The potential of mecciRNA in hepatic stellate cell to regulate progression of nonalcoholic hepatitis

MCD (methionine/choline deficient) feed and standard feed were purchased from Nantong Trophy Feed Technology Company, MCD feed contains (per 1000 g): amino acid premix (methionine free) 175.7 g, methionine 0 g, choline chloride 0 g, sucrose 431.9 g, dextrin 50 g, corn starch 150.0 g, corn oil 100.0 g, cellulose 30.0 g, mineral mix 52.4 g. All the primers used in this study were synthesized by Tsingke Company and CWBIOTECH, their sequences were shown in Supplementary Materials. All of antibodies used in western blot, including c-MYC (ab32072), SMAD2 (ab40855), SMAD3 (ab40854), THBS1 (ab267388), STAT3 (ab68153), p-STAT3 (Y705, ab267373) and GAPDH (ab8245), were purchased from Abcam.

8 weeks male C57BL/6 mice were kept in a controlled environment (24 ± 2 °C, 12/12 h day/dark cycle). And mice were randomly divided into 2 subgroups (n = 6): Group 1 was fed with standard diet for 6 weeks (control group); Group 2 was fed with MCD diet for 6 weeks (MCD group). After the treatment mentioned above, mice were fasted for 12 h before being sacrificed. The liver was excised and perfused with saline. One portion of the liver from each mouse was fixed in 4% paraformaldehyde solution for histological analysis. Another portion of the liver was used to prepare liver homogenate for the biochemical analysis.

Human hepatic stellate cells cell lines LX-2 were gifted by Dr. Xinping Huang, Guangzhou Institute of Biomedicine and Health, Chinese Academy of Sciences (purchased from the Advanced Research Center of Central South University). LX-2 was cultured in Dulbecco's Modified Eagle’s Media (Thermo Fisher Scientific, USA) supplemented with 10% fetal bovine serum (FBS, Thermo Fisher Scientific, USA) in a 5% CO2 humidified incubator at 37 °C. All the experiments were performed within 1 months of resuscitation and the cell passage was less than 3 generations from initial resuscitation (avoid activation of LX-2 cells).

Total RNAs were isolated using Trizol reagent (Invitrogen, USA), either from cultured cells or liver tissue. 2 ug of total RNA was subjected to reverse transcription using Superscript III transcriptase (Invitrogen, USA). RT-qPCR was conducted using a Bio-Rad CFX96 system (Bio-Rad, USA) with SYBR green to determine the expression level of targets of interest. In addition, miRNA cDNA Synthesis Kit (CWBIOTECH, CN) and miRNA qPCR Assay Kit (CWBIOTECH, CN) were used for miRNA detection. Expression levels of circRNAs were normalized to the expression levels of GAPDH, while small RNA RNU6 (U6) was used for miRNA. And GAPDH and MTCO2 served as the cytosolic and mitochondrial control, respectively.

We conducted this experiment according to the kit protocol (Cat. no. 37612, QIAGEN, GER). First, LX-2 cells were suspended in Lysis Buffer (selectively disrupts the plasma membrane without solubilizing it, resulting in the isolation of cytosolic proteins), and incubated for 10 min. After that, plasma membranes and compartmentalized organelles, such as nuclei, mitochondria, and endoplasmic reticulum, remained intact and were pelleted by centrifugation (1000×g, 10 min). Then, the pellet was resuspended in Disruption Buffer, repeatedly passed through a narrow-gauge needle, and recentrifuged (1000×g, 10 min) to pellet nuclei, cell debris, and unbroken cells. The supernatant (contains mitochondria) was recentrifuged (6000×g, 10 min) to pellet mitochondria. After removal of the supernatant, mitochondria are washed and resuspended in Mitochondria Storage Buffer.

We downloaded high-throughput sequencing and noncoding RNA expression profiling data (GSE134146 and GSE46300) from the GEO database. Among them, GSE134146 was used to screen for differentially expressed mecciRNAs in NASH, while GSE46300 was used to identify NASH-related genes. Differential expressions of all genes were calculated using the R package “limma”, and significance was evaluated by one-way analysis of variance (ANOVA).

We used the circMINE database (http://​hpcc.​siat.​ac.​cn/​circmine/​), which is based on 3 well-annotated databases, including miRanda [18], miRBase [19], and circBase resources [20], to predict the potential miRNAs targeted by mecciRNAs.

In addition, miRNet database (https://​www.​mirnet.​ca/​) was based on the 14 open-source databases, which could analyze and generate miRNA-mRNA network online. Specifically, the miRNA target gene data were collected from three well-annotated database miRTarBase [21], TarBase [22], and miRecords [23]. The miRNA to molecule interaction data were collected from SM2miR [24] and Pharmaco-miR [25]. The miRNA to disease interaction data were collected from miR2Disease [26] and PhenomiR [27]. The miRNA to epigenetic modifier interaction data were collected from EpimiR [28]. And, the exosomal miRNA annotation data were collected from ExoCarta [29]. It should be emphasized that Protein–Protein Interaction (PPI) database was included in the analysis during the construction of miRNA-mRNA network.

GO terms in three categories (GO: biological process, GO: cellular component and GO: molecular function) were used for pathway-enrichment analysis and biological interpretation. We used the GO enrichment analysis and visualization tool (GOrilla) to identify GO terms that were significantly enriched in the target gene list. KEGG pathway enrichment analysis was performed with clusterProfiler with a background set of all entrez IDs mapped to a KEGG pathway.

Given that searching for protein interactions and interaction networks is integral to further exploration of cellular states, biological processes and functions, relevant targets were entered into STRING (version 11.5, https://​string-db.​org/​) [30]. Searching by gene name, selecting Homo sapiens as the species and selecting suitable confidence, the purpose of in-depth study of protein–protein interactions can be achieved. Network nodes and edges represent proteins and protein–protein associations, respectively.

On the basis of the expression profile, 21 mecciRNA-related genes were clustered by the R packages. Consensus Cluster Plus with reps = 100, p Item = 0.8, and p Feature = 1 [31]. By comprehensively analyzing the consistency matrix and the consistency cumulative distribution function, the optimal partition is defined.

To differentiate immune levels between 2 clusters, we used single-sample gene set enrichment analysis (ssGSEA) to characterize 23 types of immune cells in the liver tissues, based on the specific gene signatures of immune cells [32].

The applicable statistical methods were used depending on the type of data. The student’s t-test was used for comparisons between groups. ANOVA for multiple comparisons was used to detect differences amongst the various treatments. All data from three separate experiments at least are presented as mean ± SD. Differences were considered significant for P-values less than 0.05. *P < 0.05, **P < 0.01, and ***P < 0.001.

circRNAs have become a novel research hotspot in RNA biology. They localize in specific subcellular compartments, and play different biological roles [33]. However, the functions of circRNAs in NASH is still elusive. As activation of HSCs is well regarded as the central driver of fibrosis in NASH, we decided to explore the differentially expressed circRNAs in HSCs of NASH patients [8].

Thus, we downloaded the microarray gene profiling dataset (GSE134146, including 4 patients with NASH cirrhosis and 4 patients without NASH) from Gene Expression Omnibus (GEO) for bioinformatics analysis [17]. The circRNAs with P-value < 0.05 and |Log₂FC|≥ 1.5 from t-test were identified as differentially expressed circRNAs (Fig. 1A–B). In total, 20 circRNAs were upregulated, while 8 circRNAs were downregulated in NASH (Fig. 1C). Among them, 12 circRNAs are less than 500nt in length (inclined to function as molecular scaffolds), 4 are between 500 and 1000 nt, and 12 are longer than 1000nt (inclined to function as miRNA sponges) (Fig. 1D). Consistent with previous reports, most of them (20 out of 28) are encoded by exons (Fig. 1E). These results indicated that circRNomics homeostasis of HSCs is disrupted in NASH patients. Consequently, we speculated that there might be an association between circRNA imbalance and NASH progression.

During the development of NASH, there is a constant dysfunction of mitochondria, including alterations in enzyme activities, protein expression, and signaling networks [34]. Thus, we wondered whether the mecciRNAs expression profile was altered in NASH progression.

Surprisingly, we found that half (4 out of 8) of downregulated circRNAs were encoded by the mitochondrial genome, namely hsa_circ_0089761, hsa_circ_0089762 (also known as circSCAR [17]), hsa_circ_0089763 and hsa_circ_0008882 (Fig. 2A, B). Given the majority of circRNomics are occupied by nuclear-encoded circRNAs (99.93%), it is noteworthy that mecciRNAs accounted for 14.3% of the differentially expressed circRNAs in NASH (Fig. 2C) [35]. Among them, hsa_circ_0008882 are encoded by the heavy strand of mitochondrial genome, while hsa_circ_0089761, hsa_circ_0089762 and hsa_circ_0089763 are generated from the light strand (Fig. 2D). Additionally, their sizes span a wide range, in which lengths of hsa_circ_0089761 and hsa_circ_0089763 are over 5000 nt, while hsa_circ_0089762 and hsa_circ_0008882 are less than 400 nt (Fig. 2D). Then, we used mitochondrial-cytoplasmic separation assays to determine their cellular localization. The results demonstrated that these mecciRNAs were mainly located in mitochondria, while a part of them were also presented in cytoplasm (Fig. 2E).

circRNAs could function as miRNA decoys, which are defined as miRNA sponges that bind miRNAs and prevent them from suppressing their target mRNAs [36]. Noteworthy, circRNAs are generally with relatively few binding sites for miRNAs, the idea of controlling the stability and quantity of miRNAs by circRNAs and achieving measurable effects should be considered with caution [37, 38]. However, hsa_circ_0089761 and hsa_circ_0089763 are highly possible to possess powerful miRNA-sponge capacity, resulted from their long sequence lengths (over 5000 nt). The copy number ratio between specific circRNA and many different targeted miRNAs is relatively small [39]. Hence, considering the great abundance of substrates (miRNAs), hsa_circ_0089761 and hsa_circ_0089763 are not likely to have competitive binding. On the contrary, it is reasonable to speculate that these two circRNAs might have double sponge capacity for specific miRNAs, due to the fact that hsa_circ_0089761's sequence completely contains hsa_circ_0089763's sequence (Fig. 2D).

First, utilizing the circMINE database, we predicted the miRNAs targeted by hsa_circ_0089761 and hsa_circ_0089763 [40]. According to the filter criteria (a. Score cutoff > 150; b. Energy cutoff < –7; c. Amount of specific miRNA-binding sites ≥ 3), we identified 52 and 26 mecciRNA-miRNA pairs respectively (Fig. 3A). As expected, hsa_circ_0089761 and hsa_circ_0089763 shared 26 identical miRNA targets (hsa-mir-136-5p, hsa-mir-154-3p, hsa-mir-155-3p, hsa-mir-195-3p, hsa-mir-345-5p, hsa-mir-365a-3p, hsa-mir-365b-3p, hsa-mir-374a-5p, hsa-mir-384, hsa-mir-3943, hsa-mir-4705, hsa-mir-4732-3p, hsa-mir-5003-5p, hsa-mir-5193, hsa-mir-548aa, hsa-mir-548as-3p,hsa-mir-548t-3p, hsa-mir-5583-5p, hsa-mir-6511a-3p, hsa-mir-6511b-3p, hsa-mir-664a-3p, hsa-mir-6729-3p, hsa-mir-6731-3p, hsa-mir-6758-3p, hsa-mir-7844-5p, hsa-mir-487a-3p). Then, we identified 2000 mRNAs targeted by the 26 miRNAs in miRNet database (Additional file 4: Table S1) [41]. It is worth mentioning that we also took the PPI database into account when identifying targeted mRNAs. Therefore, among these targets, there may not only be changes in mRNA or protein expression levels, but also changes in protein modification status. Next, To further enhance the reliability, we only selected the part with degree frequency > 1 for subsequent analysis (Additional file 5: Table S2, Additional file 1: Figure S1A). Finally, we successfully established a novel mecciRNA-miRNA-mRNA network in HSCs, including 2 mecciRNA nodes, 26 miRNA nodes and 1343 mRNA nodes (Additional file 2: Figure S1B).

To achieve a better understanding of the functions associated with these mecciRNAs-related genes, we performed GO Term and KEGG pathway enrichment analysis. GO analysis showed an enrichment of GO terms indicative of transcription regulation, such as transcription coactivator activity, enhancer binding, and histone modification (Fig. 3B). Meanwhile, KEGG analysis results revealed the enrichment of several pro-fibrotic signaling pathways, such as Hippo signaling pathway and TGF-β signaling pathway (Fig. 3C) [10, 42]. Strikingly, the top 20 most enriched genes contained many famous NASH-related genes, such as c-MYC, SMAD2 and SMAD3 (Fig. 3D–E) [43‐46]. Taken together, these results indicated that mecciRNAs might have the potential to regulate fibrosis-related signaling pathways in HSCs.

Death of hepatocytes could induce liver inflammation and then directly or indirectly promote the activation of HSCs [47]. Conversely, activated HSCs also further aggravated the damage and death of hepatocytes [48]. Hence, the crosstalk between HSCs and hepatocytes performs important functions in NASH progression, in which exosome-mediated intercellular communication cannot be ignored [49]. The exosomal cargo consists of proteins, lipids and ncRNAs, while many of the biologic effects of exosomes have been attributed to miRNAs [50]. Therefore, we were particularly interested in whether the changes in mecciRNA-miRNA profile of HSCs could influence the signaling pathways of hepatocytes.

First, using the exosomal miRNAs dataset (ExoCarta, a well-known database of exosomal proteins, RNA and lipids) in miRNet, we screened out 7 candidates (hsa-mir-485-3p, hsa-mir-618, hsa-mir-642a-5p, hsa-mir-1224-3p, hsa-mir-1248, hsa-mir-3529-3p and hsa-mir-3679-3p) among the 26 mecciRNA-targeted miRNAs, which had been reported that can be secreted to extracellular microenvironment through exosomes to function (Additional file 6: Table S3, Fig. 4A) [41]. Then, utilizing the same analytical methods and filter criteria, we finally identified 419 mRNAs in hepatocytes, which might be regulated by the mecciRNA-miRNA network (Additional file 7: Table S4, Fig. 4B, C).

Meanwhile, we downloaded the raw sequencing data of GSE46300 from GEO, which includes liver tissues with or without NASH, and obtained a set of differentially expressed genes between NASH and normal liver (|Log₂FC|≥ 1, P < 0.05, Fig. 4D, E) [51]. In total, 21 overlapped genes were identified (STAT3, POU5F1, TRAF3, UBE2K, CXCL5, SPAG9, MBD2, THBS1, TMOD3, NPHP1, NEDD9, COL5A3, ABCC5, FRMD6, FBXL18, TXNIP, FYTTD1, PSG4, DKK3, CYP2B6, FOXK2), in which STAT3 and THBS1 were localized at the core of protein interaction network (Fig. 4F–H). Despite several reports about their functions in liver steatosis, in-depth molecular mechanisms about STAT3 and THBS1 regulating NASH progression still deserve further study [52‐54].

Classification of specific disease based on different cellular and molecular features could deepen our understanding of this disease and even provide valuable information for treatment. Hence, in order to evaluate the effects of these 21 genes on NASH, we extracted the expression profiles of these targets from GEO46300 to construct a consensus cluster (Fig. 5A). Two subtypes (Cluster 1 and Cluster 2) were obtained with the minimum variance within the group and the maximum variance across the groups (Fig. 5B, Additional file 2: Fig. 2A–D). Meanwhile, the results of principal component analysis (PCA) confirmed that there are significant differences between the two clusters (Fig. 5C).

We then performed differential gene expression analysis across the transcriptomes of these two subtypes. In total, we identified 273 differentially expressed genes (Additional file 8: Table S5, Fig. 5D). GO analysis revealed that these genes were highly enriched for functions related to chemokine activity and immune cell chemotaxis (Table. 1, Fig. 5E). And KEGG analysis demonstrated that, in addition to the upregulation of immune cell chemotaxis-related pathways, IL-17 signaling pathway and TNF signaling pathway were also significantly activated in Cluster 1, which could facilitate the development of NAHS by exerting proinflammatory effects (Table. 2, Fig. 5F–G) [55, 56].

Considering the differences between 2 subtypes were mainly focused on immune-related pathways, we further utilized ssGSEA scores to characterize the immune cell components in the 2 subtypes [32]. Despite most of the differences did not reach statistical significance, immune cell infiltration was higher in Cluster 1 than in Cluster 2 generally, which was consistent with the GO and KEGG analysis (Fig. 5H). Noteworthy, infiltration of T helper 1 (Th1) cells and NK CD56^dim cells was significantly higher in Cluster 1, while central memory T (Tcm) cells infiltration was reduced (Fig. 5I). It is universally agreed that Th1 cells are proinflammatory cells and aggravate liver fibrosis development [57‐59]. However, the roles of NK CD56^dim cells and Tcm cells in NASH remain elusive and require further study.

In general, we established a novel immunotyping of NASH based on mecciRNA-related network, in which Cluster 1 presents increased immune cell infiltration and activation of proinflammatory signaling pathways.

First, to validate the bioinformatic conclusions, we successfully established HSC-activation model by exposing LX-2 cells (human hepatic stellate cell line) to lipopolysaccharide (LPS) at a concentration of 100 ng/ml for 48 h (Fig. 6A). We confirmed that both hsa_circ_0089761 and hsa_circ_0089763 was detected in a lower amount in activated HSCs compared to wild-type HSCs (Fig. 6B). To further validate our computational predictions, we detected the expression levels of 5 core miRNAs (as test cases) in the mecciRNA-miRNA network. The results largely conformed to our predictions that miR-642a-5p, miR-1248, miR-670-3p and miR-1224-3p was upregulated in activated HSCs, while only miR-4667-3p level was comparable between the two groups (Fig. 6C). As shown in Fig. 3D, SMAD2, SMAD3 and c-MYC are core targets in HSCs mecciRNA-miRNA network, which have been reported to play vital roles in NASH progression [43‐46]. The western blot results demonstrated the significant upregulation in expression levels of these proteins in activated HSCs as expected (Fig. 6D).

Meanwhile, a set of candidates were considered as the downstream effectors of mecciRNAs-miRNAs network in hepatocytes, in which THBS1 and STAT3 occupied key positions. As feeding mice an MCD diet is a widely accepted model to study relevant mechanisms in NASH, we established this animal model, and detected the expression levels of THBS1 and STAT3 in control group and MCD-diet group (Fig. 6E) [60, 61]. The results revealed that the expression of THBS1 was increased in MCD-diet group, despite the similar expression levels of STAT3 between the two groups (Fig. 6F). Considering PPI database was also included in mecciRNA network analysis, we further detected the expression level of phosphorylated STAT3. As shown in Fig. 6F, the results demonstrated the elevated level of phosphorylated STAT3 in MCD-diet group.

Based on these experimental results, we confirmed the reliability of mecciRNA-miRNA networks preliminarily. Additionally, our study also brought a distinct perspective on the understanding of relationship between mecciRNAs and NASH.