Deep proteome profiling of sera from never-smoked lung cancer patients

Biomedicine & Pharmacotherapy 61 (2007) 570e577 www.elsevier.com/locate/biopha

Deep proteome profiling of sera from never-smoked lung cancer patients Joseph S.K. Au a,*, William C.S. Cho a, Tai Tung Yip b, Christine Yip b, Hailong Zhu c, Wallace W.F. Leung c, Philip Y.B. Tsui c, Davy L.P. Kwok c, Simon S.M. Kwan c, Wai Wai Cheng a, Lawrence C.H. Tzang d, Mengsu Yang d, Stephen C.K. Law a a

Department of Clinical Oncology, Queen Elizabeth Hospital, 30 Gascoigne Road, Kowloon, Hong Kong b Ciphergen Biosystems, Inc., Fremont, CA 94555, USA c Research Institute of Innovative Products and Technologies, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong d Biology and Chemistry Department, City University of Hong Kong, Kowloon Tong, Hong Kong Available online 14 September 2007

Abstract Previous studies on the serum proteome are hampered by the huge dynamic range of concentration of different protein species. The use of Equalizer Beads coupled with a combinatorial library of ligands has been shown to allow access to many low-abundance proteins or polypeptides undetectable by classical analytical methods. This study focused on never-smoked lung cancer, which is considered to be more homogeneous and distinct from smoking-related cases both clinically and biologically. Serum samples obtained from 42 never-smoked lung cancer patients (28 patients with active untreated disease and 14 patients with tumor resected) were compared with those from 30 normal control subjects using the pioneering Equalizer Beads technology followed by subsequent analysis by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS). Eighty-five biomarkers were significantly different between lung cancer and normal control. The application of classification algorithms based on significant biomarkers achieved good accuracy of 91.7%, 80% and 87.5% in class-prediction with respect to presence or absence of disease, subsequent development of metastasis and length of survival (longer or shorter than median) respectively. Support vector machine (SVM) performed best overall. We have proved the feasibility and convenience of using the Equalizer Beads technology to study the deep proteome of the sera of lung cancer patients in a rapid and high-throughput fashion, and which enables detection of low abundance polypeptides/proteins biomarkers. Coupling with classification algorithms, the technologies will be clinically useful for diagnosis and prediction of prognosis in lung cancer. Ó 2007 Elsevier Masson SAS. All rights reserved. Keywords: Lung neoplasm; Proteomic profiling; Combinatorial ligands

1. Introduction Most phenotypic manifestations are believed to be related to protein expressions, which are not predictable from the knowledge of the genome alone. The ‘proteome’ reflects the state of a cell, tissue or organism more accurately and therefore from it we are more likely to discover better tumour markers for disease diagnosis and therapy monitoring. The application of mass spectrometry (MS) in proteomic research is a major * Corresponding author. Tel.: þ852 2958 3323; fax: þ852 2359 4782. E-mail address: [email protected] (J.S.K. Au). 0753-3322/$ - see front matter Ó 2007 Elsevier Masson SAS. All rights reserved. doi:10.1016/j.biopha.2007.08.017

technological breakthrough and enables high-throughput analysis of protein mixtures. Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS) was especially designed for biomarker discovery and has become a popular tool for protein pattern profiling. The concept of surface-enhanced laser desorption/ionization (SELDI) was introduced by Hutchens and Yip [1] in 1993 and this novel strategy of mass spectrometric analysis of macromolecules simplified sample extraction and facilitated effective on-probe investigation of biopolymers when compared to conventional laser desorption/ionization (e.g. MALDI). Since the presentation of this

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577

landmark work, many exciting SELDI-TOF-MS applications have been described [2e20]. Serum is more easily available than tissue and hence holds the promise of a revolution in early disease detection, predicting biological aggressiveness and prognosis, and therapeutic monitoring. However, serum is a complex biological mixture comprising hundreds of thousands of different polypeptides of an estimated dynamic range of over ten orders of magnitude (Fig. 1). In addition to the ‘‘classical serum proteins’’ (albumin, immunoglobulins, transferrin, etc.), it contains all tissue proteins (as leakage markers) plus cytokines [21]. Therefore, high-throughput proteomic analyses are greatly hampered by the huge dynamic range of concentration of different protein species, making the detection of trace proteins difficult [22]. The novel use of Equalizer Beads coupled with a combinatorial library of ligands [23] have been shown to allow access to many low-abundance proteins or polypeptides undetectable by classical analytical methods [24,25]. The population of beads has such diversity that a binding partner should exist for most proteins in a sample. Each bead has an equivalent binding capacity. High abundance proteins saturate their binding partner and excess protein is washed away, whereas trace proteins are concentrated on their specific ligands (Fig. 2). Coupled with SELDI-TOF-MS, it is a new approach for the rapid and simultaneous resolution of multiple low-abundance proteins. For lung cancer, various attempts [4,12,26,27] have been made to profile the proteomes using MS. The present study is the first to use the Equalizer Beads technology to study the proteome especially the low-abundance proteins in the

571

sera of never-smoked lung cancer patients, which is considered as a more homogeneous and distinct disease group amongst different etiological types of lung cancer [28]. As the serum proteins of lung cancer patients may be either due to the presence of the tumour itself or the host response, it may also be enlightening to compare the sera of those patients with freshly diagnosed untreated lung cancer and those patients who had the tumour already completely resected. 2. Materials and methods 2.1. Study population The study serum samples were archived specimens from 42 never-smokers with histologically confirmed adenocarcinoma of the lung. Twenty-eight patients were fresh cases without prior treatment. Fourteen patients had the tumour resected by surgery and the serum samples were taken at least 2 weeks after surgery. Serum samples taken from 30 healthy individuals serve as normal control. All lung cancer patients had been followed up till death. Clinical data on the subsequent development of distant metastasis and survival were available. 2.2. Deep proteome profiling [23] An aliquot of 180 ml serum sample was mixed with 50 ml 9 M urea 2% CHAPS 50 mM TriseHCl pH 9 in 96-well V-bottom plate (Nunc). The mixture was shaken at 4  C for 10 min. An aliquot of 200 ml of this mixture was added to 25 ml Equalizer bead in 50 ml of 50 mM TriseHCl pH 7.5 in 96-well filter plate

Fig. 1. Abundance of plasma proteins is plotted on a log scale spanning 12 orders of magnitude. Where only an upper limit is quoted, the lower end of the interval line shows an arrowhead. A significant portion of the total protein content in serum and plasma is comprised of just a few proteins. Low abundance proteins make up less than 1% of total protein content. (From N. Leigh Anderson et al. Molecular & Cellular Proteomics, 2002;1.11:845e867.)

572

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577

Fig. 2. Equalizer bead technology: Each bead has a different ligand, that has an affinity for a specific protein in the sample. (From Product Note, Ciphergen’s Deep Proteome.)

(Nunc). The plate was sealed and shaken at 4  C for 2 h. After centrifugation at 2000 rpm for 5 min to remove non-bound material, the Equalizer beads in filter plate were washed with 400 ml 50 mM TriseHCl pH 7.5 by shaking for 1 min. The wash was removed by vacuum suction. The washing was repeated one more time. Equalizer bead bound proteins were eluted with 50 ml 50% isopropanol/acetonitrile (2:1) 5% formic acid by shaking for 10 min. The eluate was collected as Bound I by centrifugation. The beads were washed with 50 ml 30% isopropanol. Wash was collected by centrifugation and pooled with Bound I. Second elution of bound proteins was with 50 ml 4 M guanidine thiocyanate 2% CHAPS 0.1 M HEPES by incubating at 90  C for 10 min. The eluate was collected as Bound II by centrifugation. The beads were washed with 100 ml 1 M urea 0.1% CHAPS 50 mM NaCl 2.5% acetonitrile 50 mM TriseHCl pH 7.5. Wash was collected by centrifugation and pooled with Bound II. Final elution was with 50 ml 9 M urea 4% CHAPS 1% acetic acid by incubating at 90  C for 10 min. Eluate was collected by centrifugation and pooled with Bound II. An aliquot of 25 ml of Bound I and 50 ml of Bound II of each sample was added to 200 ml 1 M urea 0.1% CHAPS 0.3 M KCl 0.1 M TriseHCl pH 7.5 in a bioprocessor containing IMAC30 Cu ProteinChip arrays. After shaking at room temperature for 40 min, the samples were discarded. The chips were washed with 200 ml 50 mM TriseHCl pH 7.5 for 5 min. After rinsing the chips with water, an aliquot of 1 ml sinapinic acid in 50% acetonitrile 0.5% trifluoroacetic acid was added, allowed to air dry, followed by another 1 ml of sinapinic acid. The remaining Bound I and Bound II of each sample were added to 75 ml Trisacryl Blue bead (Biosepra) in 100 ml

50 mM TriseHCl pH 7.5 in a 96-well filter plate. After shaking at 4  C for 30 min, the non-bound proteins were collected by centrifugation. The non-bound proteins were added to 75 ml Protein A bead (Biosepra) in a 96 well filter plate. After shaking at 4  C for 30 min, the non-bound proteins were collected by centrifugation. The non-bound proteins were added to 75 ml MEP (Biosepra) bead in a 96 well filter plate. After shaking at 4  C for 30 min, the non-bound proteins were collected by centrifugation. Each plate was washed with 200 ml 50 mM TriseHCl pH 7.5. Bound proteins from each plate was eluted with 75 ml 50% isopropanol/acetonitrile (2:1) 5% formic acid, 75 ml 30% isopropanol (pooled to form Bound I), 75 ml 4 M guanidine thiocyanate 1% CHAPS 0.1 M HEPES, 200 ml 50 mM TriseHCl pH 7.5 (pooled to form Bound II). An aliquot of 100 ml of Bound I and 200 ml Bound II of each sample was added to 50 ml 50 mM TriseHCl pH 7.5 in a bioprocessor containing IMAC30 Cu or Q10 chips. After shaking at room temperature for 40 min, the samples were discarded. The chips were washed with 200 ml 50 mM TriseHCl pH 7.5 for 5 min. After rinsing the chips with water, an aliquot of 1 ml sinapinic acid was added, allowed to air dry, followed by another 1 ml of sinapinic acid. All chips were read in an automated PCS4000 at 1500 nJ, averaging 20 laser shots per pixel and scanning every one out of 4 pixels. After first reading, the chips were washed with 5 ml 0.1 M citric acid, re-extracted with 1 ml 50% isopropanol/acetonitrile (2:1) 5% formic acid. The chips were re-read at 1800 nJ, averaging 30 laser shots per pixel and scanning every one out of 3 pixels.

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577

2.3. Data analysis All data were processed by Ciphergen Express 3.0 software. Peak identification involved baseline substraction, mass accuracy calibration, and automatic peak detection. The settings were as follows: for peak detection signalto-noise ratio was 3, minimal peak threshold was 10%; for cluster completion, the cluster mass was 0.5% and the signal-to-noise ratio for the second pass was 1. Comparison of relative peak intensity levels between groups was made using the ManneWhitney U-test. A sensitivity analysis of the biomarkers based on the receiver operating characteristic (ROC) curve method was performed. For prediction of different classes of subjects based on the protein peaks, different classification algorithms, namely, random tree (RT) [29], multiple layer perceptron (MLP) [30e32],

573

radial basis function network (RBF) [30e32], Bayesian belief network (BBN) [33] and support vector machine (SVM) [34] were applied. The performance of these algorithms was compared in terms of accuracy, sensitivity, and specificity calculated by the leave-one-out cross-validation (LOOCV) approach.

3. Results 3.1. Overview A total of 128 MS peak clusters were identified by Ciphergen Express. A list of the protein peaks with intensities most statistically significant different amongst disease or prognostic groups and the results of sensitivity analysis by the ROC curve method were shown in Table 1.

Table 1 Potential biomarkers for never-smoked lung cancer patients Normalized peak intensity (mean  standard error)

Specificity(%)

Classification accuracy(%)

AUC*

P value

(a) Cancer-associated biomarkers: lung cancer (n ¼ 42) versus control normal (n ¼ 30) 3836 1644  103 versus 2675  63 39 Y 86 25,425 813  44 versus 450  20 81 [ 81 34,110 2388  106 versus 3541  63 33 Y 93 45,528 230  14 versus 444  13 48 Y 93 67,955 7335  364 versus 11,858  287 38 Y 93 103,441 174  15 versus 380  18 54 Y 88 113,478 18  2 versus 52  3 65 Y 90 137,519 794  64 versus 1670  68 52 Y 93 207,201 94  12 versus 279  18 66 Y 90 272,438 14  2 versus 42  4 66 Y 86

100 80 90 87 87 83 90 87 87 87

92 81 92 90 90 86 90 90 89 86

0.91 0.87 0.96 0.94 0.94 0.91 0.93 0.92 0.93 0.88

4.82E-10 7.49E-09 4.06E-11 3.76E-11 1.85E-10 2.49E-09 6.94E-11 9.91E-10 1.23E-09 6.54E-09

(b) Activity-associated biomarkers: active lung cancer (n ¼ 28) versus control normal (n ¼ 30) 3836 1619  134 versus 2675  63 39 Y 82 34,110 2471  127 versus 3541  63 30 Y 89 45,528 241  18 versus 444  13 46 Y 89 67,955 7608  447 versus 11,858  287 36 Y 89 103,441 188  18 versus 380  18 51 Y 86 113,478 19  2 versus 52  3 63 Y 89 137,519 840  80 versus 1670  68 50 Y 93 173,784 12  2 versus 33  3 63 Y 82 207,201 103  15 versus 279  18 63 Y 89 272,438 15  2 versus 42  4 63 Y 82

100 90 87 87 83 90 87 83 87 87

91 90 88 88 84 90 90 83 88 84

0.92 0.90 0.96 0.92 0.90 0.94 0.88 0.86 0.89 0.88

6.68E-08 9.37E-09 6.47E-09 1.48E-08 1.71E-07 7.10E-09 6.68E-08 5.43E-07 9.44E-08 2.02E-07

(c) Activity-associated biomarkers: inactive lung cancer (postop.) (n ¼ 14) versus control normal (n ¼ 30) 3836 1693  162 versus 2675  63 37 Y 93 100 13,600 4001  696 versus 1570  92 155 [ 86 97 25,425 873  73 versus 450  20 94 [ 86 87 34,110 2222  191 versus 3541  63 37 Y 100 97 45,528 207  23 versus 444  13 53 Y 93 100 67,955 6788  628 versus 11,858  287 43 Y 100 87 103,441 147  23 versus 380  18 61 Y 93 90 113,478 16  2 versus 52  3 69 Y 93 90 137,519 701  107 versus 1670  68 58 Y 93 87 207,201 77  17 versus 279  18 73 Y 93 87

98 93 86 98 98 91 91 91 89 89

0.95 0.93 0.93 0.99 0.97 0.95 0.95 0.97 0.93 0.93

6.07E-07 5.74E-06 3.14E-06 1.39E-07 2.09E-07 1.02E-06 2.78E-06 5.32E-07 2.17E-06 1.91E-06

(d) Metastasis-associated biomarkers: metastatic cancer (n ¼ 20) versus control normal (n ¼ 30) 11,685 14,034  3723 versus 549  87 2456 [ 80 12,665 3584  472 versus 1295  98 177 [ 75

88 76

0.90 0.85

7.39E-07 1.73E-05

71 76

0.75 0.82

5.88E-03 3.51E-04

MW (Da)

% of change

Sensitivity(%)

93 77

(e) Prognosis-associated biomarkers: patients survive <227 days (n ¼ 17) versus survive >227 days (n ¼ 25) 7624 910  163 versus 3640  1112 75 Y 71 72 91,529 62  6 versus 105  8 41 Y 71 80 * AUC: area under receiver-operator characteristic curve.

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577

574

3.2. Lung cancer versus normal control Comparing the serum proteomic profiles of the lung cancer patients (n ¼ 42) with normal controls (n ¼ 30), 85 potential biomarkers were statistically significant (P < 0.05). The minimal P value for the top 10 biomarker candidates (MW 3836, 25,425, 34,110, 45,528, 67,955, 103,441, 113,478, 137,519, 207,201 and 272,438 Da) was 7.5109 (Table 1a). 3.3. Freshly diagnosed versus postoperative cases Comparing Tables 1b,c, eight out of the top 10 protein peaks were common for freshly diagnosed or resected lung cancer. Peaks at 173,784 and 272,438 Da were significantly decreased for those with freshly diagnosed lung cancer. Peaks at 13,600 and 25,425 Da were significantly raised for the postoperative cases. 3.4. Metastasis Comparing lung cancer patients who subsequently developed metastases (n ¼ 20) and control normal (n ¼ 30), seventy protein peaks were statistically significant. The protein peaks at 11,685 Da and 12,665 Da were most statistically significant and the mean intensities were 25.6 fold and 2.8 fold of the normal control respectively (Table 1d). 3.5. Long versus short survival Thirty protein peaks were significantly different between short survivors (less than the median of 227 days) and long survivors. The mean peak intensities of protein peaks at

7624 Da and 91,529 Da were decreased by 75% and 41% in the short survivors respectively (Table 1e). 3.6. Class-prediction As shown in Table 2, the sensitivities, specificity and accuracies based on the protein peaks to classify lung cancer (whole group, freshly diagnosed cases or resected cases) were good. Amongst the classification algorithms, SVM always achieved the highest performance in terms of accuracy. An example of the application of SVM to predict the classes of lung cancer versus normal control was illustrated in Fig. 3. Two biomarkers (45,528, 113,478 Da) were used as the predictors. Sixty percent of the samples (24 lung cancer samples and 20 normal control samples) were randomly selected as the training dataset, and the remaining samples (18 lung cancer samples and 10 normal control samples) were used for testing the accuracy of classification. Fig. 3b showed the classification results with the testing dataset, in which one normal control was misclassified as lung cancer, and three lung cancer samples were misclassified as normal control. 4. Discussion and conclusion There are different approaches to overcome the wide dynamic range of proteins in serum, including fractionation methods and depletion of abundant proteins by immunoaffinity or immobilized dyes. However, most of these methods suffer from either their low level of separation performance or from a high level of complexity with a large number of fractions to manage requiring the use of sophisticated equipments. This is the first study employing the combinatorial ligands

Table 2 The classification performance evaluated by leave-one-out cross-validation approach Random-tree

Multilayer perceptron

Radial basis function network

Bayesian belief network

(a) Lung cancer (n ¼ 42) versus normal control (n ¼ 30) Accuracy 84.7 84.7 Sensitivity 86.0 87.8 Specificity 82.8 80.6

88.9 90.5 86.7

88.9 92.5 84.4

91.7 92.9 90.0

(b) Active lung cancer (n ¼ 28) versus normal control (n ¼ 30) Accuracy 75.9 81 Sensitivity 73.3 84 Specificity 78.6 78.8

84.5 82.8 86.2

84.5 82.8 86.2

87.9 88.9 87.1

(c) Postoperative lung cancer (n ¼ 14) versus normal control (n ¼ 30) Accuracy 86.4 86.4 Sensitivity 90.0 78.6 Specificity 85.3 90.0

90.0 85.7 93.3

90.9 81.3 96.4

93.2 92.3 93.5

(d) Lung cancer with metastasis (n ¼ 20) versus normal control (n ¼ 30) Accuracy 70 78 Sensitivity 61.9 90.9 Specificity 75.9 74.4

80 85.7 77.8

74 70.6 75.8

(e) Short survivors <227 days (n ¼ 17) versus long survivors >227 days (n ¼ 25) Accuracy 61.9 69 69 Sensitivity 52.9 60 61.1 Specificity 68 77.3 75

69 57.7 87.5

Note: The best performing figures amongst the algorithms are bold.

Support vector machine

80 100 75 73.8 61.5 69

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577

Fig. 3. Class prediction by Support Vector Machine (SVM) based on two proteins (MW 113,478 Da, 45,528 Da). The solid line is the classifier. (a) Training of SVM using a random dataset of 44 samples. (b) Testing out SVM on an independent dataset of 28 samples.

library approach on lung cancer research to simultaneously increase the concentration of low-abundance proteins and decrease the concentration of high-abundance proteins from the serum samples in a simple operation while maintaining protein diversity. This convenient and innovative deep proteome technology has enabled detection of a bundle of biomarkers for lung cancer including those polypeptides/ proteins that may usually been masked by the abundant proteins. Some of these less abundant proteins may be leakage from tumour tissue and hence may serve as clinically useful tumour markers. Identification of the significantly elevated proteins is underway. Our present study focused on never-smoked lung cancer, which is a distinct and relatively homogeneous biological entity. It is almost invariably of adenocarcinoma histology and occurs more commonly in females, with younger median

575

age of diagnosis [28]. A distinct pattern of allelic imbalance, chromosomal aberrations and gene expression profiles was noted [35e37]. The incidence of EGFR mutation is higher, correlating with better response to tyrosine kinase inhibitors [38]. There is yet no known risk factor useful in predicting its occurrence and hence it is a key challenge in serum proteomics to identify biomarkers that would enable early detection, diagnosis and monitoring of disease progression for this etiological type of lung cancer endemic in East Asia. SELDITOF-MS has advanced the utilization of mass spectrometry in proteomic research by provision of the surface enhanced biochip, which allows uniform and reproducible binding and desorption of biomarkers so that quantitative assessment of individually resolved biomarkers can be made [39e42]. We have confirmed that SELDI-TOF-MS is a useful tool for the rapid detection and identification of new potential biomarker of lung cancer in serum. Our results showed that the serum proteomic profiles are highly sensitive, specific and accurate in differentiating never-smoked lung cancer patients from normal healthy individuals and to differentiate between the different prognostic subgroups. Serum proteomics may also give a better overall picture of the tumour/environment interaction that may play a key role in cancer progression. The comparison of the serum proteomes of lung cancer patients with active disease and those with resected disease is interesting. Those peaks found in the postoperative patients are unlikely to be of leakage proteins from cancerous tissue. As expected, most highly significantly raised proteins in cancer patients are host response proteins. Hence, most significant proteins in Tables 1b,c are common except for the peaks at MW 173,784 Da, 13,600 Da, 272,438 Da and 25,425 Da. These peaks may represent acute phase inflammatory proteins or leakage proteins from cancer cells. Two peaks with MW 11,685 Da and 12,665 Da were found to predict for subsequent distant metastases with accuracy of 88% and 76% respectively. Another two peaks with MW 7624 Da and 91,529 Da were found to predict for better survival (longer than the median of 227 days) with accuracy of 71% and 76% respectively. These are very interesting biomarkers and potentially useful in predicting prognosis. There have been a few published studies proteomic analysis using MS for the investigation of lung cancer proteomics and prediction of the presence or absence of cancer and different clinicohistological findings [12,26,43e49]. For example, Yanagisawa et al. [26] investigated proteomic patterns of non-small cell lung cancer lysate using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS). They reported that class-prediction models classify histology, distinguish primary tumours from metastatic lesions from other organs to the lung, and nodal involvement with 85% accuracy. In our present study, based on those significant biomarkers identified, most established classification algorithms have turned out to have quite high accuracy in class-prediction. SVM performed best in terms of accuracy and hence it should form an integral part of our biomarker discovery platform in future studies. In conclusion, we have proved the feasibility and convenience of using the Equalizer Beads technology together

576

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577

with SELDI-TOF-MS to study the deep proteome of the sera of lung cancer patients in a rapid and high-throughput fashion, and which enables detection of low abundance polypeptides/ proteins biomarkers. The technologies can be potentially applied to a very large number of clinical samples of different diseases or prognostic groups for rapid biomarker screening. For our present study, a large number of serum proteins with potential utility in diagnosis and prognosis have been discovered and these warrant a larger prospective study for confirmation. Further protein identification as well as the development of routine clinical protein assays may prove useful in assisting the clinical management of cancer patients in future. References

[17]

[18]

[19]

[20]

[21] [22]

[1] Hutchen TW, Yip TT. New desorption strategies for the mass spectrometric analysis of macromolecules. Rapid Commun Mass Spectrom 1993;7:576e80. [2] Banerjee H, Hawkins Z, Williams J, Blackshear M, Sawyer C, Cezares L, et al. Search for a novel biomarker for the brain cancer astrocytoma by using surface enhanced laser desorption/ionisation (SELDI) technique. Cell Mol Biol (Noisy-le-grand) 2004;50(6):733e6. [3] Fetsch PA, Simone NL, Bryant-Greenwood PK, Marincola FM, Filie AC, Petricoin EF, et al. Proteomic evaluation of archival cytologic material using SELDI affinity mass spectrometry: potential for diagnostic applications. Am J Clin Pathol 2002;118(6):870e6. [4] Zhukov TA, Johanson RA, Cantor AB, Clark RA, Tockman MS. Discovery of distinct protein profiles specific for lung tumors and pre-malignant lung lesions by SELDI mass spectrometry. Lung Cancer 2003;40(3):267e79. [5] Lin Z, Jenson SD, Lim MS, Elenitoba-Johnson KS. Application of SELDITOF mass spectrometry for the identification of differentially expressed proteins in transformed follicular lymphoma. Mod Pathol 2004;17(6):670e8. [6] Wu DL, Wang WJ, Guan M, Jin SB, Jin CR, Zhang YF. [Screening urine markers of renal cell carcinoma using SELDI-TOF-MS]. Zhonghua Yi Xue Za Zhi 2004;84(13):1092e5. [7] Clarke CH, Buckley JA, Fung ET. SELDI-TOF-MS proteomics of breast cancer. Clin Chem Lab Med 2005;43(12):1314e20. [8] He QY, Zhu R, Ren Y, Tam PK, Chiu JF. Serological protein profiling of neuroblastoma by ProteinChip SELDI-TOF technology. J Cell Biochem 2005;95(1):165e72. [9] Marshall FF. Serum protein profiling by SELDI mass spectrometry: detection of multiple variants of serum amyloid alpha in renal cancer patients. J Urol 2005;173(6):1919e20. [10] Sauter ER, Shan S, Hewett JE, Speckman P, Du Bois GC. Proteomic analysis of nipple aspirate fluid using SELDI-TOF-MS. Int J Cancer 2005;114(5):791e6. [11] Traub F, Feist H, Kreipe HH, Pich A. SELDI-MS-based expression profiling of ductal invasive and lobular invasive human breast carcinomas. Pathol Res Pract 2005;201(12):763e70. [12] Yang SY, Xiao XY, Zhang WG, Zhang LJ, Zhang W, Zhou B, et al. Application of serum SELDI proteomic patterns in diagnosis of lung cancer. BMC Cancer 2005;5:83. [13] Ho DW, Yang ZF, Wong BY, Kwong DL, Sham JS, Wei WI, et al. Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry serum protein profiling to identify nasopharyngeal carcinoma. Cancer 2006;107(1):99e107. [14] Liang Y, Fang M, Li J, Liu CB, Rudd JA, Kung HF, et al. Serum proteomic patterns for gastric lesions as revealed by SELDI mass spectrometry. Exp Mol Pathol 2006;81(2):176e80. [15] Pan YZ, Xiao XY, Zhao D, Zhang L, Ji GY, Li Y, et al. Application of surface-enhanced laser desorption/ionization time-of-flight-based serum proteomic array technique for the early diagnosis of prostate cancer. Asian J Androl 2006;8(1):45e51. [16] Ricolleau G, Charbonnel C, Lode L, Loussouarn D, Joalland MP, Bogumil R, et al. Surface-enhanced laser desorption/ionization time of flight mass spectrometry protein profiling identifies ubiquitin and ferritin

[23]

[24]

[25]

[26]

[27]

[28]

[29] [30] [31] [32] [33] [34] [35]

[36]

[37]

[38]

[39]

[40]

light chain as prognostic biomarkers in node-negative breast cancer tumors. Proteomics 2006;6(6):1963e75. Scarlett CJ, Saxby AJ, Nielsen A, Bell C, Samra JS, Hugh T, et al. Proteomic profiling of cholangiocarcinoma: diagnostic potential of SELDI-TOF MS in malignant bile duct stricture. Hepatology 2006;44(3):658e66. Suriano R, Lin Y, Ashok BT, Schaefer SD, Schantz SP, Geliebter J, et al. Pilot study using SELDI-TOF-MS based proteomic profile for the identification of diagnostic biomarkers of thyroid proliferative diseases. J Proteome Res 2006;5(4):856e61. Xu WH, Chen YD, Hu Y, Yu JK, Wu XG, Jiang TJ, et al. Preoperatively molecular staging with CM10 ProteinChip and SELDI-TOF-MS for colorectal cancer patients. J Zhejiang Univ Sci B 2006;7(3):235e40. Petricoin EF, Ardekani AM, Hitt BA, Levine PJ, Fusaro VA, Steinberg SM, et al. Use of proteomic patterns in serum to identify ovarian cancer. Lancet 2002;359(9306):572e7. Anderson N, Anderson N. The human plasma proteome. Mol Cell Proteomics 2002;1(11):845e67. Righetti PG, Castagna A, Antonioli P, Boschetti E. Prefractionation techniques in proteome analysis: the mining tools of the third millennium. Electrophoresis 2005;26(2):297e319. Ciphergen. Ciphergen deep proteomeÔ research services: explore the low abundance proteome. Product Note (4 September 2006). Available from http://www.ciphergen.com. Guerrier L, Thulasiraman V, Castagna A, Fortis F, Lin S, Lomas L, et al. Reducing protein concentration range of biological samples using solidphase ligand libraries. J Chromatogr B Analyt Technol Biomed Life Sci 2006;833(1):33e40. Thulasiraman V, Lin S, Gheorghiu L, Lathrop J, Lomas L, Hammond D, et al. Reduction of the concentration difference of proteins in biological liquids using a library of combinatorial ligands. Electrophoresis 2005;26(18):3561e71. Yanagisawa K, Shyr Y, Xu BJ, Massion PP, Larsen PH, White BC, et al. Proteomic patterns of tumour subsets in non-small-cell lung cancer. Lancet 2003;362(9382):433e9. Chen G, Gharib TG, Wang H, Huang CC, Kuick R, Thomas DG, et al. Protein profiles associated with survival in lung adenocarcinoma. Proc Natl Acad Sci U S A 2003;100(23):13537e42. Lam B, Lam WK, Lam CL, Ooi GC, Ho JC, Wong MP, et al. Adenocarcinoma of the lung in Chinese patients: a revisit and some perspectives from the literature. Postgrad Med J 2001;77(913):708e12. Aldous D. The continuum random tree III. Annu Probab 1993;21(1):248e89. Bishop CM. Neural networks for pattern recognition. Oxford: Oxford University Press; 1995. Haykin S. Neural networks. New York: MacMillan; 1994. Ripley BD. Neural networks and related methods for classification. J Royal Statist Soc 1996;B56:409e56. Jensen FV. Introduction to Bayesian network. New York: SpringerVerlag; 1996. Vapnik VN. Statistical learning theory. New York: John Wiley & Sons; 1998. Powell CA, Bueno R, Borczuk AC, Caracta CF, Richards WG, Sugarbaker DJ, et al. Patterns of allelic loss differ in lung adenocarcinomas of smokers and nonsmokers. Lung Cancer 2003;39(1):23e9. Miura K, Bowman ED, Simon R, Peng AC, Robles AI, Jones RT, et al. Laser capture microdissection and microarray expression analysis of lung adenocarcinoma reveals tobacco smoking- and prognosis-related molecular profiles. Cancer Res 2002;62(11):3244e50. Powell CA, Spira A, Derti A, DeLisi C, Liu G, Borczuk A, et al. Gene expression in lung adenocarcinomas of smokers and nonsmokers. Am J Respir Cell Mol Biol 2003;29(2):157e62. Pao W, Miller V, Zakowski M, Doherty J, Politi K, Sarkaria I, et al. EGF receptor gene mutations are common in lung cancers from ‘‘never smokers’’ and are associated with sensitivity of tumors to gefitinib and erlotinib. Proc Natl Acad Sci U S A 2004;101(36):13306e11. Fung ET, Thulasiraman V, Weinberger SR, Dalmasso EA. Protein biochips for differential profiling. Curr Opin Biotechnol 2001; 12(1):65e9. Wright Jr GL. SELDI proteinchip MS: a platform for biomarker discovery and cancer diagnosis. Expert Rev Mol Diagn 2002;2(6):549e63.

J.S.K. Au et al. / Biomedicine & Pharmacotherapy 61 (2007) 570e577 [41] Issaq HJ, Veenstra TD, Conrads TP, Felschow D. The SELDI-TOF MS approach to proteomics: protein profiling and biomarker identification. Biochem Biophys Res Commun 2002;292(3):587e92. [42] Weinberger SR, Dalmasso EA, Fung ET. Current achievements using ProteinChip Array technology. Curr Opin Chem Biol 2002; 6(1):86e91. [43] Xiao X, Liu D, Tang Y, Guo F, Xia L, Liu J, et al. Development of proteomic patterns for detecting lung cancer. Dis Markers 2003;19(1):33e9. [44] Li LS, Kim H, Rhee H, Kim SH, Shin DH, Chung KY, et al. Proteomic analysis distinguishes basaloid carcinoma as a distinct subtype of nonsmall cell lung carcinoma. Proteomics 2004;4(11):3394e400. [45] Oh P, Li Y, Yu J, Durr E, Krasinska KM, Carver LA, et al. Subtractive proteomic mapping of the endothelial surface in lung and solid tumours for tissue-specific therapy. Nature 2004;429(6992):629e35.

577

[46] Fujii K, Nakano T, Kanazawa M, Akimoto S, Hirano T, Kato H, et al. Clinical-scale high-throughput human plasma proteome analysis: lung adenocarcinoma. Proteomics 2005;5(4):1150e9. [47] Hood BL, Lucas DA, Kim G, Chan KC, Blonder J, Issaq HJ, et al. Quantitative analysis of the low molecular weight serum proteome using 18O stable isotope labeling in a lung tumor xenograft mouse model. J Am Soc Mass Spectrom 2005;16(8):1221e30. [48] Maciel CM, Junqueira M, Paschoal ME, Kawamura MT, Duarte RL, Carvalho Mda G, et al. Differential proteomic serum pattern of low molecular weight proteins expressed by adenocarcinoma lung cancer patients. J Exp Ther Oncol 2005;5(1):31e8. [49] Tantipaiboonwong P, Sinchaikul S, Sriyam S, Phutrakul S, Chen ST. Different techniques for urinary protein analysis of normal and lung cancer patients. Proteomics 2005;5(4):1140e9.