A total of 92 PDAC patients were included. Patient characteristics and tumor features are shown in Table S5. All patients were treatment-naive before enrollment. Eighty patients received surgery including 7 with liver metastasis. Adjuvant therapy including chemotherapy and radiotherapy were performed in 77.5% (62/80) patients receiving surgery. The 1- and 3-yr recurrence-free survival was 52.8% and 14.5%, respectively. The 1- and 3-yr patient overall survival was 59.7% and 15.9%, respectively.
Serum exo-miRNA profiles of PDAC
We stratified PDAC patients into early stage PDAC group (stage I-IIA), lymph node metastatic PDAC group (stage IIB-III) and liver metastatic PDAC group (stage IV). We collected serum samples from 12 PDAC patients (4 from each subgroup) and 4 healthy subjects to detect the exo-miRNAs using a sequencing. We compared the serum exo-miRNA profiles between different PDAC subgroups and control group (Table S6). We observed that all 3 PDAC subgroups displayed 2 significantly up-regulated miRNAs and 9 down-regulated miRNAs using a selection criteria of fold change > 2, P < 0.05 and the largest rank sum differences (Fig. 1A). Interestingly, we noted that among the 9 down-regulated miRNAs, 8 belonged to chromosome 19 microRNA cluster (C19MC) and 6 shared the same sequence (www.mirbase.org). In addition, miR-519a-5p and miR-522-5p were found to be located at the shortest subset of C19MC (length of 1270bp) named as miR-519a/522-5p.
To verify the results, we enrolled the remaining 80 PDAC patients and 40 age- (± 5 yr) and gender-matched controls (2:1). There was no significant difference in co-morbidities including hypertension, diabetes, and dyslipidemia between PDAC patients and control subjects (Table S7). We detected the PDAC-associated serum exo-miRNAs from the two groups using qRT-PCR. Finally, significantly increased serum exo-let-7g-3p and miR-490-5p and decreased miR-519a/522-5p were found in PDAC group as compared to the control group (Fig. 1B). Whereas the expression of other miRNAs had no significant differences (Fig. S1A).
Serum exo-miRNAs as biomarkers in the diagnosis of PDAC
To evaluate the diagnostic performance of the above 3 serum exo-miRNAs, we performed AUC analysis and observed that all 3 miRNAs showed AUCs > 0.75 for distinguishing 80 PDAC patients from 40 controls. There was no significant difference between the 3 miRNAs and the typical biomarker CA19-9. In contrast, CEA presented the lowest AUC (all P < 0.05, Fig. 1C).
To further establish a valid serum biomarker panel, we randomly (1:1) divided the cohort into a training set and a validation set. We next selected the cut-off value of serum exo-miRNAs by considering both diagnostic sensitivity and specificity. CA19-9 positive (> 37 U/mL) rate was 70% (28/40) in PDAC patients. We then established biomarker panels by integrating serum exo-miRNAs with CA19-9 in the training group. The panels displayed significantly enlarged AUCs in the diagnosis of PDAC than CA19-9 alone in both training and validation sets (Fig. 1D). The diagnostic sensitivity, specificity and accuracy are shown in Table S8. Approximate 70% of PDAC patients with negative CA19-9 could be diagnosed via the supplement of each of the 3 exo-miRNAs. Moreover, the diagnostic accuracy would be perfect after the combination of CA19-9 and all 3 exo-miRNAs with an AUC value of 0.99.
Serum exo-miR-519a/522-5p is associated with patient prognosis and tumor features
Survival analysis were conducted to assess the prognostic value of serum exo-miRNAs. We observed that miR-519a/522-5p could differentiate patients with high and low recurrence/death risk (Fig. 1E), whereas let-7g-3p and miR-490-5p could not (Fig. S1B). In univariate COX hazard analysis, AJCC stage III-IV, histological grade G3, vessel invasion, lymph node metastasis, liver metastasis and low expression of serum exo-miR-519a/522-5p were the risk factors for tumor recurrence and patient death (Table S9). In multivariate analysis, serum exo-miR-519a/522-5p was found to be the independent factor for both tumor recurrence and patient death (Fig. 1F).
We next analyzed the correlations between the expression of serum exo-miR-519a/522-5p and tumor features. There was a significant positive correlation between low serum exo-miR-519a/522-3p expression and lymph node/liver metastasis, neural invasion, adjacent tissue invasion, and histological grade G3 (Table S10).
Low miR-519a/522-5p level in serum exosomes links to the accumulation of miR-519a/522-3p in cancer
We detected the expression of miR-519a/522-5p in PDAC tissue and paired adjacent non-cancerous tissue using qRT-PCR (n = 32). There was a significantly higher expressed miR-519a/522-5p in PDAC tissue compared with non-cancerous tissue (P = 0.002, Fig. 2A). Then we correlated the tumor miR-519a/522-5p level with matched serum exo-miR-519a/522-5p level and found a significant negative correlation (P < 0.001, Fig. 2B).
To assess the exosome transportation of miR-519a/522-5p in PDAC cells, we examined the expression of miR-519a/522-5p in various PDAC cell lines (PANC-1, BxPC-3, MIA PaCa-2, AsPC-1 and CFPAC-1), pancreatic epithelioid cells (CCC-HPE-2) and the extracellular exosomes in the culture medium. Compared to pancreatic epithelioid cells, PDAC cells displayed significantly increased cellular miR-519a/522-5p expression (Fig. 2C) and decreased medium miR-519a/522-5p level (Fig. 2D). There was a negative correlation between cellular and medium miR-519a/522-5p levels (Fig. 2E). In addition, we compared the miR-519a/522-5p levels between PDAC cells with different malignant behaviors, based on the current understanding of the characteristics of PDAC cell lines . PDAC cell lines with high metastatic activity (AsPC-1 and CFPAC-1) presented higher cellular miR-519a/522-5p expression and lower medium miR-519a/522-5p level than those with low metastatic activity (PANC-1, MIA PaCa-2 and BxPC-3) . However, PDAC cell lines with different differentiation and proliferation exhibited no prominent difference (Fig. 2F) .
miR-519a/522-5p promotes invasiveness/metastasis of PDAC in vitro and in vivo
MiRNA mimics and inhibitors were used to evaluate the function of miR-519a/522-5p in PDAC cell lines (BxPC-3 and PANC-1), the transfection efficiency was assessed by qRT-PCR (Fig. S2A). There was no statistical significance in the relative cell survival and apoptotic rates between miR-519a/522-5p mimics/inhibitors group and the respective control group (Fig. S2B-C). In contrast, the invasion rate of the miR-519a/522-5p mimics group were significantly higher than that of the control group, whereas the miR-519a/522-5p inhibitors-transfected cells showed opposite result (Fig. 3A).
Subsequently, we constructed a stable miR-519a/522-5p-overexpressing PANC-1 cells (Fig. S2D), and established an orthotopic PDAC model in nude mice. Six weeks later, luciferase imaging system was applied to evaluate the metastasis of tumors and then the mice were sacrificed and autopsied (Fig. S2E). Compared with the control group, the overexpression of miR-519a/522-5p induced more liver metastasis (5/7 vs.1/7) (Fig. 3B-D). The further H&E and immunohistochemical analysis showed the overexpression of miR-519a/522-5p induced nodular metastatic foci in the liver (Fig. 3E).
miR-519a/522-5p could be transferred among PDAC cells via exosomes and promote invasiveness of recipient cells
To further investigate the extracellular transportation of exo-miRNA, we co-cultured untreated PANC-1 cells with the PANC-1 cells that transiently transfected with the Cy3-tagged miR-519a/522-5p for 48h. The fluorescently labeled miR-519a/522-5p was observed in the untreated PANC-1 cells through confocal microscopy. Moreover, when we added exosomes inhibitor (GW4869) into the co-culture system, the fluorescently labeled miR-519a/522-5p in the untreated PANC-1 cells was sharply weakened (Fig. 4A).
We next isolated exosomes from the serum-free medium of miR-519a/522-5p mimics, inhibitors, and control groups for further investigating whether exo-miR-519a/522-5p promotes invasiveness of recipient cells. The expression of exo-miR-519a/522-5p was assessed by qRT-PCR (Fig. S3A). The purified exosomes were co-cultured with untreated PANC-1 cells in transwell chamber for 48h. Compared with the control, increased cell invasion was observed when exo-miR-519a/522-5p mimics incubated with PANC-1 cells. In contrast, significant repression of cell invasion was found in PANC-1 cells co-cultured with exo-miR-519a/522-5p inhibitors (Fig. 4B). The uptake of exo-miR-519a/522-5p by PANC-1 cells was verified by qRT-PCR (Fig. S3B).
To improve the transfection efficiency, miR-519a/522-5p mimics were encapsulated into PANC-1-derived exosomes via electroporation. The encapsulation efficiency was approximately 20%. The transfection efficiency was greatly improved as compared to transient transfection and heavily miR-519a/522-5p loaded exosomes (elet-exo-miR-519a/522-5p) were obtained (Fig. S3C). The elet-exo-miR-519a/522-5p were co-cultured with untreated PANC-1 cells in transwell chamber for 48h. The uptake of elet-exo-miR-519a/522-5p by PANC-1 cells was verified by qRT-PCR and confocal microscopy (Fig. S3D-E). Compared with the control, the PANC-1 cells co-cultured with elect-exo-miR-519a/522-5p showed much stronger invasive potential (Fig. 4C).
miR-519a/522-5p-associated transcriptomics and metabolomics profiles in PDAC cells
We compared the differential transcriptome profiles between PANC-1 cells and stable PANC-1- miR-519a/522-5p cells (Fig. S4A). KEGG analysis revealed the significant enrichment of differential genes in cancer and metabolism pathways including proteoglycans in cancer, miRNAs in cancer, focal adhesion, adherents junction, central carbon metabolism in cancer and glycolysis/gluconeogenesis (Fig. S4B). The metabolomic analysis of the same specimens was performed to trace downstream variations of the transcriptomics. PCA plot (Fig. S4C) and PLS-DA plot (Fig. S4D) showed good sample segregation. The superclass distribution of metabolic features and differential metabolic features are shown in Fig. S4E-F. Mummichog analysis using MetaboAnalyst 4.0 showed that the metabolic features induced by miR-519a/522-5p overexpression were associated with arachidonic acid metabolism, amino sugar and nucleotide sugar metabolism, purine metabolism, 2-Oxocarboxylic acid metabolism, linoleic acid metabolism, galactose metabolism, fructose and mannose metabolism, pyrimidine metabolism, central carbon metabolism in cancer and glycolysis/gluconeogenesis metabolic pathways (Fig. S4G).
Cross-omics analysis reveals that miR-519a/522-5p drives glucose metabolic reprogramming in PDAC cells
We performed cross-omics analysis by simply overlapping the enriched pathways from metabolomics and transcriptomics and observed two common pathways as central carbon metabolism in cancer and glycolysis/gluconeogenesis. It may indicate a potential key role of glucose metabolism in miR-519a/522-5p-mediated molecules network.
Next, we calculated the differential abundance scores of metabolites and genes, which showed the tendency of them in certain pathway to be increased or decreased as compared to control group  (Fig. 5A). Among 53 metabolic pathways, 21 displayed increased levels of both genes and metabolites (right upper quadrant), indicating a consistency in metabolomics and transcriptomics data. Out of the 21 metabolic pathways, 8 were glycometabolism pathways including carbon metabolism, central carbon metabolism in cancer and glycolysis/gluconeogenesis. To concurrently visualize the regulation of both gene (right) and metabolite (left) levels in certain metabolic pathway, we performed Metabolograms analysis , which further proved the correlation between metabolomics and transcriptomics data (Fig. 5B).
The above analysis suggested an enhanced Warburg effect (i.e., increased glycolysis) after overexpression of miR-519a/522-5p in PDAC cells. Therefore, we further mapped the associated metabolites and genes to show the miR-519a/522-5p-induced metabolic reprogramming (Fig. 5C). Metabolites in glycolysis including G6P, F6P and PYR showed > 5-fold increase in abundance and genes encoding for the relevant enzymes including HK, PFK-1, ALDO, GAPDH, PGK, PGM, ENO, PKM and LDH were significantly increased. Genes encoding glucose and glutamine transporter (SLC2A1 and SLC1A5) were also significantly upregulated.
Further studies were conducted to validate whether miR-519a/522-5p could modulate glycolysis in PDAC cells. The ECAR increased as expected in both PANC-1 and BxPC-3 cells transfected with miR-519a/522-5p mimics (Fig. 5D), especially from 20 min to 60 min, which implied an elevated lactic acid formation during aerobic glycolysis.
miR-519a/522-5p directly targets SESN2 and promotes invasiveness of PDAC cells via enhancing Warburg effect
To further identify the specific pathways that directly targeted by miR-519a/522-5p, we firstly searched Targetscan (http://targetscan,org/), miRWalk (http://mirwalk.umm.uniheidelberg.de/) and miRDB (http://mirdb.org/miRDB/) databases. The potential target genes were overlapped with the down-regulated genes in our transcriptome dataset. The Venn diagram showed the overlap of 97 genes across the four gene sets (Fig. 6A). Then we integrated the glucose metabolic related gene sets from The Molecular Signatures Database (MSigDB, http://www.broadinstitute.org/gsea/msigdb) and three candidate genes were screened out (Fig. 6B). Western blotting and qRT-PCR results showed that overexpression of miR-519a/522-5p dramatically suppress SESN2 expression in PANC-1 cells (Fig. 6C-D), rather than ADIPOR2 and RNABP2 (Fig. S5A-B). Furthermore, the luciferase reporter assay indicated that overexpression of miR-519a/522-5p significantly inhibited luciferase activity in 293T cells expressing wild type SESN2 reporters, whereas the mutant abolished this effect (Fig. 6E). The above data suggested that miR-519a/522-5p may target SESN2 directly.
Furthermore, we used lentivirus expressing Flag-SESN2 to upregulate SESN2 expression in both PANC-1 cells and stable PANC-1-miR-519a/522-5p cells. The transfection efficiency was demonstrated by Western blotting (Fig. S5C). Transwell and ECAR assay showed that overexpression of SESN2 could reverse the effects of miR-519a/522-5p induced high invasion and glycolysis in PANC-1 cells (Fig. 6F-G).