BACKGROUND: Improved tests are needed for detection and management of prostate cancer. We hypothesized that differential gene expression in prostate tissue could help identify candidate blood biomarkers for prostate cancer and that blood from men with advanced prostate disease could be used to verify the biomarkers presence in circulation.
METHODS: We identified candidate markers using mRNA expression patterns from laser-capture microdissected prostate tissue and confirmed tissue expression using immunohistochemistry (IHC) for the subset of candidates having commercial antisera. We analyzed tissue extracts with tandem mass spectrometry (MS/MS) and measured blood concentrations using immunoassays and MS/MS of trypsin-digested, immunoextracted peptides.
RESULTS: We selected 35 novel candidate prostate adenocarcinoma biomarkers. For all 13 markers having commercial antisera for IHC, tissue expression was confirmed; 6 showed statistical discrimination between nondiseased and malignant tissue, and only 5 were detected in tissue extracts by MS/MS. Sixteen of the 35 candidate markers were successfully assayed in blood. Four of 8 biomarkers measured by ELISA and 3 of 10 measured by targeted MS showed statistically significant increases in blood concentrations of advanced prostate cancer cases, compared with controls.
CONCLUSIONS: Seven novel biomarkers identified by gene expression profiles in prostate tissue were shown to have statistically significant increased concentrations in blood from men with advanced prostate adenocarcinoma compared with controls: apolipoprotein C1, asporin, cartilage oligomeric matrix protein, chemokine (C-X-C motif) ligand 11 (CXCL11), CXCL9, coagulation factor V, and proprotein convertase subtilisin/kexin 6.
Prostate cancer is the second leading cause of cancer-related death in American men (1), with >90% of this cancer consisting of adenocarcinoma subtype (2). Typically, prostate adenocarcinoma undergoes slow growth; however, in a subset of men the cancer can rapidly develop, leading to multiple health problems and ultimately death. Autopsy studies have shown that many older men who died of other causes also had undetected prostate cancer (3). Roughly a million men undergo prostate needle biopsy each year [often performed on the basis of on increased blood prostate specific antigen (PSA)4], but malignancy is detected in only a fifth of these men, of whom a tenth die of prostate disease (4, 5). Better biomarkers to supplement PSA could help in diagnosis and management of prostate cancer.
The discovery of biomarkers in blood that are sensitive and specific for prostate cancer is difficult owing to the large number of proteins and protein metabolic products with potentially very low concentrations. On the other hand, measurements of expressed mRNA from prostate tissue extracts are relatively robust with the use of commercial arrays and gene amplification techniques. Also, bioinformatics tools can provide guidance on which genes encode for proteins likely to be found in blood. Analysis of the differential expression of genes in normal or benign prostatic tissue specimens compared to the gene expressions in the specimens with various forms of prostate caner provides a potential strategy for identifying candidate protein biomarkers. That is, the gene expression data could provide a treasure map to help look for specific candidate biomarkers in blood. We used this strategy in this project. Three techniques—immunohistochemistry (IHC), immunoassay, and tandem mass spectrometry (MS/MS)—were used to look for these proteins in tissue and blood. We identified 70 candidate biomarkers on the basis of mRNA changes and pursued 35 candidates that had no prior proven utility in prostate cancer. Some of these proteins had been previously investigated, so antisera were available; others had none. We used commercial antisera and immunoassay reagents when they were available. For other markers, we developed MS/MS reagents and techniques.
Validation of potential biomarkers in blood is difficult. Blood contains proteins, protein precursor forms, and protein metabolites from many tissue sources. Also, the dilution of proteins from a small organ into a large blood volume, combined with numerous blood clearance mechanisms, make many blood concentrations very low. Validation of protein biomarkers in tissue is easier because of tissue specificity and relatively higher concentrations and so is a potential intermediary step.
Two potential clinical roles for prostate biomarkers are early detection of cancer and identification of aggressive forms of cancer. For early detection, very sensitive assay systems are needed to measure the low biomarker concentrations generated by the small tumor masses. The identification of aggressive forms of cancer requires longitudinal follow-up with stable preserved blood samples. Because blood assays are not available that can measure very low concentrations of many of these proteins, and long-term stable blood samples are not available, it was impractical to fully evaluate the utility of these candidate biomarkers. Therefore we elected to determine if these markers could be measured in blood samples from men with advanced prostate cancer and whether those concentrations were different from the concentrations in blood from men with benign prostate disease and/or early cancer.
Methods and Materials
The study design is illustrated in Fig. 1. For hypothesis generation and selection of an initial candidate biomarker list, we used gene expression data generated in a discovery project funded by the Minnesota Partnership (6, 7). Briefly, from consenting patients, we procured frozen prostate tissue samples from the Mayo Clinic Prostate SPORE tissue bank. Tumor cells or adjacent nonneoplastic prostate epithelial cells from 100 patients with prostate cancer were extracted by laser capture microdissection, and mRNA was extracted, amplified, and measured on Affymetrix U133 Plus 2.0 microarray chips. We selected biomarker candidates by comparing expression profiles and filtered this candidate list to identify 70 genes with prostate tissue–specific expression, which encoded proteins localized to cellular membrane or extracellular space (see Supplemental File 1-A, which accompanies the online version of this article at http://www.clinchem.org/content/vol58/issue3). From this list, we selected 35 novel candidate biomarkers that had no prior literature or patent claims for utility in prostate disease. We performed intermediary studies to evaluate protein expression in prostate tissue using IHC and MS/MS. Where available, ELISA assays were used to measure blood concentrations of candidate markers. We developed MS/MS methods to measure blood concentrations of candidate markers that possessed favorable characteristics. Before MS/MS measurement, the blood was depleted of high-abundance proteins and glycated proteins. The depleted samples then were trypsin digested, and the target peptides were extracted with rabbit antipeptide polyclonal antibodies.
IN SILICO CANDIDATE MARKER SELECTION
We selected the initial candidate markers for differential tumor-to-normal expression or high prostate tissue–specific expression profiles. We defined tissue-specific expression using a numeric score derived from existing cross-tissue expression data sets (8). The candidates were filtered to identify markers encoding proteins localized to the extracellular space (secreted) or cellular membrane. Localization was defined by UniProt annotations (9). For candidate proteins lacking annotation, we predicted the cellular localization using an integration of SignalP, TargetP, and TMHMM (transmembrane prediction using hidden Markov models) (10). Candidate biomarkers were further filtered to select for protein targets with a high propensity for assay development by use of a multistep bioinformatic workflow (11). Briefly, the protein sequences were computationally digested with PeptideCutter (12), and tryptic peptide sequences were compared to the human proteome with BLAST (blastp) (13) to ascertain parent protein specificity. Posttranslational glycosylation and phosphorylation sites were extracted for each candidate marker from the UniProt database. This information was augmented with NetNGlyc 1.0, NetOGlyc 3.1 (14), and NetPhos 2.0 predictions (15). We determined peptide immunogenic potential on the basis of hydrophilicity, surface probability, and flexibility, using the GCG peptide structure prediction algorithm (16).
EVALUATION OF PROTEIN EXPRESSION IN TISSUE
MS/MS of tissue extracts.
We obtained fresh-frozen prostate adenocarcinoma and adjacent nondiseased tissue from 2 men with high Gleason score prostate cancer: patient no. 81 (PSA 3.5 μg/L, Gleason 9, T3aN0+) and no. 143 (PSA 7.1 μg/L, Gleason 7, T3aN0−). The main focus of these tissue extract measurements was to evaluate if the gene expressions were associated with protein expressions. Only 2 patients were evaluated because of the limited sensitivity of MS/MS to detect low concentrations even with prior fractionation. Tissue was pulverized while frozen and homogenized with a Polytron® in 1 mol/L HEPES buffer, pH 7.3, with protease inhibitors. The samples were electrophoresed on 10%–14% SDS-PAGE gels, stained with Coomassie blue, and sliced horizontally into strips. The expected biomarkers for each gel band are illustrated in Fig. 2. The gel bands were destained with 50 mmol/L Tris (pH 8.1)/50% acetonitrile, reduced with 20 mmol/L dithiothreitol/50 mmol/L Tris (pH 8.1) at 55 °C for 40 min, and alkylated with 40 mmol/L iodoacetamide at room temperature for 40 min in the dark. The proteins were digested in situ with 30 μL (0.004 μg/μL) trypsin (Promega) in 20 mmol/LTris (pH 8.1)/0.0002% Zwittergent 3-16, at 37 °C overnight, followed by peptide extraction with 60 μL of 2% trifluoroacetic acid, then 60 μL of acetonitrile. The pooled extracts were concentrated to <5 μL on a SpeedVac spinning concentrator (Savant Instruments) and then brought up in 0.1% formic acid. We used MS/MS with nanoflow liquid chromatography electrospray on a ThermoFinnigan LTQ Orbitrap Hybrid Mass Spectrometer (ThermoElectron) to look for the targeted peptides for each gel band.
We used paraffin-embedded, formalin-fixed tissue sections from 20 men with prostate cancer (17) to evaluate protein expression for 13 candidate biomarkers that had commercially available IHC antisera. For each case, we analyzed matched nondiseased and malignant tissue sections. Slides were placed in a preheated 1 mmol/L EDTA, pH 8.0, retrieval buffer for 30 min and then cooled. After the heat-activated epitope retrieval step, slides were placed on the Dako Autostainer and incubated with the candidate biomarker specific antibodies (see online Supplemental File 1-B). Staining intensity of the prostate tumor and matched nondiseased prostate tissue were scored with an ordinal scale of 0–3, with 0 representing no staining and 3 representing heavy staining. For each antiserum, we counted the number of cases in which the tumor stain intensity score exceeded the nondiseased stain intensity score, or vice versa (Table 1).
We used commercial ELISAs to measure the blood concentrations of 8 candidate biomarkers. We obtained frozen aliquots (−70 °C) of serum and EDTA samples from 50 men with advanced prostate cancer and 26 men with recent prostate biopsies from the Mayo Prostate SPORE tissue bank. Many of the men in the advanced cancer series subsequently died of prostate cancer. Blood samples from men undergoing prostate biopsy, matched by PSA concentrations, were used as controls. These controls included 13 biopsies showing no cancer and 13 biopsies showing early prostate cancer. The analytic performances of the kits were verified, and controls with known target ranges were assayed on every plate. The reagents are listed in online Supplemental File 1-C.
MS/MS BLOOD MEASUREMENTS
We used a modified version of the SISCAPA (stable isotope standards with capture by antipeptide antibodies) methodology published by Anderson and colleagues (18, 19). Serum specimens were trypsin digested, and targeted peptides were immunoextracted and quantitated by spectrometric multiple-reaction monitoring (18, 19). We used an API 5000 triple quadripole mass spectrometer (Applied Biosystems) with a higher injection volume (100 uL) than that of the published SISCAPA. Like Anderson et al., we used polyclonal rabbit antipeptide antisera; however, our antibodies did not work well with whole serum, and therefore we incorporated protein depletion before digestion. We multiplexed 5–7 antisera together to extract 1.0 mL serum and did not reuse our antipeptide extraction beads. The elution fraction from multiplexed immunoaffinity beads was evaporated on SpeedVac to approximately 100 μL, and a mixture of internal standards was added to all samples just before multiple-reaction monitoring analysis. We identified 4–5 transitions for each double-charge parent ion but chose only the optimal 2–3 transitions for monitoring to have adequate resolution to simultaneously measure the 5–7 multiplexed peptides. Table 2 shows the parameters for each endogenous peptide and the corresponding isotopically labeled peptide internal standard.
SERUM EXTRACTION AND TRYPSIN DIGESTION OF BIOMARKERS BEFORE MS/MS
To improve detection of peptides, all serum samples were either lectin-extracted or depleted of high-abundance proteins followed by trypsin digestion and immunoextraction with antipeptide beads (Fig. 1). For the first protocol, 1.0 mL sera was passed over a concanavalin A lectin column (Sigma #C7555) to separate human N-glycosylated proteins from non–N-glycosylated proteins. Glycosylated proteins were eluted from the column with a gradient of methyl α-d-mannopyranoside in 20 mmol/L Tris, 0.5 mol/L NaCl, 1 mmol/L MgCl2, 1 mmol/L MnCl2, and 1 mmol/L CaCl2, pH 7.5, trypsin digested, and serially extracted with 2 antipeptide bead sets (A and B). The second protocol depleted albumin and 13 other high-abundance proteins from a second set of 1.0-mL serum samples with MARS14 columns (Agilent Technologies). After trypsin digestion, peptides were extracted with the multiplexed antipeptide bead sets. For trypsin digestion, we added urea to obtain a final concentration of 6 mol/L. The samples were incubated with 10 mmol/L of dithiothreitol at 37 °C for 1 h followed by alkylation with 30 mmol/L iodoacetamide at room temperature for 1 h in the dark. Samples were diluted with 25 mmol/L NH4HCO3 with 0.001% Zwittergent 3-16 to 1 mol/L urea. Proteins were digested with 0.5 mg trypsin (Sigma) at 37 °C overnight at a 1:10 ratio, and the reaction was stopped by lima bean inhibitor of trypsin (Worthington).
The production of antipeptide polyclonal immuno-affinity magnetic particles is outlined in online Supplemental File 1-D. The antipeptide bead sets were designed to provide good MS/MS separation of the peptides included in each elution. Bead sets A and B each contained multiplexed antisera for 5 peptides corresponding to glycosylated proteins (Table 3). Bead set C contained 7 multiplexed antisera, including 2 antisera contained in bead set B [oxoglutarate dehydrogenase-like (OGDHL), proprotein convertase subtilisin/kexin 6 (PCSK6)]. Duplication of peptides between columns reflected uncertainty regarding the parent protein glycosylation status. Trypsin-digested samples were added to the antibody-conjugated beads and gently rocked for 4 h at room temperature. The supernatant and the first 500 μL of a PBS wash were collected and either exposed to a second bead set or stored at −80 °C for future use. Before use, each multiplexed antipeptide bead set was washed twice with the PBS buffer. The peptides were eluted with 5% acetic acid with 0.001% Zwittergen 3-16. Fresh bead sets were used for each patient sample processed.
We assessed the significance of change in IHC staining intensity between tumor and nondiseased tissue with a 1-tail sign test, excluding equivalent scoring samples, with a P value threshold of 0.01. We quantitated MS results for the tissue specimens by counting unique peptides identified for each candidate marker and analyzed the ELISA and MS sera data for group discrimination with a rank sum test and Satterthwaite t-test, with P value thresholds of 0.05. To ascertain potential marker discrimination in a subset of advanced prostate cancer sera samples, we also characterized candidate markers by the number of tumor samples with concentration values exceeding the maximum value from the control samples. To minimize false positives, only those biomarker concentrations that exceeded the highest concentration measured in the 26 control samples were considered positive on the cross-plots.
The initial selection process identified 70 candidate serum biomarkers on the basis of our bioinformatics selection criteria (see online Supplemental File 1). This list included known prostate cancer biomarkers such as PSA, α-methylacyl-CoA racemase (AMACR), and kallikrein-related pepsidase 2 (KLK2). Candidate markers were screened against the literature to identify 35 novel candidates for which utility as a prostate cancer marker was previously unreported (Table 3). Twenty-six of these candidate markers were classified as extracellular and the remaining 9 as cellular membrane proteins. The predicted molecular weights ranged from 9.3 to 251.7 kDa.
Mass spectrometry analysis of tissue extracts identified only 5 of the candidate biomarkers in the samples analyzed: apolipoprotein C1 (APOC1, gel 8), asporin (ASPN, gel 5), complement C4A (gel 8), 6-phosphogluconolactonase (PGLS, gel 7), and ribosomal protein L22–like 1 (RPL22L1, gel 8). ASPN, RPL22L1, and PGLS demonstrated some tumor-specific discrimination.
IHC analysis of prostate tumor and nondiseased tissue sections revealed that all 13 assayed proteins were expressed in at least 1 of the prostate cancer samples. Table 2 shows the assigned staining code for each sample. The aggregate counts at the bottom of each IHC column report the number of samples in which the staining code was higher or lower for the cancer tissue compared to adjacent nondiseased tissue. No marker demonstrated statistically significant differences between Gleason scores. Using a 1-tail sign test with P value <0.01, 6 markers showed statistically significant increases in staining codes for the prostate cancer tissue compared to the matched control tissue [collagen type 2, α1 (COL2A1), cartilage oligomeric matrix protein (COMP), chemokine (C-X-C motif) ligand 14 (CXCL14), ephrin A4 (EFNA4), G protein–coupled receptor 116 (GPR116), and PCSK6].
The 8 ELISA assays identified measurable concentrations of the biomarkers in a majority of samples analyzed. Statistically significant changes in 4 candidate marker concentrations between advanced prostate cancer sera and control sera were found with a rank sum test with P value <0.05 (APOC1, CXCL11, CXCL9, and COMP). Statistical significance was confirmed with a Satterthwaite t-test for APOC1, CXCL11, and CXCL9. When selecting candidate markers on the basis of >10% of advanced prostate cancer sample values exceeding the maximum value of the control samples, we found the same 3 markers identified by the Satterthwaite t-test.
For the blood-based MS/MS assays, 39 peptides were synthesized, representing 24 of the protein candidate biomarkers. The remaining 11 biomarkers were not targeted by MS/MS analysis because of difficulties synthesizing peptides or the availability of ELISA assays (Table 3). Antisera were successfully developed for 38 of 39 peptides, and 25 antibodies worked for immunoextraction. Eleven of the captured peptides resulted in weak or undetectable signal on MS/MS. Ultimately, MS/MS assays that could detect peptides in blood were obtained for 10 biomarkers.
Three multiplexed sets of MS/MS assays were run with the same sample set (from men with advanced prostate cancer and associated controls) that was used for the ELISA assays. Based on rank sum test (P < 0.05), 3 markers discriminated the advanced prostate cancer samples from associated controls. ASPN and coagulation factor V (F5) were assayed on multiplexed bead set B and PCSK6 on multiplexed bead set C. Owing to their known heterogeneity in cancer, the candidate markers were also characterized for potential signal in a subset of advanced prostate cancer patients by identifying the number of samples in which the advanced prostate cancer sera concentration values exceeded the highest control value. Increased values were identified for 3 markers in bead set A [APOF in 2/50, cadherin 7 (CDH7) in 9/50, and COL2A1 in 5/50], 4 markers in bead set B [ASPN in 26/50, β-1,3-acetylglucosaminyltransferase (B3GNT) in 1/50, F5 in 25/50, and PCSK6 in 8/50], and 3 markers in bead set C (PCSK6 in 4/50, PGLS in 7/50, and RPL22L1 in 11/50) (Fig. 3). Results are summarized in Table 3.
Conclusions and Discussion
We identified 7 novel biomarkers as statistically significant in discriminating protein concentrations in blood from men with advanced prostate cancer compared to controls: APOC1, ASPN, COMP, CXCL11, CXCL9, F5, and PCSK6. Four additional novel biomarkers were identified with increased serum concentration values in at least 10% of advanced prostate cancer samples compared to controls: CDH7, COL2A1, PGLS, and RPL22L1. These blood proteins have not been previously advocated as markers of prostate cancer. The increased expression of mRNA in prostate tissues and the biochemical characteristic predicting their presence in blood were the guiding principles for the targeted search. Although no single candidate biomarker was found in all men with advanced prostate cancer, measurements of combinations of these markers may have good clinical utility. However, verification of multivariate panels of biomarkers must await further studies that require larger numbers of patients.
Several studies have identified potential prostate cancer biomarkers on the basis of differential gene expression in tissue or cell culture (20,–,22). There is surprisingly little agreement across these studies. Part of this lack of concordance may be related to coregulation of genes causing high correlations in gene expression, where one study may select one of the correlated genes and another study may select alternate genes. The National Cancer Institute study identified and annotated 91 molecular markers with potential utility for prostate cancer, including PSA and KLK2 (20). None of our 35 novel potential biomarkers were included on this list.
Bai et al. (21) identified 37 overexpressed and 7 downregulated genes for prostate cancer based on differential expression of mRNA transcripts. None of their gene products matched our list; only 13 of their sequences had reported identities in GenBank, and the others were “expressed sequence tags.” Sardana et al. (22) identified 65 extracellular candidate tumor markers in PC3 cell culture media. Only 1 marker, COL2A1, was also on our list; however, 2 other markers, CXCL3 and C3, belonged to the same protein family as entries on our list. Yang et al. (23) used integrative genomic mining for discovery of potential blood-borne cancer biomarkers. None of their 33 markers for prostate, breast, or lung matched our list, and only 1 (CXCL9) of their larger list of 178 general markers matched our list. None of the 27 novel prostate cancer markers reported by the Genome Sciences Centre nor any of the 16 markers reported by Goulart's group were on our list (24, 25).
F5 was increased in many of our advanced prostate cancer samples. Increases in other coagulation factors (prothrombin fragments and fibrous degradation products) have been similarly reported in advanced colon cancer (26). Additionally, prostate tissue has been reported to be a rich reservoir of thrombin (27). These observations are consistent with the findings reported here and suggest that coagulation markers may be important biomarkers for prostate cancer.
It is interesting that the chemokines CXCL9 and CXCL11 were found to be overexpressed in our laser-microdissected tissues and showed discrimination for advanced prostate cancer. Localization of these markers to the prostate is expected, because chemokines are known to be constitutively expressed in epithelial cells of the male urogenital track and present in seminal fluid (28). Also, with Ewing sarcoma, CXCL9 was upregulated and was associated with tumor progression (29). Hu et al. (30) showed with the matrigel invasion assay that both CXCL12 and CXCL16 induced invasion with PC3 cell lines, but they did not report investigating the effects of CXCL9, CXCL1, or CXCL14.
The preferred samples for evaluating the utility of blood markers for discriminating aggressive prostate cancer from slow-growing prostate cancer would be blood collected at the time of diagnosis. The samples used in this study were collected many years after the diagnosis because earlier samples were not archived. Even if samples were originally saved, stability may be an issue after long storage. The candidate markers identified in this study will need further investigation to determine if they have early prognostic value. Also, further study will be needed to confirm if multivariate combinations of these markers have additional utility.
↵4 Nonstandard abbreviations:
- prostate specific antigen;
- tandem mass spectrometry;
- transmembrane prediction using hidden Markov models;
- stable isotope standards with capture by antipeptide antibodies;
- protocadherin β cluster 10;
- oxoglutarate dehydrogenase-like;
- proprotein convertase subtilisin/kexin 6;
- α-methylacyl-CoA racemase;
- apolipoprotein C1;
- coagulation factor V;
- ribosomal protein L22–like 1;
- collagen type 2, α1;
- cartilage oligomeric matrix protein;
- chemokine (C-X-C motif) ligand 14;
- ephrin A4;
- G protein–coupled receptor 116;
- kallikrein-related peptidase 2;
- cadherin 7;
Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval of the published article.
Authors' Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the Disclosures of Potential Conflict of Interest form. Potential conflicts of interest:
Employment or Leadership: None declared.
Consultant or Advisory Role: None declared.
Stock Ownership: None declared.
Honoraria: None declared.
Research Funding: E.W. Klee, NIST (MSE grant 60NANB10D00S), Minnesota Partnership for Biotechnology and Medical Genomics, NIH Mayo Clinic Prostate SPORE grant P50CA91956; O.P. Pondar, Minnesota Partnership for Biotechnology and Medical Genomics; M.K. Goodmanson, Minnesota Partnership for Biotechnology and Medical Genomics, NIH Mayo Clinic Prostate SPORE grant P50CA91956; R.B. Dyer, Minnesota Partnership for Biotechnology and Medical Genomics; E.J. Bergstrahl, NIH Mayo Clinic Prostate SPORE grant P50CA91956; H.R. Bergen III, NIH; T.J. Sebo, NIH Mayo Clinic Prostate SPORE grant P50CA91956; G.G. Klee, Minnesota Partnership for Biotechnology and Medical Genomics, NIH Mayo Clinic Prostate SPORE grant P50CA91956.
Expert Testimony: None declared.
Role of Sponsor: The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, or preparation or approval of manuscript.
- Received for publication July 28, 2011.
- Accepted for publication November 28, 2011.
- © 2012 The American Association for Clinical Chemistry