Analytical strategy

Figure 1 is a schematic representation of our experimental approach. To investigate the proteins that are related to liver cancer metastasis, we followed this workflow to profile the difference of N-glycosylated proteins in the secretome of HCC cell lines with different metastatic potential. To reduce the interference from high abundant serum proteins, the culture medium was not supplemented with serum[10,22,23]. An initial experiment showed that none of the metastatic cell lines suffered from significantly reduced viability after 24 h of culturing in the CM. Thus, a culture time of 24 h and the DMEM CM without serum were selected as the essential conditions for culturing the cells[18]. Figure 1A shows that the procedures for the collection of the secretomes. The CM was centrifuged and precipitated with TCA. Protein digestion was performed using the filter aided sample preparation (FASP) method. After the digestion, N-glycopeptides were captured by hydrazide chemistry and HILIC methods (Figure 1B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Overview of the experimental workflow.

A) The secretome was collected from the conditioned medium. B) N-glycosylated peptides were enriched using hydrazide chemistry and zic-HILIC methods. First, proteins were digested using FASP, and then the N-glycosylated peptides were captured using two methods, followed by de-glycosylation using PNGase F and LC-MS-MS analysis. C) Label-free quantitative analysis.

https://doi.org/10.1371/journal.pone.0081921.g001

Next, the enriched N-glycopeptides were processed by PNGase F to remove the glycans. The reaction was conducted in H₂¹⁶O for the hydrazide method and in H₂¹⁸O for the HILIC method. For the former, the peptide masses showed a 0.9848 Da increase due to the conversion of asparagine to aspartic acid, and for the latter, a 2.9883 Da increment was detected because of the introduction of an ¹⁸O atom during the PNGase F cleavage[10,17,24]. The deglycosylated peptides were detected using LC-ESI-MS/MS. The LC-ESI-MS/MS data were searched against a human proteome database using Proteome Discoverer software. The matched precursor ion area information was further used for label-free quantification (Figure 1C).

Mapping N-glycopeptides in the secretomes

A total of 100 μg FASP digested peptides (corresponding to 200 μg initial proteins) was used for each experiment with ~100 μL hydrazide-modified agarose beads or 2 mg zic-HILIC material used for the enrichment. Two biological replicates were conducted. The details of the protein identifications are described in Table S1. Figure 2A shows the enrichment specificity, which could be estimated based on the percentage of N-glycopeptides among all of the identified peptides, and the enrichment efficiency, which could be determined based on the number of identified unique N-glycosites. Although the hydrazide chemistry method showed higher specificity than zic-HILIC for the identification of glycopeptides (80 % vs. 30 %, as indicated by the trendline in Figure 2A), the former provided a much lower number of identified N-glycosites than the latter (280 vs. 600 on average). Compared with other studies using similar enrichment methods[10,17,21], higher numbers of N-glycosites and glycoproteins were identified here (526 unique N-glycosites and 320 unique glycoproteins from hydrazide chemistry, and 1,032 unique N-glycosites and 536 unique glycoproteins from the zic-HILIC method), of which 344 N-glycosites and 245 unique glycoproteins were identified by both methods (Figure 2B); the other N-glycosites and glycoproteins were obtained from hydrazide or zic-HILIC alone (181 vs. 687 N-glycosites and 75 vs. 291 glycoproteins, as shown in Figure 2B). Therefore, the two enrichment methods were somewhat complementary.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Summary of identified N-glycosites.

A) The number of unique N-glycosites identified and the percentage of N-glycopeptides from all of the identified peptides in each cell line. B) Overlap of N-glycosites between the different enrichment methods. C) Overlap of the N-glycosites and proteins between the different cell lines. D) Number of N-glycosites identified per protein.

https://doi.org/10.1371/journal.pone.0081921.g002

Figure 2C shows the overlap of unique N-glycosites and unique N-glycoproteins between the MHCC97H and HCCLM3 cell lines. Based on the two enrichment methods, 948 and 907 unique N-glycopeptides were detected from MHCC97L and HCCLM3 cell lines, respectively, and the degree of overlap for the N-glycosites (984 vs. 940) and N-glycoproteins (500 vs. 499) that were identified from the two metastatic cell lines was ~60 % and ~66 %, which suggests a relatively high degree of similarity between the two cell lines.

A total of 1,165 unique N-glycopeptides and 1,213 N-glycosites were recognized and mapped to 611 glycoproteins (Table S2). Approximately 50 % of the glycoproteins were identified with one N-glycosite, nearly 34 % with two or three N-glycosites, and approximately 14 % with four to nine N-glycosites. The protein with the most N-glycosites was low-density lipoprotein receptor-related protein 1 (LRP1), with 18 identified N-glycosites (Figure 2D).

Biological categorization of identified glycoproteins

To determine what subset of proteins was enriched in the secretome, we performed a prediction analysis of the protein cellular localization using Gene Ontology (GO). Approximately 65 % of the glycoproteins identified were categorized into the extracellular space and plasma membrane in the GO subcellular localization annotation (Figure 3A) (Table S3). Although 27 % of the glycoproteins have a GO subcellular localization term of cytoplasm, 75 % of these proteins were predicted to contain signal peptides. Sixty-four of these cytoplasm-localized proteins contained signal peptides, as determined by Signal P 4.0, and forty-six of them were predicted to be localized to the exosome by comparing the data against an exosome database[25]. Taken together, 90% of the identified proteins were predicted to be secreted. Moreover, the enrichment scores for the clusters of glycoproteins and secretory proteins were 195.2 and 50.2, respectively, as compared to the human proteome background by DAVID analysis (p-values: 2.52×10^-204 and 2.42×10^-63; Benjamini: 1.26×10^-201 and 3.03×10^-61) (Table S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Bioinformatics analysis of identified N-glycoproteins.

A) Cellular component annotation of identified N-glycoproteins. B) Biological functions of differentially expressed N-glycoproteins.

https://doi.org/10.1371/journal.pone.0081921.g003

The most significantly enriched biological functions are listed in Figure 3B. The top 10 functional categories, according to their p-values, that are enriched in the metastatic cell lines are shown in Figure 3B and Table S5. It can be observed that the top category contains proteins related to cell movement function, which could be involved in cancer cell migration and invasion.

Label-free quantitative analysis of N-glycopeptides

To explore the differentially expressed glycoproteins, label-free quantitative analysis was performed. Peptide quantities were estimated by precursor ions areas, which were shown in Table S1. The differentially expressed glycopeptides between the two cell lines are shown in Table S6, of which 1,096 N-glycopeptides had quantification values. To avoid the possible interference from noise, the glycopeptides were normalized by the total precursor area of the identified peptides[20,21]. The quantitative information of the 683 overlapped N-glycopeptides between the two cell lines was filtered by hierarchical clustering (Figure 4) and segregated into 10 distinct clusters, including up-regulated, down-regulated or unchanged glycopeptides (Table 1). Clusters a) and b) contain the glycopeptides with higher relative abundance levels in the MHCC97L cell line, which includes 65 and 86 glycopeptides from the two enrichment methods. Clusters i) and j) contain the glycopeptides that have higher relative abundance in HCCLM3 cells, which includes 101 and 64 glycopeptides from the two enrichment methods (Table S6). Clusters c) - h) include more than 50% of the identified glycopeptides without any significant changes in abundance between the two metastatic cell lines.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Label-free quantitative analysis.

Hierarchical clustering based on the Euclidean distance of the peak area of label-free quantification of N-glycosites showed a significant difference between the MHCC97L and HCCLM3 cell lines.

https://doi.org/10.1371/journal.pone.0081921.g004

In addition to the N-glycopeptides that were identified in both metastatic cell lines, there were 258 N-glycopeptides that were only observed in the MHCC97L cell line and 217 N-glycopeptides that were only observed in the HCCLM3 cell line. These glycopeptides belong to 385 unique N-glycoproteins. Among them, 84 % were identified with only one N-glycopeptide, while approximately 4 % of them were identified with 3 N-glycosites. Fibronectin 1 (FN1) was identified by 5 and 9 N-glycopeptides from the MHCC97L and HCCLM3 cell lines, respectively, and the quantitative analysis indicated that the relative abundance level of the former was far less than that of the latter. The 4 glycopeptides that were not identified in the MHCC97L cell line may have been of low abundance, or they could have been affected by altered FN1 glycosylation patterns under different metastatic potentials. From a biomarker perspective, the proteins that belong to clusters i) - j) in Table 1 and those that are uniquely expressed in the HCCLM3 cell line would be the most valuable, as these were up-regulated or detected in concordance with the increased HCC metastatic potential. Within these clusters, well-known hepatocellular carcinoma biomarkers such as alpha fetoprotein (AFP) and Dickkopf-related protein 1 (DKK1)[26] were identified, as well as proteins that have been suggested to be related to metastasis in other studies, for example, integrin alpha-3 (ITGA3) and tissue inhibitor of metalloproteinase-1 (TIMP-1)[21]. These glycoproteins are taken as the positive controls for our strategy, and the results indicate that the N-glycosylated secretome of the metastatic cell lines provides a potential source of disease markers.

The network of metastasis-related glycoproteins

To further elucidate the correlation of the identified N-glycoproteins with metastasis and other diseases, we assigned the glycoproteins containing up-regulated N-glycosites in the HCCLM3 cell line to different diseases using the web tool FunDO[27]. As shown in Figure 5A-B, approximately half of the 300 glycoproteins were related to cancer, liver cancer, and neoplasm metastasis (Bonferroni corrected p-values: 7.82×10^-19, 6.22×10^-7 and 2.11×10^-7). To explore the key glycoproteins and substrates that are related to HCC metastasis, a metastatic network was constructed using the differentially expressed glycoproteins (Figure 5C). The significantly regulated liver cancer-related glycoproteins that were identified by FunDo and IPA are listed in Table 2. Their gene ontology (GO) information, including cellular component and biological processes, is also shown in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Network view of the up-regulated N-glycoproteins in HCCLM3 cells.

A) The networks of the top 3 liver-related diseases. B) The number of related genes and the p-value of the top 3 liver-related diseases that are indicated in A). C) Cellular motility network. (The proteins with higher expression in HCCLM3 cells are in red (Ratio > 2), whereas the other proteins that were generated from the IPA database are not colored.).

https://doi.org/10.1371/journal.pone.0081921.g005

Validation of up-regulated glycoproteins

Among these glycoproteins containing up-regulated N-glycosites, FN1 and FAT1 were selected for further validation via western blotting analysis (Figure 6A). The protein alpha-galactosidase A (GLA) was selected as a loading control because it is expressed equally in the cell lines[28]; its ratio was also close to 1 based on the mass spectrometry (MS) data (Table S6) and the western blot results (Figure 6A). To measure the glycosylated status of the two proteins, the same experiment was repeated with the addition of PNGase F processing (Figure 6B). The results showed that for FAT1, an apparent shift to the low molecular weight region was observed. While for FN1, the shift was not significant. This could be caused by the huge molecular weight of FN1 (260 KDa), thus, the releasing of glycans from the protein sequence did not induce a discriminable shift.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Validation of the differential expression of two selected N-glycoproteins.

A) Ten micrograms secretome protein samples were separated on SDS-PAGE gels, transferred to PVDF membranes, and probed with anti-FN1 or FAT1 antibodies. B) The de-glycosylation of the same amount of secreted proteins from MHCC97L and HCCLM3 cells was performed with PNGase F cleavage for 12 h. Proteins were separated on SDS-PAGE and analyzed by western blotting.

https://doi.org/10.1371/journal.pone.0081921.g006

Cell culture and sample preparation

The human metastatic HCC cell lines MHCC97L and HCCLM3 were obtained from the Liver Cancer Institute of Zhongshan Hospital, which is affiliated with Fudan University (Shanghai, China). Their genetic backgrounds are similar because they are filial-cloned sequentially from a MHCC97 parental cell line[18,45]. The metastatic potential of the HCCLM3 cell line is higher than that of the MHCC97L cell line[46]. After orthotopic implantation of HCCLM3 tumor tissue into nude mouse liver for 35 days, widespread loco-regional and distant metastases were found 100% in lungs and abdominal wall, 80% in intra-abdominal cavity, and 70% in diaphragm[45,47]. While the pulmonary metastatic rate of MHCC97L was only 40 % [48]. Immunocytochemical studies demonstrated that the two clones were positive for AFP and the concentration of serum AFP was higher in HCCLM3-inoculated mice than that of MHCC97L[49]. The cell lines with different metastatic potential provides an important model system for the in vivo and in vitro study of HCC metastasis[50].

Approximately 1×10⁷ cells were cultured at 37 °C in 5 % CO₂ in DMEM (Hyclone, USA) supplemented with 10 % fetal bovine serum until reaching 60 ~ 70 % confluence. Cells were washed stringently and gently four times with Dulbecco’s phosphate buffered saline with calcium and magnesium (DPBS) and once with serum-free DMEM (Conditioned Medium, CM). Cells were then incubated in the CM at 37 °C for 24 h. The secretory proteins were collected and centrifuged at 2,500 ×g for 10 min (4 °C) to remove the detached cells and large debris. The resulting supernatant was collected and centrifuged for 1 h at 100,000 ×g (4 °C) to remove smaller debris and vesicles. Formic acid (FA, final concentration of 0.1 %) was immediately added to the final supernatant, which was then stored at -80 °C. The addition of FA lowered the pH (pH < 4) of the culture supernatants, thus reducing the activity of many proteases[11,22].

The proteins in the culture supernatants were extracted by trichloroacetic acid (TCA) precipitation. Thirty milliliters of culture supernatants were used for each precipitation. All of the treatments were performed at 4 °C. TCA was added to the CM solution to a final concentration of 12 %[23]. After mixing, the solution was precipitated for 4 h at 4 °C, followed by centrifugation (10,000 ×g, 10 min, 4 °C). The protein pellet was carefully washed twice with 2 mL tetrahydrofuran (THF) (pre-cooled on ice), and each wash was followed by centrifugation (10,000 ×g, 10 min, 4 °C). The final pellet was re-dissolved in 0.4 mL lysis buffer (4 % SDS, 0.1 M Tris-HCl, pH 7.6, 100 mM DTT) with a sonicator bath (30 min extraction). The concentrations of the extracted proteins were measured by a NanoDrop spectrophotometer (Thermo, USA) at 280 nm with an extinction coefficient of 1.1 absorbance units[51].

Protein digestion by FASP

Detergent was removed from the lysates, and the proteins were digested with trypsin using the FASP protocol[51], which entails the ultrafiltration of spin units with a nominal molecular weight cut-off of 30,000. Briefly, to YM-30 Microcon filter units (Millipore, USA) containing the CM protein concentrates, 200 μL 8 M urea in 0.1 M Tris/HCl, pH 8.5 (UA buffer) was added, and the units were centrifuged at 14,000 ×g at 20 °C for 15 min. This step was repeated twice. Then, 50 μL 0.05 M iodoacetamide in 8 M urea was added to the filters and the samples were incubated in the dark for 20 min. The filters were washed twice with 100 μL 8 M UA buffer followed by three washes with 100 μL 50 mM NH₄HCO₃. Finally, trypsin (Promega, USA) was added in 100 μL 50 mM NH₄HCO₃ to each filter. The protein to enzyme ratio was 100:1. The samples were incubated overnight at 37 °C, and the digested peptides were collected by centrifugation at 14,000 ×g at 20 °C for 15 min.

N-glycopeptide enrichment by hydrazide

N-glycopeptides were captured from tryptic peptide mixtures by hydrazide chemistry following the protocol of Zhang[19] with modifications. Briefly, an aliquot of 100 μg peptides was diluted with coupling buffer (50 % acetonitrile, 0.5 % acetic acid) to a final reaction environment and then oxidized with 20 mM sodium periodate (NaIO₄) at room temperature in the dark for 1 h. The excess NaIO₄ was removed after oxidation by desalting using a Sep-Pak C18 column (Waters, USA). The oxidized peptides in 800 μL coupling buffer (80 % acetonitrile, 0.5 % acetic acid) were then incubated overnight at room temperature with hydrazide beads (100 μL beads for ≤ 100 μg peptides) that were pre-washed twice with 1 mL water. Unbound peptides in the supernatant were removed by centrifugation. The beads were then washed 3 times with 2 mL water, 3 times with 2 mL 1.5 M NaCl and 3 times with 50 mM NH₄HCO₃ to remove the non-glycopeptides. The captured N-glycopeptides were released by incubating the beads with 3 μL PNGase F (500 units/μl, New England Biolabs, USA) in 100 μL 50 mM NH₄HCO₃ buffer (dissolved in ¹⁸O water) at 37 °C for 12 h. The released glycopeptides were collected from the supernatant, and the hydrazide beads were washed twice with 200 μL 80 % ACN and 0.1 % FA. The washing solutions were combined with the supernatant collected from the PNGase F incubation and dried down to a final volume of 20 μL using a SpeedVac concentrator.

N-glycopeptide enrichment by HILIC

The digested peptides were enriched with zic-HILIC media by the procedure reported by Calvano[15] with slight modifications. Briefly, first, the C8 disk was inserted into a 200 μL tip. Approximately 100 μg in-solution digested samples was dissolved in 80 % ACN, 0.5 % FA and incubated overnight at room temperature with 2 mg zic-HILIC media (Merck; particle size 10 μm) that was pre-washed twice with 1 mL coupling buffer (80 % ACN, 0.5 % FA). Then, zic-HILIC media was loaded into the 200 μL tip that was pre-filled with a C8 disk. The zic-HILIC tip was washed 5 times with 100 μL 80 % ACN, 1 % FA, 19 % H₂O, and the bound peptides were eluted 3 times with 80 μL elution buffer (99 % H₂O, 1 % FA). The eluate was dried, de-glycosylated with 3 μL PNGase F (500 units/μL, New England Biolabs) in 50 mM NH₄HCO₃ (dissolved in ¹⁸O water) at 37 °C overnight and dried down to a final volume of 20 μL using a SpeedVac concentrator.

LC MS/MS analysis

A nanoflow HPLC instrument (EASY-nLC 1000 system, Thermo Fisher Scientific, USA) was coupled on-line to a Q-Exactive mass spectrometer (Thermo Fisher Scientific, USA) with a nanoelectrospray ion source (Thermo Fisher Scientific, USA)[10]. Chromatography columns were packed in-house with Ultimate XB-C18 3 μm resin (Welch Materials, USA). The peptide mixtures were loaded onto the C₁₈-reversed phase column (10 cm length, 75 μm inner diameter) with buffer A (99.5 % water and 0.5 % FA) and separated with a 75 min linear gradient of 3 - 100 % buffer B (99.5 % acetonitrile and 0.5 % FA) at a flow rate of 350 nL/min. Including the loading and washing steps, the total time for an LC MS/MS run was approximately 90 min. The electrospray voltage was 2.0 kV. Peptides were analyzed by data-dependent MS/MS acquisition with a dynamic exclusion duration of 18 s. In MS1, the resolution was 70,000, the AGC target was 3e⁶, and the maximum injection time was 20 ms. In MS2, the resolution was 17,500, the AGC target was 1e⁶, and the maximum injection time was 60 ms. The scan range was 300 - 1400 m/z, and the 20 most intensive precursor ions were selected for MS/MS analysis.

Data analysis

The raw data were processed using the Proteome Discoverer 1.3 proteomics platform. The fragmentation spectra were searched against the RefSeq Human database (20120320) using the Mascot search engine (v 2.2.06) with the precursor and fragment mass tolerances set to 10 ppm and 20 mmu, respectively. Two missed cleavage sites were allowed, and the minimum peptide length was 7 amino acids. Variable modifications: oxidation (M), and acetylation (protein N-terminal). Fixed modification: carbamidomethyl (C). Additionally, the deamidation of asparagine together with the incorporation of an ¹⁸O atom was set as a variable modification for the assignment of N-glycosites. Peptide ions were filtered using the cut-off scores of Percolator based on p-values < 0.01. The false discovery rate (FDR) was set to 1 % for peptide identifications. For label-free quantification, precursor ions areas were extracted using Proteome Discoverer 1.3 with a 10 ppm mass precision (the experimental m/z and retention times were recorded for precursor area quantification). Ratios for each peptide were normalized by the total identified peptides, and the ratios were used for N-glycopeptide-level quantification.

Bioinformatics and statistical analysis

To evaluate the trend of protein abundance changes, hierarchical clustering was performed and a distance tree was generated using Multi Experiment Viewer (MeV)[52]. To visualize the significantly regulated N-glycosites, Statistical Product and Service Solutions (SPSS) software was used. To study gene-disease relationships, the differentially regulated genes in metastatic cell lines were assigned to different diseases based on Disease Ontology and peer-reviewed evidence from GeneRIF using the web tool FunDO (http://django.nubic.northwestern.edu/fundo)[27]. Then, Cytoscape v.2.8.2 was used to visualize gene-disease interaction networks[53]. To explore the biological functions, subcellular localization and pathways and networks of the N-glycoproteins involved, Gene Ontology (GO) annotation[54], DAVID Bioinformatics Resources[29] and Ingenuity Pathways Analysis (IPA, Ingenuity Systems, Mountain View) were employed, respectively.

Western-blotting analysis

The secretory proteins of metastatic HCC cell lines were resolved on 10 % SDS-PAGE gels, followed by transferring onto PVDF membranes (Millipore, USA). After incubation in blocking buffer (0.5 % Tween-20 in TBS, 5 % BSA) for 1 h at room temperature, membranes were blotted using antibodies against the targeted proteins for 1 h at room temperature. Membranes were then washed with TBST (TBS with 0.5% Tween-20) and incubated in 1:5,000 diluted HRP-conjugated IgG for 1 h at room temperature. After washing three times with TBST, the bands on the membrane were visualized using an ECL plus detection system[18].