Quantitative Fluorescence Imaging Analysis for Cancer Biomarker Discovery: Application to β-Catenin in Archived Prostate Specimens

The availability of structural and functional data for thousands of cellular proteins (1) has inspired widespread interest in translational studies aimed at protein expression profiling of clinical diseases and their antecedent states. A key to the implementation of this translational research is the development of techniques that offer sensitive, continuous quantification of proteins in clinical specimens. Gene microarray studies have stimulated the development of protein profiling techniques such as protein microarray analysis (2), biological mass spectrometry (3), and tissue-based fluorescence analysis (4, 5), which are applied to small tissue samples, e.g., biopsies. The application of gene microarrays to molecular profiling of cancers provided the first glimpses into the remarkable molecular heterogeneity of cancer among patients and among tumors in the same patient, and identified cancer subclasses defined by common expression patterns (6). In relation to cancer protein profiling, gene microarrays have identified numerous proteins, the expressions of which may be modified (7-10). Protein networks directly carry out cellular functions, and dysregulation of these networks via altered protein expression or posttranslational modification underlies disease.

Reverse-phase protein array analysis (RPPA; refs. 2, 6, 11-13) and automated quantitative analysis (AQUA; refs. 4, 5, 14) are two recently introduced methodologies that precisely quantify proteins and their secondary modifications in clinical specimens, i.e., slide-mounted tissues. RPPA was developed with a focus on modifications of cell-signaling cascades in disease, and simultaneously analyzes multiple lysates of specimens from one or more patients. Nanoliter volumes of lysate are arrayed onto nitrocellulose slides and probed with a primary antibody detected with a fluorophore-conjugated secondary antibody. Protein array analysis exhibits high sensitivity, precision, and linearity across a wide range of protein concentrations. AQUA was developed with a primary focus on rapid and precise quantification of protein expression in tissue microarrays (TMA) for large-scale studies of disease outcome. AQUA entails mapping subcellular compartments of tissue specimens by fluorescence antibody labeling, and assignment of the fluorescence signal from biomarkers of interest to specific compartments. The assay has produced remarkable improvements over pathologist classification of immunohistochemical staining in the prediction of population-based disease outcomes. The present study introduces a novel quantitative fluorescence imaging analysis (QFIA) procedure for the rapid and precise quantification of proteins and their secondary modifications in the context of conventional histologic features of slide specimens. The procedure complements RPPA and AQUA.

Diagnosis of prostate cancer (PC), the second leading cause of cancer deaths among U.S. men (15), is dependent on the quantification of serum prostate-specific antigen. A prostate-specific antigen test of 4 to 10 ng/mL, however, has a specificity of only 25% (16), and requires biopsy, which has a false-negative rate of 20% (17), necessitating the rebiopsy of a large majority of these men. A potential solution to this problem is based on the concept of biomarker alterations associated with premalignant field disease and cancer-induced field effect (18-20). Premalignant field disease is defined as genetic and/or epigenetic modifications of morphologically normal–appearing cells that herald an increased probability of neoplastic transformation (18, 20). In response to continued exposure to carcinogenic insult, cells at multiple foci are initiated (polyclonal disease) and undergo progression toward PC (19). In contrast, the field effect is defined as the presence of epigenetic changes in cells outside of the tumor nodules, in response to altered signaling from cancer cells (18, 20). Cells within the premalignant field or the area of field effect exhibit a normal morphology but express phenotypic modifications that may be detected as qualitative or quantitative biochemical changes (18, 20). Consequently, it may be possible to identify patients with early cancer on the basis of a positive cancer biopsy or a morphologically normal–appearing biopsy expressing biochemical field disease/effect.

An objective of the present study was to evaluate the adhesion/signaling protein β-catenin as a potential field marker for PC. β-Catenin is a 92-kDa protein encoded by a gene localized to human chromosome 3p22 (21). The protein expresses both cell adhesion and transcriptional regulation functions that are independent of each other. Impaired β-catenin degradation with increased cellular concentration of the protein contributes to the inappropriate activation of transcription factors and uncontrolled cell proliferation. Transmembrane cadherin molecules, the prime mediators of calcium-dependent cell-cell adhesion, express an intracellular catenin-binding domain (22). Binding of β-catenin or γ-catenin to this domain is required for the linkage of E-cadherin to actin filaments of the cytoskeleton. Deletion of the catenin-binding domain of E-cadherin or functional inactivation of catenins disables cell-cell adhesion even though E-cadherin retains its extracellular binding domain. In the prostate, β-catenin seems to act primarily as a cell adhesion molecule rather than a transcription factor (23). The cell-cell adhesion complex is compromised in cancer-bearing prostate glands, as evidenced by the marked down-regulation of cadherins in the cancerous epithelium (24).

Work presented in this report addressed two integrated objectives: (a) the validation of a new method, tissue-based QFIA, for fluorescence quantification of β-catenin in slide specimens of archived prostate tissues that were fixed in formaldehyde-based fixative, and (b) the evaluation of β-catenin expression as a potential field marker for PC, using archived prostate tissue specimens.

Archived Core Biopsies. The surgical pathology database of the Veterans Affairs Hospital at Omaha was searched for archived cancerous core biopsies and noncancerous biopsies of patients who were followed for at least 5 years and remained without clinical PC. Patients without clinical cancer (controls) were matched one-to-one with cancer patients on the basis of age (±5 years) and year of biopsy (±3 years). The latter was expected to minimize potential analysis variability due to differences in tissue fixation conditions or storage time. Selection of patients without clinical cancer was random with respect to noncancerous prostate conditions, i.e., no effort was made to select a particular noncancerous condition, e.g., prostatitis or benign prostatic hyperplasia. All procedures were in compliance with human studies protocols approved by the Institutional Review Board of the institution. Fifty matched pairs were identified and their paraffin blocks were retrieved from the archives. All tissue cores had been fixed in 10% neutral buffered formalin under unspecified conditions. Slides were prepared with 4-μm sections of each block containing three to six cores, and then one slide per paraffin block was stained with H&E. Two pathologists (N.A. Abrahams and D. Huang) evaluated the H&E-stained slide specimens to determine the suitability of each specimen for the study. The criteria for inclusion of a PC specimen were (a) the presence of both noncancerous and cancerous epithelium, (b) a minimum of ∼30 noncancerous acini (the large majority of specimens contained 70-100 acini), (c) cancerous areas representing at least 25% of the biopsy specimen, and (d) cancerous epithelium representing ∼80% or more of the cancerous areas. The pathologists identified two types of specimens from patients without clinical cancer, for the present study, those with a predominant glandular hyperplasia (19 specimens) and those with no apparent pathologic lesions (23 specimens). Each of these specimens contained from 30 to more than 200 acini, with the large majority containing 70 to 100 acini. Damaged specimens exhibiting cracks, fissures or fragmentation, or poor H&E staining were excluded from the study because of the possibility of labeling artifacts. Excluded specimens represented no more than 2% to 3% of the specimens retrieved from the archive. Clinical and demographic data, excluding patient identifiers, were obtained from patient records and associated with each anonymous specimen.

Core biopsies from 42 PC cases and matched controls were included in the present study for analysis of β-catenin expression. Clinical and demographic data are summarized in Table 1. At biopsy, the mean age of the cases was 67 years (range, 55-77 years) and the mean age of the controls was 67 years (range, 57-78 years); the age difference for each matched pair did not exceed 5 years. Biopsies were taken in the interval from 1991 to 1998; the difference for year-of-biopsy of each matched pair did not exceed 3 years. All control subjects were considered to be without clinical cancer at the time of biopsy because clinical follow-up including prostate-specific antigen determination 5 to 11 years later was negative for PC.

Prostatectomy Specimens of Consented Subjects for the Evaluation of Tissue Fixation Procedures and for the Preparation of Analytic Standards. In compliance with Institutional Review Board–approved protocols, tissue sections and fresh biopsy cores were collected from prostate glands of consenting subjects who had undergone prostatectomy at the University of Nebraska Medical Center or the Veterans Affairs Hospital at Omaha. Tissue specimens were collected from two benign hyperplastic (BH) glands with a predominant glandular hyperplasia and one cancer-bearing gland to evaluate different fixation procedures for QFIA, according to a standard operating procedure established by the research laboratory. Tissue blocks from one BH gland were used as standards in each analytic run. Tissue sections of 200 to 300 mg were taken from the right and left sides of the gland and transferred directly into the fixative. Sections were fixed with 4% EM grade formaldehyde, zinc-based Streck tissue fixative (Streck Laboratories, Inc.), or methacarn fixative (25) for 12, 24, 48, and 72 h. Tissues were embedded in paraffin, sectioned at 4 μm, mounted to glass slides and processed for analysis of the β-catenin fluorescence signal. Optimum fixation times yielding the brightest fluorescence signal were 24 h for Streck fixative and 48 h for formaldehyde and methacarn. Slide specimens from one BH gland fixed for 48 h in 4% formaldehyde were used as analytic standards.

Serial Sections of Prostatectomy Specimens for Comparison of β-Catenin Expression Determined by QFIA and RPPA. Validation of quantitative imaging analysis of β-catenin was implemented with 4-μm sections of prostatectomy specimens collected from two BH glands with a predominant glandular hyperplasia and eight cancer-bearing glands, which were fixed in methacarn. These previously screened cases were selected to provide a wide range of β-catenin expression. Serial sections of each specimen were mounted onto glass slides and pairs of adjacent sections were selected for study. One section of each pair was analyzed by QFIA and the matching section was analyzed by RPPA. A more distant section of each case served as an isotype control for QFIA.

A TMA for the Comparison of β-Catenin Expression in Methacarn-Fixed and Formaldehyde-Fixed Tissues as Determined by QFIA. A TMA was prepared with sections taken from three BH glands with a predominant glandular hyperplasia and seven cancer-bearing glands. Two sections were collected from each gland; one section was fixed in methacarn and the second was fixed in 4% EM grade formaldehyde (as described above). All sections were embedded in paraffin and 1.0 mm cores were punched and transferred to the blank paraffin block. Two cores each were taken from acini-rich areas, when possible, of formaldehyde-fixed and methacarn-fixed sections from each BH gland. For the cancer cases, pairs of cores were taken from areas rich in cancerous epithelium and from areas rich in noncancerous acini, again, when possible. The TMA block was sectioned at 4 μm and mounted onto glass slides. Expression of β-catenin in the tissue discs of duplicate slide preparations was analyzed by QFIA. The array permitted direct, cross-fixative comparisons of the acini of the two BH glands, the noncancerous acini of five cancer-bearing glands, and the cancerous epithelium of one cancer-bearing gland.

LNCaP Cell Standards for Quantitative Analysis. LNCaP cells (clone FGC of unknown passage) purchased from the American Type Culture Collection were cultured in RPMI medium supplemented with 10% heat-inactivated bovine serum and passaged every 5 to 7 days. Cells at passages 10 to 15 were cryopreserved in Origen Freeze Medium (Fisher Scientific). Cryopreserved cells were seeded to T-flasks, grown to 70% coverage, harvested, washed in Dulbecco's PBS with Ca⁺² Mg⁺² (PBS), and suspended at 5 × 10⁴ cells/mL PBS. Higher cell concentrations were associated with clumping during fixation and yielded poor slide specimens. EM grade formaldehyde (Polysciences, Inc.) was added to 0.5% and the suspension was fixed for 15 min at room temperature. Subsequently, the suspension was diluted with an equal volume of QFIA FIXIT (26), held overnight at 4°C and then dispensed to 4 mL cryovials and stored at −80°C. One to 3 days prior to fluorescence labeling, cryopreserved suspensions were thawed and captured to 25 mm polycarbonate filters (5 μm pores, Nucleopore; Fisher Scientific), with a glass filtration unit (Fisherbrand; Fisher Scientific). The captured cells were washed with cell adherent fluid followed by modified Saccomanno fixative and then blotted to OptiPlus barrier slides (BioGenex Laboratories). Blotted cells were sprayed with Carbofix-E (StatLab Medical Products), dried for 15 min, and the slides were stored at −20°C for at least 1 day and not more than 2 weeks, prior to use.

Analysis of slide-mounted prostate specimens was implemented as described by Paweletz et al. (11). Slide specimens to be extracted were stained with Mayer's hematoxylin and reviewed by light microscopy for structural integrity. The entire section was shaved off the glass slide using a single-edged razor blade, and deposited into 200 μL of extraction buffer consisting of a 1:1 mixture of 2× Tris-glycine SDS sample buffer and tissue protein extraction reagent (Pierce Biotechnology) plus 2.5% β-mercaptoethanol. The sample was heated for 2 h at 70°C in a heat block (Eppendorf Thermomixer 5436) and then sonicated (Branson 2110) for 10 min and heated to 95°C for 8 min. The sample was centrifuged (Fisher Micro-Centrifuge model 235V) for 5 min at maximum speed and the supernatant was stored at 5°C for a maximum of 48 h, prior to spotting onto nitrocellulose slides.

On the day extracts were arrayed, 2-fold dilutions (1 of 2 to 1 of 128) were prepared and 70 μL of undiluted sample and each sample dilution was transferred to mapped wells of a 384-well V-bottomed plate made of polypropylene (Whatman Inc.). Approximately 50 nL of each sample was spotted in triplicate onto nitrocellulose-coated glass slides. Undiluted and 2-fold serial dilutions (1 of 2 to 1 of 128) of a mouse IgG (eBioscience) solution (2.5 μg/mL) were spotted on the same slides as a quality control and reference standard. Protein arraying was implemented with an 8-pin arrayer (VP478) according to the instructions of the manufacturer (V&P Scientific, Inc.). Arrayed nitrocellulose slides were placed in a light-tight box that contained desiccant and stored at −20°C for no more than 5 days prior to protein quantification.

One day prior to antibody labeling, the protein arrays were treated with Re-Blot antibody stripping solution (Chemicon) for 15 min at room temperature, washed twice for 5 min in PBS, and then incubated overnight in blocking solution [1.2 g I-Block (Tropix) and 600 μL Tween 20 (Sigma) dissolved in 600 mL Dulbecco's PBS] at 4°C with constant rocking. Arrays were labeled with mouse anti–β-catenin primary antibody (clone 6F9); generously provided by Dr. Margaret Wheelock (Nebraska Center for Cellular Signaling, College of Dentistry, University of Nebraska Medical Center), for 2 h at room temperature and then washed twice for 10 min with TBS-T solution. Control slides were treated with mouse IgG in place of the primary antibody. All slides were treated for 30 min (in the dark, at room temperature) with IRDye 800 CW-conjugated goat anti-mouse IgG (LI-COR Biosciences) at a dilution of 1:2,500. The slides were washed twice for 10 min with TBS-T and air-dried.

Slides were scanned with the Odyssey IR imaging system (LI-COR Biosciences) at a resolution of 84 um, a sensitivity of 7.0, and a background setting of “medium.” The images were analyzed with the Odyssey software. Integral fluorescence intensity, corrected for background, was determined for each spot and the mean of each set of triplicate spots was determined.

General Procedures. Slide specimens of paraffin-embedded cores and tissue sections were deparaffinized with EZ-AR Common solution (BioGenex Laboratories) and treated for epitope recovery with EZ-AR1 (BioGenex Laboratories) according to the specifications of the manufacturer. Specimens were processed with a computer-controlled, microwave system (EZ-Retriever; BioGenex Laboratories). Slides were rinsed with tap water and then partially dried for application of labeling barriers with a PAP pen. Labeling was completed with the BioGenex autostainer optimized to obviate slide batching, which was associated with ∼20% loss of β-catenin signal. Moist slides were loaded into the autostainer programmed to soak slides for 15 min with Super Sensitive Wash Buffer, blocked for 20 min with 10% normal goat serum (Zymed Laboratories, Inc.), labeled for 1 h with mouse monoclonal antibody (1° Ab) specific for β-catenin (clone CAT-5H10; Zymed Laboratories) and then labeled for 1 h with goat anti-mouse IgG antibody (2° Ab) coupled with Alexa Fluor 568 (Molecular Probes, Inc.). This two-antibody system superseded an avidin/biotin labeling system that produced higher backgrounds and lower signal-to-noise ratios. Quantitative fluorescence labeling requires the saturation of epitopes by their corresponding antibodies. Saturating dilutions of the 1° Abs were determined by titration against β-catenin in LNCaP cells and slide preparations of a standard BH gland. The optimum dilution for saturation labeling with primary antibody was 1:100 for tissue sections and cell lines. The optimum dilution for labeling with the 2° antibodies was also 1:100. Negative controls were treated with a nonimmune mouse IgG isotype (eBioscience) adjusted to an immunoglobulin concentration the same as that of the diluted 1° Ab. The autostainer program ended by rinsing the slides with Super Sensitive Wash Buffer. All labeling procedures were completed at room temperature. Labeled specimens were mounted with ProLong Gold antifade reagent (Molecular Probes, Inc.), sealed with clear lacquer, and stored at −20°C. Evaluation of ProLong Gold antifade reagent showed no detectable photobleaching with repeated exposures of individual specimens to excitation energy, or loss of fluorescence signal during storage at −20°C for up to 2 months. Other antifade reagents including N-propylgallate (Sigma-Aldrich Co.; ref. 26) and SlowFade Antifade (Molecular Probes, Inc.) were associated with the loss of fluorescence signal due to photobleaching or storage of the slide specimens.

Archived Core Biopsies of 42 Cases and Matched Controls. Each of six sets of seven pairs of matched core specimens (a total of 42 pairs of cases and matched controls) were incorporated into two separate analyses for β-catenin labeling. The first analysis included duplicate slides of seven cancer cases and seven matched controls, six slide preparations of the tissue standard, and six slide preparations of the LNCaP cell standard (a total of 40 slides). Pairs of standard slides were distributed evenly among pairs of the core specimens in the autostainer slide racks. One slide of each pair of cores and standards served for labeling β-catenin and the other served as an isotype control. The second analysis, completed on a different day, was a replica of the first analysis, providing quantification of β-catenin in duplicate slide specimens of each case and matched control. Repeat analyses of the remaining five sets of paired cases and controls were completed according to the same design, with one change. Two cases and two controls from the set of seven pairs of specimens previously analyzed were included as another level of quality control. In the event that fluorescence analysis of the “carry-over” specimens or the standards changed by 15% or more, it would be necessary to repeat the assay and correct potential problems. Repeat assays were not required in the present study.

Instrumentation. Quantifiable images of fluorescent slide specimens were generated with a fully automated Leica DMRXA2 microscope (North Central Instruments) equipped with a Marzhauser eight-slide scanning stage (North Central Instruments) and a 150-W mercury/xenon (Hg/Xe) excitation lamp (model E7536; Hamamatsu Corp.) that offered stable illumination, essential for fluorescence quantification, and a long lifetime of ∼2,000 h. Relative to background illumination and fluorescence emission of the InSpeck standard microspheres (6 μm; excitation 505 nm/emission 515 nm; Molecular Probes, Inc.), the Hg/Xe lamp was adjusted to produce even field illumination. The microscope filter turret included a Chroma DAPI 31000 set, a Chroma FITC 41001 set, and a Chroma Texas red 41004 set for quantification of Hoechst, Alexa Fluor 488, and Alexa Fluor 568 signals, respectively. Typically, a 10× HC PLAN APO (0.40 aperture) or 20× HC PLAN APO (0.70 aperture) objective was used to capture signals from tissue sections and single cells, respectively. An E Plan 4× (0.10 aperture) objective was used to capture total signals from tissue sections matched to those analyzed by RPPA. All microscope objectives were calibrated with an “Objective Micrometer” (Fisher Scientific) with 0.01 mm divisions, and micrometer images were captured and stored as 12-bit “.tif” files for periodic calibration checks.

Twelve-bit grayscale images were captured for the quantification of fluorescence signal, with a digital black and white CCD camera (model ORCA-C4742-95-12ER; Hamamatsu Corp.) with high-resolution (1,344 × 1,024 pixels), high-quantum efficiency (>50% from 400 to 700 nm), and a dynamic range of 2,250:1. The camera was mounted via a 1.0× coupling lens that produced an 11× magnification. Live images were visualized with a 22-inch (16 × 10 aspect ratio), high-resolution (3,900 × 2,400 pixels) display monitor (Eye Smart ES-100 LCD; Meyer Instruments, Inc.). Twelve-bit digitized images were stored as “.tif” files.

All components of the automated imaging analysis system were controlled and fully programmable with Image-Pro Plus software (Meyer Instruments, Inc.) that included the Scope Pro and Advanced Fluorescence Acquisition modules. Principal features of the computer workstation (Precision 340 Workstation P4 by Dell Corporation; Meyer Instruments, Inc.) included an Intel Pentium 4 processor with 2.0 GHz front side bussing, 1 GB RAM memory (PC800), 533 MHz system bus, 32 MG, VGA graphics card, dual 80 GB hard drives (IDE 7200 rpm), USB and PCI ports, and the Windows 2000 operating system.

Image Capture. Each image capture session was initiated with a protocol to confirm consistent, system performance across analytic runs. Individual slides were prepared with each of three suspensions of InSpeck fluorescent microspheres that produced relative emissions of 1%, 10%, and 100%. Fluorescence images were generated with the 20× objective and Chroma FITC filter set and captured as 12-bit gray scale files, at exposures of 250, 25, and 3 ms, respectively, with the black and white CCD camera. The 12-bit gray scale consists of 4,096 divisions or units (gsu). Images of 200 to 400 individual microspheres per slide were segmented at a threshold of 900 gsu and measurements including mean pixel intensity (MPI) of each microsphere were transferred to an Excel sheet for analysis. Mean and SD of the MPIs for spheres from each suspension were computed and compared with the values of previous image capture sessions.

Digital images of labeled tissue specimens and the corresponding isotype controls were captured no more than 3 days after the slides were processed with the BioGenex autostainer. Positively labeled specimens of cases and controls, cell standards, or tissue standards were reviewed for fluorescence intensity; one specimen exhibiting the brightest fluorescence was selected from each group. Camera exposure time was set such that the brightest pixels in the selected specimen yielded values of ∼3,200 gsu, i.e., 80% of maximum on the 12-bit gray scale. All frames were captured at this exposure setting, typically 100 to 200 ms at a gain of “0” for both single cells and tissue sections captured with the 10× objective. Camera frames were selected to maximize the number of events (single cells, acini, or epithelial strips) captured. In addition, frames of cancerous cores were selected to provide a balanced representation of cancerous acini/epithelium (CA) and noncancerous acini (NAA). Categories of acini/epithelium included CA, NAA, and noncancerous acini (NA) of control cores and the tissue standard. Typically, 30 to 300 acini were captured in each category, and 200 to 400 single cells were captured from each slide preparation of the LNCaP cells. Images were stored as 12-bit gray scale files.

Labeled specimens matched to those specimens analyzed by RPPA were treated differently. Contiguous, nonoverlapping images of antibody-labeled specimens and isotype controls were captured with the 4× objective to encompass the entire tissue section.

Captured image files were loaded into the Image-Pro Plus environment and relevant events, i.e., individual LNCaP cells and the acini/epithelium of cores and tissue sections were segmented for quantification of fluorescence signal and event dimensions, with the “count/size” function of Image-Pro Plus. Precise segmentation of prostate epithelium was achieved by setting a lower threshold just above background emission, typically ∼230 to 250 gsu, and an upper threshold at the upper limit of the 12-bit gray scale (i.e., 4,095 gsu). In addition, an image filter set for a feret (event caliper measurement) of 10 to 15 μm excluded from measurement stromal elements, primarily endothelium, that labeled with the fluorescent reagents. Image-Pro Plus generated a high-content array of measurements that included event number, area, feret, MPI, sum of pixel intensities, and maximum and minimum pixel intensities. The algorithm for event MPI excluded the background pixels of each segmented event. The fluorescence signal and spatial features of the LNCaP cells were quantified in the same way, except that a filter was set for an area of 3 μm² and excluded fluorescent fragments from the measurement.

Quantification of the β-catenin signal in slide specimens matched to those analyzed by RPPA did not entail precise segmentation of acini, but instead, segmentation of the entire piece of tissue in the camera frame. In this way, the sum of pixel intensities corrected for background was determined for the entire tissue section for comparison to total β-catenin content of the paired tissue section as determined by RPPA. More than 99% of the β-catenin signal was present in the epithelium of these specimens.

β-Catenin Expression Under Different Analytic Conditions. The concordance of β-catenin expression in methacarn-fixed tissues analyzed by RPPA and QFIA, and in methacarn-fixed and formaldehyde-fixed tissues analyzed by QFIA were evaluated by linear regression analysis. The correlation coefficient (R value) was calculated as an indicator of the strength of the linear relationship.

β-Catenin Expression in the Archived Core Biopsies. MPIs of labeled acini/epithelium and single cells were corrected for nonspecific immunoglobulin labeling and background emission by subtracting the average MPI (AMPI) of acini/epithelium in the corresponding isotype control (typically 250-300 gsu). To determine whether the AMPI in each data set was representative, we plotted the AMPI of increasing numbers of events. AMPI was considered representative if the plot attained a steady “zero” slope, typically beginning at 30 acini or 100 cells. Arithmetic average, SD, and SE were computed for each type of event (single cell, acinus, or epithelial strip). Differences between the AMPIs of event categories (NA, NAA, CA) for each matched pair were evaluated by a paired t test with PlotIT software (Scientific Programming Enterprises).

Receiver operating characteristic curves were generated for case-associated AMPIs of CA or NAA in relation to control-associated AMPIs of NA, plotted as a function of increasing threshold AMPI (27). Fractional areas under the curve along with 95% confidence intervals (95% CI) was determined for each plot. An AUC of ≥ 0.67 indicated significant discrimination of cases (NAA or CA) and controls (NA). Sensitivities and specificities and their 95% CIs were determined at selected threshold values. Bootstrapping was used to estimate confidence intervals for the AUC and for the sensitivity and specificity at selected threshold values, taking into consideration the repeated measures design of the tests (two slides per person). Two-hundred bootstrapped samples were used to estimate the confidence intervals, resampling at the person level within a group (NA, NAA, and CA) to preserve the correlation structure from two slides within a person (27).

Precision and reproducibility, essential prerequisites for quantitative analysis, were assured with quality control procedures, including a protocol for routine evaluation of the performance of the imaging system. This protocol, based on fluorescence emission of standard fluorescent microspheres, was completed at the start of each image capture session. AMPI was determined for three sets of microspheres with relative intensities of 1%, 10%, and 100%. A variation of >5% from any previous session, for any set, would indicate the need for optical alignment, adjustment of field illumination, or replacement of the Hg/Xe lamp. Fluorescence of the standard microspheres varied <5% among the image capture sessions required for this study as exemplified by six of these system evaluation routines (Fig. 1).

Analytic reproducibility was established by the incorporation of both single-cell and tissue standards into each analytic run. Triplicate slides were labeled for β-catenin and negative controls were included for background correction. Reproducibility was exemplified by data from six analytic runs (Fig. 1). Within each run, AMPIs of individual slides differed from the means of triplicate slides by an average of 5% for LNCaP cells and 6% for acini, and across experiments, the means of triplicate slides differed from the mean of the AMPIs of all slides by an average of 7% for LNCaP cells and 10% for acini.

Reliable Quantification of β-Catenin in Slide Specimens of Methacarn-Fixed, Paraffin-Embedded Prostate Tissue. The first step in the process of validating QFIA of β-catenin in archived specimens was to examine the reliability of quantitative determinations in methacarn-fixed tissues. RPPA was selected for this study because this well-established methodology is routinely applied for the quantification of proteins in slide specimens and exhibits high precision, reproducibility, and sensitivity over a wide range of protein content (11). Because our goal was to analyze β-catenin expression in paraffin-embedded, archived specimens, we elected to implement this stage of validation with methacarn-fixed, paraffin-embedded prostate specimens. Proteins were extracted efficiently from slide specimens of tissues fixed with methacarn (28). Finally, because by far most of the β-catenin signal was expressed in prostate epithelial cells, total protein in each tissue section were analyzed, i.e., specimens were not microdissected for RPPA nor were the acini of captured images segmented for QFIA. The study design required parallel analyses by RPPA and QFIA of adjacent serial sections mounted on glass slides, maximizing agreement of protein expression in the specimens to be analyzed.

Visual inspection of the scans for one set of protein arrays, corresponding to two BH glands and three cancerous glands, revealed consistent size and intensity of the triplicate spots (Fig. 2A). The integrated fluorescence intensity of each spot was background-corrected and the mean for each triplicate was calculated. The means for each dilution of the standard and each extract were plotted against the dilution factor (Fig. 2B). For all specimens, integrated fluorescence intensity was linear over a wide range of protein concentrations, permitting reliable comparisons across samples and analyses.

The concordance of RPPA and QFIA of β-catenin was evaluated with methacarn-fixed specimens from two BH glands and eight cancerous glands, previously screened by QFIA and selected to provide a wide range of β-catenin expression. The set of tissues represented in Fig. 2 was analyzed in two separate studies, and a second set of five cancerous glands was analyzed in another study. Data from the three studies were pooled and the integrated fluorescence intensity of each tissue section, determined by QFIA, was plotted as a function of integrated intensity of the matched tissue section as determined by RPPA (Fig. 3A). The data exhibited a strong linear relationship with a correlation coefficient of 0.97, indicating that the QFIA measurements were directly proportional to the protein content of the tissue specimens.

Concordance of β-Catenin Expression in Formaldehyde-Fixed and Methacarn-Fixed Prostate Tissue Specimens, Determined by QFIA. Concordance of QFIA measurements with β-catenin content determined by RPPA was evaluated with methacarn-fixed slide specimens because methacarn is an accepted fixative for both RNA and protein analyses, and is compatible with efficient protein extraction (28). In contrast, formaldehyde, a cross-linking fixative, is incompatible with protein extraction (29). Because we were interested in applying QFIA to archived tissues that were fixed in formaldehyde-based fixatives, it was necessary to show the concordance of β-catenin expression in methacarn-fixed and formaldehyde-fixed tissues, as determined by QFIA. This was implemented with a TMA that incorporated both methacarn-fixed and formaldehyde-fixed specimens from the same surgically removed cancerous and noncancerous prostate glands. The TMA permitted a direct comparison of expression in the NA of two BH glands, the NAA of five cancer-bearing glands, and the CA of one cancer-bearing gland. To increase the stringency of this test, we included analyses of the NAA from three additional PC glands selected for relatively high expression of β-catenin. These glands were not represented in the TMA. The data from two separate analyses of the TMA specimens and a single analysis of the additional PC glands were examined by linear regression analysis. The AMPIs of acini/epithelium in tissues fixed with formaldehyde were plotted as a function of the AMPIs of acini/epithelium in tissues fixed with methacarn, and the correlation coefficient was calculated (Fig. 3B). AMPIs of formaldehyde-fixed and methacarn-fixed tissues exhibited a strong, linear relationship with a correlation coefficient of 0.86. We concluded that QFIA measurements of β-catenin expression were directly proportional with the quantity of β-catenin in the tissue sections and, consequently, QFIA may be applied to quantification of the protein in archived prostate specimens fixed in a formaldehyde-based fixative.

The preceding work (above) established the foundation for the application of tissue-based QFIA to the problem of quantifying β-catenin in archived prostate biopsies fixed in formalin, evaluating protein expression in the NAA of cancer-bearing glands as a potential field marker for the presence of PC. The expression of β-catenin was analyzed in a cross-sectional case-control study that included 42 cancer cases and 42 controls matched on the basis of age (±5 years) and year of biopsy (±3 years). Clinical and demographic data are summarized in Table 1. AMPIs of the NA of noncancer controls, and both the NAA and CA of the cancer cases, were determined with gray scale images captured from duplicate slide specimens analyzed separately in different analytic runs (Fig. 4A). Occasional prostatic intraepithelial neoplasia lesions were analyzed separately. An average of two prostatic intraepithelial neoplasia lesions was seen in specimens from each of six cancer cases. AMPIs corrected for background fluorescence are presented in Table 2. Consistent with reports (23) that β-catenin is principally a cell adhesion molecule in the prostate epithelium, fluorescence signals were seen in association with the cell membrane and not in the nucleus. Expression in CA compared with NA was significantly (P < 0.02) reduced in 37 of the 42 matched pairs of cases and controls. The grand mean ± the SE of the AMPIs of CA for all cases was 458 ± 21, compared with 674 ± 14 for NA of all controls. In relation to altered β-catenin expression in NAA as a possible field marker for PC, it was of interest to find that expression in NAA compared with NA was significantly reduced (P < 0.02) in 31 of the 42 matched pairs. The grand mean ± the SE of the AMPIs of NAA for the 31 cases was 538 ± 16 compared with 684 ± 16 for NA of the matched controls.

These observations prompted an analysis of the separation of the three groups of prostatic acini/epithelium based on the AMPI of β-catenin signaling in the acini/epithelium for each case and control. Consequently, AMPIs for the NA of the controls and the NAA and CA of the cases were pooled as three separate groups and the fraction of each group at or below the selected AMPIs was plotted against AMPI (Fig. 4B). The analysis revealed that QFIA of β-catenin produced a distinct separation of the three groups of prostatic acini. The marked separation of CA from NA supports the application of QFIA to the study of β-catenin expression in archived prostate biopsies. The separation of NAA from NA was of particular interest in relation to detecting a potential field marker for PC and suggested the possibility of detecting a significant proportion of cancer cases, with acceptable specificity. For example, none of the controls were represented at a threshold AMPI of 500 gsu, compared with 30% of cases.

These findings justified a more rigorous analysis with receiver operating characteristic plots of the same data sets, i.e., AMPIs of NA, NAA, and CA. First, we examined the classification of cases and controls based on β-catenin expression in CA and NA, respectively (Fig. 4C, curve 1). At a cut point of 500 gsu, QFIA classified 70% (95% CI, 56-84%) of cancer cases as positive, whereas classifying only 7% (95% CI, 1-13%) of controls as positive, again supporting the utility of QFIA for analysis of β-catenin expression in archived specimens. Next, we examined classifications based on β-catenin expression in NAA and NA (Fig. 4C, curve 2). At a cut point of 500 gsu, QFIA classified 24% (95% CI, 12-36%) of cancer cases as positive, with a specificity of 93% (95% CI, 87-99%). Moving the cut point to 550, we find that QFIA correctly classified 42% (95% CI, 26-57%) of the cases, with a specificity of 88% (95% CI, 80-96%). The results indicate that β-catenin expression in NAA as determined by QFIA is potentially a useful marker for clinical diagnosis of PC in biopsies that may not include existing cancerous lesions, and supports the application of QFIA to archived tissues for the identification of potential markers of disease.

The human proteome consists of >300,000 different protein species generated via alternative gene splicing and posttranslational modification from an estimated 30,000 genes (2). Proteins generate useful chemical energy and transform this energy into those functions that comprise the living organism; as a consequence, protein profiling is viewed as essential to defining disease in molecular terms. It is expected that qualitative and quantitative changes of specific proteins produce and specify disease states, offering the opportunity for prevention and early diagnosis of disease as well as individualized therapeutic intervention. This report introduces a new method, tissue-based QFIA, for quantitative analysis of proteins and secondary modifications of proteins in slide-mounted tissue sections, and shows its complementation to RPPA (2, 6, 11-13) and AQUA (4, 5, 14); published methods for quantitative profiling of proteins and their secondary modifications in small tissue samples. A key element of the present study was to discriminate the analysis of β-catenin expression in cancerous acini/epithelium (CA) and noncancerous acini (NAA) in biopsies from patients with PC, and noncancerous acini (NA) in biopsies from individuals without clinical cancer. Quality-controlled imaging analysis procedures, the automated imaging system, and powerful image analysis features of the Image-Pro Plus software permitted rapid, precise analysis of all classes of acini/epithelium.

The same analysis by RPPA or AQUA is expected to encounter significant challenges. RPPA requires laser microdissection of 2,500 to 15,000 cells for the preparation of extracts with sufficient protein content for analysis; this number of epithelial cells free of contaminating cells or stroma is readily achieved with NA or NAA. The same number of uncontaminated, cancerous epithelial cells may require many more dissections, especially in some higher-grade tumors exhibiting loss of tissue architecture and consisting of strips of relatively few, cancerous epithelial cells. Furthermore, analysis of physically separated cells such as individual stromal cells by RPPA may be impractical, although readily achievable by tissue-based QFIA. Finally, RPPA is not applicable to archived tissues fixed in formaldehyde-based fixative. RPPA has its strength in the analysis of multiple, extractable proteins in readily dissected tissue elements.

The application of AQUA to the present study would entail a different set of difficulties in comparison to RPPA. AQUA is a fully automated method that makes use of fluorescence labeling, e.g., with an anti-cytokeratin antibody mixture, to delineate tissue compartments, e.g., prostate epithelium, and to assign specific proteins to these compartments. Fluorescence discrimination of the CA and NAA compartments in the same biopsies may be a significant problem. For example, α-methylacyl-CoA racemase is expressed in PC and high-grade prostatic intraepithelial neoplasia and is an accepted PC biomarker (14). The use of this marker to determine the distribution and quantity of a protein of interest in the CA and NAA compartments would require a triple-labeling protocol to identify the epithelium (first label), with and without the α-methylacyl-CoA racemase protein (second label), i.e., CA and NAA, respectively, and to quantify the protein of interest (third label) within the compartment(s). Quantification becomes more difficult with increasing numbers of antibodies that may be associated with stearic hindrance to epitope binding. Discrimination between prostatic intraepithelial neoplasia and CA by AQUA is not likely, but is readily achieved with tissue-based QFIA. A major strength of AQUA is the high-throughput analysis of markers readily assignable to compartments delineated by fluorescence labeling.

Successful application of tissue-based QFIA to β-catenin in archived prostate biopsies fixed in formaldehyde-based fixative was in part the result of rigorous, quality-controlled procedures. Accordingly, core biopsies were selected for study, in part, because they are rapidly transferred to fixative, minimizing tissue degradation; cases and controls were matched for both age and year of biopsy; tissues were labeled with antibody reagents at epitope-saturating concentrations; fluorescence excitation was achieved with a stable Hg/Xe lamp adapted to the Leica microscope; programming of the robotic autostainer was tuned to obviate unscheduled slide batching associated with reduced fluorescence signal; performance of the imaging system was evaluated with standard fluorescent beads, before each analytic run; single cell and tissue standards were included in each analytic run; and conditions for reproducibility within and across analyses and from technologist-to-technologist were established before initiating the biomarker studies. The suitability of β-catenin as a potential field marker also contributed to the success of our field marker study. Consistent expression of this protein among individual acini or epithelial strips in the three classes of prostate epithelium facilitated significant discrimination of β-catenin measurements differing by 10% to 15%.

Dysregulation of the cadherin/catenin cell adhesion complex in PC provided the rationale for evaluating β-catenin as a potential field marker for PC. Wehbi et al. (24) analyzed the expression of cadherin proteins by immunohistochemistry in 38 PC, 14 prostatitis, and 19 benign prostatic hyperplasia specimens. The mean percentages of the areas positive for labeling with an anti–pan-cadherin antibody were 15.2%, 21.5%, and 2.6% for benign prostatic hyperplasia, prostatitis, and PC, respectively. Ninety-five percent of the PC values fell below a threshold that excluded all benign prostatic hyperplasia and prostatitis specimens combined; suggesting that altered expression of proteins in the cadherin/catenin complex may identify PC field disease/effect. A study by Kallakury et al. (21) supports this conclusion. The investigators applied immunohistochemistry to an analysis of selected proteins of the catenin/cadherin complex in PC, comparing expression in the cancerous epithelium to expression in the noncancerous epithelium of cancer-bearing prostate glands. E-cadherin, α-catenin, and p120 CTN were significantly reduced in 25%, 17%, and 45%, respectively, of 118 cancer-bearing glands. This same study, however, found significant reduction of β-catenin in only 4% of the PC cases, implying that this protein may not be a useful marker of PC. A similar study by Bismar et al. (23) also found a reduction of β-catenin in only 4% of PC cases (n = 89).

These reports seemingly contradict results obtained in the present study, which showed a significant reduction of β-catenin expression in the CA of 37 of the 42 PC cases. The apparent contradiction was resolved by the acceptance of the validity of the data of all three studies and the appreciation of substantial differences in experimental design. Kallakury et al. (21) and Bismar et al. (23) evaluated expression in the CA in relation to that in the NAA of cancer-bearing glands, whereas in the present study, we evaluated expression in the CA of biopsies from PC cases in relation to both the NAA of the same biopsies and the NA of biopsies from matched controls without clinical disease. In addition, the present study was based on continuous quantitative measurements, whereas the studies of Kallakury et al. (21) and Bismar et al. (23) were based on immunohistochemical classifications. The consequence of the latter difference is emphasized by our observation that expression of β-catenin in CA in relation to NAA of cancer-bearing glands was significantly reduced in 71% of the 42 PC cases in contrast to 4% in both immunohistochemical studies. It is interesting to note that in our study, the expression in the CA of 4 of the 42 cases was reduced relative to the NAA by >50%; a frequency of down-regulation similar to that achieved in the immunohistochemical studies.

Successful application of tissue-based QFIA to β-catenin in prostate tissues is attributed, in part, to key features of the protein. First, β-catenin exhibits bright, specific labeling in archived tissues fixed with formaldehyde-based fixative. This feature opened the tissue archives to tissue-based QFIA of the protein as a potential field marker for PC. Second, the β-catenin signal was consistent among individual acini or epithelial strips within each class (CA, NAA, or NA) of prostate epithelium evaluated. This feature precluded significant analytic variance that could mask biologically significant differences between the classes of epithelia. In agreement with published studies (23), we found β-catenin labeling, in all three classes, to be restricted to the cytoplasm in the vicinity of the cell membrane, and did not detect nuclear signals. On this basis, Bismar et al. (23) distinguish PC and colorectal cancer in which the protein is translocated to the nucleus and is detectable by immunohistochemistry. The investigators suggest the presence of different β-catenin signaling pathways in prostate and colorectal carcinogenesis.

The major finding from the application of tissue-based QFIA to β-catenin in archived prostate specimens was the generally reduced expression of β-catenin in NAA relative to NA. This difference was sufficient to produce thresholds of fluorescence intensity that detected 24% and 42% of cancer cases, with specificities of 93% and 88%, respectively. The observed sensitivities and specificities do not qualify the marker as an independent indicator of PC field disease/effect, but identify altered expression of the protein as a potentially significant contributor to a small panel of field markers, and support the feasibility of such a panel. Ultimately, the diagnostic value of this marker can be determined best by studies of tissues collected prospectively and processed by rigorously standardized procedures that may further reduce measurement variance and improve the sensitivity of the QFIA method. Such procedures are integral to the “Molecular Profiling Initiative” of the Cancer Genome Anatomy Project of the National Cancer Institute⁵

In conclusion, we have developed tissue-based QFIA, a new method for the quantification of proteins and secondary modifications of proteins in precisely selected elements of slide-mounted specimens, in the context of observable histologic features. Using RPPA, we have validated the application of QFIA to β-catenin in archived prostate tissue in formaldehyde-based fixatives. Reduced expression of β-catenin in NAA relative to the NA of matched controls is a potential field marker for PC in biopsies that miss existing cancer nodules. The observed sensitivities and specificities qualify the marker as a potentially significant contributor to a small panel of field markers, and support the feasibility of applying QFIA to the development of such a panel.

We thank Mary Ann Calero for her excellent technical support in preparing the slide-mounted paraffin sections, Doug Corum (Veterans Affairs Medical Center, Omaha, NE) for assistance in identifying and retrieving archived tissues, and the Histology Core Facility of the Eppley Institute for Cancer Research (University of Nebraska Medical Center) for paraffin-embedding the prostate tissues.