Serum samples were collected from 50 patients with NAFLD and 50 healthy individuals at The First Affiliated Hospital of Nanjing Medical University (Nanjing, China) between January and August 2018. The presence of FLD was assessed using ultrasonography. Patients were included in the study if they had confirmed fatty liver disease with no history of significant alcohol consumption, viral hepatitis, autoimmune hepatitis, metabolic diseases or hepatotoxic medication (e.g. amiodarone). Clinical information, including sex, age, body mass index and biochemical measurements were obtained from each patient. The Ethics Committees of Nanjing Medical University approved the present study and all participants provided written informed consent.

The Model Animal Research Center of Nanjing University provided male C57BL/6J mice, (age, 4–5 weeks; weight, 18–22 g; n=40) which were kept in a pathogen-free barrier facility at 22°C, with 50% relative humidity and under a 12-h light/dark cycle. The animals were randomly assigned either into the normal control (chow diet; CD) or NAFLD group (HFD; 5.56 kcal/g; fat, 58 kcal%; hydrogenated coconut oil, 54%; carbohydrate, 25.5 kcal%) for 12 weeks, as described previously (24). The mice were weighed and euthanized by cervical dislocation at the end of week 12. The intact livers were rapidly excised from the abdominal cavity and washed with PBS for primary hepatic cell culture, RNA extraction or histopathology. The Animal Experiment Ethics Committee of Nanjing Medical University approved the study (approval no. IACUC-1601176).

Fresh liver tissue samples were washed with PBS and placed in 10% buffered formalin at 4°C for 24 h. After gradient dehydration in ethanol and infiltration in xylene for 30 min, liver samples were sectioned into 5-µm slices and stained using hematoxylin and eosin (hematoxylin for 10 min; and eosin for 15 sec at room temperature) to assess organization and architecture. In addition, Masson's trichrome and Sirius red staining was performed for extracellular matrix and fibrosis evaluation using commercially available kits (Beijing Solarbio Science & Technology Co., Ltd.), according to the manufacturer's protocols. Cell apoptosis was determined using the TUNEL method, as previously described (25). Neovascularization within the liver parenchyma and nodules was also determined using CD31 staining (Abcam; cat. no. ab182981; 1:2,000; 4°C overnight). All stained slides were observed using a Zeiss Axioskop 40^® upright research light microscope with an original magnification, ×40 (Zeiss AG). Histological evaluation was performed in a blinded manner by two independent pathologists. In order to define the stage of steatosis, fibrosis, ballooning and parenchymal inflammation, NASH Activity Score (NAS) (26) was calculated.

Hepatic cells were isolated using collagenase (type IV; Sigma-Aldrich; Merck KGaA) perfusion, as described previously (27). In brief, after the portal vein was cannulated, the liver was perfused for 5 min with calcium-free HEPES buffer (0.33 mM; pH, 7.6), followed by perfusion for 8 min using collagenase (0.025%) and calcium chloride (0.075%) at 37°C. Following enzymatic digestion, the solution was centrifuged (50 × g for 2 min at 4°C) to obtain hepatic cells. Freshly isolated hepatic cells were maintained in DMEM/F12 (Gibco; Thermo Fisher Scientific, Inc.) containing 10% exosome-depleted FBS and 1% penicillin/streptomycin. The cells were incubated at 37°C in a humidified chamber with 5% CO₂ for 24 h.

Blood was collected from animals via cardiac puncture following sacrifice. Serum from animals and human subjects was obtained via centrifugation of whole blood (2,000 × g for 15 min at room temperature) and stored at −80°C. Serum biochemical markers [alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate aminotransferase (AST), albumin (ALB), globulin, γ-glutamyl transpeptidase, lactate dehydrogenase (LDH), creatine kinase (CK), hydroxybutyrate dehydrogenase, total bilirubin, cholesterol (CHOL), triglyceride (TG), high-density lipoprotein (HDL), low-density lipoprotein (LDL), lipoprotein a (Lpa), fasting blood-glucose (GLU0), creatinine (Cr) and uric acid] characterizing liver disease metabolic features were measured using a biochemical analyzer (Roche Diagnostics).

Putative target genes of the selected miRNAs (miR-135a-3p and miR-504-3p) were predicted using two databases, with a pure algorithm (miRDB: mirdb.org and TargetScan: www.targetscan.org/mamm_31) and overlapping target genes were selected for further functional analysis. In order to investigate the biological function of the predicted genes, Gene Ontology (GO) classification and Kyoto Encyclopedia Genes and Genomes (KEGG) enrichment were performed according to gene annotation using Database for Annotation, Visualization and Integrated Discovery (www.kegg.jp). Fisher's exact test, modified by the false discovery rate, was used to evaluate the statistical significance of enrichment.

EVs from in vitro hepatocellular culture medium were separated using differential centrifugation and filtration. Briefly, cell culture medium was centrifuged at 10,000 × g for 30 min at 4°C to deplete the cells and debris. Subsequently, the supernatant was filtered using a 0.22-µm pore membrane and centrifuged for 70 min at 100,000 × g at 4°C to pellet the EVs. The pellets were then suspended in 100 µl PBS. Using ExoQuick reagent (System Biosciences, LLC), EVs from mouse or human serum were extracted according to the manufacturer's protocol. In brief, the EVs were isolated by adding EV extraction reagent and centrifuging for 20 min at 10,000 × g at 4°C. The resulting EV pellets were stored at −80°C for subsequent experiments.

Using RIPA buffer (Beyotime Institute of Biotechnology), whole mouse primary hepatic cell extracts and EV proteins were prepared and measured using a BCA protein assay kit (Beyotime Institute of Biotechnology) according to the manufacturer's protocol. The extracted proteins (20 µg per lane) were separated using 10% SDS-PAGE, transferred onto PVDF membranes and blocked with 5% non-fat milk in TBS-0.05% Tween 20 for 2 h at room temperature then incubated with primary antibodies against tumor susceptibility 101 (TSG101; 1:1,000; cat. no. ab125011; Abcam), CD9 (1:1,000; cat. no. ab92726; Abcam), heat shock protein 70 (HSP70; 1:1,000; cat. no. 33-3800; Invitrogen; Thermo Fisher Scientific, Inc.), calreticulin (1:1,000; cat. no. ab92516; Abcam) and GAPDH (1:3,000; cat. no. ab8245; Abcam) overnight at 4°C. Subsequently, the membrane was incubated with secondary antibody [goat anti-rabbit IgG (HRP), cat. no. sc-2004; goat anti-mouse IgG (HRP), cat. no. sc-2005; 1:10,000; Santa Cruz Biotechnology, Inc.] for 1 h, then a SuperSignal West Pico Chemiluminescent Substrate kit (Pierce; Thermo Fisher Scientific, Inc.) was used to visualize the proteins according to the manufacturer's instructions. BandScan5.0 software (Glyko, Inc.) was used to analyze protein expression level.

In total, 5 µl EV preparation was fixed with 5 µl 4% paraformaldehyde at 4°C for 30 min and added onto formvar-coated copper grids to settle for 30 min. Subsequently, the samples were fixed with 1% glutaraldehyde for 5 min at 25°C, then stained with 2% uranyl oxalate for 5 min at 25°C and 1.8% methyl cellulose uranyl acetate for 10 min on ice in the dark, embedded in epoxy resin and polymerized at 35°C for 12 h, 45°C for 12 h and 60°C for 24 h. Residual liquid was removed from the grid with filter paper and the samples were visualized at 120 kV using the FEI Tecnai G2 Spirit Bio TWIN TEM (FEI; Thermo Fisher Scientific, Inc.). Images were acquired using an AMT 2k CCD camera (Advanced Microscopy Techniques, Corp.).

In order to identify the diameter and concentration of EVs, Nanosight NS300 system (Malvern Instruments, Ltd.) fitted with a fast video capture was used. The images of sample movement were captured for 60 sec at room temperature. In order to analyze the videos and measure particle concentrations and size distribution, the NanoSight LM10 system (NanoSight Ltd.) was used. Each sample was evaluated ≥3 times.

Total RNA was extracted from EVs and cells using TRIzol^® (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions and quantitated using the NanoDrop 2000 (NanoDrop Technologies; Thermo Fisher Scientific, Inc.). TRIzol^® LS (Invitrogen; Thermo Fisher Scientific, Inc.) was also used to extract serum RNA. A total of 1 pM cel-miR-39 RNA oligonucleotide was added to each sample as a spike-in control when extracting RNA from EVs or serum. A Taqman^® miRNA RT kit and miRNA-specific stem-loop primers (Applied Biosystems; Thermo Fisher Scientific, Inc.) were used to detect the expression levels of miRs, according to the manufacturer's instructions. In order to detect the expression levels of miRNA, the miR-X miRNA RT-qPCR SYBR kit (Clontech Laboratories, Inc.) was used for RT-qPCR with the Step One Plus Real-Time PCR system (Applied Biosystems; Thermo Fisher Scientific, Inc.) with the following program: Initial denaturation at 95°C for 3 min, followed by 40 cycles of 95°C for 10 sec and annealing and extension at each annealing temperature at 60°C for 30 sec. Relative expression level of each miRNA were normalized to cel-miR-39 and via the 2^−ΔΔCq method (28). The following primer sequences were used for RT-qPCR: cel-miR-39 forward (F), 5′-ACACTCCAGCTGGGGTCACCGGGTGTAAATC-3′ and reverse (R), 5′-CTCAACTGGTGTCGTGGAGTCGGCAATTCAGTTGAGCAAGCTGA−3′; miR-122-5p F, 5′-ACACTCCAGCTGGGTGGAGTGTGACAATGG-3′ and R, 5′-CTCAACTGGTGTCGTGGAGTCGGCAATTCAGTTGAGCAAACACC−3′; miR-129-5p F, 5′-ACACTCCAGCTGGGCTTTTTGCGGTCTGG-3′ and R, 5′-CTCAACTGGTGTCGTGGAGTCGGCAATTCAGTTGAGGCAAGCCC-3′; miR-135a-3p F, 5′-ACACTCCAGCTGGGTATAGGGATTGGAGCC-3′ and R, 5′-CTCAACTGGTGTCGTGGAGTCGGCAATTCAGTTGAGCGCCACGG-3′; and miR-504-3p F, 5′-ACACTCCAGCTGGGGGGAGTGCAGGGCAG-3′ and R, 5′-CTCAACTGGTGTCGTGGAGTCGGCAATTCAGTTGAGGAAACCCT-3′. Universal reverse primer: 5′-TGGTGTCGTGGAGTCG-3′.

The data are presented as the mean ± SEM (n≥3). The differences between two groups were analyzed using unpaired Student's t-test. Receiver operating characteristic (ROC) curves were generated based on the expression level of each miRNA to determine the utility of each miRNA for diagnosing NAFLD. GraphPad Prism 7 (GraphPad Software, Inc.) and SPSS v22.0 software (IBM Corp) were used for statistical analysis. P<0.05 was considered to indicate a statistically significant difference.

After feeding for 12 weeks, the liver from mice in the control group showed normal morphology with red, shiny and smooth appearance, whereas the liver from mice fed HFD exhibited a slightly yellow color, smooth tight membranes and soft texture (Fig. 1A). The biochemical analyses showed that mice fed HFD exhibited significantly higher AST, albumin, ALP, CHOL, HDL and LDL levels compared with in mice fed CD (Table I). In addition, the mice fed HFD exhibited notable hepatic steatosis, liver injury, hepatocellular ballooning, fibrosis, angiogenesis and cell death (Fig. 1B and Table SI). Serum ALT levels were also significantly increased in mice fed HFD compared with in mice fed CD, suggesting that a HFD induced hepatocyte damage (Fig. 1C).

TEM images showed that isolated EVs demonstrated typical morphology with an exosomal characteristic bilayer membrane (Fig. 2A). The size distribution of EVs was measured using NTA (Fig. 2B), which was within the normal range of EV sizes. Equal amounts of protein from EVs and primary hepatocytes were separated; expression levels of exosomal markers CD9, HSP70 and TSG101 were higher in EVs compared with in primary hepatocytes (Fig. 2C). In addition, calreticulin and GAPDH were expressed in primary hepatocytes, but not expressed in EVs (Fig. 2C).

According to our previous study (23), C57BL/6J mice were fed HFD for 12 weeks for induction of an NAFLD model. EVs were collected from the culture medium of hepatocytes from mice fed HFD or CD. There was a marked difference in the expression levels of several miRNAs in the hepatocyte-derived EVs from mice fed HFD compared with mice fed CD (Table II). From these miRNAs, three human and mouse homologous miRNAs were selected, with fold change >2 and P<0.05 (miR-129-5p, miR-135a-3p and miR-504-3p) for further investigation. miR-122-5p, which served as a good indicator of NAFLD in previous studies (12,15), was also included for verification.

In order to verify our previous microarray results (23), the expression levels of miR-129-5p, miR-135a-3p and miR-504-3p was measured in hepatocellular EVs using RT-qPCR. The results showed that all three miRNAs exhibited lower expression levels in hepatocellular EVs from mice fed HFD compared with in mice fed CD, showing consistency with our previous microarray results (Fig. 3B-D). The expression levels of miR-122-5p were slightly decreased in hepatocellular EVs derived from mice fed HFD (Fig. 3A). In order to investigate whether these miRNAs were present in the liver, the expression levels of candidate miRNAs in hepatocytes was investigated. miR-122-5p and miR-504-3p were decreased in primary hepatocytes from mice fed HFD, while differences in miR-129-5p and miR-135a-3p were not significant (Fig. S1). Then, it was investigated whether the selected miRNAs were differentially expressed in serum or hepatocyte-derived EVs. Expression levels of miR-504-3p, miR-135a-3p and miR-129-5p were significantly decreased in serum from mice fed HFD compared with in mice fed CD (Fig. 3F-H). However, there was increased miR-122-5p expression in serum EVs from HFD mice compared with CD mice (Fig. 3E).

In order to investigate the expression levels of selected miRNAs, 50 healthy individuals with no signs of liver disease and 50 patients with NAFLD were recruited. The clinical features of subjects are shown in Table III. The patients were sex- and age-matched. The patients with NAFLD and healthy controls did not exhibit significantly different levels of ALP, CK, hydroxybutyrate dehydrogenase, total bilirubin, total CHOL, LDL, Lpa, total protein, ALB, globulin, ALB/globulin, GLU0 and Cr. However, patients with NAFLD had notably higher AST, γ-glutamyl transpeptidase, LDH, TG and urea levels and lower HDL levels compared with healthy controls.

As EVs protect circulating miRNA from degradation by RNase, it was hypothesized that miRNA expression levels in EVs from serum would detect NAFLD progression more effectively compared with those in total serum. Using RT-qPCR and cel-miR-39 as an exogenous normalizer, it was found that miR-135a-3p, miR-129b-5p and miR-504-3p expression levels in EVs isolated from the serum of patients with NAFLD were markedly decreased compared with in the healthy controls, confirming this hypothesis. In addition, miR-122-5p in circulating EVs and serum ALT expression levels were significantly increased in patients with NAFLD (Fig. 4A-E). In order to evaluate the diagnostic potential of candidate miRNAs, ROC curve analysis was performed. miR-135a-3p exhibited the highest area under the curve (AUC; 0.849; 95% CI, 0.777–0.921), while miR-122-5p and miR-504-3p had AUC values of 0.790 (95% CI, 0.700–880) and 0.708 (95% CI, 0.606–0.810), respectively. The AUC of the traditional diagnostic indicator ALT was 0.672 (95% CI, 0.567–0.778; Fig. 4F-J). These data indicated that miR-135a-3p in EVs may be a promising diagnostic indicator and were more sensitive than ALT.

Subsequently, the presence of these miRNAs in serum from patients with NAFLD was investigated. miR-135a-3p expression levels were markedly lower, whereas those of miR-122-5p were notably increased in patients with NAFLD compared with healthy controls. However, there was no notable difference in expression levels of miR-129-5p and miR-504-3p in serum between patients with NAFLD and healthy controls (Fig. S2A-D). Thus, in NAFLD, some miRNAs were specifically enriched in EVs, and detection of serum exosomes was a more favorable diagnostic method than direct detection of serum miRNA for NAFLD.

KEGG and GO analysis were used to investigate the key biological associations, signaling pathways and molecular functions of target genes of miR-135a-3p and miR-504-3p. For miR-135a-3p, the most significantly enriched GO terms were associated with ‘platelet-derived growth factor receptor signaling pathway’, ‘cell-cell junction’ and ‘GTP binding’ (Fig. 5A-C). The significantly enriched KEGG pathways had functions associated with ‘AMP-activated protein kinase (AMPK) signaling pathway’ (Fig. 5D). The target genes of miR-504-3p were significantly enriched in the GO terms ‘nucleotide-binding, oligomerization domain-like receptor signaling pathway’, ‘glutamatergic synapse’, ‘inflammatory responses’ and ‘ion transmembrane transport’ (Fig. 5E-G). According to the KEGG enrichment analysis, the most significantly enriched pathway for the target genes of miR-504-3p was ‘glutathione metabolism’ (Fig. 5H). Furthermore, ‘lysosome’ and ‘metabolic pathways’ were also highly enriched.