Humoral Response Profiling Reveals Pathways to Prostate Cancer Progression
2007; Elsevier BV; Volume: 7; Issue: 3 Linguagem: Inglês
10.1074/mcp.m700263-mcp200
ISSN1535-9484
AutoresBarry S. Taylor, Manoj K. Pal, Jianjun Yu, Bharathi Laxman, Shanker Kalyana‐Sundaram, Rong Zhao, Anjana Menon, John T. Wei, Alexey I. Nesvizhskii, Debashis Ghosh, Gilbert S. Omenn, David M. Lubman, Arul M. Chinnaiyan, Arun Sreekumar,
Tópico(s)Monoclonal and Polyclonal Antibodies Research
ResumoThere is considerable evidence for an association between prostate cancer development and inflammation, which results in autoantibody generation against tumor proteins. This immune system-driven amplification of the autoantibody response to intracellular antigens can serve as a sensitive tool to detect low abundance serum proteomic tumor markers for prostate cancer as well as provide insight into biological processes perturbed during cancer development. Here we examine serum humoral responses in a cohort of 34 patients with either benign prostatic hyperplasia or clinically localized prostate cancer (PCa). The experimental strategy couples multidimensional liquid-phase protein fractionation of localized and metastatic prostate cancer tissue lysates to protein microarrays and subsequent mass spectrometry. A supervised learning analysis of the humoral response arrays generated a parsimonious predictor having 78% sensitivity and 75% specificity in distinguishing PCa from benign prostatic hyperplasia in a cohort of American males with elevated prostate-specific antigen. Enrichment analysis of the PCa-specific humoral signature revealed large scale immune reprogramming mediated by STAT transcription factors and the generation of autoantibodies to enzymes involved in nitrogen metabolism. Meta-analysis of independent prostate cancer gene expression data validated the presence of STAT-induced immunomodulation. Concomitant validation of elevated levels of the nitrogen metabolism pathway was obtained by direct measurement of metabolic levels of glutamate and aspartate in prostate cancer tissues. Thus, in addition to functioning as markers in prostate cancer detection, humoral response profiles can serve as powerful tools revealing pathway dysregulation that might otherwise be suppressed by the complexity of the cancer proteome. There is considerable evidence for an association between prostate cancer development and inflammation, which results in autoantibody generation against tumor proteins. This immune system-driven amplification of the autoantibody response to intracellular antigens can serve as a sensitive tool to detect low abundance serum proteomic tumor markers for prostate cancer as well as provide insight into biological processes perturbed during cancer development. Here we examine serum humoral responses in a cohort of 34 patients with either benign prostatic hyperplasia or clinically localized prostate cancer (PCa). The experimental strategy couples multidimensional liquid-phase protein fractionation of localized and metastatic prostate cancer tissue lysates to protein microarrays and subsequent mass spectrometry. A supervised learning analysis of the humoral response arrays generated a parsimonious predictor having 78% sensitivity and 75% specificity in distinguishing PCa from benign prostatic hyperplasia in a cohort of American males with elevated prostate-specific antigen. Enrichment analysis of the PCa-specific humoral signature revealed large scale immune reprogramming mediated by STAT transcription factors and the generation of autoantibodies to enzymes involved in nitrogen metabolism. Meta-analysis of independent prostate cancer gene expression data validated the presence of STAT-induced immunomodulation. Concomitant validation of elevated levels of the nitrogen metabolism pathway was obtained by direct measurement of metabolic levels of glutamate and aspartate in prostate cancer tissues. Thus, in addition to functioning as markers in prostate cancer detection, humoral response profiles can serve as powerful tools revealing pathway dysregulation that might otherwise be suppressed by the complexity of the cancer proteome. Although prostate carcinoma is the leading cancer diagnosis in American men, its early detection facilitates effective treatment modalities and improved mortality (1Jemal A. Murray T. Ward E. Samuels A. Tiwari R.C. Ghafoor A. Feuer E.J. Thun M.J. Cancer statistics, 2005.CA Cancer J. Clin. 2005; 55: 10-30Crossref PubMed Scopus (5548) Google Scholar). Although the advent of prostate-specific antigen (PSA) 1The abbreviations used are: PSA, prostate-specific antigen; BPH, benign prostatic hyperplasia; PCa, prostate cancer; STAT, signal transducers and activator of transcription; SVM, support vector machine; LOOCV, leave-one-out cross-validation; MCM, molecular concept map; CF, chromatofocusing; bis-tris, 2-[bis(2-hydroxyethyl)amino]-2-(hydroxymethyl)propane-1,3-diol; RP, reverse-phase; FGF, fibroblast growth factor. 1The abbreviations used are: PSA, prostate-specific antigen; BPH, benign prostatic hyperplasia; PCa, prostate cancer; STAT, signal transducers and activator of transcription; SVM, support vector machine; LOOCV, leave-one-out cross-validation; MCM, molecular concept map; CF, chromatofocusing; bis-tris, 2-[bis(2-hydroxyethyl)amino]-2-(hydroxymethyl)propane-1,3-diol; RP, reverse-phase; FGF, fibroblast growth factor. screening has led to earlier detection of prostate cancers (2Catalona W.J. Management of cancer of the prostate.N. Engl. J. Med. 1994; 331: 996-1004Crossref PubMed Scopus (265) Google Scholar), its lack of specificity for neoplasm has resulted in a dramatic increase in the number of subsequent prostate needle biopsies (3Jacobsen S.J. Katusic S.K. Bergstralh E.J. Oesterling J.E. Ohrt D. Klee G.G. Chute C.G. Lieber M.M. Incidence of prostate cancer diagnosis in the eras before and after serum prostate-specific antigen testing.J. Am. Med. Assoc. 1995; 274: 1445-1449Crossref PubMed Scopus (199) Google Scholar). As the population of men 65 years and older is expected to increase from 14 million in the year 2000 to 31 million by 2030 (4Brown C. Sauvageot J. Kahane H. Epstein J.I. Cell proliferation and apoptosis in prostate cancer—correlation with pathologic stage?.Mod. Pathol. 1996; 9: 205-209PubMed Google Scholar), it will be increasingly important to distinguish men with benign prostatic hyperplasia from those having neoplastic disease warranting clinical intervention. Thus, there is a compelling need to define additional clinical markers for accurate detection of prostate cancers. This situation has spawned a wide range of serum-based early detection methodologies, including protein microarrays (5Wulfkuhle J.D. Liotta L.A. Petricoin E.F. Proteomic applications for the early detection of cancer.Nat. Rev. Cancer. 2003; 3: 267-275Crossref PubMed Scopus (753) Google Scholar). Complicating this approach is the fact that potentially viable tumor biomarkers are embedded among a bounty of proteomic noise. This noise includes housekeeping and other highly abundant proteins, whereas the relatively low abundance of protein biomarker candidates confounds their detection. This span in protein concentration in a complex biofluid such as plasma or serum requires that effective detection methodologies bridge as many as 7–10 orders of magnitude in dynamic range to reliably detect those lowest concentration markers (6Anderson N.L. Polanski M. Pieper R. Gatlin T. Tirumalai R.S. Conrads T.P. Veenstra T.D. Adkins J.N. Pounds J.G. Fagan R. Lobley A. The human plasma proteome. A nonredundant list developed by combination of four separate sources.Mol. Cell. Proteomics. 2004; 3: 311-326Abstract Full Text Full Text PDF PubMed Scopus (746) Google Scholar). No such existing technology or platform offers such a broad dynamic range without implementing prefractionation strategies that may result in the loss or suppression of important biomarkers; also many high abundance proteins subjected to depletion act as carriers for low abundance biomarkers (7Granger J. Siddiqui J. Copeland S. Remick D. Albumin depletion of human plasma also removes low abundance proteins including the cytokines.Proteomics. 2005; 5: 4713-4718Crossref PubMed Scopus (180) Google Scholar). Here we show that the immune system-driven amplification of the autoantibody response to intracellular antigens can yield higher sensitivity, specificity, predictive value, and reproducibility in detecting low abundance biofluid tumor markers (8Brown D.M. Fisher T.L. Wei C. Frelinger J.G. Lord E.M. Tumours can act as adjuvants for humoral immunity.Immunology. 2001; 102: 486-497Crossref PubMed Scopus (70) Google Scholar). Early efforts have identified many gene products eliciting humoral response, including somatic alterations in p53 in 30–40% of affected patients that have been shown to predate cancer diagnosis (9Soussi T. p53 Antibodies in the sera of patients with various types of cancer: a review.Cancer Res. 2000; 60: 1777-1788PubMed Google Scholar). In other work, 60% of patients with lung adenocarcinomas exhibited humoral response to glycosylated annexins I and/or II, whereas none of the sera from non-cancer patients demonstrated such a response (10Brichory F.M. Misek D.E. Yim A.M. Krause M.C. Giordano T.J. Beer D.G. Hanash S.M. An immune response manifested by the common occurrence of annexins I and II autoantibodies and high circulating levels of IL-6 in lung cancer.Proc. Natl. Acad. Sci. U. S. A. 2001; 98: 9824-9829Crossref PubMed Scopus (270) Google Scholar). Similarly autoantibodies to the prostasome and to such antigens as PSA, prostatic acid phosphatase, HER-2/neu, p53, α-methylacyl-CoA racemase, and GRP78 have been observed in the sera of prostate cancer patients (11Mintz P.J. Kim J. Do K.A. Wang X. Zinner R.G. Cristofanilli M. Arap M.A. Hong W.K. Troncoso P. Logothetis C.J. Pasqualini R. Arap W. Fingerprinting the circulating repertoire of antibodies from cancer patients.Nat. Biotechnol. 2003; 21: 57-63Crossref PubMed Scopus (287) Google Scholar, 12Nilsson B.O. Carlsson L. Larsson A. Ronquist G. Autoantibodies to prostasomes as new markers for prostate cancer.Upsala J. Med. Sci. 2001; 106: 43-49Crossref PubMed Scopus (46) Google Scholar, 13Sreekumar A. Laxman B. Rhodes D.R. Bhagavathula S. Harwood J. Giacherio D. Ghosh D. Sanda M.G. Rubin M.A. Chinnaiyan A.M. Humoral immune response to α-methylacyl-CoA racemase and prostate cancer.J. Natl. Cancer Inst. 2004; 96: 834-843Crossref PubMed Scopus (121) Google Scholar, 14McNeel D.G. Nguyen L.D. Storer B.E. Vessella R. Lange P.H. Disis M.L. Antibody immunity to prostate cancer associated antigens can be detected in the serum of patients with prostate cancer.J. Urol. 2000; 164: 1825-1829Crossref PubMed Scopus (67) Google Scholar). Autoantibody signatures have also been identified using phage microarrays that can delineate prostate cancer patients from control individuals with >90% accuracy (15Wang X. Yu J. Sreekumar A. Varambally S. Shen R. Giacherio D. Mehra R. Montie J.E. Pienta K.J. Sanda M.G. Kantoff P.W. Rubin M.A. Wei J.T. Ghosh D. Chinnaiyan A.M. Autoantibody signatures in prostate cancer.N. Engl. J. Med. 2005; 353: 1224-1235Crossref PubMed Scopus (518) Google Scholar). However, one of the major caveats of this platform lies in the fact that most of the humoral targets identified are mimotopes that resemble the target protein in either the amino acid sequence or structure (15Wang X. Yu J. Sreekumar A. Varambally S. Shen R. Giacherio D. Mehra R. Montie J.E. Pienta K.J. Sanda M.G. Kantoff P.W. Rubin M.A. Wei J.T. Ghosh D. Chinnaiyan A.M. Autoantibody signatures in prostate cancer.N. Engl. J. Med. 2005; 353: 1224-1235Crossref PubMed Scopus (518) Google Scholar). Furthermore it is important to note that most of the proteins that elicit a humoral response are often differentiation antigens or antigens overexpressed or modified in cancer (13Sreekumar A. Laxman B. Rhodes D.R. Bhagavathula S. Harwood J. Giacherio D. Ghosh D. Sanda M.G. Rubin M.A. Chinnaiyan A.M. Humoral immune response to α-methylacyl-CoA racemase and prostate cancer.J. Natl. Cancer Inst. 2004; 96: 834-843Crossref PubMed Scopus (121) Google Scholar, 14McNeel D.G. Nguyen L.D. Storer B.E. Vessella R. Lange P.H. Disis M.L. Antibody immunity to prostate cancer associated antigens can be detected in the serum of patients with prostate cancer.J. Urol. 2000; 164: 1825-1829Crossref PubMed Scopus (67) Google Scholar). Additionally the humoral response elicited by cancers is heterogeneous. This is supported by studies from humoral response trials where among the large numbers of patients tested only a subset of patients with a specific tumor type develop a response to a specific antigen. This heterogeneity in humoral response generation necessitates the use of a multiplex panel of protein targets as autoantibody biomarkers to detect tumors with broad coverage for a large number of people. This requirement motivates our strategy of coupling comprehensive two-dimensional liquid-phase fractionation of the prostate cancer proteome to protein microarray analysis of patient sera and then mass spectrometry for the identification of proteins contributing to a discriminating multiplex humoral response to prostate cancer antigens (see Fig. 1). In this work we used the humoral response signature not only for prostate cancer detection but also for extensive analysis of pathways particularly dysregulated in prostate cancer development and progression. The Institutional Review Board of the University of Michigan Medical School approved this study. Serum samples from patients who visited the Urology Clinic for prostate cancer screening were collected prior to biopsy. The sera were banked at the University of Michigan Specialized Research Program in Prostate Cancer (Specialized Program of Research Excellence) tissue/serum bank. A total of 34 serum samples from patients who visited the clinic on 2 successive days were sequentially used for the experiments. Among these, 18 patients were biopsy-positive for prostate cancer (PCa), and 16 were negative for neoplasm but diagnosed with benign prostatic hyperplasia (BPH). The average age of all prostate cancer patients was 63.2 ± 12.8 years. For patients diagnosed with BPH, the average age was 64.8 ± 10.7 years. PSA values for the PCa and BPH groups were 7.81 ± 5.34 (2.9–20.4) and 6.79 ± 3.76 ng/ml (2.1–14.1), respectively. Detailed clinical and pathology data for this study are available in supplemental Table 1. Tissue samples obtained postsurgery from clinically localized prostate cancer patients (n = 5) and metastatic prostate cancer patients (n = 5) were used for two-dimensional liquid-phase fractionation as described below. Five patients were selected in each group to account for individual variations in tumor proteome. All chemicals were purchased from Sigma unless otherwise mentioned. For protein extraction, the tissue samples were resuspended in lysis buffer consisting of 7 m urea, 2 m thiourea, 100 mm DTT, 0.5% Bio-Lyte 3–10 (Bio-Rad), 2% octyl glucoside, and 1 mm PMSF. Samples were lysed at room temperature for 30 min followed by centrifugation at 35,000 rpm at 4 °C for 1 h. The supernatant was stored at −80 °C for future use. Sample preparation for chromatofocusing (CF) included a PD 10 column, equilibrated with 25 mm bis-tris in 6 m urea and 0.2% octyl glucoside, that was used to exchange the tissue lysate from the lysis buffer to the above buffer. We loaded 15 mg of tissue lysate in the first dimension as two separate pools; one was a mixture of the five primary disease samples, and the other was a mixture of the five metastatic disease samples described below. The CF experiment used a start buffer of 25 mm bis-tris with pH 7.1 and an elution buffer consisting of a 10-fold dilution of Polybuffer 96 and Polybuffer 74 in a ratio of 3:7 with pH adjusted to 4.0. Both buffers were prepared in 6 m urea and 0.2% octyl glucoside. Iminodiacetic acid was used to adjust the pH of both buffers. The PS-HPCF 1D column was equilibrated with the start buffer until the pH of the effluent was 7.1. Sample was applied to the column with multiple injections. Once a stable base line was achieved, the elution buffer was switched on to elute the proteins on the column in an isocratic mode. UV detection was performed at 280 nm, and the pH of the effluent was monitored using a flow-through on-line pH probe. The pH fractions were collected in 0.2 pH intervals, and 15 fractions in total were collected in the range of pH 7–4. The CF separation was stopped when the pH of the effluent reached 4. The flow rate in the CF experiment was 1 ml/min; and as fractionation was based on pH rather than time, fraction volumes ranged from 2 to 5 ml. Reverse-phase (RP)-HPLC was performed using PS-HPCF 2D (4.6 × 33-mm) columns. Solvent A was 0.1% TFA (J. T. Baker Inc.) in water, and solvent B was 0.1% TFA in acetonitrile (Burdick and Jackson, Muskegon, MI). The gradient was run from 5 to 15% in 1 min, 15% B to 25% in 2 min, 25 to 31% in 2 min, 31 to 41% in 10 min, 41 to 47% in 6 min, 47 to 67% in 4 min, finally up to 100% B in 3 min and held for another 1 min, and then back to 5% in 1 min at a flow rate of 1 ml/min. The column temperature was 40 °C higher than the ambient temperature. UV absorptions were monitored at 214 nm. RP fractions were collected with fraction volumes ranging between 100 μl and 1 ml. The fractions were dried completely by SpeedVac at 75 °C and stored at −80 °C. The fractionated proteins were resuspended in 15 μl of buffer containing PBS at pH 7.4 and protease inhibitors (Roche Applied Science) at an average protein concentration of ∼10 pg/well. The samples were transferred to a 96-well microtiter plate (MJ Research, Waltham, MA) and printed on nitrocellulose slides (Schleicher & Schuell) using a GeSim Nanoplotter2, a non-contact ink jet printer. Each spot measured ∼300 μm in diameter; spots were placed 1200 μm apart. The slides were dried for 1 h at room temperature and were either used immediately or stored at room temperature in a desiccation chamber. Nitrocelluose slides containing spotted proteins were hydrated in PBS for 10 min and blocked in PBS containing 1% BSA (Sigma) and 0.1% Tween 20 (Sigma) overnight at 4 °C. The slides were then incubated with serum (1:400) either from prostate cancer patients or from patients with BPH in probe buffer (PBS, pH 7.4, containing 1% BSA, 5 mm MgCl2, 0.5 mm DTT, 0.05% Triton X-100, and 5% glycerol) for 2 h at 4 °C. Slides were washed six times with probe buffer, each time for 5 min. They were then incubated with Alexa 647-conjugated anti-human IgG (1:2000, Invitrogen) for 1 h at 4 °C, washed with probe buffer as above, dried, and analyzed. Primary analysis, including scanning and quantification of slides, was executed with the GenePix 4000B scanner (Axon Laboratories, Inc., Foster City, CA); gridding was completed according to the manufacturer's instructions. The single red channel intensity values were calculated for each individual fraction spot. An initial round of spot checking was performed using GenePix default parameters. This was followed by a second round of curation wherein spots having any of the following characteristics were manually flagged: a diameter smaller than 300 μm, an irregular outline, localization in regions of high local background, or presence of any areas of the arrays with obvious defects. Flagged spots were seeded to −1 in raw intensity units in the subsequent analysis. The median minus background of the red channel was extracted from each array and normalized. The total feature set was filtered for dominantly negative fractions, and only those fractions with non-negative raw intensity in ≥50% of samples in the cohort were retained. Within-array standardization entailed median centering and scaling by their respective median absolute deviations. Quantile normalization was then executed to ensure the same empirical distribution across all arrays. Two-way average-linkage hierarchical clustering of an uncentered Pearson correlation similarity matrix was executed and figures were generated using Cluster and TreeView (16Eisen M.B. Spellman P.T. Brown P.O. Botstein D. Cluster analysis and display of genome-wide expression patterns.Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14863-14868Crossref PubMed Scopus (13079) Google Scholar). A supervised analysis was completed to coalesce around a subset of fractions from the 2016-element (excluding control features) humoral response arrays most predictive for class distinction across the serum samples. Array data, normalized as described above, were applied to a test statistic-based feature selection procedure calculating the F-statistic between the 18 biopsy-proven PCa and 16 BPH samples across all 2016 fractions. Different counts of the best ranking fractions by F-statistic (every count of best fractions from 5 to 100) were used to build a support vector machine (SVM) prediction model. The SVM over multiple kernel test permutations was embedded in a finite grid search of paired values of exponentially growing sequences of cost and γ. A linear kernel produced the best prediction, whose accuracy and error were calculated using leave-one-out cross-validation (LOOCV) to evaluate the performance of the models. The top ranked 20 fractions were ultimately selected from the fraction predictor based on their best performance in specificity and sensitivity and on the highest stability of recurrence at top ranks over all left-out iterations. These 20 fractions were additionally tolerant to repeated testing and small changes to model parameters and harbored a reproducible humoral response signal across multiple replicates of the serum cohort (supplemental methods and supplemental Fig. 1). All statistical analyses were performed in R 2.3.0 and SPSS. A targeted UV peak in the second dimension RP-HPLC chromatogram was collected and aliquoted into two fractions. The protein content of fractions identified by best classification performance as well as 27 fractions demonstrating no differential response between classes was digested using porcine trypsin (1:50, Promega, Madison, WI) in 1 m ammonium bicarbonate, pH 9. The digestion was performed for 16 h at 37 °C. At the end of 24 h, the trypsin activity was stopped using 3% formic acid. The peptide mixtures were separated by reverse-phase chromatography using a 0.075 × 150-mm C18 column attached to a Paradigm HPLC pump (Michrom BioResources Inc., Auburn, CA). Peptides were eluted using a 45-min gradient from 5 to 95% B (0.1% formic acid, 95% acetonitrile) where solvent A was 0.1% formic acid, 2% acetonitrile. A Finnigan LTQ mass spectrometer (Thermo Fisher Corp., Waltham, MA) was used to acquire spectra; the instrument was operated in data-dependent mode with dynamic exclusion enabled. The MS/MS spectra for the three most abundant peptide ions in full MS scan were obtained. Peak lists were generated using Bioworks 3.1 (Thermo Fisher Corp.) running TurboSequest version 27 (revision 12). The peak lists were generated using the default parameters of the software. These were then searched using the Mascot algorithm (version 1, Matrix Sciences, Boston, MA) against the composite National Center for Biotechnology Information (NCBI) human RefSeq database (downloaded on June 28, 2005, release 11 containing 27,960 proteins). The search was done using a mass tolerance of 2 Da for the precursor fragment and 0.5 Da for the daughter fragments. Furthermore all searches were performed using "no enzyme criteria," monoisotopic peptide mass, and Met + 16 as the variable modification. Confidence in peptide assignment accuracy and protein identifications was assigned via the open source Trans-Proteomic Pipeline (Institute for Systems Biology) implementing PeptideProphet, which validates peptides assigned to MS/MS spectra and subsequently combined to derive protein probabilities using ProteinProphet (version 4) (17Keller A. Nesvizhskii A.I. Kolker E. Aebersold R. Empirical statistical model to estimate the accuracy of peptide identifications made by MS/MS and database search.Anal. Chem. 2002; 74: 5383-5392Crossref PubMed Scopus (3797) Google Scholar, 18Keller A. Eng J. Zhang N. Li X.-j. Aebersold R. A uniform proteomics MS/MS analysis platform utilizing open XML file formats.Mol. Syst. Biol. 2005; 1: E1-E8Crossref Scopus (589) Google Scholar, 19Nesvizhskii A.I. Keller A. Kolker E. Aebersold R. A statistical model for identifying proteins by tandem mass spectrometry.Anal. Chem. 2003; 75: 4646-4658Crossref PubMed Scopus (3529) Google Scholar). The resulting protein lists were filtered using a ProteinProphet probability of 0.90 or higher, which corresponds to an error rate of less than 1% as estimated by ProteinProphet. Proteins identified by a single peptide with charge state of +1 were removed because of reduced confidence in these identifications derived from an ion trap mass spectrometer. Keratins were manually removed from the filtered list of proteins. All validated identifications, peptide sequences, and annotations were stored in a relational database for downstream analysis. The statistical model for testing and storing the results of associations between independent molecular concepts is as described previously (20Tomlins S.A. Mehra R. Rhodes D.R. Cao X. Wang L. Dhanasekaran S.M. Kalyana-Sundaram S. Wei J.T. Rubin M.A. Pienta K.J. Shah R.B. Chinnaiyan A.M. Integrative molecular concept modeling of prostate cancer progression.Nat. Genet. 2007; 39: 41-51Crossref PubMed Scopus (715) Google Scholar). Protein identifications from the humoral response signature were converted to Entrez Gene identifiers and batch-loaded to MCM (Oncomine) for analysis. As a control during concept enrichment, we similarly analyzed 14 of 27 fractions selected as a non-differential negative control that were sequenced as described above, and their protein content was culled with identical criteria. Any concept enriching both the differential predictor and negative control was excluded from the analysis. We extended the analysis one level with concept-to-concept enrichments for five promoter binding site concepts (see Fig. 3A, boxed). Each seeded a subset MCM enrichment on its gene list rather than on the original humoral target list. The five resulting enrichment networks were sequentially merged into a single common network. Orphaned concepts from single concept enrichment were removed. To generate an immune program under STAT control for meta-analysis with public gene expression studies, we downloaded the immune response in silico repository of 1622 genes expressed in, and classified by, multiple immune cell lineages (21Abbas A.R. Baldwin D. Ma Y. Ouyang W. Gurney A. Martin F. Fong S. van Lookeren Campagne M. Godowski P. Williams P.M. Chan A.C. Clark H.F. Immune response in silico (IRIS): immune-specific genes identified from a compendium of microarray expression data.Genes Immun. 2005; 6: 319-331Crossref PubMed Scopus (273) Google Scholar). There is a 179-gene overlap between the immune response in silico compendium and those genes under STAT control, the union of either STATx, STAT1, STAT3, or STAT5B (homodimer). This seeds the metamap analysis described in the text. An overview of the approach we took in identifying humoral targets in prostate cancer is depicted in Fig. 1. To develop a protein microarray for prostate cancer progression, we independently fractionated proteins from clinically localized and metastatic cancer tissues (n = 5, each) in two dimensions using a combination of chromatofocusing and reverse-phase chromatography. The fractionated proteins were spotted on nitrocellulose-backed glass slides and served as bait to capture potential autoantibodies found in serum. This process included pooling samples from each group separately and loading and printing separately on the same slide. This was done to maintain both a localized cancer-specific and metastatic cancer-specific proteomic signature as they may individually produce two different antigen populations, one to each signature. Thus, we believed humoral response could either target antigens from early or aggressive disease, so we maintained distinct populations. Proteins that reacted with prostate cancer sera but not the control were identified using an ion trap mass spectrometer, database search, and downstream protein informatics (see "Experimental Procedures"). The list of proteins obtained was used both to characterize the predictor and to conduct a "molecular concept" analysis for their involvement in disease processes (supplemental Fig. 2 and supplemental Tables 4–7). Approximately 2300 fractions were used to generate protein microarrays. Using this 2300-feature protein microarray, we evaluated sera from prostate cancer patients and controls. In this discovery approach we evaluated 34 serum samples consisting of 18 sera from prostate cancer patients (biopsy-positive, high PSA) and 16 from individuals with BPH (biopsy-negative for cancer, high PSA). Critically these samples constitute the clinically challenging distinction between cancer-negative (benign hyperplastic condition) and cancer-positive needle biopsy findings in the setting of elevated levels of circulating prostate-specific antigen. The 34 serum specimens were collected from consecutive patients over 2 different days in the Urology Clinic at the University of Michigan prior to prostate biopsy. We sought the pattern of differential autoantibody response that could discriminate between the benign and prostate cancer groups that was generated against specific tumor antigens represented by the cancer proteome spotted on the microarrays. Cross-validated supervised analysis implementing the non-parametric SVM was performed using the 34 samples as a training set, looking for humoral response correlates of the two-class distinction between BPH and PCa (see "Experimental Procedures"). Of the 1522 features remaining after filtering for dominantly negative fractions as the result of hybridization (see "Experimental Procedures"), a subset demonstrated differential reactivity patterns. Embedded feature reselection during LOOCV produced a 20-fraction predictor having 75% specificity (four of 16 BPH samples were misclassified) and 78% sensitivity (four of 18 prostate-cancer samples were misclassified) in d
Referência(s)