Identification of Serological Biomarkers for Early Diagnosis of Lung Cancer Using a Protein Array-Based Approach
2017; Elsevier BV; Volume: 16; Issue: 12 Linguagem: Inglês
10.1074/mcp.ra117.000212
ISSN1535-9484
AutoresJianbo Pan, Guang Song, Dunyan Chen, Yadong Li, Shuang Liu, Shaohui Hu, Christian Rosa, Daniel Eichinger, Ignacio Pino, Heng Zhu, Jiang Qian, Yi Huang,
Tópico(s)Monoclonal and Polyclonal Antibodies Research
ResumoLung cancer (LC) remains the leading cause of mortality from malignant tumors worldwide. Currently, a lack of serological biomarkers for early LC diagnosis is a major roadblock for early intervention and prevention of LC. To undertake this challenge, we employed a two-phase strategy to discover and validate a biomarker panel using a protein array-based approach. In Phase I, we obtained serological autoimmune profiles of 80 LC patients and 20 healthy subjects on HuProt arrays, and identified 170 candidate proteins significantly associated with LC. In Phase II, we constructed a LC focused array with the 170 proteins, and profiled a large cohort, comprised of 352 LC patients, 93 healthy individuals, and 101 patients with lung benign lesions (LBL). The comparison of autoimmune profiles between the early stage LC and the combined group of healthy and LBL allowed us to identify and validate a biomarker panel of p53, HRas, and ETHE1 for diagnosis of early stage LC with 50% sensitivity at >90% specificity. Finally, the performance of this biomarker panel was confirmed in ELISA tests. In summary, this study represents one of the most comprehensive proteome-wide surveys with one of the largest (i.e. 1,101 unique samples) and most diverse (i.e. nine disease groups) cohorts, resulting in a biomarker panel with good performance. Lung cancer (LC) remains the leading cause of mortality from malignant tumors worldwide. Currently, a lack of serological biomarkers for early LC diagnosis is a major roadblock for early intervention and prevention of LC. To undertake this challenge, we employed a two-phase strategy to discover and validate a biomarker panel using a protein array-based approach. In Phase I, we obtained serological autoimmune profiles of 80 LC patients and 20 healthy subjects on HuProt arrays, and identified 170 candidate proteins significantly associated with LC. In Phase II, we constructed a LC focused array with the 170 proteins, and profiled a large cohort, comprised of 352 LC patients, 93 healthy individuals, and 101 patients with lung benign lesions (LBL). The comparison of autoimmune profiles between the early stage LC and the combined group of healthy and LBL allowed us to identify and validate a biomarker panel of p53, HRas, and ETHE1 for diagnosis of early stage LC with 50% sensitivity at >90% specificity. Finally, the performance of this biomarker panel was confirmed in ELISA tests. In summary, this study represents one of the most comprehensive proteome-wide surveys with one of the largest (i.e. 1,101 unique samples) and most diverse (i.e. nine disease groups) cohorts, resulting in a biomarker panel with good performance. Lung cancer (LC) 1The abbreviations used are: LC; lung cancer; LBL, lung benign lesions; SCLC, small-cell lung cancer; NSCLC, nonsmall-cell lung cancer; CT, computed tomography; COPD, chronic obstructive pulmonary disease; TB, pulmonary tuberculosis; RC, rectal cancer; LiC, liver cancer; CC, cervical cancer; EC, esophagus cancer; GC, gastric cancer; p53, tumor protein p53; HRas, HRas proto-oncogene, GTPase; ETHE1, ETHE1 persulfide dioxygenase. 1The abbreviations used are: LC; lung cancer; LBL, lung benign lesions; SCLC, small-cell lung cancer; NSCLC, nonsmall-cell lung cancer; CT, computed tomography; COPD, chronic obstructive pulmonary disease; TB, pulmonary tuberculosis; RC, rectal cancer; LiC, liver cancer; CC, cervical cancer; EC, esophagus cancer; GC, gastric cancer; p53, tumor protein p53; HRas, HRas proto-oncogene, GTPase; ETHE1, ETHE1 persulfide dioxygenase.remains the leading cause of mortality from malignant tumors worldwide (1.Ferlay J. Soerjomataram I. Dikshit R. Eser S. Mathers C. Rebelo M. Parkin D.M. Forman D. Bray F. Cancer incidence and mortality worldwide: Sources, methods and major patterns in GLOBOCAN 2012.Int. J. Cancer. 2015; 136: E359-E386Crossref PubMed Scopus (21344) Google Scholar, 2.Siegel R. Naishadham D. Jemal A. Cancer statistics, 2013.CA Cancer. J. Clin. 2013; 63: 11-30Crossref PubMed Scopus (11524) Google Scholar). According to the World Health Organization (WHO), among the 8.8 million cancer-related deaths in 2015, LC caused 1.69 million deaths worldwide (3.Cancer. Fact Sheet. [Internet] Geneva: World Health Organization; 2017 [cited 2017 Oct 4]. Available from: http://www.who.int/mediacentre/factsheets/fs297/en/,.Google Scholar). In the most populated country China, LC alone is responsible for the mortality of 42.05 per 100,000 persons (4.Chen W. Zheng R. Zuo T. Zeng H. Zhang S. He J. National cancer incidence and mortality in china, 2012.Chin. J. Cancer Res. 2016; 28: 1-11Crossref PubMed Google Scholar). LC can be histologically categorized into two main classes: small-cell lung cancer (SCLC) and nonsmall-cell lung cancer (NSCLC). Approximately 79% of diagnosed LC is NSCLC, comprised of adenocarcinoma, squamous cell carcinoma and large cell carcinoma (5.Travis W. Brambilla E. Muller-Hermelink H. Harris C. Pathology and genetics: Tumours of the lung, pleura, thymus and heart. International Agency for Research on Cancer (IARC), Lyon2004: 9-124Google Scholar). Regardless of the great advancements in targeted therapy and immunotherapy against LC in recent years, surgical resection followed by adjunctive radiation and/or chemotherapy is still the preferred method in the treatment of NSCLC patients in early stages (e.g. I-II LC), and when surgery is performed, there is a 70% one-year survival rate if the diagnosis is made at the earliest stage (6.Goldstein S.D. Yang S.C. Role of surgery in small cell lung cancer.Surg. Oncol. Clin. N. Am. 2011; 20: 769-777Abstract Full Text Full Text PDF PubMed Scopus (7) Google Scholar). Unfortunately, most LC patients are found in late stages at the time of diagnosis. For example, more than 75% of LC patients are diagnosed at more advanced stages (7.National Lung Screening Trial Research Team Aberle D.R. Berg C.D. Black W.C. Church T.R. Fagerstrom R.M. Galen B. Gareen I.F. Gatsonis C. Goldin J. Gohagan J.K. Hillman B. Jaffe C. Kramer B.S. Lynch D. Marcus P.M. Schnall M. Sullivan D.C. Sullivan D. Zylak C.J. The national lung screening trial: Overview and study design.Radiology. 2011; 258: 243-253Crossref PubMed Scopus (829) Google Scholar). Currently, high-resolution (or low-dose) computed tomography (CT) of the chest is the only screening test shown to be efficacious at reducing mortality from early stages of lung cancer (8.National Lung Screening Trial Research Team Aberle D.R. Adams A.M. Berg C.D. Black W.C. Clapp J.D. Fagerstrom R.M. Gareen I.F. Gatsonis C. Marcus P.M. Sicks J.D. Reduced lung-cancer mortality with low-dose computed tomographic screening.N. Engl. J. Med. 2011; 365: 395-409Crossref PubMed Scopus (6913) Google Scholar, 9.Sox H.C. Screening for lung cancer with chest radiographs.JAMA. 2011; 306: 1916-1918Crossref PubMed Scopus (15) Google Scholar, 10.Shieh Y. Bohnenkamp M. Low-dose computed tomography for lung cancer screening: Clinical and coding considerations.Chest. 2017; 152: 204-209Abstract Full Text Full Text PDF PubMed Scopus (27) Google Scholar). Indeed, as reported by the National Lung Screening Trial (NLST) of randomized 53,454 high-risk, asymptomatic adults, three rounds of annual screening with low-dose CT decreased LC mortality by 20% (8.National Lung Screening Trial Research Team Aberle D.R. Adams A.M. Berg C.D. Black W.C. Clapp J.D. Fagerstrom R.M. Gareen I.F. Gatsonis C. Marcus P.M. Sicks J.D. Reduced lung-cancer mortality with low-dose computed tomographic screening.N. Engl. J. Med. 2011; 365: 395-409Crossref PubMed Scopus (6913) Google Scholar). In fact, LC was only diagnosed in 90% specificity. ELISA tests further demonstrated the potential of this biomarker panel in future clinical diagnostic test formats. All serum samples involved in this study were collected at Fujian Provincial Hospital, in Fujian Province, China, between 2014 and 2016. This cohort was comprised of 1101 serum samples collected from 162 healthy persons, 560 resident patients with LC, 153 resident patients with lung benign lesions (LBL), and 226 resident patients with other cancers. The 162 healthy persons were recruited during annual health examinations, including chest X-ray, abdominal ultrasonography, routine urinalysis, stool occult blood test, complete blood count, blood chemistries, and tumor antigen tests, such as carcinoembryonic antigen (CEA), CA199, and alphafetoprotein (AFP), to name a few. None of them showed any evidence of malignancy in the above tests. The 560 LC patients were recruited after histopathological confirmation of LC tumors. The TNM classification was used for evaluation of NSCLC staging and the VA scheme was used to classify SCLC into limited- and extensive-stages. The 153 LBL patients, including 83 pneumonia, 39 chronic obstructive pulmonary disease (COPD) and 31 pulmonary tuberculosis (TB), were recruited after accurate clinical assessment. The 226 patients with other cancers were recruited after histopathological confirmation of tumors. These patients included 34, 66, 27, 48, and 51 patients with rectal cancer (RC), liver cancer (LiC), cervical cancer (CC), esophagus cancer (EC), and gastric cancer (GC), respectively. Detailed information of each subject of this cohort is listed in supplemental Table 1. This study was approved by the Ethics Committee (i.e. IRB) of Fujian Provincial Hospital. The sera were prepared according to standard protocol. Five milliliters venous blood of each subject was collected into a 12.5 × 100 mm vacuum blood tube with diatomite coagulant, and centrifuged at 4000 rpm for 10 min at room temperature within 4 h after collection. Subsequently, sera were collected into 1.5 ml EP tubes and then stored at −80 °C until use. HuProt arrays were provided by CDI Laboratories, Inc. Each HuProt v3.0 array is comprised of 20,240 unique human full-length proteins, covering ∼75% of the human proteome. Each serum sample was diluted 1000-fold in PBS, and profiled on HuProt arrays using a standard protocol as described previously (12.Hu S. Wan J. Su Y. Song Q. Zeng Y. Nguyen H.N. Shin J. Cox E. Rho H.S. Woodard C. Xia S. Liu S. Lyu H. Ming G.L. Wade H. Song H. Qian J. Zhu H. DNA methylation presents distinct binding sites for human transcription factors.Elife. 2013; 2: e00726Crossref PubMed Scopus (238) Google Scholar, 13.Hu C.J. Pan J.B. Song G. Wen X.T. Wu Z.Y. Chen S. Mo W.X. Zhang F.C. Qian J. Zhu H. Li Y.Z. Identification of novel biomarkers for behcet disease diagnosis using HuProt array approach.Mol. Cell. Proteomics. 2017; 16: 147-156Abstract Full Text Full Text PDF PubMed Scopus (32) Google Scholar, 14.Yang L. Wang J. Li J. Zhang H. Guo S. Yan M. Zhu Z. Lan B. Ding Y. Xu M. Li W. Gu X. Qi C. Zhu H. Shao Z. Liu B. Tao S.C. Identification of serum biomarkers for gastric cancer diagnosis using a human proteome microarray.Mol. Cell. Proteomics. 2016; 15: 614-623Abstract Full Text Full Text PDF PubMed Scopus (75) Google Scholar, 15.Syed P. Gupta S. Choudhary S. Pandala N.G. Atak A. Richharia A. K. P. M. Zhu H. Epari S. Noronha S.B. Moiyadi A. Srivastava S. Autoantibody profiling of glioma serum samples to identify biomarkers using human proteome arrays.Sci. Rep. 2015; 5: 13895Crossref PubMed Scopus (35) Google Scholar). Candidate proteins identified in the HuProt array experiments were cherry-picked to fabricate the LC focused arrays in a 2 × 7 subarray format per slide. A 14-chamber rubber gasket (GraceBio Corp, Bend, OR) was mounted onto each slide to create individual chambers for the 14 identical subarrays on each slide. The subsequent assay process was identical to that described for HuProt array assay, with an exception that the volume of buffers or serum samples was reduced to 50 μl per subarray (12.Hu S. Wan J. Su Y. Song Q. Zeng Y. Nguyen H.N. Shin J. Cox E. Rho H.S. Woodard C. Xia S. Liu S. Lyu H. Ming G.L. Wade H. Song H. Qian J. Zhu H. DNA methylation presents distinct binding sites for human transcription factors.Elife. 2013; 2: e00726Crossref PubMed Scopus (238) Google Scholar). First, the median values of the foreground (Fij) and background (Bij) intensity at a give protein spot (i,j) on the protein arrays (i.e. HuProt and focused arrays) were extracted. The signal intensity (Rij) of each protein spot was defined as Fij/Bij. Because each protein is printed in duplicate on an array, Rij was averaged for each protein as Rp. Z-score of each protein on protein arrays was calculated using a method similar to the one described in our previous studies (12.Hu S. Wan J. Su Y. Song Q. Zeng Y. Nguyen H.N. Shin J. Cox E. Rho H.S. Woodard C. Xia S. Liu S. Lyu H. Ming G.L. Wade H. Song H. Qian J. Zhu H. DNA methylation presents distinct binding sites for human transcription factors.Elife. 2013; 2: e00726Crossref PubMed Scopus (238) Google Scholar). A stringent cutoff (Z ≥ 7) was used to determine the positives in this study. The sensitivity and specificity were calculated for each protein. For each comparison (LC versus negative controls), the biomarker candidates were selected with the highest discriminant ability (16.Riegelman, R. K., (2012) Studying A study & testing A test: Reading evidence-based health research 6th.Google Scholar), which is defined as Discriminantability=Sensitivity+specificity2(Eq. 1) For the focused arrays fabricated with the candidate biomarkers, the signal value for each protein was normalized by dividing the median value of negative controls for each sample. p values obtained from the t test were calculated and adjusted as false discovery rates (17.Reich M. Liefeld T. Gould J. Lerner J. Tamayo P. Mesirov J.P. GenePattern 2.0.Nat. Genet. 2006; 38: 500-501Crossref PubMed Scopus (1536) Google Scholar). The optimal cutoff value for each candidate was evaluated with two criteria: 1) at least 90% specificity and 2) the highest discriminant ability. To develop ELISA-based assays, p53, HRas, and ETHE1 proteins were purified from yeast as described previously (18.Song G. Hu C. Zhu H. Wang L. Zhang F. Li Y. Wu L. New centromere autoantigens identified in systemic sclerosis using centromere protein microarrays.J. Rheumatol. 2013; 40: 461-468Crossref PubMed Scopus (24) Google Scholar). After 50 ng of each purified protein was coated onto individual wells of an ELISA plate, each serum sample in 1:500-fold dilution was added to carry out the standard ELISA tests (18.Song G. Hu C. Zhu H. Wang L. Zhang F. Li Y. Wu L. New centromere autoantigens identified in systemic sclerosis using centromere protein microarrays.J. Rheumatol. 2013; 40: 461-468Crossref PubMed Scopus (24) Google Scholar). The immunoreactivity signals were measured by reading the A450. We employed the two-phase strategy reported in our previous studies (13.Hu C.J. Pan J.B. Song G. Wen X.T. Wu Z.Y. Chen S. Mo W.X. Zhang F.C. Qian J. Zhu H. Li Y.Z. Identification of novel biomarkers for behcet disease diagnosis using HuProt array approach.Mol. Cell. Proteomics. 2017; 16: 147-156Abstract Full Text Full Text PDF PubMed Scopus (32) Google Scholar, 14.Yang L. Wang J. Li J. Zhang H. Guo S. Yan M. Zhu Z. Lan B. Ding Y. Xu M. Li W. Gu X. Qi C. Zhu H. Shao Z. Liu B. Tao S.C. Identification of serum biomarkers for gastric cancer diagnosis using a human proteome microarray.Mol. Cell. Proteomics. 2016; 15: 614-623Abstract Full Text Full Text PDF PubMed Scopus (75) Google Scholar) to identify novel biomarkers for early LC diagnosis (Fig. 1). Briefly, in Phase I, 100 serum samples collected from 80 LC patients and 20 healthy individuals, were individually profiled on HuProt arrays. After data analysis, a total of 170 candidate proteins were identified and used to construct the LC focused arrays for Phase II validation. In Phase II, we assembled a new cohort with serum samples collected from 131 patients with early stage LC and 93 healthy subjects. Because lung benign lesions (LBL) often resemble early stage LC in imaging studies, we also included 101 LBL samples as additional negative controls. We randomly split the LC samples and negative controls (healthy + LBL) in a 2:1 ratio - two thirds were used for modeling and one third for independent validation of biomarker candidates. Eight biomarkers were validated with > 13% sensitivity at > 90% specificity. Further analysis resulted in a three-protein biomarker panel with improved sensitivity, and its performance was further tested in late stage LC and other types of cancer. Finally, this panel was converted into an ELISA-based test that yielded a performance like that observed in the array-based assays. In Phase I, we employed HuProt arrays to profile 100 serum samples collected from 80 LC patients, including 20 SCLC, 24 adenocarcinoma, 23 squamous-cell carcinoma, and 13 large-cell carcinoma, as well as 20 healthy subjects, for candidate biomarker identification (Table I; supplemental Table S1). Statistic analyses did not show any significant differences between the LC and healthy groups in terms of age, gender or smoking history composition (Table I).Table ICharacteristics of the samples in Phase IVariableLC (n = 80)Healthy (n = 20)PNo.Mean%No.Mean%Age (years)0.086Mean60.456.4Standard deviation8.511.1Sex0.223Male6682.51470.0Female1417.5630.0Smoking history (pack-years)102227.5525.0<201620.0630.0≥204252.5945.0TypeSmall Cell Lung Cancer2025.0Large Cell Lung Cancer1316.3Adenocarcinoma2430.0Squamous Cell Carcinoma2328.8 Open table in a new tab Each serum sample was diluted and individually incubated on the HuProt arrays, followed by a multiplexed detection of autoantigens that could be recognized by human autoantibodies of the IgG and IgM isotypes. Binding signals of both anti-IgG and -IgM channels were acquired, normalized, and quantified for each assay, based on which standard deviation (S.D.) was calculated (12.Hu S. Wan J. Su Y. Song Q. Zeng Y. Nguyen H.N. Shin J. Cox E. Rho H.S. Woodard C. Xia S. Liu S. Lyu H. Ming G.L. Wade H. Song H. Qian J. Zhu H. DNA methylation presents distinct binding sites for human transcription factors.Elife. 2013; 2: e00726Crossref PubMed Scopus (238) Google Scholar). Using a stringent cut off (Z score ≥ 7), positives were determined for each serum sample. For example, p53 and YARS showed respectively strong anti-human IgG and IgM signals, mostly in LC patients, but less so in healthy subjects (Fig. 2A). Sensitivity and specificity values were calculated for each protein. We chose a generous criterion (i.e. discriminant ability ≥ 60%), resulted in identification of 170 candidate proteins, 105 and 77 of which were chosen from the anti-IgG and -IgM profiles, respectively (Fig. 2B; supplemental Table S2). Functional enrichment analysis identified many cancer-relevant terms, such as regulation of apoptosis and small GTPase mediated signal transduction, as well as signaling pathways relevant to cancers such as colorectal cancer, pancreatic cancer, and thyroid cancer (FDR < 0.5) (19.Orgaz J.L. Herraiz C. Sanz-Moreno V. Rho GTPases modulate malignant transformation of tumor cells.Small GTPases. 2014; 5: e29019Crossref PubMed Scopus (18) Google Scholar, 20.Huang da W. Sherman B.T. Lempicki R.A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources.Nat. Protoc. 2009; 4: 44-57Crossref PubMed Scopus (25470) Google Scholar) (supplemental Table S3). In Phase II, we fabricated a LC focused array with the 170 candidate biomarker proteins to enable validation with a much larger cohort. We assembled a new LC cohort with serum samples collected from 131 patients with early stage LC, including 30 limited stage SCLC, 55 stage I/II adenocarcinoma, and 46 stage I/II squamous-cell carcinoma. Negative controls included 93 healthy subjects and 101 serum samples from 55 pneumonia, 26 COPD, and 20 pulmonary TB patients. Statistic analysis did not find any significant differences in age, gender or smoking history between the LC groups and negative controls (Table II; supplemental Table S1). To enable modeling and validation for biomarker identification, we randomly split each LC subgroup and negative controls in a 2:1 ratio - two thirds were used for modeling and one third for subsequent independent validation of biomarker candidates.Table IICharacteristics of the samples in Phase IIVariableEarly LC (n = 131)Healthy (n = 93)LBL (n = 101)PNo.Mean%No.Mean%No.Mean%Age (years)0.165Mean61.258.361.1Standard deviation10.08.48.8Sex0.102Male10177.16468.86968.3Female3022.92931.23231.7Smoking history (pack-years)0.10802418.32324.72827.7<202922.12425.82120.8≥207859.64649.55251.5TypeSmall Cell Lung Cancer3022.9Adenocarcinoma5542.0Squamous Cell Carcinoma4635.1 Open table in a new tab Each serum sample was profiled individually on the LC focused arrays using a similar protocol as described above. Again, both anti-IgG and -IgM profiles were obtained simultaneously. In the modeling stage, we compared the serum profiles between the LC and negative controls to identify biomarkers using stringent criteria—FDR < 0.001 and ≥ 1.20 fold-change of average signal intensity between the two groups. The analysis of IgG identified eight proteins, namely p53, ETHE1, CTAG1A, C1QTNF1, TEX264, CLDN2, NSG1, and HRas (Table III). However, the same analysis did not reveal any significant biomarkers using the anti-IgM signals. The IgG signal distributions of p53, ETHE1 and HRas in the LC and negative controls are shown as examples in Fig. 3A. Areas under the receiver operating characteristic (ROC) curves (AUCs) were calculated to assess the performance of each candidate biomarker. The AUC values of the eight proteins ranged from 0.68 to 0.81 (Table III). We next calculated the maximum discriminant ability values for each protein with a requirement of a minimum specificity of 90% (see Methods). This approach allowed us to determine the optimal cutoff values of signal intensity for each protein with the corresponding sensitivity and specificity values (Table III).Table IIIPerformance of eight biomarkers in discovery and validation stages of Phase IIProteinDiscoveryValidationAUCCut offSensitivitySpecificitySensitivitySpecificityp530.8091.20924.1%93.8%22.7%96.9%ETHE10.7851.86132.2%91.5%29.5%93.8%CTAG1A0.7841.20017.2%96.1%18.2%93.8%C1QTNF10.7631.57726.4%90.7%22.7%93.8%TEX2640.7592.08823.0%92.2%20.5%93.8%CLDN20.7441.82026.4%90.7%22.7%95.4%NSG10.7401.73527.6%91.5%29.5%92.3%HRas0.6921.98013.8%96.9%18.2%93.8% Open table in a new tab To validate these potential LC biomarkers, we compared the signal intensity of each protein between the LC and negative controls in the validation cohort. As visualized in the box plot analysis, all of them showed significantly higher signal intensities in the LC than the negative controls (supplemental Fig. S1). Three proteins, p53, ETHE1, and HRas, are shown as examples in Fig. 3A. We next applied the optimal cut-off values obtained in the modeling stage to determine the sensitivity and specificity for each protein in the validation cohort. All of the eight proteins yielded similar or better sensitivity and specificity values in the validation cohort (Fig. 3B; supplemental Fig. S1), confirming that the identified biomarkers have robust classification power for early stage LC diagnosis. We noticed that the sensitivity values of each biomarker ranged from 13.8% to 32.2%. Therefore, we attempted to identify combinatorial biomarker panels with better performance. We exhaustively evaluated the performance for all possible combinations between two and eight proteins (=253 combinations). First, we employed a binary scoring system to convert the actual signal intensity of each protein to either 1 or 0, such that 1 represented signal intensity greater than the optimal cutoff value, and 0 otherwise. Next, we evaluated the performance of every possible combination in the discovery cohort. For a given combination of n proteins, the sum of the binary scores of the n proteins was assigned to each serum sample as a summary score. If the summary score of a sample was greater than k (1 ≤ k ≤ n), the sample was called positive. The sensitivity and specificity at the best discriminant ability value were recorded for each combination. Finally, we identified the combination and its k value with the best discriminant ability by requiring a minimum specificity of 90%. As a result, the best combination, comprised of p53, ETHE1, and HRas, achieved 50.7% sensitivity at 90.7% specificity with a k value of 1. In other words, a serum sample would be scored positive when at least one (i.e. k = 1) of the three proteins showed signal intensity greater than the corresponding optimal cutoff value. When this panel was applied to the validation cohort, we obtained similar values of sensitivity and specificity (Fig. 3B), demonstrating the robustness of this panel in diagnosis of early LC. Moreover, after combing the results of the discovery and validation stages, the overall sensitivity for diagnosis of SCLC of limited stage and stage I/II adenocarcinoma, squamous cell carcinoma is 53.3%, 45.5% and 54.3%, respectively. When only high-risk smokers (i.e. ≥ 20-pack year & age > 55 years) were compared between early LC and negative controls, the performance of this biomarker panel remained almost the same at 50.0% sensitivity and 84.8% specificity. To evaluate potential value of this biomarker panel in late stage LC diagnosis, we recruited a new LC cohort of 221 serum samples, collected from 43 patients with extensive stage SCLC, 99 patients with stage III/IV adenocarcinoma, and 79 patients with stage III/IV squamous-cell carcinoma, and profiled them on the LC focused arrays. By applying this biomarker panel to analyze the obtained data set, we observed a sensitivity of 49.8%, suggesting that this biomarker panel was also useful for late stage LC diagnosis. It is known that many of the same tumor antigens can be found in patients with a wide variety of cancers, diminishing their value for accurate diagnosis of a cancer type. To evaluate the performance of this biomarker panel in other types of cancer, we profiled a cohort of 226 serum samples, collected from 34 rectal cancer (RC), 66 liver cancer (LiC), 27 cervical cancer (CC), 48 esophagus cancer (EC), and 51 gastric cancer (GC) patients, on the LC focused arrays. Interestingly, this biomarker panel could only detect 23.5%, 21.2%, 22.2%, 37.5%, and 39.2% of RC, LiC, CC, EC, and GC, respectively (Fig. 3C). This comparison indicated that this biomarker panel is clearly more sensitive in detecting LC. To transform the array-validated biomarker panel into a more clinically friendly platform, we developed an enzyme-linked immunosorbent assay (ELISA) for the three proteins. Two cohorts were assembled: one contained 226 samples randomly selected from those used in Phase II and 229 newly collected samples (see Fig. 1; supplemental Table S1). As expected, analysis of the ELISA data obtained with the samples used in the array-based assays demonstrated that all three proteins showed significantly higher signals in both early and late LC groups as compared with those in healthy and LBL groups. To ensure more rigorous tests, the 229 newly collected samples were tested in a single-blind fashion. A similar result was obtained (Fig. 4A). We next evaluated the performance of this biomarker panel with the combined ELISA data sets. The ELISA data were converted to a binary scoring system by using a cut off value of 2-S.D. above the mean of the signal intensity of the combined healthy group, following the standard ELISA protocol. Using the same criteria as described above, 49.6% and 58.8% of samples in the early and late stages of LC, respectively, were scored as positives (Fig. 4B). In contrast, only 10.3% and 13.7% of healthy and LBL samples were respectively scored as false positives. Therefore, this biomarker panel showed 49.6% sensitivity at 87.9% specificity for early LC diagnosis in the ELISA tests. Moreover, the overall sensitivity obtained in the ELISA tests for diagnosis of SCLC of limited stage and stage I/II adenocarcinoma, squamous cell carcinoma is 55.9%, 44.4% and 48.9%, respectively. Our study design possessed and displayed several strengths. First, we employed the most comprehensive human proteome (HuProt) arrays, with >75% coverage of the human proteome to improve the likelihood of finding potential
Referência(s)