293T cells and human cervical cell lines HeLa and CaSki were obtained from the Cancer Center Laboratory of Shandong University. 293T and HeLa cells were cultured in DMEM (Gibco; Thermo Fisher Scientific, Inc.) and CaSki cells were cultured in RPMI-1640 medium (Gibco; Thermo Fisher Scientific, Inc.) supplemented with 10% fetal bovine serum (FBS,; Biological Industries). All cell lines were cultured in a humidified incubator under standard culture conditions (5% CO₂, 37°C).

Cells were seeded in 10-cm diameter plates and cultured in medium supplemented with 10% exosome-depleted FBS (System Biosciences). Culture medium was collected and exosomes were obtained using serial centrifugation. The medium was centrifuged at 300 × g for 10 min and 2,000 × g for 10 min to remove cells or other debris and 10,000 × g for 20 min to remove other larger vesicles. Then the medium was centrifuged at 110,000 × g for 70 min, washed with 5 ml phosphate-buffered solution (PBS) and centrifuged at 110,000 × g for another 70 min (all steps were performed at 4°C). Finally, the exosomes were obtained from the pellet and resuspended in PBS.

HeLa cells were seeded in a 10-cm diameter dish and cultured overnight, and exosome-free serum medium with or without 10 ng/ml TGF-β1 was used to continue the culture for another 24 h. The concentration of 10 ng/ml was much higher than the level of TGF-β1 secreted by tumor cells, thus, the impact produced by TGF-β1 secreted by tumor cells was negligible (21,22). The supernatant was collected to isolate exosomes and extract a total of 1 µg RNA as input material for RNA sample preparation. Total RNA was purified by electrophoretic separation on a 15% urea denaturing polyacrylamide gel electrophoresis (PAGE) gel and small RNA regions corresponding to the 18–30 nt bands in the marker lane (14–30 ssRNA ladder marker; Takara, Japan) were excised and recovered. Then the 18–30 nt small RNAs were ligated to a 5′-adaptor and a 3′-adaptor. The adapter-ligated small RNAs were subsequently transcribed into cDNA by SuperScript II Reverse Transcriptase (Invitrogen; Thermo Fisher Scientific, Inc.) and then several rounds of PCR amplification with PCR Primer Cocktail and PCR Mix were performed to enrich the cDNA fragments. The PCR products were selected by agarose gel electrophoresis with target fragments 100–120 bp, and then purified using the QIA Quick Gel Extraction Kit (Qiagen). The library was evaluated for quality and quantitated using two methods: Checking the distribution of the fragment size using the Agilent 2100 bioanalyzer (Agilent Technologies, Inc.), and quantify the library using real-time quantitative PCR (qPCR) (TaqMan Probe). The final ligation PCR products were sequenced using the BGISEQ-500 platform. The RNA-sequence procedure and data analysis were completed by BGI (The Beijing Genomics Institute, BGI, ShenZhen, China).

Isolated exosomes diluted in PBS were gently mixed, and the mixture was inserted into the ZetaView PMX120 instrument (Particle Metrix GmbH) at 23°C through a 1-ml syringe. ZetaView 8.02.28 software (Particle Metrix GmbH) was utilized to analyze the number and the size distribution of the exosomes. Each sample was measured three times and the average value was calculated.

Isolated exosomes fixed with 2% paraformaldehyde for 5 min were dropped onto the Formvar copper carbonate grid with glow discharge for 1 min, and then it was negatively stained with 2% uranyl acetate for another 1 min. After sample drying, a HT7800 electron microscope (Hitachi) was used to photograph the grid with an acceleration voltage of 80 kV.

PKH67 Green Fluorescent Cell Linker Kits (Sigma-Aldrich; Merck KGaA) were purchased to label the lipid bilayers of exosomes. PKH67 (4 µl) was added and exosomes were isolated into 1 ml diluent C separately and incubation was carried out for 5 min at room temperature; exosomes without PKH67 staining were used for negative control. Bovine serum albumin (1 ml/5%) (Solarbio) was added to halt the staining. Then the PKH67-labeled exosome mixture was centrifuged at 110,000 × g for 2 h at 4°C. The mixture was washed with PBS and centrifuged at 110,000 × g for another 2 h to obtain pure PKH67-labeled exosomes. After resuspending the mixture in complete medium and incubating with HeLa cells at 37°C for 0, 4, 8 and 12 h, a laser confocal microscope (LSM880; Zeiss) was used to visualize the incorporation of exosomes into HeLa cells.

Protein extracted from the cells and exosomes was lysed with RIPA buffer and quantified by BCA Protein Assay Kit (Beyotime Institute of Biotechnology). Total proteins were separated on a 10% polyacrylamide gel and transferred to a polyvinylidene fluoride membrane and blocked with 5% defatted milk for 1.5 h at room temperature. The membranes were probed with the following primary antibodies at 4°C: Anti-CD63 (dilution 1:1,000, Abcam, ab59479), anti-TSG101 (dilution 1:1,000; Abcam, ab125011), anti-HSP70 (dilution 1:2,000; ProteinTech Group, Inc., 66183-1-Ig), anti-GAPDH [dilution 1:1,000; Cell Signaling Technology, Inc. (CST), #5174], anti-MGAT3 (dilution 1:1,000; ABclone, A8134), anti-E-cadherin (dilution 1:1,000; CST, #3195), anti-N-cadherin (dilution 1:1,000; CST, #13116), and anti-β-catenin (dilution 1:1,000; CST, #8480). Subsequently, the membranes were washed with triethanolamine buffered saline solution (TBS) and interacted with HRP-conjugated antibody (dilution 1:1,000; CST) for 2 h at room temperature. The membranes were detected by ECL substrate kit (Thermo Fisher Scientific Inc.) and quantified by ImageJ software (v1.8.0; National Institutes of Health, USA).

Total RNA extracted from cells or exosomes was isolated with TRIzol reagent (Thermo Fisher Scientific Inc.) according to the manufacturer's instructions and the concentration was detected by a spectrophotometer (Thermo Fisher Scientific Inc.). Then the RNA was transcribed to cDNA by using Prime Script RT Master Mix for RT-PCR (Takara). qPCR was conducted using SYBR-Green (Takara) and performed on a StepOne™ PCR amplifier (Applied Biosystems). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) or U6 were used as internal controls. The primer sequences for miRNA and mRNA were purchased from GenePharma (Shanghai, China) and are as follows: miR-663b, forward (F), 5′-GGTGGCCCGGCCGTGC-3′ and reverse (R), 5′-TATCCTTGTTGACGACTCCTTGAC-3′; U6, F, 5′-CAGCACATATACTAAAATTGGAACG-3′ and R, 5′-ACGAATTTGCGTGTCATCC-3′; MG AT3, F, 5′-TCAACCACGAGTTCGACCTG-3′ and R, 5′-CACCTTGTGGCGGATGTACT-3′; GAPDH, F, 5′-GCACCGTCAAGGCTGAGAAC-3′ and R, 5′-TGGTGAAGACGCCAGTGGA-3′.

Cells were seeded in 6-well plates until the growth density reached 90%. Then a sterile 100-µl pipette tip was used to quickly stroke the cells to form a straight wound. After washing twice with PBS, 2 ml medium without serum was added to each well. The wound healing was observed and photographed by JEM-1200 EX II electron microscope (JEOL) at 0 and 24 h, and the wound closure was evaluated by ImageJ software (v.8.0; National Institutes of Health).

Cells (HeLa: 6×10⁴ cells/200 µl for invasion and 4×10⁴ cells/200 µl for migration, CaSki: 8×10⁴ cells/200 µl for invasion and 6×10⁴ cells/200 µl for migration) in 200 µl serum-free medium were seeded in upper Transwell chambers (8-µm pore size; Corning Costar) with or without Matrigel (60 µl, 1:9 dilution in serum-free medium, BD Biosciences). In addition, 600 µl media containing 10% FBS were added to the lower compartment. After 24 h of incubation in an incubator, the chambers were washed twice by PBS, and methanol was used to fix cells for 15 min. The cells that did not pass through the upper layer were wiped off with a cotton swab and then stained with crystal violet (product no. AC0121; Beyotime) for 30 min. A total of 3–5 random fields of view were selected to photograph under the JEM-1200 EX II electron microscope (JEOL).

PicTar (https://pictar.mdc-berlin.de/), TargetScan (http://www.targetscan.org/) and miRanda (http://www.microrna.org/microrna/home.do) were used to predict miR-663b targets. The expression of the miR-663b-target gene was verified in 293T cells by dual luciferase reporter gene assay. First, we insert the 3′-UTR of wild-type (WT) and mutant (MUT) MGAT3 into the psiCHECKTM-2 plasmid (Promega Corp.). Then, psiCHECKTM-2-MGAT3-3′-UTR-WT or psiCHECKTM-2-MGAT3-3′-UTR-MUT were co-transfected into 293T cells with miR-663b mimics or its NC. After 48 h of transfection in a 96-well plate, a Dual Fluorescence Luciferase reporter gene detection system (Promega Corp.) was used to detect the fluorescence.

Lipofectamine 2000 (Invitrogen; Thermo Fisher Scientific Inc.) and Opti-MEM (Gibco; Thermo Fisher Scientific Inc.) were used for cell transfection. miR-663b mimics and inhibitor were used for upregulation and downregulation of miR-663b. miR-663b mimics NC and inhibitor NC were non-targeting and used as the respective controls; the control (CON) group was without any special treatment. The overexpression of MGAT3 was produced by subcloning PCR-amplified full-length human MGAT3 cDNA into the pCDNA3.1 plasmid (GenePharma). The empty vector was used as a blank control (pCDNA3.1 NC). After transfection in Opti-MEM for 6 h, completed culture medium was replaced and the cells were cultured for the following experiments. Total RNA and total protein were isolated from the cells 24 or 48 h after transfection for qPCR analysis and western blot analysis. In order to obtain stably silenced cell lines for subsequent experiments, we used the lentiviral vector miR663b-sponge-pLVX-AcGFP-N1-Puro to transfect HeLa and CaSki cells and cultured the cells with puromycin dihydrochloride (2 µg/ml; Amresco) for 7–10 days. The intensity and ratio of green fluorescence produced by cells under the fluorescence microscope were used to prove the efficiency of lentiviral transfection.

Statistical analysis was conducted using GraphPad Prism 8 software (GraphPad Software, Inc.). The data are expressed as the median values ± standard deviation (SD). Comparisons between the groups were analyzed by Student's t-test (unpaired, two-tailed) or one-way analysis of variance (ANOVA) followed by Tukey's post hoc test. The results are presented as the average of three experiments. Statistical significance was set at P<0.05.

First, deep RNA-seq was performed to explore the expression level of miRNAs in HeLa cell exosomes after treatment with or without 10 ng/ml TGF-β1 for 24 h. The typical goblet morphology and the size range of 30–150 nm were detected by TEM (Fig. 1A). Next, positive signs of exosomes including CD63, tumor susceptibility gene 101 (TSG101), heat shock protein 70 (HSP70) and GAPDH were confirmed by western blot analysis (Fig. 1B). NTA proved that the concentration and size of the isolated exosomes were consistent with previous reports (Fig. 1C) (23). The green fluorescent signal in the HeLa cytoplasm indicated that the purified exosomes labeled with the fluorescent membrane tracer PKH67 (green) entered the HeLa cells after incubation (Fig. 1D). Deep RNA-seq results showed that 48 miRNAs were enriched in exosomes of HeLa cells after treatment with TGF-β1 compared with the non-treated cells (negative control) (Fig. 1E), proving that TGF-β1 affects the HeLa cell exosome miRNA profile. We selected top 10 miRNAs for PCR verification in HeLa and CaSki cells. After treatment of TGF-β1 for 24 h, miR-663b was selectively enriched 4 and 3 times in the exosomes of HeLa and CaSki cells compared with the negative control group (Fig. 1F and G). These results verified that miR-663b could be selectively enriched in the exosomes of cervical cancer cells treated with TGF-β1, thus we selected miR-663b for the subsequent study.

Previous research has confirmed that miR-663b plays tumor-promoting roles in endometrial cancer, nasopharyngeal carcinoma and osteosarcoma (24–26). We collected exosomes from the TGF-β1 treatment group (Exo-miR-663b) and the negative control group (Exo-NC) in HeLa and CaSki cells to verify whether exosomal miR-663b affects migration, invasion or proliferation of cervical cancer cells. In our study, HeLa and CaSki cells were treated with exosomes from their own cells. We found exosomes from the TGF-β1 treatment group significantly enhanced HeLa and CaSki cell motility by wound healing assay (Fig. 2A and B). In addition, exosomes upon TGF-β1 exposure enhanced HeLa and CaSki cell migration and invasion as measured by Transwell assays (Fig. 2C and D).

To further explore that exosomal miR-663b is indeed involved in cervical cancer cell metastasis, we used a lentiviral vector to knock down miR-663b (Exo-miR-663b-KD). The intensity and ratio of green fluorescence were used to prove the efficiency of lentiviral transfection and the images are provided in Fig. S1. After treatment with TGF-β1, we collected exosomes from the medium of the Exo-miR-663b-KD group and wild-type group (Exo-miR-663b) (Fig. 2E). qPCR verified that the amount of miR-663b in exosomes from the Exo-miR-663b group was much higher than that in the Exo-miR-663b-KD group both in HeLa and CaSki cell lines (Fig. 2F and G). Then we observed that compared with the Exo-miR-663b treatment group, the ability of migration and invasion of HeLa and CaSki cells was attenuated when the Exo-miR-663b-KD was incubated (Fig. 2H-K). Together, our results indicated that exosomes enriched in miR-663b after TGF-β1 treatment could be transferred into new target cells to promote the metastasis of cervical cancer.

Next, we used bioinformatic tools (TargetScan, PicTar, and miRanda) to explore the potential miR-663b target genes. We identified a 7-bp binding site between MGAT3 3′-UTR and miR-663b (Fig. 3A). MGAT3, as a key glycosyltransferase in the N-glycan biosynthetic pathway (27), is considered to be a metastasis suppressor gene (28), which regulates the glycosylation step and the key tumor-suppressor gene E-cadherin to inhibit tumor invasion and metastasis (29–31). To confirm that miR-663b could regulate the putative target of MGAT3, the predicted miR-663b binding site (wild-type) or mutant sequence (mutant type) in the 3′-UTR of MGAT3 was cloned into a luciferase reporter plasmid and we detected the response to miR-663b in 293T cells. Our results showed that the co-transfection of miR-663b mimics significantly reduced the luciferase activity of the wild-type 3′-UTR of MGAT3, while the luciferase activity of the mutant 3′-UTR MGAT3 had no change (Fig. 3B).

Moreover, we directly transfected HeLa and CaSki cells with miR-663b mimics, miR-663b inhibitor, miR-663b mimics NC and miR-663b inhibitor NC, respectively. Consistent with the results of dual luciferase reporter gene assay, compared with the NC group, miR-663b mimics resulted in decreased MGAT3 protein expression in HeLa and CaSki cells (Fig. 3C and D), while treatment with miR-663b inhibitor resulted in increased MGAT3 protein expression. Above all, our results demonstrated that exosomal miR-663b could inhibit MGAT3 expression.

To further confirm the role of MGAT3 in cervical cancer cells, we explored the function of exosomal miR-663b on cell migration under the vector-mediated MGAT3 gene overexpression conditions. Wound healing and Transwell assays were performed to show that miR-663b could enhance the migration and invasion ability of HeLa and CaSki cells, while co-transfection with pcDNA3.1-MGAT3 partially alleviated this effect caused by miR-663b (Fig. 4A-D). qPCR was used to verify the overexpression efficiency of MGAT3 plasmid transfected into HeLa and CaSki cells (Fig. 4E and F). These results indicated that exosomal miR-663b promoted the metastasis ability of cervical cancer cells by inhibiting the expression of MGAT3.

As the key role in cancer recurrence and exosome-mediated tumor progression, EMT is a biological process in which epithelial cells transform into cells with a mesenchymal phenotype and cause increased invasiveness and subsequent metastasis (18,20). To understand the potential molecular mechanism of MGAT3 in inhibiting cervical cancer invasion and metastasis, we detected the level of epithelial differentiation marker E-cadherin and mesenchymal marker N-cadherin and β-catenin in cervical cancer cells. Western blot analysis verified that after transfected with pcDNA3.1-MGAT3, the expression of E-cadherin was significantly increased in HeLa cells, while the expression of N-cadherin and β-catenin were significantly decreased. In contrast, after transfection with the miR-663b mimics, the expression of N-cadherin and β-catenin were significantly increased, while the expression of E-cadherin was significantly decreased. Moreover, all of these effects caused by miR-663b could be offset by the overexpression of MGAT3 in vitro (Fig. 5A). We also observed the same results in CaSki cells (Fig. 5B).

In order to further confirm that exosomal miR-663b promotes metastasis by inhibiting the expression of MGAT3 and affecting the EMT pathway, we evaluated the expression of MGAT3 and E-cadherin after the treatment of the cell lines with Exo-miR-663b and Exo-miR-663b-KD. Compared with the Exo-miR-663b-KD group, the expression of MGAT3 in the Exo-miR-663b group was significantly decreased, accompanied by the decreased expression of E-cadherin (Fig. 5C and D). The underlying mechanism of the TGF-β1/miR-663b/MGAT3 axis in cervical cancer is shown in Fig. 5E.