This study included a total of 134 samples collected from patients with pancreatic diseases or healthy controls (HCs) who were admitted to the Nanjing Drum Tower Hospital Affiliated to Nanjing University Medical School and the First Affiliated Hospital of Soochow University between July 2018 and December 2019. Patients considered eligible for this study were those with suspected pancreatic masses and with operation indications. Endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) and/or surgical resections were performed for initial diagnosis. Plasma samples were all collected prior to EUS-FNA operations or surgical resections. All of them had not received any antitumor therapy before. Classification of PC was determined according to the 8^th edition of the American Joint Committee on Cancer (AJCC) Classification. In the discovery stage, plasma samples were selected from 30 subjects with PC (stage I/II and stage III/IV), well-differentiated P-NETs, PCLs, CP and HC (five cases in each group) from Nanjing Drum Tower Hospital. We pooled the samples from the same group into one representative sample with the purpose of increasing the efficiency of analysis and reducing the individual variations. In the validation cohort, in order to determine the diagnostic performance of candidate protein biomarkers, we enrolled another 104 subjects (18 patients with PC at stage I/II, 44 with PC at stage III/IV, 11 with PCLs, eight with well-differentiated P-NET, 13 with CP, and 10 with HC, respectively) from two hospitals. Plasma samples were prepared from EDTA-treated peripheral blood, centrifuged at 2,000×g and 10,000×g for 10 min (4°C) to remove dead cells and cell debris and stored at −80°C until subsequent analysis. The study was approved by the Ethics Committee of the Nanjing Drum Tower Hospital and the First Affiliated Hospital of Soochow University, and informed consent was obtained from all patients before the examinations.

Exosomes were isolated by ultracentrifugation in accordance with the guidelines proposed by the International Society for Extracellular Vesicles (ISEV) (12). To isolate exosomes, the plasma was thawed on ice and centrifuged at 10,000×g for 30 min and then filtered through a 0.22-um filter (MilliPore, USA). The supernatant was then ultracentrifuged at 110,000×g for 70 min. The pallet was washed in phosphate-buffered saline (PBS) and centrifuged again at 110,000×g for 70 min. The PBS was removed completely, and subsequently 60 μl ice-cold PBS was used to resuspend the exosomal pellet. All centrifugation steps were performed at 4°C.

General characterizations of exosomes were performed to demonstrate the purity of our exosome preparation. For images of exosomes at high resolution, about 5 μl of prepared exosomal suspension (the purified concentrated exosomes were mixed with an equal volume of 2% paraformaldehyde) was applied to the formvar-carbon coated grids, and the membranes were placed in a dry environment for 20 min. Then the grids were transferred to a 50 μl drop of 1% glutaraldehyde for 5 min and washed eight times with distilled waters. Samples were contrasted and embedded on ice with uranyl-oxalate for 5 min and methyl cellulose-uranyl acetate for 10 min, respectively. The excess fluid was blotted gently, and the grids were air dried for 10 min. The images were captured under the transmission electron microscopy (TEM, HT7800, Hitachi, Japan). Besides, the size distribution of exosomes was analyzed by high resolution flow cytometry using a Flow NanoAnalyzer Instrument (N30E, NanoFCM, China). The exosomal protein markers were validated by western blot analysis. We used the BCA protein assay kit (Thermo Fisher Scientific, USA) to quantify total protein concentration of exosomes according to the manufacturer’s protocol. The protein samples were denatured at 95°C for 10 min. Approximately 30 μg of protein was separated by 10% SDS-PAGE, subjected to the electrophoresis, and transferred to PVDF membranes. The membranes were then blocked with 5% skim milk for 2 h, incubated with anti-CD63 (Abcam, UK), anti-Tsg101 (Santa Cruz, TX), and anti-β-actin (Abcam, UK) overnight. Then protein bands were washed with tris-buffered saline with tween 20 (TBST) buffer three times and probed with secondary antibodies for 120 min.

Frozen exosomes were suspended on ice and 8 M urea was added. Then the samples were sonicated in ice for 2 min and incubated in ice for 30 min. We homogenized samples for 10 s every 5 min. The samples were clarified by centrifugation at 12,000×g at 4°C for 20 min. Protein was quantified with BCA protein assay kit (Thermo Fisher Scientific, USA). Approximately 30 μg of total protein was separated by 10% SDS-PAGE gel. Then separation gel was stained by CBB according to Candiano’s protocol. The stained gel was scanned by Image Scanner (GE Healthcare, USA) at the resolution of 300 dpi.

A 200 ug portion of protein from each sample was reduced and alkylated in 50 mM tris-(2-carboxyethyl) phosphine (TCEP) at 56°C for 1 h and 200 mM methyl methanethiosulfonate (MMTS) at room temperature for 1 h. The chilled acetone was then added in the protein mixtures for precipitation overnight. The solutions were centrifuged at 30,000×g for 30 min at 4°C and dissolved in 0.5 M tetraethyl-ammonium bromide (TEAB) and sonicated in ice. Subsequently, trypsin was added to the samples at a 1:20 (w/w) ratio, and the solutions were incubated for digestion at 37°C for 16 h. Finally, the collections of digested peptides were centrifuged at 12,000 rpm for 20 min.

For iTRAQ-labeling, the lyophilized samples were resuspended in 100 μl TEAB (200 mM) and then labeled by 8-plex iTRAQ reagent (SCIEX, USA) according to the instructions in the kit (PC at stage III/IV with 113 and 114 tags, PC at stage I/II with 115 tag, P-NET with 116 tag, HC with 117 tag, PCLs with 118 tag, CP with 119 tag). The labeled peptides were finally dried for further analysis.

The high pH reversed-phase liquid chromatography with a C-18 column was used to fractionate the labeled peptides. Briefly, mobile phases B (98% acetonitrile in 5 mM Ammonium formate) was used for gradient elution. The solvent gradient was set as follows: 0–25% B, 5–35 min; 25–45% B, 35–45 min. The peptides were separated at a fluent flow rate of 300 nl/min and finally lyophilized for mass spectrometry.

LC-MS/MS analysis was performed on a Nanoeasy system (Thermo Fisher Scientific, USA) with a 25 cm-long column (75 μm × 25 cm, 2 μm, 100 Å, C18 packing material, Thermo Fisher Scientific, USA) connected to a Q Exactive hybrid quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific, USA). The fractions were dissolved in loading buffer (2% acetonitrile with 0.1% FA) before analysis. Mobile phase B consisted of 98% acetonitrile and 0.1% formic acid. The elution gradient was set from 5 to 40%. Full MS scans were acquired in the mass range of 300–1200 m/z with a mass resolution of 70,000, and the AGC target value was set at 1e6. The ten most intense peaks in MS were fragmented with higher-energy collisional dissociation (HCD) with collision energy of 35. MS/MS spectra were obtained with a resolution of 17,500 with an AGC target of 1e5 and a max injection time of 22 ms. The Fusion dynamic exclusion was set for 60 s and run under positive mode.

The Proteome Discovery software (v.2.3) was used for protein identification and quantitation. All the raw data were searched against the Swiss-Prot human database (20,238 entries), with MS tolerance set at 20 ppm and MS/MS tolerance set at 0.1 Da. A global false discovery rate (FDR) was set to 0.01, and protein groups considered for quantification required at least two peptides.

We obtained the information of identified proteins from the UniProt protein sequence database. Both Gene Ontology (GO) analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were conducted for the function and pathway analysis of proteins by DAVID (https://david.ncifcrf.gov/) online tool. The STRING database was introduced for protein–protein interaction (PPI) network analysis (http://string-db.org).

The absolute expression levels of exosomal ALIX were tested using a commercial enzyme-linked immunosorbent assay (ELISA) kit (SenBeiJia Biotechnology Co., Ltd., China) based on the recommended manufacturer’s protocol. Firstly, the solubilized exosome samples were diluted five times with sample diluent, then 50 μl of the sample or standards was added to the appropriate wells in order. After the whole reaction, the optical density of each well was determined immediately using a microplate reader set to 450 nm. The concentration of exosomal ALIX was calculated in comparison with a protein standard curve.

All data were analyzed using the GraphPad Prism 6.0 statistical software (La Jolla, CA, USA) and SPSS 20.0 software (Chicago, IL, USA). The Mann–Whitney U test or Student’s t test was conducted to investigate the protein expression difference in all subjects. The expression levels of the selected biomarker candidate obtained by ELISA assays were compared using one-way ANOVA. The diagnostic value was evaluated by receiver operating characteristic (ROC) curve analysis. Differences were considered significant when P <0.05.

The study workflow was shown in Figure 1 . Firstly, the plasma-derived exosomes were isolated separately by ultracentrifugation. As shown in Figure 2 , the structure of exosomes revealed as a cup-shaped, membrane-enclosed vesicles, with a 30–150 nm size range characterized by TEM and flow cytometry ( Figures 2A, B ). The expression of exosomal biomarkers (CD63, Tsg101) was determined by western blotting using samples from six different groups of patients, which was shown in Figure 2C . And no significant difference in CD63 or Tsg101 intensity was observed among all groups of patients. Besides, exosomal biomarkers CD9, CD63, CD81, and Tsg101 were also all observed in the list of proteins identified through proteomic profiling ( Table S1 ). In summary, these results absolutely confirmed the purity of plasma-derived exosomes used in this study.

In the discovery stage, exosomal proteins were derived from all six groups and detected for iTRAQ-LC-MS/MS proteomic profiling analysis. A total of 623 proteins were identified ( Table S1 ), and 366 of them overlapped with the exosome database that was compiled in Vesiclepedia ( Figure 3A ). Moreover, based on the criteria for differentially expressed proteins (fold change >1.20 or fold change <0.83 with a relative quantification P value <0.05), 52 up-, 43 down- and six up-, 14 down-regulated proteins in exosomes were discovered from PC stage I/II and stage III/IV patients compared with healthy groups. Besides, 73, 105, and 248 proteins were found to be differentially expressed in exosomes from P-NET, PCLs, and CP groups versus HC group ( Figure 3B ). In addition, 22 proteins showed significant differential expression between stage I/II and stage III/IV PC patients. Apart from this, exosomes from P-NET, PCLs, and CP groups had differential expression of 33, 57, and 332 proteins in comparison with PC, respectively ( Figure 3C ). When we compared the identified proteins between PC and other pancreatic diseases groups, we noticed an overlap of 16 proteins, which were presented in Table 1 .

Bioinformatic analysis was performed to reveal the functional characterization of the differentially expressed proteins. GO annotations indicated that protein binding, cell part, and cellular process were the most enriched terms in the molecular function (MF), cellular component (CC), and biological process (BP) ( Figure 4A ). KEGG pathway analysis showed that many reported signaling pathways associated with tumorigenesis and metastasis in PC, including transcriptional mis-regulation in cancer, endocytosis, and NF-kappa B signaling pathway were all included in the top 20 enriched pathways ( Figure 4B ). Moreover, an intensity heatmap was also utilized to depict the proteins that were significantly differentially expressed among the six groups and identified in the Uniprot and Vesiclepedia database (150 proteins in total) ( Figure 4C ).

Among all differentially expressed proteins, there were a total of 16 proteins that overlapped between PC and other pancreatic diseases groups. Four proteins were significantly higher in PC compared with other pancreatic diseases, including PDCD6IP (also known as ALIX), GPRC5B, SDCBP, and IST1 ( Figure 5A ). To reveal the relationships of the 16 proteins, we reconstructed the interaction networks using the STRING protein–protein interaction (PPI) database. The STRING PPI analysis yielded a highly clustered network containing 14 nodes with 16 edges (clustering coefficient: 0.738, enrichment p-value < 0.001). As shown in Figure 5B , among the four upregulated proteins in PC group, three were interacted with each other (ALIX, SDCBP and IST1), and the fold change of ALIX expression in PC vs. other pancreatic diseases was higher compared with both SDCBP and IST1 ( Table 1 ). Based on the above results, we subsequently selected exosomal ALIX as the candidate biomarker for further analysis.

Late diagnosis and lack of effective treatments are the main reasons for poor prognosis of PC. Hence, we focused on investigating the diagnostic utility of exosomal ALIX as a novel biomarker for PC with plasma samples from patients with PC, other pancreatic diseases, and healthy controls. The expression level of exosomal ALIX was evaluated by ELISA, and ROC curve analysis was performed to assess the diagnostic value of ALIX. A total of 104 individuals were enrolled in this clinical cohort, including 62 patients with PC (stage I/II, n = 18; stage III/IV, n = 44), 11 patients with PCLs, eight patients with well-differentiated P-NET, 13 patients with CP, and 10 healthy subjects. At the same time, the general clinicopathological features were observed and summarized. We noticed that the exosomal ALIX expression was significantly higher in advanced PC than early PC ( Figure 6A ). Moreover, ALIX expression was significantly elevated in all PC patients compared with patients with other pancreatic diseases in total or HC ( Figure 6B ). Interestingly, we also found that the expression of exosomal ALIX in PC patients was higher compared with other pancreatic diseases, including PCLs (P = 0.0037), CP (P = 0.0043) and P-NET (P = 0.1132) separately, though there was no statistically significant difference between PC and P-NET ( Figure 6C ). Next, we analyzed the correlations between clinical characteristics and exosomal ALIX expression level in PC patients by univariate analysis, and the results showed that ALIX expression was significantly associated with TNM stage (I/II vs. III/IV) and distant metastasis (P < 0.05) ( Table 2 ). Then, we explored the value of exosomal ALIX alone or in combination with serum CA199 in differential diagnosis of pancreatic diseases ( Table 3 ). For one thing, ALIX presented an acceptable diagnostic efficacy with an AUC value of 0.730 between PC and other pancreatic diseases, which was a liter lower than CA199 (0.891). For another, when differentiating PC patients at stage I/II from stage III/IV, ALIX had a slightly higher AUC value (0.768) than CA199 (0.756). We further combined exosomal ALIX with CA199 to estimate their diagnostic performance in differentiating PC vs. other pancreatic diseases, as well as early vs. late PC. Consequently, the combination of these two markers for the diagnosis of PC versus other pancreatic diseases yielded a promising AUC of 0.910, with a sensitivity of 90.6% and specificity of 83.9% at the cut-off point ( Table 3 , Figure 6E ). They also improved the diagnostic efficacy between patients at stage I/II and stage III/IV with the AUC value of 0.872, which was significantly higher than CA199 alone ( Table 3 , Figure 6D ). Therefore, we suggested that exosomal ALIX had the capability to detect PC, and the combination of ALIX and CA199 can further improve the diagnostic value, especially in distinguishing early and late PC.