Combined analysis of secreted proteins and glycosylation identifies prognostic features in cholangiocarcinoma

Abstract

Secreted proteins are overexpressed in cholangiocarcinoma and actively involved in promoting metastatic spread. Many of these proteins possess one or more sites of glycosylation and their various glycoforms have potential utility as prognostic or diagnostic biomarkers. To evaluate the effects of secretome glycosylation on patient outcome, we elucidated the glycosylation patterns of proteins secreted by parental and metastatic cholangiocarcinoma cells using liquid chromatography-mass spectrometry. Our analysis showed that the secretome of cholangiocarcinoma cells was dominated by fucosylated and fucosialylated glycoforms. Based on the glycan and protein profiles, we evaluated the combined prognostic significance of glycosyltransferases and secretory proteins. Significantly, genes encoding fucosyltransferases and sialyltransferases showed favorable prognostic effects when combined with secretory protein-coding gene expression, particularly thrombospondin-1. Combining these measures may provide improved risk assessment for cholangiocarcinoma and be used to indicate stages of disease progression.

Graphical Abstract

Introduction

Cholangiocarcinoma (CCA) is often diagnosed at a stage when regional metastasis has occurred, with a long-term survival rate of about 5–7%. Surgical resection is the dominant form of potentially curative treatment but only a small proportion of cases are deemed resectable, recurrence is common, and the median survival after resection is estimated to be less than three years (Endo et al., 2008). Improved strategies for early detection and prevention of metastasis are needed but the molecular basis of the disease remains elusive.

The metastatic cascade is comprised of a series of steps required for tumors to successfully dissociate from the primary tumor mass and produce a secondary tumor. Putative targets have been identified along the cascade in an effort to delay or limit metastasis. Secretory proteases have been shown to be directly involved in increasing the metastatic phenotypes of cancer cells through a number of molecular mechanisms, including angiogenesis, release of growth factors, in addition to cell invasion (Devy et al., 2009; Joyce et al., 2004; Parish et al., 2001). Indeed, selective and transient loss of the basement membrane that surround the tumor mass is characteristic of invasive carcinoma (Amenta et al., 2003; Kodama et al., 2005; Spaderna et al., 2006). However, despite the promise of protease inhibitor-based therapy against cancer, matrix metalloproteinase inhibitors (MMPIs) failed to reduce metastatic tumor burden in clinical trials and exhibited adverse side effects (Coussens et al., 2002). More recent studies have enhanced the understanding that certain proteases are responsible for generating products that suppress cell migration and invasion (Dufour and Overall, 2013; López-Otín and Matrisian, 2007). It is also now recognized that natural inhibitors of proteases, specifically metalloproteinase inhibitor 1 (TIMP1), plasminogen activator inhibitor 1 (PAI1), and protein C inhibitor, can have pro-tumorigenic effects independent of their inhibitory activity (Asanuma et al., 2007).

Approximately 14% of the human proteome is predicted to be transported out of the cell (Keerthikumar, 2016), with extracellular matrix (ECM) proteins and proteolytic enzymes being classic examples. Proteins destined for secretion are subject to glycan processing as they fully mature prior to export via the secretory pathway. The glycans that modify proteins can be large in size and exhibit conformational flexibility, influencing protein structure, protein activity, substrate recognition, and receptor binding (Borrok et al., 2012; Helle et al., 2007; Noach et al., 2017; Park et al., 2020; Peng et al., 2017; Zhou and Tsai, 2009). In particular, aberrant expression of secreted proteins has been shown in CCA, including laminin γ2, a major component of the extracellular matrix, and lipocalin 2, an iron-trafficking protein (Aishima et al., 2004; Srisomsap et al., 2010). However, it remains unclear whether metastatic CCA have acquired a unique set of secreted proteins to facilitate cell-matrix interactions and whether secreted protein functions are modulated by specific post-translational modifications such as glycosylation.

A greater understanding of the secretory pathway will help identify biomarkers with predictive and prognostic significance and therapeutic targets for prevention of cancer metastasis. To assess the impact of glycan modifications of the secretome on cancer progression, we compared the glycosylation profiles of secreted proteins from highly metastatic CCA cells to those from their parental cells. This analysis elucidated a subset of CCA-derived glycoproteins that were used to associate glycogenes and secreted protein-coding genes with overall survival. The data highlight the need to discern glycan structures and glycosylation sites when evaluating the role of secreted proteins in metastatic CCA. Changes arising from glycan expression of secreted proteins in CCA may contribute to the enhanced capability of disseminated tumors to invade basement membrane barriers and support cell growth.

Materials & Methods

Cell culture.

Human cholangiocarcinoma cells (KKU-213A, KKU-213B) were obtained from the Japanese Collection of Research Bioresources Cell Bank (JCRB, Osaka, Japan) and used to establish highly metastatic cholangiocarcinoma cells (KKU-213AL5, KKU-213BL5) as previously described (Saentaweesuk et al., 2018; Sripa et al., 2020; Uthaisar et al., 2016). Cells were grown in Ham’s F12 media supplemented with 10% (v/v) fetal bovine serum, 100 U/mL penicillin, and 100 μg/mL streptomycin and maintained at 37°C in a humidified incubator with 5% CO₂.

Collection of secreted proteins.

Cells were incubated in triplicates in serum-free Ham’s 12 media for 24 hours. The conditioned media were collected and centrifuged at 2,000 × g for 10 min. Proteins in the media were concentrated from 40 mL to 200 μL of 100 mM ammonium bicarbonate by buffer exchange using Vivaspin 6 centrifugal concentrators (Vivaproducts, Littleton, MA). Protein concentration was measured using an Implen NanoPhotometer P300 (Implen, Westlake Village, CA) according to the manufacturer’s protocol and equalized across all fractions.

Preparation of N-glycans.

A fraction (25 μL) of each concentrated protein sample was mixed to a final volume of 100 mM ammonium bicarbonate in 5 mM dithiothreitol and heated for 10 s at 100°C to thermally denature the proteins. To cleave N-glycans from membrane proteins, 2 μL of peptide N-glycosidase F (New England Biolabs, Ipswich, MA) were added and samples were incubated in a microwave reactor (CEM Corporation, Matthews, NC) at 37°C for 10 min at 20 watts. After addition of chilled ethanol (1:4), samples were placed in −80°C overnight and centrifuged for 20 min at 21,000 × g to precipitate residual deglycosylated proteins. The supernatant containing the released N-glycans was collected and dried. N-Glycans were purified by solid phase extraction using a porous graphitized carbon (PGC) matrix (Grace, Columbia, MD). Eluted fractions were dried in vacuo.

Preparation of glycopeptides.

The remaining fraction (175 μL) of each concentrated protein sample was denatured in 8 M urea at 55°C, reduced with 18 mM dithiothreitol, alkylated with 27 mM iodoacetamide, diluted to 1 M urea, and incubated with 2 μg trypsin at 37°C overnight. Following incubation, 80% of the digested proteins were dried in vacuo and reconstituted in 1 mL of 1% (v/v) trifluoroacetic acid and 80% (v/v) acetonitrile in water. Glycopeptides were enriched by solid-phased extraction using iSPE-HILIC cartridges (Nest Group, Southborough, MA). Cartridges were conditioned with acetonitrile and 0.1% (v/v) trifluoroacetic acid in water, followed by 1% (v/v) trifluoroacetic acid and 80% (v/v) acetonitrile in water. Samples were loaded onto the column and washed with 1% (v/v) trifluoroacetic acid and 80% (v/v) acetonitrile in water. Enriched products were eluted with 0.1% (v/v) trifluoroacetic acid in water and dried in vacuo prior to mass spectrometric analysis.

Preparation of peptides.

The remaining 20% of digested proteins were subjected to proteomic analysis. Peptides were enriched using C18 cartridges (Agilent Technologies, Santa Clara, CA). Cartridges were washed and conditioned with acetonitrile followed by 0.1% (v/v) trifluoroacetic acid in water. Samples were introduced to the column, washed with 0.1% (v/v) trifluoroacetic acid in water, and eluted with a solution of 80% (v/v) acetonitrile and 0.1% (v/v) trifluoroacetic acid in water. Peptides were dried in vacuo prior to mass spectrometric analysis.

Glycan LC-MS/MS.

Glycan samples were reconstituted in nanopure water and analyzed using a nano-LC-Chip-QTOF-MS/MS system (6520, Agilent Technologies). Samples were maintained at 6°C and introduced to the MS with a microfluidic chip, which consists of enrichment and analytical columns packed with PGC and a nanoelectrospray tip. A binary gradient was applied at a flow rate of 0.4 μL/min: (A) 3% (v/v) acetonitrile and 0.1% (v/v) formic acid in water and (B) 90% (v/v) acetonitrile in 1% (v/v) formic acid in water. MS spectra were acquired at 1.5 s per spectrum over a mass range of m/z 600–2000 in positive ionization mode. Mass inaccuracies were corrected with reference mass m/z 1221.991.

Collision-induced dissociation (CID) was performed with nitrogen gas using a series of collision energies (V_collision) dependent on the m/z values of the N-glycans, based on the equation:

$${\text{V}}_{\text{collision}}=\text{slope}\left(m/z\right)+\text{offset},$$

where the slope and offset were set at (1.8/100 Da) V and −2.4 V, respectively.

Peptide LC-MS/MS.

Purified peptides (1 μg) were loaded using 2% (v/v) acetonitrile and 0.1% (v/v) trifluoroacetic acid in water and analyzed using a reverse-phase Michrom Magic C18AQ column (200 μm, 150 mm) coupled with a Q Exactive Plus Orbitrap mass spectrometer through a Proxeon nano-spray source (Thermo Fisher Scientific, Waltham, MA). A binary gradient was applied at 0.3 μL/min using (A) 0.1% (v/v) formic acid in water and (B) 0.1% (v/v) formic acid in acetonitrile. Per acquisition, the instrument was run in data-dependent mode as follows: spray voltage, 2.0 kV; ion transfer capillary temperature, 250°C; MS scan range, m/z 350–2000; precursor resolution, 70,000; MS automatic gain control, 1e6; MS maximum injection time, 30 ms; precursor ion isolation window, 1.6 m/z; MS/MS scan range, m/z 200–2000; MS/MS automatic gain control, 5e4; MS/MS maximum injection time, 50 ms; product ion resolution, 17,500; higher-energy collisional dissociation (HCD) collision energy, 27.

Glycopeptide LC-MS/MS.

Purified glycopeptides (1 μg) were loaded using 2% (v/v) acetonitrile and 0.1% (v/v) trifluoroacetic acid in water and analyzed using a reverse-phase Michrom Magic C18AQ column (200 μm, 150 mm) coupled with a Q Exactive Plus Orbitrap mass spectrometer through a Proxeon nano-spray source (Thermo Fisher Scientific). A binary gradient was applied at 0.3 μL/min using (A) 0.1% (v/v) formic acid in water and (B) 0.1% (v/v) formic acid in acetonitrile. Per acquisition, the instrument was run in data-dependent mode as follows: spray voltage, 2.0 kV; ion transfer capillary temperature, 250°C; MS scan range, m/z 700–2000; precursor resolution, 70,000; MS automatic gain control, 1e6; MS maximum injection time, 30 ms; precursor ion isolation window, 1.6 m/z; MS/MS scan range, m/z 200–2000; MS/MS automatic gain control, 5e4; MS/MS maximum injection time, 50 ms; product ion resolution, 17,500; fixed first mass, 130 m/z; stepped collision energy, HCD stepped collision energy, 17, 27, 37.

Glycan data analysis.

N-Glycan compounds were identified with an in-house retrosynthetic library of all possible glycan compositions according to accurate mass (Kronewitter et al., 2009). Signals above a signal-to-noise ratio of 5.0 were filtered and deconvoluted using MassHunter Qualitative Analysis B.06.01 (Agilent Technologies). Deconvoluted masses were compared to theoretical masses using a mass tolerance of 20 ppm and a false discovery rate of 0.6%. Abundances were determined by integrating ion counts for observed glycan masses. Relative abundances were determined by normalizing abundances to the summed peak areas of all glycans detected. Statistical evaluation of significant glycan abundance changes was performed using an unpaired, two-tailed Student’s t-test.

Glycopeptide data analysis.

Raw data were exported using Xcalibur 4.0 (Thermo Fisher Scientific). Proteins were identified using Byonic 2.7.4 (Protein Metrics, Cupertino, CA) against the reviewed Swiss-Prot human protein database with sample-specific parameters as determined from Preview (Protein Metrics): mass tolerance of 5 ppm for the precursor and 10 ppm for fragment ions; carbamidomethylation of cysteine as a fixed modification; oxidation of methionine, dioxidation of tryptophan, deamidation of asparagine and glutamine, acetylation of the protein N-terminus, N-terminal pyroglutamate formation of glutamine and glutamate, N-terminal ammonia loss of cysteine, and N-glycosylation of asparagine as variable modifications; two missed cleavage sites. Identifications were filtered with a 1% false discovery rate, |Log Prob| > 2, and Delta Mod > 10. Site-specific quantitation was performed using the area under the curve of the extracted ion chromatogram which was normalized to the total protein concentration.

Peptide data analysis.

Raw data were exported using Xcalibur 4.0 (Thermo Fisher Scientific). Proteins were identified using MaxQuant 1.5.7.4 (Cox and Mann, 2008) against the reviewed Swiss-Prot human protein database (20,406 entries) with the following parameters: carbamidomethylation of cysteine as a fixed modification; oxidation of methionine, deamidation of asparagine and glutamine, dioxidation of methionine and tryptophan, and acetylation of the protein N-terminus as variable modifications; specific cleavage, trypsin; two missed cleavage sites. Identifications were filtered with a 1% false discovery rate and |Log Prob| > 2. Label-free quantification analysis was employed and protein intensities were normalized to the total intensity of each sample.

Data availability.

The raw data used in this study have been deposited to the MassIVE repository and are available online at ftp://massive.ucsd.edu/MSV000091721. All other data are available upon reasonable request.

RNA isolation and qRT-PCR analysis.

Cells were harvested, washed twice with PBS, and resuspended in RNAlater (Life Technologies, Carlsbad, CA). Total RNA were extracted using RNeasy plus mini kit (Qiagen, Germantown, MD) and the quantity and quality of RNA were determined using a Qubit Fluorometer (Life Technologies) and TapeStation 2200 (Agilent Technologies) following the manufacturer’s protocol. Total RNA were reverse transcribed to cDNA using iScript Reverse Transcription Supermix (Bio-Rad Laboratories, Hercules, CA) following the manufacturer’s protocol. Predesigned human glycosylation and extracellular matrix and cytoskeleton PrimePCR plates (Bio-Rad Laboratories) were used for real-time PCR using the CFX96 Touch Real-Time PCR detection system (Bio-Rad Laboratories) and the analysis was performed using CFX Manager 3.1 (Bio-Rad Laboratories). Expression was normalized to that of the reference gene, GAPDH. Enrichment networks were prepared using ExpressAnalyst 3.0 (Zhou et al., 2019).

Clinical analysis.

Fragments per kilobase million (FPKM) normalized transcriptomic data together with the clinical data were obtained from The Cancer Genome Atlas (TCGA) research network under project IDs TCGA-CHOL and TCGA-PANCAN and stratified by the median (Goldman et al., 2020; Weinstein et al., 2013; Xu et al., 2019). Kaplan-Meier analysis, correlation analysis, and statistical analysis were performed using Prism 9.0.2 (GraphPad Software, San Diego, CA). For comparison of survival differences between groups, a log rank (Mantel-Cox) test was used. For comparison of gene expression between sample types (normal vs. tumor, tumor vs. metastasis), a Welch’s t-test was used. All P-values are two-sided.

Results

Expression levels of secretory protein-coding genes are correlated with survival of patients with cholangiocarcinoma

A total of 2072 gene entries classified as extracellular or secreted proteins were obtained from the human proteome database, Swiss-Prot. Cholangiocarcinoma cases from The Cancer Genome Atlas (TCGA) database were stratified by high (quantile 2, q2) or low (quantile 1, q1) gene expression and compared using Kaplan-Meier analysis. Among the curated secretory protein-coding genes, expression of 1782 genes were accessible from the TCGA-CHOL cohort (Farshidfar et al., 2017) (Table S1). Differences between survival curves were evaluated using the log rank test with a minimum sample size of 10 per group. Significantly, low expression of 22 genes and high expression of 29 genes corresponded with worse survival (P < 0.05) (Figures S1–S2). Genes that showed statistically significant correlations with survival were enriched mainly in cancer, cytokine-cytokine receptor interaction, PI3K-Akt signaling, focal adhesion, neuroactive ligand-receptor interaction, and ECM-receptor interaction pathways (Figure 1A). The majority of these genes encode secretory protein isoforms that are predicted to be N- and/or O-glycosylated, often at more than one glycosylation site (Table 1). This analysis suggests that the expression levels of secreted glycoproteins may impact cholangiocarcinoma patient outcomes.

Figure 1.

Secreted protein-coding genes enriched in cholangiocarcinoma and their protein products. (A) Enriched gene sets from the KEGG database that correlated with survival in cases from the TCGA-CHOL cohort. Nodes are colored according to P-value and sized according to number of genes. Nodes with overlapping genes are connected by edges. (B-C) Quantitative RT-PCR analysis of extracellular matrix and adhesion protein-coding genes. Expression in KKU-213A, KKU-213AL5, KKU-213B, and KKU-213BL5 were normalized to that of GAPDH. Annotated dots represent genes that were upregulated more than threefold (upper threshold) or downregulated more than threefold (lower threshold) in metastatic CCA compared to their respective parental CCA cells. (D-E) Proteins secreted by KKU-213A and KKU-213AL5, grouped according to their localization and ordered according to intensity: b, blood component; e, extracellular matrix; o, other or unknown. (F) Common extracellular matrix (ECM) proteins identified in KKU-213A and KKU-213AL5 grouped by protein type and presented as percent change (KKU-213A to KKU-213AL5).

Table 1.

Secretory protein-coding genes correlated with survival in cholangiocarcinoma

Gene Name	Correlated with Worse Survival	P-value (Log Rank)	Number of Predicted Glycan Sites
			N-Linked	O-Linked
ADAMTS8	q1	0.019	4	27
ANTXR2	q1	0.011	2	4
ARSF	q1	0.039	3	12
CD163	q1	0.036	5	14
COL4A3	q1	0.025	1	92
COL4A4	q1	0.0021	2	69
DEFA6	q1	0.043	0	2
GDF5	q1	0.013	1	19
GH1	q1	0.015	0	4
GUCA2A	q1	0.015	0	0
HAPLN1	q1	0.028	2	0
LAMC1	q1	0.037	11	80
LILRA2	q1	0.037	7	8
LILRB1	q1	0.026	3	15
MXRA5	q1	0.046	15	213
NOV	q1	0.016	1	15
NPPC	q1	0.0078	0	4
OLR1	q1	0.0054	2	2
PM20D1	q1	0.00042	2	11
PTEN	q1	0.039	0	32
SIGLEC1	q1	0.011	12	10
XCL2	q1	0.048	0	14
AOC1	q2	0.046	4	10
C1QBP	q2	0.002	3	7
CFHR3	q2	0.0033	3	8
CLPSL2	q2	0.024	0	0
COL7A1	q2	0.041	3	90
CRLF1	q2	0.037	3	9
CST1	q2	0.014	0	0
CXCL17	q2	0.021	0	5
DEFA3	q2	0.0099	0	0
EDN2	q2	0.043	0	17
ENTPD6	q2	0.015	2	10
ITLN2	q2	0.022	0	1
KLK3	q2	0.0096	1	0
KLK12	q2	0.038	2	2
LEFTY1	q2	0.036	1	9
LOXL1	q2	0.004	0	34
LY6G5C	q2	0.024	0	0
MDK	q2	0.013	0	0
MSMB	q2	0.029	1	0
NPW	q2	0.04	0	9
PROS1	q2	0.038	3	12
RLN3	q2	0.033	0	7
SMPDL3B	q2	0.03	3	2
SSC4D	q2	0.04	0	21
TECTB	q2	0.044	3	0
THPO	q2	0.046	6	38
TNFSF12	q2	0.031	1	2
TNFSF13	q2	0.0055	1	0
WFDC10B	q2	0.017	0	0

Extracellular matrix proteins are enriched in metastatic CCA cell models

To determine whether the expression of secreted proteins is altered in relation to CCA progression, metastatic CCA cells, KKU-213AL5 and KKU-213BL5, and their respective parental cells, KKU-213A and KKU-213B, were established as described previously (Sripa et al., 2020; Uthaisar et al., 2016). In comparing the transcriptomes of KKU-213A to KKU-213AL5, a number of genes encoding ECM and adhesion proteins exhibited changes above the three-fold threshold, among which a greater proportion of genes were expressed in higher abundances in the highly metastatic than in the parental group (Figure 1B). Specifically, upregulated genes in KKU-213AL5 compared to KKU-213A included MMP9, LAMA3, COL12A1, TNC, LAMB1, and LAMB3 (Figure 1C). Extracellular protein-coding genes were more resistant to expression level changes in the KKU-213B and KKU-213BL5 pair. These results indicate that a subset of extracellular protein-coding genes may be associated with more aggressive CCA.

We next isolated proteins in the conditioned media of parental and metastatic CCA cells, particularly KKU-213A and KKU-213AL5, to globally inspect the extracellular milieu of secreted proteins. KKU-213A and KKU-213AL5 showed about 89% similarity in the types of proteins secreted out of the cell (Table S2). All identified proteins were classified according to extracellular location as annotated in the Human Protein Atlas (Uhlén et al., 2019). Of the 236 proteins identified, 87 (37%) were distinguished as proteins that are secreted into blood while 31 (13%) were characteristic of proteins that are secreted to the ECM and 118 (50%) that are secreted locally, to an unknown location, or as part of extracellular exosomes (Figure 1D,E). Among proteins that were not blood-bound, eight major proteins comprised nearly a third of the secreted proteins from KKU-213A by abundance, including thrombospondin-1, β2 microglobulin, CCN family member 1, 14-3-3 protein sigma, heat shock protein 90β, laminin α5, calreticulin, and agrin. Similarly, seven major non-blood-bound proteins comprised a third of the secreted proteins from KKU-213AL5 by abundance, including thrombospondin-1, β2 microglobulin, calreticulin, CCN family member 1, amyloid-beta precursor protein, heat shock protein 90β, and laminin α5. In general, a modest relative increase in the abundance of basement membrane proteins laminin, perlecan, agrin, and ladinin as well as ECM degradative enzymes was observed in KKU-213AL5 compared to KKU-213A (Figure 1F). This analysis identified the types of CCA-derived proteins that were released into the surrounding medium during cell culture.

Compositional diversity of N-glycans from proteins secreted by CCA

The secreted pool of proteins from KKU-213A and KKU-213AL5 was comprised of a diverse array of glycans. Oligomannose and undecorated complex/hybrid type glycans (afucosylated and asialylated) were minor components of the N-glycans released from secreted proteins while purely fucosylated and fucosialylated (fucosylated and sialylated) glycans were the most abundant (Figure 2A,B; Table S3). Common to KKU-213A and KKU-213AL5, Hex₅HexNAc₄Fuc₁NeuAc₁, Hex₅HexNAc₄Fuc₁NeuAc₂, and Hex₅HexNAc₄Fuc₁ were more than two-fold higher in abundance than any other N-glycan observed in the secreted pool. Comparatively, the glycan profile from proteins secreted by KKU-213B and KKU-213BL5 was distinguished from the secreted glycan profile of KKU-213A and KKU-213AL5, representing differences expected from tumor origin heterogeneity (Figure S3–S4). Both KKU-213A/KKU-213AL5 and KKU-213B/KKU-213BL5 pairs were similar in that decorated glycan structures were highly abundant in their secretomes, among which asialylated, fucosylated glycans were the most abundant. In comparing metastatic to their parental cells, similar relative levels of fucosylation were observed in both pairs (Figure 2C). Accounting for all fucosylated glycans, those with one or more fucose residue comprised 78% of the total glycans identified in KKU-213A, 83% in KKU-213AL5, 75% in KKU-213B, and 83% in KKU-213BL5. A greater proportion (~70:30) of mono-fucosylated structures was observed over those with more than one fucose residue in KKU-213A and KKU-213AL5. This ratio was more evenly distributed (~50:50) in KKU-213B and KKU-213BL5.

Figure 2.

N-glycosylation of secreted proteins in parental and metastatic CCA. (A-B) Extracted compound chromatogram of glycans released from the pool of proteins secreted by KKU-213A and KKU-213AL5. Each peak corresponds to a unique glycan compound and is colored according to glycan type. Glycans are named according to composition: Hex_HexNAc_Fuc_NeuAc. Cpd, compound. (C) Ratio between mono-fucosylated (Mono-Fuc) glycan structures and those that displayed more than one fucose residue (Multi-Fuc) in the secreted proteins of KKU-213A, KKU-213AL5, KKU-213B, KKU-213BL5. (D) Quantitative RT-PCR analysis of genes encoding proteins from the N-glycan biosynthesis/catabolism pathways. Expression in KKU-213A, KKU-213AL5, KKU-213B, and KKU-213BL5 were normalized to that of GAPDH. Annotated dots represent genes that were upregulated more than threefold (upper threshold) or downregulated more than threefold (lower threshold) in metastatic CCA compared to their respective parental CCA cells. (E) (F) Number of glycosylation sites observed in the secretome of KKU-213A and KKU-213AL5 that were occupied with oligomannose, afucosylated and asialylated (aFuc & aSia), fucosylated (Fuc), fucosialylated (Fuc & Sia), or sialylated N-glycans. (G-H) Combined number of glycoforms observed in KKU-213A and KKU-213AL5 with oligomannose, afucosylated and asialylated (aFuc & aSia), fucosylated (Fuc), fucosialylated (Fuc & Sia), or sialylated (Sia) N-glycan. (I-J) Number of glycoforms observed at each occupied glycosylation site of THBS1 and CLU from KKU-213A and KKU-213AL5. Pie charts indicate the distribution of site-specific glycoforms.

In support of the observed glycan changes, expression of a subgroup of genes that encode proteins involved in the biosynthesis of complex/hybrid glycans were mostly unaltered when comparing parental to metastatic CCA (Figure 2D). A notable difference between the cell populations was observed in the expression of the β-galactoside α2,3-sialyltransferase-coding gene, ST3GAL1, which was lower in both KKU-213A (5.7-fold) and KKU-213B (12-fold) in comparison to KKU-213AL5 and KKU-213BL5, respectively. In addition to the biosynthetic pathway, we analyzed genes that encode proteins involved in the catabolism of N-glycans. An increase in the expression of the sialidase-encoding gene, NEU1, was observed in both KKU-213A (4-fold) and KKU-213B (4.9-fold) compared to KKU-213AL5 and KKU-213BL5, respectively.

Based on the number of unique glycan compositions, glycans from KKU-213B- and KKU-213BL5-secreted proteins were more similar to one another than those from KKU-213A- and KKU-213AL5-secreted proteins (Figure S4). While equal amounts of secreted proteins were analyzed from KKU-213A and KKU-213AL5, the latter secreted proteins with a greater variety of glycan structures, mainly attributable to fucosylation and sialylation (Figure 2E). In general, similar types of glycan compositions were expressed in KKU-213B and KKU-213BL5 as those in KKU-213A and KKU-213AL5, suggesting that these glycans may associate specifically with secreted proteins in CCA.

Site-specific heterogeneity of glycosylated proteins secreted by CCA

Theoretical N-glycan compositions were then compiled into a library to bracket the search for intact glycopeptides (Kronewitter et al., 2009). Secreted proteins originating from KKU-213A and KKU-213AL5 presented glycosylation sites that were most frequently occupied with fucosylated glycans (Figure 2F; Table S4). The majority of identified glycoforms across all secreted proteins were comprised of fucosylated glycan structures (Figure 2G–H). Among glycoproteins secreted by both KKU-213A and KKU-213AL5, metalloproteinase inhibitor 1, prosaposin, thrombospondin-1, clusterin, and galectin-3-binding protein were heavily glycosylated, displaying at least ten unique glycoforms within one or more glycosylated site. Notably, thrombospondin-1 (THBS1), clusterin (CLU), and galectin-3-binding protein (LGALS3BP) displayed extreme heterogeneity, with CLU bearing up to over 100 glycoforms (Figure 2I–J).

Combined prognostic significance of glycosyltransferase- and secretory protein-coding genes

From the TCGA-CHOL cohort, FUT4 was among the highest expressed fucosyltransferase (FUT)-coding genes in CCA tissue while FUT9 was among the lowest expressed (Figure 3A). Compared to normal tissue, the expression of FUT genes (FUT1–FUT11) was significantly upregulated in CCA with the exception of FUT6, which remained unchanged (Figure S5). A stratified analysis of fucosyltransferases was performed with the cohort divided into equal quantiles of low (q1) or high (q2) expression of FUT1–FUT11, which excluded consideration of FUT5 and FUT9 due to limited data. A five-year Kaplan-Meier analysis showed that tumors with low (q1) expression of FUT4 correlated with reduced survival than with high (q2) expression (P = 6.4e-4) (Figure 3B). In contrast to FUT4, varied expression of FUT1, FUT2, FUT3, FUT6, FUT7, FUT8, FUT10, and FUT11 did not significantly correlate with survival (Figure S6). Interestingly, we identified plasma α-fucosidase (FUCA2) as among glycoproteins secreted by CCA cells, which is responsible for cleaving α1,6-linked fucose residues transferred onto the N-acetylglucosamine (GlcNAc) residue at the reducing end of N-glycans by FUT8 and is itself glycosylated (Table S4). FUCA2 expression was negatively correlated with FUT4 (P = 7.9e-3) and FUT11 (P = 2.2e-2) expression in human tissues (Figure 3C–D).

Figure 3.

Combined prognostic significance of fucosyltransferase- and fucose-bearing secretory protein-coding genes. (A) Expression levels of fucosyltransferase genes in cholangiocarcinoma cases from the TCGA-CHOL cohort. Data are presented as mean ± SD. FPKM, fragments per kilobase million. (B) Kaplan-Meier analysis correlating FUT4 expression with survival. The P-value was determined using a log rank test. q, quantile. Pairwise correlation between (C) FUCA2 and FUT4 expression and between (D) FUCA2 and FUT11 expression in cholangiocarcinoma cases from the TCGA-CHOL cohort. The outermost edges of the shaded area represent the 95% confidence bands. r, Pearson correlation coefficient. (E) Workflow for stratification of cases into quantiles (q) based on the expression of fucosyltransferase- and fucose-bearing secretory protein-coding genes (n = 36). (F-J) Kaplan-Meier analysis showing correlation of secretory protein-coding genes with survival based on concurrent low (q1) or high (q2) expression of fucosyltransferase-coding genes. P-values were determined using a log rank test.

A similar analysis was conducted of cases stratified by quantiles of low (q1) or high (q2) expression of 14 secreted glycoproteins that exhibited glycoforms with multiple fucose residues in both KKU-213A and KKU-213AL5 jointly with low (q1) or high (q2) expression of genes encoding fucosyltransferases (Figure 3E). Expression of secreted multi-fucosylated protein-coding genes alone did not show significant correlations with survival (Table S1). However, low (q1) THBS1 expression was correlated with reduced survival compared to high (q2) THBS1 when FUT2 expression was concurrently low (q1) (P = 4.1e-2) (Figure 3F). In cases with high (q2) FUT2 expression, a significant association of THBS1 with survival was not observed (P = 0.49). Likewise, low (q1) THBS1 and low (q1) FUT10 expression was correlated with reduced survival compared to high (q2) THBS1 and low (q1) FUT10 expression (P = 4.7e-2) (Figure 3G). THBS1 expression was uncorrelated with survival when FUT10 expression was high (q2) (P = 0.60). Low (q1) AGRN expression was correlated with reduced survival compared to high (q2) AGRN expression when the expression of FUT7 was below the median (q1) (P = 4.4e-2). A correlation with survival was not observed when FUT7 expression was above the median (q2) (P = 0.47) (Figure 3H). Similarly, subjects with low (q1) ALCAM and low (q1) FUT7 expression had worse survival than those with high (q2) ALCAM and low (q1) FUT7 expression (P < 5.0e-2) (Figure 3I). Furthermore, LGALS3BP expression was correlated with survival only when FUT4 expression was high (q2) (P = 2.8e-2) (Figure 3J). We verified that the difference in survival was not due to correlations in expression between THBS1 and FUT2, THBS1 and FUT10, AGRN and FUT7, ALCAM and FUT7, or LGALS3BP and FUT4 (Figure S7). This analysis highlights the potential effect of α1,3/4-fucosylation on the prognostic value of proteins secreted by CCA cells.

Besides fucosyltransferase-coding genes, β-galactoside α2,3-sialyltransferase (ST)- and α2,6-sialyltransferase-coding genes were highly expressed in CCA tissue with ST6GAL1 showing the highest expression (Figure 4A). Compared to normal tissue, the expression of ST6GAL1, ST3GAL3, and ST3GAL6 was significantly downregulated in CCA while the expression of ST6GAL2 and ST3GAL4 remained unchanged (Figure S8). ST6GAL1, ST6GAL2, ST3GAL3, ST3GAL4, and ST3GAL6 expression were not associated with overall survival (Figure S9). A group of 12 secreted glycoproteins were identified that bear sialic acid in both KKU-213A and KKU-213AL5 and their coding genes were stratified by quantiles of low (q1) or high (q2) expression jointly with low (q1) or high (q2) expression of relevant β-galactoside α-sialyltransferase-coding genes (Figure 4B). Expression of secreted sialylated protein-coding genes alone did not show significant correlations with survival (Table S1). Notably, when ST6GAL2 expression was high (q2), low (q1) THBS1 expression was significantly correlated with reduced survival compared to high (q2) THBS1 expression (P = 4.8e-2) (Figure 4C).

Figure 4.

Combined prognostic significance of sialyltransferase- and sialic acid-bearing secretory protein-coding genes. (A) Expression levels of sialyltransferase genes in cholangiocarcinoma cases from the TCGA-CHOL cohort. Data are presented as mean ± SD. FPKM, fragments per kilobase million. (B) Workflow for stratification of cases into quantiles (q) based on the expression of sialyltransferase- and sialic acid-bearing secretory protein-coding genes (n = 36). (C-E) Kaplan-Meier analysis showing correlation of secretory protein-coding genes with survival based on concurrent low (q1) or high (q2) expression of sialyltransferase-coding genes. P-values were determined using a log rank test.

THBS1 expression did not correlate with survival when ST6GAL2 expression was low (q1) (P = 0.39). In addition, low (q1) THBS1 expression was correlated with reduced survival with high (q2) ST3GAL6 expression (P = 4.0e-2) than high (q2) THBS1 expression (Figure 4D). Compared to low (q1) FUCA2 expression, high (q2) FUCA2 expression was correlated with reduced survival when the expression of ST3GAL6 was above the median (q2) (P = 3.3e-2). FUCA2 expression was uncorrelated with survival when the expression of ST3GAL6 was below the median (q1) (P = 0.94) (Figure 4E). We verified that the difference in survival was not due to correlations in expression between THBS1 and ST6GAL2, THBS1 and ST3GAL6, or FUCA2 and ST3GAL6 (Figure S10). Collectively, the data show that the prognostic potential of secreted proteins may depend on the presence of specific glycoforms.

Using the findings from the TCGA-CHOL cohort, we analyzed the combined prognostic value of genes related to the fucosylation and sialylation of glycoproteins and those encoding secretory proteins in a larger cohort of patients with liver hepatocellular carcinoma (TCGA-LIHC; 370 cases). Among fucosyltransferase-coding genes, the highest expressed in primary tumor tissues was FUT6 and the lowest expressed was FUT9 (Figure S11). High (q2) expression of FUT4 and FUT11 was significantly correlated with reduced survival when compared to low (q1) expression (Figure S12). Among sialyltransferase-coding genes, the highest expressed was ST6GAL1 and the lowest was ST6GAL2 (Figure S11). High (q2) expression of ST3GAL4 was correlated with reduced survival when compared to low (q1) expression (P = 4.0e-3). In contrast, low (q1) expression of ST6GAL1 was correlated with reduced survival when compared to high expression (P = 2.7e-2) (Figure S12). Among the fucosylated secretory proteins that correlated with survival in CCA, a significant correlation with survival was found in hepatocellular carcinoma between low (q1) and high (q2) expression of AGRN (P = 8.1e-3) and FUCA2 (P = 1.0e-3). When stratified by low (q1) or high (q2) expression of fucosyltransferase-coding genes, the expression of AGRN was correlated with survival only with high FUT1 (P = 1.0e-2), high FUT2 (P = 1.5e-2), high FUT3 (P = 4.8e-2), low FUT4 (P = 4.9e-2), low FUT6 (P = 1.9e-2), low FUT7 (P = 1.5e-2), or high FUT8 (P = 1.0e-2) expression (Figure S13). The expression of ALCAM was correlated with survival only with high FUT1 (P = 1.6e-2), low FUT4 (P = 3.4e-2), or low FUT10 (P = 4.2e-2) expression (Figure S14). The expression of LGALS3BP was correlated with survival only with low FUT1 (P = 1.7e-2), high FUT2 (P = 2.9e-2), high FUT3 (P = 3.2e-2), low FUT6 (P = 5.0e-4), high FUT8 (P = 2.8e-2), or low FUT11 (P = 4.8e-2) expression (Figure S15). The expression of THBS1 was correlated with survival only with low ST6GAL1 (P = 4.7e-2) or high ST3GAL3 (P = 4.4e-2) expression (Figure S16). The expression of FUCA2 was correlated with survival with low (P = 1.6e-2) and high (P = 3.1e-2) ST3GAL6 expression (Figure S17). Correlations of expression were weak between fucosyltransferase- or sialyltransferase-coding genes and AGRN (r = 0.12–0.51), ALCAM (r = −0.053–0.18), LGALS3BP (r = 0.21–0.32), THBS1 (r = −0.19–0.037), and FUCA2 (r = −0.016) (Figure S18–S22). Fucosylated and sialylated secretory proteins may be implicated in hepatocellular carcinoma as well as in cholangiocarcinoma.

Fucosylation and sialylation as prognostic markers for metastasis

To evaluate the prognostic significance of FUT or ST genes and secreted protein genes in metastasis, we applied our analysis to the pan-cancer dataset (TCGA-PANCAN; 12,839 samples), which compares 33 tumor types. Excluding samples with recurrent tumor, additional primary tumor, primary blood derived cancer, and additional metastatic cancer, we compared gene expression data from 10,593 primary tumor and 396 metastatic samples. The expression of FUT and ST genes were differentially expressed in metastatic samples. Specifically, FUT7 (P = 4.7e-2), FUT10 (P = 4.9e-27), FUT11 (P = 3.5e-10), ST6GAL1 (P = 7.2e-13), ST3GAL3 (P = 2.5e-69), ST3GAL4 (P = 4.1e-140), and ST3GAL6 (P = 8.0e-137) expression was significantly upregulated in metastatic compared to non-metastatic cancer while FUT1 (P = 3.5e-114), FUT2 (P = 1.5e-169), FUT3 (P = 3.4e-128), FUT4 (P = 4.8e-14), FUT5 (P = 9.9e-42), FUT6 (P = 3.2e-195), FUT9 (P = 2.0e-147), and ST6GAL2 (P = 1.2e-55) expression was significantly downregulated (Figure S23–S24). Across the metastatic samples in the TCGA-PANCAN cohort, FUT and ST genes were significantly correlated with overall survival. High (q2) expression of FUT2 (P = 3.1e-2), FUT6 (P = 2.8e-2), and FUT11 (P = 2.4e-3) correlated with worse survival in comparison to low (q1) expression (Figure 5A–G). In contrast, high (q2) expression of FUT7 (P = 1.5e-2), FUT8 (P = 1.2e-3), ST6GAL1 (P = 4.7e-3), and ST3GAL6 (P =2.4e-3) correlated with better survival in comparison to low (q1) expression.

Figure 5.

Correlation of fucosylation and sialylation related genes with overall survival in metastasis. Kaplan-Meier analysis of metastatic cases from the TCGA-PANCAN dataset stratified by high (q2) or low (q1) expression of genes encoding (A-E) fucosyltransferase, (F-G) sialyltransferase, or (H-I) glycosylated secretory proteins. P-values were determined using a log rank test.

Genes encoding secreted proteins that we identified as fucosylated and sialylated in CCA were differentially expressed in metastasis. Compared to non-metastatic samples, LGALS3BP (P = 3.9e-47), TIMP1 (P = 3.6e-39), HSPG2 (P = 3.8e-20), PSAP (P =3.5e-40), DAG1 (P = 1.8e-37), and FUCA2 (P = 5.4e-5) expression was significantly upregulated in metastatic samples while THBS1 (P = 6.6e-12), ALCAM (P = 5.9e-74), CLU (P = 1.5e-72), ULBP2 (P = 1.1e-27), and SERPINA1 (P = 3.2e-43) expression was significantly downregulated (Figure S25). Within the metastatic samples alone, secreted protein genes significantly correlated with survival. High (q2) expression of TIMP1 (P = 9.4e-3) and SERPINA1 (P < 1.0e-4) correlated with better survival (Figure 5H–I). Fucosylated and sialylated secretory proteins represent potential targets to improve the prognosis of metastatic cancer.

Discussion

Extracellular proteins are often glycosylated prior to localization. We have previously catalogued membrane-associated glycans and glycoproteins from CCA (Park et al., 2020), which revealed dominance of extended oligomannose type N-glycans. The secreted glycoproteome exhibited more processed N-glycans in comparison to that of the membrane, significantly favoring fucosialylated complex type N-glycans over oligomannose type N-glycans. Although both membrane and secreted proteins traverse through the secretory pathway, variations in residence times, export pathway, protein fold, or ER-localized quality control may lead to differences in glycosylation profiles.

In tissues and sera from CCA patients, the total levels of fucosylation were observed to be elevated when compared to healthy controls (Betesh et al., 2017; Indramanee et al., 2012). Prior studies have shown the significance of terminal fucosylation in promoting the progression of CCA. Inhibition of fucosylation by FUT1 suppressed the migration, invasion, and adhesion of CCA cells (Indramanee et al., 2019). Similarly, inhibition of FUT8 expression suppressed the proliferation and migration of CCA (Chen et al., 2022). However, based on a gene expression-based survival analysis, FUT4 expression in CCA was positively correlated with survival. A favorable prognostic effect of FUT4 has also been suggested in head and neck squamous cell carcinoma and colorectal and rectal cancer (Hao et al., 2020). In acute myeloid leukemia, FUT4 was a favorable prognostic factor in patients who received chemotherapy but not in those who received allogenic hematopoietic stem cell transplantation (Dai et al., 2020). While confirmation of this analysis in CCA will require larger cohorts, multiple factors may govern the expression of FUT4, including promoter methylation (Li et al., 2012). Thus, it is unclear whether CCA patients with high FUT4 expression and extended survival in the TCGA-CHOL cohort also exhibited enhanced levels of fucosylation. Furthermore, studies are needed to evaluate the relationship between the expression of FUT4 and secreted proteins in CCA (Yang et al., 2012). As we showed with glycoproteomic analysis of CCA secreted proteins, certain glycosylation sites are heavily fucosylated over others. FUT4 encodes a specific fucosyltransferase that transfers an α1,3/4-linked fucose to the GlcNAc of an N-acetyllactosamine (LacNAc) on glycoproteins forming tumor-associated Lewis epitopes. It will be important to classify which proteins are altered by FUT4 expression in CCA patients to further investigate the significance of the Lewis determinants. Concerning N-glycosylation, given the higher levels of fucosylation in secreted than membrane proteins, differential FUT4 expression may have higher relevance to secreted proteins.

We observed that glycosylated proteins secreted by CCA cells include extracellular matrix components (laminin α3, α5, β1, β2, β3, γ1, γ2, agrin, basement membrane-specific heparan sulfate proteoglycan core protein), matrix-binding proteins (thrombospondin-1, galectin-3-binding protein), hydrolases (tissue-type plasminogen activator, complement C1s subcomponent, cathepsin D, plasma α-_L-fucosidase), protease inhibitors (metalloproteinase inhibitor 1, metalloproteinase inhibitor 2, α1-antitrypsin), growth factors (progranulin), and transporters (prosaposin, clusterin). In particular, THBS1 has been proposed to have dual effects on cancer progression. Domain-specific and cell-specific binding to multiple membrane-associated proteins stimulates signal transduction pathways that support or inhibit metastasis. THBS1 is most commonly recognized as a potent endogenous angiogenesis inhibitor, upon binding to the cell surface receptor CD36. In addition, The carboxy-terminal domain of THBS1 has been shown to mediate cell attachment and contribute to cancer cell motility (Kosfeld and Frazier, 1992; Taraboletti et al., 1987). The C-terminal consists of antiparallel beta sheets connected to an unstructured calcium-binding region and is the region of THBS1 that binds CD47, which leads to immunosuppression (Huang et al., 2017). Among CCA cases, high THBS1 expression was correlated with longer survival only with low FUT10, high ST6GAL2, or high ST3GAL6 expression. Although the effect of N-glycosylation on secreted proteins is not currently well understood, our analysis shows that the combination of secreted protein and glycosyltransferase gene expression may have improved prognostic utility than either secreted protein or glycosyltransferase gene expression alone.

In eukaryotes, approximately one third of proteins encoded by the genome are secreted or integrated into cell membranes. Of note, while we profiled N-linked glycosylation, extensive processing of glycans on secreted and membrane proteins by fucosyltransferases and sialyltransferases occur in both N- and O-linked glycans. Analysis of specific secreted protein glycoforms may be used as markers to detect risk of CCA progression. Additional studies are warranted to identify the mechanisms that regulate the secretion of glycoproteins in metastatic CCA. Taken together, this study contributes to the growing understanding of the types and role of glycans associated with the secretome.

Data availability

The raw data used in this study have been deposited to the MassIVE repository and are available online at ftp://massive.ucsd.edu/MSV000091721.