The Complexity and Dynamics of the Tissue Glycoproteome Associated With Prostate Cancer Progression

The complexity and dynamics of the immensely heterogeneous glycoproteome of the prostate cancer (PCa) tumor microenvironment remain incompletely mapped, a knowledge gap that impedes our molecular-level understanding of the disease. To this end, we have used sensitive glycomics and glycoproteomics to map the protein-, cell-, and tumor grade–specific N- and O-glycosylation in surgically removed PCa tissues spanning five histological grades (n = 10/grade) and tissues from patients with benign prostatic hyperplasia (n = 5). Quantitative glycomics revealed PCa grade–specific alterations of the oligomannosidic-, paucimannosidic-, and branched sialylated complex-type N-glycans, and dynamic remodeling of the sialylated core 1- and core 2-type O-glycome. Deep quantitative glycoproteomics identified ∼7400 unique N-glycopeptides from 500 N-glycoproteins and ∼500 unique O-glycopeptides from nearly 200 O-glycoproteins. With reference to a recent Tissue and Blood Atlas, our data indicate that paucimannosidic glycans of the PCa tissues arise mainly from immune cell–derived glycoproteins. Furthermore, the grade-specific PCa glycosylation arises primarily from dynamics in the cellular makeup of the PCa tumor microenvironment across grades involving increased oligomannosylation of prostate-derived glycoproteins and decreased bisecting GlcNAcylation of N-glycans carried by the extracellular matrix proteins. Furthermore, elevated expression of several oligosaccharyltransferase subunits and enhanced N-glycoprotein site occupancy were observed associated with PCa progression. Finally, correlations between the protein-specific glycosylation and PCa progression were observed including increased site-specific core 2-type O-glycosylation of collagen VI. In conclusion, integrated glycomics and glycoproteomics have enabled new insight into the complexity and dynamics of the tissue glycoproteome associated with PCa progression generating an important resource to explore the underpinning disease mechanisms.


In Brief
Integrated glycomics and glycoproteomics with reference to the established Tissue Atlas were used to uncover the cell-, protein-, site-, and structurespecific N-and O-glycosylation in prostate cancer tissues spanning five disease grades relatively to benign prostatic hyperplasia control tissues. Dynamic cell-specific changes in the paucimannosylation and oligomannosylation of known bone marrow-and prostatederived glycoproteins, respectively, and increased N-glycan branching and core 2-type O-glycosylation of known extracellular matrix glycoproteins were found to be key changes associated with prostate cancer progression.
The complexity and dynamics of the immensely heterogeneous glycoproteome of the prostate cancer (PCa) tumor microenvironment remain incompletely mapped, a knowledge gap that impedes our molecular-level understanding of the disease. To this end, we have used sensitive glycomics and glycoproteomics to map the protein-, cell-, and tumor grade-specific Nand O-glycosylation in surgically removed PCa tissues spanning five histological grades (n = 10/grade) and tissues from patients with benign prostatic hyperplasia (n = 5). Quantitative glycomics revealed PCa grade-specific alterations of the oligomannosidic-, paucimannosidic-, and branched sialylated complex-type N-glycans, and dynamic remodeling of the sialylated core 1-and core 2-type O-glycome. Deep quantitative glycoproteomics identified~7400 unique N-glycopeptides from 500 N-glycoproteins and 500 unique O-glycopeptides from nearly 200 O-glycoproteins. With reference to a recent Tissue and Blood Atlas, our data indicate that paucimannosidic glycans of the PCa tissues arise mainly from immune cell-derived glycoproteins. Furthermore, the grade-specific PCa glycosylation arises primarily from dynamics in the cellular makeup of the PCa tumor microenvironment across grades involving increased oligomannosylation of prostate-derived glycoproteins and decreased bisecting GlcNAcylation of N-glycans carried by the extracellular matrix proteins. Furthermore, elevated expression of several oligosaccharyltransferase subunits and enhanced N-glycoprotein site occupancy were observed associated with PCa progression. Finally, correlations between the protein-specific glycosylation and PCa progression were observed including increased site-specific core 2-type Oglycosylation of collagen VI. In conclusion, integrated glycomics and glycoproteomics have enabled new insight into the complexity and dynamics of the tissue glycoproteome associated with PCa progression generating an important resource to explore the underpinning disease mechanisms.
Prostate cancer (PCa) is the second leading cause of cancer death in men worldwide (1). PCa is a highly heterogeneous disease leading to a variety of clinical outcomes spanning latent to highly aggressive disease forms (2). The PCa cells are surrounded by a heterogenous stroma comprising a complex extracellular matrix (ECM) and different types of cells, including immune cells, fibroblast cells, and endothelial cells, which dynamically modulate the tumor phenotype (3).
The gold standard for PCa diagnosis and prognosis is still based on a subjective histological examination of prostate tissue. PCa tissues are by convention classified into five grades, i.e., grade 1 to 5 (G1-G5) based on the Gleason score (GS) of the architectural patterns of the prostate adenocarcinoma tissue (4,5). The inherent variability of tumor sampling associated with the intratumoral heterogeneity is a severe limitation for accurate PCa prognostication (6). As recently reviewed (7), PCa biospecimens, including serum, urine, and tissues, have repeatedly been subjected to a variety of omicstype analyses with the aim to find novel quantitative molecular signatures to more accurately detect and stratify PCa. We have recently performed grade-specific proteome profiling of PCa tissues spanning G1 to G5 (8). In short, we identified key biological processes being altered during PCa progression, including cell adhesion, osteoblast differentiation, and blood coagulation. We reported in that study a panel of 11 proteins able to stratify patients with low and high PCa grades. Despite the growing list of reported candidate biomarkers, none has replaced the blood prostate-specific antigen (PSA) currently used as a biomarker for PCa screening despite its relatively low specificity and limited prognostic value (9,10). From  Building evidence indicates that not only the protein expression but also the glycans decorating proteins are altered in PCa (11)(12)(13). Such molecular aberrations generate distinct signatures associated with PCa that may be used for more accurate PCa diagnosis and prognosis by being more quantitative features that are less prone to inconsistent sampling (14). PSA glycoform profiling has been shown to provide better accuracy in detecting PCa compared with the conventional protein-level analysis of PSA in blood (15), which demonstrates the potential of using glycoprofiling to discover new and improve existing PCa biomarkers.
Glycans cover all cell surfaces including the surface of PCa tumor cells. In fact, the glycocalyx forms the interface between the tumor cells and the microenvironment, thus enabling communication between cells within the tumor and the ECM (16,17). It has been demonstrated that glycans mediate cellular events associated with hallmarks in cancer, such as invasion and metastasis, immune evasion, activation of endothelial cells, and induction of angiogenesis (18). Aberrant glycosylation occurring in both the cancer cells and the surrounding cells facilitates a dynamic and complex glycanmediated crosstalk within the tumor microenvironment of importance to PCa onset and progression (19).
Characterization of protein glycosylation remains analytically challenging because of the nontemplate-driven biosynthesis dictated by competing glycosyltransferases and glycoside hydrolases, which produce extensive structural diversity of structurally related glycoforms. It is widely recognized that comprehensive glycoprofiling requires multiple orthogonal analytical approaches (20,21). MS-based glycomics and glycoproteomics have emerged as powerful analytical tools to quantitatively map the glycan fine structures and the site-specific microheterogeneity and macroheterogeneity of glycoproteins at scale in complex biological samples (22)(23)(24)(25). Indeed, both glycomics and glycoproteomics have already been utilized to profile different features of protein glycosylation in PCa. Glycomics profiling revealed that core fucosylation and α2,3-sialylation are elevated features in PCa sera relative to benign prostatic hyperplasia (BPH) sera (26). Elevation of triantennary and tetraantennary N-glycans on serum proteins was also found to be predictors of castration-resistant PCa (27). In addition, Nglycomics revealed that highly branched sialylated N-glycans of urinary proteins are elevated features associated with highgrade PCa (GS = 9) relative to low-grade PCa (GS = 6) (28).
In separate efforts aiming to identify diagnostic and prognostic glycoprotein biomarkers for PCa, peptides were profiled using conventional proteomics after de-N-glycosylation of proteins extracted from tissue, urine, and serum obtained from PCa patients and control cohorts (29)(30)(31). A panel of serum and urinary glycoproteins with the potential to detect PCa, predict the PCa grade, and discriminate between aggressive and nonaggressive PCa tissues was reported. Furthermore, intact glycopeptide profiling of two PCa cell lines, LNCaP and PC3, representing cell line models for androgen-dependent and androgen-independent PCa, respectively, demonstrated a higher site-specific fucosylation and altered site occupancy of glycoproteins in PC3 cells (32). Finally, we recently used glycoproteomics to identify more than 900 N-and O-glycopeptides from urine glycoproteins, of which a panel of 56 N-glycopeptides was able to discriminate PCa and BPH (33). Concertedly, these studies have considerably advanced our knowledge of protein glycosylation in PCa, but the glycan fine structures, their site-specific microheterogeneity and macroheterogeneity, and their dynamics and cellular origins within the complex tumor environment during PCa progression remain unexplored.
To this end, we here apply integrated glycomics and glycoproteomics to investigate a clinically annotated collection of surgically removed fresh PCa tissues spanning five grades (G1-G5) and BPH tissues as a relevant clinical reference. This is the first study to report both on the glycan fine structures and the site-and cell-specific expression and dynamics of Nand O-glycosylation during PCa progression.

Ethics
The study was approved by the ethics review board of the Faculdade de Medicina da Universidade de São Paulo under the protocol no. 2.695.126. Written informed consents were obtained from all participants. All experiments were performed in accordance with the approved guidelines and regulations.

Patient Cohort, Tissue Collection, Storage, Isolation, and Protein Extraction
All prostate surgical specimens were examined by a trained uropathologist. Fresh tissues were collected after radical prostatectomy from a total of 55 patients, which included 10 patients from each PCa grade (G1-G5) and five individuals treated surgically for BPH. The PCa grades were determined using the modified Gleason grading system proposed by the International Society of Urological Pathology (34). The samples included in this study were only used for discovery phase research. Details of the sample cohort are available in our previous study, where label-free quantitative proteomics profiling was carried out on the same samples (8) and can be also found in supplemental Table SA. Fresh prostate tissues (~60 mg) stored in RNAlater for improved stability were washed in 80% (v/v) cold acetonitrile (ACN) and resuspended in an extraction buffer containing 8 M urea, 10 mM DTT, 1 mM sodium fluoride, 1 mM sodium orthovanadate, and a protease inhibitor cocktail (1:10). Prostate tissues were rapidly lysed (2 min, 30 Hz with a 5 mm stainless steel bead) using a TissueLyser (Qiagen). Protein concentrations were determined by the Qubit fluorimetric detection method (Thermo Fisher) and immediately used for glycomics and glycoproteomics.

Glycan Release
Extracted proteins (20 μg) were reduced using 10 mM DTT, 30 min, 30 • C, and alkylated using 40 mM iodoacetamide 30 min in the dark. The proteins were blotted on a primed 0.45 μm polyvinylidene fluoride membrane (Merck Millipore). The N-and O-glycans were sequentially released as described (35). Briefly, protein spots were stained using Direct Blue (Sigma-Aldrich), excised, and transferred to a flat bottom polypropylene 96-well plate (Corning Life Sciences), blocked with 1% (w/v) polyvinylpyrrolidone in 50% (v/v) methanol and washed with MilliQ water. The N-glycans were then released using 10 U recombinant Elizabethkingia miricola N-glycosidase F (Promega; V4831, 10 U/ μl), 16 h, and 37 • C. The detached glycans were reduced with 1 M sodium borohydride in 50 mM aqueous potassium hydroxide for 3 h at 50 • C. The reaction was stopped using glacial acetic acid, and the glycans were desalted using strong cation exchange/C18 and porous graphitized carbon (PGC) solid phase extraction (SPE). The same protein spots were then incubated with 0.5 M sodium borohydride in 50 mM potassium hydroxide at 50 • C for 16 h to release and reduce the O-linked glycans by reductive β-elimination. The O-glycans were desalted as described previously.

Glycan Analysis
PGC-LC-MS/MS-based glycan analysis was performed on an Ul-tiMate 3000 HPLC system (Dionex, Sunnyvale, CA, USA) interfaced with a linear trap quadrupole Velos Pro ion trap (Thermo Scientific, San Jose, CA, USA). The glycans were loaded on a PGC HPLC capillary column (Hypercarb KAPPA; 5 μm particle size, 200 Å pore size, 180 μm inner diameter × 100 mm length; Thermo Scientific) operated at 50 • C with a constant 4 μl/min flow rate. About 10 mM aqueous ammonium bicarbonate solution, pH 8 (solvent A), and 10 mM aqueous ammonium bicarbonate in 70% ACN (solvent B) were used as mobile phases using the following gradient for N-glycans: 8 min at 2.6% B, 2 min from 2.6 to 13.5% B, 45 min from 13.5 to 55% B, 15 min from 55 to 70% B, 1 min from 70 to 98% B, 5 min at 98% B, 1 min from 98 to 2.6% B and 6 min at 2.6% B as well as O-glycans: 5 min at 2.8% B, 33 min from 2.8 to 30% B, 5 min from 30 to 98% B, 5 min at 98% B, 2 min from 98 to 2.8% B, and 5 min at 2.8% B.
The electrospray ionization source was operated in negative-ion mode with source potential of +3.2 kV. Full MS scans were acquired using a mass range m/z 570 to 2000 for N-glycans and m/z 380 to 1800 for Oglycans, 3 microscans, m/z 0.25 full width at half maximum (FWHM) resolution, 5 × 10 4 automatic gain control (AGC), and 50 ms accumulation. For the MS/MS settings, m/z 0.35 FWHM resolution; 2 × 10 4 AGC, 300 ms accumulation, and 2 m/z precursor ion window. The five most abundant precursors in each full scan spectrum were selected for collision-induced dissociation-based MS/MS using normalized collision energy of 33% with an activation Q of 0.250 and 10 ms activation.

Glycan Data Analysis
Glycan precursor ions were extracted using RawMeat, version 2.1 (Vast Scientifi; www.vastscientific.com) (36). Common contaminants and redundant precursors were manually removed. The m/z of the extracted monoisotopic precursor ions was converted to mass [M-H]and searched against GlycoMod (http://www.expasy.ch/tools/ glycomod) to identify putative monosaccharide compositions. The mass tolerance was set to 0.5 Da. Only compositions containing Hex, HexNAc, dHex, NeuAc, NeuGc, and glycans reported in the Uni-CarbKB database were considered. The glycan fine structures were manually elucidated using monoisotopic mass and the PGC LC elution and MS/MS fragment patterns. Extracted ion chromatogrambased glycan quantification was performed using Skyline, version 19.1, as described (37). GlycoWorkBench, version 2.1 (https://code. google.com/archive/p/glycoworkbench/) was used to aid the manual annotation of glycan fragment spectra and generate the glycan cartoons. Xcalibur, version 2.2 (Thermo Scientific) was used to browse and annotate raw LC-MS/MS data.

TMT Labeling of Peptides
A pooled peptide reference sample containing peptides from all samples was generated for quantitative comparison across multiple tandem mass tag (TMT)-10plex experiments. The peptide samples were randomly combined in six TMT-10plex sets. Each set included at least one sample from each PCa grade and BPH. The reference sample was consistently labeled with the 126 Da reporter ion channel. In total, 54 peptide samples were labeled with TMT for quantitative proteomics, including five BPH samples, 10 PCa samples each from G1 to G4, and nine PCa samples from G5 (supplemental Table SB).
Peptides originating from 25 μg protein extracted from each sample were dissolved in 100 μl 100 mM triethylammonium bicarbonate buffer and labeled individually with 0.23 mg TMT-10plex mass tags (Thermo). In brief, the TMT10plex tags (0.23 mg) in 41 μl neat anhydrous ACN were added to the peptide samples and incubated for 1 h at room temperature. Reactions were quenched using 8 μl 5% (v/v) hydroxylamine, 15 min at room temperature. After TMT labeling, all peptide samples were mixed 1:1 (w:w), desalted using hydrophilic-lipophilic balance SPE cartridges (Waters), and dried. Aliquots of the TMTlabeled peptide mixtures from the six TMT-10plex experiments were analyzed without any prefractionation by LC-MS/MS (supplemental Table SB).

Glycopeptide Prefractionation
Enriched glycopeptides, de-N-glycosylated peptides, and nonmodified peptides were resuspended in 50 μl 25 mM ammonium bicarbonate for high pH prefractionation using Oligo R2 reversed phase SPE microcolumns packed on supporting C18 discs (Empore) in standard p10 pipette tips. The columns were washed three times with 50 μl of 100% ACN and 50 μl of 25 mM aqueous ammonium bicarbonate. Samples were loaded on the columns followed by two washing steps with 50 μl of 25 mM aqueous ammonium bicarbonate. The peptides were eluted in three fractions, i.e., fraction 1: 10% ACN in 25 mM aqueous ammonium bicarbonate; fraction 2: 20% ACN in 25 mM aqueous ammonium bicarbonate, and fraction 3: 60% ACN in 25 mM aqueous ammonium bicarbonate, and dried. The peptides from the three fractions were resuspended in 0.1% (v/v) formic acid (FA) for separate LC-MS/MS analysis (supplemental Table SB).

Peptide Profiling
TMT-tagged peptides were loaded on a trap column (2 cm × 100 μm inner diameter) custom packed with ReproSil-Pur C18 AQ 5 μm resin (Dr Maisch, Ammerbuch-Entringen, Germany). Approximately 1 μg of total peptide was injected per LC-MS/MS run. The peptides were separated at 250 nl/min on an analytical column (Reprosil-Pur C18-Aq; 25 cm × 75 μm, 3 μm ID; Dr Maisch, Ammerbuch-Entringen, Germany) using an UltiMate 3000 RSLCnano System. The mobile phases were 99.9% ACN in 0.1% (both v/v) aqueous FA (solvent B) and aqueous 0.1% (v/v) FA (solvent A). The gradient was 2 to 30% B over 100 min, 30 to 50% B over 18 min, 50 to 95% B over 1 min, and 9 min at 95% B. The nanoLC was connected to a Q-Exactive HF-X Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific) operating in positive-ion mode. The Orbitrap was used to acquire the full MS scan with an AGC of 3 × 10 6 ions and 50 ms accumulation. Full MS scans were acquired at high-resolution 60,000 FWHM at m/z 200 with an m/z range of 350 to 1800. The 20 most abundant precursor ions were selected from each MS full scan using data-dependent acquisition and fragmented utilizing higherenergy collision-induced dissociation (HCD) fragmentation with a normalized collision energy of 35%. Only multicharged precursors (Z ≥ 2) were selected for fragmentation. Fragment spectra were acquired at 45,000 resolution with an AGC of 1 × 10 5 and 90 ms accumulation using a precursor isolation window of m/z 1.0 and a dynamic exclusion of 30 s after a single isolation and fragmentation of a given precursor ion.

Analysis of Intact N-Glycopeptides
The HCD-MS/MS spectra of intact glycopeptides were searched with Byonic, version 2.6.46 (Protein Metrics, CA, USA) (39) using 10/ 20 ppm as the precursor and product mass tolerance, respectively. Cys carbamidomethylation (+57.021 Da) and TMT (+229.163 Da) at N terminus and lysine (K) were considered fixed modifications. Trypsinspecific cleavages were considered with a maximum of two missed cleavages per peptide. The following variable modifications were included: Met oxidation (+15.994 Da) and N-glycosylation of sequonlocalized Asn with a predefined glycan database of 309 mammalian Nglycans without sodium adducts to which the paucimannose M3F (Man3GlcNAc2Fuc1) was manually added or O-glycosylation of Thr/ Ser with a predefined O-glycan database containing 78 common mammalian O-glycans without sodium adducts available within Byonic. The MS/MS spectra were searched against a protein database composed of all reviewed UniProtKB human proteins (supplemental Table SB). All searches were filtered to <1% false discovery rate at the protein level and 0% at the peptide level by using a decoy database (40). Only glycopeptides identified with high score ID were considered (PEP 2D scores <0.001) (33). Glycopeptides identified with low confidence, in the reverse database and contaminant database, were excluded. Intact glycopeptides were quantified using the Report Ion Quantifier as a node in Proteome Discoverer, version 2.2 (Thermo Scientific). The abundances of the reporter ion intensities from the MS/MS scans were extracted from the QuantSpectra table. Glycopeptides were manually grouped by summing the reporter ion intensities from peptide spectral matches with the same UniProtKB identifier, same glycosylation site within the protein, and same glycan composition. The abundances of the unique glycopeptides from each channel were first normalized by dividing each unique glycopeptide reporter ion intensity by the reference reporter intensity within that specific experiment and further normalized by the total ratio intensity of each channel to correct for any variation in the total yield during the labeling reactions.

Protein Identification and Glycosylation Site Analysis
For protein identification and quantification as well as glycosylation site analysis, the LC-MS/MS raw files were imported into MaxQuant, version 1.5.2.8 (41). The database search engine Andromeda (42) was used to search the HCD-MS/MS spectra against a database composed of all reviewed UniProtKB human proteins (downloaded July, 2017; 20,201 entries) with a precursor ion tolerance of 4.5 ppm and product ion tolerance of 20 ppm. Report fragment ions 10plex TMT were included in the quantification settings, and enzyme specificity was set to trypsin with a maximum of two missed cleavages. Cys carbamidomethylation (+57.021 Da) was considered a fixed modification, and Met oxidation (+15.994 Da), deamidation of Asn (N) and Gln (Q) (+0.984 Da), and protein N-terminal acetylation (+42.010 Da) were selected as variable modifications. Five variable modifications per peptide were allowed. All identifications were filtered to achieve 1% peptide spectral match and 1% protein false discovery rate. The quantitation of the identified proteins was determined by reporter ion intensities using at least one razor/unique peptide (supplemental Table SB). The individual protein abundances were normalized across experimental sets by dividing each protein reporter ion intensity by the reference reporter intensity within that specific experiment. The data were further normalized by the total ratio intensity of each channel to correct for any variation in total yield during the labeling reactions.
Peptides identified as potential contaminants or appearing in the reverse database were excluded for the glycosylation site analysis. Only peptides containing deamidated Asn within conserved Nglycosylation motifs, i.e., NxS/T/C, where x s P, were considered. The de-N-glycopeptides were normalized by the reference channel and further normalized by the total ratio intensity of each channel.
The Human Tissue and Blood Atlas, both part of the Human Protein Atlas (43)(44)(45), were used to annotate the cellular origin of the identified proteins. Sets of elevated proteins within relevant tissues (which included tissue enriched, group enriched, and tissue enhanced) including in the bone marrow (total of 534 proteins) and the prostatic tissue (total of 120 proteins) were extracted from the Tissue Atlas. The list of ECM proteins was obtained from the Human Blood Atlas, under the section the human secretome and category secreted to ECM (total of 234 proteins).

Experimental Design and Statistical Rationale
In total, 55 biological replicates from BPH (n = 5) and PCa patients (G1-G5, n = 10/grade) were investigated. No technical replicates were acquired. The BPH served as a clinically relevant reference. For glycome profiling and glycosylation site analysis, statistical significance was assessed using unpaired two-tailed Student's t tests with p < 0.05 as the confidence threshold. For glycoproteome profiling, the statistical significance of the summed intensities of the glycopeptides grouped based on glycan classes or tissue origin was assessed using unpaired two-tailed Student's t tests with p < 0.05 as the confidence threshold. Correlation analysis was carried out using Pearson correlation coefficient and tested for significance using t distribution tests with p < 0.05 as the confidence threshold.

Integrated Glycomics and Glycoproteomics Profiling of PCa and BPH Tissues
In this study, the protein N-and O-glycosylation were investigated from fresh prostate tissue collected after radical prostatectomy from age-matched cohorts of 50 PCa patients spanning five grades of the disease (G1-G5, n = 10/grade) and a group of BPH patients (n = 5) included as a clinical reference (Fig. 1A). Multiple analytical approaches were used to quantitatively profile: (1) the N-and O-glycan fine structures, including their composition, branching, and linkage isomers using PGC-LC-MS/MS-based glycomics of released N-and O-glycans (Fig. 1B); (2) the protein carriers, N-and O-glycosylation sites, and their microheterogeneity obtained using label-assisted glycoproteomics of intact N-and O-glycopeptides; and (3) the N-glycan occupancy at each site (macroheterogeneity) using proteomics profiling of formerly Nglycosylated peptides (hereafter referred to as de-N-glycopeptides) adjusted by the protein level (Fig. 1C).
In short, glycomics identified 180 N-glycans spanning 80 unique compositions and 17 O-glycans spanning 13 O-glycan compositions (supplemental Tables S1 and S2 and supplemental Datas S1 and S2), glycoproteomics identified 7447 unique intact N-glycopeptides (herein defined as a unique glycan composition at a unique site on a unique glycoprotein) and 522 unique intact O-glycopeptides covering 540 N-glycoproteins and 178 O-glycoproteins, respectively (supplemental Tables S3 and S4). Furthermore, site occupancy analysis established the occupancy of 2101 N-glycosylation sites covering 721 N-glycoproteins and the relative protein level of 3308 proteins (supplemental Tables S5 and S6) across all samples (Fig. 1D).

Heterogeneous Cell-and Site-Specific N-Glycosylation in PCa and BPH Tissues
The N-glycome of the investigated tissue samples was found to comprise mainly complex-type biantennary and triantennary sialoglycans with and without fucosylation (83.4%) and a lower amount of oligomannosidic-type N-glycans (11.6%) and paucimannosidic-type N-glycans (4.8%, average relative abundance of all replicates) (Fig. 2, A and B). LC separation of the α2,6and α2,3-linked sialoglycans was achieved, and MS/MS signatures confirmed the presence of core fucosylation, bisecting GlcNAc, LacdiNAc, and NeuGc and other important glycan fine structural features (supplemental Fig. S1, supplemental Table S1, and supplemental Datas S1 and S2).
The glycoproteomics data enabled deep glycoproteome coverage comparable to recent glycoproteomics studies (23)(24)(25). In total, 1223 unique glycopeptides, corresponding to 16.4% of the data set, were observed across all 54 samples (supplemental Fig. S2A), and 6887 unique glycopeptides (92.5% of the data set) were detected across all patient groups (supplemental Fig. S2B). Importantly, quantitative glycoproteome profiling achieved by TMT labeling and detection of intact glycopeptides is getting more common in glycoproteomics as elegantly demonstrated in a recently published study tracking the changes in protein N-glycosylation during mouse myogenesis and muscle development (46). Key advantages of TMT-based quantitative glycoproteomics include enhanced reproducibility, quantitative accuracy, sensitivity, and sample throughput enabled by the ability to multiplex the glycopeptide analyses relevant when dealing with larger sample cohorts (47). Complex/hybrid, oligomannose, paucimannose, and chitobiose core glycopeptides covered 82.4%, 15.6%, 1.4%, and 0.6% of the glycoproteome, respectively (Fig. 2C). The glycoproteomics and glycomics data showed excellent agreement as demonstrated by the high correlation (r = 0.827) of the glycans detected by the two data sets (data not shown). Moreover, 70% of all identified intact glycopeptides carried glycans observed in the N-glycome profile. Nearly 70% of the glycosites carried more than one glycan (average of 6.8 glycan/site; Fig. 2D, panel I), whereas 60% of the proteins were observed with only one occupied glycosite (Fig. 2D, panel II and supplemental Table S3).
Extensive microheterogeneity and macroheterogeneity of the identified glycoproteins arising from the various cellular origins were observed (Fig. 2E). The ECM glycoproteins were particularly heterogenous displaying more than two glycans per site and often more than five occupied glycosites per protein. The proteins from prostate and bone marrow origin carried fewer glycosites per protein but displayed high microheterogeneity. The data allowed for a detailed assessment of the site-specific microheterogeneity and macroheterogeneity of the identified glycoproteins with tissue information as demonstrated by analysis of key glycoproteins from the three tissue origins (Fig. 2F). The prostate-specific PAP carried few occupied glycosites (N94, N220, and N333) that displayed high microheterogeneity of all three sites. Conversely, the ECM glycoprotein LAMA5 carried 16 occupied glycosites displaying limited heterogeneity; e.g., N143, N959, N1330, N1529, N2209, and N3107 carried only a single glycan type. Finally, the bone marrow-derived LTF contained three occupied sites, two of them with high glycan heterogeneity (N156 and N497) and one with low heterogeneity (N642) (supplemental Table S3).
We then compared our site-specific glycoprofiling data of PAP, LAMA5, and LTF to the literature. In agreement with our data, N94, N220, and N333 were previously found to be utilized glycosylation sites of seminal fluid PAP (54). We identified 199 N-glycans of PAP adding significantly to the 21 Nglycans previously reported from that protein. Similarly, N156, N497, and N642 were previously reported glycosylation sites of milk LTF (55). In line with our data, complex-type N-glycans were found to decorate LTF; Yu et al.

Dynamics of the N-Glycoproteome During PCa Progression
The protein N-glycosylation dynamics during PCa progression was investigated using three levels of analysis. First, quantitative N-glycan distribution analysis revealed that many N-glycans are differentially expressed across the PCa grades relative to BPH (t test, p < 0.05) (Fig. 3A). Low-grade PCa tissues (G1-G4) were dominated by a high expression of paucimannosidic-and monoantennary complex-type N-glycans relative to BPH tissues, whereas high-grade PCa tissues (G5) were dominated by highly branched complex-type N-glycans. Oligomannosidic-, hybrid-, and biantennary complex-type N-glycans, some of which carried LacdiNAc features, showed a lower expression in G5 relative to BPH tissues. Bisecting GlcNAc-containing glycans were relatively lowly expressed in PCa tissues regardless of grade, whereas the α2,6or α2,3-sialylation and core fucosylation were uniformly expressed glycosignatures across the examined tissue samples (supplemental Table S1).
Previous studies have reported that the N-acetylglucosaminyltransferase (GnT)-III, which catalyses bisecting GlcNAcylation, is associated with tumor suppression by inhibiting growth factor signaling, tumor development, and migration (61)(62)(63). The antennary branching of N-glycans is catalyzed by GnT-IV and GnT-V (64). Lange et al. (65) reported that the GnT-V-mediated glycan branching is elevated in metastatic PCa and is a potential marker for poor PCa prognosis. Gao et al. (66) reported higher mRNA levels and activity of GnT-V in PCa cell lines compared with nontumorigenic prostatic cells. Elevation of triantennary and tetra-antennary N-glycan in serum was also shown to be a signature associated with castration-resistant PCa patients (67).
N-glycolylneuraminic acid (Neu5Gc)-containing N-glycans were also identified in our analysis of prostate tissue as evidenced by LC-MS/MS (supplemental Data S1). Most notably, a significant lower abundance of the α2,6-linked monovalent and divalent Neu5Gc-containing biantennary corefucosylated glycans were identified in PCa G4 and G5 compared with BPH (supplemental Table S1). Humans are deficient in the hydroxylase that converts CMP-Neu5Ac to CMP-Neu5Gc, but Neu5Gc can be metabolically incorporated into human tissues via dietary intake (e.g., red meat) (68). Neu5Gc has previously been associated with inflammation and risk of cancer (69,70), but little is known about its role in PCa.
Supporting the prostate-specific elevation of oligomannose observed in this study, spatial distribution analysis of detached glycans in cancer tissues using MALDI-MS imaging previously demonstrated that oligomannosidic-type glycans are primarily expressed in the tumor region of late stage of ovarian cancer (74) and PCa (75,76). Furthermore, enhanced oligomannosylation of serum glycoproteins in advanced PCa detected by antioligomannose antibodies was reported (77). Finally, downregulation of the Golgi α-mannosidase I responsible for processing of oligomannose glycans was associated with metastatic cholangiocarcinoma and increased migratory and invasive capabilities (78).
Correlation analysis between the N-glycome and N-glycoproteome data sets was then carried out to infer the glycan fine structural features decorating each glycosylation site. The intact glycopeptides carrying complex-type glycan compositions showed significant correlation with the matched glycan structures from the glycome data set (supplemental Table S7). Correlations were found for several structural features including glycan types, number of antennas, and the presence of bisecting GlcNAcylation and LacdiNAc motifs. We observed that bisecting GlcNAc is an enriched signature of glycans carried by the ECM glycoproteins, whereas prostatederived glycoproteins are enriched in LacdiNAc glycans (Fig. 3C). Supporting these findings, previous studies reported that LacdiNAc glycans decorate prostate-specific glycoproteins such as PSA (79,80) and that bisecting GlcNAc is a glycan feature of ECM glycoproteins (81)(82)(83).
Recapitulating the downregulation of bisecting GlcNAc and LacdiNAc N-glycans in PCa relative to BPH, bisecting GlcNAc-and LacdiNAc-containing glycopeptides of ECM and prostate-derived glycoproteins, respectively, showed decreased abundances in PCa. Furthermore, elevation of highly branched glycans decorating ECM-derived glycoproteins was observed in advanced PCa (G5), recapitulating the grade-specific changes observed in the glycome data set.

Altered Expression of Oligosaccharyltransferase and N-Glycan Site Occupancy During PCa Progression
Aiming to also map the dynamics of the glycosylation sites and their occupancy levels in PCa progression, we identified 2101 de-N-glycopeptides from 721 N-glycoproteins following peptide:N-glycosidase F digestion (Fig. 4A). In total, 360 of the formerly N-glycosylated proteins were also identified in the proteome data set (Fig. 4B), which was used to normalize the de-N-glycopeptide level and determine the site occupancy. The N-glycan site occupancy was compared across PCa grades and BPH, and a total of 389 de-N-glycopeptides belonging to 187 N-glycoproteins showed altered occupancy (supplemental Table S5).
The utilization of glycosylation sites relates to the first step of the biosynthesis of N-glycoproteins involving the transfer of a lipid-linked N-glycan precursor (Glc3Man9GlnNAc2) to nascent polypeptides by the large oligosaccharyltransferase (OST) complex (84). The proteome data revealed that two OST subunits, RPN1 (P04843) and RPN2 (P04844), were significantly elevated in PCa grades (G1-G4) providing a molecular basis for the increased site occupancy in PCa (Fig. 4C). In support, another key OST subunit, OSTC (Q9NRP0), was also found to be overexpressed in PCa grades G1 and G3. The catalytic OST subunits, STT3A (P46977), STT3B (Q8TCJ2), and DAD1 (P61803), were detected below the level of quantitation (supplemental Table S6).
Clustering analysis was performed using the set of de-Nglycopeptides that showed altered site occupancy and was consistently identified in all replicates (Fig. 4D). Two main clusters were observed: cluster 1 grouped peptides containing less occupied glycosylation sites in PCa relative to BPH, whereas cluster 2 comprised peptides with elevated site occupancy. Motif sequence analysis of cluster 1 and cluster 2 de-N-glycopeptides was performed to explore the OST substrate specificity in light of the occupancy data using the peptide region immediately C-terminal to the N-glycosylation site. Cluster 1 de-N-glycopeptides showed an enrichment of FIG. 3. N-glycosylation remodeling associated with prostate cancer (PCa) progression. A, depiction of the N-glycome alteration associated with PCa progression (PCa G1-G5 versus BPH, Student's t test, p < 0.05, supplemental Table S1). Graphs are based on the Log2fold change of the relative abundance between each PCa grade and BPH. The individual glycans were grouped according to their glycan type and other structural features. The most abundant glycan structure from each class is depicted next to each graph. B, weakly altered tissuespecific glycosylation across PCa progression as determined by intact glycopeptide profiling. Decreased oligomannosylation was observed in the bone marrow-derived glycoproteins from PCa G3 compared with BPH, whereas increased oligomannosylation was observed in prostatederived glycoproteins from PCa G5 compared with BPH (Student's t test, p < 0.05). C, distribution of intact glycopeptides in PCa and BPH grouped by cellular origin and glycan feature that showed significant correlation with the glycome data set (Pearson, p < 0.05) (supplemental Table S7). The heat map shows the average of the summed relative abundance of intact glycopeptides from all replicates (n = 5 BPH; n = 10/G1-G4, and n = 9/G5). BPH, benign prostatic hyperplasia, ECM, extracellular matrix. hydrophilic amino acids, such as cysteine (C), arginine (R), and glutamic acid (E), whereas hydrophobic amino acids, such as leucine (L) and phenylalanine (F), were over-represented in cluster 2 de-N-glycopeptides. Accordingly, the hydropathy of the OST recognition region of the de-N-glycopeptides upregulated in PCa (cluster 2) was found to be significantly higher than the sites experiencing less occupancy in PCa (cluster 1) (Fig. 4F). Our data point to that the increased expression of relatively hydrophobic sequons and the elevation of OST subunits concertedly increase the site occupancy during PCa progression, which are findings that require further investigations.
Other mechanisms besides expression of OST subunits have been demonstrated to impact the N-glycosylation site occupancy. The OST substrate recognition and glycosylation transfer efficiency are modulated by the oxidoreductase activity of the yeast OST subunits Ost3p and Ost6p (85). Furthermore, the STT3B-OST-specific subunit TUC3, found to be downregulated in PCa tissues, is known to reduce the OST-mediated N-glycan transfer efficiency (86). In the same study, knockdown of TUC3 in PCa cell lines led to increased proliferation, invasion, and in vivo tumor growth, indicating that this subunit may act as a tumor suppressor in PCa. Finally, the expression of RPN2 was previously linked to other types of cancer, such as breast cancer, colorectal cancer, and lung cancer, but hitherto not to PCa (11).

Altered Protein O-Glycosylation During PCa Progression
The O-glycome profile of the PCa tissue revealed a total of 17 structures covering 13 compositions (supplemental Table S2). Core 1-and 2-type structures were prominent Oglycans across all investigated samples (Fig. 5A). Quantitative analysis showed that sialylated core 1 glycans were significantly reduced, whereas sialylated core 2 structures were elevated during PCa progression (Fig. 5B).
Relatively low microheterogeneity and macroheterogeneity of the O-glycoproteome data was observed (Fig. 5D). Most Oglycosites (~80%) carried only a single glycan, and most Oglycoproteins (60%) were identified with a single utilized glycosylation site. Sequence motif analysis showed a position-unspecific enrichment of proline around the glycosite (Fig. 5E) in agreement with a previous report demonstrating enhanced O-glycosylation at Pro-rich sites (87).
Correlation analysis between the O-glycome and O-glycoproteome showed that the glycans of 13 O-glycopeptides quantitatively correlated with 5 O-glycan structures (supplemental Table S8). For example, the O-glycopeptide from the high abundant glycoprotein collagen VI (P12111; supplemental Table S6) correlated with the abundant sialylated core 2 O-glycan. Moreover, this glycoform was confidently identified at T1627 in all 54 replicates (Fig. 5F and supplemental Tables S4 and S8). This glycoform recapitulated the changes in the O-glycome by displaying a PCa gradespecific elevation relative to BPH (Fig. 5G).
Core 2-branched O-glycans are synthesized by the core 2 β1,6-N-acetylglucosaminyltransferase (C2GNT; Q02742), which catalyses the transfer of N-acetylglucosamines onto mucin-type core 1 O-glycans. This protein was not identified in the proteomics data set. C2GNT expression reportedly is high in PCa cells as shown by immunohistochemistry of formalin-fixed paraffin-embedded tissues and correlates with PCa progression (90). It was also reported that high expression of C2GNT is associated with poor prognosis and higher risk of recurrence after radical prostatectomy (91). Finally, core 2 O-glycans were found to play important roles in the evasion of natural killer cell immunity, thus favoring PCa metastasis (92). Our data add to these observations by showing a sitespecific elevation of core 2 O-glycosylation of collagen VI during PCa progression. The role of core 2 O-glycosylation of collagen VI in PCa progression remains to be studied. CONCLUSION We are the first to report on the complexity and dynamics of the tissue glycoproteome during PCa progression. Integrated use of MS-driven glycomics and glycoproteomics with valuable tissue annotation reference libraries enabled sensitive and quantitative insight into the cell-, protein-, and sitespecific glycosylation across the PCa grades (Fig. 6). This study forms a valuable resource to explore how the dynamic remodeling of the glycoproteome functionally impacts PCa progression, thus, ultimately contributing to an improved molecular level understanding the disease mechanisms and BPH (Student's t test, *p < 0.05, **p < 0.01, ***p < 0.001). Other known OST subunits (e.g., STT3A/B and DAD) were identified below the limit of quantitation. D, Euclidean clustering analysis of sites consistently identified with altered N-glycan occupancy in PCa. The heat map plots the average site occupancy (adjusted for protein level) that were identified without missing values (n = 5 BPH; n = 10/G1-G4 and n = 9/G5). Two major clusters were identified: (i) cluster 1 sites showed a reduced occupancy, whereas (ii) cluster 2 sites showed a higher site occupancy in G1-G5 relative to BPH (Student's t test, *p < 0.05, **p < 0.01, ***p < 0.001). E, sequence motif analysis of cluster 1 and cluster 2 sites. Residues immediately C terminal to the N-glycosylation site were explored for enriched motifs. Sequence logos were probability weighted (Kullback-Leiblerhe divergence). The height of the amino acid residue corresponds to their probability times their log-odd scores. F, hydropathy scores of cluster 1 and cluster 2 sites. BPH, benign prostatic hyperplasia; OST, oligosaccharyltransferase; OSTC, oligosaccharyltransferase complex. fueling ongoing efforts to develop effective diagnostic and prognostic markers for improved patient outcome. Acknowledgments -We thank Drs Christopher Ashwood and Hannes Hinneburg for assistance with the acquisition and analysis of the glycome data. We thank Dr Ardeshir Amirkhani and Dr Edward S. X. Moh for assistance with the acquisition of glycoproteomics data.