|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 25, No 35 (December 10), 2007: pp. 5578-5583 © 2007 American Society of Clinical Oncology. DOI: 10.1200/JCO.2007.13.5392 Panel of Serum Biomarkers for the Diagnosis of Lung Cancer
From the Departments of Radiology, Biostatistics and Bioinformatics, and Pharmacology and Cancer Biology, Duke University Medical Center, Durham, NC Address reprint requests to Edward F. Patz Jr, MD, Department of Radiology, Duke University Medical Center, Box 3808, Durham, NC 27710; e-mail: patz0002{at}mc.duke.edu
Purpose Currently, a blood test for lung cancer does not exist. Serum biomarkers that could aid clinicians in making case management decisions would be enormously valuable. We used two proteomic platforms and a literature search to select candidate serum markers for the diagnosis of lung cancer. Methods We initially assayed six serum proteins, four discovered by proteomics and two previously known to be cancer associated, on a training set of sera from 100 patients (50 with a new diagnosis of lung cancer and 50 age- and sex-matched controls). Classification and Regression Tree (CART) analysis selected a panel of four markers that most efficiently predicted which patients had lung cancer. An independent, blinded validation set of sera from 97 patients (49 lung cancer patients and 48 matched controls) determined the accuracy of the four markers to predict which patients had lung cancer.
Results Four serum proteins—carcinoembryonic antigen, retinol binding protein, Conclusion This panel of four serum proteins is valuable in suggesting the diagnosis of lung cancer. These data may be useful for treating patients with an indeterminate pulmonary lesion, and potentially in predicting individuals at high risk for lung cancer.
Lung cancer continues to be a significant worldwide public health issue. Although advances in noninvasive imaging have improved our ability to detect lung cancer, more than 75% of lung cancer patients present with advanced stage of disease when therapeutic options are limited.1 Even those patients who present with clinical stage I lung cancer have at best a 60% 5-year survival rate, signifying that a large percentage of all stage I patients have undetectable metastatic disease at the time of presentation.1 These statistics underscore the need for improvements in early detection strategies and more accurate molecular staging of tumors. Recently, low-dose spiral computed tomography (CT) has been proposed as an early detection screening tool.2 However, despite its high sensitivity, the specificity of CT in lung cancer detection is poor. In one trial, nodules were detected in more than 70% of participants while less than 4% actually had lung cancer.2 Because of the possibility of lung cancer, all patients with an indeterminate pulmonary nodular opacity require sequential follow-up studies to assess growth. This imaging strategy is inadequate as it results in delayed diagnosis, up to 30% of resected lesions suspicious for cancer are proved to be benign, and the cost is prohibitive.3-5 Biomarkers that define high-risk patients, and that suggest which patients with nodules have lung cancer, would enhance diagnostic capabilities, complement imaging studies, and have immediate clinical benefit for lung cancer detection. This study was initiated as a first step in the effort to discover a novel panel of serum biomarkers for lung cancer. After an initial discovery phase, we focused on four serum proteins and evaluated their utility to accurately distinguish patients with a new diagnosis of lung cancer from matched control patients without lung cancer, with the goal of using these markers to treat patients with suspected lung cancer.
Protein Selection We used two different methodologies to highlight serum proteins that are expressed differentially between patients with lung cancer versus age and sex matched individuals without lung cancer. The platforms used were two-dimensional difference gel electrophoresis (2D-DIGE) and Matrix-assisted laser desorption/ionization-time of flight mass spectrometry (MALDI-TOF MS). All sera were collected under a protocol approved by our institutional review board and all patients provided written informed consent.
For the 2D-DIGE study, serum proteins from 10 patients with newly diagnosed non–small-cell lung cancer (NSCLC) were compared with those from 10 individuals without cancer (Appendix, online only). Analysis of the DIGE gels revealed four proteins (transferrin, fibrinogen β chain, retinol binding protein [RBP], and haptoglobin) whose levels differed by at least 1.5-fold between the two groups (P
For the MALDI-TOF MS experiment, serum samples from an independent group of 18 NSCLC patients and 18 control individuals were first subjected to solution-phase isoelectric focusing using a Rotofor Cell (Bio-Rad, Hercules, CA), which fractionates each sample into 20 tubes. Each fraction was then analyzed by MALDI-TOF MS. As described in the Appendix (online only), spectra generated from each of the 20 fractions for each sample were then combined to form 36 composite spectra.7 Comparison of NSCLC spectra with control spectra found a statistically significant peak at m/z 50,430 that was differentially represented in NSCLC serum compared with control serum. We identified the protein as Two well-studied cancer biomarkers, carcinoembryonic antigen (CEA) and squamous cell carcinoma antigen (SCC) were also added to the biomarker panel. Although numerous studies have demonstrated the value of each of these biomarkers in lung cancer diagnosis and treatment follow-up when used in conjunction with other biomarkers and/or clinical data, neither is effective alone.10,11
Patient Samples
Biomarker Assays The serum levels of all proteins constituting our biomarker panel (ie, transferrin, RBP, haptoglobin, AAT, CEA, and SCC) were determined in all specimens in the training and validation sets using commercially available enzyme-linked immunosorbent assays (ELISAs) according to the manufacturers' instructions (Appendix Table A2, online only). The assays for each biomarker in the training set were conducted at the same time, and the assays for each biomarker in the blinded validation set were conducted at the same.
Data Analysis The resulting classification tree was tested, without knowledge of true diagnosis, on an independent validation set of biomarker data from 97 patients, comprised of 49 serum samples from patients with a new diagnosis of lung cancer and 48 age- and sex-matched controls. Each sample was independently assigned to a terminal node based on their ELISA results, and each node is associated with a probability of malignancy derived from the analysis of the training set. Once patients in the validation set were assigned to a terminal node, the code was broken, and the accuracy of the markers was determined.
Selection of Candidates for the Biomarker Panel We used two different platforms to detect differentially expressed serum proteins from patients with or without lung cancer, with the aim of capitalizing on the tendency of each platform to identify groups of proteins possessing different physicochemical properties. We identified transferrin, RBP, and haptoglobin from the 2D-DIGE experiment and AAT from the MALDI-TOF MS experiment. We supplemented this group of biomarkers with two proteins, CEA and SCC, whose serum levels are known to vary depending on lung cancer status but have insufficient sensitivity and specificity as stand-alone diagnostic assays or in combination with a variety of other proteins.11,14,15
Training Set Selection of a Panel of Four Serum Markers
Independent Blinded Validation Set Using the Four Serum Markers We then tested this same classification tree on a separate, independent blinded set of 49 lung cancer and 48 nonlung cancer sera. The classification tree correctly classified 35 (71.4%) of 49 lung cancer sera and 32 (66.7%) of 48 noncancer control sera (sensitivity, 77.8%; specificity, 75.4%). Fifty-seven percent of all lung cancer cases (28 of 49) were binned in terminal nodes 4, 5, and 7, while only 6% of control cases (three of 48) were in these nodes. If a patient was assigned to one of these terminal nodes, they had a 90% probability (27 of 30) of having lung cancer. The probability of lung cancer for each of the seven terminal nodes is presented in Table 2. Sixty-two percent of stage I patients (10 of 16), 66% of stage II patients (two of three), 63% of stage III patients (12 of 19), and 100% of stage IV patients (11 of 11) were accurately assigned to a lung cancer node, as shown in Figure 2A. The distribution of histology according to the terminal nodes is shown in Figure 2B.
Although correct classification rates with the test set were lower overall than with the training set, individual nodes varied in their ability to classify sera in the test set. As presented in Table 2, nodes 1, 5, and 7 performed equally well or better with the test set as compared with the training set. Together, these three nodes correctly classified 20 (83%) of 24 noncancer sera and 21 (84%) of 25 cancer sera in the test set. Nodes 2, 3, 4, and 6 performed better with the training set, with correct classification rates for the test set ranging from 35% to 70%. The fact that the most accurate nodes in both the training and test sets are determined by CEA levels alone or in combination with RBP, suggests that these two biomarkers could serve as the foundation of a very powerful lung cancer serum diagnostic test.
Lung cancer accounts for more cancer deaths than any other malignancy. Despite advances in diagnostic capabilities and treatment, lung cancer mortality has not significantly changed over the past several decades. Most patients present with inoperable disease when therapeutic options including chemotherapy and radiotherapy are rarely curative. Screening studies for lung cancer with chest radiographs and sputum cytology have failed to show that this approach will decrease the number of patients who die from the disease.16 More recent trials with CT have found it can detect small tumors, but no study has shown that this will reduce lung cancer mortality, the ultimate goal of a screening program.17 These trials also illustrate a number of challenging issues with imaging as a solitary early detection strategy for lung cancer. It has been recommended that all high-risk individuals (ie, smokers and former smokers) be screened by CT.2,18 However, only a small percentage of individuals currently classified as high-risk develop lung cancer, and even some of these cancers may have an indolent phenotype (ie, there is an overdiagnosis bias).3,17,19,20 Thus, defining the optimal population to be screened remains unresolved. The high false-positive rate in CT screening trials dictates that a large number of individuals undergo follow-up studies. More troublesome is that in some CT screening trials, up to 30% of resected nodules were benign; thus a number of individuals underwent unnecessary thoracotomy, with all of the morbidity and mortality associated with this procedure.3 While several strategies to evaluate indeterminate lesions have been suggested, there is no evidence-based consensus as to the appropriate treatment of these patients. To address this fundamental limitation of CT, its low specificity, we wanted to develop a novel panel of serum markers that can distinguish lung cancer from benign lesions. The concept of biomarkers is founded on the biologic properties of cancer as a systemic disease. As a malignancy develops, it secretes proteins required for growth and metastasis and sheds cells into the circulation. The host responds by inducing changes in tissue architecture and vasculature in the microenvironment of the incipient tumor, as well as systemically mounting an immunologic defense. This consists of innate and adaptive responses with migration of inflammatory cells including macrophages, histiocytes, and lymphocytes into the tumor, and the production of autoantibodies.21-24 Thus, we predict that a combination of tumor-expressed and host response proteins can be used to develop a profile of cancer for clinical screening. Because lung cancer is a heterogeneous disease, a panel of markers that covers the broad clinical phenotype is needed. We used several different approaches to derive a set of differentially expressed serum proteins that could be employed to screen patients. Although we initially assayed for seven proteins, the CART analysis of the training set indicated that four markers in particular—CEA, RBP, SCC, and AAT—were sufficient to classify 88% of patients with cancer and 82% of patients without cancer correctly. The classification scheme produced by CART analysis assigns each patient a probability of malignancy based on the terminal node into which he or she falls. In the training set, if a patient fell into one of three terminal nodes—bins 4, 5, or 7—there was a 92% chance that the patient had lung cancer. In the validation phase of the study, the classification tree correctly classified 71.4% of the lung cancer patients and 66.6% of the noncancer control patients. Again, bins 4, 5, and 7 proved to be key in lung cancer prediction: 57% of all lung cancer cases were binned in these nodes, while only 6% of control cases were in these nodes. If a patient was assigned to one of these terminal nodes, they had a 90% chance of having lung cancer. The protein markers found in this study have all been associated with lung cancer, although none of them hold sufficient diagnostic power for patient treatment independently. In combination, however, they appear to have clinical utility. The most immediate scenario in which this panel could be used is when an indeterminate pulmonary nodule is detected on imaging studies, whether detected in a screening trial or performed for other indications. Those patients with a low-risk clinical panel who do not fall into a malignant terminal node could be followed with imaging studies at intervals determined by the risk probability of the terminal node. Patients with a high-risk clinical profile and a malignant terminal node would require immediate intervention. In this study, more than 60% of lung cancer patients fell into one of three malignant terminal nodes with a more than 90% probability of cancer. In these patients a positron emission tomography scan could then be performed for staging, followed by surgery or a biopsy, depending on other clinical factors. After the appropriate confirmatory clinical trials, one could also envision that this panel may be used in a number of scenarios. For example, the panel could be used before imaging to define high-risk patients. Those patients with a high-risk clinical profile in combination with the panel would go on to a CT. Those whose test results suggest a low probability of cancer would be re-evaluated with the serum markers during their routine follow-up. While additional markers should enhance our diagnostic accuracy, the current panel warrants further consideration in the appropriate setting. One must recognize that the terminal nodes represent probabilities that any individual will have lung cancer, and these data should be interpreted cautiously. Further studies in more patients are required to refine the diagnostic algorithms with the expectation that an efficient, optimal diagnostic strategy will improve patient outcomes.
The author(s) indicated no potential conflicts of interest.
Conception and design: Edward F. Patz Jr, Michael J. Campa, Elizabeth B. Gottlin, James E. Herndon II Financial support: Edward F. Patz Jr Administrative support: Edward F. Patz Jr, Michael J. Campa, Xiang Rong Guan Provision of study materials or patients: Edward F. Patz Jr, Michael J. Campa, Elizabeth B. Gottlin, Xiang Rong Guan Collection and assembly of data: Edward F. Patz Jr, Michael J. Campa, Elizabeth B. Gottlin, Irina Kusmartseva, Xiang Rong Guan, James E. Herndon II Data analysis and interpretation: Edward F. Patz Jr, Michael J. Campa, Elizabeth B. Gottlin, Irina Kusmartseva, James E. Herndon II Manuscript writing: Edward F. Patz Jr, Michael J. Campa, Elizabeth B. Gottlin, James E. Herndon II Final approval of manuscript: Edward F. Patz Jr, Michael J. Campa, Elizabeth B. Gottlin, Irina Kusmartseva, Xiang Rong Guan, James E. Herndon II
Methods Two-dimensional difference gel electrophoresis. Before two-dimensional difference gel electrophoresis (2D-DIGE) analysis, serum samples (10 µL each) from 10 patients with newly diagnosed non–small-cell lung cancer and from 10 individuals without cancer were depleted of six abundant proteins using a VivaPure SEPPRO Mixed 6 Kit (Vivascience, Edgewood, NY) according to the manufacturer's instructions. Depleted serum samples were diluted into buffer consisting of 8 M urea, 2 M thiourea, 20 mmol/L Tris-HCl, pH 8.5, and 4% (w/v) 3-[3-cholamidopropyl-dimethylammonio]-1-propane-sulfonate (CHAPS) and the proteins labeled with Cy3 or Cy5. Dye labeling efficiency was normalized by randomly assigning five cancer and five control sera to Cy3 and the remaining five cancer and five noncancer sera to Cy5. A pooled internal standard, consisting of equal volumes of each of the 20 depleted serum samples, was labeled with Cy2. Equal amounts of protein from all three samples (150 µg each cancer, control, and pool) were then mixed and subjected to 2D-gel electrophoresis using 13-cm immobilized pH gradient strips (Amersham Biosciences, Piscataway, NJ), pH 3-10, for the first dimension and 12% (w/v) polyacrylamide gels for the second dimension. The gels were imaged on a Typhoon 9400 Variable Mode Imager (Amersham) at the 3 excitation and emission wavelengths specific for each of the CyDyes. All gels were analyzed using DeCyder 5.0 (Amersham) software. After the gel images were cropped to exclude peripheral gel artifacts arising from the immobilized pH gradient strip and the dye front, intragel analysis was performed using DeCyder Difference In-gel Analysis software (Amersham Biosciences). Approximately 2,000 spots were autodetected in each gel and these were subsequently filtered with manually selected filter exclusion parameters. Spot maps of each filtered gel were saved and imported into biologic variation analysis software for inter-gel matching and statistical analyses. Using an independent student's t-test to score differences in spot intensity between the cancer and control sera, protein spots that were found to have a t-score of 0.1 or less were excised and submitted to the University of North Carolina Duke Proteomics Center (Chapel Hill, NC) for protein identification. Solution-phase isoelectric focusing and matrix-assisted laser desorption/ionization-time of flight mass spectrometry analysis. Before isoelectric focusing (IEF), 2 mL of each serum sample was depleted of salt by overnight dialysis (7,000 MWCO) against distilled, deionized (DD) water and then centrifuged to remove insoluble material. The dialyzed serum was diluted to 18 mL with DD water and carrier ampholytes (pH 3-10 range) were added to a final concentration of 2% (w/v). IEF was carried out at 12 W constant power using a Rotofor Prep IEF Cell (Bio-Rad, Hercules, CA) according to the manufacturer's instructions. At the completion of the Rotofor run, 20 fractions (0.5 to 1.5 mL each; average protein concentration approximately 1.5 mg/mL) were harvested into microcentrifuge tubes and stored at –80°C until use. Matrix-assisted laser desorption/ionization-time of flight mass spectrometry analysis (MALDI-TOF MS) analysis of IEF-fractionated serum was carried out using a conventional dried droplet protocol using a saturated solution of sinapinic acid (SA; Sigma Chemical Company, St Louis, MO) in 45% (v/v) acetonitrile and 0.1% (v/v) trifluoroacetic acid as the matrix. SA was obtained in crystalline form and was used without further purification. One microliter of each Rotofor fraction was diluted with 5 µL SA matrix solution and 1.2 µL of the resulting mixture deposited on the MALDI sample stage. External calibration of MALDI-TOF mass spectra was achieved by using a preset calibration equation that was defined in separate, but similar, MALDI-TOF MS experiments on proteins of known molecular weight. All MALDI-TOF MS mass spectra were acquired on a Voyager DE-PRO Biospectrometry Workstation (PerSeptive Biosystems, Framingham, MA) in the linear mode using a nitrogen laser (337 nm). Mass spectra were collected in the positive-ion mode using an acceleration voltage of 25 kV and a delay of 475 ns. The grid voltage, guide wire voltage, and low mass gate were set to 94.0%, 0.15%, and 1,000.0 m/z, respectively.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Mountain CF: Revisions in the International System for Staging Lung Cancer. Chest 111:1710-1717, 1997[CrossRef][Medline] 2. Swensen SJ, Jett JR, Hartman TE, et al: CT screening for lung cancer: Five-year prospective experience. Radiology 235:259-265, 2005 3. Sone S, Li F, Yang ZG, Honda T, et al: Results of three-year mass screening programme for lung cancer using mobile low-dose spiral computed tomography scanner. Br J Cancer 84:25-32, 2001[CrossRef][Medline] 4. Manser R, Dalton A, Carter R, et al: Cost-effectiveness analysis of screening for lung cancer with low dose spiral CT (computed tomography) in the Australian setting. Lung Cancer 48:171-185, 2005[CrossRef][Medline] 5. Mahadevia PJ, Fleisher LA, Frick KD, et al: Lung cancer screening with helical computed tomography in older adult smokers: A decision and cost-effectiveness analysis. JAMA 289:313-322, 2003 6. Gabay C, Kushner I: Acute-phase proteins and other systemic responses to inflammation. N Engl J Med 340:448-454, 1999 7. Wang MZ, Howard B, Campa MJ, et al: Analysis of human serum proteins by liquid phase isoelectric focusing and matrix-assisted laser desorption/ionization-mass spectrometry. Proteomics 3:1661-1666, 2003[CrossRef][Medline] 8. Ljujic M, Nikolic A, Divac A, et al: Screening of alpha-1-antitrypsin gene by denaturing gradient gel electrophoresis (DGGE). J Biochem Biophys Methods 68:167-173, 2006[CrossRef][Medline] 9. Zelvyte I, Wallmark A, Piitulainen E, et al: Increased plasma levels of serine proteinase inhibitors in lung cancer patients. Anticancer Res 24:241-247, 2004 10. Molina R, Agusti C, Mane JM, et al: CYFRA 21-1 in lung cancer: Comparison with CEA, CA 125, SCC and NSE serum levels Int J Biol Markers 9:96-101, 1994[Medline] 11. Schneider J: Tumor markers in detection of lung cancer. Adv Clin Chem 42:1-41, 2006[Medline] 12. Breiman L, Friedman J, Olshen R, et al: Classification and Regression Trees. Pacific Grove, CA, Wadsworth, 1984 13. Steinberg D, Golovnya M, Tolliver D, in: CART for Windows User Guide. San Diego, CA, Salford Systems, 2002 14. Tas F, Aydiner A, Topuz E, et al: Utility of the serum tumor markers: CYFRA 21.1, carcinoembryonic antigen (CEA), and squamous cell carcinoma antigen (SCC) in squamous cell lung cancer. J Exp Clin Cancer Res 19:477-481, 2000[Medline] 15. Kulpa J, Wojcik E, Reinfuss M, et al: Carcinoembryonic antigen, squamous cell carcinoma antigen, CYFRA 21-1, and neuron-specific enolase in squamous cell lung cancer patients Clin Chem 48:1931-1937, 2002 16. Marcus PM, Bergstralh EJ, Fagerstrom RM, et al: Lung cancer mortality in the Mayo Lung Project: Impact of extended follow-up. J Natl Cancer Inst 92:1308-1316, 2000 17. Bach PB, Jett JR, Pastorino U, et al: Computed tomography screening and lung cancer outcomes. JAMA 297:953-961, 2007 18. Henschke CI, Yankelevitz DF, Libby DM, et al: Survival of patients with stage I lung cancer detected on CT screening. N Engl J Med 355:1763-1771, 2006 19. Patz EF Jr, Goodman PC, Bepler G: Screening for lung cancer. N Engl J Med 343:1627-1633, 2000 20. Patz EF: Jr. Lung cancer screening, overdiagnosis bias, and reevaluation of the Mayo Lung Project. J Natl Cancer Inst 98:724-725, 2006 21. Petersen RP, Campa MJ, Sperlazza J, et al: Tumor infiltrating Foxp3+ regulatory T-cells are associated with recurrence in pathologic stage I NSCLC patients. Cancer 107:2866-2872, 2006[CrossRef][Medline] 22. Zhong L, Coe SP, Stromberg AJ, et al: Profiling tumor-associated antibodies for early detection of non-small cell lung cancer. J Thorac Oncol 1:513-519, 2006[CrossRef][Medline] 23. Welsh TJ, Green RH, Richardson D, et al: Macrophage and mast-cell invasion of tumor cell islets confers a marked survival advantage in non-small-cell lung cancer. J Clin Oncol 23:8959-8967, 2005 24. Condeelis J, Pollard JW: Macrophages: Obligate partners for tumor cell migration, invasion, and metastasis. Cell 124:263-266, 2006[CrossRef][Medline] Submitted July 16, 2007; accepted September 12, 2007.
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2007 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
1-antitrypsin, and squamous cell carcinoma antigen—were collectively found to correctly classify the majority of lung cancer and control patients in the training set (sensitivity, 89.3%; specificity, 84.7%). These markers also accurately classified patients in the independent validation set (sensitivity, 77.8%; specificity, 75.4%). Remarkably, 90% of patients who fell into any one of three groupings in the CART analysis had lung cancer. 
.05; Appendix Table A1). We chose to exclude fibrinogen β chain in the biomarker panel because its involvement in coagulation and status as an acute phase protein would likely influence its concentration in serum irrespective of the presence of lung cancer.
