Integrative Analysis of N-Linked Human Glycoproteomic Data Sets

Apr 11, 2012 - Integrative Analysis of N-Linked Human Glycoproteomic Data Sets. Reveals ..... SDS-PAGE, and labeling was visualized via Western blot...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/jpr

Integrative Analysis of N-Linked Human Glycoproteomic Data Sets Reveals PTPRF Ectodomain as a Novel Plasma Biomarker Candidate for Prostate Cancer Theodore E. Whitmore,† Amelia Peterson,‡ Ted Holzman,§ Ashley Eastham,∥ Lynn Amon,§ Martin McIntosh,§ Adrian Ozinsky,† Peter S. Nelson,⊥ and Daniel B. Martin*,# †

Institute for Systems Biology, 1441 N. 34th St., Seattle, Washington 98103, United States Department of Chemistry, University of Wisconsin, Madison, Wisconsin, United States § Division of Public Health Sciences and ⊥Division of Human Biology, Fred Hutchinson Cancer Research Center, 1100 Fairview Avenue, Seattle, Washington 98109, United States ∥ Analytical & Formulation Sciences, Amgen Inc., Seattle, Washington 98119, United States # Seattle Cancer Care Alliance, 825 Eastlake Avenue East, P.O. Box 19023, Seattle, Washington 98109, United States ‡

S Supporting Information *

ABSTRACT: In an attempt to identify prostate cancer biomarkers with greater diagnostic and prognostic capabilities, we have developed an integrative proteomic discovery workflow focused on N-linked glycoproteins that refines the target selection process. In this work, hydrazide-based chemistry was used to identify N-linked glycopeptides from 22Rv1 prostate cancer cells cultured in vitro, which were compared with glycopeptides identified from explanted 22Rv1 murine tumor xenografts. One hundred and four human glycoproteins were identified in the former analysis and 75 in the latter, with 40 proteins overlapping between data sets. Of the 40 overlapping proteins, 80% have multiple literature references to the neoplastic process and ∼40% to prostatic neoplasms. These include a number of well-known prostate cancerassociated biomarkers, such as prostate-specific membrane antigen (PSMA). By integrating gene expression data and available literature, we identified members of the overlap data set that deserve consideration as potential prostate cancer biomarkers. Specifically, the identification of the extracellular domain of protein tyrosine phosphatase receptor type F (PTPRF) was of particular interest due to the direct involvement of PTPRF in the control of β-catenin signaling, as well as dramatically elevated gene expression levels in the prostate compared to other tissues. In this investigation, we demonstrate that the PTPRF E-subunit is more abundant in human prostate tumor tissue compared to normal control and also detectable in murine plasma by immunoblot and ELISA. Specifically, PTPRF distinguishes between animals xenografted with the 22Rv1 cells and control animals as early as 14 days after implantation. This result suggests that the ectodomain of PTPRF has the potential to function as a novel plasma or tissue-based biomarker for prostate cancer. The workflow described adds to the literature of potential biomarker candidates for prostate cancer and demonstrates a pathway to developing new diagnostic assays. KEYWORDS: PTPRF, glycosylation, biomarkers, prostate cancer, proteomics, hydrazide, plasma



INTRODUCTION Prostate cancer (PCa) remains the most prevalent non-skin cancer in men over the age of 50.1 There are approximately 200,000 new cases diagnosed in the United States with the disease affecting 1 in 6 men during their lifetime. Prostate cancer remains an important public health concern in Western countries and an emerging malignancy in developing nations (reviewed in ref 2). While the routine use of blood-based screening for prostate specific antigen (PSA) has contributed significantly to the early detection, treatment, and survival rates of patients with prostatic neoplasms, elevated PSA levels do not actually discriminate between indolent and aggressive forms of © 2012 American Chemical Society

prostate cancer or from non-malignant conditions such as benign prostatic hyperplasia (BPH).3,4 As a result of this poor discrimination of disease status, many needless biopsies and surgical procedures are performed on patients with benign prostate abnormalities5 and the benefit of PSA screening remains controversial.6−9 Thus, there remains a clear need for additional prostate cancer biomarkers with greater diagnostic and prognostic specificity.10,11 Received: September 15, 2010 Published: April 11, 2012 2653

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

xenograft was developed from an outgrowth following castration-induced regression of the parental, androgendependent CWR22 xenograft and subsequent tumor relapse in an androgen-deficient environment.30 22Rv1 is the human prostate adenocarcinoma epithelial cell line derived from the relapsed and serially propagated CWR22R xenograft.29 Although 22Rv1 cells express a functional androgen receptor, they are androgen-independent and grow well in androgen-free medium.33,34 22Rv1 cells produce tumors in mice with characteristics and morphology analogous to the parental CWR22R xenograft and secrete measurable PSA. In the present investigation, we have characterized the glyco-proteome of 22Rv1 cells both in vitro and in vivo to use as a starting point in a verification analysis. A number of candidates were identified using an integrative analysis, one of which is shown to be highly sensitive and specific for disease in the mouse xenograft model.

Biomarker discovery and validation in plasma is a complicated process due to the complexity and dynamic range of the plasma proteome, the challenges of finding proteins with tissue specificity, and the variables that arise from patient heterogeneity. Most plasma and serum proteome discovery efforts have largely failed to yield useful biomarker targets.12−16 Efforts employing extensive fractionation17 may be a better fit for serum and plasma biomarker discovery and have been shown to identify and validate proteins for pancreatic cancer.18 However, the latter are very large-scale efforts. To address the challenges of identifying useful biomarkers for serum diagnostics, we have applied an integrative proteomic approach that focuses discovery efforts exclusively on the prostate tissue producing the biomarker candidates and further focuses on a subclass of proteins that may be of particular utility for diagnostic assays. Herein we target N-linked glycoproteins measurable in cultured human prostate cancer cells and the xenografts derived from implanting those cancer cells into immunocompromised mice. The capture, identification, and quantification of N-linked glycoproteins are emerging proteomics techniques that readily identify proteins primarily destined for the plasma membrane and extracellular regions. Due to their proximity to blood circulation, these proteins have the potential of serving as accessible biomarkers or even therapeutic targets. While others have used lectin-based methodologies to study this subproteome (reviewed in refs 19 and 20), we have focused on the selective capture of glycosylated proteins using hydrazide chemistry to reduce sample complexity and increase the dynamic range of the MS analysis.21,22 Herein we employ two such strategies that have evolved from this technology. For soluble or secreted glycoproteins such as those found in blood plasma, the first strategy entails peptide capture on hydrazidederivatized agarose resin, followed by enzymatic release with PNGase-F.21−26 The second strategy is designed primarily for the capture of membrane-associated glycoproteins and uses a membrane-impermeable reagent, biocytin hydrazide, which reacts with aldehydes generated from the oxidation of extracellular sugars. Following derivatization and trypsin digestion, the biotinylated membrane-specific glycopeptides are captured on streptavidin-agarose and released with PNGase-F.27 In addition to focusing on membrane and secreted glycoproteins, we sought to streamline the development of a pipeline for discovery and validation of potential targets by utilizing a prostate xenograft model system that allows target discovery both in vitro and in vivo. Human xenograft models have an advantage over human studies by minimizing factors that would otherwise hinder biomarker discovery, such as the genetic and clinical heterogeneity of tumor samples.28 Additionally this process clearly sidesteps the dynamic range issues that arise using discovery proteomics in body fluids as well as the questions of origin of the protein biomarkers. In the case of xenografts, the proteins identified are unequivocally tumorderived given their assignment to human protein identifiers. In this study, we have xenografted 22Rv1 cells originally derived from human prostate cancer xenografts CWR22 and CWR22R.29 The CWR22 prostate tumor xenograft was established in 1996 from a primary human prostate adenocarcinoma (Gleason sum 9). An androgen-sensitive tumor, the xenograft produces PSA at levels detectable in the bloodstream of mice in correlation to their castrationresponsive tumor burden.30−32 The CWR22R prostate tumor



EXPERIMENTAL SECTION

22Rv1 Xenograft Mouse Model

All mouse work was carried out according to IACUC regulations as approved by the Fred Hutchinson Cancer Research Center (FHCRC) animal use committee. For this study, male NOD SCID mice were injected in their flanks with 7.5 × 105 22Rv1 cells in 50% HBSS/50% Matrigel (BD Biosciences, San Jose, CA) or mock-injected with 50% HBSS/ 50% Matrigel as control. Plasma and tumors (when present) were collected at defined time points and analyzed from 14 to 30 days post-injection. Flash frozen tumors and plasma were stored at −80 °C until use. Analysis of 22Rv1 N-Linked Membrane Glycoproteome

Unless otherwise noted, all cell culture media, reagents, and supplements were purchased from GIBCO (Invitrogen, Grand Island, NY). Labeling reagents were purchased from SigmaAldrich (St. Louis, MO). For all cell-culture-based experiments, unless otherwise mentioned, 22Rv1 human prostate cancer cells (American Type Culture Collection, Manassas, VA) were grown in 175 cm2 dishes to 80% confluence in phenol-redcontaining RPMI 1640 media supplemented with fetal bovine serum (FBS, 10% v/v) and penicillin/streptomycin (P/S, 1% v/ v). Using a method adapted from Wollscheid et al.,27 8 dishes of 22Rv1 prostate cancer cells (1 × 108 cells) were washed on the dish with labeling buffer (FBS, 1/1000 v/v, in phosphate buffered saline (PBS), pH 6.5) after aspiration of culture media. Washed cells were treated with sodium m-periodate solution in labeling buffer (2.5 mM) for 20 min at RT to oxidize cis-diol groups on the terminal sugar of membrane glycans. Following washing with labeling buffer, biocytin hydrazide (Biotium, Hayward, CA) in labeling buffer (6.5 mM) was applied to the cells for 30 min to label oxidized sugars. The reaction was quenched for 15 min with 2 vol of tyrosine hydrazide in labeling buffer (25 mM) before washing with labeling buffer. Hypotonic lysis buffer (10 mM Tris, 2 mM magnesium chloride, 10 mM potassium chloride, and 1 mM EDTA) supplemented with protease inhibitors and sodium cyanoborohydride reductant (5 mM) was added to still culture dishadherent cells, and cells were removed and combined by vigorous scraping into a precooled Dounce homogenizer. Cells were lysed with 25 strokes, and debris was pelleted by centrifugation. The pelleted cell debris was subjected to a second round of lysis. An equal volume of membrane preparation buffer (280 mM sucrose, 50 mM MES, 450 mM 2654

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

al.23 Briefly, 1 mg of protein was diluted in denaturing buffer to a concentration of 2 mg/mL. The proteins were reduced, alkylated, and digested with trypsin prior to desalting. Terminal glycan sugar residues were oxidized with sodium m-periodate, and the sample was desalted again. The eluate pH was adjusted to 5.5 with sodium acetate and incubated in 50 μL of Affi-Prep Hz Hydrazide Support (Bio-Rad, Hercules, CA) overnight at RT. The glycopeptide-coupled support was washed extensively and sequentially with sodium chloride (1.5 M), water, dimethylformamide/pyridine/water (50/10/40 v/v/v), dimethylformamide, and ammonium bicarbonate (25 mM, pH 7.5). The support was resuspended in a minimum volume of ammonium bicarbonate (25 mM, pH 7.5) with 3 μL of PNGaseF and incubated overnight at 37 °C. Cleaved peptides were collected and dried under reduced pressure.

sodium chloride, 10 mM magnesium chloride) was added to the combined supernatants from both rounds of lysis and incubated at 4 °C, 10 min. The membrane fraction was pelleted via ultracentrifugation at 100,000g, 4 °C for 1 h. The resulting pellet was disrupted in a minimal amount of sodium carbonate dibasic (100 mM, pH 11) and incubated for 30 min on ice. A second ultracentrifugation for 30 min pelleted the crude membrane fraction for glycopeptide enrichment. Membrane pellets were homogenized by tip sonication in SDS (0.1% v/v) in ammonium bicarbonate until clear. Insoluble debris was removed via centrifugation. TPCK trypsin (Pierce, Rockford, IL), 15 μg per dish of cells, was added, and samples were sonicated via microtip at 4 °C for 1 min in 3 × 20 s bursts. Following 1 h of incubation at 37 °C, samples were reduced with 100 mM Tris(2-carboxyethyl) phosphine buffered to pH 8 with ammonium bicarbonate (50 mM), at a final concentration of 2 mM for 30 min at 4 °C with agitation. Iodoacetamide (500 mM, pH 8 in ammonium bicarbonate) was added to a final concentration of 10 mM and incubated 30 min at 4 °C in the absence of light. The alkylation reaction was quenched with dithiothreitol (500 mM, pH 8 in ammonium bicarbonate) added to a final concentration of 50 mM and incubated 30 min at 4 °C. Typsin digestion was then completed with the addition of another 15 ug TPCK trypsin/dish of cells and overnight incubation at 37 °C. Digestion was confirmed by SDS-PAGE, and labeling was visualized via Western blot. Washed streptavidin beads (Ultralink Streptavidin Plus, Pierce) were combined with the digested membrane fraction and protease inhibitors to quench enzyme activity. The slurry was incubated with constant agitation for 1−3 h at RT and then overnight at 4 °C. Successful depletion of glycosylated peptides was confirmed by Western dot blot probed with StreptavidinHRP (Zymed, South San Francisco, CA). Following depletion, the beads were washed with sodium chloride (5 M), incubated with SDS in ammonium bicarbonate (2% v/v, 65 °C, 30 min), and then aggressively washed with hypotonic lysis buffer, ammonium bicarbonate (100 mM), sodium chloride (5 M), sodium carbonate dibasic (100 mM, pH 11), and ammonium bicarbonate (25 mM). The streptavidin beads were resuspended in a minimal volume of ammonium bicarbonate (25 mM, pH 7.5) with PNGaseF (New England Biolabs, Ipswich, MA) and incubated overnight at 37 °C. Cleaved peptides were collected and dried under reduced pressure.

Strong Cation Exchange (SCX)

All samples were subjected to SCX cleanup prior to mass spectrometry analysis to reduce the potential for contamination from preservatives in the commercial PNGaseF preparation. Dried peptides were cleaned up via strong cation exchange using an in-house prepared fritted column packed with polysulfethyl A resin (PolyLC, Columbia, MD) connected to a syringe pump (Harvard Apparatus, Holliston, MA). Briefly, peptides were reconstituted in 25% v/v acetonitrile containing 1% v/v acetic acid and loaded onto a preconditioned column at 15 μL/min. The bound peptides were washed with loading buffer and then eluted in 400 mM ammonium acetate in 25% v/v acetonitrile in a single fraction. The eluate was dried and peptides washed twice in methanol (50% v/v) prior to resuspension in 1% v/v acetonitrile with formic acid (0.1% v/ v) for reversed phase LC−MS/MS. LC−MS/MS

All analyses utilized a Thermo Scientific LTQ ion trap coupled to an Agilent 1100 Nano-LC through an electrospray source. Peptides were desalted in 1% acetonitrile in 0.1% formic acid− water on a fused silica fritted capillary precolumn packed with Magic C18Aq RP silica (2 cm × 75 μm i.d., 5 μm, 200 Å; Michrom Bioresources, Auburn, CA) and separated over a 10 cm Magic C18Aq RP analytical column (75 μm i.d., 5 μm, 100 Å) using a 30 min gradient from 1% to 35% acetonitrile in 0.1% formic acid−water. Data were collected in data-dependent mode with 5 data-dependent MS/MS scans per full MS scan (m/z 250−2000) in centroid mode. Data-dependent CID MS/ MS scans were collected at 35% normalized collision energy with dynamic exclusion enabled. The dynamic exclusion parameters were as follows: mass width, m/z 3; repeat count, 1; repeat duration, 30 s; exclusion list size, 50; and exclusion duration, 180 s. The data were searched with X!Tandem (version 2007.07.01.2) allowing a maximum of two missed cleavages, +57 fixed modification on cysteines, potential +15.99 modification on methionines, potential +1 modification on asparagines, and ±3 Da error on parent monoisotopic mass. The results were scored with PeptideProphet (version 3, April 2004), and proteins were inferred with ProteinProphet (TPP version 3.4 SQUALL rev. 0, Build 200712110953). Proteins were selected with a minimum ProteinProphet probability of 0.9 and with peptides each having a minimum PeptideProphet probability of 0.75. This corresponds to a calculated theoretical error rate of approximately 1%. The MS2 spectra from the mouse xenograft data were searched against the combined International Protein Index (IPI) protein sequence databases (in the FASTA format) for human (version 3.29, June 2007),

Analysis of N-Linked Explanted 22Rv1 Culture Supernatant and Bulk Tumor Glycoproteomes

For analysis of secreted proteins from cell culture supernatant, 22Rv1 cells were grown under standard conditions until ∼75% confluent. Culture dishes were washed three times with PBS (pH 7.2) before cells were placed in FBS-free RPMI 1640 with 1% v/v P/S for 24 h. The media was collected, filtered of any cellular debris (Steriflip, Millipore, Bellerica, MA), and concentrated with multiple buffer exchanges into 100 mM ammonium bicarbonate using a stirred cell concentrator (Millipore) with a 10 kDa NMWL ultrafiltration membrane. Frozen bulk tumor from xenograft experiments was ground in liquid nitrogen in a cooled mortar and pestle to a fine powder. The powder was solubilized by tip sonication at 4 °C in 400 mM ammonium bicarbonate solution containing 8 M urea and 0.1% SDS (denaturing buffer). Insoluble debris was removed via centrifugation. Protein samples from culture supernatant or bulk tumor were subjected to glycopeptide enrichment as described by Tian et 2655

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

Figure 1. Workflow used for three proteomic studies. Cultured 22Rv1 cells were studied using biocytin hydrazide to capture extracellular membrane glycoproteins (left). Glycopeptides from serum-free culture media from these cells was captured using hydrazide agarose (center). The same technique was used on a protein lysate made from xenografted 22Rv1 cells (right).

plasma was then concentrated and buffer exchanged into Tris buffer (10 mM) via centrifugal filtration (NMWL 3 kDa filter, Millipore, Billerica, MA). Human control total protein tissue lysates of prostate tumor, adult normal prostate, and universal normal tissues were purchased from BioChain Institute (Hayward, CA) and Protein Biotechnologies (San Diego, CA). Twenty to thirty micrograms of protein lysate or depleted plasma were resolved by SDS-PAGE (NuPAGE, Invitrogen, Carlsbad, CA), electrophoretically transferred onto PVDF membranes (Immobilon-P, Millipore), blocked in nonfat milk, 5% v/v in Tris buffered saline with 0.05% Tween 20, (TTBS, Sigma, St. Louis, MO), and incubated with primary and respective HRP-conjugated secondary antibodies. Chemiluminescence detection was carried out using Western blotting detection reagents and high-performance chemiluminescence film from Amersham (GE Healthcare, Piscataway, NJ).

mouse (version 3.29, June 2007), and bovine and porcine trypsins. The MS2 spectra from the cell culture data were searched against the combined IPI protein sequence databases for human (version 3.29, June 2007) and bovine and porcine trypsins. Complete annotated data is provided in Supplemental Tables 1 and 2. Antibodies and ELISA Reagent

Monoclonal rat anti-human LAR (PTPRF) antibody and affinity-purified goat anti-human PTPRF antibody were purchased from R&D Systems (Minneapolis, MN). Both antibodies were made against the extracellular domain of LAR using purified NSO-derived recombinant human LAR (rhLAR; aa 27−1251). Purified NSO-derived rhLAR protein (aa 27−1251) was obtained as a prerelease reagent from R&D Systems. ImmunoPure mouse anti-goat IgG (H+L) horseradish peroxidase (HRP)-conjugated antibody (minimal cross-reactivity to mouse, human, and rabbit proteins) was purchased from Pierce (Rockford, IL). ELISA reagents including reagent diluent, substrate solution, stop solution, ELISA plates, and plate sealers were purchased from R&D Systems.

Enzyme-Linked Immunosorbent Assay (ELISA)

A quantitative ELISA was developed for identification of the PTPRF ectodomain using purified NSO-derived rhLAR as standard. Briefly, depleted control plasma and depleted plasma from xenograft tumor-burdened mice were diluted to 100 ng/ mL and incubated for 2 h on a plate coated with monoclonal rat antihuman LAR antibody. Non-depleted plasma samples were diluted 1:100 and incubated as above. Plates were washed 3 times with PBST (0.05% Tween 20 in phosphate buffered saline, pH 7.4 (Sigma)) and incubated 2 h with affinity-purified goat anti-LAR antibody. Following washing (as before), plates were incubated 45 min with affinity-purified mouse anti-goat IgG-HRP conjugated antibody, washed, and incubated 20 min with substrate before addition of stop solution to quench the reaction. The optical density of each well was determined

Western Blot

For analyses by Western blot, protein lysates from xenograft tissues were prepared using a tissue protein extraction reagent (T-PER, Pierce) supplemented with protease inhibitors (Halt; Pierce) based on the manufacturer’s protocol. Mouse plasma was depleted of the high-abundant proteins albumin, IgG, and transferrin using a MS-3 multiaffinity removal spin cartridge from Agilent (Santa Clara, CA) In brief, 25−30 μL of mouse plasma was loaded per use followed by washing, eluting, and column regeneration exactly according to the manufacturer’s instructions using buffers provided in the kit. The depleted 2656

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

Table 1. N-Linked Glycoproteins Identified in Overlapping 22Rv1 Xenograft Tumor Tissue and Cell Culture Data Sets accession no.a

proteinb

gene symbolc

functional classd

P12821 Q13740 P25311 P50895 P35613 Q9BY67 Q5ZPR3 Q08722 P08962 P10909 P16870 P07339 Q14126 P42892 Q04609 P13726 P10253 P15586 Q8NBJ4 P14625 P56199 P17301 P06756 P05556 P11047 O95297 O15031 P10586 P08195 Q13433 Q8NBI5 Q8IWA5 Q96NT5 P02786 P07996 P01033 Q9HD45 Q13641 O14773 P19075

Angiotensin-converting enzyme CD166 antigen Zinc-α-2-glycoprotein Lutheran blood group glycoprotein Basigin Cell adhesion molecule 1 CD276 antigen Leukocyte surface antigen CD47 CD63 antigen Clusterin Carboxypeptidase E Cathepsin D Desmoglein-2 Endothelin-converting enzyme 1 Glutamate carboxypeptidase 2 Tissue factor Lysosomal α-glucosidase N-Acetylglucosamine-6-sulfatase Golgi membrane protein 1 Endoplasmin Integrin α-1 Integrin α-2 Integrin α-V Integrin β-1 Laminin subunit γ-1 Myelin protein zero-like protein 1 Plexin-B2 Receptor-type tyrosine-protein phosphatase F 4F2 cell-surface antigen heavy chain Zinc transporter ZIP6 Solute carrier family 43 member 3 Choline transporter-like protein 2 Proton-coupled folate transporter Transferrin receptor protein 1 Thrombospondin-1 Metalloproteinase inhibitor 1 Transmembrane 9 superfamily member 3 Trophoblast glycoprotein Tripeptidyl-peptidase 1 Tetraspanin-8

ACE ALCAM AZGP1 BCAM BSG CADM1 CD276 CD47 CD63 CLU CPE CTSD DSG2 ECE1 FOHL1 (PSMA) F3 GAA GNS GOLM1 HSP90B1 ITGA1 ITGA2 ITGAV ITGB1 LAMC1 MPZL1 PLXNB2 PTPRF SLC3A2 SLC39A6 SLC43A3 SLC44A2 SLC46A1 TFRC THBS1 TIMP1 TM9SF3 TPBG TPP1 TSPAN8

Protease Binding protein Binding protein Receptor Binding protein Binding protein Binding protein Receptor Receptor Binding protein Protease Protease Binding protein Protease Protease Receptor Enzyme Enzyme Protein Binding protein Receptor Receptor Receptor Receptor Receptor ligand Binding protein Receptor Receptor Transporter Transporter Protein Transporter Transporter Receptor ligand Receptor ligand Binding protein Binding protein Binding protein Protease Receptor ligand

a Accession number according to UniProtKB/SwissProt. bComplete IPI lists and total numbers of peptide sequences identified are provided for both the 22Rv1-derived xenograft tumor tissue and cell culture data sets in Supporting Information. cOfficial human gene symbol according to HUGO Gene Nomenclature Committee (HGNC). dInferred from UniProtKB/Swiss-Prot.

methods were used. In the first, designed as an analysis of soluble media proteins, cells grown to near confluence were washed with PBS and placed in serum-free media that was collected after 24 h. After media concentration, proteins were trypsin-digested, and hydrazide-derivatized agarose resin was used to capture glycopeptides following oxidation as described.23 The second analysis using cultured 22Rv1 cells used an alternative method to capture plasma membrane glycoproteins through biotinylation with biocytin hydrazide.27 This reagent was covalently coupled to extracellular membrane proteins after a short period oxidization of extracellular sugar residues on live cells. After protein extraction and trypsin digestion, derivatized glycopeptides were captured on streptavidin beads. In both methods captured proteins were enzymatically released with PNGase-F and prepared for MS analysis. The latter method was employed as it provides

immediately using a microplate reader at 450 nm. Analyses were performed in triplicate with background subtraction.



RESULTS

Identification of N-Linked Glycoproteins from Human Prostate Cancer Tissues

The objective of this study was to enhance the discovery and specificity of prostate cancer biomarkers by use of N-linked glycocapture technology in a cell- and tissue-specific xenograft model, the 22Rv1 cell line.29 We analyzed this cell line under two distinct environments: under in vitro culture conditions and after xenograft into NOD-SCID mice. For both analyses, a workflow was employed based on the capture of N-linked glycopeptides on a bead using hydrazide chemistry and release of peptides by enzymatic deglycosylation with PNGase-F (Figure 1). For the studies of cultured cells, two different 2657

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

additional specificity because culture media containing bovine plasma is washed away and glycoproteins of the tumor cell extracellular membrane are preferentially captured away from other cellular glycoproteins. Glycopeptides were captured from 22Rv1 tumor xenografts using a similar method as that described for preparation of culture supernatant.23 All proteins were extracted in a denaturing buffer, reduced and alkylated, and digested to peptides. Glycopeptides were captured on hydrazide agarose and washed, and deglycosylated peptides were released using PNGase-F. Because the proteins identified from 22Rv1 cells cultured in vitro were to be compared with those identified in the xenograft tumor, the proteins identified from cultured 22Rv1 cells and defined media by biocytin hydrazide and hydrazide agarose, respectively, were combined into a single list. The LC−MS/MS analyses for the cell culture-derived data set produced approximately 200 unique and N-linked glycopeptides from 104 distinct glycoproteins (Supplemental Table 1 and 2) as determined by database searching with X! Tandem35 and filtering using a ProteinProphet36 probability cutoff of 0.9. A parallel analysis was performed on the explanted tumor from one of the mice xenografted with the 22Rv1 cells. The excised tumor was taken at 28 days following xenoplantation and flash frozen in liquid nitrogen. This analysis produced 194 unique N-linked glycopeptides representing 75 distinct human-specific glycoproteins (Supplemental Tables 3 and 4). Also identified were 70 murine-specific proteins and 26 proteins with assigned peptides that did not permit certain determination of the species of origin (Supplemental Tables 5 and 6). We sought to identify potential biomarker glycoproteins that could be used for diagnostic purposes from the 22Rv1 xenograft tissues. To provide specificity for this targeting, we chose to restrict further analysis of the xenograft-identified human proteins to only those that were also identified in the cultured cells where there was no opportunity for interference from murine proteins. This also eliminated proteins produced exclusively as a result of the in vitro culture conditions. Merging of the tissue culture and xenograft data sets revealed an overlap of 40 human glycoproteins (Table 1). On the basis of the UniProtKB/Swiss-Prot database, the majority of these proteins were functionally classified as binding proteins/receptor ligands (40%), cell membrane receptors (25%), or proteases/enzymes (20%) (Table 1). The distribution of cellular localization is similar to that previously reported for other glycoprotein-based studies,22 with 35 of the 40 proteins having annotated locations of the plasma or cellular membrane, extracellular matrix, or extracellular space (Figure. 2). The human glycoproteins identified in both the tissue culture and xenograft data sets include many strong candidates for potential clinical biomarkers. Prostate-specific antigen was not detected, likely because it has only a single asparagine glycosylation site that is immediately bounded by trypsincleavable arginine and lysine residues. Over 80% of the proteins in Table 1 have multiple references to neoplastic processes and metastasis, and over 40% of this group has specific references to prostatic neoplasms. The overlapping human proteins identified in both culture and xenograft experiments were further stratified for prostate specificity through their relative mRNA expression levels in various human tissues using the publically available GNF GeneAtlas U133A-Array database (http:// biogps.gnf.org). The mRNA for nine of the proteins is

Figure 2. Subcellular location of N-linked glycoproteins identified in overlapping 22Rv1-derived xenograft and cell culture data. Inferred from UniProtKB/SwissProt.

expressed at levels above median in normal prostate tissue, as well as a number of other tissues, indicating some level of prostate-specific expression. These are zinc-α-2-glycoprotein (AZGP1), CD47, Carboxypeptidase E (CPE), Lysosomal αglucosidase (GAA), golgi membrane protein 1 (GOLM1), folate hydrolase (prostate-specific membrane antigen) (FOLH1, PSMA), receptor-type tyrosine-protein phosphatase F (PTPRF), transmembrane 9 superfamily member 3 (TM9SF3), and tetraspanin-8 (TSPAN8). For three of these nine proteins, namely, PSMA, AZGP1, and PTPRF, the mRNA levels are, however, expressed far above median levels in the prostate, suggesting more specific tissue restriction. Of these, both PSMA37−42 and AZGP143−45 are well-known and, in the case of PSMA, well studied diagnostic markers for prostate cancer. Until now, however, no direct associations have been made between PTPRF, prostate cancer, and potential use as a diagnostic or prognostic marker. Given these interesting biological connections to the prostate and the success of our target generation in identifying proteins with annotation to malignant transformation in prostate cancer, we developed a set of assays to measure PTPRF abundance in our xenografted animals. Analysis of PTPRF Expression in 22Rv1 Cells and Xenografted Animals by Immunoassays

Using commercially available antibodies for PTPRF, we found that the extracellularly cleaved domain (E-subunit) of the PTPRF was clearly visible in 22Rv1 cells and 22Rv1 xenografts, but not in mouse tissues. The protein was also detectable in the plasma of animals xenografted with 22Rv1 cells, but not in control mouse plasma. Using conventional immunoblot, we first evaluated cultured 22Rv1 cells as well as a collection of lysates from other prostate cancer cell lines. Immunoblots of the cell lysates (Figure 3A) showed a fragment of ∼139 kD in all cases. This fragment corresponds to the extracellular cleaved domain of PTPRF, which has been reported to be generated by the sequential activities of furin and α-secretase,46 discussed below. While the bands are less intense in the non-androgenresponsive PC3 and DU145 cells (Figure 3A), the correct molecular weight of the visualized bands suggests but does not 2658

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

Figure 4. Western blot analysis of PTPRF E-subunit expression in xenograft lysates and human and mouse tissues. (A) Lanes 1−4: PTPRF, E-subunit is seen at the expected molecular weight in 22Rv1 xenograft tumor tissue lysates from four individual animals, 28−30 days post-injection with 7.5 × 105 22Rv1 cells. (B) Lane 1: human universal tissue lysate; lane 2: human liver tissue lysate; lane 3: normal human prostate tissue lysate; lane 4: human benign prostatic hyperplasia (BPH) tissue lysate; lane 5: human prostatic adenocarcinoma tumor tissue lysate; lane 6: normal human breast tissue; lane 7: human breast carcinoma tumor tissue lysate; lane 8: mouse universal tissue lysate; lane 9: mouse liver tissue lysate.

Figure 3. Western blot analysis of PTPRF E-subunit expression in prostate cancer cell lines and culture supernatants. (A) Expression of PTPRF in prostate cancer cell lines. PTPRF was observed in lysates from 22Rv1, LNCaP + R1881, LNCaP − R1881, PC3, and DU145 cells (lanes 1−5). (B) PTPRF was also observed in culture supernatants from 22Rv1 and LNCaP cells (lanes 1 and 2). Arrows = PTPRF; E-subunit, ∼139 kDa.

prove the presence of this protein. PTPRF is shown to be released into the media in 22Rv1 cells and LNCaP cells (Figure 3B) as expected. The difference in size of the fragments likely is the result of differences in either furin and α-secretase that produce it in these two cell lines or due to a secondary proteolytic activity present in LNCaP cells. PTPRF was not detected in a murine fibroblast line (Supplemental Figure 1). Immunoblotting of lysates made from explanted 22Rv1derived xenograft tumors also demonstrated the presence of a signal for the E-subunit of PTPRF (Figure 4A). A number of commercially available human tissues were simultaneously examined including a universal human lysate, normal prostate, benign prostatic hyperplasia (BPH), prostatic adenocarcinoma, and both normal and cancerous breast tissue lysates. As shown in Figure 4B, the Western blot results indicate far greater expression in the human prostate and breast tumor tissues versus normal prostate and breast tissue controls. The increased signal of the PTPRF E-subunit in prostatic adenocarcinoma was also clearly different from the signal present in the BPH protein lysate. Also notable on this blot (lanes 8 and 9), murine tissue lysates produced no appreciable signal for PTPRF. We then sought to investigate if the PTPRF E-subunit could be detected in plasma from mice grafted with 22Rv1 cells. Plasma was pooled from two sets of 3 animals implanted with 22Rv1 cells and sacrificed 30 days after xenografting . Because of differences in growth rates, the estimated tumor volumes ranged from 192 to 819 mm3, but tumors could be clearly classified as small or large. At the time of sacrifice, the average tumor size was approximately 300 mm3 in pool 1 and 700 mm3 in pool 2. The plasma was subjected to abundant protein depletion using a commercial column containing monoclonal antibodies directed against albumin, IgG, and transferrin. Western blot analyses of depleted plasma (Figure 5) demonstrated a clear signal from the PTPRF E-subunit at the expected molecular weight in the plasma from the 22Rv1

Figure 5. Western blot detection of PTPRF E-subunit in depleted 22Rv1 xenograft plasma. Lanes 1 and 2: depleted plasma from control mice; lanes 3 and 4: depleted plasma from 22Rv1 xenograft prostate tumor model mice. All samples are pools of 3 individual animals each with plasma pooled from animals with tumors of similar sizes. In lane 3, plasma is from mice with average tumor volume of 302.7 ± 96 mm3. In lane 4, plasma is from mice with average tumor volume of 708.7 ± 109.5 mm3. Clusterin-α was used as the plasma loading control.

xenograft animals versus no signal in the respective normal mouse plasma control. A biological replicate using a second cohort of animals, injected at a later date and with a larger pool size of 10 animals each, yielded identical findings with a clear detection of the PTPRF E-subunit in the 22Rv1 xenografted animals versus no detection in the control mice (Supplemental Figure 2). ELISA Quantitation of PTPRF in Plasma

We confirmed and elaborated further on the Western blot results through the development of a quantitative enzymelinked immunosorbent assay (ELISA) for the PTPRF Esubunit. This assay employed a monoclonal “capture” and polyclonal “detection” anti-PTPRF antibody together with extracellular domain-specific, NSO-derived recombinant human PTPRF (rhPTPRF; aa 27−1251) protein for standardization. The assay was linear and sensitive down to a concentration of approximately 100 pg/mL and was highly specific for PTPRF 2659

dx.doi.org/10.1021/pr201200n | J. Proteome Res. 2012, 11, 2653−2665

Journal of Proteome Research

Article

Figure 6. ELISA detection of PTPRF E-subunit in non-depleted 22Rv1 xenograft plasma. (A) Fourteen days post-injection of NOD SCID mice with 7.5 × 105 22Rv1 cells and mock injection of control mice. (B) Twenty-one days post-injection of NOD SCID mice with 7.5 × 105 22Rv1 cells and mock injection of control mice. Error bar = standard deviation of measurements done in triplicate. BDL = below detection limit.

cultured and xenografted 22Rv1 human prostate cancer cells. The workflow used was successful in identifying proteins with literature annotations indicating involvement in malignant transformation, with many of the targets having known associations with prostate cancer. In the set of human glycoproteins seen in both the tissue culture and xenograft data sets, 80% have multiple references to neoplastic processes and metastasis, and 40% have specific references to prostatic neoplasms. In addition, many of these proteins carry annotations for cellular processes particularly relevant to oncogenesis and metastasis. For instance, one-third of the proteins on the overlapping list are directly involved in the biological processes of cell adhesion, migration, and/or proliferation. Examples of the proteins involved in cell adhesion/migration with direct relevance and references to prostate cancer include CD166 antigen (ALCAM),47 thrombospondin-1 (THBS1),48,49 and the integrins α-2 (ITGA2) and β1 (ITGB1), which form the integrin complex α-2/β-1 integrin.50 Proteins that are annotated to be involved in cell proliferation in prostate cancer include metalloproteinase inhibitor 1 (TIMP1),51,52 immunoglobulin superfamily member 8 (IGSF8),53 and cell adhesion molecule 1 (CADM1),54 as well as the apoptosis/anti-apoptosis associated proteins, clusterin (CLU)55,56 and endoplasmin (HSP90B1).57 A subset of the identified proteins have previously been proposed for use as diagnostic markers. This is particularly useful for demonstrating the relevance of our analysis, as changes in their expression levels have been correlated with prostate cancer progression. This group of proteins includes CD166 antigen (ALCAM), 5 8 zinc-α-2-glycoprotein

(Supplemental Figure 3). There was no PTPRF detected in the control mouse samples. PTPRF was first quantified in the abundant-protein-depleted plasma from the pooled control and xenografted animals, which originated from two biological replicate studies described in the Western blot analysis above. Here, comparable levels of the PTPRF E-subunit were detected in depleted plasma from the xenografted animals in both studies (Supplemental Figure 4). We then sought to determine if significant levels of the PTPRF E-subunit could also be detected in non-depleted plasma. We found that the assay was indeed successful using non-depleted plasma and therefore unaffected by the presence of murine albumin, IgG, and transferrin (Figure 6A). These studies were performed using the plasma of individual mice at 2 and 3 weeks post-xenografting. After 2 weeks (Figure 6A), circulating levels of PTPRF E-subunit were detected and measured in plasma at ∼9−72 ng/mL (average ∼33 ng/mL, 8 individual mice). At three weeks post-injection (Figure 6B), the circulating level of PTPRF E-subunit had increased approximately 10-fold to ∼128−855 ng/mL (average ∼305 ng/mL, 8 individual mice). This increase in the circulating levels of the PTPRF E-subunit was determined to be statistically significant with a two-sample, unequal variance t test and a p value of