N-Linked Fucosylated Glycans Are Biomarkers for Prostate Cancer with a Neuroendocrine and Metastatic Phenotype

Prostate cancer continues to be a significant cause of morbidity and mortality in US men as the most diagnosed noncutaneous cancer and the leading cause of cancer death (1). Serum PSA testing has been widely deployed in prostate cancer screening but unfortunately suffers from poor sensitivity and specificity, thus leading to overdiagnosis and overtreatment (2–6). Among patients who receive definitive treatment with prostatectomy or radiation therapy, 20% to 40% of patients undergoing radical prostatectomy and 30% to 50% of patients undergoing radiation therapy experience recurrence within 10 years (7–9). Although localized and regional prostate cancer has a 100% 5-year survival rate, patients with distant metastasis have a 5-year survival rate of 32%, with the lymph nodes, bones, lung, liver, and brain as the most common sites of metastasis (10).

Androgen deprivation therapy is the standard of care for patients with metastatic disease, which eventually leads to the development of castration-resistant prostate cancer (CRPC) that limits androgen deprivation therapy effectiveness (11, 12). This is compounded by the phenomenon that treatment-induced neuroendocrine prostate cancer (t-NEPC) emerges in up to ∼20% of patients with CRPC (13). t-NEPC is an especially lethal variant of prostate cancer characterized by increased visceral metastatic burden compared with conventional CRPC, as well as a markedly reduced median survival time of less than 1 year (13–16). Together, these findings suggest that our existing tools do not accurately distinguish aggressive and indolent prostate cancer at the time of diagnosis, which leads to both inappropriate overtreatment of indolent disease and undertreatment for potentially lethal prostate cancer. Thus, this highlights an urgent need for more specific, easily detectable prostate cancer biomarkers.

Abnormal glycosylation has been classified as a hallmark of tumorigenesis (17, 18). For prostate cancer, the study of N-linked glycosylation, in which a sugar conjugate is added cotranslationally to asparagine residues, has been reviewed extensively across the spectrum of sample types and disease severity (4, 19–21). Fucosylation, the addition of fucose sugars to N-glycans by a family of fucosyltransferases (FUT), plays an important role in cancer progression and immune response (22–25). Fucosylation has been associated with aggressive prostate cancer (24, 26), and older studies with limited numbers of tissues have indicated the presence of different Lewis carbohydrate antigens like sialyl-Lewis-a, sialyl-Lewis-x, and the di-fucosylated sialyl-Lewis-y (Le^y) structure in metastatic prostate cancer (27–30). However, the structures of specific glycan compositions and their association with poor outcomes in patients with prostate cancer are poorly defined. To explore the role that fucosylated N-glycans may play in the development of lethal prostate cancer, we evaluated independent cohorts of human prostate cancer tissue specimens for their composition of fucosylated N-glycans using N-glycan mass spectrometry imaging (MSI; 31–34) and analyzed how the abundance of these N-glycans correlated with outcomes.

This retrospective study was approved by the Washington University in St. Louis Institutional Review Board (#202011019). The study was conducted in accordance with the Declaration of Helsinki. These specimens included a combination of formalin-fixed, paraffin-embedded (FFPE) prostatectomy specimens as well as primary small-cell carcinomas and metastases to lymph nodes and bone. The clinical characteristics of the patients who participated in this study are provided in Supplementary Table S1.

Radical prostatectomy specimens were banked in the Institutional Review Board–approved Cancer Tissue Bank and Database at University of Texas Health San Antonio (#20050234HR). Long-term outcomes were determined by chart review, and outcomes were confirmed by a urologic oncologist. Patients with prostate cancer were classified into two categories: men whose prostate cancer had not recurred after treatment after a minimum of 5 years of follow-up (termed no evidence of disease) and men who had documented metastatic disease after primary treatment. Men who had evidence of biochemical recurrence but no evidence of metastatic disease were not part of the dataset. To prepare the tissue microarrays (TMA), FFPE tissue blocks were sectioned and prepared on charged glass slides and stained with hematoxylin and eosin (H&E). The slides were examined and marked by a genitourinary pathologist (DAT). Tumor tissue was subsequently cored and used to construct the TMAs. Noncancerous, adjacent normal prostate tissue was used as a control in each TMA. The clinical characteristics of the patients used in this study are provided in Supplementary Table S2. Two no evidence of disease and five metastatic disease specimens were excluded from the analysis, as cores for those patients were absent from the TMA. Grading of prostate tumors was based on the 2014 World Health Organization/International Society of Urological Pathology (ISUP) consensus (35).

A TMA using prostate tumor tissues derived from the LuCaP series of patient derived xenografts (PDX) has been previously described (36) and was obtained from the University of Washington. 

MALDI matrix α-cyano-4-hydroxycinnamic acid (CHCA) was obtained from Sigma-Aldrich. Tissue-tack glass histology slides, acetonitrile and water (HPLC grade), citraconic anhydride, and xylenes (histological grade) were obtained from Thermo Fisher Scientific. Recombinant peptide N-glycosidase F (PNGase F PRIME) was obtained from N-zyme Scientific.

Immunohistochemistry was performed on 5-μm thick, formalin fixed paraffin embedded (FFPE) whole tissue sections. For chromogranin A and synaptophysin staining, slides were deparaffinized and conditioned with CC1 conditioning solution according to manufacturer specifications and stained on the Ventana Benchmark Ultra automated platform (Roche Diagnostics. Anti-Chromogranin A (LKH210; Roche) was used prediluted and incubated with slides at 37°C for 32 minutes and Anti-Synaptophysin (MRQ-40; Cell Marque) was incubated with slides at 37°C for 36 minutes. For AR staining, unstained slides were shipped to PhenoPath, PLLC for immunohistochemical staining. Stained slides were returned and interpreted by a board-certified pathologist (JKS). Anti-Blood Group Lewis Y monoclonal antibody (CGYJ031; Creative Biolabs) was diluted 1:100 and incubated overnight at 4°C using deparaffinized and antigen retrieved tissue slides. All slides were digitized with a NanoZoomer 2.0-HT System using NDP.scan 2.5 software for acquisition and NDP.view software for analysis (Hamamatsu Photonics).

Slides stained for androgen receptor (AR), chromogranin A (CgA), and synaptophysin (SYN) were graded for expression using the following schemes. AR: 0 (none), 1 (weak), 2 (moderate), 3 (strong). CgA: 0 (none), 1 (focal granular staining in rare cells), 2 (focal granular staining in more than rare single cells or extensive granular staining in scattered cells, 3 (diffuse). SYN: 0 (none), 1 (weak but patchy), 2 (moderate to strong staining but patchy), 3 (moderate to strong but diffuse).

N-glycan matrix-assisted laser desorption/ionization–MSI (MALDI-MSI) was performed on FFPE tissue slides using a standardized protocol (33) used recently for analysis of prostate cancer (31) and multitumor TMA (37) FFPE tissues. Antigen retrieval of dewaxed slides was done in citraconic anhydride buffer, pH 3, for 30 minutes in a decloaking chamber at 95°C. PNGase F PRIME enzyme at 0.1 µg/µL was applied at 10 psi and 45°C using an M5 sprayer (HTX Technologies; ref. 33). The slides were incubated in prewarmed humidity chambers for 2 hours at 37°C for release of N-glycans. After 2 hours of PNGaseF digestion, the M5 sprayer was used to apply 7 mg/mL α-cyano-4-hydroxycinnamic acid (CHCA) matrix in 50% ACN/0.1% trifluoroacetic acid at 10 psi and 79°C (33). The prepared slides were desiccated at room temperature until analysis. For detection of released N-glycans, a Solarix dual-source 7T MALDI-Fourier transform ion cyclotron resonance (FTICR) mass spectrometer and a timsTOF Flex MALDI-quadrupole time-of-flight (QTOF) mass spectrometer were used (Bruker Corporation), operated in positive ion mode, as previously described (37, 38). Data were collected with a laser spot size of 20 to 25 µm, 300 laser shots per pixel, 40 µm raster, and mass ranges of 700 to 4,000 m/z. For each tissue, all acquired spectra were imported to SCiLS Lab software (Bruker Corporation) for processing and visualization. N-glycan spectra were normalized to total ion count. Spectra were annotated by matching glycan peak m/z values to an established in-house N-glycan database (31, 32, 38). Structural assignments and glycan compositions were made using multiple prior characterizations by reversed-phase LC/MS-MS, MALDI-TOF-MS/MS, and alternate enzymatic release experiments (37–39).

From the cumulative N-glycan composition data obtained across the prostate adenocarcinoma and small-cell/neuroendocrine tissues, a panel of 53 fucosylated N-glycans was selected to represent a fucose scoring panel. The structures for all 53 glycans are provided in Supplementary Table S3. Only glycans that colocalized to tumor regions based on pathology annotation of H&E and IHC stains were scored. These tumor glycans were selected on structural criteria, the primary one being that each glycan contained at least two fucose residues (i.e., either a core and outer arm fucose or two outer arm fucoses). Summarized in Supplementary Fig. S1, other modifications included addition of a sulfate, 1 to 2 sialic acids, and 1 to 3 polylactosamines [a repeat of galactose (Gal) and N-acetylglucosamine (GlcNAc)]. Most structures had three or four branched antennae, with or without a bisecting GlcNAc. Paucimannose (Man; Man_1-3GlcNAc₂), high-mannose (Man_5-9GlcNAc₂), and singly fucosylated N-glycans were not included in the score as all tumor types contained these species. For scoring, the intensity values for each of the 53 glycans in each tissue were extracted using SCiLS software and were scored as present (i.e., adding +1 to the score) if isotopic peak spectra were present at a signal to noise ratio above 2 and glycan pixel images were visible specifically in the tumor regions. Therefore, the score range for the specimens in this study can have a theoretical range from 0 to 53, with each point corresponding to the presence of a hyper-fucosylated glycan in the tissue specimen.

Two independent datasets were used for validation of the MALDI imaging data. Clinical–pathologic data and genomic data were downloaded from the cBioPortal for cancer genomics (http://www.cbioportal.org/; refs. 40, 41). First, a dataset of 494 primary prostate cancers (genomics data for 489 cancers) was obtained from the “Prostate Adenocarcinoma PanCancer Atlas” dataset from The Cancer Genome Atlas (refs. 42, 43). Second, a dataset of 444 CRPC specimens from 429 patients was obtained from the “Metastatic Prostate Adenocarcinoma SU2C/PCF Dream Team” dataset (44). Copy-number alteration (CNA) data were obtained as amplification (AMP), gain (GAIN), shallow deletion (HETLOSS), and deep deletion (HOMDEL) for all 11 FUTs (FUT1–FUT11) involved in N-glycan fucosylation. Tumor samples were selected if they had a gain or amplification for any of the 11 FUT genes. For analyses involving histology and location of metastatic tumors, tumors with increased copy number for any of the 11 FUT genes were further filtered by subtracting out any tumor that had a shallow or deep deletion for any of the FUT genes (N = 74). These data were further restricted to include only those FUT enzymes (FUT1, FUT3, FUT4, FUT5, FUT6, and FUT9) that can synthesize Le^y antigen (N = 50). These tumors were then compared with tumors that were diploid for all the 11 FUT genes (N = 43). Similarly, RNA expression for FUT1 to FUT11 was obtained in fragments per kilobase of transcripts per million mapped reads format from the metastatic prostate adenocarcinoma dataset as above and matched with exposure to the androgen receptor signaling inhibitors (ARSI) abiraterone and enzalutamide (naïve or exposed; N = 195) as well as AR CNAs (N = 208).

Continuous variables were reported as the median (±95% confidence interval), and categorical variables were reported as proportions, unless otherwise specified. Comparisons of continuous variables were performed using nonparametric Kruskal–Wallis or Mann–Whitney tests. Categorical variables were analyzed with Fisher exact or χ² tests. Statistical analyses were performed by using Prism 9.1.0 (GraphPad Software; RRID:SCR_002798). The significance of the receiver operator curves (ROC) was assessed by testing the null hypothesis that the AUC equaled 0.5. Two-tailed statistical tests were performed where applicable, and P < 0.05 was considered to indicate a statistically significant difference.

The genomic data generated in this study are publicly available at http://www.cbioportal.org/. MALDI data generated in this study are available upon request from the corresponding author.

Previously, our group used MALDI-MSI to identify N-linked glycans that were unique to prostate cancer (31, 37, 45), including hormone-sensitive and hormone-refractory tumors (32). In this current study, we hypothesized that prostate cancer with histologic small-cell/neuroendocrine features would have a unique N-linked glycan signature compared with prostate cancer with histologic features of conventional adenocarcinoma. An institutional panel of primary and metastatic prostate cancer tissues with phenotypic small-cell/neuroendocrine histology was selected along with a panel of primary and metastatic prostate cancer tissues with adenocarcinoma histology (Supplementary Table S1). Corresponding immunostaining of synaptophysin (SYN), chromogranin A (CgA), and the AR were done for each tissue. The tissue set was then analyzed by N-glycan MALDI-MSI to determine which N-glycan species were present in the tumors. Interestingly, we initially noted in a prostate cancer specimen with both adenocarcinoma and small-cell carcinoma that these histologies had distinct glycan profiles, as illustrated with a segmentation map of the specimen in Fig. 1A. Paucimannose and high-mannose glycans were a common tumor biomarker, as previously described (Fig. 1B; refs. 31, 33). Other expected branched tri- and tetra-antennary N-glycans with one or more sialic acids common to prostate adenocarcinomas were also detected (Fig. 1C; refs. 37, 45). However, the small-cell carcinoma expressed a class of fucosylated (n = 3–9 fucoses) tri- and tetra-antennary N-glycans (Fig. 1D). The overall configurations of these “hyper-fucosylated” N-glycans are shown in Supplementary Fig. S1 and are defined as those glycans with four base tri- and tetra-antennary glycan structures that progressively have from 3 to 9 fucose residues added. The enrichment of these hyper-fucosylated glycans in small-cell/neuroendocrine histology relative to adenocarcinoma histology is shown in Supplementary Fig. S2. The distinct spatial expression patterns of glycans in foci of adenocarcinoma and small-cell carcinoma were supported by H&E (Fig. 1E and F) and IHC for SYN, a neuroendocrine marker that was positive in the small-cell carcinoma (Fig. 1G), and the AR, which was negative in the small-cell carcinoma but positive in the adenocarcinoma (Fig. 1H). To add an additional structural level to these findings, we observed that a subset of the fucosylated N-glycans in Supplementary Table S1 had a Lewis-y (Le^y) tetrasaccharide antigen motif in which two fucoses are linked to a Gal and GlcNac via the linkage α-Fuc-(1→2)-β-Gal-(1→4)-(α-Fuc-[1→3])-GlcNAc. IHC enrichment of Le^y in the small-cell component confirmed the MSI (Fig. 1I; Supplementary Fig. S3).

Analyses of additional prostate cancer specimens further supported these findings. For reference, a Gleason grade group 5 (GG5) adenocarcinoma expressed high-mannose glycans but no evidence of hyper-fucosylated glycans (Fig. 2A; Supplementary Fig. S4). In contrast, a de novo prostate small-cell carcinoma metastasis to a lymph node (Fig. 2B; Supplementary Fig. S3) and a mixed ISUP GG5 adenocarcinoma and small-cell carcinoma primary tumor (Fig. 2C; Supplementary Fig. S5) expressed both high-mannose and complex hyper-fucosylated tetra-antennary glycans. Intriguingly, we observed that there was a class of adenocarcinomas that also expressed a subset of hyper-fucosylated glycans that were present in the small-cell carcinoma samples. In one example, a GG5 adenocarcinoma that metastasized to a lymph node expressed a hyper-fucosylated tetra-antennary glycan (Fig. 2D). The presence of these hyper-fucosylated N-glycans correlated with the presence of IHC expression of the neuroendocrine marker CgA (Supplementary Fig. S6).

These findings prompted us to investigate fucosylation patterns in the prostatectomy specimens and correlation with neuroendocrine IHC markers. As described in the “Materials and Methods” section, we selected 53 fucosylated N-glycans with at least 2 fucose residues (structures in Supplementary Table S3) to perform a hierarchical clustering approach to determine if subsets of adenocarcinomas had similar fucosylation patterns to the tumors with small-cell histology. Based upon the presence or absence of the individual fucosylated N-glycans in each tissue, this allowed assigning each tissue a numerical value for the number of fucosylated glycans detected, which served as the “fucose score.” Using this numerical fucose score, we discovered that the specimens with small-cell histology had the highest quantity of hyper-fucosylated N-glycans (Fig. 3A and B). Related to our above observation, there was a subset of adenocarcinomas that clustered with small-cell specimens that had higher abundance of the hyper-fucosylated N-glycans. We investigated the expression of the neuroendocrine markers CgA and SYN as a function of fucosylated glycan presence in both the small-cell and adenocarcinoma specimens. We discovered that higher staining in both instances was significantly associated with abundance of hyper-fucosylated N-glycans (Fig. 3C and D). Interestingly, not only was this phenomenon seen in specimens that were small-cell/neuroendocrine–based upon routine histologic features but also adenocarcinoma specimens without any identifiable small-cell/neuroendocrine features. This was further validated by AR staining, in which tumors that were AR-negative (i.e., pure small-cell specimens) had significantly higher abundance of hyper-fucosylated N-glycans compared with the AR-positive specimens (Fig. 3E).

Because the presence of neuroendocrine features in adenocarcinomas has been associated with adverse clinical features and poor outcomes, we wanted to determine if the presence of fucosylated N-glycans in histologic prostate adenocarcinomas was associated with poor outcomes using a TMA of prostatectomy specimens (Supplementary Table S2). We profiled the TMA for expression of the panel of N-glycans described above (Fig. 4A). We confirmed our previous findings that the presence of paucimannose and high-mannose glycans was general tumor markers in each of the cores. However, we discovered that cores of adenocarcinomas from prostatectomy specimens who later developed metastatic disease had higher abundance of hyper-fucosylated N-glycans than those that did not (Fig. 4B). Interestingly, despite this observation, no significant correlations were observed between fucosylated N-glycan abundance and tumor grade, stage, or serum PSA at prostatectomy (Supplementary Fig. S7). A ROC analysis based upon the ability of the fucose score (i.e., the number of detected hyper-fucosylated glycans to predict metastatic disease) revealed an AUC of 0.702 (P = 0.0002; Fig. 4C). Because we identified that a subset of fucosylated glycans containing the Le^y motif were enriched in small-cell carcinomas, we performed the same ROC analysis but restricted it to those hyper-fucosylated glycans that harbored the Le^y motif. This resulted in a greater and more significant AUC (0.783; P < 0.0001; Fig. 4D). These findings suggested a potentially important role for fucosylated N-glycans in metastatic disease.

Because there are 11 FUTs that fucosylated N-glycans, we investigated the presence of CNAs in genes encoding these enzymes in primary prostate tumors compared with metastatic prostate tumors using previously published datasets (42–44). We discovered that although there were significant enrichments in deletions of FUT3 to FUT7 (P < 0.001), there were more robust and significant enrichments in copy number of all 11 FUT genes (P < 0.0001; Fig. 4E). We further interrogated the metastatic prostate cancer dataset to identify if enrichments of the FUT genes were enriched in small-cell/neuroendocrine versus adenocarcinoma histology. Interestingly, not only was there significant (P < 0.05) enrichment of FUT copy number in metastatic tumors with small-cell/neuroendocrine histology (Fig. 4F) but the difference was more significant (P < 0.01) when copy number was restricted to the FUT with Le^y synthetic ability (Fig. 4G).

The class of drugs known as ARSIs includes enzalutamide and abiraterone and are commonly used in patients with metastatic prostate cancer. Using the metastatic prostate cancer dataset described above, we examined the effects of ARSI exposure on gene expression of FUT1 to FUT11. Interestingly, there were no differences in FUT gene expression in tumors from patients who were naïve to or exposed to ARSI (Supplementary Fig. S8A). Next, we investigated the effects of AR CNAs on FUT gene expression. Although there were significant decreases in expression of FUT2 (P < 0.05), FUT4 (P < 0.01), FUT8 (P < 0.05), and FUT10 (P < 0.01) in tumors with higher copy numbers of AR, these differences were no longer significant (q < 0.05) after correcting for multiple comparisons (Supplementary Fig. S8B). As these are genomic and transcriptomic datasets, we ascribed functional significance to these findings by leveraging MALDI N-glycan imaging of a TMA with tumors from LuCaP PDX models that were grown from advanced primary and metastatic prostate tumor specimens that were also subject to castration to investigate the effects of androgen deprivation on fucosylation (36). Supporting the findings above, there were no significant differences in the number of detected fucosylated N-glycans (i.e., the fucose score) in tumors that were naïve versus resistant to castration (Supplementary Fig. S8C). Together, these findings suggest that fucosylation of N-glycans may not be regulated by AR signaling in metastatic prostate cancer.

Next, we wanted to determine if fucosylated N-glycan profiles were differently expressed between primary and metastatic sites. Using the LuCaP PDX TMA, we discovered that the tumor cores consistently expressed mannosylated N-glycans (Fig. 5A), concordant with previous analyses. However, we observed that liver (i.e., visceral) metastases had the highest abundance of fucosylated N-glycans relative to primary prostate tumors and metastases to lymph nodes and bone (Fig. 5A and B). This is a clinically relevant finding, as the presence of liver metastases in patients with prostate cancer is one of the most lethal sites of disease and associated with very poor outcomes with a median survival of 10 to 14 months (46).

To validate the discovery that liver metastases were enriched with hyper-fucosylated N-glycans, we first analyzed FUT copy number in metastatic prostate tumors as a function of tumor origin. We discovered that liver metastases had a significantly higher copy number of Le^y synthesizing FUT genes (P < 0.01; Fig. 5C). As an additional level of validation, we obtained tissue specimens from a small cohort of patients (n = 3) with small-bowel neuroendocrine tumors. This patient cohort was selected, as small bowel neuroendocrine tumors frequently metastasize to the liver, thus giving us an opportunity to profile hyper-fucosylated N-glycan expression in not only the primary tumor but in corresponding liver metastases from the same patient. Interestingly, the fucose score was significantly higher in liver metastases versus the primary small-bowel tumors from all three patients (Fig. 5D). Together, these findings indicate that hyper-fucosylated N-glycans may be biomarkers for progression to lethal metastatic disease in prostate cancer and particularly the most aggressive clinical phenotype of prostate cancer (i.e., its likelihood to metastasize to the liver).

Prior studies of the molecular mechanisms of NEPC have highlighted genomic, epigenetic, and transcriptional alterations that promote disease progression (13, 47, 48). However, such studies are limited by the relative lack of tissue specimens to capture the temporal evolution of the disease and paucity of patient-derived preclinical models (49). The goal of this study was to leverage retrospective FFPE prostate cancer specimens with long-term outcome data and our MALDI-MSI platform to identify glycan biomarkers with prognostic clinical utility for prostate cancer. Consistent with prior studies of N-glycan expression in prostate cancer, we identified enrichment of branched tri- and tetra-antennary N-glycans (31, 37, 45) as well as subsets of these glycans with hyper-fucosylation in prostate cancer that correlated with expression of the neuroendocrine markers SYN and CgA (50).

As previously established, aberrant glycosylation can be an indicator of cancer and predictor of progression (17, 18, 51). N-glycan fucosylation is observed (37) and associated with aggressive features and worse prognosis in melanoma (52), pancreatic ductal adenocarcinoma (53), lung cancer (54), colorectal cancer (55), cholangiocarcinoma (56), hepatocellular carcinoma (57), glioblastoma (58), and breast cancer (59). Prior studies by our group and others have defined the glycomic landscape of prostate cancer progression, including CRPC (4, 19, 21, 32). For conventional adenocarcinoma, tri- and tetra-antennary branched N-glycans usually contain 1 to 3 fucoses. Fucoses can be added to N-glycans by FUT enzymes with different anomeric linkage specificities within the glycan structural backbone: outer arm α1,2 (FUT1 and FUT2); outer arm/Le^y α1,3 or α1,4 (FUT3–FUT7 and FUT9–FUT11); and core α1,6 (FUT8; refs. 23, 60). The Le^y antibody recognizes two fucoses linked on an antenna in α1,2 configurations on GlcNAc and Gal, and clearly many of the NEPC N-glycans possess this structure. It is unlikely to be the only source of the complex outer arm fucosylations identified; hence, the roles of the other outer arm (FUT3–FUT7/FUT9–FUT11) remain to be determined, most likely for the sialylated versions of them that have potential sialyl- Lewis-x or Lewis-a antigen motifs.

The colocalization of immunostaining of the neuroendocrine markers SYN and CgA with Le^y antigen and the hyper-fucosylated N-glycans in tissues with small-cell/neuroendocrine histology is consistent with what has been described for the glycosylation of SYN and other glycoproteins in the brain synapse and synaptic vesicles (61). As described herein for the small-cell/neuroendocrine specimens, the brain N-glycome is characterized by an abundance of high-mannose, bisecting fucose and hyper-fucosylated branched N-glycans (61). The synaptic vesicle contains many highly glycosylated glycoproteins, and a recent glycopeptidomic study in mouse brain has determined that multi-fucosylation (three or more) is an integral modification of SYN and other abundant synaptic glycoproteins (62). N-glycan imaging of the normal brain also identifies hyper-fucosylated tetra-antennary structures, which appear to be a normal feature of brain glycosylation (63). Thus, it is feasible that NEPC and other neuroendocrine tumors could be expected to have the same type of hyper-fucosylated N-glycans present as we demonstrate in this study.

The enrichment of N-glycan fucosylation in metastases to the liver compared with other sites is an intriguing observation that may provide insights into the heterogeneity of disease mechanisms between primary and metastatic sites. We have evaluated the N-glycosylation of adjacent nontumor regions of liver and lung tissues, published recently in a N-glycome atlas study of 15 tissues (37). There is no apparent N-glycosylation pattern that is unique to the liver or lung. In addition, we have also discovered that HCCs commonly express tetra-antennary N-glycans with 0 to 3 fucose residues common in the prostate adenocarcinoma tissues, but HCC tumors have neither the higher numbers of fucose residues (i.e., >3) nor the branched bisecting multi-fucosylated structures found in NEPC (64, 65). In terms of the fucose score criteria described herein, most HCC tumors would have fucose scores of 2 to 4 and normal liver tissue would have a fucose score of 0. It is currently unclear if hyper-fucosylation emerges as a primary tumor evolves or if there is clonal selection of hyper-fucosylated prostate cancer cells in the development of liver metastases. A recent study of prostate-specific membrane antigen expression in a cohort of patients with metastatic disease as well as in patient-derived xenograft models demonstrated overall decreased prostate-specific membrane antigen levels in hepatic metastases relative to primary lesions, independent of AR positivity or NEPC phenotype (66). These findings underscore the distinct biology of metastatic lesions, in a manner that is mediated by site-specific alterations as by modulatory interactions with the tumor microenvironment. Currently, we do not have an obvious explanation for this phenomenon, and we intend to expand these studies to evaluate the role of fucosylated N-glycans in cellular interactions between metastatic tumors and the liver.

It is also intriguing to speculate on the mechanistic role of N-glycan fucosylation in NEPC and metastatic prostate cancer. As discussed above, increased fucosylation is not specific to NEPC but present in multiple cancers. Interestingly, higher levels of N-glycan fucosylation are associated with worse outcomes in multiple cancers, suggesting that fucosylation may serve similar functions in NEPC and metastatic prostate cancer. First, fucosylation is associated with proliferation. Studies in cancer cell lines and xenograft models demonstrate that pharmacologic and genetic inhibition of fucose synthesis and glycan fucosylation inhibits cell proliferation and induces apoptosis (67, 68). Interestingly, Le^y antigen has been linked to increased cancer cell proliferation as well as activation of the mitogenic pathways PI3K/Akt (69) and EGFR-induced ERK/MAPK (ref. 70). Second, fucosylation is associated with invasion and metastasis. Terminal fucosylation, in particular, affects the synthesis of selectin ligands that regulate cellular interactions with the endothelium that are important for processes such as lymphocyte homing and tumor cell migration (71, 72). For example, knockdown of FUT3 in colorectal cancer cells inhibited liver metastasis in xenograft models (73). Third, fucosylation in tumor cells is implicated in resistance to chemotherapy. For example, increased FUT4, FUT6, and FUT8 expression in liver cancer cell lines are responsible for a multidrug resistance phenotype through the PI3K/Akt signaling pathway (74). Fourth, there are numerous studies that have linked fucosylation with immunity. Interestingly, fucosylation is a vital factor to normal immune function, as it is involved in the development and function of multiple types of immune cells (75). In particular, much attention has been given to FUT8, which mediates core N-glycan fucosylation. FUT8 is vital for development, viability, and activation of CD4⁺ T cells (76–78). However, conflicting data are present, indicating that fucosylation may actually inhibit immune function. Core fucosylation by FUT8 may actually induce CD8+ T-cell exhaustion by stabilizing cell-surface PD1 (79). In addition, FUT8-mediated core glycosylation of the B7H3 immune checkpoint protein can suppress T-cell activity in triple-negative breast cancers (80). Regardless of the potentially conflicting data about the role of fucosylation in tumor immunity, more studies are needed in this space, including comparing terminal with core fucosylation in immune-competent models of metastatic prostate cancer.

Many tools have been and are being developed to predict metastatic disease and future outcomes. In particular, the Decipher test is a 22-gene genomic classifier assay based on biopsy and prostatectomy tumor tissue that has been used as a prognostic tool for risk stratification of patients with localized prostate cancer. This test has been included in the National Comprehensive Cancer Network guidelines to potentially guide treatment decisions in certain cohorts of patients with prostate cancer (81). In a comprehensive systematic review of the evidence for Decipher in prostate cancer decision making, the AUC calculations for metastasis in most studies were in the range of 0.7 to 0.8 (82). In the present study, we find that this potentially novel biomarker, the fucose score, is both independent of clinical risk factors (grade, stage, and PSA) and has a moderate AUC of 0.7. Future considerations will include use of clinical cohorts to combine traditional clinical risk factors and the fucose score. Furthermore, combination of the fucose score with genomic tests, such as Decipher, may provide even further risk stratification for metastatic potential of primary prostate cancer, allowing for early diagnosis of potentially lethal disease.

Considering its mechanistic importance, fucosylation represents a promising target for therapeutics. Indeed, selective inhibitors of fucosylation have been developed (83, 84) that have demonstrated the potential to suppress tumor growth in in vitro and in vivo models of multiple cancers, including prostate cancer (59, 85). Although the sole clinical trial to date demonstrated preliminary evidence of antitumor activity, these clinical investigations for fucosylation inhibitors were terminated due to thromboembolic events (86). Continued studies of the structural and functional machinery of glycosylation and its alterations in cancer represent important opportunities for new and more effective treatments.

K. Bejar reports grants from the National Center for Advancing Translation Sciences of Health under Award Number T32TR004545 during the conduct of the study. D.A. Troyer reports personal fees from Provia Biologics outside the submitted work. R.J. Leach reports grants from the NIH during the conduct of the study. E. Corey reports grants from AstraZeneca, K36, MacroGenics, Bayer Pharmaceuticals, Forma Theraputics, Foghorn, KronosBio, AbbVie, Gilead, Janssen Research, Zenith Epigenetics, Genentech, and GSK outside the submitted work. R.R. Drake reports that the research group is part of a MUSC-Bruker Center of Excellence in Clinical Glycomics program and that Bruker makes the mass spectrometers they use for MS imaging of the glycans in the tissues, as well as that he, nor his laboratory, does not receive any compensation from this, or “free” items per se, that would constitute a conflict of interest or financial relationship. No disclosures were reported by the other authors.

J.E. Ippolito: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J.P. Hartig: Data curation, formal analysis, investigation, visualization, writing–review and editing. K. Bejar: Data curation, formal analysis, investigation, writing–review and editing. H. Nakhoul: Formal analysis, writing–original draft, writing–review and editing. J.K. Sehn: Formal analysis, investigation, methodology, writing–review and editing. C. Weimholt: formal analysis, investigation, methodology, writing–review and editing. G. Grimsley: Data curation, formal analysis, investigation, visualization, writing–review and editing. E. Nunez: visualization, writing–review and editing. N.A. Trikalinos: Data curation, validation, writing–review and editing. D. Chatterjee: Data curation, validation, writing–review and editing. E.H. Kim: Resources, funding acquisition, writing–original draft, writing–review and editing. A.S. Mehta: Resources, formal analysis, methodology, writing–review and editing. P.M. Angel: Resources, formal analysis, funding acquisition, investigation, visualization, methodology, project administration, writing-review and editing. D.A. Troyer: Data curation, formal analysis, writing–review and editing. R.J. Leach: Resources, data curation, writing–review and editing. E. Corey: Resources, data curation, writing–review and editing. J.D. Wu: Resources, funding acquisition, writing–review and editing. R.R. Drake: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing-review and editing.

We would like to thank the patients who generously donated tissue that made this research possible. Support is provided by Department of Defense W81XWH-19-1-0795 (J.E. Ippolito, J.D. Wu, R.R. Drake), NIH R01CA258960 (J.E. Ippolito, E.H. Kim), R01CA282022 (J.E. Ippolito, E.H. Kim, P.M. Angel, R.R. Drake), R01CA212409 (J.D. Wu, R.R. Drake), the American Cancer Society TLC-23-1180705-01-TLC (R.J. Leach, P.M. Angel), the Prostate Cancer Foundation (J.E. Ippolito), Siteman Cancer Center (J.E. Ippolito), and the Barnes Jewish Hospital Foundation (J.E. Ippolito). The characterization and maintenance of the LuCaP PDX models was supported by the Pacific Northwest Prostate Cancer SPORE (P50CA97186) and the P01 NIH grant (P01CA163227). The graphical abstract figure was created using Biorender.com.