Trizol reagent (Invitrogen, USA) was used to extract the total RNA in NPC cell lines. Then the reverse transcription was continued referring to the protocol by High Capacity cDNA Reverse Transcription Kit (Biosystems). The cDNA was utilized for real-time PCR via Mx-3000P Quantitative PCR System (Stratagene). GAPDH and U6 were selected as control group. The following were the primer sequences for different genes: miR-125a-5p forward, 5′-GGTCA TTCCCTGAGACCCTTTAAC-3′; reverse, 5′-GTGCAGG GTCCGAGGT-3′. TAZ forward, 5′-ACCCACCCACGATGACCCCA-3′; reverse, 5′-GCACCCTAACCCCAGGCCAC-3′; GAPDH, forward, 5′-GGAGCCAAAAGGGTCATCAT-3′; reverse, 5′-GTGATGGCATGGACTGTGGT-3′. U6 forward, 5′-CTCGCTTCGGCAGCACA-3′; reverse 5′-AACGCTTCACGATTTGCGT-3′. The relative mRNA level was normalized by the expression of GAPDH, and the relative miR level normalized by expression of U6. The independent experiments were performed 3 times, each of which used 3 samples independently.

The TAZ gene was queried by the miRNA target prediction sites TargetScan (Human) (http://www.targetscan.org/), miRbase (http://www.mirbase.org/) and miRDB (http://mirdb.org/miRDB). The scores and strength of binding sites for predicted target were detected. The miRNAs overlapping in more than 2 searches were selected for further studies.

The clinical information and also microRNA expression levels for clinical samples in NPC and HNSCC were acquired from the GEO (Gene Expression Omnibus) database (NCBI/GEO/GSE32960) and TCGA database (https://tcga-data.nci.nih.gov/tcga/). All the data were analyzed with GraphPad Prism.

NPC cell lines such as HONE-1, 6-10B, and 5-8F were maintained in our lab and cultured in RPMI 1640 medium (GIBCO) with 10% FBS, supplemented with 100 U/ml penicillin and 100 mg/ml streptomycin. The keratinocyte serum-free medium (KSFM, Invitrogen, Carlsbad, CA, USA) with human recombinant epidermal growth factor (rEGF) was used to culture the NP-69 cells and NP-460 cells which were immortalized nasopharyngeal epithelial cell line. The 293 T cells were maintained in our lab and were cultured for luciferase reporter assay.

MSC were cultured and the exosomes were isolated from the conditioned medium using Exosome isolation Kit—Serum and Plasma (Exiqon) according to protocol, the procedures of which were performed in previous studies (22). The electron microscopy and nanoparticle tracking technology (Nanosight™) were used for identification for exosomes.

The pGL3-TAZ and pGL3-mutTAZ reporter genes were both established by Bio-Asia (China). The luciferase reporters and the miR-125a mimics were both transfected in 293T cells. The activity of the reporter protein was detected by a luciferase assay kit (Promega) after transfected for 48 h with reference to the instructions.

The paraffin-embedded tissues were sliced and heated for further investigation. After deparaffinized, rehydrated and antigen retrieving, the slides were blocked with goat serum and incubated with primary antibody (Abcam) at 25°C for 4 h. After the usage of a secondary antibody and DAB (3,3′-diaminobenzidine) substrate solution, the picture were captured. Matrix-associated vascular channels including the vasculogenic mimicry structure were detected by PAS staining. Images for all the slides were captured by Leica DM 2500 microscope.

Matrigel were coated in 96 wells plates (50 μl/well), and NPC cells (6-10B and 5-8F) were seeded in the wells (3 × 10⁴ cells per well) and cultured for 18 h by RPMI 1640 medium without FBS. Images were took after 36 h using Leica DM 2500 microscope randomly.

The miR-125a mimics, miR-125a inhibitors and corresponding negative control were purchased from Genepharma (Shanghai, China). Lentiviruses expressing miR-125a, inhibiting miR-125a and the scrambled control were constructed by Genechem (Shanghai, China). The transfecting with lentivirus were performed at MOI = 10 in the NPC cells and at MOI = 10 in the human mesenchymal stem cells.

NPC cells were seeded in the top chamber cultured by serum-free media. The bottom chamber was filled with RPMI 1640 medium containing 15% FBS. After 36 h (6–10 B) or 24 h (5–8 F) of incubation, the cells were simply removed and fixed with 4% paraformaldehyde in the top chamber for 20 min, and then stained with giemsa solution for 20 min. Five fields per well were photographed at random.

A total of six BALB/c nude mice (male, 4–5 weeks old) were purchased in the SLAC Laboratory Animals (Shanghai), and were randomly divided into 2 groups. After disinfection, stably transfected cells (1 × 107 cells/ml, 0.2 ml of cell suspension) were subcutaneously injected at the left groin. Tumor growth was detected, and the diameters of long (a) and short (b) for each tumor were measured every 3 days. Tumor volumes were calculated as V = 0.5 × a × b² and then the tumor growth curves were generated.

The cultured cells were harvested using RIPA cell lysis buffer. All the protein lysates in different groups were loaded and separated by SDS-PAGE, and the SDS PAGE gel were transferred to a polyvinylidene difluoride (PVDF) membrane. The blots were incubated with primary antibodies against TAZ (Proteintech; China), LAMB2 (Proteintech; China), N-cadherin (Cell Signaling Technology; USA), E-cadherin (Cell Signaling Technology; USA), MMP2 (Cell Signaling Technology; USA), MMP9 (Cell Signaling Technology; USA), and GAPDH (Abcam; UK). The enhanced chemiluminescence (ECL, Millipore, USA) was utilized for visualizing the protein bands. The intensity of the protein bands was analysed via ImageJ 1.52a software (National Institutes of Health, USA) and normalized to the bands of GAPDH.

NPC cells were cultured in RPMI 1640 medium containing 15% FBS in 6-well plates at 37 °C for 24 h. Wounds were generated by scratching with 200 ul sterile pipet tips, and washed 3 times with phosphate-buffered saline (PBS). After wounds were made, cells were cultured in RPMI 1640 medium containing 1% FBS. Then images were captured at the time 0 h and 36 h after wounding.

All statistical data were analyzed using GraphPad Prism 7.00 software. Student’s t-test was used for calculating significance between two groups, and one-way ANOVA was used for multiple comparisons. Kaplan–Meier survival curves were also analyzed via log-rank tests using GraphPad Prism 7.00 software. All tests were two-sided, and p-values <0.05 were considered statistically significant. All data are presented as the mean ± standard error of the mean. Bar graphs represent mean and SD or mean and SEM across three independent experiments.

To identify miRNA candidates involved in NPC initiation and progression, we analyzed and compared microRNA expression levels between NPC tissue and normal nasopharynx tissue using the Gene Expression Omnibus (GEO) database (NCBI/GEO/GSE32960). Heatmap (23) ( Figures 1A, B ) and volcano plot ( Figure 1C ) visualization of gene expression were constructed, and we found that miR-125a significantly decreased in NPC tissue samples compared to normal nasopharynx samples. An absolute fold-change cut-off of ≥1.5 was used for statistical analysis. Indeed, miR-125a was significantly downregulated in 79.10% NPC patients (n = 311). In the rest of NPC patients, miR-125a expression level also exhibited a tendency of declination, although the differences failed to reach the statistical significance ( Figures 1D, E ). In addition, patients with high expression of mir-125a had higher survival rate as well as better prognosis ( Figure 1F ). The TCGA database validated the downregulation of mir-125a in HNSCC ( Figure 1G ). Moreover, the normal nasopharynx cell lines displayed the higher miR-125a expression than NPC cell lines ( Figure 1H ).

To sum up, these results indicated that the low level of miR-125a is related to development of NPC cells.

We showed VM structures in NPC tissue by double staining with PAS and CD34, in line with previous studies in other neoplasms. VM structures could be observed in NPC tissues, which were negative for vascular endothelial marker CD34 but positive for PAS ( Figure 2A ). We also found that the high graded of NPC stage is correlated with the higher level of VM structure, indicating that VM structures are related to advanced-stage and poor prognostics ( Figure 2A ). Then we conducted VM formation assay and found that over-expression of miR-125a hampered VM-forming abilities ( Figure 2B ). In contrary, inhibition of miR-125a promoted the formation of VM in NPC cells ( Figure 2B ).

As the transcriptional signature of VM is associated deeply to migration ability and Epithelial-to-Mesenchymal Transition (EMT) (24), we therefore detected the invasion and migration characteristics ( Figures 2C, D ) and EMT-related protein level ( Figure 2E ) of NPC cells. We found that upregulation of miR-125a decreased the invasion and migration ability of NPC cells ( Figures 2C, D ). Furthermore, results for western blot assays also showed that after overexpression of miR-125a, the expression level of mesenchymal marker N-cadherin was significantly downregulated, whereas a moderate elevation was observed in E-cadherin, which is a well-defined epithelial marker, indicating that miR-125a was a suppressor for EMT in NPC cells. Accordingly, migration marker MMP9 and VM marker LAMB2 was simultaneously decreased after miR-125a transfection, suggesting the inhibitory role of miR-125a on NPC migration and VM formation ( Figure 2E ). These results demonstrated that low level of miR-125a triggered formation of VM and may play a vital role in tumor progression in NPCs.

To determine the target of miR-125a in NPC, we firstly detected the differentially expressed genes using TargetScan (http://www.targetscan.org/), miRwalk (http://mirwalk.umm.uni-heidelberg.de/), miRDB (http://mirdb.org/) and the TCGA database for prediction, and 25 genes were selected ( Figure 3A ).

We subsequently identified the candidate target-genes according to the following criteria: (1) metastasis-related genes in NPC based on LinkedOmics (http://www.linkedomics.org/) (25); and (2) had the definite function of tumor suppressor based on literature review. Finally we selected TAZ to be a potential target of miR-125a. TAZ (tafazzin) is negatively associated to adherens junction and cell adhension moleculars according to the Gene Set Enrichment Analysis ( Figure 3B ), which indicated that TAZ has the positive correlation with metastasis.

Kaplan–Meier survival curves in HNSCC were constructed using the TCGA clinical data, and low TAZ level were prognostic of better overall survival in HNSCC patients ( Figure 3C ). TAZ activation is widespread in many human tumors, and closely related to EMT progression (26). The luciferase activity in miR-125a-transfected 293T cells decreased significantly, which demonstrated that miR-125ap directly targeted TAZ ( Figures 3D, E ).

Furthermore, the effects of the knockdown of TAZ resembled the effects of miR-125a overexpression, which decreased the protein levels of E-cadherin, LAMB, MMP9, and N-cadherin ( Figure 3F ).

To verify that TAZ plays a crucial role along the tumor-suppressive biological progression of miR-125a, we performed a rescue experiment by over-expressing both miR-125a and TAZ in NPC cells.

After rescuing the expression of TAZ, the number of VM tunnel reduced compared to that in group of TAZ-high expression ( Figure 4A ). To further confirm the opposing effects of miR-125a and TAZ in NPC cells, we conducted transwell assay ( Figure 4B ) and wound healing assay ( Figure 4C ). The results indicated that invasion and migration showed a significant reduction when co-transfecting miR-125a and TAZ in NPC cells.

Altogether, these results show that TAZ has pro-metastasis roles, which expression was directly suppressed by miR-125a.

As the MSCs are known to package miRs into exosomes (18), we hypothesized that after transducting miR-125a in MSC, MSCs would package miR-125a into exosomes and release these exosomes (EXO-miR-125a), which is of therapeutic importance.

To validate the hypothesis, MSCs was transfected with miR-125a or scramble vector, and then exosomes were harvested from the medium and detected by electron microscopy ( Figure 5A ). Western blot verified further that the nanoparticles contained the exosome markers TSG101 and CD9 but absent for endoplasmic membrane marker calnexin ( Figure 5B ). The expression level of miR-125a in MSC exosomes was confirmed by PCR, and the level of miR-125a was increased after transfection ( Figure 5C ).

Transwell assays and wounding assays were performed for NPC cells pre-treated with exosomes for 48 h. The results indicated that EXO-miR-125a decreased migration of NPC cells ( Figures 5D, E ). The VM formation capacity was reduced by miR-125a-transfected MSC exosomes ( Figure 5F ), which is in accordance with the VM inhibitory capability of miR-125a. Furthermore, the results of western blot showed that EXO-miR-125a decreased the invasion markers (N-cadherin and MMP9) and VM formation marker LAMB2 in 5–8F and 6–10B cell lines ( Figure 5G ).

To extend our findings in vivo, we established the mouse subcutaneous xenograft model of 6–10B cells ( Figure 6A ). Xenografted mice were randomized to receive EXO-NC or EXO-miR-125a treatment.

The tumor burden of mice in the EXO-miR-125a group was much smaller than that in EXO-NC group ( Figure 6B ), and similarly, the solid tumor weight taken at day of sacrifice was lower ( Figure 6C ).

Moreover, CD34/PAS staining indicated that EXO-miR-125a treatment hampered the VM formation abilities ( Figure 6D ), and TUNEL assay demonstrated that EXO-miR-125a enhanced the level of apoptosis of tumor cells. These results suggesting that exosomal miR-125a, plays a role in the anti-tumor efficacy in vivo.