Surface-Enhanced Laser Desorption/Ionization Timeof-Flight Mass Spectrometry Serum Protein Profiling to Identify Nasopharyngeal Carcinoma

Transcription

1 99 Surface-Enhanced Laser Desorption/Ionization Timeof-Flight Mass Spectrometry Serum Protein Profiling to Identify Nasopharyngeal Carcinoma David Wing Yuen Ho, MPhil 1 Zhen Fan Yang, Ph.D. 1 Birgitta Yee-Hang Wong, MBBS 1 Dora Lai-Wan Kwong, MBBS 2 Jonathan Shun-Tong Sham, MD 2 William Ignace Wei, MS 1 Anthony Po Wing Yuen, MS 1 1 Department of Surgery, University of Hong Kong, Pokfulam, Hong Kong, China. 2 Department of Clinical Oncology, University of Hong Kong, Pokfulam, Hong Kong, China. BACKGROUND. Diagnosis of nasopharyngeal carcinoma (NPC) at an early disease stage is important for successful treatment and improving the outcome of patients. The use of serum protein profiles and a classification tree algorithm were explored to distinguish NPC from noncancer. METHODS. Serum samples were applied to metal affinity protein chips to generate mass spectra by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS). Protein peak identification and clustering were performed using the Biomarker Wizard software. Proteomic spectra of serum samples from 50 NPC patients and 54 noncancer controls were used as a training set and a classification tree with 6 distinct protein masses was generated by using Biomarker Pattern software. The validity of the classification tree was then challenged with a blind test set including another 20 NPC patients and 25 noncancer controls. RESULTS. The software identified an average of 93 mass peaks/spectrum and 6 of the identified peaks were used to construct the classification tree. The classification tree correctly determined 83% (123 of 149) of the test samples with 83% (58 of 70) of the NPC samples and 82% (65 of 79) of the noncancer samples. In a combination of the serum protein profiles with Epstein Barr (EBV) nuclear antigen 1 (EBNA1 IgA) test, the diagnostic sensitivity and specificity were increased to 99% and 96%, respectively. CONCLUSIONS. The results suggest that SELDI-TOF-MS serum protein profiles could discriminate NPC from noncancer. The combination of serum protein profiles with an EBV antibody serology test could further improve the accuracy of NPC screening. Cancer 2006;107: American Cancer Society. KEYWORDS: biomarker, classification and regression tree (CART), nasopharyngeal carcinoma (NPC), surface-enhanced laser desorption/ionization (SELDI), mass spectrometry. Supported by the Research Grant Council of the Hong Kong Special Administrative Region (Project No. HKU 7501/03M), a Michael Kadoorie Cancer Genetics Grant, and a small project grant from the University of Hong Kong. Address for reprints: Anthony Po Wing Yuen, MS, Department of Surgery, Queen Mary Hospital, 102 Pokfulam Road, Hong Kong, China; Fax: (011) ; pwyuen@hkucc.hku.hk Received October 25, 2005; revision received January 14, 2006; accepted February 21, Nasopharyngeal carcinoma (NPC) is one of the most common cancers in Hong Kong. 1 It ranks fourth for males and tenth for females, and accounts for 5% of all new cancer cases in Hong Kong. 2 In addition, it has a tendency to affect middle-aged people more than most other cancers, and this results in a substantial loss of the working life in the community. The current effective treatment for NPC is achieved by radiotherapy, but the outcome is related to the extent of the disease. Patients with the tumor confined within the nasopharynx can achieve a 5-year overall survival rate of 85% (Ho Stages I and II); however, those with distant dissemination can only achieve 55% (Ho Stages III and IV) American Cancer Society DOI /cncr Published online 17 May 2006 in Wiley InterScience (

2 100 CANCER July 1, 2006 / Volume 107 / Number 1 Unfortunately, the symptoms of early-stage NPC are usually minimal and innocuous, and the nasopharynx cannot be visualized by the patients and family physicians. Many NPC patients are diagnosed at Stages III-IV, when the patients visit the otorhinolaryngologist. Therefore, early detection and diagnosis of NPC is crucial for a better outcome of the patients. 4 Routine clinical methods of examination for nasopharyngeal diseases, such as the use of nasoendoscopy, however, are not applicable as a screening tool because they can only be done by an otorhinolaryngologist and are not cost effective. Serological tests, detecting the antibodies to Epstein Barr virus (EBV), such as viral capsid antigen (VCA) immunoglobulin A (IgA), early antigen (EA) IgA, and Epstein Barr nuclear antigen (EBNA1) IgA have been used routinely as serological screening markers in high-risk populations. Nevertheless, none of them has proven to be a standalone and reliable assay due to either low sensitivity or specificity. 5,6 Therefore, identification of additional biomarkers is important for the early detection and management of this disease. The rapid development in clinical proteomics, which detects the differences of protein patterns in tissue and body fluid specimens between patients and healthy people, presents an alternative path for biomarker discovery. Two-dimensional gel electrophoresis is a gold standard method that is able to resolve thousands of proteins and provides the highest resolution. However, it is labor-intensive and timeconsuming, and thus is not easily applied in the clinical setting. Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF- MS) is a recent breakthrough in proteomic research. This technique was developed by Ciphergen Biosystems (Fremont, CA). This system detects proteins that are selectively adsorbed onto the surface of a protein chip array after removing impurities by washing with buffer. The system has been shown to be sensitive, rapid, and reproducible. 7 The application of this innovative technique has been successful in the discovery of biomarkers in prostate, breast, ovarian, bladder, and liver cancers Furthermore, the construction of a classification tree using the artificial intelligence algorithms for processing the SELDI data improves the accuracy in differentiating prostate, ovarian, bladder, and liver cancers from the noncancer groups Therefore, the objective of this study was to explore whether a discriminatory protein pattern identified by SELDI protein profiling coupled with an artificial intelligence data analysis algorithm could effectively differentiate NPC from noncancer. TABLE 1 Demographics of the NPC Patients and Controls Included in the Study for the Classification and Regression (CART) Analysis Groups Number Stages Number of EBNA1 IgA Sex (M/F) Median Age (Range), Years NPC 43 I and II 37 29/14 49 (22-90) 27 III and IV 22 19/8 48 (32-75) Noncancer Normal 43 NT 25/18 46 (21-77) BED /13 46 (21-79) NPC indicates nasopharyngeal carcinoma; EBNA1, Epstein-Barr nuclear antigen 1; IgA, immunoglobulin A; M, male; F, female; NT, not tested; BED, benign ear, nose and throat diseases. MATERIALS AND METHODS Serum Samples This study was approved by the Ethics Committee of the University of Hong Kong. All NPC patients and noncancer (a group of people with no evidence of cancer) donors involved in the study signed an agreement form consenting to the donation of their specimens. All patients and controls were Chinese. Serum samples were collected from 70 patients with NPC from the year 2003 to 2004 before radiotherapy and 79 noncancer controls. There were 43 patients with early disease (Stages I and II, American Joint Committee on Cancer [AJCC] staging system) and 27 patients with advanced disease (Stages III and IV). Of these 79 noncancer individuals, 36 had benign ear nose and throat diseases (BED) including symptoms of epistaxis (18), tinnitus (6), ear blockage (5), enlargement of neck mass (4), and family history of NPC (3). They were confirmed clinically negative for NPC by nasoendoscopy. The other 43 noncancer individuals were healthy blood donors without ear, nose, and throat diseases. The demographics of the NPC patients and controls are shown in Table 1. These two groups were comparable in sex and age (P value for sex and age,.22 and.32, respectively). Blood was collected in a serum separator tube and allowed to clot for 30 minutes at room temperature. After centrifugation, serum samples were aliquoted to small volumes (50 L) and stored at 70 C until use. The serum levels of EBNA1 IgA of NPC patients and BED patients were also determined. Serum Protein Profiling Three types of chips (Ciphergen Biosystems) with surface chemistry of hydrophobic (H50), ionic (CM10), and metal binding (IMAC3) were initially evaluated to determine which affinity chemistry provided the best serum protein profiles. The IMAC3 chip was found to

3 SELDI-TOF-MS in Nasopharyngeal Carcinoma/Ho et al. 101 attain the highest number of protein peaks ( 93 peaks) among the chips tested. Therefore, it was suitable for this work and used throughout the study. Serum samples were thawed and briefly centrifuged (5 minutes, 10,000 revolutions per minute [rpm]) and pretreated before loading. To 5 L of each serum sample, 5 L of a solution containing 8 mol/l urea and 10 g/l CHAPS in 1 phosphate-buffered saline (PBS) [ph 7.2] was added. The mixture was vortexmixed and incubated at 4 C for 30 minutes. After incubation, the diluted serum mixture was mixed with 108 L binding/washing buffer containing 1 PBS, 0.25% Triton X-100, and 250 mm sodium chloride. Before spotting, an 8-spot IMAC3 array was incubated with 10 L of 100 mm copper sulfate for 10 minutes, repeated once, and finally washed with high-performance liquid chromatography (HPLC) water to remove the unbound copper. The chip was assembled to a bioprocessor (Ciphergen Biosystems) and washed with the binding/washing buffer 3 times for equilibration. One hundred microliters of diluted serum sample were spotted onto each IMAC3 array spot and the bioprocessor was sealed and shaken on a platform shaker at a speed of 250 rpm for 1 hour. The diluted samples were discarded and the bioprocessor was washed 3 times with 200 L of binding/washing buffer with shaking for 5 minutes, followed by 2 quick rinses with 500 L of HPLC-grade water. The chip was removed from the bioprocessor and allowed to air dry. One microliter saturated solution of sinapinic acid (Sigma Chemical, St. Louis, MO) in 50% acetonitrile volume/volume (v/v) and 0.5% v/v trifluoroacetic acid (Sigma Chemical) was added twice to each spot, with the array surface air-dried between each sinapinic acid application. The captured proteins in the arrays were detected by a PBS-II c reader (Ciphergen Biosystems) and the TOF mass spectra were then generated by averaging 61 laser shots in a positive mode with an intensity of 245 and a detector sensitivity of 7. Mass accuracy was calibrated externally using the All-in- One peptide mass standard (Ciphergen Biosystems). The data were analyzed by ProteinChip Software (version 3.1; Ciphergen Biosystems). Data Analysis The entire dataset was randomly separated into training and test datasets before analysis. A training set consisted of spectra data from 50 patients with NPC and 54 noncancer controls to build up the classification tree. The discriminatory ability of the classification algorithm was then challenged with a blind test dataset consisting of another spectra data of 20 patients with NPC and 25 noncancer controls. All spectral data were normalized by total ion current after background subtraction. The range of peak masses was analyzed between ,000 daltons (Da) because the majority of resolved protein/peptides were found in this range. The molecular masses from Da were excluded from analysis because they were mainly the signal noises of the energy-absorbing molecule (EAM). The Biomarker Wizard (Ciphergen Biosystems) was subsequently used to make peak detection and clustering across all spectra in the training set with the following settings: signal/noise (first pass): 4; minimum peak threshold: 10%; mass error: 0.3%; and signal/noise (second pass): 2 for the ,000 Da mass range. Corresponding peaks in the spectra from the test set were likewise identified using the clustering data from the training set by the same software. The spectral data of the training set were then exported as spreadsheet files and then further analyzed by the Biomarker Pattern Software (BPS) (version 4.0; Ciphergen Biosystems) to develop a classification tree. Classification and Regression Tree (CART) Analysis The BPS is an implementation of CART, which is a statistical tool developed by Breiman et al. 19 BPS uses the peak information generated from the training set of samples to build the classification tree. CART consists of two steps: tree construction and tree pruning. The tree construction process initially splits up spectral data from the training set into 2 nodes, using 1 rule at a time in the form of a question. The splitting decision in this case is based on the intensity of a peak or cluster. Samples go to the left daughter node if their intensities are equal to or less than the cutoff intensity value. The splitting process is then continued until no further gain in classification is achieved and terminal nodes are produced. Classification of terminal nodes is decided by the group of samples (i.e., NPC and noncancer) that represents the majority of samples in that node. In the second step, the classification tree is cut down to a desired size using tree-cost complexity pruning. Statistical Analysis Calculation of sensitivity, specificity, positive predictive value, and negative predictive value was performed using Instat (version 3.0) for windows (Graph- Pad Software, San Diego, CA). Comparisons of parameters between the NPC group and noncancer control group were performed using the Mann Whitney U-test. A P value.05 was considered statistically significant.

4 102 CANCER July 1, 2006 / Volume 107 / Number 1 FIGURE 1. Mass spectra from the sera of 3 patients with nasopharyngeal carcinoma (NPC) compared with the sera of 6 noncancer control group patients (3 from normal patients and 3 from patients with benign ear, nose, and throat disease [BED]) ranging from ,000 mass-to-change ratio (m/z). The box indicates a peak with an average mass of (A) 6811 daltons (Da) and (B) 10,272 Da that were overexpressed in the NPC group compared with the noncancer group. RESULTS Spectra from 104 serum samples of patients with NPC and controls were acquired in the training set. The protein peaks were identified with masses from ,000 Da, and 93 peak clusters or common peaks were generated from the identified peaks using the Biomarker Wizard Software. It was found that most of the peaks were detected between ,000 Da, and they were considered the most useful for protein profiling. Peak data from the training set were saved and exported for pattern recognition by the BPS. A classification tree was created from the training set to discriminate the disease and control groups. It was noted that the overexpressed protein peaks at 6811 and 10,272 Da were included as primary splitters in all classification trees generated during CART analysis and they almost separated the majority of the samples into 2 groups. Figure 1 represents the spectral views showing these 2 overexpressed protein peaks in FIGURE 2. Relative expression levels of 6811 and 10,272 daltons (Da) protein masses in the sera of the nasopharyngeal carcinoma (NPC) patients compared with that from the noncancer group. Black bars indicate the mean normalized intensity. (A) Expression levels of 6811 Da protein in the NPC patients and the noncancer group were and , respectively (P.05). (B) Expression levels of 10,272 Da protein in the NPC patients and the noncancer group were and , respectively (P.001). the NPC sera when compared with the control sera. From the quantitative point of view, the average normalized intensities of these 2 peaks were approximately 2-fold higher in the NPC sera compared with that in the noncancer sera (Fig. 2). These differences were statistically significant, with P values less than.05 and.001 for the peaks at 6811 and 10,272 Da, respectively. Of the many classification trees generated by adjusting the settings of Gini, costs, advance, and testing of BPS software, the most optimal classification tree with the lowest error cost was eventually established. The selected classification tree used 6 splitters with distinct masses of 6692, 6811, 6862, 7979, 9176, and 10,272 Da, respectively, and classified the cases into 7 terminal lymph nodes (Fig. 3). The error rate of the generated classification tree was estimated through a process of cross-validation. Performance of the generated classification tree is summarized in Table 2 for the training and test sets. The classification algorithm had an accuracy rate of 81% in the training set (Table 2). The validity and accuracy of the classification tree algorithm were then evaluated by attempting to classify blind cases correctly in the test set, which con-

5 SELDI-TOF-MS in Nasopharyngeal Carcinoma/Ho et al. 103 FIGURE 3. Diagram of a decision tree for the classification of the nasopharyngeal carcinoma (NPC) and noncancer samples in the training dataset. The circles indicated the primary nodes and the squares were the terminal nodes. The mass value in the root nodes was followed by the intensity value. TABLE 2 Performance of the Classification Tree Analysis of NPC in Training and Test Sets Sensitivity,% Specificity, % Accuracy rate, % Training set 80 (41/50)* 84 (45/54) 81 (85/104) Test set 85 (17/20) 80 (20/25) 82 (37/45) Overall 83 (58/70) 82 (65/79) 83 (123/149) NPC indicates nasopharyngeal carcinoma. *Numbers in parentheses denote the number of correctly classified samples of the total number of samples in the group. sisted of 20 sera from NPC patients, 8 sera from healthy individuals, and 17 sera from BED patients (distinct from the training set). The algorithm correctly classified 82% (37 of 45) of the testing samples with a sensitivity of 85% (17 of 20) and specificity of 80% (20 of 25) (Table 2). Eighty-six percent (37 of 43) of the NPC patients with Stages I and II, and 78% (21 of 27) of the patients with Stages III and IV were detected by the classification algorithm, with no statistical significance. The sensitivity, specificity, and accuracy rates of the EBNA1 IgA serology test in the present study were 83% (58 of 70), 83% (30 of 36), and 84% (88 of 106), respectively, which were comparable to that of the serum protein profiles shown in Table 3. The combined results of the serology test (EBNA1 IgA) and serum protein profiles of patients are shown in Table 3. The combination of both tests increased the sensitivity to 99% (69 of 70), with at least 1 of the 2 tests being positive, and a specificity of 96% (24 of 25) when both tests were negative. Because the positive predictive rate of a screening/diagnostic test changes according to the prevalence of the disease in the population at risk in various localities (in Hong Kong there are about 5.6 million Chinese age 14 years at risk for NPC; the annual incidence rate is about 36 per 100,000), the projected positive predictive rate and false-negative rate of NPC screening in this study were calculated according to the prevalence of NPC in Hong Kong and the data in Table 3 (Table 4). When both tests were used together, the sensitivity to predict the risk of NPC by concordant positive results was 4.7 (840/180) and 5.6 (840/150) times higher than that of the serum EBNA1 IgA test alone and serum protein profiling alone, respectively; the risk of missing NPC in the screening test (false-negative test) was 12 (17/1.4) and 12 (17/1.2) times lower than that of the EBNA1 IgA test alone or protein profiling alone. DISCUSSION There may be 2 types of serum protein biomarkers that are related to cancers. One type may be proteins

6 104 CANCER July 1, 2006 / Volume 107 / Number 1 TABLE 3 EBNA1 IgA and Serum Protein Profiles EBNA1 IgA protein profiles EBNA1 IgA protein profiles EBNA1 IgAprotein profiles EBNA1 IgAprotein profiles Total NPC BED Total EBNA1, Epstein-Barr nuclear antigen 1; IgA, immunoglobulin A; EBNA1 IgA, when the antibody level exceeds the cutoff value of the reference sera; EBNA1 IgA, when the antibody level equals or is lower than the cutoff value of the reference sera; protein profiles, when the serum protein profile is classified as nasopharyngeal carcinoma at the terminal node by the classification tree algorithm; protein profiles, when the serum protein profile is classified as noncancer at the terminal node by the classification tree algorithm; NPC, nasopharyngeal carcinoma; BED, benign ear, nose and throat diseases. TABLE 4 Projected Positive Predictive Rate and False-Negative Rate (Risk of Missing) for Screening NPC in the Hong Kong Population Screening Tests Positive Predictive Rate of NPC in Screening Every 100,000 Population Risk of Missing NPC With False-Negative Result (%) EBNA1 IgA 180 EBNA1 IgA Protein profiles 150 Protein profiles 7 17 Concordant positive results (protein profiles, EBNA1 IgA ) 840 Discordant results (protein profiles, EBNA1 IgA or protein profiles, EBNA1 IgA ) 360 Concordant negative results (protein profiles, EBNA1 IgA-) NPC, nasopharyngeal carcinoma; EBNA1, Epstein-Barr nuclear antigen 1; IgA, immunoglobulin A; EBNA1 IgA, when the antibody level exceeds the cutoff value of the reference sera; EBNA1 IgA, when the antibody level equals or is lower than the cutoff value of the reference sera; protein profiles, when the serum protein profile is classified as NPC at the terminal node by the classification tree algorithm; protein profiles, when the serum protein profile is classified as noncancer at the terminal node by the classification tree algorithm. or peptides secreted from the viable cancer cells or released from the dead cancer cells into the circulation. Alternatively, cancer cells may stimulate host responses that lead to an aberrant increase or decrease of normal proteins in abundance, such as cytokines, chemokines, and inflammatory proteins. Therefore, protein expression profiles reflect the presence of these 2 types of biomarkers in cancer and a clinical test that can simultaneously analyze multiple protein profiles may provide a more versatile means of cancer diagnosis. A recent finding showed that a classification tree successfully distinguished head and neck squamous cell cancer (HNSCC) from normal with a sensitivity and specificity of 83% and 100%, respectively. 20 Three protein masses of 5064, 13,881, and 15,139 Da were the primary splitters of the classification tree that segregated HNSCC from noncancer (Table 5). Other cancers, such as hepatocellular carcinoma (HCC), could be detected at an early stage by the combination of -fetoprotein (AFP), des-gamma carboxprothrombin, and GP73 tests with SELDI protein patterns with a sensitivity of 75% and specificity of 92%, respectively. 18 In contrast, the serum AFP test alone could achieve only a low sensitivity for the detection of HCC. The protein masses used for constructing the classification tree were 5808, 8939, 9501, and 11,735 Da in that study (Table 5). Clearly, the protein profiles as well as the protein masses used for the classification trees in these 2 studies were different from each other, despite using the same type of protein chip. In our study we identified 37 protein peaks, which showed a statistical difference between the NPC group and the noncancer group (data not shown). However, it was noted that not all of these significant peaks were used in the classification tree algorithms, except the peaks at 6811 and 10,272 Da. It was suggested by Wadsworth et al. 20 that the CART approach examined all the possible protein peak combinations from the input spectral data and made the best classification tree in which a statistically nonsignificant protein peak included was also important for the classification algorithm after stratification. Although 4 of the 6 peaks (6692, 6862, 7979, and 9176 Da) shown in Figure 3 had no statistical difference between the 2 groups of sera, they became crucial to the classification tree to delineate subsets of groups that had been stratified by the significant peaks of 6811 and 10,272 Da. By the same

7 SELDI-TOF-MS in Nasopharyngeal Carcinoma/Ho et al. 105 TABLE 5 Important Peaks for the Establishment of Decision Trees by BPS in Different Types of Cancers Types of Cancer Types of Specimen Types of Chip Used Peaks Selected in Decision Trees m/z Reference Prostate cancer Serum IMAC3 4475, 5074, 5382, 7024, 7820, 8141, HCC Serum IMAC3 5808, 8939, 9501, HNSCC Serum IMAC3 5064, 13881, Pancreatic cancer Serum IMAC3 3816, 3955, 6430, 7466, 7562, NPC Serum IMAC3 6692, 6811, 6862, 7979, 9176, Present finding BPS indicates biomarker patterns software; HCC, hepatocellular carcinoma; HNSCC, head and neck squamous cell cancer; NPC, nasopharyngeal carcinoma; m/z, mass-to-change ratio. proteomic technique using the same protein chip, we developed a classification tree from serum protein profiles to differentiate NPC patients from noncancer controls (Table 5). With the SELDI-TOF-MS serum profiling in combination with CART, we achieved an overall sensitivity of 83% and a specificity of 82% for the detection of NPC. It was interesting to note that the protein masses used by Wadsworth et al. 20 to differentiate head and neck cancer patients from normal controls were distinct from those we found in our study for distinguishing NPC patients. Other cancers, such as prostate cancer and pancreatic cancer, also exhibited unique peaks in the classification trees for the segregation of cancers and normal (Table 5). 14,21 The differences of serum protein patterns in various cancers suggest that a particular cancer might be identified by its specific protein phenotypes, which could be used for the differential diagnosis of the disease after careful validation. Although the EBV serology test has been practiced for many years, its usefulness as a universal screening test for NPC is still questionable because of the problems of either low specificity or sensitivity. In the report by Tsang et al., 22 the sensitivities were 89% for the VCA IgA test, but only 63% for the EA IgA test. Therefore, the EA antibody test alone is not sensitive enough to be used for screening purposes. Conversely, the VCA IgA test is highly sensitive but has more than a 50% false-positive rate. 23 Another possible biomarker for NPC is the measurement of serum EBV DNA. Nevertheless, the sensitivity and specificity of the EBV DNA assay cannot be repeated by different investigators in different laboratories. Recent studies demonstrated a low sensitivity and high false-positive rate using this assay. 6,24 Our previous study also showed that serum EBV DNA could only detect 61% of NPC patients with local recurrence. 25 Therefore, the diagnostic value of EBV DNA needs further verification. In the absence of a reliable EBV antibody test or EBV DNA assay with both high sensitivity and specificity, a multiple-marker approach may have a greater potential to improve the accuracy in NPC screening. In this study the combination of a serologic EBNA1 IgA test and serum protein profiles improved the diagnostic performance by increasing the positive predictive rate and reducing the false-negative rate, which markedly surpassed that of each test alone. The current clinical management policy of patients based on the EBNA1 IgA test is to perform nasoendoscopy and random biopsy of the nasopharynx for all patients with a positive serology test, and reassure without nasoendoscopy for patients with a negative serology test. When the serum EBNA1 IgA test and protein profiles are used together, we can reduce the false-negative rate from 17% to 1.4%, and at the same time manage the patients better according to the risk of NPC with the following algorithm: 1) In the high-risk group with both tests showing concordant positive results, nasoendoscopy with routine random nasopharyngeal biopsy is recommended. 2) In the low-risk group with both tests showing concordant negative results, a reliable exclusion of the cancer is recommended. These patients can be reassured without routine nasoendoscopy if they have no suspicious symptoms. 3) In the moderate-risk group with discordant results of only 1 of the 2 tests being positive, nasoendoscopy without routine nasopharyngeal biopsy is recommended unless a suspicious lesion is observed in the nasopharynx. However, these patients are recommended for reassessment at a 6-month frequency. Although the present study and others have demonstrated the potential clinical applications of serum protein profiles coupled with an artificial intelligence data analysis algorithm in screening various cancers with high sensitivity and specificity, the SELDI- TOF-MS technique is still a new tool in which further validation of its clinical applications and refinement in both hardware technology and software data analysis are needed. One major concern is that molecules constituting most of the reported protein patterns are epiphenomena of cancer and these proteins do not originate from the tumor cells but are produced and

8 106 CANCER July 1, 2006 / Volume 107 / Number 1 released into the circulation by other organs in response to the presence of cancer. 26 To overcome this criticism, identification of the proteins is needed in order to understand their source and relation to the underlying pathology. 27 The protein masses of 6811 and 10,272 Da in the present study will be identified in order to study the potential roles of these proteins in NPC. Another major concern of the SELDI-TOF-MS technique is the problem of the reproducibility of results. Significant variations in protein peaks in prostate cancer and ovarian cancer have been identified by different research groups. In addition, data of the same protein peaks are not easily reproduced among laboratories. The problem in reproducibility and discrepancy might be explained by the hypothesis that the SELDI-TOF-MS technique is too sensitive to the experimental details and a small error will be magnified in every subsequent step. 28 These problems could be solved by following stringent experimental protocols, including specimen collection, handling, shipping, and standardizing laboratory procedures. 28,29 In our study the recommended protocol was followed strictly at every step and the protein peaks were selected in the optimal range of ,000 Da. In addition, quality control materials, such as serum/ plasma reference standards, should be developed and used to monitor the instrumental bias and process reproducibility over a continuous time period. 29 Furthermore, the problem of reproducibility can be further improved by upgrading the instrumentation of SELDI-TOF-MS, 29 for instance, by the replacement of a more sophisticated hybrid quadrupole-tof mass spectrometer (QqTOF MS). One recent study reported that the mass spectra acquired using the high-resolution QqTOF MS were both 100% sensitive and specific in the detection of ovarian cancer. 30 However, its application in screening NPC has not been reported yet. Finally, because the serum protein profiles and generation of a classification tree require the application of bioinformatics (i.e., spectral data management and analysis) and the vast amount of spectral data generated by the SELDI technique require an implementation of advanced data management and analysis strategies, and research groups have developed and applied a variety of data handling tools, it is necessary to unify and standardize the data management and analysis methods for future clinical applications. The annual incidence of NPC in the years was less than 1/100,000 among Caucasians, whereas it was greater than 20/100,000 among southern Chinese. 31 Therefore, NPC is peculiar in racial and geographic distribution. As a result, although EBV infection is recognized as a common etiological factor for NPC among different regions, the serum protein patterns may be diverse. Therefore, the application of our present finding in other populations apart from southern Chinese needs to be validated. In conclusion, our results demonstrate that serum protein profiling using SELDI-TOF-MS coupled with a learning classification algorithm could discriminate NPC from noncancer. To our knowledge, this is the first study to demonstrate the use of serum protein profiling to detect NPC. In addition, the use of a combination of serum protein profiles with EBV serology tests could markedly increase the overall diagnostic accuracy and optimize the algorithm in the follow-up management of diagnostic endoscopy and biopsy. Furthermore, based on our recent finding showing that the combination of the serological EBV antibody test with a panel of methylation markers could increase the diagnostic accuracy of NPC, 32 there is a potential of combining the serum protein profiles, EBV serological tests, and DNA markers for the screening of this disease with an even higher sensitivity and specificity. REFERENCES 1. Ho J. Nasopharyngeal carcinoma (NPC). Adv Cancer Res. 1972;15: Hong Kong Cancer Registry. Cancer Incidence and Mortality in Hong Kong, Hong Kong: Hospital Authority; Teo P, Yu P, Lee WY, et al. Significant prognosticators after primary radiotherapy in 903 nondisseminated nasopharyngeal carcinoma evaluated by computer tomography. Int J Radiat Oncol Biol Phys. 1996;36: Cheng SH, Tsai SY, Yen KL, et al. Concomitant radiotherapy and chemotherapy for early-stage nasopharyngeal carcinoma. J Clin Oncol. 2000;18: Sheen TS, Ko JY, Chang YL, et al. Nasopharyngeal swab and PCR for the screening of nasopharyngeal carcinoma in the endemic area: a good supplement to the serologic screening. Head Neck. 1998;20: Chan KH, Gu YL, Ng F, et al. EBV specific antibody-based and DNA-based assays in serologic diagnosis of nasopharyngeal carcinoma. Int J Cancer. 2003;105: Vlahou A, Schellhammer PF, Mendrinos S, et al. Development of a novel proteomic approach for the detection of transitional cell carcinoma of the bladder in urine. Am J Pathol. 2001;158: Malik G, Ward MD, Gupta SK, et al. Serum levels of an isoform of apolipoprotein A-II as a potential marker for prostate cancer. Clin Cancer Res. 2005;11: Carter D, Douglass JF, Cornellison CD, et al. Purification and characterization of the mammaglobin/lipophilin B complex, a promising diagnostic marker for breast cancer. Biochemistry. 2002;41: Zhang Z, Bast RC Jr., Yu Y, et al. Three biomarkers identified from serum proteomic analysis for the detection of early stage ovarian cancer. Cancer Res. 2004;64: Wright GL Jr., Cazares LH, Leung SM, et al. Proteinchip(R) surface enhanced laser desorption/ionization (SELDI) mass spectrometry: a novel protein biochip technology for detection of prostate cancer biomarkers in complex protein mixtures. Prostate Cancer Prostatic Dis. 1999;2:

9 SELDI-TOF-MS in Nasopharyngeal Carcinoma/Ho et al Paradis V, Degos F, Dargere D, et al. Identification of a new marker of hepatocellular carcinoma by serum protein profiling of patients with chronic liver diseases. Hepatology. 2005;41: Qu Y, Adam BL, Yasui Y, et al. Boosted decision tree analysis of surface-enhanced laser desorption/ionization mass spectral serum profiles discriminates prostate cancer from noncancer patients. Clin Chem. 2002;48: Adam BL, Qu Y, Davis JW, et al. Serum protein fingerprinting coupled with a pattern-matching algorithm distinguishes prostate cancer from benign prostate hyperplasia and healthy men. Cancer Res. 2002;62: Petricoin EF, Ardekani AM, Hitt BA, et al. Use of proteomic patterns in serum to identify ovarian cancer. Lancet. 2002; 359: Li J, Zhang Z, Rosenzweig J, Wang YY, Chan DW. Proteomics and bioinformatics approaches for identification of serum biomarkers to detect breast cancer. Clin Chem. 2002;48: Zhang YF, Wu DL, Guan M, et al. Tree analysis of mass spectral urine profiles discriminates transitional cell carcinoma of the bladder from noncancer patient. Clin Biochem. 2004;37: Schwegler EE, Cazares L, Steel LF, et al. SELDI-TOF MS profiling of serum for detection of the progression of chronic hepatitis C to hepatocellular carcinoma. Hepatology. 2005;41: Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and Regression Trees. Belmont, CA: Wadsworth; Wadsworth JT, Somers KD, Cazares LH, et al. Serum protein profiles to identify head and neck cancer. Clin Cancer Res. 2004;10: Yu Y, Chen S, Wang LS, et al. Prediction of pancreatic cancer by serum biomarkers using surface-enhanced laser desorption/ionization-based decision tree classification. Oncology. 2005;68: Tsang RK, Vlantis AC, Ho RW, et al. Sensitivity and specificity of Epstein-Barr virus IgA titer in the diagnosis of nasopharyngeal carcinoma: a three-year institutional review. Head Neck. 2004;26: Low WK, Leong JL, Goh YH, Fong KW. Diagnostic value of Epstein-Barr viral serology in nasopharyngeal carcinoma. Otolaryngol Head Neck Surg. 2000;123: Leung SF, Lo YM, Chan AT, et al. Disparity of sensitivities in detection of radiation-naive and postirradiation recurrent nasopharyngeal carcinoma of the undifferentiated type by quantitative analysis of circulating Epstein-Barr virus DNA1, 2. Clin Cancer Res. 2003;9: Wei WI, Yuen AP, Ng RW, et al. Quantitative analysis of plasma cell-free Epstein-Barr virus DNA in nasopharyngeal carcinoma after salvage nasopharyngectomy: a prospective study. Head Neck. 2004;26: Diamandis EP. Mass spectrometry as a diagnostic and a cancer biomarker discovery tool: opportunities and potential limitations. Mol Cell Proteomics. 2004;3: Petricoin EF III, Liotta LA. Counterpoint: the vision for a new diagnostic paradigm. Clin Chem. 2003;49: Seibert V, Ebert MP, Buschmann T. Advances in clinical cancer proteomics: SELDI-ToF-mass spectrometry and biomarker discovery. Brief Funct Genomic Proteomic. 2005;4: Liotta LA, Lowenthal M, Mehta A, et al. Importance of communication between producers and consumers of publicly available experimental data. J Natl Cancer Inst. 2005;97: Conrads TP, Fusaro VA, Ross S, et al. High-resolution serum proteomic features for ovarian cancer detection. Endocr Relat Cancer. 2004;11: Parkin DM, Whelan SL, Ferlay J, et al. Cancer Incidence in Five Continents. Volume 7. Lyon, France: IARC; 1997: Wong TS, Kwong DL, Sham JS, et al. Quantitative plasma hypermethylated DNA markers of undifferentiated nasopharyngeal carcinoma. Clin Cancer Res. 2004;10: