Autoantibody Signatures in Prostate Cancer
http://www.100md.com
《新英格兰医药杂志》
ABSTRACT
Background New biomarkers, such as autoantibody signatures, may improve the early detection of prostate cancer.
Methods With a phage-display library derived from prostate-cancer tissue, we developed and used phage protein microarrays to analyze serum samples from 119 patients with prostate cancer and 138 controls, with the samples equally divided into training and validation sets. A phage-peptide detector that was constructed from the training set was evaluated on an independent validation set of 128 serum samples (60 from patients with prostate cancer and 68 from controls).
Results A 22-phage-peptide detector had 88.2 percent specificity (95 percent confidence interval, 0.78 to 0.95) and 81.6 percent sensitivity (95 percent confidence interval, 0.70 to 0.90) in discriminating between the group with prostate cancer and the control group. This panel of peptides performed better than did prostate-specific antigen (PSA) in distinguishing between the group with prostate cancer and the control group (area under the curve for the autoantibody signature, 0.93; 95 percent confidence interval, 0.88 to 0.97; area under the curve for PSA, 0.80; 95 percent confidence interval, 0.71 to 0.88). Logistic-regression analysis revealed that the phage-peptide panel provided additional discriminative power over PSA (P<0.001). Among the 22 phage peptides used as a detector, 4 were derived from in-frame, named coding sequences. The remaining phage peptides were generated from untranslated sequences.
Conclusions Autoantibodies against peptides derived from prostate-cancer tissue could be used as the basis for a screening test for prostate cancer.
Limitations of the prostate-specific antigen (PSA) test for the early detection of prostate cancer1 indicate the need for other means of screening for this neoplasm. The finding that patients with cancer produce autoantibodies against antigens in their tumors2,3,4,5,6,7 suggests that such autoantibodies could have diagnostic and prognostic value.4,5,8,9,10 For example, mutant forms of the p53 protein elicit anti-p53 antibodies in 30 to 40 percent of patients with various types of cancers.11 Recently, we found that patients with prostate cancer produce antibodies against -methylacyl-coenzyme A racemase,12 an overexpressed protein in epithelial cells in prostate cancer.13,14,15 This autoantibody had 72 percent specificity and 62 percent sensitivity in detecting prostate cancer.12 The use of additional prostate-cancer antigens could improve the sensitivity and specificity of an autoantibody-based screening test for prostate cancer.
Here we report the use of a technique that combines phage-display technology with protein microarrays to identify and characterize new autoantibody-binding peptides derived from prostate-cancer tissue. A similar approach has been used to identify selected antigens for the diagnosis of breast cancer.16 This emerging area of research, termed "cancer immunomics," allows a global analysis of the autoantibodies against antigens in a neoplasm (see the Glossary for definitions of terms).
Glossary
Methods
Populations and Samples
This study, which was approved by the institutional review board of the University of Michigan Medical School, started in March 2003 and ended in December 2004. It had discovery, training, and validation phases. All serum samples, unless otherwise indicated, were obtained from patients in the University of Michigan Health System. Written informed consent was obtained from all patients.
The tissue and serum bank at the University of Michigan Specialized Research Program in Prostate Cancer has collected more than 2000 serum samples since 1995. Of these, 331 samples, which were collected from 1995 to 2004, met the following eligibility criteria: they were obtained immediately before surgery from patients with biopsy-proven, clinically localized prostate cancer who were at least 40 years old and who had received no previous prostate-cancer therapy. Of the 331 samples, 150 were randomly chosen for the discovery, training, and validation phases of the study. Four samples were excluded because proteins had precipitated in the serum vials. Twelve samples that fulfilled the criteria were obtained from the Dana – Farber Cancer Institute for the validation phase in order to explore the robustness of the assay across institutions. These samples were chosen randomly from 236 samples that met the eligibility criteria.
The University of Michigan Clinical Pathology Laboratories provided 159 control samples of serum from men between the ages of 46 and 83 years who had no history of cancer. These samples were collected from 2001 to 2004 in three independent collection periods. An additional 55 samples (designated as the "other" category) were collected from patients after prostatectomy (14 samples) or from men with either advanced hormone-refractory prostate cancer (11) or lung cancer (30, randomly selected from 74 samples obtained from male patients). Details regarding the serum samples that were used in our "other" cohort are in the Supplementary Appendix (available with the full text of this article at www.nejm.org).
In the discovery phase (biopanning and 2304-element microarrays), 39 prostate-cancer samples and 21 control samples were used. The training phase involved the use of 59 prostate-cancer samples and 70 control samples. To evaluate the phage-peptide detectors that we developed in the discovery and training phases, we used an independent validation set of 60 prostate-cancer samples (48 from the University of Michigan and 12 from the Dana – Farber Cancer Institute) and 68 control samples. In addition, 55 samples in the "other" category were assessed exclusively in the validation phase.
In the 257 prostate-cancer samples and control samples (which included the training and validation sets [Table 1]), the median levels of PSA were 6.3 ng per milliliter (range, 0.1 to 46.3) and 1.7 ng per milliliter (range, 0.1 to 24.5), respectively.
Table 1. Clinical and Pathological Information for the Training and Validation Samples.
Autoantibody Profiling
By iterative biopanning (see Glossary) of a phage-display library derived from prostate-cancer tissues, we developed phage protein microarrays and used them to develop an autoantibody signature to distinguish samples with prostate cancer from those of controls. Details concerning the construction of phage-display libraries and preparation of the phage-protein microarrays are given in the Supplementary Appendix and shown in Figure 1.
Figure 1. Schematic Representation of the Development of Phage-Protein Microarrays to Characterize Autoantibody Signatures in Prostate Cancer (Discovery Phase).
A cDNA library was constructed from a pool of total mRNA isolated from prostate-cancer tissue obtained from six patients. After digestion, the cDNA library was inserted into the T7 phage vector. The T7 fusion vectors were then packaged into T7 phages to generate a prostate-cancer cDNA T7 phage-display library. To enrich the library with clones of peptides reacting with human serum from patients with clinically localized prostate cancer and not with serum from controls, several cycles of affinity selections (biopanning) were performed. Briefly, the phage libraries were preadsorbed onto purified IgGs from the control pool of serum samples (from 10 patients) to remove nonspecific clones. Next, the precleared phage libraries were enriched for cancer-specific peptides with the use of a pool of IgGs purified from the serum of 19 patients with prostate cancer. The bound phages were eluted and propagated by infecting bacterial cells. After five rounds of biopanning, enriched prostate-cancer–specific peptide clones were cultured onto LB agar plates. A total of 2304 single colonies, including T7 empty phage clones as negative spots and antihuman IgG as positive spots, were randomly picked and propagated into 96-well plates. Phage-clone lysates were then printed onto coated glass slides with the use of a robotic spotter to create a phage-protein microarray. Cy5 (red fluorescent dye)–labeled antihuman antibody was used to detect IgGs in human serum that were reactive to peptide clones, and a Cy3 (green fluorescent dye)–labeled antibody was used to detect the phage capsid protein in order to normalize for spotting. Thus, if a phage clone carries a peptide that is reactive to human IgG, after scanning, this spot will be yellow in color; otherwise, the spot will remain green, representing an unreactive clone. A total of 31 samples (20 from patients with cancers and 11 from controls) were tested on the 2304 phage-peptide microarray. Analysis of these 31 samples identified 186 phage peptides with the highest level of differentiation between cancers and controls, which were then used to develop focused microarrays for analyses in the subsequent training and validation phases.
Statistical Analysis
Statistical analyses were performed with SPSS software, version 11.5, and R 2.0. Details are given in the Supplementary Appendix.
Results
Development of Phage-Protein Microarrays (Discovery Phase)
To develop a phage-display library of prostate-cancer peptides, we isolated messenger RNA (mRNA) from prostate-cancer tissue obtained from six patients with clinically localized disease (Table 1 of the Supplementary Appendix). After the insertion of the complementary DNA (cDNA) fragments into the T7 phage system, peptides that were encoded by the prostate cancer cDNA were expressed and displayed on the surface of the phage fused to the C-terminal of the capsid 10B protein of the phage. This surface complex functioned as bait to capture autoantibodies in serum.
Serum samples from 39 patients with prostate cancer and 21 controls were selected randomly from the University of Michigan serum cohorts for the discovery phase. To enrich the library for peptides that bind specifically to autoantibodies in patients with prostate cancer, we carried out successive rounds of selection (Figure 1). The procedure entailed removal of irrelevant phages from the library with the use of protein A/G beads that were coated with a pool of IgG antibodies isolated from 10 randomly selected control serum samples (Figure 1, and Table 2 of the Supplementary Appendix). These beads were incubated with the phage particles, and the supernatant, which contained unbound phage particles, was collected. These residual phages, now freed of phage particles that bind to irrelevant antibodies in normal serum, were enriched for phage particles that express prostate-cancer–specific peptides by incubation with protein A/G beads coated with a pool of IgG antibodies from 19 patients with prostate cancer who were chosen randomly from the University of Michigan serum collection (Figure 1, and Table 2 of the Supplementary Appendix). Finally, the adherent phages were eluted from the beads and propagated in bacterial cells.
We carried out five such rounds of purification (biopanning). Phage clones, each bearing a single fusion peptide derived from the prostate-cancer cDNA library, were selected randomly from the purified library to generate protein microarrays on coated glass slides with the use of a robotic spotter. Once in a microarray format, the enriched phage clones were used to test serum for autoantibodies against prostate cancer peptides.
Initially, we selected 2304 individual phage clones, including 5 empty clones as negative controls, from the enriched phage library and constructed a high-density protein microarray (Figure 3 of the Supplementary Appendix). To decrease the complexity of subsequent validation studies and develop a focused array, we randomly selected 20 serum samples from patients with cancer and 11 samples from controls from the respective University of Michigan collections (Table 3 of the Supplementary Appendix) and screened the high-density phage microarray. Of the 20 samples from patients with prostate cancer, 19 contained antibodies that reacted with phage-peptide clones on the microarrays, whereas only 1 of 11 controls had such antibodies (Figure 3 of the Supplementary Appendix). After normalization of all values obtained by the scanner, we selected phage-peptide clones that yielded a ratio of Cy5 to Cy3 greater than 1.2 in at least one of the serum samples. This analysis identified 186 phage-peptide clones that reacted with serum samples from patients with prostate cancer. These clones, along with negative-control phage clones, were used to construct a smaller, focused protein microarray for subsequent screening of serum samples (training and validation phases).
Identification of a Phage-Peptide Detector (Training Phase)
Figure 2 shows the training and validation phases of this study. A total of 257 serum samples from 119 patients with clinically localized prostate cancer and 138 controls, plus 55 samples in the "other" category, were tested on the 186-element focused arrays (Table 1 and Table 4 through Table 9 of the Supplementary Appendix). In the training phase, we analyzed 59 samples from patients with prostate cancer and 70 control samples (Figure 2). An algorithm17 was used to evaluate whether the sample size we profiled was sufficient to build a classifier for clinical diagnosis. On the basis of this algorithm and with the use of the specific measures of our phage-protein microarray data, we found that a mean (±SD) of 46±40 samples from 1000 simulation runs would be sufficient to ensure with 95 percent confidence that the probability of the misclassification of a future sample would be less than 0.15. Figure 3 depicts representative scanned arrays. To create a "class detector," we used a nonparametric-pattern-recognition approach that consisted of a genetic algorithm combined with k-nearest neighbor to select a subgroup of "informative" phage peptides based on leave-one-out cross-validation on the training samples. We identified a panel of 22 phage-peptide clones that could best distinguish serum samples from patients with prostate cancer from those of controls, with 97.1 percent specificity (2 of 70 control samples were misclassified) and 88.1 percent sensitivity (7 of 59 prostate-cancer samples were misclassified) in the training set (Table 10 and Table 11 of the Supplementary Appendix). Figure 4A shows a heatmap of the results with the 22 phage-peptide clones in the training set.
Figure 2. Overview of the Strategy Used for the Development and Validation of Autoantibody Signatures to Identify Prostate Cancer (Training and Validation Phases).
Figure 3. Autoantibody Signatures in Prostate Cancer.
Representative images of phage-protein microarrays demonstrate the difference in immunoreactivity between samples from patients with prostate cancer and those from controls. Yellow spots represent immunoreactivity in serum samples, and green spots no reactivity. Circled spots represent five phage-peptide clones: 5'-UTR_BMI1 (1), eIF4G1 (2), RPL13a (3), BRD2 (4), and RPL22 (5).
Figure 4. Supervised Analyses and Validation of Autoantibody Signatures in Prostate Cancer.
Representations are shown of heatmaps of 22 phage peptides analyzed for immunoreactivity across 129 training samples (Panel A) and for an independent validation set of 128 serum samples from patients with prostate cancer and from controls (Panel B). Individual peptide clones were represented in rows, whereas serum samples were represented in columns. Yellow indicates positive immunoreactivity, and black or blue no immunoreactivity. In Panel C, the performance of the 22-phage-peptide detector is compared with prostate-specific antigen (PSA) in the validation set. Receiver-operating-characteristic curves are based on multiplex analysis of the 22-phage-peptide biomarkers and serum PSA from a total of 128 samples (60 from patients with prostate cancer and 68 from controls). The red line indicates the 22-phage-peptide detector, and the green line indicates the PSA test. Panel D shows the performance of the 22-phage-peptide detector in patients with PSA levels between 4 and 10 ng per milliliter. The samples were a subgroup of the 128-sample validation group, with a total of 42 samples (22 from patients with cancer and 20 from controls). Panel E shows the performance of the 22-phage-peptide detector in patients with PSA levels between 2.5 and 10 ng per milliliter. The samples were a subgroup of the 128-sample validation group, with a total of 51 samples (28 from patients with cancer and 23 from controls).
Validation of the Phage-Peptide Detector
With the use of a 22-phage-peptide detector derived from the training phase, we applied a weighted voting scheme to classify samples in the independent validation set (128 patients) as either prostate cancer or control (Figure 2). In total, 8 of 68 serum samples from controls and 11 of 60 samples from patients with prostate cancer were misclassified in this validation set (Figure 4B, and Table 12 of the Supplementary Appendix). These results yielded a specificity of 88.2 percent (95 percent confidence interval, 78 to 95) and a sensitivity of 81.6 percent (95 percent confidence interval, 70 to 90). Similar performance criteria were observed with the use of different class-prediction models and a second randomization of the data set (Section II of the Supplementary Appendix).
We next calculated receiver-operating-characteristic curves for the 22-phage-peptide detector and PSA levels in the validation set. Different cutoff values of weighted voting scores were used as threshold points to plot the true positive rate against the false positive rate for the prediction model. The ability of the panel of 22 phage peptides to discriminate between prostate-cancer samples and control samples was significant (P<0.001), with an area under the curve equal to 0.93 (95 percent confidence interval, 0.88 to 0.97) (Figure 4C). The area under the curve for PSA was 0.80 (P<0.001; 95 percent confidence interval, 0.71 to 0.88). This result was expected, since these patients were identified primarily by elevated PSA levels. Among patients with PSA levels of 4 to 10 ng per milliliter in the validation set, the phage-peptide detector had significant discriminatory power (P<0.001) as compared with PSA (P=0.50) in distinguishing serum samples from patients with prostate cancer from those of controls. The area under the curve was 0.93 (95 percent confidence interval, 0.86 to 1.00) for the phage-display method and 0.56 (95 percent confidence interval, 0.38 to 0.74) for PSA (Figure 4D). When the lower limit of PSA was decreased to 2.5 ng per milliliter, the discriminatory power of the phage-peptide profile was maintained (P<0.001), with an area under the curve of 0.94 (95 percent confidence interval, 0.88 to 1.00), whereas that for PSA decreased slightly to 0.50 (95 percent confidence interval, 0.33 to 0.66) (Figure 4E, and Section III of the Supplementary Appendix).
We also used the 22-phage-peptide detector to test the 55 serum samples in the "other" category. Among the samples obtained from patients after prostatectomy, 5 of 14 samples were classified as prostate cancer, and 3 of 11 samples from patients with hormone-refractory disease were classified as prostate cancer (Table 13 of the Supplementary Appendix). Of the 30 samples from patients with lung adenocarcinoma, 9 were classified as having prostate cancer, which suggests some cross-reactivity of autoantibodies across tumor types.
Characterization of the Phage-Peptide Detector
The panel of 22 phage-peptide clones was sequenced. Of these, five were in-frame and within known expressed sequences (Table 16 of the Supplementary Appendix). These five included bromodomain-containing protein 2 (BRD2), eukaryotic translation initiation factor 4 gamma 1 (eIF4G1), ribosomal protein L22 (RPL22), ribosomal protein L13a (RPL13a), and hypothetical protein XP_373908 [GenBank] . Except for hypothetical protein XP_373908 [GenBank] , these structures were derived from intracellular proteins involved in regulating either transcription or translation. The remaining 17 phage-peptide clones were either in untranslated regions of expressed genes or out of frame in the coding sequence of known genes. These clones may express peptides that are structurally similar to peptides in expressed proteins but are unrelated or weakly related at the protein-sequence level (Table 17 and section IV of the Supplementary Appendix).
To determine whether the four in-frame phage-peptide clones (Figure 5A) are deregulated in prostate cancer, we performed a meta-analysis of publicly available data regarding gene expression in prostate cancer.18,19,20,21,22,23,24 These analyses and a preliminary immunoblot analysis suggested that the four in-frame phage epitopes are overexpressed in prostate cancer (Figure 5B and Figure 5C, and sections V and VI of the Supplementary Appendix).
Figure 5. Meta-Analysis of Gene Expression of Humoral-Immune-Response Candidates.
Panel A shows a heatmap representation of the immunoreactivity for four in-frame phage-peptide clones assessed across 257 serum samples. Yellow indicates immunoreactivity, and black or blue no immunoreactivity. Panel B shows relative levels of gene expression of in-frame phage-peptide clones assessed with the use of publicly available DNA microarray data on the Web site of Oncomine (www.oncomine.org). The first author of each DNA microarray study is provided, as are P values for each comparison — for example, benign versus localized prostate cancer (PCA) and PCA versus metastatic prostate cancer (MET). Panel C shows immunoblot validation of the overexpression of humoral-response candidates at the protein level in prostate cancer. Panel D shows expression of the humoral-response candidate eIF4G1 in prostate cancer by immunofluorescence staining. Subpanel 1 displays clinically localized prostate cancer (left) adjacent to a benign gland (right). Subpanel 2 displays magnifications of a single prostate-cancer gland. Stains for eIF4G1 (red), E-cadherin (green), and nuclei (blue) were used. The scale bar represents 5 μm. Panel E shows a histogram of staining intensity from immunohistochemical analysis. The open boxes represent benign tissue cores, and the black boxes represent tumor cores.
Discussion
In this study, we used protein microarrays to identify autoantibodies against tumor antigens in patients with prostate cancer. Specifically, we constructed phage-protein microarrays in which peptides derived from a prostate-cancer cDNA library were expressed as a prostate-cancer – phage fusion protein. The phage-protein microarrays were screened to identify phage-peptide clones that bind autoantibodies in serum samples from patients with prostate cancer but not in those from controls.
The use of PSA-based screening for prostate cancer has risen dramatically since its introduction in the late 1980s.25,26 However, reliance on PSA for the detection of early prostate cancer is still unsatisfactory, especially because of a high rate of false positive results27 — as high as 80 percent.28,29 This rate results in many unnecessary prostate biopsies.30 To circumvent this and other problems of screening for prostate cancer, we have begun to evaluate the use of autoantibody signatures to detect prostate cancer. By relying on multiple immunogenic prostate-cancer peptides, this approach may be an improvement over a single biomarker such as PSA.
Serologic analysis of recombinant cDNA expression libraries of human tumors with autologous serum (SEREX) has demonstrated that antibodies in the serum of patients with cancer can be used to isolate new tumor antigens.3,4 This technique, however, relies on one-step screening without affinity-selection steps and requires a large volume of serum to screen phage clones blotted onto membrane filters. The SEREX approach has limited clinical use, since it is not conducive to the analysis of hundreds of serum samples of small amounts.
By taking advantage of combinatorial screening and high throughput analysis of autoantibody repertoires, we developed a technique that overcomes the disadvantages of SEREX for cancer diagnosis. However, like gene-expression profiling and pattern-recognition approaches with serum proteomics, our method may have the limitations of background signals, sample-selection bias, and limited reproducibility.31 To minimize these problems, immunoreactivity for each phage peptide was measured in relation to an internal control signal detected by antibody against phage capsid proteins. The discriminatory power of autoantibody signatures was validated by reshuffling and analyzing the training and validation sets with the use of various class-prediction models. The reproducibility of the assay was investigated by the use of experimental assays both within and among the arrays. The difference between duplicate peptides in the same array was less than 5 percent for 98 percent of the spots. Analyses of repeated experiments with the use of the same serum samples revealed that the results were very consistent, with a correlation coefficient greater than 98 percent.
The autoantibody signature was detected in only 5 of 14 serum samples from patients who had undergone prostatectomy and in 3 of 11 serum samples from patients with hormone-refractory disease — a finding suggesting that the autoantibody profile is attenuated on removal of the "immunogen" or treatment with antiandrogens, chemotherapeutic agents, or both. Of 30 serum samples from patients with lung cancer, 9 were classified as prostate cancer. This result is in contrast to the more than 80 percent sensitivity for prostate cancer with the use of the phage-peptide system, suggesting that the autoantibody profile has relative specificity for prostate cancer (P<0.001 by the proportion test).
Our results were consistent across a range of clinical and pathological features, including PSA level, Gleason grade, stage, and presence or absence of PSA recurrence (Table 18 of the Supplementary Appendix), with sensitivities and specificities ranging from 80 to 90 percent in discriminating between patients with prostate cancer and controls. This diagnostic performance was maintained in the intermediate ranges of PSA (i.e., 4 to 10 ng per milliliter or 2.5 to 10 ng per milliliter).
Autoantibody signatures may be useful in combination with initial PSA screening; our data show that the 22-phage-peptide detector significantly adds to the diagnostic power of PSA alone (P<0.001). The use of such a "supplementary" autoantibody panel might be important at PSA levels of 10 ng per milliliter or less. This additional discriminatory power could improve the information used to make decisions about biopsy of the prostate.
The sequences in the phage display system are probably those of relatively short peptides rather than full-length proteins, since the cDNAs were enzyme-digested and fragmented before ligation into the phage vector. The average stretch of peptides in the 22-phage-peptide detector was 53±34 amino acids, with a maximum length of 134 residues and a minimum of 11 residues.
Four of the phage clones representing known proteins — BRD2, eIF4G1, RPL13a, and RPL22 — were substantially more reactive with serum from patients with prostate cancer than with that from controls (Figure 5A). Both meta-analysis and immunoblot analysis of tissue extracts suggest that all four proteins are deregulated in prostate tumors. Immunofluorescence of tissue sections and immunohistochemical analysis of tissue microarrays showed that eIF4G1 was overexpressed in prostate-cancer epithelial tissue, as compared with benign epithelial tissue (Figure 5D and Figure 5E, and Figure 5 and Section VI of the Supplementary Appendix).
Of the 22 phage peptides identified in this study, 17 are not present in peptide stretches in known proteins. These 17 peptides may be weakly homologous to known proteins or may have no distinct homology to the primary sequences of known proteins and thus may be "mimotopes" (i.e., stretches of amino acids that mimic an antigen but are not homologous at the sequence level).
We have not tested the phage-microarray system for screening for prostate cancer; this requires extension and confirmation in community-based screening cohorts. It will be important to evaluate the autoantibody signatures associated with prostate cancer in patients with prostatitis, autoimmune conditions, and other diseases. Although the technique is promising, how it will perform in prospective and multiinstitutional studies remains to be determined.
Supported by National Cancer Institute grants (UO1CA111275, to Drs. Chinnaiyan, Rubin, Ghosh, Sanda, and Wei, and P50 CA69568, to Drs. Pienta, Chinnaiyan, Montie, Rubin, Sanda, and Ghosh); National Institutes of Health grants (R01 CA8241901-A1, to Dr. Sanda, and R01GM72007-01, to Dr. Ghosh); a grant (RSG-02-179-MGO, to Drs. Chinnaiyan, Rubin, and Ghosh) from the American Cancer Society; a grant (to Dr. Chinnaiyan) from the V Foundation; grants (W81XWH-04-1-0886, to Dr. Wang; DMAD17-03-0105, to Dr. Sreekumar, and PC040517, to Dr. Mehra) from the Department of Defense; and a grant (5P30CA46592) from the Cancer Center Bioinformatics Core. Dr. Chinnaiyan is a Pew Biomedical Scholar.
We are indebted to Dr. Saravana M. Dhanasekaran for providing samples of prostate-cancer mRNA, to Dr. Paul L. Fox of the Lerner Research Institute in Cleveland for providing anti-RPL13a antibody, to Jason Harwood and Srilakshmi Bhagavathula of the Prostate Tissue and Informatics Core for the collection of serum samples and integration of associated clinical information, to Drs. David Beer and Guoan Chen for providing serum samples from patients with lung cancer as control samples for nonprostate cancer, to Drs. Francesca Demichelis and Jeremy Taylor for providing additional statistical consultation, to Jeffrey Fielhauer for his help with sequence analysis and preparation of the manuscript, and to the staff of the microscopy and image analyses laboratory at the University of Michigan.
Source Information
From the Departments of Pathology (X.W., J.Y., A.S., S.V., R.S., D. Giacherio, R.M., A.M.C.), Biostatistics (R.S., D. Ghosh), Urology (J.E.M., K.J.P., J.T.W., A.M.C.), and Internal Medicine (K.J.P.), and the Comprehensive Cancer Center (A.S., S.V., J.E.M., K.J.P., J.T.W., A.M.C.), University of Michigan Medical School, Ann Arbor; and Beth Israel – Deaconess Medical Center (M.G.S.), Dana – Farber Cancer Institute (P.W.K.), and Brigham and Women's Hospital (M.A.R.), Harvard Medical School — all in Boston.
Drs. Wang and Yu contributed equally to this article.
Address reprint requests to Dr. Chinnaiyan at the Department of Pathology and Urology, University of Michigan Medical School, 1301 Catherine St., MSI 4237, University of Michigan, Ann Arbor, MI 48109, or at arul@umich.edu.
References
Thompson IM, Pauler DK, Goodman PJ, et al. Prevalence of prostate cancer among men with a prostate-specific antigen level 4.0 ng per milliliter. N Engl J Med 2004;350:2239-2246.
Sahin U, Tureci O, Schmitt H, et al. Human neoplasms elicit multiple specific immune responses in the autologous host. Proc Natl Acad Sci U S A 1995;92:11810-11813.
Chen YT, Gure AO, Tsang S, et al. Identification of multiple cancer/testis antigens by allogeneic antibody screening of a melanoma cell line library. Proc Natl Acad Sci U S A 1998;95:6919-6923.
Stockert E, Jager E, Chen YT, et al. A survey of the humoral immune response of cancer patients to a panel of human tumor antigens. J Exp Med 1998;187:1349-1354.
Brichory FM, Misek DE, Yim AM, et al. 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;98:9824-9829.
Minenkova O, Pucci A, Pavoni E, et al. Identification of tumor-associated antigens by screening phage-displayed human cDNA libraries with sera from tumor patients. Int J Cancer 2003;106:534-544.
Zhong L, Peng X, Hidalgo GE, Doherty DE, Stromberg AJ, Hirschowitz EA. Antibodies to HSP70 and HSP90 in serum in non-small cell lung cancer patients. Cancer Detect Prev 2003;27:285-290.
Old LJ, Chen YT. New paths in human cancer serology. J Exp Med 1998;187:1163-1167.
Nilsson BO, Carlsson L, Larsson A, Ronquist G. Autoantibodies to prostasomes as new markers for prostate cancer. Ups J Med Sci 2001;106:43-49.
Mintz PJ, Kim J, Do KA, et al. Fingerprinting the circulating repertoire of antibodies from cancer patients. Nat Biotechnol 2003;21:57-63.
Soussi T. p53 Antibodies in the sera of patients with various types of cancer: a review. Cancer Res 2000;60:1777-1788.
Sreekumar A, Laxman B, Rhodes DR, et al. Humoral immune response to alpha-methylacyl-CoA racemase and prostate cancer. J Natl Cancer Inst 2004;96:834-843.
Luo J, Zha S, Gage WR, et al. Alpha-methylacyl-CoA racemase: a new molecular marker for prostate cancer. Cancer Res 2002;62:2220-2226.
Jiang Z, Woda BA, Rock KL, et al. P504S: a new molecular marker for the detection of prostate carcinoma. Am J Surg Pathol 2001;25:1397-1404.
Rubin MA, Zhou M, Dhanasekaran SM, et al. alpha-Methylacyl coenzyme A racemase as a tissue biomarker for prostate cancer. JAMA 2002;287:1662-1670.
Fernandez-Madrid F, Tang N, Alansari H, et al. Autoantibodies to annexin XI-A and other autoantigens in the diagnosis of breast cancer. Cancer Res 2004;64:5089-5096.
Fu WJ, Dougherty ER, Mallick B, Carroll RJ. How many samples are needed to build a classifier: a general sequential approach. Bioinformatics 2005;21:63-70.
Dhanasekaran SM, Barrette TR, Ghosh D, et al. Delineation of prognostic biomarkers in prostate cancer. Nature 2001;412:822-826.
LaTulippe E, Satagopan J, Smith A, et al. Comprehensive gene expression analysis of prostate cancer reveals distinct transcriptional programs associated with metastatic disease. Cancer Res 2002;62:4499-4506.
Luo JH, Yu YP, Cieply K, et al. Gene expression analysis of prostate cancers. Mol Carcinog 2002;33:25-35.
Luo J, Duggan DJ, Chen Y, et al. Human prostate cancer and benign prostatic hyperplasia: molecular dissection by gene expression profiling. Cancer Res 2001;61:4683-4688.
Singh D, Febbo PG, Ross K, et al. Gene expression correlates of clinical prostate cancer behavior. Cancer Cell 2002;1:203-209.
Welsh JB, Sapinoso LM, Su AI, et al. Analysis of gene expression identifies candidate markers and pharmacological targets in prostate cancer. Cancer Res 2001;61:5974-5978.
Rhodes DR, Yu J, Shanker K, et al. ONCOMINE: a cancer microarray database and integrated data-mining platform. Neoplasia 2004;6:1-6.
Jemal A, Tiwari RC, Murray T, et al. Cancer statistics, 2004. CA Cancer J Clin 2004;54:8-29.
Dennis LK, Resnick MI. Analysis of recent trends in prostate cancer incidence and mortality. Prostate 2000;42:247-252.
Schmid HP, Prikler L, Semjonow A. Problems with prostate-specific antigen screening: a critical review. Recent Results Cancer Res 2003;163:226-231.
Catalona WJ, Smith DS, Ratliff TL, et al. Measurement of prostate-specific antigen in serum as a screening test for prostate cancer. N Engl J Med 1991;324:1156-1161.
Makinen T, Tammela TL, Stenman UH, et al. Second round results of the Finnish population-based prostate cancer screening trial. Clin Cancer Res 2004;10:2231-2236.
Cohen L, Fouladi RT, Babaian RJ, et al. Cancer worry is associated with abnormal prostate-specific antigen levels in men participating in a community screening program. Cancer Epidemiol Biomarkers Prev 2003;12:610-617.
Ransohoff DF. Bias as a threat to the validity of cancer molecular-marker research. Nat Rev Cancer 2005;5:142-149.(Xiaoju Wang, Ph.D., Jianj)
Background New biomarkers, such as autoantibody signatures, may improve the early detection of prostate cancer.
Methods With a phage-display library derived from prostate-cancer tissue, we developed and used phage protein microarrays to analyze serum samples from 119 patients with prostate cancer and 138 controls, with the samples equally divided into training and validation sets. A phage-peptide detector that was constructed from the training set was evaluated on an independent validation set of 128 serum samples (60 from patients with prostate cancer and 68 from controls).
Results A 22-phage-peptide detector had 88.2 percent specificity (95 percent confidence interval, 0.78 to 0.95) and 81.6 percent sensitivity (95 percent confidence interval, 0.70 to 0.90) in discriminating between the group with prostate cancer and the control group. This panel of peptides performed better than did prostate-specific antigen (PSA) in distinguishing between the group with prostate cancer and the control group (area under the curve for the autoantibody signature, 0.93; 95 percent confidence interval, 0.88 to 0.97; area under the curve for PSA, 0.80; 95 percent confidence interval, 0.71 to 0.88). Logistic-regression analysis revealed that the phage-peptide panel provided additional discriminative power over PSA (P<0.001). Among the 22 phage peptides used as a detector, 4 were derived from in-frame, named coding sequences. The remaining phage peptides were generated from untranslated sequences.
Conclusions Autoantibodies against peptides derived from prostate-cancer tissue could be used as the basis for a screening test for prostate cancer.
Limitations of the prostate-specific antigen (PSA) test for the early detection of prostate cancer1 indicate the need for other means of screening for this neoplasm. The finding that patients with cancer produce autoantibodies against antigens in their tumors2,3,4,5,6,7 suggests that such autoantibodies could have diagnostic and prognostic value.4,5,8,9,10 For example, mutant forms of the p53 protein elicit anti-p53 antibodies in 30 to 40 percent of patients with various types of cancers.11 Recently, we found that patients with prostate cancer produce antibodies against -methylacyl-coenzyme A racemase,12 an overexpressed protein in epithelial cells in prostate cancer.13,14,15 This autoantibody had 72 percent specificity and 62 percent sensitivity in detecting prostate cancer.12 The use of additional prostate-cancer antigens could improve the sensitivity and specificity of an autoantibody-based screening test for prostate cancer.
Here we report the use of a technique that combines phage-display technology with protein microarrays to identify and characterize new autoantibody-binding peptides derived from prostate-cancer tissue. A similar approach has been used to identify selected antigens for the diagnosis of breast cancer.16 This emerging area of research, termed "cancer immunomics," allows a global analysis of the autoantibodies against antigens in a neoplasm (see the Glossary for definitions of terms).
Glossary
Methods
Populations and Samples
This study, which was approved by the institutional review board of the University of Michigan Medical School, started in March 2003 and ended in December 2004. It had discovery, training, and validation phases. All serum samples, unless otherwise indicated, were obtained from patients in the University of Michigan Health System. Written informed consent was obtained from all patients.
The tissue and serum bank at the University of Michigan Specialized Research Program in Prostate Cancer has collected more than 2000 serum samples since 1995. Of these, 331 samples, which were collected from 1995 to 2004, met the following eligibility criteria: they were obtained immediately before surgery from patients with biopsy-proven, clinically localized prostate cancer who were at least 40 years old and who had received no previous prostate-cancer therapy. Of the 331 samples, 150 were randomly chosen for the discovery, training, and validation phases of the study. Four samples were excluded because proteins had precipitated in the serum vials. Twelve samples that fulfilled the criteria were obtained from the Dana – Farber Cancer Institute for the validation phase in order to explore the robustness of the assay across institutions. These samples were chosen randomly from 236 samples that met the eligibility criteria.
The University of Michigan Clinical Pathology Laboratories provided 159 control samples of serum from men between the ages of 46 and 83 years who had no history of cancer. These samples were collected from 2001 to 2004 in three independent collection periods. An additional 55 samples (designated as the "other" category) were collected from patients after prostatectomy (14 samples) or from men with either advanced hormone-refractory prostate cancer (11) or lung cancer (30, randomly selected from 74 samples obtained from male patients). Details regarding the serum samples that were used in our "other" cohort are in the Supplementary Appendix (available with the full text of this article at www.nejm.org).
In the discovery phase (biopanning and 2304-element microarrays), 39 prostate-cancer samples and 21 control samples were used. The training phase involved the use of 59 prostate-cancer samples and 70 control samples. To evaluate the phage-peptide detectors that we developed in the discovery and training phases, we used an independent validation set of 60 prostate-cancer samples (48 from the University of Michigan and 12 from the Dana – Farber Cancer Institute) and 68 control samples. In addition, 55 samples in the "other" category were assessed exclusively in the validation phase.
In the 257 prostate-cancer samples and control samples (which included the training and validation sets [Table 1]), the median levels of PSA were 6.3 ng per milliliter (range, 0.1 to 46.3) and 1.7 ng per milliliter (range, 0.1 to 24.5), respectively.
Table 1. Clinical and Pathological Information for the Training and Validation Samples.
Autoantibody Profiling
By iterative biopanning (see Glossary) of a phage-display library derived from prostate-cancer tissues, we developed phage protein microarrays and used them to develop an autoantibody signature to distinguish samples with prostate cancer from those of controls. Details concerning the construction of phage-display libraries and preparation of the phage-protein microarrays are given in the Supplementary Appendix and shown in Figure 1.
Figure 1. Schematic Representation of the Development of Phage-Protein Microarrays to Characterize Autoantibody Signatures in Prostate Cancer (Discovery Phase).
A cDNA library was constructed from a pool of total mRNA isolated from prostate-cancer tissue obtained from six patients. After digestion, the cDNA library was inserted into the T7 phage vector. The T7 fusion vectors were then packaged into T7 phages to generate a prostate-cancer cDNA T7 phage-display library. To enrich the library with clones of peptides reacting with human serum from patients with clinically localized prostate cancer and not with serum from controls, several cycles of affinity selections (biopanning) were performed. Briefly, the phage libraries were preadsorbed onto purified IgGs from the control pool of serum samples (from 10 patients) to remove nonspecific clones. Next, the precleared phage libraries were enriched for cancer-specific peptides with the use of a pool of IgGs purified from the serum of 19 patients with prostate cancer. The bound phages were eluted and propagated by infecting bacterial cells. After five rounds of biopanning, enriched prostate-cancer–specific peptide clones were cultured onto LB agar plates. A total of 2304 single colonies, including T7 empty phage clones as negative spots and antihuman IgG as positive spots, were randomly picked and propagated into 96-well plates. Phage-clone lysates were then printed onto coated glass slides with the use of a robotic spotter to create a phage-protein microarray. Cy5 (red fluorescent dye)–labeled antihuman antibody was used to detect IgGs in human serum that were reactive to peptide clones, and a Cy3 (green fluorescent dye)–labeled antibody was used to detect the phage capsid protein in order to normalize for spotting. Thus, if a phage clone carries a peptide that is reactive to human IgG, after scanning, this spot will be yellow in color; otherwise, the spot will remain green, representing an unreactive clone. A total of 31 samples (20 from patients with cancers and 11 from controls) were tested on the 2304 phage-peptide microarray. Analysis of these 31 samples identified 186 phage peptides with the highest level of differentiation between cancers and controls, which were then used to develop focused microarrays for analyses in the subsequent training and validation phases.
Statistical Analysis
Statistical analyses were performed with SPSS software, version 11.5, and R 2.0. Details are given in the Supplementary Appendix.
Results
Development of Phage-Protein Microarrays (Discovery Phase)
To develop a phage-display library of prostate-cancer peptides, we isolated messenger RNA (mRNA) from prostate-cancer tissue obtained from six patients with clinically localized disease (Table 1 of the Supplementary Appendix). After the insertion of the complementary DNA (cDNA) fragments into the T7 phage system, peptides that were encoded by the prostate cancer cDNA were expressed and displayed on the surface of the phage fused to the C-terminal of the capsid 10B protein of the phage. This surface complex functioned as bait to capture autoantibodies in serum.
Serum samples from 39 patients with prostate cancer and 21 controls were selected randomly from the University of Michigan serum cohorts for the discovery phase. To enrich the library for peptides that bind specifically to autoantibodies in patients with prostate cancer, we carried out successive rounds of selection (Figure 1). The procedure entailed removal of irrelevant phages from the library with the use of protein A/G beads that were coated with a pool of IgG antibodies isolated from 10 randomly selected control serum samples (Figure 1, and Table 2 of the Supplementary Appendix). These beads were incubated with the phage particles, and the supernatant, which contained unbound phage particles, was collected. These residual phages, now freed of phage particles that bind to irrelevant antibodies in normal serum, were enriched for phage particles that express prostate-cancer–specific peptides by incubation with protein A/G beads coated with a pool of IgG antibodies from 19 patients with prostate cancer who were chosen randomly from the University of Michigan serum collection (Figure 1, and Table 2 of the Supplementary Appendix). Finally, the adherent phages were eluted from the beads and propagated in bacterial cells.
We carried out five such rounds of purification (biopanning). Phage clones, each bearing a single fusion peptide derived from the prostate-cancer cDNA library, were selected randomly from the purified library to generate protein microarrays on coated glass slides with the use of a robotic spotter. Once in a microarray format, the enriched phage clones were used to test serum for autoantibodies against prostate cancer peptides.
Initially, we selected 2304 individual phage clones, including 5 empty clones as negative controls, from the enriched phage library and constructed a high-density protein microarray (Figure 3 of the Supplementary Appendix). To decrease the complexity of subsequent validation studies and develop a focused array, we randomly selected 20 serum samples from patients with cancer and 11 samples from controls from the respective University of Michigan collections (Table 3 of the Supplementary Appendix) and screened the high-density phage microarray. Of the 20 samples from patients with prostate cancer, 19 contained antibodies that reacted with phage-peptide clones on the microarrays, whereas only 1 of 11 controls had such antibodies (Figure 3 of the Supplementary Appendix). After normalization of all values obtained by the scanner, we selected phage-peptide clones that yielded a ratio of Cy5 to Cy3 greater than 1.2 in at least one of the serum samples. This analysis identified 186 phage-peptide clones that reacted with serum samples from patients with prostate cancer. These clones, along with negative-control phage clones, were used to construct a smaller, focused protein microarray for subsequent screening of serum samples (training and validation phases).
Identification of a Phage-Peptide Detector (Training Phase)
Figure 2 shows the training and validation phases of this study. A total of 257 serum samples from 119 patients with clinically localized prostate cancer and 138 controls, plus 55 samples in the "other" category, were tested on the 186-element focused arrays (Table 1 and Table 4 through Table 9 of the Supplementary Appendix). In the training phase, we analyzed 59 samples from patients with prostate cancer and 70 control samples (Figure 2). An algorithm17 was used to evaluate whether the sample size we profiled was sufficient to build a classifier for clinical diagnosis. On the basis of this algorithm and with the use of the specific measures of our phage-protein microarray data, we found that a mean (±SD) of 46±40 samples from 1000 simulation runs would be sufficient to ensure with 95 percent confidence that the probability of the misclassification of a future sample would be less than 0.15. Figure 3 depicts representative scanned arrays. To create a "class detector," we used a nonparametric-pattern-recognition approach that consisted of a genetic algorithm combined with k-nearest neighbor to select a subgroup of "informative" phage peptides based on leave-one-out cross-validation on the training samples. We identified a panel of 22 phage-peptide clones that could best distinguish serum samples from patients with prostate cancer from those of controls, with 97.1 percent specificity (2 of 70 control samples were misclassified) and 88.1 percent sensitivity (7 of 59 prostate-cancer samples were misclassified) in the training set (Table 10 and Table 11 of the Supplementary Appendix). Figure 4A shows a heatmap of the results with the 22 phage-peptide clones in the training set.
Figure 2. Overview of the Strategy Used for the Development and Validation of Autoantibody Signatures to Identify Prostate Cancer (Training and Validation Phases).
Figure 3. Autoantibody Signatures in Prostate Cancer.
Representative images of phage-protein microarrays demonstrate the difference in immunoreactivity between samples from patients with prostate cancer and those from controls. Yellow spots represent immunoreactivity in serum samples, and green spots no reactivity. Circled spots represent five phage-peptide clones: 5'-UTR_BMI1 (1), eIF4G1 (2), RPL13a (3), BRD2 (4), and RPL22 (5).
Figure 4. Supervised Analyses and Validation of Autoantibody Signatures in Prostate Cancer.
Representations are shown of heatmaps of 22 phage peptides analyzed for immunoreactivity across 129 training samples (Panel A) and for an independent validation set of 128 serum samples from patients with prostate cancer and from controls (Panel B). Individual peptide clones were represented in rows, whereas serum samples were represented in columns. Yellow indicates positive immunoreactivity, and black or blue no immunoreactivity. In Panel C, the performance of the 22-phage-peptide detector is compared with prostate-specific antigen (PSA) in the validation set. Receiver-operating-characteristic curves are based on multiplex analysis of the 22-phage-peptide biomarkers and serum PSA from a total of 128 samples (60 from patients with prostate cancer and 68 from controls). The red line indicates the 22-phage-peptide detector, and the green line indicates the PSA test. Panel D shows the performance of the 22-phage-peptide detector in patients with PSA levels between 4 and 10 ng per milliliter. The samples were a subgroup of the 128-sample validation group, with a total of 42 samples (22 from patients with cancer and 20 from controls). Panel E shows the performance of the 22-phage-peptide detector in patients with PSA levels between 2.5 and 10 ng per milliliter. The samples were a subgroup of the 128-sample validation group, with a total of 51 samples (28 from patients with cancer and 23 from controls).
Validation of the Phage-Peptide Detector
With the use of a 22-phage-peptide detector derived from the training phase, we applied a weighted voting scheme to classify samples in the independent validation set (128 patients) as either prostate cancer or control (Figure 2). In total, 8 of 68 serum samples from controls and 11 of 60 samples from patients with prostate cancer were misclassified in this validation set (Figure 4B, and Table 12 of the Supplementary Appendix). These results yielded a specificity of 88.2 percent (95 percent confidence interval, 78 to 95) and a sensitivity of 81.6 percent (95 percent confidence interval, 70 to 90). Similar performance criteria were observed with the use of different class-prediction models and a second randomization of the data set (Section II of the Supplementary Appendix).
We next calculated receiver-operating-characteristic curves for the 22-phage-peptide detector and PSA levels in the validation set. Different cutoff values of weighted voting scores were used as threshold points to plot the true positive rate against the false positive rate for the prediction model. The ability of the panel of 22 phage peptides to discriminate between prostate-cancer samples and control samples was significant (P<0.001), with an area under the curve equal to 0.93 (95 percent confidence interval, 0.88 to 0.97) (Figure 4C). The area under the curve for PSA was 0.80 (P<0.001; 95 percent confidence interval, 0.71 to 0.88). This result was expected, since these patients were identified primarily by elevated PSA levels. Among patients with PSA levels of 4 to 10 ng per milliliter in the validation set, the phage-peptide detector had significant discriminatory power (P<0.001) as compared with PSA (P=0.50) in distinguishing serum samples from patients with prostate cancer from those of controls. The area under the curve was 0.93 (95 percent confidence interval, 0.86 to 1.00) for the phage-display method and 0.56 (95 percent confidence interval, 0.38 to 0.74) for PSA (Figure 4D). When the lower limit of PSA was decreased to 2.5 ng per milliliter, the discriminatory power of the phage-peptide profile was maintained (P<0.001), with an area under the curve of 0.94 (95 percent confidence interval, 0.88 to 1.00), whereas that for PSA decreased slightly to 0.50 (95 percent confidence interval, 0.33 to 0.66) (Figure 4E, and Section III of the Supplementary Appendix).
We also used the 22-phage-peptide detector to test the 55 serum samples in the "other" category. Among the samples obtained from patients after prostatectomy, 5 of 14 samples were classified as prostate cancer, and 3 of 11 samples from patients with hormone-refractory disease were classified as prostate cancer (Table 13 of the Supplementary Appendix). Of the 30 samples from patients with lung adenocarcinoma, 9 were classified as having prostate cancer, which suggests some cross-reactivity of autoantibodies across tumor types.
Characterization of the Phage-Peptide Detector
The panel of 22 phage-peptide clones was sequenced. Of these, five were in-frame and within known expressed sequences (Table 16 of the Supplementary Appendix). These five included bromodomain-containing protein 2 (BRD2), eukaryotic translation initiation factor 4 gamma 1 (eIF4G1), ribosomal protein L22 (RPL22), ribosomal protein L13a (RPL13a), and hypothetical protein XP_373908 [GenBank] . Except for hypothetical protein XP_373908 [GenBank] , these structures were derived from intracellular proteins involved in regulating either transcription or translation. The remaining 17 phage-peptide clones were either in untranslated regions of expressed genes or out of frame in the coding sequence of known genes. These clones may express peptides that are structurally similar to peptides in expressed proteins but are unrelated or weakly related at the protein-sequence level (Table 17 and section IV of the Supplementary Appendix).
To determine whether the four in-frame phage-peptide clones (Figure 5A) are deregulated in prostate cancer, we performed a meta-analysis of publicly available data regarding gene expression in prostate cancer.18,19,20,21,22,23,24 These analyses and a preliminary immunoblot analysis suggested that the four in-frame phage epitopes are overexpressed in prostate cancer (Figure 5B and Figure 5C, and sections V and VI of the Supplementary Appendix).
Figure 5. Meta-Analysis of Gene Expression of Humoral-Immune-Response Candidates.
Panel A shows a heatmap representation of the immunoreactivity for four in-frame phage-peptide clones assessed across 257 serum samples. Yellow indicates immunoreactivity, and black or blue no immunoreactivity. Panel B shows relative levels of gene expression of in-frame phage-peptide clones assessed with the use of publicly available DNA microarray data on the Web site of Oncomine (www.oncomine.org). The first author of each DNA microarray study is provided, as are P values for each comparison — for example, benign versus localized prostate cancer (PCA) and PCA versus metastatic prostate cancer (MET). Panel C shows immunoblot validation of the overexpression of humoral-response candidates at the protein level in prostate cancer. Panel D shows expression of the humoral-response candidate eIF4G1 in prostate cancer by immunofluorescence staining. Subpanel 1 displays clinically localized prostate cancer (left) adjacent to a benign gland (right). Subpanel 2 displays magnifications of a single prostate-cancer gland. Stains for eIF4G1 (red), E-cadherin (green), and nuclei (blue) were used. The scale bar represents 5 μm. Panel E shows a histogram of staining intensity from immunohistochemical analysis. The open boxes represent benign tissue cores, and the black boxes represent tumor cores.
Discussion
In this study, we used protein microarrays to identify autoantibodies against tumor antigens in patients with prostate cancer. Specifically, we constructed phage-protein microarrays in which peptides derived from a prostate-cancer cDNA library were expressed as a prostate-cancer – phage fusion protein. The phage-protein microarrays were screened to identify phage-peptide clones that bind autoantibodies in serum samples from patients with prostate cancer but not in those from controls.
The use of PSA-based screening for prostate cancer has risen dramatically since its introduction in the late 1980s.25,26 However, reliance on PSA for the detection of early prostate cancer is still unsatisfactory, especially because of a high rate of false positive results27 — as high as 80 percent.28,29 This rate results in many unnecessary prostate biopsies.30 To circumvent this and other problems of screening for prostate cancer, we have begun to evaluate the use of autoantibody signatures to detect prostate cancer. By relying on multiple immunogenic prostate-cancer peptides, this approach may be an improvement over a single biomarker such as PSA.
Serologic analysis of recombinant cDNA expression libraries of human tumors with autologous serum (SEREX) has demonstrated that antibodies in the serum of patients with cancer can be used to isolate new tumor antigens.3,4 This technique, however, relies on one-step screening without affinity-selection steps and requires a large volume of serum to screen phage clones blotted onto membrane filters. The SEREX approach has limited clinical use, since it is not conducive to the analysis of hundreds of serum samples of small amounts.
By taking advantage of combinatorial screening and high throughput analysis of autoantibody repertoires, we developed a technique that overcomes the disadvantages of SEREX for cancer diagnosis. However, like gene-expression profiling and pattern-recognition approaches with serum proteomics, our method may have the limitations of background signals, sample-selection bias, and limited reproducibility.31 To minimize these problems, immunoreactivity for each phage peptide was measured in relation to an internal control signal detected by antibody against phage capsid proteins. The discriminatory power of autoantibody signatures was validated by reshuffling and analyzing the training and validation sets with the use of various class-prediction models. The reproducibility of the assay was investigated by the use of experimental assays both within and among the arrays. The difference between duplicate peptides in the same array was less than 5 percent for 98 percent of the spots. Analyses of repeated experiments with the use of the same serum samples revealed that the results were very consistent, with a correlation coefficient greater than 98 percent.
The autoantibody signature was detected in only 5 of 14 serum samples from patients who had undergone prostatectomy and in 3 of 11 serum samples from patients with hormone-refractory disease — a finding suggesting that the autoantibody profile is attenuated on removal of the "immunogen" or treatment with antiandrogens, chemotherapeutic agents, or both. Of 30 serum samples from patients with lung cancer, 9 were classified as prostate cancer. This result is in contrast to the more than 80 percent sensitivity for prostate cancer with the use of the phage-peptide system, suggesting that the autoantibody profile has relative specificity for prostate cancer (P<0.001 by the proportion test).
Our results were consistent across a range of clinical and pathological features, including PSA level, Gleason grade, stage, and presence or absence of PSA recurrence (Table 18 of the Supplementary Appendix), with sensitivities and specificities ranging from 80 to 90 percent in discriminating between patients with prostate cancer and controls. This diagnostic performance was maintained in the intermediate ranges of PSA (i.e., 4 to 10 ng per milliliter or 2.5 to 10 ng per milliliter).
Autoantibody signatures may be useful in combination with initial PSA screening; our data show that the 22-phage-peptide detector significantly adds to the diagnostic power of PSA alone (P<0.001). The use of such a "supplementary" autoantibody panel might be important at PSA levels of 10 ng per milliliter or less. This additional discriminatory power could improve the information used to make decisions about biopsy of the prostate.
The sequences in the phage display system are probably those of relatively short peptides rather than full-length proteins, since the cDNAs were enzyme-digested and fragmented before ligation into the phage vector. The average stretch of peptides in the 22-phage-peptide detector was 53±34 amino acids, with a maximum length of 134 residues and a minimum of 11 residues.
Four of the phage clones representing known proteins — BRD2, eIF4G1, RPL13a, and RPL22 — were substantially more reactive with serum from patients with prostate cancer than with that from controls (Figure 5A). Both meta-analysis and immunoblot analysis of tissue extracts suggest that all four proteins are deregulated in prostate tumors. Immunofluorescence of tissue sections and immunohistochemical analysis of tissue microarrays showed that eIF4G1 was overexpressed in prostate-cancer epithelial tissue, as compared with benign epithelial tissue (Figure 5D and Figure 5E, and Figure 5 and Section VI of the Supplementary Appendix).
Of the 22 phage peptides identified in this study, 17 are not present in peptide stretches in known proteins. These 17 peptides may be weakly homologous to known proteins or may have no distinct homology to the primary sequences of known proteins and thus may be "mimotopes" (i.e., stretches of amino acids that mimic an antigen but are not homologous at the sequence level).
We have not tested the phage-microarray system for screening for prostate cancer; this requires extension and confirmation in community-based screening cohorts. It will be important to evaluate the autoantibody signatures associated with prostate cancer in patients with prostatitis, autoimmune conditions, and other diseases. Although the technique is promising, how it will perform in prospective and multiinstitutional studies remains to be determined.
Supported by National Cancer Institute grants (UO1CA111275, to Drs. Chinnaiyan, Rubin, Ghosh, Sanda, and Wei, and P50 CA69568, to Drs. Pienta, Chinnaiyan, Montie, Rubin, Sanda, and Ghosh); National Institutes of Health grants (R01 CA8241901-A1, to Dr. Sanda, and R01GM72007-01, to Dr. Ghosh); a grant (RSG-02-179-MGO, to Drs. Chinnaiyan, Rubin, and Ghosh) from the American Cancer Society; a grant (to Dr. Chinnaiyan) from the V Foundation; grants (W81XWH-04-1-0886, to Dr. Wang; DMAD17-03-0105, to Dr. Sreekumar, and PC040517, to Dr. Mehra) from the Department of Defense; and a grant (5P30CA46592) from the Cancer Center Bioinformatics Core. Dr. Chinnaiyan is a Pew Biomedical Scholar.
We are indebted to Dr. Saravana M. Dhanasekaran for providing samples of prostate-cancer mRNA, to Dr. Paul L. Fox of the Lerner Research Institute in Cleveland for providing anti-RPL13a antibody, to Jason Harwood and Srilakshmi Bhagavathula of the Prostate Tissue and Informatics Core for the collection of serum samples and integration of associated clinical information, to Drs. David Beer and Guoan Chen for providing serum samples from patients with lung cancer as control samples for nonprostate cancer, to Drs. Francesca Demichelis and Jeremy Taylor for providing additional statistical consultation, to Jeffrey Fielhauer for his help with sequence analysis and preparation of the manuscript, and to the staff of the microscopy and image analyses laboratory at the University of Michigan.
Source Information
From the Departments of Pathology (X.W., J.Y., A.S., S.V., R.S., D. Giacherio, R.M., A.M.C.), Biostatistics (R.S., D. Ghosh), Urology (J.E.M., K.J.P., J.T.W., A.M.C.), and Internal Medicine (K.J.P.), and the Comprehensive Cancer Center (A.S., S.V., J.E.M., K.J.P., J.T.W., A.M.C.), University of Michigan Medical School, Ann Arbor; and Beth Israel – Deaconess Medical Center (M.G.S.), Dana – Farber Cancer Institute (P.W.K.), and Brigham and Women's Hospital (M.A.R.), Harvard Medical School — all in Boston.
Drs. Wang and Yu contributed equally to this article.
Address reprint requests to Dr. Chinnaiyan at the Department of Pathology and Urology, University of Michigan Medical School, 1301 Catherine St., MSI 4237, University of Michigan, Ann Arbor, MI 48109, or at arul@umich.edu.
References
Thompson IM, Pauler DK, Goodman PJ, et al. Prevalence of prostate cancer among men with a prostate-specific antigen level 4.0 ng per milliliter. N Engl J Med 2004;350:2239-2246.
Sahin U, Tureci O, Schmitt H, et al. Human neoplasms elicit multiple specific immune responses in the autologous host. Proc Natl Acad Sci U S A 1995;92:11810-11813.
Chen YT, Gure AO, Tsang S, et al. Identification of multiple cancer/testis antigens by allogeneic antibody screening of a melanoma cell line library. Proc Natl Acad Sci U S A 1998;95:6919-6923.
Stockert E, Jager E, Chen YT, et al. A survey of the humoral immune response of cancer patients to a panel of human tumor antigens. J Exp Med 1998;187:1349-1354.
Brichory FM, Misek DE, Yim AM, et al. 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;98:9824-9829.
Minenkova O, Pucci A, Pavoni E, et al. Identification of tumor-associated antigens by screening phage-displayed human cDNA libraries with sera from tumor patients. Int J Cancer 2003;106:534-544.
Zhong L, Peng X, Hidalgo GE, Doherty DE, Stromberg AJ, Hirschowitz EA. Antibodies to HSP70 and HSP90 in serum in non-small cell lung cancer patients. Cancer Detect Prev 2003;27:285-290.
Old LJ, Chen YT. New paths in human cancer serology. J Exp Med 1998;187:1163-1167.
Nilsson BO, Carlsson L, Larsson A, Ronquist G. Autoantibodies to prostasomes as new markers for prostate cancer. Ups J Med Sci 2001;106:43-49.
Mintz PJ, Kim J, Do KA, et al. Fingerprinting the circulating repertoire of antibodies from cancer patients. Nat Biotechnol 2003;21:57-63.
Soussi T. p53 Antibodies in the sera of patients with various types of cancer: a review. Cancer Res 2000;60:1777-1788.
Sreekumar A, Laxman B, Rhodes DR, et al. Humoral immune response to alpha-methylacyl-CoA racemase and prostate cancer. J Natl Cancer Inst 2004;96:834-843.
Luo J, Zha S, Gage WR, et al. Alpha-methylacyl-CoA racemase: a new molecular marker for prostate cancer. Cancer Res 2002;62:2220-2226.
Jiang Z, Woda BA, Rock KL, et al. P504S: a new molecular marker for the detection of prostate carcinoma. Am J Surg Pathol 2001;25:1397-1404.
Rubin MA, Zhou M, Dhanasekaran SM, et al. alpha-Methylacyl coenzyme A racemase as a tissue biomarker for prostate cancer. JAMA 2002;287:1662-1670.
Fernandez-Madrid F, Tang N, Alansari H, et al. Autoantibodies to annexin XI-A and other autoantigens in the diagnosis of breast cancer. Cancer Res 2004;64:5089-5096.
Fu WJ, Dougherty ER, Mallick B, Carroll RJ. How many samples are needed to build a classifier: a general sequential approach. Bioinformatics 2005;21:63-70.
Dhanasekaran SM, Barrette TR, Ghosh D, et al. Delineation of prognostic biomarkers in prostate cancer. Nature 2001;412:822-826.
LaTulippe E, Satagopan J, Smith A, et al. Comprehensive gene expression analysis of prostate cancer reveals distinct transcriptional programs associated with metastatic disease. Cancer Res 2002;62:4499-4506.
Luo JH, Yu YP, Cieply K, et al. Gene expression analysis of prostate cancers. Mol Carcinog 2002;33:25-35.
Luo J, Duggan DJ, Chen Y, et al. Human prostate cancer and benign prostatic hyperplasia: molecular dissection by gene expression profiling. Cancer Res 2001;61:4683-4688.
Singh D, Febbo PG, Ross K, et al. Gene expression correlates of clinical prostate cancer behavior. Cancer Cell 2002;1:203-209.
Welsh JB, Sapinoso LM, Su AI, et al. Analysis of gene expression identifies candidate markers and pharmacological targets in prostate cancer. Cancer Res 2001;61:5974-5978.
Rhodes DR, Yu J, Shanker K, et al. ONCOMINE: a cancer microarray database and integrated data-mining platform. Neoplasia 2004;6:1-6.
Jemal A, Tiwari RC, Murray T, et al. Cancer statistics, 2004. CA Cancer J Clin 2004;54:8-29.
Dennis LK, Resnick MI. Analysis of recent trends in prostate cancer incidence and mortality. Prostate 2000;42:247-252.
Schmid HP, Prikler L, Semjonow A. Problems with prostate-specific antigen screening: a critical review. Recent Results Cancer Res 2003;163:226-231.
Catalona WJ, Smith DS, Ratliff TL, et al. Measurement of prostate-specific antigen in serum as a screening test for prostate cancer. N Engl J Med 1991;324:1156-1161.
Makinen T, Tammela TL, Stenman UH, et al. Second round results of the Finnish population-based prostate cancer screening trial. Clin Cancer Res 2004;10:2231-2236.
Cohen L, Fouladi RT, Babaian RJ, et al. Cancer worry is associated with abnormal prostate-specific antigen levels in men participating in a community screening program. Cancer Epidemiol Biomarkers Prev 2003;12:610-617.
Ransohoff DF. Bias as a threat to the validity of cancer molecular-marker research. Nat Rev Cancer 2005;5:142-149.(Xiaoju Wang, Ph.D., Jianj)