Profiling the pancreatic cancer secretome with metabolic glycoengineering

TL;DR

This study uses a new method to better identify proteins secreted by pancreatic cancer cells, which could help find new biomarkers for detection and monitoring.

Contribution

A metabolic glycoengineering strategy is introduced to enhance secretome profiling and biomarker discovery in pancreatic cancer.

Findings

MGE-LC–MS/MS identified unique secreted and EV-associated proteins in pancreatic cancer cells.

ManNAc analogs enhance extracellular vesicle production, improving secretome profiling.

The method successfully enriched PC-derived glycoproteins from mouse plasma for biomarker discovery.

Abstract

Profiling the secretome for biomarkers offers an attractive, minimally invasive strategy to detect and monitor cancer. Several challenges, however, must be overcome, including the broad dynamic range of biomolecules in the secretome and the requirement for selective detection of tumor-associated markers. Here, we employed a metabolic glycoengineering (MGE) strategy, using 1,3,4-O-Bu3ManNAz, an azido-tagged, bio-orthogonal metabolic precursor of sialic acid, to label the glycome of pancreatic near-normal and cancer cells to improve conventional LC–MS/MS proteomics–based biomarker discovery. By using this “MGE-LC–MS/MS” approach that incorporates MGE enrichment into conventional LC–MS/MS proteomics, we identified several unique proteins from the secretomes of cancer cells evaluated in vitro. In addition to proteins known to be secreted, we identified several putatively intracellular, non–N-glycosylated proteins, such as β-glucocerebrosidase and paladin, linked to pancreatic cancer (PC) as well as proteins associated with extracellular vesicles (EVs) in PC, such as dCTP pyrophosphatase 1. The identification of EV-associated proteins was consistent with our discovery that ManNAc analogs used in the MGE-LC–MS/MS workflow enhance EV production, creating a more complete secretome profile of PC cells. Pointing toward clinical relevance, we used MGE-LC–MS/MS to enrich PC-derived glycoproteins from plasma harvested from mice bearing xenografted human pancreatic tumors, unambiguously demonstrating that this approach can interrogate the secretomes of cancer cells for biomarker discovery. Finally, we discovered that MGE dramatically improved the production of EVs, which both aids in biomarker discovery (this study) and holds potential to facilitate biomanufacturing of these nascent drugs.

Figures

Full text

A lack of biomarkers for early stage pancreatic cancer (PC) contributes to a poor 5-year survival rate of less than 11% (1, 2). The only Food and Drug Administration–approved serum biomarker for this disease is the carbohydrate antigen 19-9, which has limited specificity and sensitivity (3, 4, 5, 6, 7). Motivated by compelling links between aberrant glycosylation and cancer (8, 9, 10, 11), we reasoned that biomolecular engineering strategies that manipulate cellular glycans could facilitate the discovery of biomarkers with increased specificity and sensitivity. In particular, we exploited metabolic glycoengineering (MGE), a chemical biology technique that takes advantage of the relaxed substrate tolerance of certain glycosylation pathway enzymes to incorporate non-natural monosaccharides into cellular glycomes (12, 13, 14, 15, 16, 17, 18), facilitating the discovery of cancer biomarkers.

MGE has been used to image primary tumors and selectively target drugs to tumors (19, 20, 21, 22, 23, 24, 25, 26). MGE has also been applied to secretome analyses in some cell types (e.g., neurons (27) and prostate stromal cells (28)), but, to our knowledge, not for secreted biomarker discovery in PC. Another advantage of analyzing the secretome is the presence of extracellular vesicles (EVs), which contain nuclear, cytosolic, and other proteins (29, 30, 31, 32, 33). The presence of cytosolic and nuclear proteins in the serum through EVs that traffic those proteins into the extracellular milieu (34, 35, 36) adds another dimension and diversity to biomarker discovery not available through cell surface or cell lysate analyses, the usual endpoints analyzed in MGE-based diagnostic and therapeutic strategies (17, 37, 38).

A general challenge in pursuing diagnostic biomarkers is that small, early stage cancers release low levels of biomolecules, which limits detection even with highly sensitive techniques, such as LC–MS/MS (39, 40, 41). Here, we tackle this limitation by using MGE to both install bio-orthogonal azido groups into glycoproteins to increase the capture of secreted azido-labeled glycoconjugates as well as augment the production of EVs (Fig. 1). Our approach involves treating PC cells or tumor-bearing animals with the high-flux azido-derivatized N-acetylmannosamine analog, 1,3,4-O-Bu_3_ManNAz (42, 43, 44, 45, 46), which is converted to the corresponding azido-tagged sialic acid “Sia5Az” and biosynthetically incorporated into sialoglycoproteins in place of natural sialic acid. The azido-labeled sialoglycoproteins can be biotinylated or captured on appropriately functionalized resins via bio-orthogonal azide–alkyne cycloaddition conjugation “click” reactions (18, 47, 48, 49). When biotinylated, the azido-tagged proteins can then be captured through streptavidin–biotin affinity and identified using LC–MS/MS. In the past, we used this approach to identify 55 azido-tagged glycoproteins from SW1990 human PC cell lysates (50), and other investigators (e.g., Haun et al. (37)) identified 954 cell surface proteins from PC cells. In the current study, we extend these efforts to PC secretome analysis.Figure 1Schematic overview of MGE-enriched biomarker detection. (A) Normal and cancer cells secrete proteins that can be identified by mass spectrometry analysis; one of these proteins, which illustrates a putative cancer biomarker, is circled. The premise of this study is that low-abundance proteins (such as this one) that could serve as cancer biomarkers often are difficult to detect against a background of proteins secreted from normal cells. (B) Our strategy to overcome this pitfall is to metabolically label the secretome using 1,3,4-O-Bu_3_ManNAz to selectively install azide groups into the glycans of putative cancer biomarkers. These azide-labeled proteins can be captured using alkyne-conjugated resin (or first be biotinylated with alkyne-conjugated biotin and then captured with streptavidin-conjugated resin), whereas nonlabeled material is removed from subsequent analysis. By removing nonlabeled proteins through enrichment of MGE-labeled species that we hypothesize over-represent cancer-derived proteins, this study demonstrates the feasibility of using an MGE-based strategy for the identification of putative pancreatic cancer biomarkers. MGE, metabolic glycoengineering.

In the present study, we deployed MGE to probe the secretome of PC cells to develop an improved biomarker detection strategy for this disease. For context, carbohydrate antigen 19-9 (the only Food and Drug Administration–approved PC biomarker) has limited cancer specificity because it can be elevated in non-neoplastic conditions, such as acute and chronic pancreatitis, hepatitis, and biliary obstruction (1, 4, 5, 51, 52, 53). In addition to improving specificity, we also sought to increase sensitivity of secretome analysis by using MGE to capture low-abundance azido-labeled proteins and EVs that are difficult to detect using conventional LC–MS/MS (Fig. 1A). Coupling MGE with LC–MS/MS (Fig. 1B) enabled us to identify 28 proteins exclusively present in the secretome of the Capan-2 PC cell line and 69 proteins from patient-derived PC cells (P198 and JD13D). Building on these in vitro results, we captured low-abundance human proteins from the plasma of mice transplanted with human pancreatic tumors. Among the proteins we identified were several that were already associated with cancer, which validates our strategy for biomarker discovery. Significantly, we also identified proteins that were not previously linked to PC, demonstrating the potential of MGE-based secretome analysis for the discovery of novel biomarkers.

Results and discussion

MGE labels secreted proteins in pancreatic cells

Mammalian cells metabolize azido-derivatized N-acetylmannosamines (54), including our “high-flux” butanoyl-modified analog 1,3,4-O-Bu_3_ManNAz (46, 50, 55), to the corresponding non-natural sialic acids, which are incorporated into numerous cell-surface sialoglycans (Fig. 1). By contrast, relatively little is known about azido-analog incorporation into secreted glycoproteins; therefore, we began this study by evaluating the secretome of a commercially available near-normal pancreatic cell line (human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing [hTERT HPNE] cell line) in comparison to a cancerous pancreatic cell line (Capan-2) to determine the feasibility of this strategy to identify biomarker candidates. Specifically, we supplemented the growth media of each cell line with 1,3,4-O-Bu_3_ManNAz for up to 96 h, harvested the cell culture supernatants, and probed for azido-labeled secreted proteins using far-Western blots (Fig. 2). These experiments confirmed that azido-modified monosaccharide analogs were successfully incorporated into secreted proteins, and, from a practical standpoint, established kinetic parameters for subsequent experiments in patient-derived cells.Figure 2Coomassie staining and far-Western blot analysis of 1,3,4-O-Bu_3_ManNAz-enriched secreted proteins in hTERT HPNE and Capan-2 cells. The cells were either untreated controls (−) or treated with 200 μM (+) 1,3,4-O-Bu_3_ManNAz for (A) 8, (B) 24, (C) 48, or (D) 96 h. The secreted proteins were precipitated from cell culture supernatant and then run in parallel on two gels under identical conditions. One gel was stained with Coomassie Brilliant Blue to detect all proteins present in the secretome. The other sample was subject to far-Western analysis by biotinylating the azido-tagged glycoproteins using click chemistry conjugation with alkyne–PEG–biotin, after which the samples were run on SDS-PAGE, electroblotted onto a nitrocellulose membrane, and probed for biotinylated proteins with streptavidin-conjugated horseradish peroxidase. hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line.

Secretome labeling at an early time point (8 h, Fig. 2A) was consistent with our previous finding that metabolic incorporation of non-natural ManNAc analogs into sialoglycoproteins was measurable as soon as 2 to 4 h after the start of media supplementation, followed by a sustained increase in signal over the next 2 to 3 days (56, 57). At the early time point of 8 h, azido labels were disproportionately incorporated into high–molecular weight proteins with increasing labeling of low–molecular weight proteins at later time points (e.g., 24 h, Fig. 2). One explanation for this labeling pattern is that proteolytic degradation reduced the size of secreted proteins over time or media depletion of 1,3,4-O-Bu_3_ManNAz. Overall, the reduced labeling observed at longer time points (e.g., 96 h compared with 8–24 h) was also consistent with depletion of the azido-modified analog from the culture medium between 24 and 72 h (58). Furthermore, the faster loss of azido-labeled proteins from secretome samples from the Capan-2 cells compared with the near-normal cell line is consistent with secretion of neuraminidase by cancer cells as reported by others (59, 60) and observed in this study, where neuraminidase 1 (sialidase-1) was present at higher levels for the P198 and JD13D cells compared with the hTERT HPNE cell line (Table S1). We did not attempt to remedy the time-dependent loss of signal by periodic supplementation of the culture medium with 1,3,4-O-Bu_3_ManNAz to support indefinite expression of non-natural sialoglycoconjugates (61) or include neuraminidase inhibitors. Instead, we supplemented the medium only once to more realistically reflect conditions for biomarker discovery in clinical settings where one-time dosing is most feasible. Overall, these experiments established two key points: first, they showed that MGE successfully labeled secreted proteins, and second, they optimized conditions for subsequent experiments. Specifically, they showed that unlike cell surface protein labeling, where the highest signal typically is observed from 48 to 72 h after media supplementation, labeling of secreted proteins is optimal at shorter periods (e.g., 8–24 h).

Metabolic azido labeling of N-glycans is cell line specific

The fundamental premise of our MGE-based biomarker discovery strategy is that stronger azido labeling occurs in cancer cells compared with their normal counterparts because hypersialylation is a general characteristic of human cancers (62, 63, 64, 65). At first, it was unexpected that the secretome of the near-normal hTERT HPNE cells was more strongly labeled than the Capan-2 secretome (Fig. 2). A possible explanation is that, as mentioned above, enhanced proteolytic degradation or desialylation occurs because of increased levels of proteases and neuraminidases in the secretomes of the cancer cells. Another explanation is that the metabolic incorporation of azido-tagged sialic acids in the 1,3,4-O-Bu_3_ManNAz-treated Capan-2 cells was lower than in the near-normal pancreatic cells despite expectations to the contrary. Indeed, we found that some types of breast cancer cells unexpectedly incorporated ManNAc analogs into sialoglycans at lower rates than near-normal cells (66, 67). All in all, the noticeably lower metabolic labeling of the Capan-2 compared with the hTERT HPNE secretome (Fig. 2) has precedent in other types of cancer.

The elevated signal from the near-normal cells indicated that global measurements of MGE-labeled secretome proteins could have minimal diagnostic or prognostic value. By contrast, a comparison of the far-Western blots with the corresponding Coomassie-stained gels (Fig. 2) showed that different size fractions of the secreted proteins were selectively labeled in the two cell lines. For example, the intense high–molecular weight band(s) observed in the far-Western blots at early time points did not correspond to commensurate staining in the Coomassie gels, indicating that selective azido labeling of high–molecular weight proteins occurred during early time points. Conversely, pronounced bands in the Coomassie-stained gels at lower molecular weights were virtually absent in the far-Western blots; this selective lack of labeling of low–molecular weight proteins was particularly pronounced for the Capan-2 PC cell samples.

Azido-modified ManNAc analogs primarily label N-linked glycans

The incorporation of ManNAc analogs into N-linked glycans is “well known,” and there are also reports that non-natural sialosides can be incorporated into mucin-type, O-linked glycans (68, 69). Accordingly, we next evaluated whether azido labeling of glycoproteins in 1,3,4-O-Bu_3_ManNAz-treated Capan-2 and hTERT HPNE cells was primarily N- or O-linked by treating cell lysate samples with PNGase F, which cleaves N-glycans, and Protein Deglycosylation Mixture II, which cleaves both N- and O-glycans. As seen in far-Western blots for Capan-2 (Fig. 3A) and hTERT HPNE cells (Fig. 3B), signal was virtually eliminated when the cell lysates were treated with PNGase F, demonstrating that the azido-labeled monosaccharides were primarily, if not entirely, incorporated into N-glycans. To confirm this result, Protein Deglycosylation Mixture II almost entirely ablated azido labeling, consistent with this product’s ability to remove N-glycans from proteins. Together, these experiments established that azido analog incorporation in 1,3,4-O-Bu_3_ManNAz-treated Capan-2 and hTERT HPNE cells is primarily into N-linked glycans. These results also rule out significant nonenzymatic incorporation of azido-modified monosaccharides attributed to conjugation of the peracetylated ManNAz (Ac_4_ManNAz) to thiols (e.g., cysteine) found in proteins (70, 71). An implication of this finding is that, similar to 1,3-di-O-acetyl-N-azidoacetylgalactosamine and 1,6-di-O-propionyl-N-azidoacetylgalactosamine (70, 71) 1,3,4-O-Bu_3_ManNAz minimizes nonenzymatic protein conjugation associated with peracetylated ManNAc analogs, making it a superior labeling agent for MGE applications.Figure 3Far-Western blot analysis of the 1,3,4-O-Bu_3_ManNAz-enriched proteome in Capan-2 and hTERT-HPNE cells. (A) Capan-2 or (B) hTERT-HPNE samples were either untreated control cells (−) or cells incubated with 200 μM 1,3,4-O-Bu_3_ManNAz (+) for 8 to 96 h. Secreted proteins were precipitated from FBS-free cell culture supernatant and divided into aliquots that were subject to no glycosidase, PNGase F, or Protein Deglycosylation Mix II. Each sample was then separated by PAGE under identical electrophoretic conditions. The gels were first stained with SYPRO Protein Gel Stain to confirm equivalent protein loading and then subjected to far-Western analysis by biotinylating the azido-tagged glycoproteins using click chemistry conjugation with DBCO–PEG4–biotin, running the proteins on an SDS-PAGE gel, electroblotting them onto a nitrocellulose membrane, and probing for biotinylated proteins with streptavidin-conjugated horseradish peroxidase. FBS, fetal bovine serum; hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line.

Mass spectrometry identification of azido-labeled proteins

We next used LC–MS/MS to identify azido-labeled proteins from the 1,3,4-O-Bu_3_ManNAz-treated Capan-2 and hTERT HPNE cells that constitute biomarker candidates. In the first set of these “MGE-LC–MS/MS” experiments, we adapted methods we previously developed to capture azido-labeled N-glycosylated peptides from cell lysates (50) and then selectively released N-glycan–modified peptides with PNGase. As our far-Western gels demonstrated, PNGase treatment enriches azido-bearing glycopeptides found in the secretome (Fig. 3). Importantly, this workflow avoids releasing proteins nonspecifically or nonenzymatically labeled by azido-modified analogs (70, 71) and is consistent with how, as discussed above, 1,3,4-O-Bu_3_ManNAz largely avoids such nonspecific labeling. LC–MS/MS analysis of the resulting azido-labeled, N-glycan–modified peptides identified 47 proteins that were unique to the hTERT HPNE secretome, 28 unique to the Capan-2 secretome, and 18 common to both cell lines (Fig. 4A). The identification of a larger number of labeled proteins (i.e., 47) from the hTERT HPNE secretome compared with the Capan-2 secretome (i.e., 28) was consistent with the far-Western blots that showed stronger labeling in the hTERT HPNE cells (Fig. 2). Overall, several previously reported cancer-associated proteins, including the carcinoembryonic antigen–related cell adhesion molecule 5 (72, 73) and mesothelin (74, 75) were identified from the Capan-2 samples, which highlights the ability of MGE-LC–MS/MS (Fig. 1B) to identify MGE-labeled, cancer-associated proteins present in the secretome.Figure 4Secretome proteins identified by MGE-LC–MS/MS. (A) A Venn diagram representation of proteins identified from the secretomes of 1,3,4-O-Bu_3_ManNAz-treated hTERT HPNE and Capan-2 cell lines. (B) Pathway analysis in Database for Annotation, Visualization and Integrated Discovery (DAVID) bioinformatics resources shows the number of proteins associated with each of the listed pathways. Potential glycosites are noted in Table S2. hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line; MGE, metabolic glycoengineering.

Of note, two of the 75 total identified proteins—suprabasin (SBSN) and the family with sequence similarity 3 protein family member C (FAM3C)—have no known N-glycosylation sequons but nevertheless were enriched through MGE-LC–MS/MS (Table S2). First, SBSN is an O-glycosylated protein whose high expression is correlated with poor prognosis across multiple cancers (76, 77, 78). This protein is also aberrantly elevated in the bone marrow and is a candidate biomarker of advanced disease state in myelodysplastic syndromes (79, 80). One possible explanation for the isolation of SBSN during MGE-LC–MS/MS is through azido labeling of its O-glycan(s); however, this is unlikely to be the sole factor given our N-glycan–based enrichment strategy. We posit that this protein may have been more captured and identified because of its abundant presence in PC cell–derived EVs (81, 82), a possibility we explore in more detail later in this report.

The second protein identified from MGE-LC–MS/MS enrichment, despite not being N-glycosylated, FAM3C, was found exclusively in the Capan-2 cell line, consistent with its links to tumor progression across multiple cancers, including gastric, pancreatic, and breast cancer (83, 84, 85). Similar to SBSN, FAM3C can be O-glycosylated (86), making metabolic glycan labeling a possible explanation for its identification through MGE-LC–MS/MS. However, an equally plausible explanation is that because FAM3C has a GG-type lectin motif (87) (www.proteinatlas.org/ENSG00000196937-FAM00000196933C), its carbohydrate-binding ability allows it to piggyback on other glycoproteins enriched by MGE-LC–MS/MS. Then upon PNGase F treatment, this protein is released along with legitimately captured, azido-modified proteins. Another explanation is that—similar to the speculative capture of SBSN via EV azido labeling—FAM3C was also enriched by this mechanism.

Metabolic pathway analysis confirms cell line–specific azido labeling of proteins

In addition to identifying biomarker candidates, the MGE-LC–MS/MS data provided mechanistic insights through pathway analysis of the azido-labeled proteins from the Capan-2 and hTERT HPNE secretomes. In particular, the Database for Annotation, Visualization and Integrated Discovery (DAVID), a publicly available high-throughput functional annotation tool, was coupled with Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis to identify the key metabolic signaling pathways (88, 89). This combined strategy identified a total of 21 metabolic pathways, biological processes, or disease conditions linked to the azido-labeled proteins present in the secretomes of the hTERT HPNE and Capan-2 cell lines (Fig. 4B). Seven of these pathways were represented by proteins from both cell lines, six were represented by proteins only identified in the hTERT HPNE line, and eight were represented by proteins only identified in the Capan-2 line.

We briefly highlight two of these pathways to illustrate how the azido-labeled secretome can provide insights into cancer. First, the “complement and coagulation cascade” pathway unique to Capan-2 cells was represented by three proteins: complement factor 1; plasminogen activator, tissue type; and plasminogen activator and urokinase. Recent publications show that this pathway, when activated in the tumor microenvironment, is cancer promoting in many types of cancer (90, 91) and plays specific roles in PC (92); MGE-LC–MS/MS illustrates how this oncogenic pathway can be illuminated in cancer cells. By contrast to the complement and coagulation cascade, the lysosomal pathway was populated by proteins in both cell lines but nonetheless was preferentially represented in Capan-2 cells, where 10 MGE-identified proteins were identified in contrast with only five in hTERT HPNE cells (Fig. 4B). The lysosomal pathway–associated proteins unique to the Capan-2 secretome were β-glucocerebrosidase (GBA), cathepsin H, cathepsin C, legumain, lipase A, and tripeptidyl peptidase 1 (TPP1). Mechanistically, a functional lysosome mediates autophagy and micropinocytosis, critical processes in the degradation of misfolded proteins and damaged organelles for rapidly proliferating cancer cells (93, 94). Interestingly, an ManNAc supplementation–based approach also identified protein folding chaperones in breast cancer cells (67), suggesting that this biochemical quality control system—which relies on glycans to identify misfolded proteins—might broadly depend on sialylation to modulate its activity in cancer.

GBA exemplifies cell line–specific, azido-labeled biomarker candidates

One of the lysosomal pathway proteins mentioned above that was selectively detected through MGE labeling in Capan-2 cells was GBA (Fig. 4A), an enzyme that converts glucosylceramide to ceramide and glucose (95). This protein attracted our attention because it is linked to other types of cancer (96) and its transcript levels are elevated in PC patients (97); GBA has not yet been associated with this disease at the protein level, however. In the current work, Western blot analysis showed that GBA was present in the secretomes (Fig. 5A) and cell lysates (Fig. 5B) of both the near-normal and cancerous pancreatic lines; these results indicated that GBA itself would have minimal diagnostic value using conventional proteomics methods. Far-Western analysis of GBA, by contrast, showed azido labeling in 1,3,4-O-Bu_3_ManNAz-treated Capan-2 cells (Fig. 5C) but not in hTERT HPNE cells. Accordingly, GBA represents an enticing MGE-based biomarker candidate because, despite being made in both normal and cancer cells, it was uniquely azido labeled, enriched, and identified in the cancer line.Figure 5Selective MGE labeling of GBA secreted from Capan-2 cancer cells. Western blots of secreted proteins (A) and cell lysate (B) showing levels of secreted and intracellular GBA, respectively, in cells treated with the azido-analog 1,3,4-O-Bu_3_ManNAz and the control analog 1,3,4-O-Bu_3_ManNAc. These results showed that GBA was both expressed and secreted by each cell line, albeit at higher levels in the cancer Capan-2 line. (C) Western blot analysis for secreted GBA before and after streptavidin–biotin enrichment. Secreted proteins from control (0 μM) or 1,3,4-O-Bu_3_ManNAz-treated (200 μM) hTERT-HPNE and Capan-2 cells. Input represents secreted protein before enrichment; unbound represents eluted proteins (i.e., non–azido-labeled proteins) from streptavidin–biotin enrichment; and bound represents captured proteins (i.e., azido-labeled proteins) from the streptavidin–biotin enrichment step; as indicated in the outlined box, azido-labeled secretome GBA was only present in the Capan-2 line. GBA, β-glucocerebrosidase; hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line; MGE, metabolic glycoengineering.

Overall, although the exact mechanism for the cell line– and protein-specific incorporation of azido-labeled sialic acids remains to be delineated, the pilot experiments described above (Figure 2, Figure 3, Figure 4, Figure 5) verified that metabolic labeling with 1,3,4-O-Bu_3_ManNAz combined with the bio-orthogonal capture of secreted proteins was a feasible method to selectively isolate and identify cancer-associated proteins for biomarker discovery. These results provided a foundation to extend this strategy from the hTERT HPNE and Capan-2 lines to profile the secretomes of patient-derived primary cancer cells, as described next.

Detection and DAVID analysis of secreted proteins from patient-derived PC cells using conventional LC–MS/MS proteomics

The MGE-LC–MS/MS results outlined above provided a foundation for integrating MGE into conventional LC–MS/MS workflows to improve the detection and identification of cancer-specific proteins. Accordingly, we next compared the two approaches in patient-derived PC cells. First, to provide a baseline for comparison with MGE-enriched samples, we analyzed the secretomes of P198 and JD13D patient-derived PC cells using conventional tandem mass tag quantitative LC–MS/MS proteomics. In these experiments, the hTERT HPNE line was used as a control to provide direct comparison with the less sensitive proteomics method used in the pilot studies described above for the Capan-2 PC cell line. In these next experiments, we used a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific) to maximize the number of proteins identified. To briefly recap how increasingly sensitive instrumentation has facilitated proteomics, more than a decade ago, representative experiments identified ∼250 to 600 (29) to ∼900 (30) secreted proteins from cultured cancer cells; recently, ∼2000 to 3000 proteins can be routinely identified (33). Our current experiments reflected this increased sensitivity, with a total of 2163 different proteins identified in aggregate from the three pancreatic cell lines tested. Of this full set of proteins, the large majority (2127 of 2163) were identified in the secretomes of all three lines (Fig. 6). Of relevance for biomarker discovery, there were 11 proteins found in secretomes of both the P198 and JD13D cells that were not present in the hTERT HPNE samples; this set of proteins constitutes PC biomarker candidates.Figure 6Venn diagram of conventional LC–MS/MS analysis. The numbers of secretome proteins from patient-derived P198 and JD13D cancer cell lines and the near-normal hTERT HPNE line are shown (the full names and the complete list of proteins are given in Table S1). Potential glycosites are noted in Table S3. hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line.

The DAVID bioinformatics tool was coupled with Online Mendelian Inheritance in Man disease analysis (88, 89) to analyze the 11 cancer-associated proteins identified in the patient-derived cell lines (Fig. 6). These analyses did not associate any known genetic mutations in these 11 proteins with PC. The implications of this finding are threefold. First, these proteins may comprise undiscovered biomarker candidates not yet in the Online Mendelian Inheritance in Man database. Alternatively, these proteins may contribute to cancer through epigenetic mechanisms that result in neo- or increased expression in cancer. Finally, post-translational processing (e.g., changes in glycosylation) rather than classical genetic (e.g., point mutations resulting in direct functional changes) or epigenetic mechanisms may be involved; this third possibility is most relevant to the MGE-based biomarker discovery strategy described in this study.

Unlike the 11 cancer-specific proteins exclusively found in the secretomes of the P198 and JD13D cell lines, several of the 2127 proteins identified in the secretomes of all three cell lines have been linked to cancer. Examples include genes encoding serine/threonine kinase 1 (98), epidermal growth factor receptor (99, 100), epithelial cell adhesion molecule (101), cadherin 1 (102), catenin beta 1 (103), and palladin (PALLD) (104). To discuss one of these examples briefly, PALLD, a cytoskeletal protein involved in the control of cell shape and motility, is associated with PC-associated fibroblasts and pancreatic adenocarcinoma cells; specifically, overexpression and mutations increase risk for familial and sporadic PCs (105). For example, a cancer-associated mutation in PALLD changes a proline to a serine, increasing the invasiveness and metastasis of PC cells (104). Importantly, the identification of PALLD (and the other examples listed above: serine/threonine kinase 1, epidermal growth factor receptor, epithelial cell adhesion molecule, cadherin 1, and catenin beta 1) in the secretomes of the near-normal hTERT HPNE cells as well as the P198 and JD13D cancer cells illustrates how a lack of cancer selectivity is a fundamental limitation of cancer biomarker discovery using conventional LC–MS/MS.

Coupling MGE with LC–MS/MS increases cancer-selective biomarker detection in patient-derived PC cell lines

Moving forward, once a foundation describing the secretomes of the hTERT HPNE, P198, and JD13D cell lines was established using conventional LC–MS/MS (Fig. 6), we incorporated MGE into the workflow to profile the secretomes of these three lines after 1,3,4-O-Bu_3_ManNAz treatment and enrichment using bio-orthogonal capture methods (Fig. 1).

Because the two patient-derived cell lines (P198 and JD13D) evaluated by conventional LC–MS/MS (Fig. 6) had not previously been tested in MGE experiments, we evaluated the ability of 1,3,4-O-Bu_3_ManNAz to install azido-tagged glycans onto the cell surface before conducting MGE-LC–MS/MS analysis and successfully showed incorporation (Fig. S1). Subsequent MGE-LC–MS/MS analyses identified a total of 2146 proteins present in total, which was slightly fewer (99.2%) than the 2163 identified by conventional LC–MS/MS across all three cell lines. Of these 2146 proteins, 2017 were common to all three lines (Fig. 7). The larger number of proteins selectively identified in only one or two of the lines indicated that the MGE-based protocol provided increased power to discriminate cell lines from each other because there was a collective total of 129 cell line–specific proteins identified compared with only 37 using conventional LC–MS/MS. Importantly, there were 69 MGE-enriched proteins in the secretomes of both cancer cell lines that were absent from the hTERT HPNE cell line when identified by using MGE-LC–MS/MS (Fig. 7 and Table S4) compared with only 11 obtained through conventional LC–MS/MS proteomics. This ∼6-fold larger set of proteins, comprising biomarker candidates, demonstrates the value of combining MGE enrichment with LC–MS/MS.Figure 7Venn diagram of secretome proteins identified by MGE-LC–MS/MS analysis. The numbers of secretome proteins identified in patient-derived P198 and JD13D cancer cell lines and the near-normal hTERT HPNE line after MGE-based enrichment are shown (the full name and complete list of proteins is given in Table S1). N-glycosite analysis was performed to identify proteins that had no potential for N-glycosylation. Proteins that meet this criterion were marked with an asterisk. Potential glycosites are noted in Table S4. hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line; MGE, metabolic glycoengineering.

As discussed previously, PALLD is an example of a cancer-associated protein that cannot be used as a biomarker using conventional LC–MS/MS because it was ubiquitously identified across both normal and cancerous cell lines. Many proteins that share this quality are among the 98 cancer-specific biomarker candidates identified with MGE-LC–MS/MS in the P198 and JD13 cells (Fig. 7). Importantly, however, a subset of the proteins identified in all cell lines using conventional LC–MS/MS (Table S1) were uniquely identified in the two cancer lines using MGE-LC/MS/MS (Table S3). At a biochemical level, these findings suggested that PC cell lines have an ability not found in normal cells to selectively incorporate non-natural, azido-labeled monosaccharides into their glycans. We discounted that selective incorporation of this type was mediated by cancer cell–selective sialyltransferase expression by profiling transcript levels of sialic acid–processing genes in cancer (Capan-2) and near-normal (hTERT HPNE) cells (Fig. S2). Only two genes, the ST6GALNAC1 and ST6GALNAC2 sialyltransferases, were differentially expressed between the two cell lines, showing enhanced levels in the cancer cells. However, because these enzymes add sialosides to O-linked glycans, which were not labeled by 1,3,4-O-Bu_3_ManNAz treatment, their differential expression could not explain the cell line–specific MGE-LC–MS/MS identification of proteins we observed.

In another initially puzzling result, several proteins identified using MGE-LC–MS/MS from patient-derived PC cells were not N-glycosylated, similar to SBSN or FAM3C identified from hTERT HPNE/Capan-2 results (Fig. 4A). Unlike SBSN or FAM3C, these proteins were neither O-glycosylated nor did they have carbohydrate-binding domains or were they known to be secreted into the extracellular milieu, such as dCTP pyrophosphatase 1 (DCTPP1) (106). To explain this set of nonintuitive findings, we hypothesized that EVs could be enriched via azido-modified surface glycoproteins, thereby allowing cargo proteins to be identified from the 1,3,4-O-Bu_3_ManNAz-treated samples.

MGE enhances EV secretion, facilitating sensitive cancer biomarker profiling

To evaluate the putative enrichment of cargo proteins from azido-labeled EVs, we incubated the model pancreatic ductal adenocarcinoma (PDAC) Kras^LSL.G12D/+^, p53^R172H/+^, Pdx^Cretg/+^ (KPC) cell line with 1,3,4-O-Bu_3_ManNAz at 250 μM for 48 h in EV-free media before collecting the supernatant, isolating the EVs, and performing nanoparticle tracking analysis (NTA). As a notable aside, 1,3,4-O-Bu_3_ManNAz dramatically increased the production of EVs by ∼20.3-fold, an effect mimicked by the non–azido analog 1,3,4-O-Bu_3_ManNAc with an ∼17.3-fold increase in this murine cell line. In confirmation of these unexpected results, EVs were isolated from 1,3,4-O-Bu_3_ManNAz-treated, metastatic PC SW1990 cells and counted using VideoDrop, which has a measurable size range of 80 to 500 nm that ensures that the NTA quantification was not an artifact of cellular debris. Instead, the VideoDrop results confirmed an increase in EV production of ∼5.5-fold in human cells (Fig. 8A). To evaluate EVs prepared from a second 1,3,4-O-Bu_3_ManNAz-treated human cell line (patient-derived JD13D cells), we used DCTPP1 (identified above in the MGE-LC–MS/MS analysis of patient-derived cell lines) as a model to ascertain whether nonglycosylated proteins could be detected as cargo in these azido-modified vesicles. Western blot analysis against DCTPP1 was conducted on both cell and purified EV lysates as well as parallel samples treated with 1,3,4-O-Bu_3_ManNAz to confirm that this protein’s expression was not altered by this analog. Equal levels of DCTPP1 expression were found in both control and MGE-treated EVs, demonstrating that DCTPP1 is generally present in PC EVs and not directly impacted by MGE with 1,3,4-O-Bu_3_ManNAz (Fig. 8B). This result is also consistent with prior publications, which detected DCTPP1 in the EVs of PC (107).Figure 8Tributanoylated ManNAc analogs increase EV production in pancreatic cancer cells. (A) The numbers of EVs harvested from human pancreatic cancer (SW1990) cells incubated without analog or with 250 μM 1,3,4-O-Bu_3_ManNAz were counted via VideoDrop. The error bars presented are one SD from the mean with an N = 3. ∗∗p < 0.01 as calculated by a Student’s t test. (B) Western blot comparing cell lysate and purified EV lysate from JD13D cells either treated with 250 μM 1,3,4-O-Bu_3_ManNAz or volumetric equivalents of ethanol as control. DCTPP1 was selected because it was identified by MGE-LC–MS/MS (Fig. 7) despite lacking an N-glycan sequon. The absence of ApoA1, a well-known EV-negative marker for EVs purified by ultracentrifugation, indicated that the final two lanes contained protein solely derived from JD13D EVs (168). (C) EVs harvested from murine pancreatic cancer (KPC) cells incubated with the indicated media supplements were counted using NTA and graphed with error bars that are one SD from the mean with an N = 3. ∗p < 0.05 and ∗∗p < 0.01 as calculated by a Student’s t test. DCTPP1, dCTP pyrophosphatase 1; EV, extracellular vesicle; MGE, metabolic glycoengineering; NTA, nanoparticle tracking analysis.

The ability of MGE-LC–MS/MS to enrich nonglycosylated EV cargo proteins, exemplified by DCTPP1, and enhance EV production (Fig. 8), speculatively explains otherwise-puzzling aspects of the DAVID pathway analysis presented above (e.g., in Fig. 7). In particular, the pathway analyses that distinguished cancer cells (P198 and JD13D) from the near-normal cells (hTERT HPNE) showed selective enrichment of nonglycosylated cytosolic and nuclear proteins. Several of these proteins were connected to the pyrimidine pathway by which cancer cells support de novo biosynthesis of nucleotides needed for DNA replication and RNA production (108, 109). These include DNA polymerase epsilon 3; RNA polymerase I subunit C; RNA polymerase II subunit H; and carbamoyl-phosphate synthetase 2, aspartate transcarbamylase, and dihydroorotase.

To delineate the mechanism responsible for the robust increase in EV secretion observed upon treatment with tributanoylated ManNAc analogs, we next sought to clarify which component of these compounds drives this phenotype. Each analog introduces three variables: ManNAc-dependent flux into the sialic acid pathway, the azido functional group, and intracellularly released butyrate produced by nonspecific esterases. Any of these factors could theoretically contribute to altered EV output. Accordingly, we designed a comprehensive experiment to compare the EV production in KPC cells treated with no analogs, 1,3,4-O-Bu_3_ManNAc, 1,3,4-O-Bu_3_ManNAz, 1,3,4-O-Bu_3_GlcNAc, and two combinations of sodium butyrate (NaBu) plus ManNAc to discriminate the roles of carbohydrate identity, azido modification, and short-chain fatty acid (SCFA) release. Only the tributanoylated ManNAc–based analogs, 1,3,4-O-Bu_3_ManNAc and 1,3,4-O-Bu_3_ManNAz, supported significant increases in EV production, demonstrating that neither the azido group alone nor the presence of butyrate or ManNAc at matched molar concentrations recapitulated this effect (Fig. 8C). Inclusion of the GlcNAc analog controlled for monosaccharide specificity; the results, where the analog had a measurable but muted effect compared with the ManNAc-based analogs, were consistent with reports that GlcNAc 2-epimerase can convert this sugar to ManNAc (110), which can be phosphorylated by GNE and feed sialic acid biosynthesis (111, 112, 113). Finally, providing NaBu and ManNAc either at concentrations equimolar to esterase-processed analogs (750 μM NaBu + 250 μM ManNAc) or at biologically active but poorly cell-permeant concentrations reported in the literature (e.g., 2.5 mM NaBu + 25 mM ManNAc (114, 115, 116, 117)) resulted in no significant increase in EV production. The slight, nonsignificant elevation observed at high doses implies that intracellular uptake and coordinated flux of the butyrate–hexosamine conjugate, rather than independent exposure to either component, triggers enhanced EV secretion.

MGE-mediated enhanced EV secretion linked to changes in tetraspanin expression

To investigate the mechanisms underlying the pronounced increase in EV secretion observed upon treatment with 1,3,4-O-Bu_3_ManNAz, we next assessed whether this analog altered EV biogenesis or physical characteristics in a manner that could explain this effect. Transmission electron microscopy (TEM) revealed no differences in vesicle circularity or size between control and analog-treated samples, indicating that 1,3,4-O-Bu_3_ManNAz does not induce EV overproduction by altering vesicle morphology or generating aberrant vesicle subtypes (Fig. 9A). We then used ExoView, an affinity-based platform that captures and quantifies EVs via canonical tetraspanin markers (CD9, CD63, and CD81), to evaluate whether analog supplementation modulated the expression of tetraspanins known to regulate EV trafficking and functional output (118, 119, 120). Notably, treatment with 1,3,4-O-Bu_3_ManNAz shifted the tetraspanin coexpression pattern of KPC–EVs, with a marked decrease in CD63-positive vesicles and a corresponding increase in CD9 double-positive vesicles (Fig. 9B). Surprisingly, the decrease in CD63-positive EVs and increase in CD9-positive EVs did not correlate with the number of potential N-glycosylation sites (CD63 possesses three N-glycans, whereas CD9 carries only one), indicating that differential sialylation of tetraspanins alone does not account for this divergence (121, 122).Figure 9Tributanoylated ManNAc analogs impact EV tetraspanin expression. (A) Transmission electron microscopy (TEM) images of EVs collected from KPC cells either treated with volumetric equivalents of ethanol as control (left) or 250 μM 1,3,4-O-Bu_3_ManNAz (right). The EVs displayed similar morphology and size, demonstrating that MGE does not alter EV morphology. (B) ExoView phenotyping of EV tetraspanin markers (CD9, CD63, and CD81) of EVs secreted by ethanol-treated (control) or 250 μM 1,3,4-O-Bu_3_ManNAz-treated KPC cells. Tetraspanin association with EVs captured using CD81 or CD9 as representative markers was graphed as percentages of the bound population. (C) Comparison of the population percentages of EV tetraspanin (CD9, CD63, and CD81) between control and 250 μM 1,3,4-O-Bu_3_ManNAz-treated groups. The error bars presented are SD, N = 5. ∗∗∗p < 0.001 as calculated by a Student’s t test. EV, extracellular vesicle; MGE, metabolic glycoengineering.

Prior studies have shown that CD63 surface expression is regulated by polylactosamine structures on the protein, where increased polylactosamine enhances CD63 localization to the plasma membrane (123). As sialylation caps polylactosamine extension, the elevated sialylation induced by 1,3,4-O-Bu_3_ManNAz provides a mechanistic rationale for the observed decrease in CD63. However, reduced CD63 cannot readily explain the substantial increase in EV secretion. Functionally, CD63 is predominantly associated with late endosomal compartments and intraluminal vesicle formation within multivesicular bodies. The reduction in CD63-positive populations suggests that MGE alters classical late endosomal maturation or the endosomal sorting complex required for transport–dependent pathways. Reports that CD63 glycosylation modulates its interactions within the Golgi and influences cargo sorting (e.g., CXCR4) suggest that altered CD63 sialylation could modify its trafficking function independently of expression level (124). Furthermore, sialylated CD63 has been linked to increased cancer cell proliferation and metastatic behavior, processes frequently associated with elevated EV release, implying that qualitative changes to CD63 function may favor EV production (125, 126).

In parallel, the striking increase in CD9 expression is consistent with an orthogonal mechanism driven by the intracellular release of SCFAs from tributanoylated ManNAc analogs. CD9 is enriched in plasma membrane–derived EVs and early endosomal compartments (127), and multiple studies have proposed that CD9 directly contributes to EV biogenesis through both endosomal sorting complexes required for transport–dependent and –independent pathways (127, 128, 129, 130). The butyrate groups of 1,3,4-O-Bu_3_ManNAz are cleaved by nonspecific esterases, generating intracellular butyrate, a known histone deacetylase inhibitor (114). Because CD9 expression is negatively regulated by histone deacetylation, analog-derived SCFAs likely promote CD9 upregulation through epigenetic relaxation. Thus, concurrent CD9 upregulation and CD63 functional modulation provide a coherent model in which tributanoylated ManNAc treatment rewires the tetraspanin landscape to favor EV production. The shift from late endosomal CD63+ vesicles toward plasma membrane/early endosomal CD9+ vesicles, coupled with enhanced total tetraspanin expression across all markers (Fig. 9C), indicates that MGE activates multiple parallel biogenesis pathways rather than amplifying a single dominant route. In particular, we hypothesize that the combined effects of SCFA-mediated CD9 induction and sialylation-dependent alterations in CD63 function create an EV-permissive state. Because these mechanisms rely on conserved metabolic and epigenetic features, they may operate across cell types but are likely amplified in cancer cells, which often exhibit elevated histone acetylation (131, 132, 133) and polylactosamine biosynthesis (134, 135, 136).

MGE-LC–MS/MS enrichment of secretome proteins has a broad dynamic range

The inherent complexity, coupled with the broad dynamic range of the diverse biological molecules present in the serum, poses a challenge to biomarker discovery from plasma samples. The problem is compounded for early cancer diagnosis when biomarkers are present at levels too low to detect against background, nontarget proteins. We postulated that MGE could provide a solution by enriching for low abundance, bio-orthogonally tagged biomarkers. We did not want, however, to lose the ability to simultaneously detect abundant biomarker candidates. Accordingly, we quantified the enrichment of biomarker candidates across a several-order-of-magnitude concentration range by comparing proteins that were differentially enriched (or depleted) by MGE-LC–MS/MS compared with conventional LC–MS/MS in three categories. These categories included (i) enrichment where a protein was absent from the alternative method of analysis; (ii) enrichment by 3σ or greater; and (iii) enrichment by twofold or greater but less than 3σ. As shown in Figure 10A, where the numbers of each protein assigned to each category are plotted to scale, the three cell lines show roughly similar responses with only a relatively small fraction of proteins enriched by either conventional or MGE-based LC–MS/MS in each cell line (i.e., 10.1% [217/2145] in hTERT HPNE cells; 7.24% [155/2140] in P198 cells; and 6.76% [146/2159] in JD13D cells). This depiction shows that in each cell line there was a set of enriched proteins (i.e., they were more abundant when identified by MGE-LC–MS/MS compared with LC–MS/MS) and a set of depleted proteins (i.e., they were more abundant when identified by LC–MS/MS). Nevertheless, there were some notable differences; for example, a comparison of the near-normal hTERT HPNE cell line with the P198 cancer line showed a bias toward enrichment in the MGE-LC–MS/MS samples in the cancer cells, whereas the JD13D cells showed a bias toward the identification of a larger number of proteins by conventional LC–MS/MS than either of the other two cell lines.Figure 10Relative protein abundance and multivariate analysis. (A) The relative abundance of proteins with statistically significant enrichment or depletion is shown based on three levels (>2-fold; >3σ; and complete depletion or enrichment, which means the protein was not detected by the alternative method). MvA plots for (B) hTERT HPNE, (C) P198, and (D) JD13D cells show fold change as a result of coupling MGE to LC–MS/MS versus conventional LC–MS/MS. Red symbols on the graph indicate proteins with >2-fold enrichment, blue symbols indicate >2-fold depletion in MGE-LC–MS/MS versus conventional LC–MS/MS, and yellow symbols indicate changes of 3σ or greater. Insets at the top and bottom of each panel show proteins identified only by MGE-LC–MS/MS (top) or LC–MS/MS (bottom). hTERT HPNE, human telomerase reverse transcriptase–immortalized human pancreatic nestin-expressing cell line; MGE, metabolic glycoengineering; MvA, multivariate analysis.

We next generated multivariate analysis (MvA) plots to depict differentially enriched proteins based on the abundance of each protein (Fig. 10; the hTERT HPNE, P198, and JD18D cell lines are shown in B, C, & D, respectively). The MvA plots graphically depict the differences between measurements obtained through conventional LC–MS/MS proteomics and MGE-LC–MS/MS by transforming the data as the log_2_ ratio of the protein abundances and then plotting these values on the x-axis against the ratio of their mean abundance on the y-axis. A protein was deemed to be differentially enriched (or depleted) if the fold change was greater than two with a p value less than 0.05 (137). The MvA plots compare proteins partially enriched in each cell line by either MGE-LC–MS/MS (represented by positive ratio numbers on the y-axis) or conventional LC–MS/MS (represented by negative ratio numbers on the y-axis) along with proteins completely enriched by each method (shown in insets at the top and bottom of each panel), shown in Figure 10, B–D. Overall, the data points depicting enriched (or depleted) for each line are distributed from left to right on the x-axis, indicating that MGE-LC–MS/MS is not noticeably biased toward the detection and identification of low- or high-abundance proteins, making this method suitable for biomarker discovery across a broad range of concentrations.

In vivo comparison of LC–MS/MS with MGE-LC–MS/MS for secretome analysis of serum proteins from mice bearing human pancreatic tumors

Based on the promising in vitro results described previously where MGE facilitated the selective enrichment and identification of putative biomarkers for PC using cell culture experiments, we investigated the ability of MGE-LC–MS/MS to identify secreted proteins in immunocompromised mice transplanted with human pancreatic tumors. As a first step in this process, we analyzed the tumors to ensure metabolic incorporation of 1,3,4-O-Bu_3_ManNAz into the implanted cancer cells; this step was necessary because this ManNAc analog has not been tested in vivo previously. Tumor-bearing animals were dosed with 1,3,4-O-Bu_3_ManNAz through intraperitoneal injection once daily for 17 days, at which time the animals were sacrificed, and then the tumor and plasma proteins were harvested. The harvested tumors were homogenized under liquid nitrogen (N_2_), and proteins were extracted using T-PER reagent, biotinylated with alkyne–PEG–biotin, analyzed using far-Western blot analysis, and identified using LC–MS/MS. For comparison, conventional LC–MS/MS was performed on samples from vehicle-control animals. In both approaches, Coomassie-stained gels showed that sets of multiple, roughly similar proteins were isolated (Fig. 11A). Far-Western analysis, however, showed negligible signal from control LC–MS/MS samples. This is as expected because proteins do not have the endogenous azido groups needed for bio-orthogonal labeling in untreated animals; by contrast, there was strong metabolic labeling observed in samples from 1,3,4-O-Bu_3_ManNAz-treated animals (Fig. 11B).Figure 11Analysis of proteins from mice bearing transplanted human pancreatic tumors. Proteins taken directly from homogenized tumors were analyzed by Coomassie-stained gels (A) or far-Western blots (B); similar analysis was conducted for plasma-derived proteins (C & D, respectively). Each lane represents a sample extracted from a different mouse. For example, the first four lanes of (C) represent separate plasma samples from each of the four mice within the vehicle control cohort. Proteins from plasma samples were identified by conventional LC–MS/MS from control animals and by MGE-LC–MS/MS from 1,3,4-O-Bu_3_ManNAz-treated animals; the numbers of murine and human proteins are depicted in (E) and (F), respectively. MGE, metabolic glycoengineering.

This azido labeling of proteins in mice treated with 1,3,4-O-Bu_3_ManNAz was expected based on literature reports of similar experiments done with Ac_4_ManNAz, a similar but less efficient azido analog that also installs Sia5Az into mammalian glycans (46, 138). What remained unknown from previous studies was whether tumor-derived azido-labeled proteins could be detected in circulation in the tumor-bearing animals. To address this question, we analyzed plasma samples from each group of animals. Coomassie gels verified the presence of plasma proteins in all samples (Fig. 11C), which was an expected result because blood is well known to contain many proteins. The far-Western blots (Fig. 11D), by comparison, showed robust labeling selectively in samples obtained from the 1,3,4-O-Bu_3_ManNAz-treated animals. These blots, however, could not distinguish between tumor-derived human proteins and host murine proteins. Therefore, to make this determination, we next analyzed samples using conventional LC–MS/MS for the vehicle-control samples and MGE-LC–MS/MS for the 1,3,4-O-Bu_3_ManNAz-treated animals.

By using mass spectrometry (MS) coupled with bioinformatics analyses, we identified a total of 122 murine proteins in plasma; 113 of these proteins were identified using LC–MS/MS in samples from the control animals. In the MGE-LC–MS/MS experiments, we identified 47 azido-labeled proteins from the 1,3,4-O-Bu_3_ManNAz-treated animals, with 38 of these proteins represented in both sets (Fig. 11E and Table S5). Interestingly, eight of the nine murine proteins detected in the mice exclusively by MGE-LC–MS/MS (β-2-glycoprotein 1 precursor [Apoh], L-lactate dehydrogenase A chain isoform 2 [Ldha], complement component C9 precursor [C9], plasma kallikrein preproprotein [Klkb1], enolase 1B retrotransposed [Eno1b], glyceraldehyde-3-phosphate dehydrogenase isoform 1 [Gapdh], catalase [Cat], fermitin family homolog 3 [Fermt3], and transketolase [Tkt]) were associated with EVs as compiled in the ExoCarta database (http://exocarta.org/index.html). As we have seen in prior in vitro experimentation, although many of these proteins themselves are not expected to be N-glycosylated and normally have intracellular rather than cell surface or secreted localization, EVs contain surface N-glycans that can be azidomodified upon 1,3,4-O-Bu_3_ManNAz supplementation, putatively facilitating isolation of these cargo proteins. Considering that EV proteins are typically low in abundance in the plasma and can be masked by other high-abundance plasma proteins (139), the ability of MGE to enrich EV proteins in vivo, thereby facilitating their detection and identification in the presence of other high-abundance blood proteins, highlights the merit of this approach.

Overall, the number of human proteins identified by MS analyses was lower than the number of murine proteins (30 versus 122; Fig. 11F and Table S6), with 29 identified by conventional LC–MS/MS in samples from untreated control animals and six identified using MGE-LC–MS/MS from 1,3,4-O-Bu_3_ManNAz-treated animals. Although the MGE-LC–MS/MS approach resulted in the detection of ∼4-fold fewer proteins than conventional LC–MS/MS, the successful and unambiguous identification of secreted human tumor–derived proteins from in vivo rodent plasma samples is unprecedented and constitutes a new tool to profile the secretome. All six of the tumor-derived proteins identified through MGE (heat shock cognate 71 kDa protein isoform 1; keratin, type II cytoskeletal 5; actin, cytoplasmic 1; keratin, type I cytoskeletal 10; keratin, type II cytoskeletal 1; and filaggrin-2 (FLG2) are associated with EVs as listed in the ExoCarta database (http://exocarta.org/index.html) and several publications (140, 141, 142, 143).

Of these six human tumor proteins identified by MGE-LC–MS/MS in vivo, five were also identified by conventional LC–MS/MS. The only protein exclusively identified by MGE-LC–MS/MS in vivo was FLG2, whose full name filaggrin is derived from filament aggregating protein, a protein that binds to keratin fibers in epithelial cells and has been linked to skin disorders, such as eczema (144) and atopic dermatitis (145). Filaggrin can have opposing effects in cancer; its links to vitamin D can provide protection, whereas its ability to disrupt epithelial barriers promotes metastasis (146). Although FLG2 has not yet been tied to PC in the primary literature, the Human Protein Atlas (147) describes it as being expressed at moderately high levels in a subset of pancreatic carcinomas. In addition, excess FLG2 in keratinocytes is removed by secretion in EVs to prevent premature death (148). Based on our current findings, PC cells may employ a similar mechanism, providing a foundation for developing FLG2 as a biomarker for this type of cancer.

Concluding comments

This study establishes several benefits of incorporating MGE into cancer biomarker discovery (Fig. 1). First, pilot studies with the near-normal hTERT HPNE and Capan-2 cancer lines confirmed cell line–specific incorporation of azido-tagged glycans in 1,3,4-O-Bu_3_ManNAz-treated near-normal hTERT HPNE and cancerous Capan-2 cells (Fig. 2), with MS-based proteomics demonstrating that dozens of MGE-enriched secretome proteins could be identified using our MGE-LC–MS/MS strategy. Of note, a larger number of proteins were identified by MGE-based analysis in the near-normal hTERT HPNE cells compared with the Capan-2 cancer line (Fig. 4A), but this ultimately is not a problem for biomarker discovery, provided that cancer-associated proteins can also be identified selectively. This was the case in our study, evidenced by 28 proteins uniquely identified in the secretomes of the Capan-2 cells (Fig. 4).

Based on the promising results from the hTERT HPNE and Capan-2 cell lines, we expanded our analysis to clinically relevant, primary patient-derived lines in cell culture and then to a murine xenograft model. Throughout these studies, we compared the MGE-based LC–MS/MS strategy with conventional LC–MS/MS proteomics. In vitro, we observed cell type–specific labeling of secreted proteins, and perhaps most important, we demonstrated that the MGE-LC–MS/MS strategy enhanced the identification of cancer-associated proteins as illustrated by GBA, PALLD, and the ∼6-fold higher number of cancer-specific proteins identified by this method (Fig. 7) compared with conventional LC–MS/MS (Fig. 6). We then gained insight into the unexpected identification of cancer biomarkers previously unassociated with the secretome by demonstrating that the MGE-LC–MS/MS workflow generates ∼10-fold higher EV secretion (Fig. 8) associated with altered expression of canonical tetraspanins (Fig. 9). We posit that the increased production of tumor-associated EVs upon 1,3,4-O-Bu_3_ManNAz treatment enables the enrichment of low-abundance biomarker candidates that would otherwise be present at undetectably low levels. This functionality has not yet been observed via other methods of EV glycoengineering (149).

An important and novel aspect of this study is an in vivo proof-of-principle demonstration for MGE-LC–MS/MS biomarker discovery of secretome proteins. Despite the overall lower number of proteins subsequently identified in vivo, we nevertheless successfully enriched and identified azido-labeled human tumor–associated proteins from the plasma of tumor-bearing mice (Fig. 11). In addition to confirming individual proteins (e.g., DCTPP1) by MGE-LC–MS/MS in PC EVs, we identified cancer-related metabolic pathways that were selectively responsive to MGE labeling. In particular, the lysosome and pyrimidine pathways used for the biosynthesis of metabolites critical for the proliferation of cancer cells were selectively enriched and identified through MGE-LC–MS/MS biomarker discovery. We emphasize that although it is too early to project whether the specific biomarker candidates identified in this study will ultimately translate to the clinic, we nonetheless find these results exciting because they indicate that metabolic bio-orthogonal labeling of sialoglycoproteins can enhance the detection of these putative cancer biomarkers in primary, patient-derived cell lines as well as in vivo.

Finally, from a technology development standpoint, the MGE-based approach described in this report adds a versatile and powerful tool that complements, and extends, a decade of efforts to glycoengineer EVs. Briefly, in 2015, Hung and Leonard (150) engineered the EV-associated protein Lamp2b to be highly glycosylated to avoid proteolytic degradation. Another approach included genetic modulation of glycogenes, for example, FUT7 and FUT9, to create sialyl Lewis X–expressing EVs (149). Conversely, in other cases, glycogenes have been knocked out of EV-producing cells to attenuate glycosylation (151); deglycosylation may benefit EVs by enhancing cellular uptake (152). More recently, the Guo group has developed chemoenzymatic methods to install azido-modified sialic acids onto EVs (153, 154). By using 1,3,4-O-Bu_3_ManNAc, our approach achieves the same objective in a simple and cost-effective manner, and in the process, dramatically increases EV production, a benefit not found in any other glycoengineering (or other) approach.

Experimental procedures

Cell lines

The pancreatic cell lines, hTERT HPNE E6/E7st (American Type Culture Collection [ATCC] CRL-4037), Capan-2 (ATCC HTB-80), and SW1990 (ATCC CRL-2172), were obtained from the ATCC. Patient-derived PC cells, P198 and JD13D, were obtained from the PancXenoBank Division of Gastrointestinal and Liver Pathology, Department of Pathology, The Johns Hopkins School of Medicine, Institutional Review Board number NA_00001584. Authentication of the patient-derived PC cells was done by PancXenoBank Division of Gastrointestinal and Liver Pathology. KPC murine PDAC cells, originally derived from a genetically engineered mouse model (155), were a gift kindly provided by Dr Lei Zheng (UT Health San Antonio MD Anderson Mays Cancer Center). Mycoplasma contamination was tested by the Fragment Analysis Facility (Johns Hopkins University School of Medicine), and contamination with other bacteria was monitored using the Molecular Probes Cell Culture Contamination Detection Kit (Thermo Fisher Scientific). None of the cell lines used were classified as commonly misidentified lines.

Cell culture

All cells were cultured in a humidified incubator maintained at 37 °C with 5.0% CO_2_. Capan-2 cells were cultured in McCoy’s 5a modified medium (Thermo Fisher Scientific) supplemented with 10% (v/v) fetal bovine serum (FBS), whereas the hTERT HPNE cell line was cultured in 75% (v/v) Dulbecco’s modified eagle medium (DMEM; Thermo Fisher Scientific) with 25% M3 supplement (InCell), 750 ng/ml puromycin, and 10 ng/ml human recombinant epidermal growth factor. These media contain a final concentration of 5.0% (v/v) FBS. The P198, JD13D, SW1990, and KPC cell lines were cultured in high-glucose DMEM (Thermo Fisher Scientific) supplemented with 10% (v/v) FBS.

Media composition was kept constant with the P198, JD13D, SW1990, and KPC experimental conditions to avoid metabolic confounding. Unless otherwise noted, all culture media were supplemented with 1.0% (v/v of a 100x stock solution) antibiotic–antimycotic (AA, 10,000 units penicillin, 10 mg/ml streptomycin, 25 μg/ml amphotericin B; MilliporeSigma) (156).

MGE of cells with 1,3,4-O-Bu3ManNAz

The metabolic precursors, 1,3,4-O-Bu_3_ManNAz and 1,3,4-O-Bu_3_ManNAc, were synthesized and characterized as published (42, 46). The compounds were purified using column chromatography and stored at −20 °C before use. Stock solutions (100 mM in ethanol) were made as needed and stored at 4 °C for a maximum of 6 weeks. Addition of metabolic precursor to culture medium was performed in parallel with controls, ethanol, which was always less than 0.1% v/v. In our previous MGE experiments, this level of ethanol is biocompatible with the cells and does not affect any of the glycan-related endpoints being evaluated (46).

Far-Western analysis of azido-analog incorporation into N- and O-linked glycans

Capan-2 and hTERT HPNE cells were seeded at appropriate concentrations to achieve a final cell count of 1,200,000 cells per well in 6-well tissue culture plates after 8, 24, 48, and 96 h when supplemented with 1,3,4-O-Bu_3_ManNAz at 200 μM. The proteins were extracted from cell lysate following resuspension in a 1:50 ratio of protease inhibitor cocktail (Promega, G6521) in radioimmunoprecipitation assay (RIPA) lysis buffer (MilliporeSigma, R0278-50ML) and placed in −20 °C until analysis. The proteome sample content was measured using a bicinchoninic acid (BCA) assay and equilibrated to 10 μg aliquots. Protein samples were divided into three groups: a control group with no glycosidase treatment, PNGase F (New England Biolabs [NEB], P0705S) treated, and Protein Deglycosylation Mixture II (NEB, P6044S)–treated samples following the manufacturer’s instructions. After deglycosylation, the protein samples were incubated with WS DBCO Biotin (Vector Laboratories, CCT-A116) for 1.0 h. The samples were provided with a Laemmli solution supplemented with 10% 2-mercaptoethanol and stored at −80 °C until analyzed by SDS-PAGE and far-Western analysis.

Purification of secretome proteins

Cells were seeded at 5,000,000 cells/dish in 150 mm tissue culture dishes with 30 ml of their respective medium and were incubated overnight. The following day, the media were replaced with FBS- and AA-free medium supplemented with 0 (controls) or 200 μM of 1,3,4-O-Bu_3_ManNAz or 1,3,4-O-Bu_3_ManNAc. In the initial experiments with the hTERT HPNE and Capan-2 cells, the cell culture supernatants from each plate were collected at 8, 24, 48, and 96 h and, in subsequent experiments with hTERT HPNE, P198, and JD13D cells, at 24 h. The supernatants were immediately snap-frozen in liquid N_2_ and stored at −80 °C until analyzed. Secreted proteins in the supernatant were precipitated using the trichloroacetic acid–sodium deoxycholate method (157).

Coomassie gel staining and far-Western blot analysis

Precipitated secretome proteins were dissolved in 50 mM Tris–HCl (pH 8) with 1.0% SDS in PBS, with each sample containing between 5 and 300 μg of secreted protein as determined by UV–visible spectroscopy at 280 nm; this method was used because BCA analysis can be unreliable in the presence of 1.0% SDS. The proteins were biotinylated by copper-catalyzed click chemistry reaction with biotin–PEG4–alkyne and purified using the Click-iT Protein Reaction Buffer kit (Thermo Fisher Scientific, C10276) following the manufacturer’s instructions. Proteins were separated on 10% SDS-PAGE gels that were run in duplicate with 5 to 30 μg of a sample (depending on the experiment) loaded per lane. One of each set of two gels was stained with Coomassie stain (BioRad; 1610803) overnight following the manufacturer’s protocols, destained in water, and imaged the following day. The proteins from the other gel were transferred onto a nitrocellulose membrane, blocked in 5% (w/v) bovine serum albumin in Tris-buffered saline with 0.1% (v/v) Tween-20 (TBST) for 1 h, and probed with streptavidin–horseradish peroxidase (HRP) (Cell Signaling, 3999S) in 5% (w/v) bovine serum albumin in TBST used at a ratio of 1:10,000 or 1:150,000 for proteins extracted from in vivo samples or in vitro, respectively. The membrane was washed with TBST (3 × 10 min), and chemiluminescent substrate (Thermo Fisher Scientific, 34095) was added, and the blots were imaged using a GE Healthcare ImageQuant LAS 4000 series instrument.

SYPRO Ruby staining and far-Western blot analysis

Proteome proteins were dissolved in Laemmli solution supplemented with 10% 2-mercaptoethanol and boiled at 100 °C for 10 min before they were flash cooled on ice. Each sample contained 10 μg of secreted protein as determined by the BCA assay. The proteins were separated on 10% SDS-PAGE gels. Gels for protein expression were stained using SYPRO Ruby Protein Gel Stain (Thermo Fisher Scientific, S12001) as per the manufacturer’s instructions. Briefly, gels were placed in a clean container with 100 ml of fix solution (50% methanol, 7% acetic acid in ultrapure water [>18 MΩ·cm resistance]) for 30 min before being placed in 60 ml of SYPRO Ruby Protein Gel Stain overnight. The gels were then washed using 100 ml ultrapure water (>18 MΩ·cm resistance) for 30 min before imaging. Gels for far-Western analysis were transferred onto a nitrocellulose membrane using an iBlot 2 Dry Blotting System (Invitrogen, IB21001), blocked in EveryBlot Blocking Buffer (Bio-Rad; 12010020 and 1201002) for 5 min, and probed with streptavidin–HRP (Cell Signaling, 3999S) in EveryBlot Blocking Buffer at a ratio of 1:20,000 for 1.0 h. The membrane was washed with TBST (3 × 10 min), and chemiluminescent substrate (Thermo Fisher Scientific, 34095) was added, and the blots were imaged using a GE Healthcare ImageQuant LAS 4000 series instrument.

LC–MS/MS analysis of Capan-2 and hTERT HPNE cells

Precipitated proteins (200 μg) from the secretome of untreated control cells were dissolved in 7.0 M urea in 50 mM Tris–HCl (pH 8.0), reduced by adding DTT to a final concentration of 5.0 mM, then incubated for 30 min. Next, the proteins were alkylated by adding iodoacetamide to a final concentration of 15 mM and incubating in the dark for 30 min. A 20 μg aliquot of MS-grade Trypsin/Lys C Mix (Promega) was added to the reduced and alkylated proteins, and the mixture was incubated overnight according to the manufacturer’s protocol, then particulate matter was removed by centrifugation. The supernatant containing the peptides was incubated in PNGase F (NEB, P0705S) constituted in Glyco buffer 2 using protocols supplied by NEB and then incubated overnight at 37 °C with mixing using a head-over-head shaker. The deglycosylated peptides were analyzed by LC–MS/MS as described below.

MGE-LC–MS/MS analysis of Capan-2 and hTERT HPNE cells

Precipitated proteins (350 μg) from the secretomes of 1,3,4-O-Bu_3_ManNAz-treated Capan-2 and hTERT HPNE cells were dissolved in PBS and reacted with biotin–PEG4–alkyne using the Click-iT Protein Reaction Buffer kit (C10276) following the manufacturer’s instructions. Next, 200 μg of the biotinylated proteins were dissolved in 7.0 M urea in 50 mM Tris–HCl (pH 8), reduced by adding DTT to a final concentration of 5.0 mM, and incubated for 30 min. The proteins were alkylated by adding iodoacetamide to a final concentration of 15 mM, followed by incubation in the dark for 30 min. The proteins were digested overnight at 37 °C using 20 μg of MS-grade Trypsin/Lys-C Mix based on the manufacturer’s instructions. The peptides were recovered by centrifugation to remove the particulate matter.

The biotinylated glycopeptides were captured on streptavidin magnetic beads (Dynabeads M-280 Streptavidin; Thermo Fisher Scientific) using the manufacturer’s protocols. Briefly, the beads were washed three times with 1.0 ml of PBS by gently inverting the tube several times, then separating the magnetic beads from the supernatant on a magnetic stand for 5.0 min. The samples were added to the beads and allowed to bind for 1 h at room temperature with gentle mixing using a head-over-head shaker. The beads were washed five times with 1.0 ml of PBS and twice with 1.0 ml of 100 mM ammonium carbonate to remove nonbiotinylated peptides.

For the initial experiments with the Capan-2 and hTERT HPNE cells, after the capture of the azido-modified biotinylated glycopeptides on the streptavidin beads, the beads were incubated in PNGase F (NEB, P0705S) constituted in Glyco buffer 2 following NEB protocols overnight at 37 °C in a head-over-head shaker. For subsequent experiments with hTERT HPNE, P198, and JD13D cell lines and plasma proteins from in vivo experiments, the glycopeptides were released from the beads by incubation with Protein Deglycosylation Mix II (NEB, P6044S) following NEB protocols. PNGase F released N-linked glycans only, whereas enzymes in the Protein Deglycosylation Mix II remove N-linked glycans, short O-linked glycans, and selected long O-linked glycans from the proteins (in practice, as demonstrated in the results of this study, incorporation of 1,3,4-O-Bu_3_ManNAz-derived azido-monosaccharides into glycans was minimal; so for all intents and purposes, the two cleavage methods were identical in this study). For all experiments, the beads were separated on a magnetic stand, the supernatant was collected, and the released peptides were subjected to LC–MS/MS analysis as described below.

LC–MS/MS analysis

For the initial experiments with samples obtained from the Capan-2 and hTERT HPNE cell lines, as well as plasma proteins obtained from later in vivo experiments, the deglycosylated peptides were desalted by using SCX phase (PolyLC, Inc) liquid chromatography, washed with 200 μl of 0.1% TFA, eluted with 30 μl of 60% acetonitrile/0.1% TFA, and then 30 μl of 80% acetonitrile/0.1% TFA, followed by analysis using an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific). For follow-up experiments with hTERT HPNE, P198, and JD13D samples, the deglycosylated peptides from treated and control samples were labeled with tandem mass tag reagent (Thermo Fisher Scientific, 90110), mixed together, then purified with SCX phase chromatography, fractionated as described above, and analyzed using a high-resolution Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific).

Database searching

All MS/MS results (Tables S1, S5, & S6) were analyzed using Mascot 2.6.2 (Matrix Science), which was preset to search the RefSeq2017_83_Human_170917 or RefSeq2017_83_mus_musculus_190523 database with a fragment ion mass tolerance of 0.020 Da, parent ion tolerance of 5.0 ppm while assuming trypsin digestion. Deamidation of asparagine and glutamine, as well as oxidation of methionine and carbamidomethyl of cysteine, was specified in Mascot as modifiable variables.

Criteria for protein identification and data analysis

Scaffold 4.10.0 (Proteome Software, Inc) or Proteome Discoverer 2.3 (Thermo Fisher Scientific) was used to validate MS/MS results. Peptide identifications were accepted if they were verifiable at >95% probability by the Peptide Prophet algorithm with Scaffold delta-mass correction (158). Likewise, protein identifications were accepted if they could be established at >95% probability with a minimum of one peptide being recognized and a maximum of one missed cleavage site by the Protein Prophet algorithm (159). Proteins containing similar peptides that could not be differentiated by MS/MS analysis alone were grouped based on the principles of parsimony.

Analysis of fold change

Proteome data derived from the Proteome Discoverer 2.3 software were imported with Partek Genomics Suite 7.0 (Partek, Inc) for protein annotation. Discoverer assigns one or more detected mass spectra to a National Center for Biotechnology Information RefSeq identifier, representing a single protein isoform, and for each identifier, the median of these multiple spectra values was calculated to produce a single log_2_ abundance value representing that protein. Only those spectra with a unique, single, RefSeq, and Proteome Discoverer isolation interference value of <30% were accepted for further evaluation. The values of the data were quantile-normalized to minimize experimental noise, and variation was log_2_ converted for statistical analysis. To determine the relative protein expression levels of the different samples (analog-treated and control samples) in Partek, their quantile normalized log_2_ values underwent two-tailed one-way ANOVA t test analysis wherein the ratio of the protein’s abundance in treated relative to untreated samples was reported as a fold change. To characterize these fold changes, the log_2_ values underwent standard deviation analyses, allowing us to set appropriate fold-change thresholds of differential enrichment. The abundance of each protein obtained in treated and untreated samples was compared to obtain linear fold-change values, and those proteins whose linear fold changes were greater than two and had p values less than 0.05 were deemed to be differentially enriched by the MGE approach. To graphically depict and examine differentially enriched proteins using MvA, these values were imported into Spotfire DecisionSite with Functional Genomics v9.1.2 (TIBCO Spotfire) to obtain MvA plots.

Bioinformatics analysis

The official gene IDs of LC–MS/MS–identified proteins were exported to the DAVID bioinformatics resources at https://david.ncifcrf.gov/summary.jsp for Gene Ontology (cellular component) and Kyoto Encyclopedia of Genes and Genomes pathway analyses. For N- and O-glycosite prediction, the FASTA sequence of the proteins was obtained from UniProt (160) and exported to SPRINT-Gly (161).

Western blot analysis of GBA in secreted proteins

Enrichment of the biotinylated secretome samples for GBA was carried out using Dynabeads M-280 Streptavidin. The beads (100 μl) were washed with 1 × 50 mM Tris with 0.1% SDS, pH 7.5, 1x TBS with 2.0 M urea, and 1x TBS and incubated with 800 μg of sample in 1.0 ml of 50 mM Tris with 0.1% SDS, pH 7.5. After 1 h, the unbound fractions were collected, and the beads were boiled in 50 μl of Laemmli solution supplemented with 10% 2-mercaptoethanol at 90 °C for 10 min before being flash cooled on ice to elute the bound glycoproteins. Western blots were probed with anti-GBA monoclonal antibodies (1:1000 dilution, Abcam, ab55080) 1 h at room temperature with gentle agitation. Following washes with TBST (3 × 10 min), membranes were incubated with goat anti-mouse IgG-HRP secondary antibody (1:10,000 dilution in TBST; Thermo Fisher Scientific, 31430) for 1 h at room temperature.

Western blot analysis of GBA in cell lysates

Primary cells were cultured in 6-well plates at 300,000 cells/well overnight. The following day, the medium was aspirated and replaced with an FBS-, AA-free medium containing 200 μM of analog (1,3,4-O-Bu_3_ManNAz or 1,3,4-O-Bu_3_ManNAc) in ethanol or an equivalent volume of ethanol vehicle. After 24 h, the medium was aspirated, and the cells were washed with PBS. Then, the cells were lysed in 30 μl of RIPA buffer containing protease inhibitor cocktails on ice for 20 min and centrifuged at 16,000g for 15 min at 4.0 °C, and the supernatants containing the cell lysates were collected. Aliquots containing 10 μg of protein were loaded on 4% to 15% gel for SDS-PAGE separation for GBA analysis. Western blots were performed as described for the secreted streptavidin-captured sialoglycoproteins in Western blot analysis of GBA in secreted proteins.

Transcript analysis

The following methodology was used to acquire the data presented in Figure S2. Cells were seeded in 150 mm dishes at 5,000,000 cells/well overnight in their respective media. The following day, the medium was aspirated and replaced with FBS-, AA-free medium containing 200 μM of 1,3,4-O-Bu_3_ManNAz in ethanol or an appropriate volume of ethanol vehicle. After 24 h, the cells were snap-frozen in liquid N_2_ and stored at −80 °C until analysis. Total RNA isolation and complementary DNA synthesis were carried out as described previously (162) whereas the qRT–PCRs were performed for each gene analyzed using previously reported primer pairs (66). Amplification conditions and data analysis were performed as described (163). The transcript level for each gene was expressed as relative transcript abundance, which was normalized to a housekeeping gene, β-actin, to keep the data consistent between the multiple plates that were run individually. The values were plotted on a log_10_ scale because of the large dynamic range of the qRT–PCR data. A relative transcript abundance value of 1 × 10^-6^ (bottom of scale) is equivalent to a Ct value of 35 (the cutoff Ct for the experiment). All experiments and PCR analyses were performed in biological triplicate.

Flow cytometry assays

Primary cells were seeded at 300,000 cells/well in a 6-well plate overnight in their respective medium. The following day, the medium was aspirated and replaced with FBS-, AA-free medium containing 20 μM of 1,3,4-O-Bu_3_ManNAz or an appropriate volume of ethanol vehicle. After 24 h, the medium was aspirated, the cells were washed three times with PBS, then treated with FBS-, AA-free medium containing 16.5 μM dibenzocyclooctyne-conjugated Cy5 (MilliporeSigma, 777374) for copper-free click chemistry conjugation reactions. After 1 h, the cells were harvested using Corning Cell Stripper Dissociation Reagent (Corning, 25056Cl), washed three times with PBS, then resuspended in PBS for fluorescence-activated cell sorting. The experiment was performed in triplicate on a Cytoflex (Beckman Coulter) flow cytometer.

Production, isolation, and characterization of EVs

The production, isolation, and Western blot characterization of EVs were adapted from published protocols from the Liu group (164, 165) and others (166); these methods, as used in this report, are briefly summarized below:

Production of EVs in KPC and SW1990 cells

KPC PDAC C57Bl6/J cells and SW1990 PDAC were obtained from the ATCC. The cells were seeded at 10^6^ cells in a T75 flask (Thermo Fisher Scientific) maintained in either RPMI1640 media (Thermo Fisher Scientific) or DMEM (Thermo Fisher Scientific) for KPC and SW1990, respectively. Both media were supplemented with 10% (v/v) FBS and 1.0% (v/v of a 100x stock solution) AA (10,000 units penicillin, 10 mg/ml streptomycin, 25 μg/ml amphotericin B; MilliporeSigma). For collection of EVs, the FBS of seventh passage cells was replaced with 10% (v/v) exosome-depleted FBS (Gibco) once the cell density in the flask reached 7 × 10^6^ cells. For collection of MGE EVs, the cells were incubated in 250 μM of 1,3,4-O-Bu_3_ManNAc or 1,3,4-O-Bu_3_ManNAz along with the exosome-depleted FBS for 48 h before EV isolation, as described below.

Isolation of EVs

Conditioned culture medium (30 ml) underwent centrifugation for 10 min at 300g to remove cells, followed by centrifugation at 2000g for 10 min at 4 °C to remove dead cells and large debris. The supernatant (15 ml) was subsequently concentrated by centrifugation at 4000g for 20 min using an Amicon Ultra-15 filter column with an Ultracel-100 membrane (MilliporeSigma). Concentrated EVs (approximately 250 μl) were isolated by size-exclusion chromatography using qEV columns (iZON, IC1-70) following the manufacturer’s instructions. In brief, a 0.5 ml fraction of the concentrated EVs was loaded at the column's top and eluted with PBS, with the EV-rich fractions then pooled. The purified EVs were further concentrated using an Amicon 100 kDa column. The concentration of the EVs was measured using an NTA instrument (ZetaView; Particle Metrix) with a 488-nm laser and ZetaView 8.04.02 software or a Corning VideoDrop (VD-1000).

Western blot analysis of proteins in the production cell and EVs

Proteins were extracted from cell pellets or purified EVs by resuspending samples in RIPA lysis buffer (Sigma–Aldrich, R0278-50ML) with protease inhibitor cocktail (Promega, G6521) at a 1:50 ratio. Lysates were stored at −20 °C until analysis. Total protein concentration was measured using a BCA assay. Aliquots containing 20 μg of protein were then loaded on 4% to 15% gel for SDS-PAGE separation and then transferred onto a nitrocellulose membrane using an iBlot 2 Dry Blotting System (Invitrogen, IB21001). Membranes were blocked in EveryBlot Blocking Buffer (12010020) for 5 min and then incubated with ApoA1 polyclonal antibody (1:1000 dilution; Thermo Fisher Scientific, PA5-29557) and DCTPP1 polyclonal antibody (1:10,000 dilution; Thermo Fisher Scientific, A305-794A-T) for 1 h at room temperature with gentle agitation. Following washes with TBST (3 × 10 min), membranes were incubated with goat anti-rabbit IgG-HRP secondary antibody (1:10,000 dilution in TBST; Thermo Fisher Scientific, 31460) for 1 h at room temperature. Protein bands were visualized using a chemiluminescent substrate (Thermo Fisher Scientific, 34095) and imaged using a GE Healthcare ImageQuant LAS 4000 series instrument.

TEM of EVs

EV samples were prepared for TEM using negative staining. Samples (8 μl) were adsorbed onto glow-discharged (EMS GloQube) ultrathin carbon-coated 400-mesh copper grids (EMS CF400-Cu-ultrathin) by floating for 2 min at room temperature. Grids were rinsed in three drops of TBS (5 s each), then negatively stained by floating on two consecutive drops of 0.75% (w/v) aqueous uranyl acetate. Excess stain was immediately removed by aspiration. Grids were air-dried and imaged using a Hitachi 7600 TEM operating at 80 kV, equipped with an AMT XR80 8-megapixel CCD camera. Representative images were captured at appropriate magnifications to visualize EV morphology and size distribution.

ExoView tetraspanin phenotyping of EVs

EV tetraspanin profiles were characterized using a Mouse EV-C-TETRA ExoView Tetraspanin kit and an ExoView TMR100 scanner (NanoView Biosciences) according to the manufacturer's instructions, as previously described (167). Briefly, purified KPC EVs (10 μl) were mixed with 40 μl of incubation buffer provided in the kit. The diluted sample (50 μl total volume) was loaded onto ExoView R100 chips at room temperature for 16 h. Following incubation, chips were washed with incubation buffer to remove unbound particles, then incubated with a fluorescently labeled antibody cocktail of anti-mouse CD81, CD63, and CD9 (1:1200 dilution in blocking solution) for 1 h. After washing with wash buffer, all chips were scanned using the ExoView R100 scanner, which combines interferometric reflectance imaging (single-particle interferometric reflectance imaging) for single-particle detection with three-channel fluorescence imaging for phenotypic characterization.

Data were analyzed using NanoViewer software (version 2.8.10; NanoView Biosciences). Individual EVs captured on each tetraspanin spot were counted, and colocalization of fluorescent signals was quantified to determine the percentage of each EV tetraspanin marker.

In vivo experiments

Patient-derived pancreatic tumors were obtained postsurgery from the Johns Hopkins Hospital via PancXenoBank (JHU094). Tumors were then resected and implanted as xenografts in the backs of 4-week-old female Foxn1^nu^ athymic nude mice. These mice were maintained as a live tumor bank, which conserved the in vivo properties of patient-derived tumors. When xenograft tumors reached 100 mm^3^ in size, they were excised and cut into 1 to 2 mm^3^ pieces. Tumor pieces were dipped in 50% Matrigel culture medium (Thermo Fisher Scientific) and then implanted onto the backs of nude mice. The pockets were closed using a 7-0 nylon microfilament suture. Three weeks postimplantation, the mice were randomly divided into two groups (n = 4). For one group, each of the mice received a daily dose of 500 mg/kg/d of 1,3,4-O-Bu_3_ManNAz in 200 μl of 5.0% ethanol/95% PBS (v/v) via intraperitoneal administration, whereas for the second group, each mouse was injected with 200 μl of the corresponding vehicle control. After 17 days, mice were sacrificed by isoflurane inhalation, the blood was collected by cardiac puncture, and the plasma was extracted and stored at −80 °C until analyzed. The plasma proteins were processed and analyzed by LC–MS/MS as described for the initial experiments for the Capan-2 and hTERT HPNE cell lines. The tumors were harvested and immediately snap frozen in liquid N_2_, homogenized at liquid N_2_ temperatures, proteins were extracted using T-PER tissue protein extraction reagent (Thermo Fisher Scientific, 78510), and analyzed by far-Western blot analysis as described above. The animal experiments were conducted in accordance with a protocol approved by the Johns Hopkins University Animal Care and Use Committee and in compliance with the Association for Assessment and Accreditation of Laboratory Animal Care guidelines.

Data availability

MS data for all proteins detected in vitro are available in Table S1. The analyzed transcripts of identified proteins for canonical N-glycosylation sites can be found in Tables S2, S3, and S4 for Figures 4, 6, and 7, respectively. MS data for murine proteins detected in vivo can be found in Table S5, whereas human proteins identified in vivo can be found in Table S6. Flow cytometry characterization of 1,3,4-O-Bu_3_ManNAz-treated P198 and JD13D cells can be found in Figure S1, whereas transcript profiling of genes of the sialic acid metabolic pathway in Capan-2 and hTERT HPNE cells can be found in Figure S2. VideoDrop data presented in Figure 8 and ExoView analysis presented in Figure 9 can both be found as Figure S3 and Folder S1, respectively. Folder S1 is available at the following link: Nanoview Report-KPC AND KPC-A. Any additional data are available from the authors upon request.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.