Discovery and Verification of Head-and-neck Cancer Biomarkers by Differential Protein Expression Analysis Using iTRAQ Labeling, Multidimensional Liquid Chromatography, and Tandem Mass Spectrometry*S

Multidimensional LC-MS/MS has been used for the analysis of biological samples labeled with isobaric mass tags for relative and absolute quantitation (iTRAQ) to identify proteins that are differentially expressed in human head-and-neck squamous cell carcinomas (HNSCCs) in relation to non-cancerous head-and-neck tissues (controls) for cancer biomarker discovery. Fifteen individual samples (cancer and non-cancerous tissues) were compared against a pooled non-cancerous control (prepared by pooling equal amounts of proteins from six non-cancerous tissues) in five sets by on-line and off-line separation. We identified 811 non-redundant proteins in HNSCCs, including structural proteins, signaling components, enzymes, receptors, transcription factors, and chaperones. A panel of proteins showing consistent differential expression in HNSCC relative to the non-cancerous controls was discovered. Some of the proteins include stratifin (14-3-3σ); YWHAZ (14-3-3ζ); three calcium-binding proteins of the S100 family, S100-A2, S100-A7 (psoriasin), and S100-A11 (calgizarrin); prothymosin α (PTHA); l-lactate dehydrogenase A chain; glutathione S-transferase Pi; APC-binding protein EB1; and fascin. Peroxiredoxin2, carbonic anhydrase I, flavin reductase, histone H3, and polybromo-1D (BAF180) were underexpressed in HNSCCs. A panel of the three best performing biomarkers, YWHAZ, stratifin, and S100-A7, achieved a sensitivity of 0.92 and a specificity of 0.91 in discriminating cancerous from non-cancerous head-and-neck tissues. Verification of differential expression of YWHAZ, stratifin, and S100-A7 proteins in clinical samples of HNSCCs and paired and non-paired non-cancerous tissues by immunohistochemistry, immunoblotting, and RT-PCR confirmed their overexpression in head-and-neck cancer. Verification of YWHAZ, stratifin, and S100-A7 in an independent set of HNSCCs achieved a sensitivity of 0.92 and a specificity of 0.87 in discriminating cancerous from non-cancerous head-and-neck tissues, thereby confirming their overexpressions and utility as credible cancer biomarkers.

Annually about 500,000 cancer-related deaths are estimated in the United States alone; of these ϳ13,000 are attributed to head-and-neck squamous cell carcinoma (HNSCC), 1 making it the sixth most common cause of cancer deaths (1). A lack of biomarkers for early detection and risk assessment is clearly reflected by the fact that more than 50% of all HNSCC patients have advanced disease at the time of diagnosis (2). The 5-year survival rate of HNSCC patients is less than 50%, and the prognosis of advanced HNSCC cases has not changed much over the past 3 decades (except in a few advanced centers) (2,3). Conceivably improvement in understanding the steps leading to tumorigenesis will provide the ability to identify and predict malignant progression at an earlier stage of HNSCC lesions, in turn leading to more effective treatment and reduction of morbidity and mortality. Currently there are no clinically established tumor markers available to facilitate the diagnosis or prognosis of head-and-neck cancer. It is expected that identification of novel protein markers or therapeutic targets will ultimately improve patient care and survival.
Proteomics combined with MS has become a powerful paradigm for the examination of proteins in a global manner in the postgenomics era, and thereby the discovery of cancer biomarkers and drug targets. Although transcriptomics provides the tool for unraveling gene expression networks, proteomics links these networks to protein products and provides further insight into posttranslational modifications that regulate cellular functions, thereby complementing genomics analyses (for a review, see Ref. 4). Identification of differentially expressed proteins in HNSCCs using proteomics revealed that expression patterns of proteins may have some predictive power for clinical outcome and personalized risk assessment (4 -11).
Differential tagging with isotopic reagents, such as ICAT (12) or the more recent variation that uses isobaric tagging reagents (iTRAQ, Applied Biosystems, Foster City, CA), followed by multidimensional LC and MS/MS analysis is emerging as one of the more powerful methodologies in the search for disease biomarkers. Our recent studies using both ICAT and iTRAQ reagents resulted in identification and relative quantification of proteins leading to discovery of potential cancer markers for endometrial cancer (13)(14)(15)(16)(17). Herein we have extended the use of iTRAQ labeling in combination with multidimensional LC-MS/MS analysis to head-and-neck cancer for comparison of protein profiles of HNSCC and noncancerous head-and-neck tissues in an attempt to identify potential biomarkers, as well as to identify in a global fashion molecular pathways that are deregulated in head-and-neck cancer, which in turn should aid in drug target discovery. The iTRAQ experiments were performed on resected HNSCC and non-cancerous tissue homogenates. The rationale for using whole tissue homogenates as opposed to laser capture microdissection-procured tumor cells has been discussed previously (16). Major advantages in the analysis of tissue homogenates are that the relevant proteins are much more abundant in the tissues of interest than in bodily fluids, and there is an automatic link between a protein that is differentially expressed and the tumor itself. Such a link would need to be demonstrated if the differentially expressed proteins were to be discovered in a bodily fluid, e.g. blood, as every tissue or organ can potentially discharge into blood. Furthermore the tumor microenvironment plays an important role in cancer progression (18), and the examination of protein expression in tissues from a homogenate of different cell types takes into account the contributions of the tumor microenvironment.
The protein expression profiles of HNSCCs were compared with non-cancerous head-and-neck tissues (controls) using the iTRAQ labeling technique in combination with multidimensional LC-MS/MS analysis. In the iTRAQ technology, tagging is on primary amines, thus potentially allowing the tagging of most tryptic peptides. The multiplexing ability afforded by the iTRAQ reagents, which are available in four different tags, is ideally suited for our study because it provides us with a means to perform a proteomics analysis of both paired and non-paired non-cancerous (histologically normal) head-andneck tissues while simultaneously comparing them against the cancer samples. This strategy helps to identify proteins that might be differentially expressed due to manifestation of field cancerization (8,19) in clinically normal mucosa and may be useful in designing strategies for risk prediction of disease recurrence or second primary tumor development.
Some of the overexpressed proteins that we identified in the tissues by the iTRAQ technology and LC-MS/MS analysis were confirmed by immunohistochemistry and Western blotting. These approaches ensured that the proteins selected demonstrate a consistent pattern of overexpression in HNSCCs and greatly increased the confidence of the observations stemming from iTRAQ analysis. Besides their potential utility as biomarkers for HNSCC, these proteins also provide valuable insight into the still unknown molecular networks and mechanisms that govern the normal-tomalignant conversion of the epithelium.

EXPERIMENTAL PROCEDURES
Samples and Reagents-Head-and-neck cancer tissues were retrieved from an in-house, dedicated, research head-and-neck tissue bank with approval from the Human Ethics Committee of All India Institute of Medical Sciences, New Delhi, India. With patient consent, surgically resected specimens of HNSCCs and paired non-cancerous tissues (each taken from a distant site) were collected and banked from patients undergoing curative surgery at the Department of Otorhinolaryngology, All India Institute of Medical Sciences. Normal tissues with no evidence of cancer (non-paired non-cancerous controls) were collected from patients attending the Dental Outpatient Department of All India Institute of Medical Sciences for tooth extraction after consent of the patients. After excision, tissues were flash frozen in liquid nitrogen within 20 min of devitalization and stored at Ϫ80°C until further use; one tissue piece was collected in 10% formalin and embedded in paraffin for histopathological analysis. The clinical and pathological data were recorded in a predesigned form. These included clinical tumor, node, and metastasis staging (based on the International Union Against Cancer's classification of malignant tumors (20)); site of the lesion, histopathological differentiation; age; and gender of the patients.
The histologic diagnosis for each sample was reconfirmed using microscopic examination of a hematoxylin-and eosin-stained frozen section of each research tissue block. The tissue from the mirror face of the histologic section was then washed three times in ϳ1 ml of PBS with a mixture of protease inhibitors (1 mM 4-(2-aminoethyl)benzenesulfonyl fluoride, 10 M leupeptin, 1 g/ml aprotinin, and 1 M pepstatin) as described previously (13). The washed tissue was then homogenized in 0.5 ml of PBS with protease inhibitors using a handheld homogenizer. These homogenates were then flash frozen in liquid nitrogen and stored at Ϫ80°C until use. Samples were thawed and clarified by centrifugation, and the protein concentration was determined by a Bradford-type assay using Bio-Rad protein quantification reagent.
The iTRAQ experiments were performed in five sets of four samples each. A pool of non-paired non-cancerous head-and-neck tissue homogenates was used as a control in each set of experiments: equal amounts of total protein from the lysates of six non-cancerous samples (non-paired controls) were pooled to generate a common reference "control sample" against which all the HNSCC samples were compared. Each sample contained 200 g of proteins. Trypsin digestion and labeling were performed according to the manufacturer's (Applied Biosystems) protocol; however, as we were using double the manufacturer's recommended amounts, we used two individual vials of each reagent for labeling each sample. iTRAQ labeling was performed as follows: control (non-paired non-cancerous pool), iTRAQ reagent 114; two cancer samples, iTRAQ reagents 115 and 117; individual non-cancerous tissue sample (paired or non-paired sample), iTRAQ reagent 116. A total of five iTRAQ sets were analyzed resulting in 10 cancer (five buccal mucosa and five tongue) samples and two paired non-cancerous plus three non-paired non-cancerous samples being compared with the control sample. The paired noncancerous samples originated from patients with cancer that were resected from sites a minimum of 2 cm away from the advancing edge of the cancer. Each iTRAQ set was analyzed with one run each of on-line 2D LC-MS/MS and off-line 2D LC-MS/MS analyses.
Strong Cation Exchange (SCX) Separation Conditions-For the off-line 2D LC-MS/MS analysis, each set of labeled samples was first separated by SCX fractionation using an HP1050 high performance liquid chromatograph (Agilent, Palo Alto, CA) with a 2.1-mm-internal diameter ϫ 100-mm-length Polysulfoethyl A column packed with 5-m beads with 300-Å pores (The Nest Group, Southborough, MA) as described previously (17). A 2.1-mm-internal diameter ϫ 10-mmlength guard column of the same material was fitted immediately upstream of the analytical column. Separation was performed as described previously (17). Briefly each pooled sample set was diluted with the loading buffer (15 mM KH 2 PO 4 in 25% acetonitrile, pH 3.0) to a total volume of 2 ml, and the pH was adjusted to 3.0 with phosphoric acid. Samples were then filtered using a 0.45-m syringe filter (Millipore, Cambridge, Ontario, Canada) before loading onto the column. Separation was performed using a linear binary gradient over 1 h. Buffer A was identical in composition to the loading buffer; Buffer B was Buffer A containing 350 mM KCl. Fractions were collected every 2 min using an SF-2120 Super Fraction Collector (Advantec MFS, Dublin, CA) after an initial wait of 2 min to accommodate the void volume. This resulted in a total of 30 SCX fractions per sample set. These fractions were dried by speed vacuuming (Thermo Savant SC110 A, Holbrook, NY) and resuspended in 30 l of 0.1% formic acid each.
For the on-line 2D LC-MS/MS analysis, an SCX cartridge (BioX-SCX, LC Packings, Amsterdam, The Netherlands) was plumbed upstream of the reverse phase (RP) desalting cartridge and analytical column. This SCX cartridge was connected through a second valve on the Switchos unit as shown in Fig. 1. Samples were separated on this SCX cartridge using 10-l step elutions with increasing concentration of ammonium acetate (10 mM, 50 mM, 100 mM, 150 mM, 200 mM, 250 mM, 300 mM, 350 mM, 500 mM, and 1 M). Each step elution was loaded onto the RP desalting column using the switching program as shown in Fig. 1 where the eluting peptides were desalted before loading onto the analytical column that was subsequently brought in line with the desalting column. The flow path used for these steps was designed to ensure that there was never any flow reversal through either of the cartridges (SCX or RP). Separation on the RP analytical column was effected as described for the second stage of the off-line LC-MS/MS analysis described below.
LC-MS/MS Run Conditions-The SCX fractions from 6 to 30 were analyzed by nano-LC-MS/MS using the LC Packings Ultimate instrument fitted with a 10-l sample loop. Samples were loaded, using a microliter pick-up mode, onto a 5-mm RP C 18 precolumn (LC Packings) at 50 l/min and washed for 4 min before switching the precolumn in line with the separation column. The separation column used was either a 75-m-internal diameter ϫ 150-mm-length PepMap RP column from LC Packings packed with 3-m C 18 beads with 100-Å pores or an in-house equivalent packed with similar beads (Kromasil; The Nest Group). The flow rate used for separation on the RP column was 200 nl/min with the following gradient: 0 min, 5% B; 10 min, 5% B; 15 min, 15% B; 125 min, 35% B; 145 min, 60% B; 150 min, 80% B; 160 min, 80% B; 162 min, 5% B; 188 min, stop.
Samples were analyzed on a QSTAR Pulsar i mass spectrometer (Applied Biosystems/MDS SCIEX, Foster City, CA) in information-dependent acquisition mode with the scan cycles set up to perform a 1-s MS scan followed by five MS/MS scans of the five most abundant peaks for 2 s each. Every fourth scan the peak that was closest in intensity to the threshold of 10 counts was selected for MS/MS. Data acquisition was performed without any repetitions and with a dynamic exclusion of 30 s. Relative protein abundances were determined using the MS/MS scans of iTRAQ-labeled peptides (17). The iTRAQlabeled peptides fragmented under CID conditions to give reporter ions at 114.1, 115.1, 116.1, and 117.1 Th. The ratios of peak areas of the iTRAQ reporter ions reflect the relative abundances of the peptides and, consequently, the proteins in the samples. Larger, sequence information-rich fragment ions were also produced under these conditions and gave the identity of the protein from which the peptide was derived.
Data Analysis-The software used for data acquisition was Analyst QS 1.1 (Applied Biosystems/MDS SCIEX). Data were analyzed using ProteinPilot (17,21), and the database searched was the Celera human database (human KBMS, November 9, 2004) with a total of 178,243 entries; both were provided by Applied Biosystems. Identified proteins were grouped by the software to minimize redundancy. All peptides used for the calculation of protein ratios were unique to the given protein or proteins within the group; peptides that were common to other isoforms or proteins of the same family that were reported separately were ignored. The ProteinPilot cutoff score used was 1.3, which corresponds to a confidence limit of 95%. Userdefined options used included: (i) cysteine alkylation, (ii) trypsin digestion, (iii) no special factors, (iv) human proteins, (v) biological modifications, and (vi) thorough identification search.
Statistical Analysis-The average iTRAQ ratios from different runs were calculated for each protein in the off-line and on-line analyses. Thereafter the iTRAQ ratios for each protein in the two analyses were averaged. Proteins that were selected for further analysis met the following criteria: 1) detection in Ն6 of the 10 cancer samples, Ն50% of which showed differential expression Ն1.5-fold relative to the control sample, and/or 2) known to be of interest from other studies. These proteins are listed in Table I along with two housekeeping proteins (to contrast the performance of the potential biomarkers).
Biomarker Panel Analysis-To identify a panel of best performing proteins that can distinguish between HNSCC and non-cancerous tissues, each protein in Table I was individually assessed for its ability to discriminate between normal and cancer samples by evaluating its receiver operating characteristic (ROC) curve based on the iTRAQ ratios. Plotting ROC curves and calculating the area under the curve (AUC) and other attributes were performed using the ROCR package within the R statistical computing environment (The R Project for Statistical Computing). Proteins giving the highest AUC values were selected for biomarker panel analysis and used as input variables into a naïve Bayes model implemented in Java (Sun Developer Network) using the WEKA package (24). Given a sample i that has iTRAQ ratios (or IHC scores; see below) in the vector x(i), the naïve Bayes model has the form p͑i ϭ cancer͉x͑i͒͒ ϭ p͑cancer͒ ϫ p͑x͑i͉͒͒͑i ϭ cancer͒p͑x͑i͒͒ where p(i ϭ cancer͉x(i)) is the probability that i is a cancer sample given its x(i) values. This is the posterior probability and is calculated using Bayes' theorem. A value Ն0.5 is considered a positive hit. p(x(i)͉i ϭ cancer) is the probability that within the cancer samples, x(i) exists within them. p(cancer) is the probability of i being a cancer sample; this is the prior probability. p(x(i)) is the probability of i occurring and is a normalization factor. Nine trials of 3-fold crossvalidation were used for each biomarker panel input into the naïve Bayes model. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for each trial.
Verification of Candidate Potential Cancer Markers by Immunohistochemistry-The three best performing proteins from the above biomarker panel analysis were selected for immunohistochemical ver-ification using an independent, larger sample set. Antibodies against these three biomarkers were available commercially (Santa Cruz Biotechnology Inc., Santa Cruz, CA). Each antibody was first optimized with respect to dilution and the use of microwave heating in citrate buffer (0.01 M, pH 6.0) to expose the antigen ("antigen retrieval"). Paraffin-embedded sections (5 m) of human HNSCCs (25 cases) and paired head-and-neck non-cancerous tissues from these patients (25 samples) as well as non-paired non-cancerous head-andneck tissues (10 samples) were collected on gelatin-coated slides. For histopathological analysis, representative sections were stained with hematoxylin and eosin; immunostaining was done on serial sections as described previously (25). Following the application of a protein blocker for 10 min, deparaffinized tissue sections were first incubated with the primary antibodies for 1 h at room temperature or for 16 h at 4°C followed by the respective secondary antibody conjugated with biotin. The primary antibody was detected using the streptavidin-biotin complex (DAKO LSABϩ kit, DAKO Cytomation, Glostrup, Denmark) and diaminobenzidine as chromogen. Slides were washed with three times TBS (0.1 M, pH ϭ 7.4) after every step. Finally the sections were counterstained with Mayer's hematoxylin and mounted with DPX mountant. In the negative controls, the primary antibody was replaced by non-immune mouse IgG of the same isotype to ensure specificity. HNSCC tissue sections with known immunopositivity for specific proteins were used as positive controls in each batch of sections analyzed (25).
Evaluation of Immunohistochemical Staining-The immunopositive staining was evaluated in five areas. Sections were scored as positive if epithelial cells showed immunopositivity in the cytoplasm, plasma membrane, and/or nucleus when judged independently by two scorers who were blinded to the clinical outcome, i.e. the slides were coded, and the pathologists did not have prior knowledge of the local tumor burden, lymphonodular spread, and grading of the tissue samples while scoring the immunoreactivity. First, a quantitative score was performed by estimating the percentage of immunopositive stained cells: 0, Ͻ10% cells; 1, 10 -30% cells; 2, 30 -50% cells; 3, 50 -70% cells; and 4, Ͼ70% cells. Second, the intensity of staining was scored by evaluating the average staining intensity of the positive cells (0, none; 1, weak; 2, intermediate; and 3, strong). Finally a total score (ranging from 0 to 7) was obtained by adding the quantitative score and the intensity score for each of the 60 sections. The immunohistochemical data were subjected to statistical analysis as described above for the MS results.
Western Blot Analysis of Proteins in HNSCCs and Normal Tissues-Whole cell lysates were prepared from five HNSCCs and five non-cancerous head-and-neck tissues. Frozen tissue samples were homogenized and lysed in a buffer containing 50 mM Tris-Cl (pH 7.5), 150 mM NaCl, 10 mM MgCl 2 , 1 mM ethylenediamine tetraacetate (pH 8.0), 1% Nonidet P-40, 100 mM sodium fluoride, 1 mM phenylmethanesulfonyl fluoride, and 2 l/ml protease inhibitor mixture (Sigma). Protein concentrations were determined using the Bradford reagent (Sigma), and equal amounts of proteins (80 g/lane) from the HNSCCs and non-cancerous tissues were resolved on a 12% SDSpolyacrylamide gel. The proteins were then electrotransferred onto PVDF membranes. After blocking with 5% nonfat powdered milk in TBS (0.1 M, pH ϭ 7.4), blots were incubated with the respective primary antibodies (1:200 dilution) at 4°C overnight. The protein abundance of ␣-tubulin was used as a control for protein loading and was determined with mouse monoclonal anti-␣-tubulin antibody (Clone B7, Santa Cruz Biotechnology Inc.). Membranes were incubated with the respective secondary antibody, horseradish peroxidase-conjugated rabbit/goat/mouse anti-IgG (goat anti-rabbit IgG, 1:5000; rabbit anti-goat IgG, 1:4000; or rabbit anti-mouse IgG, 1:2000; DAKO Cytomation), and diluted with 1% bovine serum albumin for 2 h at room temperature. After each step, blots were washed three times with 0.2% Tween, TBS. Protein bands were detected by the enhanced chemiluminescence method (Santa Cruz Biotechnology Inc.) on X-Omat film.

RESULTS
The off-line and on-line 2D LC-MS/MS analyses collectively resulted in the identification of a total of 811 non-redundant proteins. Only a few of these proteins displayed consistent differential expression in the HNSCC samples (measured in Ն6 of the 10 samples and with Ն50% showing Ն1.5-fold differential expression relative to the control sample) that warranted further analysis. These proteins, all confidently identified with Ն2 peptide matches (except adenomatous polyposis coli protein (APC)-binding protein end-binding protein 1 (EB1) and superoxide dismutase (manganese)), are given in Table I along with two structural proteins, actin and ␤-2tubulin, as controls. (See supplemental Table S1 for peptide sequences and coverage and supplemental Figs. S1 and S2 for the CID spectra of the single peptide identifications.) As the nano-LC analyses were performed on 25 SCX fractions, the acquired data files were searched in two groups out of necessity (the version of ProteinPilot software available at that time was incapable of handling a large number of data files, each with a large amount of data): fractions 6 -15 were searched in one group, and fractions 16 -30 were searched in a second group. The ProteinPilot result files from these two halves were then exported into an Excel spreadsheet where the proteins of interest from the two searches were combined by averaging the ratios for the protein in each sample. It is noteworthy that each of the ratios reported by searching either half of the fractions is itself comprised of the ratios from multiple peptides identified in the given protein. ProteinPilot automatically only includes unique and high confidence matches of peptides for any particular protein in the ratios reported (i.e. it excludes those that are shared between different isoforms of any protein or low confidence matches to peptides). These averaged ratios from the off-line and on-line analyses were then again averaged and reported in Table I. Of all the individual expression ratios (two off-line and one on-line) that were used in the calculations of the ratios reported, 56.4% varied by less than 10% from their respective average shown, and 82.0% varied by less than 20%. It is reassuring that the expression ratios from different analyses and separate handling were comparable. Nine proteins that did not meet the cutoff criteria stated above, cytokeratin 14, polybromo-1D, PKM2, annexin A1, nucleophosmin 1, Hsp27, cystatin B, GRP 94, and MARCKS, were also included in Table I for further analysis as these had been reported in head-and-neck cancer or are of biological relevance in cancer. The HNSCCs analyzed included five squamous cell carcinomas (SCCs) of buccal mucosa and five SCCs of the tongue. The rationale for the choice of these two SCC types was to determine whether there are site-specific protein expressions or not. The best performing proteins that can differentiate between HNSCC and non-cancerous tissues were identified by determining the individual ROC curves of the proteins in Table I (as described under "Experimental Procedures"). The three proteins with the highest AUC values, YWHAZ, stratifin, and S100-A7, are listed in Table II together with their individual and collective values of merit, including sensitivity (cancer samples correctly identified as cancer samples) and specificity (normal samples correctly identified as normal samples). As a panel, the three best performing biomarkers achieved a sensitivity of 0.92 and a specificity of 0.91 in discriminating HNSCC from non-cancerous head-andneck tissues (Table II and Fig. 2a). A number of proteins, e.g. prothymosin ␣ and APC-binding protein EB1, were predominantly overexpressed in SCCs of buccal mucosa (Table I) and showed some promise in differentiating between SCCs of buccal mucosa and the tongue; however, as the number of samples is small, this possibility will need to be fully investigated in a future study involving more samples of both types.
Verification of candidate protein biomarkers is a necessary step in moving from the initial discovery to possible application. The panel of three best performing biomarkers identified by MS analysis, YWHAZ, stratifin, and S100-A7, were chosen for verification in a different and larger set of HNSCCs and non-cancerous head-and-neck tissues. Verification exercises included immunohistochemical (Fig. 3) and Western blot analyses (Fig. 4) at the protein level as well as RT-PCR analysis (Fig. 5) at the mRNA level. All verification results support the above MS findings. In the immunohistochemical analysis, the biomarker panel of YWHAZ, stratifin, and S100-A7 achieved a sensitivity of 0.92 and a specificity of 0.87 (Table III and Fig.  2b) in discriminating HNSCCs from non-cancerous head-andneck tissues. The paired non-cancerous head-and-neck tissues obtained from HNSCC patients might have altered protein expressions prior to histological changes. To investigate this possibility, the non-cancerous tissues were segregated into paired and non-paired groups and evaluated separately with the HNSCCs. Significantly the panel of the three biomarkers, YWHAZ, stratifin, and S100-A7, appears to perform better in discriminating HNSCC tissues against the nonpaired non-cancerous head-and-neck tissues (sensitivity, 0.96; specificity, 0.96) than against the paired non-cancerous tissues (sensitivity, 0.92; specificity, 0.83) (see Table IV). These results appear to support the notion of protein expression alterations prior to histological changes and caution the use of only paired samples.
The specificity of the three-biomarker panel, YWHAZ, stratifin, and S100-A7, for head-and-neck cancer was assessed, as a first step, in three other types of cancers, esophageal squamous cell carcinoma (ESCC), invasive ductal carcinoma of the breast, and ovarian cancer, by means of immunohistochemical analyses (on five tumor samples each) and Western  FIG. 3. Immunohistochemical verification of iTRAQ-discovered potential cancer markers YWHAZ, stratifin, and S100-A7 in HNSCCs and non-cancerous head-and-neck tissues. Positive staining is brown and is intense in HNSCCs. The left panel shows the non-cancerous (histologically normal) tissues, and the right panel depicts the HNSCC tissue sections. A, the HNSCC sample shows intense cytoplasmic and nuclear staining for YWHAZ, whereas the normal mucosa shows no detectable immunostaining. B, the HNSCC tissue section shows cytoplasmic staining for stratifin in tumor cells, whereas the normal mucosa shows no detectable immunostaining. C, the HNSCC tissue section shows intense cytoplasmic staining for S100-A7 in tumor cells, whereas the normal mucosa shows no detectable immunoreactivity. All panels show ϫ200 magnifications.

FIG. 4. Western blot analyses of YWHAZ, stratifin, and S100-A7 in HNSCCs and paired non-cancerous head-and-neck tissues.
Equal amounts of protein lysates from HNSCCs and paired noncancerous head-and-neck tissues were used. See the text for details. The panels show increased expression of YWHAZ (i), stratifin (ii), and S100-A7 (iii) in HNSCCs (C1-C3) as compared with paired noncancerous head-and-neck tissues (N1-N3). ␣-Tubulin (iv) was used as the loading control.  N1  N2  N3  C1  C2  C3   N1  N2  N3  C1  C2  C3   N1  N2  N3  C1  C2  C3   N1  N2  N3  C1  C2  C3 FIG. 5. RT-PCR analyses of YWHAZ, stratifin, and S100-A7 in HNSCCs and non-cancerous head-and-neck tissues. i shows increased levels of YWHAZ transcripts in HNSCCs (C1-C3) as compared with the non-cancerous head-and-neck tissues that show basal levels (N2 and N3) and no detectable level (N1) of YWHAZ transcripts. ii shows increased levels of stratifin transcripts in HNSCCs (C1-C3) as compared with the non-cancerous head-andneck tissues that show a basal level (N3) and no detectable level (N1 and N2) of stratifin transcripts. iii shows increased levels of S100-A7 transcripts in HNSCCs (C1-C3) as compared with the non-cancerous head-and-neck tissues that show a basal level (N3) and no detectable level (N1 and N2) of S100-A7 transcripts. ␤-Actin (iv) was used as a control for normalizing the quantity of RNA used. be good specificity against breast and ovarian cancers although weaker discriminatory power against ESCC, which shares with head-and-neck/oral squamous cell carcinoma in being both squamous cell carcinomas and proximal in anatomical locations.

DISCUSSION
Multidimensional LC-MS/MS has been used for the analysis of clinical samples of HNSCCs labeled with isobaric mass tags (iTRAQ) to identify proteins that are differentially expressed in head-and-neck cancer in relation to non-cancerous head-and-neck tissues. The expression ratios were consistent between the on-line and off-line 2D LC-MS/MS methods used, demonstrating that the methodologies were rugged and reproducible even though the conditions and details used in peptide elution from the SCX columns in the two methods were different. Expectedly the number of proteins identified by the on-line analysis (431) was lower than the number identified by the off-line analysis (580) because of the lower capacity of the SCX cartridge used in the former (50 ng of total peptides versus up to 1 mg in the latter). However, the on-line analysis was advantageous in terms of shorter data acquisition time and lower amounts of total sample required.
Development of HNSCC is a multistep process that often involves field cancerization, a phenomenon in which not only the site of the primary tumor, but the entire mucosa of the upper aerodigestive tract, is prone to undergoing malignant transformation or progression at multiple sites (19). It is now evident that molecular changes underlying field cancerization are not localized to areas with altered histology but may persist beyond the histological border of precancerous lesions; a large fraction of the carcinogen-exposed field may harbor molecular aberrations without presenting clinical or morphological symptoms (Ref. 8 and references therein). Identification of proteins with altered expression as a manifestation of field cancerization is important in identification of biomarkers for prediction of risk of recurrence as well as for development of second primary tumors in patients treated for HNSCC. Thus, the selection of normal controls for HNSCCs in a differential expression analysis, including the current study, is not straightforward and requires careful planning. To address this issue, we included two types of non-cancerous histologically confirmed normal tissues in our analysis: 1) non-cancerous tissues obtained from HNSCC patients from a site distant to the tumor and 2) normal tissues obtained from individuals with no evidence of cancer or precancerous lesions. In a recent proteomics study, Roesch-Ely et al. (8) investigated changes in protein expressions occurring in different stages of tumorigenesis and field cancerization in HNSCCs. A number of reported differentially expressed proteins, including calgizarrin, stratifin, histone H4, and cystatin A, were also identified in this current study. To our best knowledge, this study is the first reporting differential expression of calmodulin-like protein 5, polybromo-1D, APC-binding protein EB1, ␣ 1 -antitrypsin precursor, carbonic anhydrase I, mast cell tryptase ␤ III, histone H2B.1, L-plastin, and peptidylprolyl isomerase A in HNSCC.
Among the differentially expressed proteins identified, no single protein emerged as a unique marker for HNSCC. However, a panel of three best performing biomarkers, YWHAZ, stratifin, and S100-A7, performed satisfactorily as determined by both MS and immunohistochemistry on independent sets of samples. Significantly YWHAZ has been identified previously by us to be overexpressed in oral cancer at the mRNA level and has subsequently been verified using immunohistochemical analysis (26). This serves as an independent validation of and complements the current results. Furthermore YWHAZ has also been reported to be overexpressed in stomach cancer (27) and in breast and prostate tumor model systems (28,29). More importantly, YWHAZ is not overexpressed in endometrial and lung cancer tissue samples (18,30), thus illustrating the fact that this protein can provide some selectivity in discriminating among different cancer types.
Stratifin has been reported to be overexpressed in HNSCC. A recent proteomics study reported a 3.6-fold stratifin overexpression (31), thus corroborating the results obtained in this study. A second independent study also showed stratifin overexpression in the range of 2.8 -9.1-fold in cancer samples (32). In addition, a study of 300 patients with pancreatic ductal adenocarcinoma showed stratifin overexpression in 82% of primary infiltrating adenocarcinomas, whereas another 15% showed weak immunopositivity. Overexpression of stratifin correlated with poor prognosis (33). Interestingly and significantly, stratifin was reported to be down-regulated in HNSCCs by Roesch-Ely et al. (8), whereas we observed consistent overexpression of stratifin in iTRAQ and in IHC verification analysis. The HNSCCs in the study of Roesch-Ely et al. (8) were from the German population with tobacco smoking and alcohol consumption being the major risk factors, whereas the clinical samples in this study and in Lo et al. (11,31) and Chen et al. (32) are from Asian populations where, in addition, chewing tobacco and/or betel quid and bidi smoking are important risk factors. These differences in the risk factors may account for the observed variations in stratifin expression and warrant in-depth investigation in a larger study.
14-3-3 proteins recognize phosphoserine/threonine-containing motifs used by a variety of signal transduction pathways to bind over 200 target proteins that play important roles in the regulation of various cellular processes, including mitogenic and cell survival signaling, cell cycle control and apoptotic cell death, epithelial-mesenchymal transition, and cell adhesion, invasion, and metastasis (34). The involvement of 14-3-3 proteins in the regulation of oncogenes and tumor suppressor genes points to a potential role in tumorigenesis (35); multiple pathways can be targeted by modulation of these proteins, underscoring their potential as candidate drug targets. Although it might be argued that 14-3-3 proteins are, therefore, too pleiotropic to be targets for therapeutic inhibition, it has been shown that simultaneous inactivation of all 14-3-3 proteins sensitizes cancer cells to DNA-damaging agents. Selective inactivation of stratifin leads to an increased sensitivity toward cancer chemotherapeutic agents. Recent studies have shown that stratifin forms homodimers, whereas YWHAZ forms homodimers and also heterodimers with other isoforms (36,37). Stratifin has been extensively investigated; by contrast, YWHAZ remains largely unexplored. It is noteworthy that the potential success of strategies aimed at modulating 14-3-3 availability in the cell for cancer therapy is demonstrated in studies showing that reducing cellular 14-3-3 increases chemosensitivity (38,39).
Prothymosin ␣, found to be overexpressed here in HNSCC, has also been implicated in several other cancers and in lymph nodes and tonsils (40 -48). It will be interesting to determine whether its expression in lymph nodes of HNSCC patients would correlate with locoregional spread of the disease and be a determinant of disease prognosis. S100-A7, a small calcium-binding protein of the S100 protein family, originally identified in psoriatic keratinocytes, is up-regulated in abnormally differentiating keratinocytes, in squamous carcinomas of different organs, and in a subset of breast tumors (49 -53). Incidentally S100-A7 was also identified in oral premalignant epithelia by microarray analysis and proposed to be a marker for invasion (49). It has been hypothesized to play a role in breast tumor progression by promoting angiogenesis and enhancing the selection of cells that overcome their anti-invasive function (53). This hypothesis has also been suggested to explain why S100-A7 expression is high in high grade or estrogen receptor-negative tumors as these are associated with increased hypoxia and reactive oxygen species, a scenario in which the angiogenic effects of S100-A7 are most important. It is noteworthy that increased hypoxia and reactive oxygen species also occur in head-andneck tumors and might explain the observed changes in S100-A7 expression here. Another study in breast cancer showed that BRCA1 is a transcriptional repressor of S100-A7; BRCA1 and c-Myc form a complex on S100-A7 promoter, and BRCA1-mediated repression is dependent on a functional c-Myc (54). Furthermore BRCA1 mutations in tumors abrogate the repression of S100-A7. In the absence of BRCA1, S100-A7 is induced by topoisomerase II poison and etoposide and increases the cellular sensitivity to etoposide, suggesting a mechanism for BRCA1-mediated resistance to etoposide (54). Incidentally BRCA1 alterations have been reported in head-and-neck cancer (55,56). However, a correlation, if any, between BRCA1 alterations and S100-A7 expression in head-and-neck cancer remains to be demonstrated.
Calgizzarin (S100-A11) has also been previously linked with cancer and was reported as a potential marker for endometrial cancer (14). Likewise S100-A2, which shows overexpression in HNSCC, is also known to be overexpressed in other forms of cancer, such as non-small cell lung cancer and uterine leiomyoma (57,58). It has been demonstrated that calgizzarin plays an antiapoptotic role in an uterine leiomyosarcoma cell line (58). We have reported previously that pyruvate kinase M2 is overexpressed in endometrial cancers (14,17). Several studies suggest that PKM2 is present primarily in a dimeric form in tumors and is useful as a biomarker in their early detection (59 -63). PKM2 overexpression in tumor cells is explained on the basis of its key role in the generation of ATP in the glycolytic pathway. Under hypoxic conditions that are typical for tumors, this pathway is a critical route by which tumors satisfy the higher energy requirements needed for proliferation (for reviews, see Refs. 64 and 65).
Two of the more interesting proteins discovered in this current HNSCC study are the APC-binding protein EB1 and polybromo-1D. EB1, initially discovered as a protein that binds APC at its C-terminal region, has also been shown to bind tubulin and associate with the microtubules that form the mitotic spindle during mitosis (66). Thus EB1 participates in microtubule-dependent processes, including intracellular vesicle trafficking, organization of organelles within the cell, and even cell migration (66). The EB1 interaction with APC is of particular interest as APC is a tumor suppressor whose inactivation leads to a significantly enhanced level of susceptibility for malignant transformation in colorectal cancer (66). One possible proposed explanation for the mechanism of action of EB1 in esophageal squamous cell carcinoma is that overexpression of EB1 correlates with the nuclear accumulation of ␤-catenin and affects the interaction between APC and ␤-catenin, indirectly causing the activation of the ␤-catenin/ T-cell factor pathway (67). It is, therefore, possible that overexpression of EB1 in this study could be the first evidence for the same process occurring in HNSCC.
Polybromo-1D (PB1), also known as BRG1-associated factor 180 (BAF180), is a relatively new member of the SWI/ SNF-B (PBAF) chromatin-remodeling complex that is a homolog of the yeast Rsc protein complex, which is required for progression through mitosis (68). In fact, antibodies against BAF180 localize to the kinetochores during mitosis (68). The fact that both PB1 and EB1 are known to be involved with mitosis is also noteworthy but requires further investigation to ascertain whether a direct relationship between the two exists. Other studies have shown that in yeast Rsc can act as an activator as well as a suppressor of transcription and that it can be functionally linked with the PKC pathway (22,69). Additionally it was shown that temperature-sensitive mutants of one of the proteins (Nps1) in the Rsc complex, when placed at the restrictive temperature, can be rescued by the overexpression of not only the yeast homolog of PKC, PKC1 (as well as other proteins downstream of the PKC1 signal pathway), but also by Bim1p, which is the yeast homolog of EB1 (22).
In addition to the possibility of this potentially significant link between PB1 and EB1, there is also independent evidence suggesting that PB1 is a tumor suppressor and that this activity is found in lung cancer but not in breast cancer (23). This was verified when transfection of BAF180 gene into breast tumor cell lines, possessing a truncated version of the same gene, resulted in growth inhibition (23). Other members of this complex that have been associated with cancer include hSNF5/INI1 and BRG1 itself. HSNF5/INI1 mutations have been found in malignant rhabdoid tumors, whereas mutations in BRG1 have been noted in various cell lines including carcinomas of the breast, lung, pancreas, and prostate. Implication of other members of the PBAF complex in suppression of various cancers, in addition to the above evidence that suggests that PB1 may be a tumor suppressor itself, makes PB1 an exciting discovery in our study. In light of the suggested tumor suppressor role of PB1, our observed lower expression of this protein in the HNSCC samples is consistent with expectations and warrants in-depth investigation of its role in head-and-neck tumorigenesis.
In conclusion, the use of iTRAQ labeling of head-and-neck cancers combined with LC-MS/MS has led to the discovery of several novel, differentially expressed proteins in these tumors. A panel of the three best performing biomarkers achieved a sensitivity of 0.92 and a specificity of 0.91. This performance was verified using immunohistochemistry on a larger, independent set of clinical samples of HNSCCs. A preliminary analysis of the specificity of this panel shows good discriminatory power against breast and ovarian cancers although weaker power against esophageal squamous cell carcinoma. Some of the most interesting new potential biomarkers warrant additional in-depth studies. * This work was supported, in whole or in part, by the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.