Most prostate cancers are treated, although more than 80% remain clinically insignificant and fewer than 3% are fatal. This retrospective study of 240 radical prostatectomy cases with comprehensive follow-up was a search for reliable markers of prostate cancer prognosis evaluable on biopsy specimens to enable minimization of unnecessary treatment, morbidity, and costs. Representative cancer and benign tissue from each prostatectomy specimen was made into tissue microarrays and stained with antibodies targeting 20 gene sequences. Traditional clinical and pathologic prognosticators and the 20 antibody stains were correlated with patient outcomes. By univariable analysis 4 of 20 antibodies (STMN1/stathmin 1, CYP4Z1/cytochrome p450-4z1, CDH1/E-cadherin, and Hey2), Gleason score, perineural invasion, and apical involvement were statistically significant outcome predictors for biopsy tissue. By multivariate analysis, Gleason score, Hey2, and CYP4Z1 were independently predictive. STMN1 and CDH1 were not independent of Gleason score but remain useful because marker interpretation is objective and Gleason scores often differ for biopsy and prostatectomy specimens.
Prostate cancer is the most common nonskin cancer among men in the United States. There were 2.2 million cases in 2010. More than 200,000 new cases are diagnosed annually.1 The incidence increases with advancing age and increasing life expectancy. More than 80% of prostate cancers are clinically indolent. Only 2.8% of prostate cancers (1 in 36) prove fatal. Before the 1980s, only patients who had signs or symptoms (approximately 20% of all cases) were diagnosed. In the early 1980s, the serum prostate-specific antigen (PSA) test was introduced to monitor patients with prostate cancer for disease progression. By the late 1980s, the serum PSA test was approved by the US Food and Drug Administration as a screening test to detect asymptomatic, early-stage prostate cancers, resulting in an explosive, mass diagnosis of clinically insignificant, “screen-detected,” cases.
Screening PSA is widely accepted as being a relatively poor, nonspecific test, which results in false-positive and false-negative results, necessitating biopsy for patients with abnormal screening PSA results to establish a more definitive diagnosis. Since the prevalence of prostate cancer is so high, biopsies triggered by an elevated screening PSA level detect many occult cancers that may or may not have caused the initial elevated serum PSA level and that may or may not be clinically relevant. The biopsy procedure implemented with screening PSA in the 1980s typically comprised a single core of tissue from each of the 2 lateral prostate lobes. Since the average prostate measures 4.0 × 2.0 × 3.0 cm and each biopsy core is approximately 1 cm long and 0.1 cm/1 mm in diameter, 2 cores represent less than 0.05% of the overall gland volume. Even with such limited sampling, approximately 1 in 6 men were found to have prostate cancer at biopsy. With today’s standard sampling of 12 cores from multiple defined areas of the gland, it is estimated that 1 in 2 men will be diagnosed with prostate cancer during their lifetimes.2
There is currently no single reliable method and no commercially available test to distinguish aggressive from indolent prostate cancer at biopsy. As such, millions of men with clinically insignificant disease are treated and over-treated to avoid undertreating the small minority of potentially life-threatening cases. In the United States in 2010, approximately 48 men were treated for every life saved, at a cost of more than $600 million3,4 and serious side effects, particularly impotence, urinary incontinence, bladder outlet obstruction, cystitis, and prostatitis.
The development of an accurate test predictive of prostate cancer prognosis that can be performed on biopsy tissue, at diagnosis, would enable physicians to reliably distinguish the minority of patients who need and warrant radical therapy for aggressive disease from patients who can be safely “watched,” dramatically reducing unnecessary treatment, complications, and health care costs. Traditional clinical and pathologic indicators of prostate cancer prognosis that clinicians currently rely on to make management decisions (serum PSA level at diagnosis, patient age, race, family history, pathologic Gleason score, TNM stage, apical involvement, and perineural invasion) are widely recognized as nonabsolute determinants of tumor biology since cancers of similar grade and stage in patients of similar age and race, with similar PSA values, often behave very differently.5,6
Advances in molecular cytogenetic research, including genome-wide association studies detecting germline single nucleotide polymorphisms, expression array studies, and linkage analysis, have enabled the detection of genetic mutations that cause and affect prostate cancers. We hypothesize that submicroscopic variations in DNA damage accumulated over time in different prostate cancer cell lines likely determine variable outcomes, despite similar clinical and pathologic features. Most of the early published prostate cancer molecular cytogenetic studies focused on identifying genetic mutations associated with and predisposing to the development of prostate cancer7–20 in an attempt to better understand prostate cancer genesis and to find a more sensitive and specific diagnostic test than the serum PSA. More recent reports in the English literature describe evaluations of whether specific genetic mutations are directly associated with prostate cancer prognosis in an attempt to better define appropriate and case-specific prostate cancer treatment.
A number of molecular events and cytogenetic markers are gaining attention as potential prognosticators for prostate cancer.5,21–30 These include chromosomal copy number variations, particularly chromosome 8q; overexpression of genes such as MYC; proliferation markers such as Ki-67; apoptosis-related proteins such as bcl-2 and MDM2; growth factors such as epidermal growth factor receptor, vascular endothelial growth factor, and transforming growth factor-β1; adhesion molecules such as E-cadherin and CD44; oncogenes such as HER-2/neu; tumor suppressors genes such as p53; transcriptional repressors such as EZH2; PTEN; cyclin-dependent kinases and their inhibitors such as p27, cyclin D1, cyclin B1, p16INK4a, and p21; F-box protein such as skp2; Cox-2; caveolins; zinc-α2-glycoprotein (AZGP1); α-methylacyl coenzyme A racemase; MCM7; MMP-2; MMP-9; AR; cathepsin-D; hepsin and PIM1; laminin receptor; and the TMPRSS2/ERG fusion gene. Few, however, have demonstrated clinical usefulness, and none has, as yet, evolved into an established, commercially available, practical test that can reliably determine disease behavior at the time of diagnosis.31,32
The purpose of this study was to search for reliable molecular markers of prostate cancer prognosis that can accurately stratify patients for risk (and ultimately dictate appropriate therapy) with greater reliability than, and independent of, traditional clinical and pathologic prostate cancer prognostic indicators. Markers were evaluated using labeled proprietary antibodies directed against the transcription-translation protein products of gene sequences on tissue microarrays. Antibodies were selected over genetic sequencing because of the greater cost-effectiveness and availability of immunostains compared with genetic testing in most community hospital tissue laboratories, the shorter turnaround time, and the ability to visually identify the targeted cell population. Furthermore, studying protein products of gene sequences rather than DNA sequences themselves avoids the distraction of mutations and pathways that do not translate into biologic effects.
Materials and Methods
Accreditation and Compliance
The study was approved by the Sharp Healthcare Institutional Review Board, San Diego, CA, and was performed in full compliance with all ethical and federal standards for human subject research.
This was a retrospective study of men with prostate cancer treated with radical prostatectomy, with or without additional adjuvant therapy, as clinically appropriate. Cases were excluded if the patients had received any nonsurgical prostate cancer treatment (such as hormone deprivation, radiation, or chemotherapy) before radical prostatectomy. For inclusion, all prostatectomy glass slides and tissue blocks had to be available for review, and patients had to have undergone comprehensive clinical follow-up. Follow-up was determined to be comprehensive if all postoperative PSA results were available at the standard intervals of testing for at least 5 years from surgery in cases that remained disease-free following prostatectomy, or for as long as possible, but including intervals less than 5 years, in cases that progressed following surgical treatment. Standard intervals for postprostatectomy serum PSA testing were defined as serial PSA tests every 3 months for the first year, every 6 months for the second year, and annually for life thereafter, based on protocols followed by the study patients’ primary care physicians, urologists, and oncologists.
Cases were identified by searching the Tumor Registry data of Sharp Healthcare, San Diego, for patients with prostate cancer initially treated with radical prostatectomy at Sharp Grossmont Hospital, La Mesa, CA, or Sharp Memorial Hospital, San Diego. Each case identified was assigned a study case number in the sequence of accrual. All glass slides and paraffin blocks for each identified case were retrieved from archives. All patient identifiers were removed and replaced with the study case number to maintain integrity of protected health information.
Dates of Cohort Prostatectomies
Since the hospital laboratories involved in this study store slides and blocks for only 10 years pursuant to federal regulations, tissues from prostatectomy cases before 1999 were no longer available when this study began in 2009. To meet the inclusion criterion of a minimum of 5 years follow-up for indolent cases, prostatectomies performed after December 2005 were not included, except for sporadic cases having aggressive outcomes.
Clinical Follow-up Data
Follow-up information was obtained by reviewing all available patient records in multiple databases, including hospital and physicians’ office medical records, telephonic interviews with patients’ physicians, and, when necessary, patients themselves. For each case, patient age at diagnosis, race, highest preoperative PSA level, date of diagnosis, date of prostatectomy, dates and values of all postoperative PSA tests, types and dates of any adjuvant treatment following prostatectomy, response to such adjuvant therapy, and date of last available follow-up were recorded on a data collection form designated “Data Sheet I.” Follow-up intervals were calculated from date of prostatectomy to last encounter. All clinical follow-up data for all cases were recorded cumulatively on an Excel spreadsheet (Microsoft, Redmond, WA).
Classification of Clinical Outcomes
Clinical outcomes were classified into 4 categories as follows, based on discussions with urologists, medical oncologists, and radiation oncologists: (1) disease-free, all postsurgical serum PSA levels at the expected intervals of follow-up remained at 0.1 ng/mL (0.1 μg/L) or less; (2) biochemical failure, PSA level more than 0.1 ng/mL (0.1 μg/L) but less than 0.4 ng/mL (0.4 μg/L); (3) local recurrence, PSA level 0.4 ng/mL (0.4 μg/L) or greater, and/or palpable or radiographic mass in the prostatic bed; or (4) distant metastasis, radiographic or biopsy-proven prostate cancer outside the pelvis.
Clinical Outcome Events Evaluated
Patients were determined to be disease-free or to have progressive disease. For patients with progressive disease, the disease event (biochemical failure, local recurrence, or metastasis) and the time to first onset of the disease event were documented.
Classification of Response to Postoperative Adjuvant Therapy
For patients who received postsurgical adjuvant therapy (radiation, hormones/hormone-deprivation, and/or chemotherapy), response to adjuvant therapy was classified into 3 categories, as follows: (1) complete, posttherapy serum PSA levels reverted to and remained at 0.1 ng/mL (0.1 μg/L) or less; (2) partial, PSA level temporarily dropped to 0.1 ng/mL or less but increased again and/or PSA level dropped following adjuvant treatment but never returned to 0.1 ng/mL (0.1 μg/L) or less; or (3) failed, continued increase in PSA level despite adjuvant therapy.
All prostatectomy specimens included in this study were processed in 1 of 2 Sharp Healthcare histology laboratories, Sharp Grossmont and Sharp Memorial hospitals. All Sharp Healthcare laboratories use the same protocols and manuals for specimen handling and processing. As such, radical prostatectomy specimens in our study were examined grossly in a consistent manner, and sampling was uniform irrespective of the site of specimen processing. For each case, all prostatectomy slides were reviewed by 1 experienced surgical pathologist (S.M.) to ensure uniform evaluation, avoiding interobserver variability. The histologic features evaluated included Gleason grade and score, all elements of the pathologic TNM stage (as determined by using the AJCC Staging Manual33), capsular invasion, apical involvement, and perineural invasion. For each study case, these data were recorded on a form designated “Data Sheet II.” All pathologic data for all cases were recorded cumulatively on an Excel spreadsheet.
Selection of Tissue for Microarray Preparation
For each case, a representative area of the highest Gleason grade cancer and a representative area of benign prostate were identified and labeled on glass slides. The paraffin tissue blocks corresponding to the selected slides were used to make tissue microarrays.
Cancer and benign tissue cores of 0.6 mm diameter (identified as cores “A” and “B,” respectively) were punched from each selected donor tissue block using a microarray apparatus, inserted into recipient paraffin block channels, and identified by case number to create tissue microarray blocks. Each microarray block contained cores from a maximum of 100 cases arranged in a 10 × 10 array format with a position orientation core of placental tissue on 1 corner. Microarray maps were constructed to enable identification of the cores in each microarray.
Microarray Slide-Antibody Incubation
For the study, we selected 20 proprietary antibodies directed against the protein products of 20 targeted gene sequences thought to be potential prognostic indicators for prostate cancer, based on previously reported protein translation of genes expressed in breast cancer.34 For each of the antibodies, optimal dilutions were selected by testing a small titer tissue array of prostate cancers with positive and negative antibody controls. Dilutions were determined as optimal when staining was robust, reproducible, and consistent. Each antibody, at optimal dilution Table 1 and Table 2, was incubated with tissue slides cut from each microarray block such that every antibody was tested on every tissue core.
Tissue microarray sections were deparaffinized and dehydrated by submerging 3 times in xylene for 10 minutes each, followed by rinsing 3 times in 100% ethanol and twice in 95% ethanol, and then microwaving the tissue, and boiling for 11 minutes in 10 μmol/L buffered citrate (pH 6.0). Slides were cooled to room temperature and rinsed in distilled water followed by phosphate-buffered saline and dipped in 0.03% hydrogen peroxide followed by rinsing with phosphate-buffered saline. They were then stained with dilutions of antibodies in DAKO diluent (DakoCytomation, Carpinteria, CA) for 1 hour at room temperature. Secondary antibody was applied for 1 hour. Staining was visualized using the EnVision staining kit (DakoCytomation) according to the manufacturer’s instructions.35 The appropriate location of staining in the cell was defined for each antibody (Table 2).
Stains were interpreted using a 4-tier scoring system as follows: 0, 0% to less than 10% staining in appropriate tissue in an appropriate location; 1, no target tissue present (unsuccessful core); 2, weak positive staining in greater than 10% of appropriate tissue in an appropriate location; or 3, strong positive staining in greater than 10% of appropriate tissue in an appropriate location.
Examples of how antibodies were scored, based on staining presence, location, and intensity, are shown in Image 1, Image 2, Image 3, and Image 4.
Microarray core, study core 261A. Negative immunostain result (score 0) (Hey2/S0456, ×200).
In the final data analysis, staining scores were converted to a binary scale of positive or negative (or no information, if instructive tumor tissue was absent), by grouping scores of 2 and 3 into 1 category, “positive staining,” and keeping the original 0 score category representing “negative staining.” This approach allowed for an objective and reliably reproducible “black and white” evaluation of marker status, minimizing interobserver and intraobserver variability.
For all final statistical analyses, positive stains were given a numeric score of 1 and negative stains, a numeric value of 0.
Markers present in the paired benign and malignant patient samples were compared by using the McNemar test with the continuity correction (2-tailed). The clinical and pathologic indicators assessed and the scores of the 20 antibody stains for each study case were correlated with patient outcomes by using Cox analyses of outcome incidence in univariable and multivariable analysis. Results were considered statistically significantly predictive when the P value was less than .05. Hazard ratios (e^ coefficients) were calculated to assess how and to what degree indicators were predictive of outcome. The greater the deviation of the e^ coefficient from 1.0, the greater the likelihood of an outcome event occurring. A positive e^ coefficient (>0.0) correlated with an increased risk of event outcome, and a negative e^ coefficient (<0.0) correlated with a decreased risk of event outcome. Kaplan-Meier estimates were constructed for antibody markers found to be significantly predictive by Cox analyses. The Kaplan-Meier estimate figures show the proportion of outcome-free cases at various time points. Predictability of adjuvant treatment failure for each indicator was calculated by using logistic regression.
Of the 20 antibodies, 4 were found to be statistically significantly predictive of outcome in univariate analysis. A 4-antibody model was constructed to demonstrate risk for varying outcomes, based on results of these 4 antibodies evaluated together, and Kaplan-Meier curves for outcomes based on results of the 4-antibody model were evaluated. The 4-antibody model is a statistical model for predicting the likelihood that a patient will be outcome/event-free at any given time from date of prostatectomy. The model used in this study is based on models that have proven successful in reliably predicting odds of outcome events in other studies.34,35 Each score was given a value determined by using the terms given to the scores for negative or positive staining (0 or 1, respectively) in the univariate Cox model for each marker. This approach allows each marker to be individually weighted and ignores any nonindependence of the marker. The sum of the 4 scores for each individual case determines the overall risk.
The English literature was searched for articles from 1990 to 2011, with focus on key words “prostate cancer,” “molecular markers,” and “prognosis.” Articles before 1990 were included if referenced elsewhere or highly regarded. The search also included publications in reference lists of these articles and online Web pages, as indicated.
We found 304 cases of radical prostatectomy performed for prostate cancer at Sharp Grossmont and Sharp Memorial hospitals during the targeted study period from 1999 onward that had archival tissue blocks and slides available for review. The slides of all 304 cases were reviewed, the selected paraffin blocks were processed into tissue microarrays, and the micro-array slides were stained with the 20-antibody panel. Five tissue microarrays were constructed from the paired benign and malignant tissue cores from these 304 cases. Comprehensive clinical follow-up information could be obtained for 240 of the 304 cases. These 240 cases constituted the final study cohort.
Patients ranged in age from 50 to 87 years, with a mean of 70 years. Of the 240 patients, 156 (65.0%) remained disease-free following surgery for follow-up periods ranging from 5 years (60 months) to 11 years, 10 months (142 months), with an average follow-up duration of 7 years, 7 months (91 months). In 84 (35.0%) of the 240 cases, disease progressed. All 84 cases that progressed had at least biochemical failure; 15 had only biochemical failure. In 69 of the 84 cases, local recurrence additionally developed, and 24 of these 69 cases progressed to distant metastasis. The follow-up interval for the progressive group ranged from 3 months to 10 years, 10 months (130 months), with an average follow-up period of 6 years, 3 months (75 months). No patient in this study died of prostate cancer during the follow-up period. Several died of other causes.
The Gleason grades and scores ranged from a low Gleason score of 2 + 2 = 4 to a high of 5 + 5 = 10. Most were scores of 6 and 7, grades 3 and 4. The preoperative serum PSA values ranged from a low of 1.1 ng/mL (1.1 μg/L) in a patient operated on for a palpable “suspicious” nodule to a high of 61.4 ng/mL (61.4 μg/L).
Of the 84 progressive cases, 61 patients had adjuvant therapy, including radiation and hormone/hormone-deprivation therapy for biochemical failure and/or local recurrence and docetaxel with or without mitoxantrone chemotherapy for metastasis. Of the 61 treated cases, 27 had a complete response to adjuvant therapy, and 22 had a partial response; in 11 cases, therapy failed. In 1 case the response was unknown.
As shown in the gene/marker cluster diagram Figure 1, benign tissue microarray samples clustered together and malignant samples clustered together with respect to the molecular markers/gene sequences they express. The benign and malignant samples separate very cleanly, as demonstrated by the blue bar on the right. This adds confidence that the results of the molecular profiling are valid. Within the malignant sample group, the markers that showed the greatest variability are shown in Figure 2.
By univariate analysis of the clinical and pathologic indicators of prognosis Table 3, Gleason score, primary and secondary Gleason grades, highest Gleason grade, pathologic T stage, TNM stage (AJCC Staging Manual33), and preoperative PSA were all statistically significant predictors of time to first failure, time to biochemical failure, local recurrence, and distant metastasis. Perineural invasion was significantly predictive of time to first failure, biochemical failure, and local recurrence. Apical involvement was significant in predicting time to first failure and biochemical failure. Positive margins were statistically predictive of biochemical failure, local recurrence, and metastasis. Patient age at diagnosis and tertiary Gleason grade (other than highest Gleason grade) were not predictive of outcome.
By univariate analysis, 4 of the 20 antibodies showed statistically significant outcome predictability Table 4. Positivity of staining for STMN1 was statistically significant in predicting biochemical failure. Positivity for Hey2 was statistically significant in predicting time to first failure, biochemical failure, local recurrence, and distant metastasis and in predicting “untreatable” disease (failure to respond to adjuvant therapy). Absence of CYP4Z1 (negative staining) was statistically significant in predicting time to first failure and biochemical failure. Negative staining for CDH1 was significant in predicting local recurrence.
For each marker that proved to be statistically significant in determining an outcome event, Kaplan-Meier estimate curves were constructed for each outcome. Similarly, curves were constructed for the cumulative 4-antibody model. The Kaplan-Meier estimates show the proportion of patients with positive and negative marker stain results who will likely be outcome-free at any point in time. These results are shown in Figure 3, Figure 4, Figure 5, Figure 6, Figure 7, Figure 8, Figure 9, Figure 10, Figure 11, Figure 12, and Figure 13.
Marker/gene sequence cluster diagram for all tissue cores, benign and malignant, for all 304 cases studied. Gene sequences/molecular markers are represented along the x-axis (in the vertical rows). Individual tissue cores are represented along the y-axis (in the horizontal rows). Red indicates increased marker expression compared with other markers in the same case. Green indicates decreased marker expression compared with other markers in the same case. Black is the midpoint of expression. Gray indicates no tissue for the case. Cores with similar marker expression cluster together. Benign and malignant cores separate neatly, as demonstrated by the moving bar graph on the right. This adds confidence that the markers are reliable.
In multivariate analysis considering the pathologic indicators evaluable on biopsy (Gleason score, perineural invasion, and apical involvement) and the 4 markers statistically significant by univariate analysis, Gleason score was independently significant in predicting time to first failure, biochemical failure, local recurrence, metastasis, and treatment failure; Hey2 was independently significant in predicting time to first failure, biochemical failure, local recurrence, distant metastasis, and treatment failure; and CYP4Z1 was independently significant in predicting time to first failure and biochemical failure Table 5.
Kaplan-Meier estimate curve demonstrating distant metastasis–free probability for a 4-antibody model. The 4 antibodies were stathmin 1, cytochrome p450-4z1, Hey2, and E-cadherin.
Antibodies S0456, EP1972-1, S0725M, and S5073, directed against the protein products of genes Hey2, STMN1, CYP4Z1, and CDH1, respectively, are statistically significant predictors of prostate cancer prognosis. Hey2 and STMN1, when present/positive, predict disease progression and aggression. Positivity for CYP4Z1 and CDH1 predict more indolent behavior. Hey2 and CYP4Z1 are independently predictive of all other indicators evaluable on biopsy. CDH1 and STMN1 are not independent of Gleason score but are still potentially clinically useful as Gleason is subjective, whereas marker interpretation is objective. Furthermore, Gleason score on biopsy is often not fully representative of the final Gleason score at prostatectomy.
These 4 markers hold promise for a future reliable test, which can be performed on biopsy tissue, for prostate cancer prognosis and therapeutic response to separate patients with aggressive disease who need treatment from the vast majority with indolent disease who can be safely watched. This testing will dramatically reduce overtreatment, treatment-related comorbidities, and health care costs. Hey2 holds particular promise as a marker of aggressive disease resistant to all forms of postsurgical adjuvant therapy. This raises the possibility of Hey2 performing a prognostic (and subsequent therapeutic) role in prostate cancer management, similar to that of HER-2/neu in breast cancer. These findings warrant a large-scale validation study to determine the clinical usefulness and reproducibility of Hey2, CDH1, CYP4Z1, and STMN1 as reliable molecular cytogenetic markers of prostate cancer prognosis and therapeutic response. The long-range potential of such research includes the production of a commercially available, cost-effective, reliable, prognostic, antibody kit and, ultimately, monoclonal antibody therapies for detection and treatment, respectively, of aggressive, potentially fatal, prostate cancer.
We express our deepest gratitude to the Grossmont Hospital Foundation, the partnership of Sharp Healthcare; the staff at Sharp Grossmont and Sharp Memorial histology laboratories; Clarient; Anna Landin, tumor registrar, Sharp Grossmont Hospital; Rebecca Owens, Clarient; and numerous urologists, oncologists, physicians, and staff who allowed access to patients’ records.
Supported in part by funding from the Grossmont Hospital Foundation.