HCC tissues were collected from the Eastern Hepatobiliary Surgery Hospital (Shanghai, China) with informed consent obtained from the 36 patients for PDC culture and further research from August 2013 to June 2015. These patients with a median age of 49 years (range, 37–74) consisted of 30 males and 6 females (Table I). PDCs were established and cultured by 3D Medicine Inc. (Shanghai, China) using standard procedures. In brief, fresh tissues from partial tumor of HCC patients were washed with Dulbeccos modified Eagles medium (DMEM)/F12 medium (Gibco; Thermo Fisher Scientific, Inc., Waltham, MA, USA) to remove excess blood. The tissues were minced into 2-mm pieces and incubated with DMEM/F12 medium supplemented with 5% fetal bovine serum (FBS; Gibco; Thermo Fisher Scientific, Inc.). When the cells reached 80% confluence, they were trypsinized and prepared for subculture. Subsequently, the medium was changed every 3 days.

Cells in the logarithmic growth phase were seeded on 6-well plates (5×10⁵ cells/well) and incubated for 24 h to obtain a 90% confluent monolayer. The wound healing assay was performed by creating scratches using 200-µl pipette tips, followed by gentle washing three times with phosphate-buffered saline (PBS). Cells were cultivated in serum-free medium for 24 h, and the images of the wound gaps were captured at 0, 6 and 24 h after the wound was made. Wound healing ability was determined by measuring the change in the scraped area using Image-Pro Plus 6.0 software (Media Cybernetics, Inc., Rockville, MD, USA). Each experiment was repeated three times.

Exosomes were collected from PDCs during passages 10–15. Briefly, when the cells reached 80% confluence, the culture medium was replaced with fresh serum-free DMEM/F12, and the cells were cultured for an additional 48 h. Subsequently, the supernatants were centrifuged at 1,000 × g for 5 min and 4,000 × g for 5 min to remove cellular debris, and then collected in 50-ml tubes for storage at −80°C. Exosomes were isolated using a 3D Medicine exosome isolation kit (CFDA license no. Hu min xie bei 20170019). Briefly, supernatants were brought to room temperature, passed through a 0.45-µm filter, and then a 0.22-µm filter. 3D-TC reagent was added to the supernatants at a 1:2 ratio and mixed by inverting the tubes several times. The mixture was incubated overnight at 4°C, and centrifuged at 4,700 × g for 30 min at 4°C to obtain the precipitated exosomes. The isolated exosomes were resuspended in 200 µl PBS.

Exosomes were fixed with 2% glutaraldehyde (Sigma-Aldrich; Merck KGaA, Darmstadt, Germany). One drop of the sample was loaded onto Formvar^® coated copper grids (Agar Scientific, Ltd., Stansted, UK), and was dried at room temperature for 10 min. The grids were washed with ultrapure water three times and negatively stained with 2.5% uranyl acetate (pH 7.0; SPI-Chem; Structure Probe, Inc., West Chester, PA, USA), followed by methyl cellulose/uranyl acetate (pH 4.0; Sigma-Aldrich; Merck KGaA). Exosome particles were visualized using an H-600 electron microscope (Hitachi, Ltd., Tokyo, Japan) at 100 kV accelerating voltage.

Cellular protein was extracted from PDCs using radioimmunoprecipitation assay lysis buffer (Thermo Fisher Scientific, Inc.). Exosomal protein was extracted from exosomes using 3D Medicine protein lysis buffer. The protein lysate was then centrifuged at 4°C for 15 min, and the supernatant was quantified using a BCA kit (Pierce; Thermo Fisher Scientific, Inc.). Protein (70 µg/well) was electrophoresed by SDS-PAGE on 10% gels and transferred to a polyvinylidene difluoride membrane (EMD Millipore, Billerica, MA, USA). The membranes were blocked using 3% bovine serum albumin (BSA; Thermo Fisher Scientific, Inc.) at 4°C for 12 h and then incubated with CD9 antibody (diluted at 1:1,000; cat. no. sc-13118; Santa Cruz Biotechnology, Inc., Dallas, TX, USA). Horseradish peroxidase-conjugated anti-mouse IgG was used as the secondary antibody (diluted at 1:5,000; cat. no. sc-2380; Santa Cruz Biotechnology, Inc.). Bands were visualized using enhanced chemiluminescence (Thermo Fisher Scientific, Inc.). Comparison of the grey signal intensity bands on the western blots were performed using software Adobe Photoshop CS6. Each experiment was repeated three times.

RNA from exosomes was isolated using miRNeasy Serum/Plasma kit (Qiagen, Inc., Valencia, CA, USA) according to the manufacturers protocol. Briefly, 700 µl QIAzol was added to 200 µl exosomes, vortexed and incubated. Then 90 µl chloroform was added to the mixture for further vortexing and incubation, and the mixture was centrifuged at 4°C for 15 min. The upper aqueous phase was transferred to a new tube, and two volumes of 100% ethanol were added. Following vortexing, the solution was loaded to a spin column and centrifuged at 8,000 × g for 15 sec. The flow-through was discarded, and the spin column was rinsed twice with wash buffer, air-dried and the recovered exosomal RNA was eluted using 15 µl RNase-free water. Total RNA from PDCs was isolated using TRIzol (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturers protocol. RNA quality and concentration were analyzed using the NanoDrop ND-100 Spectrophotometer (Thermo Fisher Scientific, Inc., Wilmington, DE, USA).

Small RNA library preparation for high throughput sequencing was performed using the protocols and reagents from the NEBNext Small RNA Library Prep Set for Illumina (New England BioLabs, Inc., Ipswich, MA, USA). The small RNA library prepared from PDCs was fractionated by DNA length on a 2% TAE agarose gel for the excision of the 150-bp band. The small RNA library was purified using NucleoSpin Gel and PCR Clean-Up kit (Qiagen, Shanghai, China). The purified library was analyzed for purity using an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA) and then subjected to single-ended strand-specific sequencing on the Illumina HiSeq X10 platform (Illumina, Inc., San Diego, CA, USA).

Following trimming of adaptor sequences, reads were mapped to hg19 using the Burrows-Wheeler Aligner 0.7.12-r1039 and a number of reads mapped to miRNAs from miRBase v21 were calculated. The expression levels were normalized using DESeq2 version 1.20.0. The principal component analysis (PCA) and the contribution of each gene to the dominant direction were performed as described by Huang et al (13). miRNAs that met all the following conditions were considered differentially expressed: i) P<0.05 between two groups; ii) the absolute value of contribution was >0.03; and iii) the average expression level of groups with higher mean expression was >100 reads per kilobase per million mapped reads (RPKM).

The data were presented as the mean ± SD of three independent experiments. The different migration rates between the fast and slow groups were assessed by Students t-test. For differentially expressed miRNAs, the Students t-test was performed to calculate the P-value. Pearsons correlation analysis of miRNA expression between PDCs and their exosomes was performed. The target genes of differentially expressed miRNAs were analyzed using miRTarBase (http://mirtarbase.mbc.nctu.edu.tw) instead of frequently used computational tools, such as TargetScan, miRanda, PicTar and DIANA-microT. The overall survival (OS) in patients with HCC was analyzed using the Kaplan-Meier method and log-rank test. Fishers test and Chi-square test were used to analyze the clinical characteristic differences between the fast and slow groups. P<0.05 was considered to indicate a statistically significant difference.

HCC tissues obtained from surgical resection were cultured to establish PDCs. All cell lines were free of contamination by bacteria or mycoplasma. Short tandem repeat (STR) analysis revealed that all PDCs were derived from the corresponding tissue samples and the STR loci profiles were all unique when compared with each other (data not shown). The growth rate was estimated by the passage time, and the PDCs with a passage time of >20 days were excluded. Considering the potential similarity of PDCs derived from the same patient, all selected PDCs were derived from different patients. In total, 36 PDC cultures were selected for further use in migration assays.

The clinical information of the patients is presented in Table I. The 21 PDCs with a passage time of ≤7 days were classified as having a fast growth rate, and the 15 PDCs with a passage time of >7 days were classified as having a slow growth rate. Hepatitis B virus (HBV) DNA integration in PDCs was examined by polymerase chain reaction (PCR) amplification of the virus gene, and 20 PDCs tested positive, while 5 tested negative for HBV. Serological analysis for HBV antigen and antibody of the derived patient was also performed. HBV surface antigen was detected in 32 patients, indicating a chronic HBV infection in 89% of patients. Edmondson-Steiner grading indicated poor differentiation in the majority of the patients. Quantity and distribution of microvascular invasion (MVI) was evaluated and classified as M0-MVI for no visible MVI, M1-MVI for low-risk MVI, and M2-MVI for high-risk MVI. Other clinical information, including TNM stage and prognosis, was unavailable.

To identify miRNAs that may play a role in the metastasis of liver tumors, cell migration rates were determined using an in vitro wound healing assay. The migration rate was quantified as the wound closure percentage at 24 h after the scratch wound was made. Among the 36 PDCs, the top 10 and bottom 10 PDCs according to the migration rate were designated as the fast- and slow-migrated groups, respectively. No association was observed between PDC migration rate and patient clinical characteristics, such as the growth rate and HBV infection status in the patients. The wound closure percentages were all ≥24.8% in the fast-migrated group, and ≤7.5% in the slow-migrated group. Statistically, the mean and median values were 32.5 and 28.0% in the fast-migrated group, and 2.7 and 3.6% in the slow-migrated group (Fig. 1), respectively, with a significant difference in wound closure between the groups (P=1.8×10⁻⁷). In order to exclude the influence of the proliferation differences of HCC PDCs on the cellular migration, the 20 HCC PDCs in the fast-migrated group and the slow-migrated group were resuscitated and detected in serum-free medium. When cells were cultivated in serum-free medium for 24 h, the results revealed that the proliferation rates in the fast-migrated group and the slow-migrated group were not significant (data not shown).

Exosomes were isolated from the supernatant of PDCs. Fig. 2A presents PDC-derived exosomes visualized by electron microscopy. Exosomes appeared as typical spherical vesicles ranging from 50–150 nm in diameter. Protein expression of known exosome marker CD9 was confirmed by western blot analysis in PDC-derived exosomes and in the PDCs (Fig. 2B upper image). The gray signal intensities of CD9 were presented in Fig. 2B (lower image). The CD9 blot produced two protein bands, with the upper band presumably due to glycosylation modification.

Deep sequencing of small RNA generated an average of 13 million and 9 million total reads for PDC and PDC-derived exosome samples, respectively. Of the total reads, >97% were successfully mapped to the reference genome and the sequencing data was adequate for further analysis.

In PDC samples, miRNA was the most abundant class of small RNAs, representing 47.5% of the entire cellular small RNA component. The percentage of miRNAs among all small RNAs varied from 32.3 to 55.9% from sample to sample, with no significant difference between the fast- and slow-migrated groups (Fig. 3A). By contrast, in exosome samples miRNAs accounted for only 3.0% of all small RNA molecules. Similarly, no significant difference in the exosome miRNA proportion was observed between the two groups. The most abundant class of small RNAs present in the exosomes was unannotated small RNA (Fig. 3A).

Considering the marked difference in the composition of the small RNA population between cellular and exosomal samples, the read length distribution of the samples was investigated. The PDC samples produced a predominant peak at 23 nucleotides and an additional peak at 32 nucleotides, while exosomes had a major peak at 32 nucleotides, and tiny peaks from 22–25 nucleotides (Fig. 3B). miRNAs are 21–25 nucleotides, which corresponded to the major peak in cellular small RNAs and the minor peaks in the exosomal components. The distribution of read length corresponded to the percentile of miRNAs present in cellular and exosomal RNA samples.

Next it was also investigated whether specific miRNAs were upregulated or downregulated in the fast-migrated group compared with the slow-migrated group. PCA was performed and the contribution of each gene to the dominant direction was calculated. The unsupervised PCA clustered the samples into two groups (separated by the blue line in Fig. 4A), which generally accorded with the migration rate of each sample. To focus on the miRNAs with relatively high expression, miRNAs with mean expressions of <100 RPKM were removed. Each miRNA with a Students t-test result of P<0.05 between groups, and an absolute value of contribution of >0.03 was considered to be differentially expressed. Six exosomal miRNAs were differentially expressed between the fast- and slow-migrated groups. In the fast-migrated group, five miRNAs (miR-140-3p, miR-30d-5p, miR-29b-3p, miR-130b-3p and miR-330-5p) were downregulated, and one miRNA (miR-296-3p) was upregulated (Fig. 4B). The detailed parameters for the six miRNAs are listed in Table II.

To reveal the potential roles of the differentially expressed miRNAs in HCC, their target genes were analyzed using miRTarBase (http://mirtarbase.mbc.nctu.edu.tw). The prediction principle for computational tools predominantly depends on the complementarity between miRNA seed region and the 3-untranslated region of their target genes. The majority of miRNAs imperfectly bind to the complementary sites, thus various computational methods have been generated for miRNA target prediction, and the resulting lists of candidate target genes from different algorithms often do not overlap. However, miRTarBase only includes experimentally validated miRNA-target interactions, which provides more reliable data for subsequent pathway analysis. To further reveal the potential roles of these miRNAs, the miRNAs and their target genes were analyzed using the miRTarBase database. In summary, 1652 non-repetitive genes were potential targets of the six differentially expressed miRNAs. Pathway enrichment analysis was performed using DAVID 6.8 (https://david.ncifcrf.gov/) for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome analysis.

In KEGG pathway analysis, the top 10 pathways were predominantly cancer-associated pathways, including ‘pancreatic cancer’, ‘prostate cancer’ and ‘pathways in cancer’. The second highest scoring pathway was ‘focal adhesion’ (Fig. 5A), which is involved in the regulation of cell migration (14). In the Reactome analysis, the highest scoring pathway was ‘extracellular matrix (ECM) organization’, which was consistent with the ‘focal adhesion’ pathway identified by KEGG analysis. Other Reactome pathways, including ‘signaling by PDGF’ and ‘non-integrin membrane-ECM interactions’ were also associated with cell migration (Fig. 5B). The genes associated with the ‘focal adhesion’ pathway that were predicted as targets of the six exosomal miRNAs were comprised of upstream cytokine-receptor interaction molecules, including ECM components (COL1A1, COL5A2, LAMA and LAMC1), growth factors (PDGFC, VEGFA and IGF1), receptor tyrosine kinases (ERBB2 and PDGFRA) and integrins (ITGA5 and ITGB1), and downstream factors (PTEN and PIK3CA), cell proliferation genes (CCND1 and CCND2), cell survival genes (BCL2 and BAD) The targets of the six differentially expressed miRNAs in the ‘focal adhesion’ pathway and their associated miRNAs are presented in Fig. 6.

The expression of individual miRNAs was compared between exosomes and their corresponding PDCs. Generally, the association analysis highlighted a strong correlation (R²=0.83) between normalized miRNA expression in PDCs and exosomes, which indicated that exosomes reflect the miRNA composition of the derived cells. As for the six differentially expressed miRNAs, normalized exosomal miRNA levels were also correlated with the miRNA levels of PDCs as well (Fig. 7A). Considering the average miRNA expression levels were evaluated above, the expression levels of the 6 miRNAs in individual PDC samples and the paired exosome samples were further analyzed, and a similar trend of high correlation of miRNA expression between PDCs and their corresponding exosomes was observed (data not shown).

In line with the finding that exosome miRNAs may reflect the miRNA composition of the PDCs that the exosomes were derived from, tumor tissue-derived miRNA levels and the clinical information of liver tumors were obtained from TCGA for validation analysis. Among the 372 available records, only 4 patients were categorized as M1 for definitive distal metastasis; therefore, it was impossible to study the association between miRNA levels and metastasis directly. Considering distal metastasis was positively correlated with poor clinical prognosis, the association between OS and miRNA levels were analyzed. The patients were divided by the expression level of a certain differentially expressed miRNA, with the top quartile and bottom quartile patients set as the low and high miRNA expression groups, respectively. The miRNA expression level and corresponding OS was analyzed using the Kaplan-Meier method (Fig. 7B). Patients with high expression of the three differentially expressed miRNAs, miR-30d, miR-140 and miR-29b, exhibited improved OS (P=0.0351, P=0.0465 and P=0.0672, respectively). Analysis of the clinical characteristics, including tumor stage, pathological T, N and M stage, detected no differences between the low and high miRNA expression groups (Tables III–V), suggesting that the result is not a secondary effect caused by the differences in the clinical characteristics between the two groups. Moreover, we analyzed the difference between the cirrhosis of these patients and survival from TCGA database. The following result revealed that there was no difference between cirrhosis and cum survival (P=0.773), indicating that our result concerning miRNAs and survival is not an epiphenomenon from their differences in underlying cirrhosis (data not shown).