Discovery proteomics defines androgen-regulated glycoprotein networks in prostate cancer cells, as well as putative biomarkers of prostatic diseases

Supraphysiologic androgen (SPA) inhibits cell proliferation in prostate cancer (PCa) cells by transcriptional repression of DNA replication and cell-cycle genes. In this study, quantitative glycoprotein profiling identified androgen-regulated glycoprotein networks associated with SPA-mediated inhibition of PCa cell proliferation, and androgen-regulated glycoproteins in clinical prostate tissues. SPA-regulated glycoprotein networks were enriched for translation factors and ribosomal proteins, proteins that are known to be O-GlcNAcylated in response to various cellular stresses. Thus, androgen-regulated glycoproteins are likely to be targeted for O-GlcNAcylation. Comparative analysis of glycosylated proteins in PCa cells and clinical prostate tissue identified androgen-regulated glycoproteins that are differentially expressed prostate tissues at various stages of cancer. Notably, the enzyme ectonucleoside triphosphate diphosphohydrolase 5 was found to be an androgen-regulated glycoprotein in PCa cells, with higher expression in cancerous versus non-cancerous prostate tissue. Our glycoproteomics study provides an experimental framework for characterizing androgen-regulated proteins and glycoprotein networks, toward better understanding how this subproteome leads to physiologic and supraphysiologic proliferation responses in PCa cells, and their potential use as druggable biomarkers of dysregulated AR-dependent signaling in PCa cells.

coupled to changes in protein glycosylation in PCa cells, in particular, how androgen levels influence changes in the glycosylation of growth factor receptors that harbor N-and O-linked oligosaccharides or intracellular proteins that contain the O-linked N-acetylglucosamine (O-GlcNAc) moiety 42,43 . The LNCaP cell line, which models PCa, was selected for the glycoprotein profiling experiment because the inhibition of SPA-mediated proliferation in this context is well-documented 30 . As shown previously [21][22][23] , physiologic levels of androgen (i.e., 1 nM R1881) stimulated maximal proliferation of LNCaP cells, whereas SPA (i.e., 10 nM R1881) attenuated their proliferation (Supplemental Fig. 1A,B). These findings showed that androgenic responses in LNCaP cells are biphasic and dose-dependent, justifying their use as an experimental model for studying how the inhibition of SPA-mediated proliferation influences protein glycosylation in PCa cells.
As an orthogonal approach to verify the observed changes in protein expression, we used semi-quantitative western blots. Proteins were selected for verification based upon the commercial availability of western blot-grade antibodies. They were DNA-dependent protein kinase catalytic subunit (PRKDC), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), sodium/potassium-transporting ATPase subunit beta-1 (ATP1B1), and clathrin heavy chain 1 (CLTC) (Fig. 1C, Supplemental Fig. 2). With the exception of PRKDC at 10 nM R1881, western blots across the doses of androgen tested were roughly concordant with protein abundance changes as determined by dMS (Table 1). These findings showed that the glycoproteomic profiling experiment has the power to detect changes in the expression of microsomal glycoproteins in androgen-treated LNCaP cells.
To identify molecular associations between the androgen-mediated proliferation responses in LNCaP cells and specific biological pathways, we evaluated the enrichment and de-enrichment of glycoproteins across treated samples. Biological pathways that were over-or under-represented across the tested androgen concentrations were identified using the WEB-based Gene SeT AnaLysis Toolkit (WebGestalt) program ( Fig. 2A). Notably, ten of the top-ranked biological pathways were conserved across each experimental sample (i.e., 0, 0.1, 1.0, 10 nM R1881), highlighting similarities in the molecular composition of lectin-enriched proteomes profiled in the glycoproteomic experiment ( Fig. 2A). TreeView visualization of lectin-enriched proteomes shows differences in the enrichment of biological pathways in experimentally-treated samples ( Fig. 2A) 46 . The biological pathways included PI3K-Akt signaling, Proteoglycans in cancer, Leukocyte transendothelial migration, Tight junctions, AMPK signaling, Glycolysis/Gluconeogenesis, Phagosome, Pathways in cancer, Estrogen signaling, and Prostate cancer ( Fig. 2A). Despite conservation of the biological pathways, different clusters of protein-protein interactions (PPIs) were detected at each dose of androgen, highlighting potential differences in the glycosylation status of PPIs (Supplemental Fig. 3). Importantly, androgens caused dose-dependent changes in the expression of components of specific biological pathways. For example, components of the Glycolysis/Gluconeogenesis pathway were more commonly expressed at higher doses of androgen ( Fig. 2A) Fig. 6). Moreover, only ten clusters (i.e., c2-c7, c15, c22, c23, and c26) from our dataset matched the twelve theoretical clusters (i.e., T4, T6, T9, T12, T19, T20,  T21, and T26) (Fig. 2B) (Supplemental Fig. 6). Given that the eight glycoprotein clusters were associated with androgen-mediated inhibition of proliferation, we tested glycoprotein clusters for functional relationships with a specific biological process in response to SPA. WebGestalt analyses identified overrepresented biological pathways in five of the eight glycoprotein clusters (i.e., c2, c5, c7, c22, and c26) (Fig. 2C) (Supplemental Excel file 2). The biological pathways included SRP-dependent cotranslational protein targeting to membrane (i.e., clusters 2 and 26), Negative regulation of ubiquitin protein ligase activity (i.e., cluster 5), Positive regulation of DNA damage response-p53 class mediator, and Neutrophil degranulation (Fig. 2C). These results suggest that SPA-mediated

ENTPD5 is an androgen-regulated glycoprotein in LNCaP cells and is differentially expressed
in clinical prostate tissue. We sought to leverage the glycoproteomic profiling strategy to analyze clinical prostate tissues for differences in glycoprotein expression between BPH, localized PCa, and metastatic PCa samples and, where available, paired NAT. A total of 37 fresh-frozen prostate tissue samples, which represented 29 cases (8 NAT, 10 BPH, 13 localized PCa, and 6 mPCa samples), were collected and processed for the glycoproteomic profiling experiments (Supplemental Table II-IV) (Fig. 3) (Supplemental Figs. 7-9). Our initial glycoproteomic profiling experiments of PCa tissues and paired NAT (i.e., R90197381-N/T, R42521246-N, R93776568-N/T, R03188233-N/T) involved extensive peptide fractionation using strong cation exchange (SCX), high-performance liquid chromatography (HPLC) (e.g., 21 SCX fractions). This fractionation strategy resulted in the quantification of > 2000 proteins from a single tissue sample (Supplemental Fig. 10). However, due to experimental design constraints, the remaining tissue samples were fractionated using SCX spin columns (e.g., 7 SCX fractions). Although this strategy reduced the number of quantified proteins by ~ threefold, meaningful glycoprotein expression data were obtained from these prostate tissue samples (Supplemental Fig. 10, Supplemental Excel file 3).
To identify glycoprotein signatures that distinguish between to BPH, PCa, and mPCa, we compared glycoprotein expression across all tissue samples. Supervised hierarchical clustering uncovered sets of glycoproteins that are upregulated in BPH relative to localized PCa and mPCa and, conversely, signatures of sets that are upregulated in localized PCa and mPCa relative to BPH (Fig. 3C). For example, lymphocyte-derived immunoglobulins were more abundant in BPH samples than PCa and mPCa samples (Fig. 3C, lower image), consistent with the notion that inflammation underlies the initiation and progression of BPH 47,48 . In contrast, glycoproteins involved in proteasome function (e.g. PSMD4) 49,50 , N-linked glycosylation (e.g., ENTPD5) 51-54 , glycan metabolism (e.g., NAGA) 55 , and receptor trafficking (e.g., LAMP1) 56 were upregulated primarily in localized PCa and mPCa ( Fig. 3C, upper image). These results show that our glycoproteomic profiling experiments have the power to detect differences in glycoprotein expression in clinical prostate tissue specimens.
Given that androgen-regulated gene-expression programs are frequently dysregulated in early-stage PCa 3,57 , we decided to use androgen-regulated glycoproteins in LNCaP cells as biomarkers to guide the discovery of candidate androgen-regulated glycoproteins in clinical prostate tissue samples (Fig. 3D). Supervised hierarchical clustering of glycoproteins from LNCaP, as well as BPH, localized PCa, and mPCa samples, uncovered many glycoprotein clusters between samples (Fig. 3D). We focused on LNCaP glycoproteins whose levels changed in response to androgens (i.e., androgen-mediated increases or decreases) and that were upregulated in localized PCa and/or mPCa samples. One such protein was ENTPD5. Its overall levels were regulated by androgens in LNCaP cells and were higher in both localized PCa and mPCa samples than in BPH samples (Fig. 3D). Moreover, levels of glycosylated ENTPD5 were consistently higher in PCa tissues than in NAT (Fig. 3D). Notably, they were highest in cells treated with 1 nM androgen (Fig. 4B), the concentration of androgen that induced maximal proliferation of LNCaP cells, and undetectable in cells treated with 10 nM androgen, the dose that antagonized the proliferation of LNCaP cells (Supplemental Fig. 1). Cluster analysis revealed that levels of glycosylated β-catenin (CTNNB1), which is also regulated by androgens in LNCaP cells 58 , were higher in localized PCa and mPCa samples than BPH samples (Fig. 3D). These findings support the validity of identifying candidate androgenregulated glycoproteins in LNCaP prostate-tumor cells based on their overexpression in cancerous prostate tissue.
ENTPD5 promotes the proliferation of cancer cells and is frequently overexpressed in cancerous tissues 51,60 . To verify that androgens regulate the expression of ENTPD5 in prostate-tumor cells, LNCaP cells in AD growth medium were challenged with various concentrations of androgen (24 h), generated whole-cell lysates, and performed Western blot analysis of ENTPD5. Overall protein levels increased up to 1 nM androgen, but were lower at 10 nM androgen (Fig. 4A, left panel). We also probed for glycosylated ENTPD5, using lectin-enriched microsomes. Exposure to up to 1 nM androgen led to increased glycosylation of ENTPD5, whereas exposure to 10 nM androgen caused a noticeable reduction in glycosylation (Fig. 4A, right panel). At the 10 nM dose, the reduction in glycosylated ENTPD5 was greater than the reduction of total ENTPD5, showing that this modification is suppressed by SPA in LNCaP cells. Overall, the results verified that ENTPD5 is an androgen-regulated glycoprotein in LNCaP cells.
These findings prompted us to determine whether ENTPD5 expression is also transcriptionally regulated by androgens, which would support its designation as an ARG in LNCaP cells. Androgen exposure caused a dose-dependent increase in ENTPD5 gene-transcripts at up to the 1 nM concentration, and a slight reduction in ENTPD5 gene expression was observed at 10 nM androgen by qPCR analyses (Fig. 4B, Supplemental  Fig. 11). Importantly, the KLK3 gene, which is a direct target of AR-dependent transcription and a canonical ARG in androgen-responsive prostate cancers [61][62][63] , showed the same biphasic transcriptional response to androgen (Fig. 4B). This result prompted us to search for genomic AR binding sites in ENTPD5; their presence would provide further evidence that it is a downstream target gene of AR. Thus, genomic AR binding sites identified in a previous chromatin immunoprecipitation sequencing (ChIP-Seq) study were re-examined in androgen (i.e., R1881)-treated LNCaP cells 4 . Notably, ChIP-seq signals at ENTPD5 localized to introns 2 (chr14:74,482,139-74,482,531, 375 base pairs) and 6 (chr14:74,458,755-74,459,147, 125 base pairs) (Supplemental Fig. 12A) 4 . Although the MEME motif program failed to identify canonical androgen response element (ARE) sequences (e.g., AGA ACA NNNTGT TCT ) 64 in either DNA segment, a 15-bp consensus motif, CCASBAN-NYCC AGC Y, which was the longest and most abundant (7 copies), was detected in both introns (Supplemental Fig. 12A). The Tomtom motif comparison tool showed that this sequence has strong homology to consensus motifs in ETS-family transcription factors ELK4 and ETS1, as well as to motifs in the ladybird homeobox 2 www.nature.com/scientificreports/ (LBX2) transcription factor (Supplemental Fig. 13B). Notably, AR and ETS1 physically interact and coregulate a subset of ARGs in LNCaP cells 65 . Interestingly, evaluation of clinical prostate cancer datasets at the cBioPortal for Cancer Genomics showed that ENTPD5 is infrequently amplified (~ 3%) in a subset of CRPCs ( Fig. 4C) 66,67 , which suggest ENTPD5 overexpression may have some role in the progression of late-stage PCa. Our findings show that ENTPD5 is an ARG that is transcriptionally regulated, either directly or indirectly, through complex interactions between AR and auxiliary transcription factors in LNCaP cells.

Discussion
The recent success of bipolar androgen therapy (BAT), which restores ADT sensitivity to a subset of CRPCs, has spurred greater clinical interest in the treatment of CRPC with SPA [68][69][70][71] . Additionally, a recent study showed that SPA suppressed proliferation in patient-derived xenografts of CRPC 72 . SPA-mediated inhibition of the proliferation of PCa cells is caused by AR-dependent transcriptional mechanisms involving the repression of the transcription factors that underlie cell growth (e.g., c-MYC, E2F) 35 , the upregulation of cell-cycle inhibitors (e.g., p27, p21, Skp2) 73,74 , the induction of terminal differentiation (e.g., APRIN, PLZF) [75][76][77][78][79][80] , the repression of DNA replication genes 81 , the repression of AR and AR variants 37,81,82 , and activation of cell senescence through the repression of E2F-regulated genes 36,39 . Transcription-independent mechanisms underlying SPA-mediated inhibition of proliferation by PCa cells include activation of the DNA double-strand break damage response (DDR) 83 and the inhibition of DNA licensing by AR stabilization at pre-replication complexes during M phase 34,84 . In this glycoproteomics study of LNCaP cells, we found that SPA coordinates the expression of glycoproteins involved in the biological pathways SRP-dependent protein cotranslational targeting to membrane, Negative regulation of ubiquitin protein ligase activity, and Positive regulation of DNA damage response. How glycosylation status is coupled to SPA-mediated inhibition of PCa-cell proliferation will require further exploration. Given the enrichment of ribosomal proteins among the SPA-enriched glycoprotein networks, and the fact that ribosomal proteins are targeted for O-GlcNAcylation during protein translation and stress-granule (SG) formation 85-87 , we speculate that they are O-GlcNAcylated. Given that O-GlcNAcylation regulates translation initiation, stabilizes nascent polypeptide chains during cotranslation, and triggers SG disassembly for the translation of stress mRNAs [88][89][90] , it is possible that SPA elicits a stress response in PCa cells that antagonizes proliferation through the formation of stalled translation preinitiation complexes and SGs 91 . Previous studies showed that SPA induces dsDNA breaks in PCa cells 70,83 , suppresses the gene expression proteins that drive DDR and homologous recombination in CRPC patient-derived xenografts (PDXs) 72 , and mediates an extreme response to BAT in a patient who harbors a germline missense mutation in the serine/threonine protein kinase-encoding ATM gene and a frameshift mutation in the breast cancer gene BRCA2 92 . As reported in the current study, SPA enriches for the biological pathway that represents Positive regulation of DNA damage response in LNCaP cells, providing further experimental support for previously reported physical and functional interactions among AR, topoisomerase II (TOP2), and the DNA repair machinery in human PCa 83   localized PCa, and 6 metastatic PCa) were clustered using Pearson correlation distances between averages using GenePattern. Magnified clusters exemplify those with increased protein expression in cancer vs. BPH samples (i.e., top right panel) and those with increased protein expression in BPH vs. cancer samples (i.e., bottom right panel). Expression scale for fold expression was 0 (i.e., black color, undetectable expression) up to 5 (i.e., bright red, saturated expression). (D) Cluster analysis of glycoproteins in LNCaP cells, BPH, localized PCa and metastatic PCa tissue normalized to externally spiked BSA. The glycoproteins in LNCaP cells (i.e., 0, 0.1, 1, and 10 nM R1881-treated cells) and glycoproteins from BPH, localized PCa, and metastatic PCa samples (i.e., 37 tissue samples, including: 10 BPH; 5 localized PCa; 6 metastatic PCa -100 μg and 8 paired TT/NAT samples-50 μg) were clustered using Pearson correlation distances between averages using GenePattern. Magnified cluster represents ENTPD5 expression in LNCaP cells and metastatic tissue samples. The expression scale for the clustergram denotes the relative protein abundance, ranging from 0 for no protein expression (black) to saturation (red) at fivefold expression. Clustered androgen-regulated glycoproteins in LNCaP cells whose expression increased in metastatic tissue samples are highlighted in the magnified clustergram.  www.nature.com/scientificreports/ degradation of proteins that are cotranslated proteins by the UPS 89 . Our findings justify further scientific exploration of whether a functional relationship exists between O-GlcNAcylation, the UPS, and SPA-mediated inhibition of PCa cell proliferation. The elucidation of androgen-regulated gene-expression programs that control the proliferation of PCa cells has been the focus of many research studies seeking to shed light on how defects in the expression of ARGs contribute to the development and/or progression of human PCa 27,28,97-100 . We have expanded beyond this genomic perspective and shown, for the first time, how androgens elicit dose-dependent changes in glycoprotein expression in LNCaP PCa cells. Androgens are known to transcriptionally regulate protein glycosylation pathways in PCa cells, directly through AR-dependent mechanisms 101 . Twenty-five ARGs in the protein glycosylation pathway have been shown to encode enzymes that act at different steps of the hexamine biosynthesis pathway (HBP) 102 , N-and O-glycan biosynthesis, and chondroitin sulfate (CS) synthesis 103 . Our glycoproteomic profiling experiment shows that the enzyme ENTPD5, which promotes N-glycosylation and ER protein folding, and also contributes to the Warburg effect in PCa cells 51 , is an androgen-regulated glycoprotein in LNCaP cells. Moreover, we present experimental data showing that ENTPD5 expression is regulated transcriptionally (AR-dependent) and potentially post-transcriptionally (glycosylation) by androgens in this cell type. Of note, the overexpression of ENTPD5 correlates with AKT activation in primary tumors, and ENTPD5 expression is required for the glycosylation of growth factor receptors (e.g., EGFR, Her2, IGF-IRβ) in PCa cells 51 . In LNCaP cells, levels of glycosylated ENTPD5 were reduced following SPA treatment, suggesting that this modification might be functionally coupled to SPA-mediated inhibition of proliferation. Thus, it is possible that a reduction in glycosylated ENTPD5 leads to a reduction in glycosylated growth factor receptors and to a subsequent reduction in growth factor-mediated proliferation signals in PCa cells. Experiments to clearly establish a functional relationship between ENTPD5 glycosylation and SPA-mediated inhibition of proliferation in PCa cells requires further scientific inquiry.
Although this glycoproteomics study uncovered novel androgen-regulated proteins in both PCa cells and clinical prostate tissues, it has several experimental limitations. Firstly, it does not define site-level mass spectrophotometric determination of N-linked and O-linked residues. The LWAC method enriched for detergentsolubilized microsomal glycoproteins in LNCaP cells and clinical tissue under non-denaturing conditions. Thus, the glycoproteomic output might have been contaminated with non-glycosylated proteins that bound to bonafide N-and O-linked glycosylated proteins through piggyback interactions. The incorporation of methods using solid phase extraction of glycopeptides (SPEG) followed by either the enzymatic release of N-linked glycopeptides 104,105 or the chemical release or O-linked glycopeptides 106,107 would facilitate site-level mass spectrophotometric determination of N-linked and O-linked residues in lectin-affinity purified glycoprotein samples respectively. Secondly, the accuracy of resection of tissue sections was controlled by a uni-core cutting tool. The use of laser capture microdissection (LCM) methods would improve the accuracy of this aspect of the study. For example, our comparative glycoprotein expression analyses of BPH and PCa samples was limited because BPH samples are heterogeneous in cellular composition, and the samples likely contain smooth muscle cells, fibroblasts, and both secretory and basal epithelial cells. LCM methods would afford greater accuracy in the resection of epithelial cells from these heterogenous samples so that true expression differences in androgen-regulated glycoproteins between epithelial cell-types between BPH and PCa samples could be determined. Thirdly, the number of available clinical samples used for glycoprofiling limited the power of the study. Increasing the number of samples would increase the power to detect candidate biomarkers in diseased tissue samples. Notwithstanding these experimental limitations, we anticipate that the glycoproteomic findings presented here will provide new insights into how androgens regulate glycoprotein networks in PCa cells. In addition, this subpopulation of proteins might represent a rich resource of candidate biomarkers of cellular diseases that affect the prostate gland.

Methods
Materials. LNCaP cells were from American Type Culture Collection; Dulbecco's Phosphate Buffered Solution, phenol red-deficient RPMI 1640 media, 10X Glutamax, and 10X penicillin and streptomycin were from Invitrogen; normal and charcoal stripped fetal bovine serum were from Hyclone Laboratories (Logan, UT) (Invitrogen); protease inhibitor cocktail tablets and dithiothreitol (DTT) were from Thermo Scientific Pierce; Wheat germ agglutinin and Concanavalin A-agarose beads were from Vector Laboratories Inc. (Burlingame, CA). Tissue biopsy punch tools, sugars, solvents (i.e., non-organic and organic), and all other chemicals were from Sigma-Aldrich. Western blot antibody reagents included: a rabbit polyclonal antibody to prostate-specific antigen (PSA) from DAKO (catalog #A0562)(Carpinteria, CA), a mouse monoclonal to Hemagglutinin A (HA) (catalog #2367), and rabbit polyclonal antibodies to DNA-dependent protein kinase catalytic subunit (PRKDC) (catalog #4602) and glyceraldehyde 3-phosphate dehydrogenase (GAPDH)(catalog #2118) were from CST (Danvers, MA), mouse monoclonal antibodies to sodium/potassium (Na/K) ATPase (ATP1B1)(catalog #sc-21712) and AR (catalog # AR441) from Santa Cruz Biotechnology (Santa Cruz, CA), a mouse monoclonal antibody to clathrin heavy chain (CLTC)(catalog #610499) from BD Biosciences (San Jose, CA), a mouse monoclonal antibody to ectonucleoside triphosphate diphosphohydrolase 5 (ENTPD5)(catalog # 743512) from R&D Systems (Minneapolis, MN). The compound methyltrienolone (R1881) was purchased from Perkin Elmer (Waltham, MA), and Enzalutamide (catalog #S1250) was purchased from Selleckchem (Houston, TX). Sequence-grade trypsin (catalog #V5113) was purchased from Promega (Madison, WI). The bicinchoninic acid (BCA) protein assay kit (catalog #23228), Slide-A-Lyzer dialysis cassettes (catalog #'s 66373, 66380), Dulbecco's Phosphate Buffered Solution (PBS)(catalog #14040141), SuperScript® III First-Strand Synthesis kit (catalog #18080400), Oligofectamine (catalog # 12252011), Lipofectamine 2000 Transfection Reagent (catalog # 11668027), and 4-12% Bis-Tris gels (catalog #NP0336BOX), and the CyQUANT Cell Proliferation Assay Kit (catalog # C7096) were from ThermoFisher Scientific (Waltham, MA ). C18 (catalog #SEM SS18V) and strong cation exchange Clinical samples. 37 fresh-frozen prostate tissue samples, representing 29 human subjects, were accrued from academic and commercial sources. The first set of tissue samples was derived from four human subjects and consisted of paired samples of tumor tissue (TT) and normal adjacent tissue (NAT) accrued by the University of Iowa Tissue Core (Iowa City, IA) (Supplemental Table III). The second set of tissue samples was obtained from fifteen subjects and consisted of five paired TT and NAT samples, and 10 BPH samples accrued from Proteogenex (Culver City, CA). The third set of tissue samples was from 10 subjects and composed of TT accrued from Bioserve Biotechnologies LTD (Beltsville, MD). Aside from 2 of the TT samples, all others had a Gleason score of ≥ 7. This included 9 stage-II tumor samples, 8 stage-III tumor samples, and 3 stage-IV samples (Supplemental Table II and III). All tissue sample hematoxylin and eosin (H&E) stained slides were evaluated by a certified clinical pathologist, and regions of TT and NAT were denoted on H&E slide to guide the resection of frozen tissue samples. For all BPH samples, the entire tissue sample was processed for glycoprotein extraction and processing for mass spectrometry analysis. Ethical approval and consent to participate: Tumor samples were obtained under informed consent after approval by the University of Iowa Institutional Review Board: IRB#200907702 and #201103721 protocol. All data collection, processing, and consenting process were executed after approval by the IRB at the University of Iowa. Also, all methods were performed in accordance with the relevant guidelines and regulations of the IRB.
Tissue protein extraction. A 5 mm Uni-core cutting tool was used to resect TT and NAT samples (Fig. 3, Supplemental Fig. 7). Cored as well as BPH samples were suspended in ice-cold PBS (containing CaCl 2 and MgCl 2 ), briefly vortexed, and centrifuged at 4 °C at 557 × g for 5 min to remove non-tissue contaminants, including optimum cutting temperature (OCT) medium, blood, and cellular debris. Each sample was resuspended in membrane extraction buffer (MEB) (20 mM Tris, 150 mM NaCl, 0.1 mM CaCl 2 , 0.1 mM MnCl 2 , 1 × Halt Protease Inhibitor Complex, 10 mM DTT, 5 mg/ml Digitonin) and rotated end-over-end overnight at 4 °C. The samples were centrifuged at 1200 × g for 8 min at 4 °C and the collected supernatants were centrifuged for another 1 h at 100,000 × g at 4 °C. The supernatants were collected and quantified by silver-stained gel analysis; detergent-solubilized LNCaP whole-cell lysates were used as references for standard for protein quantification (Supplemental Figs. 8 and 9).

Enrichment of glycosylated microsomal proteins from LNCaP prostate-tumor cells. Large-scale
experiment. LNCaP prostate-tumor cells grown in 500 cm tissue culture plates for 72 h in androgen-depleted (AD) growth medium (phenol-red free RPMI 1640 + 10% charcoal-stripped FBS) were exposed to vehicle (EtOH) or synthetic androgen R1881 at 0.1, 1, and 10 nM for 24 h. Cells were washed twice with ice-cold PBS, scraped from the plates, and centrifuged for 5 min at 1800 rpm at 4 °C. The supernatants were decanted, and cell pellets were resuspended into hypotonic lysis buffer (HLB) (10 mM Hepes, 1.5 mM MgCl 2 , 10 mM KCl, pH 7.9, 5 mM DTT-Sigma, and 1 × Halt Protease Inhibitor Complex). The hypotonic samples were incubated on ice for 10 min and subjected to nitrogen cavitation (i.e., 100 psi) for 5 min. Nitrogen-cavitated samples were centrifuged at 600 × g for 20 min at 4 °C to pellet out intact nuclei and unbroken cells. The supernatants were collected and centrifuged at 100,000 × g for 3 h at 4 °C. The supernatant (i.e., cytosolic protein fraction) was removed and the crude microsomal pellet, which contained intact organelles and membrane microsomes, was solubilized in MEB. The samples were rotated end-over-end for 16 h at 4 °C, and then subjected to 100,000 × g centrifugation for 1 h at 4 °C to remove detergent-insoluble particulate matter. Collected supernatants were quantified by silverstained gel analysis, using detergent-solubilized LNCaP whole-cell lysates as a reference standard for protein quantification (Supplemental Fig. 8). The digitonin-solubilized samples (i.e., 10 mg protein) were incubated and rotated end-over-end overnight with a mixture of wheat germ agglutinin (WGA) and concanavalin A (ConA) agarose beads (Vector Laboratories Inc., Burlingame, CA) in MEB at 4 °C. Non-specific, non-glycosylated proteins were removed by three consecutive washes with MEB. Glycosylated proteins were competitively eluted by incubating each sample with MEB supplemented with 500 mM N-acetyl-D-glucosamine, 200 mM α-methyl mannose, 200 mM α-methyl glucose, and 200 mM α-D-mannose for 30 min at 4 °C. Supernatants were collected and loaded into 10 kDa cutoff dialysis cassettes and subjected to overnight dialysis in urea buffer (UB) (8 M Urea, 50 mM Tris, and 100 mM β-mercaptoethanol, pH 8.5). Collected supernatants were quantified by silverstained gel analysis, using detergent-solubilized LNCaP whole-cell lysates as a reference standard for protein quantification (Supplemental Fig. 9).
Small-scale experiment. LNCaP cells were treated and processed for glycoprotein enrichment exactly as described in the large-scale experiment above, except that LNCaP cells were grown in 10 cm 2 tissue culture dishes for this experiment. Liquid chromatography mass spectrometry. Desalted, tryptic peptide samples were dissolved in mass spectrometry loading buffer (1% acetic acid, 1% acetonitrile) and analyzed by nanoliquid chromatography-tandem mass spectrometry using an Agilent 6520 Accurate-Mass Quadropole Time-of-Flight mass spectrometer interfaced with an HPLC Chip Cube. The samples were loaded onto an Ultra High Capacity Chip (500 nL enrichment column, 75 μm × 150 mm analytical column). LC-MS/MS analysis was performed using a 120min gradient ranging from 1.5 to 32% buffer C (100% acetonitrile, 0.8% acetic acid). Full MS (MS1) data was acquired using a mass range of 400-1250 m/z and acquisition rate of 1 spectra/second. From these data, an ion preferred list was generated with Agilent MassHunter Qualitative Software, with the settings of 400-1250 m/z, 2 + and 3 + charge states, and spectra with 2 or more ions. Directed mass spectrometry (dMS) was performed with the following settings: a maximum of 10 ions per cycle, a narrow isolation width (~ 1. Post search. The data were filtered based upon the Spectrum Mill Forward-Reverse Score threshold of 1.2, Rank 1-2 score threshold of 2, Score threshold of 3 and %SPI threshold of 30. The spectral intensity of the identified protein was normalized to the ratio of the tumor over normal experiments of identified bovine serum albumin carboxymethylated cysteine (C = 58.0055 AMU) peptides total intensities for equal sample loading. Fig. 2A. Enriched network pathway analyses were performed using the WEB-based Gene SeT AnaLysis Toolkit "WebGestalt". Spectrum Mill identified proteins in vehicle and androgen-treated groups (i.e., 0, 0.1, 1.0, and 10 nM R1881) were uploaded individually into the WebGestalt program. UniProt IDs were subjected to enriched pathway analyses with KEGG Pathway as the function database and default parameters 108 . The enriched network pathways and computed pvalues < 0.05 were calculated as −log(p-value) and transformed as a heatmap in the TreeView software.

Supervised clustering of lectin-enriched proteome between LNCaP cells and clinical tissue samples.
GenePattern 109 supervised clustering was performed on protein intensity values ranging from 1.03 × 10 4 to 2.77 × 10 9 . The values were log transformed and uploaded into GenePattern for hierarchical clustering. The columns (samples) were fixed and the rows (identified proteins) were clustered using Pearson correlation as the clustering criterion with pairwise complete-linkage as the hierarchical clustering method. The exported clustered results were visualized in the TreeView software with the maximum value capped to 5 on the scale bar to easily visualize changes in protein expression across samples.