An N-glycoproteomic site-mapping analysis reveals glycoprotein alterations in esophageal squamous cell carcinoma

The eight pairs of tumor tissues and adjacent normal tissues at least 5 cm from the center of the tumor used for N-glycoproteomic site-mapping and proteomics analysis in this study were obtained from patients diagnosed with ESCC who underwent surgical resection at the Shanxi Cancer Hospital (Taiyuan, China). The five pairs of tissues used in UEA-I lectin affinity chromatography were obtained from the same source as above. There was no preoperative chemotherapy, radiotherapy or other therapies done on these patients. All the recruited patients gave their informed consent before they participated in our study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of the Shanxi Medical University (Approval No. 2017LL037). Details of clinicopathological features of cases are summarized in Additional file 1: Table S1.

After the samples were ground into cell powder in liquid nitrogen, four times the volume of lysis buffer (1% protease inhibitor, 3 μM TSA, 1% TritonX-100, 10 mM dithiothreitol, 50 mM NAM and 2 mM EDTA) was added and the samples were sonicated on ice using an ultrasonic processor. An equal volume of Tris-saturated phenol (pH 8.0) was added. Then, the mixture was further vortexed for 5 min. After centrifugation (4 °C, 10 min, 5,000 g), the upper phenol phase was transferred to a new centrifuge tube. Proteins were precipitated by adding at least four volumes of ammonium sulfate-saturated methanol and incubated at − 20 °C for at least 6 h. After centrifugation at 4 °C for 10 min, the supernatant was discarded. The remaining precipitate was washed with ice-cold methanol, followed by ice-cold acetone for three times. The protein was redissolved in 8 M urea and the protein concentration was determined with BCA kit according to the manufacturer’s instructions. Then the samples were reduced with 5 mM dithiothreitol for 30 min at 56 °C, iodoacetamide was added to make the final concentration 11 mM and samples were incubated at room temperature for 15 min in darkness. Finally, the urea concentration was diluted to less than 2 M. Trypsin was added at 1:50 trypsin-to-protein mass ratio overnight for the first digestion and 1:100 trypsin-to-protein mass ratio for 4 h for the second digestion to make the digestion more sufficient.

The tryptic peptides were fractionated into fractions by high pH reverse-phase HPLC using Agilent 300Extend C18 column (5 μm particles, 4.6 mm ID, 250 mm length). Firstly, the peptides were separated with a gradient of 8–32% acetonitrile (pH = 9.0) for more than 60 min into 60 fractions. Then, the peptides were combined into 18 fractions and dried by vacuum centrifuging. All of the above processes, LC–MS/MS and subsequent bioinformatics analyses were performed in Jingjie PTM Biolabs (Hangzhou in China). The peptides were dissolved in 0.1% formic acid (solvent A) and directly loaded into a reversed-phase analytical column (75 μm i.d, 25 cm length). The gradient consisted of a gradual increase from 5 to 20% of solvent B (0.1% formic acid in 90% acetonitrile) for more than 40 min, 20% to 35% in 12 min and climbing to 80% in 4 min, followed by being held at 80% for the last 4 min. Above all were operated at a constant flow rate of 700 nl/min on EASY-nLC 1000 UPLC system. The peptides were submitted to NSI source and analyzed by tandem mass spectrometry (MS/MS) in Orbitrap Fusion (Thermo). The m/z scan range was 350–1550 for full scan, and Orbitrap was used to detect intact peptides at a resolution of 60,000. Then, peptides were selected and analyzed by MS/MS using NCE setting as 35 and Orbitrap was used to detect the fragments at a resolution of 30,000. Automatic gain control (AGC) was set at 5E4. Fixed first mass was set as 100 m/z.

Maxquant search engine (v1.5.2.8) was used to retrieve the mass spectrometer data. Parameter settings were as follows. The database was SwissProt Human (20,130 sequences), and the reverse decoy database was added to calculate the false positive rate (FDR) caused by random matching. A common contamination database was used to eliminate the influence of contaminated proteins. Trypsin/P was specified enzyme allowing up to 2 missing cleavages. The minimum required peptide length was set to seven amino acids and the maximum modification number of the peptide was set to 5. In first search, the mass tolerance for precursor ions was set as 20 ppm. In main search, the mass tolerance for precursor ions was set as 4.5 ppm. The mass tolerance for secondary fragment ions was set as 0.02 Da. Carbamidomethyl on Cys was specified as fixed modification. The variable modification was defined as oxidation on Met, deamidation 18O (N), deamidation (NQ), and acetylation of the N-terminal of the protein. The quantitative method was set to TMT-10plex. FDR of proteins and PSM identification was adjusted to 1% and minimum score for modified peptides was set > 40.

The annotation of Gene Ontology (GO) of proteins comes from the UniProt-GOA database (http://​www.​ebi.​ac.​uk/​GOA/​). Firstly, the system converted protein ID to UniProt ID, then used UniProt ID to match GO ID, and extracted corresponding information from UniProt-GOA database based on GO ID. If there was no according protein information in UniProt-GOA database, the InterProScan soft based on protein sequence would be used to annotate protein’s GO function. The proteins were then classified according to cellular component, biological process and molecular function. Kyoto Encyclopedia of Genes and Genomes (KEGG) database was used to annotate protein signaling pathway. Firstly, we used KEGG online service tools KAAS to annotate KEGG database description of proteins. Then KEGG mapper, KEGG online service tools was used to map the annotation result on the KEGG pathway database. Wolfpsort, an updated version of PSORT/PSORT II, was applied to predict subcellular localization.

Software MoMo (motif-x algorithm) was applied to analyze the motif characteristics of N-glycosylation sites. Among them, the peptide sequences of 10 amino acids upstream and downstream of all the N-glycosylation site were analyzed. And all the database protein sequences were used as background database parameter. When the number of peptides in a specific sequence was more than 20 and the p value was less than 0.00000 1, it was considered that the characteristic sequence was a motif of modified peptide. The calculation method is as follows: (the number of peptide identified with a certain motif/the number of peptide identified with Ng)/(the number of peptide identified with this motif in the database/the number of theoretical peptide in which Ng occurs in the database).

Sufficient tissue samples were fully ground in liquid nitrogen. The protein lysate (Beyotime, #P0013) and protease inhibitors were added to the samples. The samples were incubated on ice for 1 h and centrifuged at 4 °C for 20 min. The supernatant was total protein solution and protein concentration was measured using the BCA method. 50 µg total-protein was taken as input and 2,000 µg total-protein was mixed with 100 µl biotinylated UEA-I lectin. Next, the mixture was fixed to 1 ml with Phosphate-Buffered Saline (PBS) containing 1 mM CaCl2, MgCl2 and MnCl2. After incubation overnight at 4 °C on rotating shaker, 50 µl agarose streptavidin beads (Vector laboratories, #SA-5010) were added to the mixture and continued to be incubated for 4 h. Then the beads were washed with PBST (PBS mixed with 0.05% tween-20) five times after centrifugation, protein loading was added and the samples were boiled at 100 °C for 5 min, which followed by the western blotting.

All protein samples were separated on 8% SDS-PAGE by electrophoresis and transferred to PVDF membranes. After being blocked with 5% skim milk for 1 h, the membranes were incubated with the primary antibodies at 4 °C overnight, and incubated with the secondary antibody at room temperature for 2 h. Antibodies used were as follows: ITGB1 (1:1,000, Proteintech, #12594-1-AP), CD276 (1:1,000, Proteintech, #66481-1-lg), and GAPDH (1:20,000, Proteintech, #60004-1-lg). IRDye 800CW conjugated goat anti-mouse (or rabbit) IgGs was used for protein detection using the Odyssey infrared imaging system. Finally, signals were digitized by ImageJ software (v1.52i).

The Mfuzz package (version 2.36.0) based on the open-source statistical language R (version 3.6.3) was used to perform soft-clustering analysis. The fuzzy c-means algorithm, one of the most widely used soft-clustering method, provided by Mfuzz package, is more noise robust and a priori pre-filtering of genes/proteins can be avoided, so it avoided the exclusion of biologically relevant genes/proteins from the data analysis [20, 21]. FCM clustering uses Euclidean distance as the distance metric, and demands two main parameters (c = number of clusters, m = fuzzification parameter). Different from k-means clustering, each element has a set of membership coefficients corresponding to the degree of being in a given cluster by FCM clustering.

For the quantitative information of N-glycosites, we removed the N-glycosites containing missing values in the raw data at first, to make sure that each gene could be quantified in each sample. The quantitative ratio of N-glycosites was calculated in each sample pair, and the quantitative ratios were log2 transformed and then z-score was normalized for each pair of sample, making the mean zero and standard deviation one. The normalized-data made sure that N-glycosites with similar temporal patterns were close in Euclidean space [22]. The transformed profiles were then clustered using the Mfuzz package. For this analysis, the optimal values of c and m were derived. Final clustering was done with the parameters c = 15 and m = 1.48. Then we screened out the groups related to the degree of LNM, set the appropriated membership coefficients to make the remaining genes in these groups had a similar trend of change (membership ≥ 0.30).

Figure 1 showed the workflow for the global N-glycoproteomics and proteomics analyses of the ESCC tumor and matched normal tissues. ESCC tissue and paired normal tissue samples from the same patient were used to reflect the level of different individuals. Sixteen protein samples from eight patients were assigned to two assays with TMT labeling and digested with trypsin for further analyses. In the N-glycoproteomics analysis, deglycosylated peptides after PNGase F treatment were actually measured instead. This de-N-glycoproteomics was aimed at profiling glycosylated proteins without focusing on the glycan structure, equaling to the N-glycosylation site mapping. A detailed description of the clinical characteristics of the analyzed samples was presented in Additional file 1: Table S1. To avoid unexpected changes of glycoproteins caused by the amount variation of protein samples, we firstly investigated the global proteomic profile in eight paired tumor and normal tissues using TMT labeling-based quantitative proteomic assays. Principal Component Analysis (PCA) and Pearson correlation analysis showed a clear distinction between the ESCCs and normal esophageal tissues (Fig. 2A, B). Relative standard deviation (RSD) was less than 20% in two groups, representing the ideal repeatability (Fig. 2C). The number of according peptides/proteins in basic statistical analysis of mass spectrometry was shown in Fig. 2D. These results of quality control process indicated that our proteomics analysis system was robust. Altogether, 6,856 proteins were identified from the selected samples, among which 6,113 proteins were quantified. The mean value of the 6,113 protein quantitative ratios was 1.11, indicating that there is no obvious change in protein expression levels. With fold change value > 1.5 and p-value < 0.05 as threshold, totally 596 proteins were up-regulated while 495 proteins were down-regulated (Fig. 3A). All these data were listed in Additional file 1: Table S2. Proteins showing significant up/down-regulation were described in Fig. 3B. In terms of cellular component, the up-regulated proteins were mainly localized in chromosome, nuclear lumen, spliceosomal complex and intracellular organelle lumen, while most of the down-regulated proteins were found in the extracellular region, extracellular matrix and vesicle (Fig. 3C). Accordingly, KEGG-based enrichment analysis showed that the prominent pathways enriched in the up-regulated proteins included spliceosome, DNA replication, RNA transport, transcriptional mis-regulation in cancer, cell cycle, base excision repair and mismatch repair processes. By contrast, protein expression in the cellular pathways including ECM-receptor, focal adhesion, metabolism (e.g., protein digestion and absorption; tyrosine, histidine, phenylalanine, and tryptophan metabolism; drug metabolism-cytochrome P450), interaction and proteoglycans processes was decreased (Fig. 3D). Based on the proteomics data, the following N-glycoproteomic site-mapping data was normalized by corresponding protein expression.

As one of classical post-transcriptional modifications (PTM), glycosylation has been proved to be essential for regulating cellular functions. Although the N-glycoproteomics profile has been revealed in some cancer types, it remains unreported in ESCC. In this study, the quality control results of N-glycoproteomics showed that our optimized methods were highly confident. Most identified peptides were distributed in size of 8–20 amino acids, which was consistent with the common pattern after trypsin treatment (Fig. 4A). Mass error of most spectrograms was within 5 ppm, which indicated that the accuracy of the mass spectrometer was reliable and the qualitative analysis of protein would not be affected by the large mass deviation (Fig. 4B). The results exhibited reliable reproducibility and accuracy in LC–MS/MS by both Principal Component Analysis and Pearson correlation analysis (Fig. 4C, D). In total, N-glycoproteomics data quality control indicated that the analysis system was robust.

To characterize the possible specific sequence, motif analysis was performed to discover the likelihood of amino acid types being over- or under-represented surrounding the N-glycosylation sites. Seven significantly enriched motifs were identified. Motif logos according to the score ranking included N-X-T, N-C-S, R-XXXXXX-N-X-S, N-A-S, R-N-X-S, N-X-S, N-G-X (X represents a random amino acid residue; N, T, A, G, C, S, R represent asparagine, threonine, alanine, glycine, cystine, serine, arginine, respectively) (Fig. 4E). To validate whether there were position-specific amino acids surrounding N-glycosylation sites, the amino acid frequencies in the sequences flanking N-glycosylation sites were assessed using motif model (Fig. 4F). In addition to canonical N-linked glycopeptide sequon of N-X-T/S, we found that cysteine residue at + 1 and + 3, glycine residue at  − 1 and + 1, arginine residue at − 1, − 2 and − 7, and alanine residue at + 1 position were overrepresented. Interestingly, despite N-X-C and N-X-V have been discovered to be atypical N-glycosylation sequons [23], cystine and valine were found to be depleted at + 3 position of N-glycosylation site in our study. The motif score, relative frequency and other detailed information of these motif were demonstrated in Additional file 1: Table S3.

On average, a total of 1,839 N-glycosylation sites in 1,021 glycoproteins were identified in total, of which 1,588 sites in 901 glycoproteins were quantitative. Figure 5A demonstrated the distribution of the number of identified N-glycosylation sites in each glycoprotein. Among all identified glycoproteins, approximately 62.8% (641/1021) contained only one N-glycosylation site, 20.3% (207/1021) possessed two N-glycosylation sites, and 5.5% (56/1,021) contained more than four N-glycosylation sites. Notably, six glycoproteins were found to have more than nine N-glycosylation sites as follows, LDL receptor-related protein 1 (23 sites), laminin subunit alpha-2 (14 sites), fibrillin-1 (13 sites), laminin subunit gamma 1 (12 sites), tenascin C (10 sites) and laminin subunit alpha 5 (10 sites).

To explore the relations and difference between the global N-glycoproteins and proteins, we compared datasets of the quantified N-glycoproteomic site-mapping and the proteomics. It was demonstrated that the quantified proteins were more than glycoproteins for ESCC cases. Moreover, the numbers of quantified proteins had no trend while the numbers of quantified glycoproteins increased with the progression of lymph node metastasis (Fig. 5B). The overlapping part in Fig. 5C represents the protein groups identified in both N-glycoproteomics and proteomics. In our results, a total of 542 overlapped proteins were identified. In other words, only 60.2% (542/901) of the glycosylated proteins identified in N-glycoproteomic site-mapping analysis were also found in the proteomics analysis. In fact, some glycosylated proteins, such as surface receptors and secreted proteins, show a relatively low abundance, which makes it difficult to detect them in the proteomics analysis. This was supported by the results that the intensity of deglycoside proteins only identified in the N-glycoproteomics data was lower than that of the corresponding proteins found in the global proteomics data (Fig. 5D).

A total of 716 differentially expressed N-glycosylation sites (with a cut-off change of 1.5-fold and p-value < 0.05) in 441 proteins were characterized in our N-glycoproteomic site-mapping analysis. We found that 512 N-glycosylation sites in 326 glycoproteins were significantly up-regulated and 204 N-glycosylation sites in 115 glycoproteins were significantly down-regulated (Fig. 5E). It is well known that the glycosylation modification change may be attributed the changes of total protein expression. Thus, we mainly focused on the 542 glycoproteins that had quantitative information in both proteomics and N-glycoproteomics.

We evaluated the N-glycoproteomic ratios of 1,107 N-glycosites in 542 proteins which were normalized by their corresponding protein amount. A significant correlation between the N-glycosite ratios and protein ratios (Pearson r = 0.84) was discovered, demonstrating that the changes of N-glycosylation modification were probably caused by the protein abundance change. A total of 264 differentially expressed N-glycosylation sites in 179 proteins were identified, in which 243 N-glycosylation sites in 162 glycoproteins were significantly up-regulated and 21 N-glycosylation sites in 18 glycoproteins were significantly down-regulated. The differentially expressed glycoproteins and glycosylation sites were listed in Additional file 1: Table S4. Twenty-four glycosylation sites of 20 proteins were consistently up-regulated in all samples and nineteen glycosylation sites of 17 proteins were consistently down-regulated in more than four samples (Fig. 5F). Among them, procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2 (PLOD2), collagen alpha-3 (VI) chain (COL6A3), peptidyl-prolyl cis–trans isomerase (FKBP10) and Tenascin (TNC) had two or more differential glycosylation sites in all samples. The number of differentially expressed N-glycosylation sites that were identified in individuals of both experiments was exhibited in Fig. 5G.

Subcellular localization analysis revealed that the majority of differentially expressed glycoproteins (fold change > 1.5 and p value < 0.05) distributed in the extracellular region (53.6%), plasma membrane (22.4%) and endoplasmic reticulum (10.6%). However, only a few glycoproteins were localized in the cytoplasm (2.2%) and nucleus (3.4%). It was highly consistent with the biological function of glycoproteins, completely different from the protein localization in proteomics (Fig. 6A). Then we performed enrichment analysis of the differentially expressed glycoproteins to identify the significantly Gene Ontology terms and KEGG pathways. Most up-regulated glycoproteins were enriched in cell-substrate-related metabolic processes including collagen metabolic process, receptor metabolic process, protein activation cascade. For the KEGG pathway enrichment, some known glycosylation-affected cancer-associated pathways including ECM-receptor interaction, focal adhesion, phagosome and lysosome were enriched in up-regulated glycoproteins. Interestingly, some amino acid metabolism-related pathways including lysine degradation, tyrosine metabolism, phenylalanine metabolism, valine, leucine and isoleucine degradation were glycosylated and the corresponding glycoproteins were significantly up-regulated in ESCC (Fig. 6B). Three N-glycosites (N63, N209 and N696) were identified in Procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2 (PLOD2) that were reported to catalyze the hydroxylation of lysyl residues in collagen-like peptides. Other up-regulated glycoproteins related to amino acid metabolism included procollagen-lysine, 2-oxoglutarate 5-dioxygenase 1 and 3 (PLOD1 and PLOD3), amine oxidase copper containing 3 (AOC3), interleukin 4 induced 1 (IL4I1), 3-hydroxy-3-methylglutaryl-CoA synthase 1(HMGCS1) and so on. In contrast, these up-regulated pathways in N-glycoproteomic site-mapping were found to be down-regulated in proteomics, indicating these pathways can be actually regulated via glycosylation modification in ESCC. A number of the down-regulated terms were associated with peptidase regulator activity. The KEGG pathway analysis showed ECM-receptor interaction, focal adhesion and PI3K-Akt signaling pathway were negatively enriched in our results (Fig. 6C). Besides that, differential glycosylated proteins in some amino acid metabolism-related pathways were described in Fig. 6D.

To further identify the potential promising biomarkers in ESCC, we further validated the 179 significantly altered glycoproteins between ESCC and paired normal tissues. Firstly, 97.7% of the proteins (i.e., 175 out of 179) have been identified and validated in other proteomics studies of ESCC. Extensive literature searching uncovered evidence that 53 (i.e., 29.6%) of the 179 glycoproteins were directly related to ESCC. Moreover, literature review and glycosites database search hit 60 (33.5%) and 174 (97.2%) glycoproteins, respectively (Additional file 1: Table S5). These data at least provided the evidence that the overlapped glycoproteins probably could serve as ESCC markers.

Previously, we also systematically profiled label-free quantitative proteomes and N-glycoproteome of two ESCC cell lines. The corresponding method and quality control results were shown in the previous study [24]. After normalization by protein quantification, a total of 1,100 N-glycosylation sites in 575 glycoproteins were identified (Additional file 1: Table S6). We integrated significantly altered glycoproteins at ESCC tissue level with the identified glycoproteins at ESCC cellular level. A total of 88 glycoproteins were identified to be overlapped (Additional file 1: Table S5). The identified overlapped proteins may be truly ESCC specific.

Also, we previously used lectin affinity chromatography and LC/MS/MS analysis to elucidate the glycosylation patterns in ESCC cell lines. The lens culinaris agglutinin (LCA; mannose, glucose, core fucose-binding) and ulex europaeus agglutinin 1 (UEA-I; fucose-binding) were applied [24]. The 707 and 956 glycoproteins were identified respectively (Additional file 1: Tables S7 and S8). We performed a combined analysis of 179 significantly altered glycoproteins in this study and lectin enriched glycoproteins in the previous study. A total of 21 and 40 glycoproteins were identified to be overlapped respectively (Additional file 1: Table S5). These data perhaps shed light on the rough estimate of glycan patterns and candidate biomarkers of ESCC for distinguishing cancer patients from healthy individuals.

Next, we used the lectin affinity enrichment followed by western blot to further validate the biomarkers for ESCC. We chose PRNP, PSAP, CD276, LAMP1, ITGB1 in this phase based on the following considerations: (1) In the N-glycoproteomic site-mapping analysis at tissue level, they stood out as glycoproteins with significantly altered expression between ESCC and paired normal tissue (fold change > 1.5 after normalization by overall protein quantification, Fig. 7A); (2) The glycoproteins were identified in both N-glycoproteomics and lectin based mass spectrometry analysis at ESCC cellular level; (3) All the glycoproteins were identified in glycosite database; (4) All of them were predicted as secretory proteins by the program SignalP. The values predict a high likelihood of detecting the proteins remotely in the blood stream and may act as biomarkers in ESCC; (5) To the best of our knowledge, they have not been tested as glycoprotein markers for ESCC. UEA-I lectin enrichment followed by western blot showed significantly increased fucosylated levels of ITGB1 and CD276 in ESCC compared with paired normal tissue, while protein expression levels change to a lesser extent than fucosylation levels (Fig. 7B, C). Gray-scale analysis by ImageJ software also showed that the expression level of normalized fucosylated ITGB1 and CD276 in tumor tissues was significantly higher than those in normal tissues (Fig. 7D). The validation results indicated that elevated fucosylated ITGB1 and CD276 contributed to the occurrence and development of ESCC, which might be potential biomarkers in ESCC.

As the lymph node status is one of the main factors affecting the clinical stage and survival of ESCC patients, we were interested in the dynamic process of glycoprotein expression during ESCC development, especially in lymph node metastasis (LNM). To investigate the types of dynamic change, we performed dynamic clusters analysis on the basis of glycoprotein quantity. We captured a series of glycoprotein clusters whose fold change increased or decreased with the progression of lymph node metastasis.

After further study about the dysregulated glycoprotein clusters in cancer progression, totally fifteen clusters were discovered in our data. Clusters #3, #7 and #9 represented a series of glycoproteins with expression change (Fig. 8A). Clusters #3 and #7 represented a series of glycoproteins whose fold-changes tend to decrease as the disease develops, while #9 was glycoprotein cluster with increasing fold-changes. The enrichment pathways and interaction networks of these glycoproteins were exhibited in Fig. 8B and C. The glycoproteins of Clusters #9 were enriched in HIF-1 signaling pathway (INSR, IGF1R, TFRC), RAP1 signaling pathway (INSR, IGF1R, ITGB3), and lysine degradation pathway (PLOD3 and COLGALT1), indicating that these pathways might play an important role in the LNM process in ESCC. Furthermore, the glycoproteins of Clusters #3 and #7 were enriched in cell adhesion molecules (CAMs), complement and coagulation cascades, and leukocyte trans-endothelial migration. The decreased glycosylation levels of CAMs may promote tumor cells to detach from the primary tumor and facilitate tumor metastasis. For leukocyte migration and complement system, low glycosylation levels may affect their immune surveillance and clearance capabilities of cancer cells.