double 18 O labeling of the carboxyl terminus. Protein identification was performed using probability-based Mowse (molecular weight search) scores (p Ͻ 0.05) and the MudPIT algorithm of the MASCOT search engine. Information derived from MS spectra and database searches was exported into Microsoft Excel and XML file formats, respectively, for further 16 O/ 18 O quantification analysis. The false discovery rate of protein identification was determined by searching the MASCOTgenerated decoy database (using the same parameters described above) and was adjusted to false discovery rate Ͻ3% for each experiment. The resulting list of distinguishable proteins was generated by excluding identifications obtained solely from shared peptides and by including proteins containing at least one distinct peptide with a score higher than the identity threshold in MASCOT (but that was not shared with other identified proteins). Protein quantification was performed using the resulting distinguishable protein data set.

Protein Quantification Pipeline
The quantification pipeline is shown in Fig. 1B. This pipeline involves three main steps.
Step 1: Localization of Identified Peptides on the MALDI Plate-From the exported MASCOT search results (XML files), each rank 1 peptide with an ion score Ͼ15 was analyzed to locate the well corresponding to the MALDI MS spectra. The correspondent wells for peptides identified in the MALDI-TOF/TOF analysis were defined as the wells where TOF/TOF measurements were performed. The correspondent wells of peptides detected via ESI were determined from the ESI chromatographic retention times. In general, the retention time of each identified peptide, minus the delay time for sample collection on the MALDI plate (chromatographic offset time), was divided by 10 (because of the 10-s collection time per well) to translate the retention time into the well information. The well with the maximum ion intensity for a particular compound was identified by scanning forward and backward in 4-well regions surrounding the correspondent well and submitted as the apex of a particular ion chromatography (the submitted well).
Step 2: Peak Pairing, Relative Abundance Calculation, and Summation of Three Consecutive Well Fractions-The m/z values and intensities of the paired monoisotopic peaks (I 0 , I 2 , and I 4 for the 16 O 2 -, 16 O/ 18 O-, and 18 O 2 -labeled peptides, respectively) were determined for each located well. Only paired peaks containing all isotopic peaks (I 0 , I 2 , and I 4 ) were selected for further analysis. To minimize interference from overlapping peaks in the peptide quantification, the paired peaks with front neighbor ions (I Ϫ1 or I 0 Ϫ 1) displaying more than 30% of the intensity of I 0 were filtered out. Finally, we determined the sum of the peak areas spanning Ϯ1 well surrounding the submitted well for a particular ion displaying a 0.2-Da mass tolerance of the calculated mass of the identified peptide. The relative abundance of the identified peptide was then calculated based on the theoretical isotopic distribution, which was computed using the Isotopic Pattern Calculator as described previously (26).
Step 3: Protein Grouping and Abundance Evaluation-Peptides that had been identified and quantified via multidimensional fractionations were then combined and grouped by Swiss-Prot entry name. To improve the reliability of protein identification and quantification, shared and carboxyl-terminal peptides were filtered out during quantitative analysis. A Dixon's test (using a critical Q value corresponding to a 95% confidence level) was applied to remove the outlier ratios of peptides. Protein abundance ratios and standard deviations were then calculated. For proteins containing only one or two quantified distinct peptides, the protein ratios and their associated deviations were directly averaged without further consideration. To account for errors in sample preparation, the protein abundance ratios for each experiment were readjusted via global median normalization in which the individual protein ratios were divided by the median value of all quantified protein ratios in each experiment.

Immunohistochemical Staining
IHC staining analyses were performed using an automatic immunohistochemical staining device according to the manufacturer's instructions (Bond TM , Vision Biosystems, Mount Waverley, Victoria, Australia) and as reported previously (46). Consecutive sections (5 m thick) of formalin-fixed, paraffin-embedded specimens from 10 OSCC patients were stained with various antibodies using the Envision kit (Dako Corp., Carpinteria, CA). Immunohistochemical analyses were performed using specific antibodies against ubiquitin cross-reactive protein (UCRP) (a rabbit polyclonal antibody; kindly donated by Dr. . The intensities and percentages of positive staining of the target cells were determined by pathologists (Ying Liang and Chuen Hsueh) and used for quantitative scoring. Staining intensity was graded using four scores with 0 representing a negative stain and 1, 2, and 3 indicating weak, moderate, and strong staining, respectively. Scores were then multiplied by the percentage of positively stained cells to obtain the final protein expression score. The final expression scores were classified into four groups, including negative staining (scores of 0), weak staining (scores of 10 -70), moderate staining (scores of 80 -170), and strong staining (scores Ն180).

Western Blot Analysis
Cell extracts were prepared as described previously (47), and protein concentrations were determined using the Bradford protein assay reagent (Bio-Rad). Samples (30 g of protein/lane) were separated by 8 or 15% SDS-PAGE, transferred to PVDF membranes (Millipore Corp.), and probed using primary antibodies against the candidates of interest as described previously (43,44). For analyzing IFN-stimulated gene expression by Western blot, the OC3 or SCC4 cells were washed twice with PBS and then incubated in fresh medium with or without IFN-␤ (PeproTech Inc.) for 24 h.

Functional Annotation and Network Analysis
Differentially expressed proteins detected via quantitative proteomics analysis were functionally classified according to Gene Ontology biological process using ProteinCenter TM (Proxeon Biosystems, Odense, Denmark). Network analyses of protein candidates and the ratios of their expression in tumor and non-tumor cells (obtained from five independent experiments) were performed using the MetaCore TM analytical suite version 4.7 (GeneGo, Inc., St. Joseph, MI) and compared using p values Ͻ0.01 as statistical metrics. The statistical significance of the identified networks was based on p values, which are defined as the probability that a given number of proteins from the input list will match a certain number of gene nodes in the network.

Transwell Migration Assay
Cell migration was assayed in 24-well Transwell chambers (using an 8-m-pore filter) (Costar, Corning Inc., NY). The OC3 cells were suspended in 300 l of serum-free DMEM and Keratinocyte-SFM in a 1:2 ratio and were treated with or without IFN-␤ (20 units/ml) (PeproTech Inc.). The cells were then inserted into the upper chamber, while the lower chambers were filled with 600 l of serum-free DMEM and Keratinocyte-SFM in a 1:2 ratio containing 10 g/ml fibronectin (Sigma). After a 6-h incubation at 37°C, the chambers were gently washed twice with PBS, fixed in methanol, and stained with Giemsa. The numbers of cells that traversed the filter to the lower chamber were counted at a 400ϫ magnification in six fields per filter using NIH Image J software (version 1.4g, National Institutes of Health, Bethesda, MD). Results were expressed as means of cell number Ϯ S.D. Statistical analysis was performed using two-sided, unpaired Student's t test with p values less than 0.05 considered significant.

RESULTS
Experimental Design and Sample Preparation-We designed an 18 O labeling-based protein identification and quantification approach with two integrated MS systems to identify dysregulated proteins in OSCC tumor cells and adjacent nontumor epithelia (Fig. 1A) in which a computational pipeline for integration of the ESI and MALDI tandem MS data was developed (Fig. 1B). The epithelial cells of the primary OSCCs and their adjacent non-tumorous tissues were dissected by LCM and used for protein extraction. Fig. 2A shows representative images of tumor and non-tumor epithelia sections before and after LCM. The quality and quantity of proteins extracted from the three pairs of microdissected samples (T1/N1, T2/N2, and T3/N3) were examined by SDS-PAGE followed by silver staining (Fig. 2B). The results revealed that although the T2/N2 pair seems similar, a difference in protein bands ranging from 36.5 to 66 kDa could be detected in the tumor and non-tumor parts of T1/N1 and T3/N3 pairs. For example, when compared with T1, N1 shows an additional prominent band around 50 kDa and the lack of a band around 45 kDa. In addition, the protein profiles of the three pairs are not similar to each other, indicating the heterogeneous nature of the patient sample pairs. Equal amounts of extracted proteins were then trypsin-digested and labeled in 16 (Fig. 3A, upper panel). These fractions were then subjected to simultaneous second dimensional on-line LC-ESI and off-line LC-MALDI analyses. Protein identification was performed using WarpLC software. In each SCX fraction of sample 1, the number of proteins identified using the MAS-COT algorithm (by ESI alone or by integrated ESI-MALDI analysis) are summarized in Fig. 3A, lower panel. The number of unique proteins identified using the integrated ESI-MALDI strategy increased by ϳ45-100% in each SCX fraction as compared with the number of proteins identified using ESI-IT alone. Approximately 33-60% of the total unique peptides identified in ESI-MALDI mode (in six independent experiments) could only be determined by MALDI-TOF/TOF analysis (Fig. 3B). The benefit of using MALDI-TOF MS spectra to obtain quantitative information is that this technique generates less complex spectra and higher impurity tolerance than the ESI MS system (48 -50). In addition, the femtomolar sensitivity and the 1:10 dynamic range of quantification can be achieved by 18 O labeling in MALDI-TOF MS measurement (51). Therefore, the abundance of peptides identified using integrated ESI and MALDI MS systems was calculated using MALDI-TOF MS spectra, thereby generating more accurate mass measurements and higher resolution than is possible with conventional ESI-IT MS.
Quantification of ESI-MALDI-identified Peptides by the MALDI-TOF MS Spectra-The reliability of MS-based quantification, especially for 16 O/ 18 O-labeled peptides, depends highly on the resolution and accuracy of the MS spectrum (52). As mentioned earlier, the spectra generated in MALDI-TOF MS are of sufficient quality for quantifying 16 O/ 18 O-labeled peptides in contrast to the spectra generated in traditional ESI-IT MS (52). However, the integrated ESI-MALDI MS system can be challenging with regard to the alignment of ion features generated from the ESI and MALDI MS peptide measurements (using MALDI MS spectra). To address this issue, we developed a computational pipeline to determine the matched MALDI-TOF MS spectra of all identified peptides (Fig. 1B). The performance of this alignment-and-quantification pipeline was first evaluated by calculating the percentage of the quantified peptides. As shown in Fig. 3B (blank bar), more than 85% of the identified peptides (with ion scores Ͼ15) could be quantified. The accuracy of protein quantification was then evaluated using equally mixed model samples of 16T1/18T1 (a mixture containing equal amounts of 16 Oand 18 O-labeled peptides prepared from the microdissected tumor cells of sample 1). In this experiment, 283 proteins were quantified, and their log-transformed protein ratios could be fitted into a Gaussian distribution with a mean of 0.0661 (about 1.05 in original scale) and two standard deviations of 0.70267 (Fig. 3C). Wherein ϳ95% of the proteins displayed fold changes in the range of Ϫ0.64 to 0.77 (mean Ϯ 2S.D.; equal to 0.64 -1.7 in original scale). The correlation between protein identification and quantification was then assessed using a swap-labeled sample set (16T1/18N1 and 16N1/18T1) in which 388 proteins were simultaneously identified and quantified in both samples with a linear correlation coefficient of ϳ0.87 (Fig. 3D). The representative MALDI-TOF MS spectra and the corresponding ESI spectra obtained via LC-ESI/ MALDI MS analysis of a specific SCX-separated fraction from 16T1/18N1 are shown in Fig. 3E. These findings illustrate the different resolutions of the two MS systems while simultaneously indicating the change in 16 O ion features relative to the 18 O-labeled features. In summary, these results demonstrate the feasibility of our integrated ESI-MALDI strategy for use in 16 O/ 18 O-labeled protein identification and quantification. This method was subsequently used to identify differentially expressed proteins among the three pairs of microdissected OSCC samples.
Identification of Differentially Expressed Proteins in OSCC Tissue Specimens-Using MudPIT scoring (the MASCOT algorithm), 747, 787, 524, 626, and 600 unique proteins were identified within 16N1/18T1, 16T1/18N1 (sample 1), 16T2/ 18N2 (sample 2), 16N3/18T3, and 16T3/18N3 (sample 3), respectively, with corresponding false determination rates of 2.72, 1.06, 2.19, 1.39, and 1.39 (Fig. 4A). With regard to the reliability of protein identification, proteins identified by peptides with scores higher than the identity threshold that were not shared with other proteins were selected as distinguishable proteins and included in further quantification analysis. In all, 572, 593, 386, 439, and 466 proteins (of the 615, 648, 406, 466, and 486 distinguishable proteins, respectively) were quantified (Fig. 4A). In summary, 1233, 1035, and 977 unique, distinguishable, and quantified proteins, respectively, were identified from the three pairs of OSCC specimens (as determined by five independent experiments). Detailed identifica-tion and quantification information for the 977 quantified proteins is available in supplemental Tables S2-1 (protein list) and S2-2 (peptide list). The MS/MS spectra and the correspondent fragment assignments of the single distinct peptidebased protein identifications are summarized in supplemental Fig. S1. Distributions of the normalized protein ratios are displayed as box plot diagrams in Fig. 4B. The 95% distribution of protein ratios determined in the equally mixed model sample (16T1/18T1) fall within the 0.64 -1.7 range (Fig. 3C); thus, proteins displaying an average T/N ratio higher than 2 or lower than 0.5 were selected for further consideration. Among these proteins, those displaying an average fold change of Ն2.5 in tumor versus non-tumor parts in at least three experiments were chosen as potential candidates for future analysis. Considering the limited identification rate of shotgun proteomics as well as the inherent heterogeneity of OSCC, distinguishable proteins that were detected in two of the five experiments but that displayed average fold changes greater than 5 or less than 0.2 were also selected as potential candidates. After analysis of the 977 quantified proteins, 53 upregulated and 27 down-regulated candidates were identified and were classified by biological process category using Gene Ontology (GO) (Tables I and II (Table I and

references cited therein).
Validation of Candidates by Immunohistochemical Staining and Western Blotting-Commercially available antibodies were used in Western blot analyses to examine the expression of eight up-regulated candidates in three paired oral biopsies. The results revealed increased expression of seven proteins (UCRP, fascin, GBP1, ANXA3, HSP47, STAT1, and FLNA) in two of the three tumor biopsies (Fig. 5A). We then examined the expression of the eight up-regulated candidates and four additional proteins (thymidine phosphorylase, mitochondrial superoxide dismutase, filamin B, and carbonic anhydrase II) in The number of total proteins identified is shown at the top of each bar. B, the ratio distributions of proteins quantified in each experiment are presented as box plot diagrams. The mean value of all ratios (E), the percentages of ratio data points, and the minimum and maximum data points (Ϫ) are indicated.  (126), larynx (127) 10 paired OSCC specimens by immunohistochemical staining (supplemental Fig. S2). The staining results were evaluated by two pathologists and scored as either negative, weak, moderate, or strong expression. The specificity of each antibody was verified by Western blot analysis using protein extracts from an oral cancer cell line (supplemental Fig. S3). A representative staining pattern for one paired tissue section (case 9) per protein as well as the scoring results for the 12 candidates are shown in Fig. 5, B and C, respectively. Comparison of staining scores from tumor and non-tumor counterparts revealed that, with the exception of ANAX3, 11 of the 12 candidates were significantly overexpressed (in more than eight of the 10 paired specimens) in tumor cells. The representative MS and MS/MS spectra used for the identification and quantification of these validated proteins are shown in supplemental Fig. S4. Collectively these observations demonstrate the consistency in results obtained from MS-based identification/quantification and our immunohistochemical validation. In addition, these results indicate the feasibility of using this technology platform to discover aberrantly expressed proteins in microdissected OSCC cells. MetaCore Analysis of Altered Signaling Pathways in OSCC-To determine which biological networks are affected by the dysregulated proteins, the 80 MS-identified candidates were analyzed using MetaCore (version 4.7) (53). The analysis revealed six significantly altered pathways (p Ͻ 0.001) in OSCC lesions, including pathways related to keratin filament remodeling, IFN-␣/␤ signaling, non-junctional endothelial cell contact, antiviral actions of IFNs, GPIb-IX-V-dependent platelet activation, and tetraspanin contributions to integrin-mediated cell adhesion (Table III). Data obtained for two prominent pathways (keratin filaments involved in cytoskeleton remodeling (Ϫlog p value ϭ 19.364) and type I IFN signaling (Ϫlog p value ϭ 5.695)) are shown in Fig. 6.
Up-regulation of IFN-␤-mediated Signaling Pathway in OSCC-The results described above suggested that the type I IFN signaling pathway was significantly altered in OSCC lesions. We determined that 10 of the 53 up-regulated candidates were members of the IFN-stimulated gene (ISG) family (54 -56), including SYWC, IFM1, STAT1, TYPH, UCRP, MX1, GBP1, GBP2, PML, and AMPL (Table I). Therefore, it is possible that an upstream regulator of this pathway (e.g. IFN-␤) would be up-regulated in OSCC as well. This notion was confirmed by immunohistochemical staining, which clearly revealed the overexpression of IFN-␤ in all 10 pairs of OSCC tissue sections (Fig. 7A). To explore the possible biological role of IFN-␤ in OSCC, the effects of IFN-␤ on cell proliferation and expression of two ISGs (UCRP and STAT1) were investigated in OSCC cells. Our results revealed that IFN-␤ treatment for 1-3 days had a marginal effect (ϳ10 -20%) on the growth of OC3 cells (supplemental Fig. S5). In contrast, IFN-␤ treatment significantly enhanced the expression of UCRP and STAT1 in OC3 cells (Fig. 7B), consistent with the prediction generated by GeneGo Map. The UCRP protein, also referred Fold changes in target protein expression in tumor (T) and non-tumor (N) cells. Freq., frequency of target up-or down-regulated proteins in detectable samples; ND, not detected; NQ, detected but not quantified.
c Target proteins that were dysregulated in tumor cells as determined using genomics (G), proteomics (P), IHC, and Western blot (WB) approaches.  c Target proteins that were dysregulated in tumor cells as determined using proteomics (P) and IHC approaches.

Quantitative Proteome Analysis of Microdissected OSCC
to as ISG15, is known to play a critical role in the IFNmediated immune response against antiviral infection (57,58). Through a mechanism called ISGylation, UCRP, like ubiquitin, conjugates with a variety of cellular proteins that modulate diverse cellular functions such as RNA processing, stress response, metabolism, cytoskeleton organization, and regulation (59,60). We found that IFN-␤ treatment also stimulated UCRP expression in another OSCC cell line (SCC4) and significantly enhanced the conjugation of UCRP with cellular proteins (via ISGylation) (Fig. 7C). Notably ISGylation was concomitantly enhanced in three OSCC tumor tissues that overexpressed UCRP as compared with adjacent non-tumor controls (Fig. 7C). Finally the cell motility of OC3 oral cancer cells increased significantly in response to IFN-␤ treatment (Fig. 7D). Collectively these results demonstrate that the IFN-␤-mediated signaling pathway was up-regulated in OSCC lesions studied and that IFN-␤ stimulates UCRP expression, ISGylation, and migration of OSCC cells. cancer in more detail, a larger cohort comprising 49 paired OSCC specimens (17 buccal cancers, 18 tongue cancers, 10 gum cancers, three hard palate cancers, and one mouth floor cancer) was surveyed by immunohistochemical staining again. The results showed that 48 of 49 (98.0%) were negative and 44 of 49 (89.8%) were moderately or strongly positive for UCRP staining in adjacent normal and tumor parts, respectively (supplemental Table S4). As OSCC is a subtype of head-and-neck cancer, thus we further examined UCRP expression using a head-and-neck tissue microarray chip containing 60 head-and-neck squamous cell carcinoma tissues and three normal gingival tissues. As shown in Fig. 8, UCRP was not detected in normal gingival tissues; however, moderate to strong staining of UCRP was detected in nine of 12 tissue sections from patients with cheek cancer, nine of 15 tissue sections from patients with tongue cancer, and six of 18 tissue sections from patients with larynx cancer. In addition, all three tissue sections from patients with upper jaw cancer exhibited weak UCRP staining, whereas UCRP was not detected in tissue sections from patients with nose cancer. These results indicate that UCRP is highly expressed in OSCC lesions at different sites within the oral cavity.

DISCUSSION
Laser capture microdissection is often used in conjunction with MS-based protein identification technology to assist with the discovery of tumor-associated molecules in tissue specimens containing various types of cells (27,31,(61)(62)(63)(64). However, relatively few studies have used these techniques to identify OSCC/HNSCC-associated proteins in tissue specimens from patients with OSCC/HNSCC (15,17,29). For example, Melle et al. (29) used ProteinChip technology and 2D gel electrophoresis to examine the up-regulation of annexin V in microdissected HNSCC tissues. In addition, Baker et al. (17) used LC-MS/MS to identify ϳ100 unique proteins in sets of normal and cancerous microdissected tongue specimens and used immunohistochemistry to demonstrate the downregulation of cytokeratin (CK) 13 and the up-regulation of heat-shock protein 90 in tumor cells. Another recent study used LC-MS/MS to analyze the proteins extracted from microdissected formalin-fixed, paraffin-embedded tissue sections of normal oral epithelium as well as well, moderately, and poorly differentiated oral cancers. The authors identified 391 and 866 total proteins in the normal oral epithelia and in well differentiated oral cancer tumors, respectively (15). This study explored the relative distribution of identified proteins in tissue samples by counting the peptide numbers of each protein detected. In addition, the authors validated the expression of cytokeratins 4 and 16, desmoplakin, desmoglein 3, and vimentin proteins using immunohistochemistry (15). Although these studies identified and confirmed several OSCC/HNSCC-associated proteins, a more global view of the changes in protein expression in microdissected OSCC cells and adjacent non-tumor epithelial cells can be achieved using a precise quantification approach. Here we describe a quantitative technology platform that combines 18 O labeling, comprehensive 2D LC separation, and integrated ESI-MALDI MS/MS measurements. This technique was successfully used to identify and quantitate 977 proteins in microdissected samples from three pairs of freshly resected OSCC specimens. Among the MS-quantified proteins, we identified 53 up-regulated and 27 down-regulated proteins with fold changes Ն2.5. The reliability of the MS-based protein identification/quantification platform was confirmed via immunohistochemical validation experiments, which revealed that more than 90% of the 12 up-regulated proteins were overexpressed in OSCC tumor cells (Fig. 5 and supplemental Fig. S2). In addition, the reliability of this platform was confirmed by other immunohistochemical studies showing that five additional candidates (K1C16, TENA, FINC, TSP1, and NDRG1) were overexpressed in OSCC tissues (Table I). Finally 12 of the 53 upregulated proteins (SYWC, K1C16, STAT1, TYPH, LDHA, K1C17, GRP78, SODM, ITB4, TPM4, K1C14, and GTR1) appeared to be overexpressed in OSCC tissues as determined by counting the peptides detected in microdissected OSCC samples from formalin-fixed, paraffin-embedded tissues sections (Table I and Ref. 15). To our knowledge, this is the largest quantitative proteomics data set of microdissected OSCC specimens reported to date.  At present, we cannot exclude the potential bias of our findings due to the inclusion of only male subjects in this study (16 patients ; three for MS experiments, 10 for IHC staining, and three for Western blotting) (supplemental Table S1) and the small number of patients (n ϭ 3) for the MS-based biomarker discovery. It is noted, however, that the majority of oral cancer patients in Taiwan are male (85-93%) (65,66). In addition, among the 53 up-regulated candidates identified from the three pairs of OSCC samples used here, more than 20 candidates have been confirmed and/or rediscovered to be overexpressed in OSCC tissues using a larger number of samples in the present study or by other groups (Ref. 15 and Table I and references cited therein). This observation indicates that, although potential bias may result from using specimens from a limited number of male patients, the quantitative proteome data set generated here can be useful for searching potential OSCC biomarkers.
Recently Siu and co-workers (16,39) used an isobaric mass tag (iTRAQ (isobaric tags for relative and absolute quantitation)) labeling method and LC-MS/MS to identify several differentially expressed proteins in OSCC. The authors identified a total of 811 non-redundant proteins in tumor tissues and described a panel of proteins displaying consistently differential expression in tumors relative to non-cancerous controls. Several up-regulated candidates (e.g. FSCN1, LDHA, SODM, S10A2, and K1C14) were also identified in this study (Table I and Ref. 16). The authors also used immunohistochemistry analysis to identify a panel of up-regulated proteins (e.g. 14-3-3 (stratifin), 14-3-3 , and calcium-binding protein S100A7) that could serve as potential markers for discriminating OSCC from non-cancerous tissues (16,39). In our study, 14-3-3 and 14-3-3 (but not S100A7) were shown in five independent experiments to have T/N ratios ranging from 1.9 to 3.68 and from 1.33 to 3.14, respectively, in three of five experiments (supplemental Table S2). When a more stringent filter (T/N Ն2 in three of five experiments and an average ratio Ն2.5) was applied, these proteins were not included in our final candidate list. The differential protein expression patterns determined using different stable isotope labeling techniques were consistent, suggesting that stable isotope-dependent quantitative proteomics methods are reliable and feasible quantitative platforms.
Most (38 of 53) of the proteins that were up-regulated in tumor cells contributed to tumor-related biological processes such as cell proliferation, communication, defense, organization and biogenesis, and cell motility. The biological implications of these differentially expressed protein candidates were extracted using the MetaCore data mining suite. The most significant biological network identified in this analysis was the cytoskeleton remodeling-keratin filament pathway with a Ϫlog p value of 19.364. The EPIPL and CK 6, 14, 16, and 17 proteins of this pathway were found to be up-regulated in oral cancer cells, whereas the EVPL, PEPL, and CK 4, 13, 15, and 19 proteins of this pathway were found to be down-regulated (Table III). Previous studies have shown that the transcription of CK6/16/17 can be induced in keratinocytes during wound healing (67,68). In patients with OSCC, the loss of CK 13 or 19 expression may increase recurrence and enhance invasiveness (69). Although the molecular mechanisms of CK expression and regulation in tumor development remain unclear, the alterations in CK expression were consistent among different analytical platforms, highlighting the importance of keratin expression and regulation in oral cancer progression. A systematic investigation of keratin filament regulation may shed light on the development of epithelia and on the relative malignancy of tumor cells.
The second significant network identified in this study was the type I IFN signaling pathway (Ϫlog p value of 5.695). Two key downstream regulators of this pathway (STAT1 and UCRP/ISG15) were identified and quantified in all five experiments with fold changes in expression in tumor and nontumor cells ranging from 3.94 to 8.37 (STAT1) and 8.71 to 38.99 (UCRP/ISG15), respectively. Hundreds of ISG proteins have been previously determined using genomics and proteomics approaches (54,55). Our current study showed that 10 of the 53 up-regulated candidate proteins belong to the ISG family (Table I), and increased mRNA or protein levels of seven proteins (SYWC, IFM1, STAT1, TYPH, UCRP, MX1, and GBP2) have been detected in OSCC via previous genomics or proteomics studies (9,10,70). Of these up-regulated ISG proteins, four (STAT1, UCRP, GBP1, and TYPH) were confirmed to be overexpressed in OSCC tissues via immunohis- tochemical staining (in at least eight of the 10 tissue pairs examined) (Fig. 5B). Most notably, IFN-␤ (the key upstream regulator of this pathway) was also confirmed to be overexpressed in the tumor cells of all OSCC tissue pairs (Fig. 7A). Although previous genomics and proteomics studies have confirmed the dysregulation of type I IFN signaling components in OSCC/HNSCC (9,10,70), it remains unclear whether the components of this pathway are systemically altered in OSCC/HNSCC and whether this pathway contributes to OSCC/HNSCC progression. Our use of quantitative proteomics approaches, biological network analyses, Western blot analyses, and immunohistochemistry analyses provides strong evidence that this pathway is significantly enhanced in OSCC tumor cells.
Interferons are categorized as two types (type I (IFN-␣ and IFN-␤) and type II (IFN-␥)). These molecules are multifunctional cytokines that possess antiviral, antiproliferative, and immunomodulatory activities (71)(72)(73). Type I IFNs are known to inhibit the growth of a variety of cancer cells, and this inhibition can be mediated, at least in part, by the Jak-STATmediated cell death pathway (74,75). However, other studies have shown that some cancer cells are resistant to IFN-␣/␤mediated antiproliferation, which may be attributed to the deregulation of the Jak-STAT, NF-B, and phosphatidylinositol 3-kinase/AKT pathways in these cells (76 -78). In this study, we found that OSCC cells respond to IFN-␤ by activating downstream target genes and increasing protein ISGylation but that these cells are resistant to IFN-␤-mediated inhibition of cell growth (Fig. 7, B and C, and supplemental Fig. S5). Interestingly we also found that the migration ability of OSCC cells was enhanced after exposure to IFN-␤ (Fig.  7D). Collectively these observations suggest that the overexpression of IFN-␤ in OSCC tissues may have unexpected and profound effects on OSCC cells. This intriguing possibility warrants further investigation.
The TYPH protein is overexpressed in a wide variety of solid tumors. This protein can be induced by several cytokines, including IFNs, and contributes to angiogenesis (79,80). In addition, STAT1 is a key regulator of the IFN signaling pathway and is known to be overexpressed in OSCC tissues (70,81). UCRP/ISG15, a critical molecule in the IFN-mediated immune response against antiviral infection, was recently identified as a novel tumor marker candidate in bladder and breast cancers (82,83). We show here that UCRP/ ISG15 was highly expressed in OSCC lesions at different sites of the oral cavity (Figs. 7 and 8, supplemental Fig. S2, and supplemental Table S4). Previous studies have shown that UCRP/ISG15 interferes with the ubiquitin/proteasome pathway and alters the sensitivity of tumor cells to camptothecin, an antitumor drug. This interference presumably involves ISGylation, which modifies the functions of various cellular proteins (83)(84)(85). We detected increased levels of UCRP/ISG15 and its protein conjugates in IFN-␤-stimulated OSCC cells as well as in three oral tumor tissues studied (Fig. 7C). Approximately 200 ISG15-conjugated proteins (ISCPs), which are known to participate in modulating diverse cell functions, have been identified in eukaryotic cells using a combination of double affinity purification and MSbased proteomics approaches (59,60). In addition, we detected 16 differentially expressed ISCPs in OSCC tissues, including 13 up-regulated (SYWC, STAT1, FLNA, FLNB, MX1, GBP1, EPIPL, LDHA, ANXA3, HSP47, PML, ACTN1, and AMPL) and three down-regulated (HSP71, SERA, and 6PGD) proteins (Tables I and II). Collectively these findings raise the intriguing possibility that overexpressed IFN-␤, UCRP/ISG15, and ISCPs might modify certain properties of OSCC cells, such as their sensitivity to chemotherapy. Further studies are needed to explore this possibility.