The Fibrotic Effects of LINC00663 in Human Hepatic Stellate LX-2 Cells and in Bile Duct-Ligated Cholestasis Mice Are Mediated through the Splicing Factor 2-Fibronectin

Hepatic fibrosis can develop into cirrhosis or even cancer without active therapy at an early stage. Long non-coding RNAs (lncRNAs) have been shown to be involved in the regulation of a wide variety of important biological processes. However, lncRNA mechanism(s) involved in cholestatic liver fibrosis remain unclear. RNA sequence data of hepatic stellate cells from bile duct ligation (BDL) mice or controls were analyzed by weighted gene co-expression network analysis (WGCNA). Based on WGCNA analysis, a competing endogenous RNA network was constructed. We identified LINC00663 and evaluated its function using a panel of assays, including a wound healing assay, a dual-luciferase reporter assay, RNA binding protein immunoprecipitation and chromatin immunoprecipitation. Functional research showed that LINC00663 promoted the activation, migration and epithelial–mesenchymal transition (EMT) of LX-2 cells and liver fibrosis in BDL mice. Mechanistically, LINC00663 regulated splicing factor 2 (SF2)-fibronectin (FN) alternative splicing through the sponging of hsa-miR-3916. Moreover, forkhead box A1 (FOXA1) specifically interacted with the promoter of LINC00663. In summary, we elaborated the fibrotic effects of LINC00663 in human hepatic stellate LX-2 cells and in bile duct-ligated cholestasis mice. We established a FOXA1/LINC00663/hsa-miR-3916/SF2-FN axis that provided a potential target for the diagnosis and targeted therapy of cholestatic liver fibrosis.


Introduction
HF is a repair response of the body after chronic liver injury [1]. It has various causes and is manifested by the deposition of extracellular matrix (ECM) components [2]. Several studies have shown that in both parenchymal and cholestatic liver injury, activated hepatic stellate cells (HSCs) are the main source of ECM [3]. In normal liver, HSCs remain in a nonproliferating, quiescent state. When liver injury occurs, HSCs are activated and converted to myofibroblasts, which are characterized by the expression of α-smooth muscle actin (α-SMA) and collagen, and have proliferative, contractile, inflammatory and chemotactic effects [4,5]. Activated HSCs can produce and deposit more collagen and fibronectin (FN) in the ECM to promote HF [6]. Therefore, we will focus on the mechanism of HSC activation in cholestatic liver fibrosis. Transforming growth factor-β (TGF-β) is an effective cytokine that can activate HSCs to differentiate into myofibroblast-like cells that promote fibrosis and is therefore widely used in activation models of HSCs in vitro [7].
Fibronectin is a macromolecular glycoprotein that constitutes the extracellular matrix and is secreted by HSCs after activation [8]. The pre-mRNA of FN is thought to undergo alternative splicing to produce variants such as extra domain A-FN (EDA-FN), which is crucial in fibrosis [9]. Splicing factor 2 is a clear and found earlier predominant factor in the

Data Collection and Preprocessing
A GSE34640 dataset was downloaded from the Gene Expression Omnibus (GEO) database [17]. In this study, we selected eight samples, including five quiescent mouse HSCs and three bile duct ligation-activated HSCs. Unqualified samples were inspected and rejected. Data were normalized in a limma package of R software (version 3.6.1). Probes were annotated by Affymetrix annotation files. Finally, 25% of genes were screened out to construct a co-expression network using an analysis of variance calculation.

WGCNA
Soft threshold power (β) was selected by a pickSoftThreshold function based on scalefree topology criteria. We used this β to create a weighted adjacency matrix. Then it was converted to a topology matrix (TOM). According to the TOM-based dissimilarity (1-TOM) measurement, genes were used as a hclust function to indicate hierarchical clustering. After hierarchical clustering, strongly linked genes were assigned to the same module. The module eigengene represents the gene expression profiles and average gene expression level of the module. Module membership (MM) represents the relationship between genes and modules. Gene significance (GS) can be considered as the relationship between a single gene and clinical information [15]. The genes of the most important modules were selected to proceed with the analysis.

Identification and Functional Annotation of Key Genes in Co-Expression Networks
We selected the top 5% of genes in the significant module as key genes that had high connectivity in each candidate module. These genes were submitted to the DAVID database [18,19] for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses. There are three main categories of GO enrichment analysis: molecular function, biological process and cellular component. A value of p < 0.01 shows that the difference is statistically significant. The KEGG database can analyze the metabolic pathways and functions of intracellular genome products. A value of p < 0.05 indicates that the difference is statistically significant.

Identification of Hub Genes
The protein-protein interaction (PPI) network was established by key genes using STRING [20] and visualized by Cytoscape [21]. Then, we used the mcode [22] software to screen hub genes in the PPI network (selection criteria: mcode scores > 5, degree cut-off = 2, node score cut-off = 0.2, max depth = 100 and k-score = 2). We selected genes with a degree > 10 as hub genes.

Construction of ceRNA Network
We used online prediction tools, such as Targetscan [23] and miRDB [24], to predict miRNAs that could bind to target mRNAs, and RegRNA 2.0 [25] to predict lncRNAs that could bind to miRNAs. Based on the ceRNA hypothesis [26], lncRNA acts as a miRNA sponge, binds directly to miRNAs, and indirectly regulates the function of mRNAs. Therefore, lncRNAs and mRNAs negatively regulated by miRNAs were selected to construct a ceRNA network. Verification was performed on activated LX-2 cells.

RNA Extraction and Quantitative Real-Time Polymerase Chain Reaction
Total RNA was extracted from cultured cells or mouse tissues using TRIzol (Invitrogen, Waltham, MA, USA). We subsequently used a Nanodrop 2000 to analyze the concentration and purity of RNA. U6 and hsa-miR-3916 levels were detected using a miRNA assay system (Takara Biotechnology Co., Ltd., Dalian, China). To detect mRNA expression, a ReverTra Ace ® qPCR RT Kit (Toyobo, Osaka, Japan) was used to synthesize cDNA using 1 µg total RNA with a reaction volume of 10 µL. Real-time quantitative PCR (RT-qPCR) analysis was performed using THUNDERBIRD ® SYBR ® qPCR Mix (Toyobo). To normalize the data, β-actin or U6 were used as reference genes. Each reaction was performed three times. A 2 −∆∆Ct method was used to calculate relative gene expression levels. Table 1 lists the primers used.

Dual-Luciferase Reporter Assay
In the LINC00663 target hsa-miR-3916 and hsa-miR-3916 target gene SF2 luciferase reporter assay, LX-2 cells were cotransfected with pmirGLO report vectors containing wildtype (WT) LINC00663, mutant (MUT) LINC00663 or WT-SF2, MUT-SF2 and hsa-miR-3916 mimics or miR-NC. In the FOXA1 target LINC00663 luciferase reporter assay, HEK293T cells were cotransfected with a pmirGLO report vectors containing WT-LINC00663, LINC00663 promoter region mutation and FOXA1 or NC plasmids. Forty-eight hours after transfection, we used a dual luciferase reporter assay system (Promega, Madison, WI, USA) to measure luciferase activity. Using Renilla luciferase activity as a baseline, the data were normalized.
2.11. RNA Fluorescence In Situ Hybridization Assay 5 fam-labeled LINC00663 probes (GenePharma) were used for RNA fluorescence in situ hybridization (FISH) assays in LX-2 cells according to the manufacturer's protocol. The probe sequence was 5 -CGTTCTTTCACGTCCCTAGCTGTATTCACTCTCCCTG-3 . Confocal images were captured using a Nikon confocal microscope.

RNA Immunoprecipitation
We performed RNA immunoprecipitation (RIP) experiments using a Magna RIP TM RNA-Binding Protein Immunoprecipitation Kit (Millipore) according to the manufacturer's protocol. The antibody used for RIP was 5 µg of anti-Ago2 (Millipore, 03-110). Purified RNAs were detected by RT-qPCR. Gene-specific primers are presented in Table 2.

Chromatin Immunoprecipitation
PROMO [27,28] and JASPAR [29] were used to predict transcription factors that could bind to the LINC00663 promoter region. Chromatin immunoprecipitation (ChIP) experiments were performed using a SimpleChIP ® Plus Enzymatic Chromatin IP Kit (CST) according to the manufacturer's instructions. The antibody used for ChIP was 5 µg of anti-FOXA1 (Abcam, ab170933). Purified DNA was detected by RT-qPCR. Primers used for PCR in ChIP experiments were the forward sequence 5 -ACAGAGCCAGGTGAAACAGAG-3 and the reverse sequence 5 -AACGATTGGGCG-CTCTAAGG-3 .

Animal Studies
Mice experiments were approved by the Institutional Ethics Committee of China Medical University. In a BDL-induced mice liver fibrosis model, 20 Balb/c male mice aged between 8 and 9 weeks were randomly divided into four groups: sham operation (Sham, n = 5); BDL operation (BDL, n = 5); BDL operation treated with injection of pcDNA3.1-NC (BDL + pcDNA, n = 5); and BDL operation treated with injection of pcDNA3.1-LINC00663 (BDL + pcDNA-LIN, n = 5). Each plasmid sample was injected through the tail vein (2 × 10 6 cells per mouse) after a surgical operation [30]. All mice were killed after 21 days [31]. Liver samples and sera were collected for the analysis of liver function and fibrotic index.

Hydroxyproline Assay and Serum ALT and AST
Hydroxyproline in liver samples was measured using hydroxyproline detection kits (Nanjing Jiancheng Biochemical Institute, Nanjing, China) to detect total collagen content. Serum ALT and AST were tested by an automatic analyzer (Hitachi 7600, Hitachi, Tokyo, Japan).

Wound Healing Assay
Scratches were made with a 200 µL pipette tip 48 h post-transfection and growth media was replaced with serum-free DMEM. Cell migration at 0 h and 48 h at the same location were recorded by taking photos. The percentage of wound healing was quantified as (0-h scratch area-48-h scratch area)/0-h scratch area × 100%.

Statistical Analysis
All data were normalized using a Shapiro-Wilk test. Normal distribution data were analyzed using a t-test, while non-normal data were analyzed using a Mann-Whitney U test. A two-tailed p-value < 0.05 was considered statistically significant.

Data Preprocessing and Co-Expression Network Construction
We downloaded eight raw data sets from the GEO database. After probe conversion and the removal of duplicate genes, a total of 21,747 genes were obtained. Subsequently, we used analysis of variance to calculate and screen out the top 25% of genes (n = 5437) for co-expression network construction. We chose to define the adjacency matrix β = 10 based on scale-free topology criteria. As a result, the index reached 0.84 (Supplementary Figure S1A). In view of a dynamic tree cutting method, 5437 genes were separated into 15 modules (Supplementary Figure S1B).

Identification of Significant Modules
A darkolivegreen module (r = 0.99, p < 0.05) with 1304 genes and darkgrey module (r = −0.89, p < 0.05) with 1072 genes were strongly correlated with cell status after the cellular state was introduced into the weighted network ( Figure 1A). The following analysis calculated the GS and MM of the darkolivegreen module (r = 0.97, p < 0.05, Figure 1B) and the darkgrey module (r = 0.82, p < 0.05, Figure 1C). Therefore, these two modules were identified as important modules.

Functional Annotation
Information on 116 key genes is shown in Supplementary Table S2. For GO analysis, genes were predominantly rich for extracellular protein matrix, extracellular space, ex-

Functional Annotation
Information on 116 key genes is shown in Supplementary Table S2. For GO analysis, genes were predominantly rich for extracellular protein matrix, extracellular space, extracellular exosomes, protein binding and protein homodimerization activity (Supple-

Functional Annotation
Information on 116 key genes is shown in Supplementary Table S2. For GO analysis, genes were predominantly rich for extracellular protein matrix, extracellular space, extracellular exosomes, protein binding and protein homodimerization activity (Supplementary  Table S3). For KEGG analysis, genes were concentrated primarily through ECM-receptor interaction and cell-cycle pathways (Supplementary Table S4).

PPI Network Construction and Hub Gene Identification
A PPI network is shown for 116 key genes contained in 116 nodes and 157 edges ( Figure 1D). Using the mcode [22] software, we eventually obtained seven genes with degree > 10 ( Table 3). Of these, FN attracted our attention due to it being the highest degree. Fibronectin is a macromolecular glycoprotein that constitutes the extracellular matrix and is secreted by HSCs after activation [8]. The pre-mRNA of FN is thought to undergo alternative splicing to produce variants such as extra domain A-FN (EDA-FN), which is crucial in fibrosis [9]. Splicing factor 2 is a clear and found earlier predominant factor in the alternative splicing of FN pre-mRNA, and the level of SF2 expression directly affects the amount of EDA-FN produced [10][11][12]. Therefore, we selected SF2 as a downstream target gene.

Construction and Validation of ceRNA Network
In TGF-β-treated LX-2 cells, both mRNA ( Figure 2A) and protein ( Figure 2B) levels of α-SMA and COL1A1 were significantly increased, showing that cells were successfully activated. Since the activation levels of the activation markers for 48 h were better than that of activation for 24 h, we used a model of activation for 48 h in the following experiments: We verified the expression of SF2, FN, and EDA-FN in activated LX-2 cells and found that the expression of all three was upregulated in activated LX-2 cells compared to control cells ( Figure 2C,D, p < 0.05). Through miRNA target prediction websites, we found that hsa-miR-3916 bound to SF2 and that the expression of hsa-miR-3916 was downregulated in activated LX-2 cells compared to control cells ( Figure 2E, p < 0.05). Similarly, we found that LINC00663 bound to hsa-miR-3916 and the expression of LINC00663 was upregulated in activated LX-2 cells compared to control cells ( Figure 2F, p < 0.05). Taken together, we constructed and verified a LINC00663/hsa-hsa-miR-3916/SF2-FN ceRNA network. Furthermore, subcellular fractionation analysis showed that LINC00663 were localized both in the cytoplasm and nucleus ( Figure 2G).

LINC00663 Regulated the Activation of LX-2 Cells
To study the role of LINC00663 in the activation of LX-2 cells, activated LX-2 were transfected with si-LINC00663 and pcDNA-LINC00663. The expressio LINC00663 was decreased in cells of the si-LINC00663 group compared to those o si-NC group ( Figure 3A, p < 0.05), and increased in the pcDNA-LIN group compar the pcDNA group ( Figure 3B, p < 0.05). The levels of α-SMA and COL1A1 were low the si-LINC00663 group compared to the si-NC group ( Figure 3C,D, p < 0.05) and higher in the pcDNA-LIN group compared to the pcDNA group ( Figure 3E,F, p < These results showed that LINC00663 regulated the activation of LX-2 cells.

LINC00663 Regulated the Activation of LX-2 Cells
To study the role of LINC00663 in the activation of LX-2 cells, activated LX-2 cells were transfected with si-LINC00663 and pcDNA-LINC00663. The expression of LINC00663 was decreased in cells of the si-LINC00663 group compared to those of the si-NC group ( Figure 3A, p < 0.05), and increased in the pcDNA-LIN group compared to the pcDNA group ( Figure 3B, p < 0.05). The levels of α-SMA and COL1A1 were lower in the si-LINC00663 group compared to the si-NC group ( Figure 3C,D, p < 0.05) and were higher in the pcDNA-LIN group compared to the pcDNA group ( Figure 3E,F, p < 0.05). These results showed that LINC00663 regulated the activation of LX-2 cells.

LINC00663 Regulated Epithelial-Mesenchymal Transition and Migration of LX-2 Cells
To study the role of LINC00663 in epithelial-mesenchymal transition (EMT) and migration of LX-2 cells, activated LX-2 cells were transfected with si-LINC00663 and pcDNA-LINC00663. The levels of desmin and vimentin were reduced while that of E-cad was increased by LINC00663 knockdown ( Figure 4A, p < 0.05). The expression of desmin and vimentin was increased and that of E-cad was reduced when LINC00663 was overexpressed ( Figure 4B, p < 0.05). To further validate the hypothesis that LINC00663 promoted migration, wound healing assays were employed. As shown in Figure 4C,D, LINC00663 knockdown alleviated migration and overexpression of LINC00663 aggravated migration of LX-2 cells (p < 0.05). These data suggested that LINC00663 regulated the EMT and migration of LX-2 cells.

LINC00663 Regulated Epithelial-Mesenchymal Transition and Migration of LX-2 Cells
To study the role of LINC00663 in epithelial-mesenchymal transition (EMT) and migration of LX-2 cells, activated LX-2 cells were transfected with si-LINC00663 and pcDNA-LINC00663. The levels of desmin and vimentin were reduced while that of Ecad was increased by LINC00663 knockdown ( Figure 4A, p < 0.05). The expression of desmin and vimentin was increased and that of E-cad was reduced when LINC00663 was overexpressed ( Figure 4B, p < 0.05). To further validate the hypothesis that LINC00663 promoted migration, wound healing assays were employed. As shown in Figure 4C,D, LINC00663 knockdown alleviated migration and overexpression of LINC00663 aggravated migration of LX-2 cells (p < 0.05). These data suggested that LINC00663 regulated the EMT and migration of LX-2 cells.

Overexpression of LINC00663 Aggravated BDL-Induced Hepatic Fibrosis In Vivo
To investigate the effect of LINC00663 in BDL-induced hepatic fibrosis, mice were given pcDNA-LINC00663 (pcDNA-LIN) or a negative control (pcDNA). We found that liver tissues treated with pcDNA-LIN had a higher expression of LINC00663 compared with those of the pcDNA group ( Figure 5A, p < 0.05). Liver hydroxyproline levels (Figure 5B) and alanine aminotransferase and aspartate transaminase serum levels (Table 4) in pcDNA-LIN-injected mice were also increased compared to BDL-induced mice injected with pcDNA. However, the administration of pcDNA-LIN exacerbated BDL-induced hepatic fibrosis, as indicated by macroscopic examination, HE and Masson staining, IHC and WB of α-SMA and COL1A1 ( Figure 5C,D). These data suggested that LINC00663 might exacerbate BDL-induced hepatic fibrosis.

Overexpression of LINC00663 Aggravated BDL-Induced Hepatic Fibrosis In Vivo
To investigate the effect of LINC00663 in BDL-induced hepatic fibrosis, mice were given pcDNA-LINC00663 (pcDNA-LIN) or a negative control (pcDNA). We found that liver tissues treated with pcDNA-LIN had a higher expression of LINC00663 compared with those of the pcDNA group ( Figure 5A, p < 0.05). Liver hydroxyproline levels ( Figure 5B) and alanine aminotransferase and aspartate transaminase serum levels (Table 4) in pcDNA-LINinjected mice were also increased compared to BDL-induced mice injected with pcDNA. However, the administration of pcDNA-LIN exacerbated BDL-induced hepatic fibrosis, as indicated by macroscopic examination, HE and Masson staining, IHC and WB of α-SMA and COL1A1 ( Figure 5C,D). These data suggested that LINC00663 might exacerbate BDL-induced hepatic fibrosis.

LINC00663 Regulated Activation, EMT, and Migration of LX-2 Cells through Hsa-miR
Using RegRNA 2.0, we found two possible binding sites for LINC00663 hsa-miR-3916. We constructed reporter vectors in which potential binding sites i sequence for LINC00663 were all or individually mutated (MUT3-LINC00663 w mutant type of both binding sites; Figure 6A). As expected, cotransfectio MUT3-LINC00663 with a hsa-hsa-miR-3916 mimic did not repress luciferase ac ( Figure 6A, p < 0.05). To study the role of hsa-hsa-miR-3916, activated LX-2 cells transfected with a hsa-hsa-miR-3916 mimic or inhibitor. The expression hsa-hsa-miR-3916 increased in the hsa-hsa-miR-3916 mimic group and decreased i hsa-hsa-miR-3916 inhibitor group compared to the miR-NC group ( Figure 6B Table 4. Serum ALT and AST levels of mice in each group (mean ± SD, n = 7).

LINC00663 Regulated SF2-FN Alternative Splicing by Sponging Hsa-miR-3916
Using Targetscan and miRDB, we identified a potential binding site for hsa-miR-3916 and SF2. We constructed a reporter vector with a mutated potential binding site for SF2 sequences ( Figure 7A). As expected, co-transfection of MUT-SF2 and a hsa-miR-3916 mimic failed to suppress luciferase activity ( Figure 7A, p < 0.05). To investigate whether LINC00663 regulated SF2, FN and EDA-FN expression, activated LX-2 cells were transfected with si-LINC00663 and pcDNA-LIN. LINC00663 knockdown reduced SF2 and EDA-FN expression ( Figure 7B,C, p < 0.05), whereas LINC00663 overexpression promoted SF2 and EDA-FN expression ( Figure 7D-E, p < 0.05) in activated LX-2 cells; however, neither of these regulated the expression of FN, indicating that LINC00663 regulated SF2-FN alternative splicing ( Figure 7D,E, p > 0.05). To further demonstrate that LINC00663 regulated SF2-FN alternative splicing by sponging hsa-miR-3916, we performed a rescue assay in activated LX-2 cells. A miR-3916 inhibitor rescued the reduction of SF2 and EDA-FN in cells regulated by LINC00663 ( Figure 7F,G, p < 0.05), whereas a hsa-miR-3916 mimic abolished the elevation of SF2 and EDA-FN in LINC00663-regulated cells (Supplementary Figure S2E,F, p < 0.05). RNA immunoprecipitation assays were performed in LX-2 cells using Ago2 antibody to investigate the presence of interactions between LINC00663, hsa-miR-3916, and SF2. These three molecules were found to be significantly enriched in the anti-Ago2 group compared to the IgG group ( Figure 7H, p < 0.05) implying a competitive regulatory relationship between LINC00663, hsa-miR-3916, and SF2.

LINC00663 Regulated by the Transcription Factor FOXA1
We used PROMO and JASPAR to predict the binding site of FOXA1 to the LINC00663 promoter region ( Figure 8A). Wild and mutated promotor sequences were constructed in a pGL3-basic vector. A dual-luciferase activity test showed that co-transfection of wild-LINC00663 and pGL3-FOXA1 increased luciferase activity ( Figure 8B, p < 0.05) illustrating that FOXA1 bound to the promoter region of LINC00663. Furthermore, FOXA1 expression was upregulated in activated LX-2 cells compared to a control group ( Figure 8C,D, p < 0.05). To investigate the relationship between FOXA1 and LINC00663, activated LX-2 cells were transfected with si-FOXA1 or pEGFP-FOXA1. We found that FOXA1 upregulated LINC00663 ( Figure 8E,F, p < 0.05). Moreover, we performed ChIP assays in LX-2 cells using FOXA1 antibody. We found that the LINC00663 promoter was significantly enriched using anti-FOXA1 compared to IgG control antibody, suggesting a direct binding relationship between FOXA1 and the LINC00663 promoter region ( Figure 8G, p < 0.05). Together, these results identified an important regulatory axis whereby LINC00663 that was regulated by FOXA1 sponged hsa-miR-3916 and regulated SF2-FN alternative splicing expression in cholestatic liver fibrosis ( Figure 8H). performed ChIP assays in LX-2 cells using FOXA1 antibody. We found that the LINC00663 promoter was significantly enriched using anti-FOXA1 compared to IgG control antibody, suggesting a direct binding relationship between FOXA1 and the LINC00663 promoter region ( Figure 8G, p < 0.05). Together, these results identified an important regulatory axis whereby LINC00663 that was regulated by FOXA1 sponged hsa-miR-3916 and regulated SF2-FN alternative splicing expression in cholestatic liver fibrosis ( Figure 8H).

Discussion
In recent years, with the widespread recognition of cholestatic liver fibrosis and the gradual increased use of serum anti-mitochondrial antibody, the incidence of PBC and PSC has shown a rapid upward trend. The incidence of PBC in China increased from 2.16/100,000 in 1991 to 2000 to 8.99/100,000 in 2011-2020; PBC cases have increased worldwide [32]. At present, epidemiological data on PSC in China is lacking, but data from Northern Europe and North America have shown that the incidence of PBC has been steadily increasing year after year [33].
The hypothesis that is the basis for ceRNAs is that lncRNAs can affect gene expression by competing with mRNAs for binding to miRNAs [34]. Research reports have highlighted how lncRNAs, such as lncRNA-p21 [35], GAS5 [36] and PVT1 [37], can sponge target miRNAs to influence the activation of HSCs, which may occupy an important regulatory place in the HF process. In the present study, we explored the role of LINC00663 in the progression of HF for the first time. As expected, our data revelated that LINC00663 was overexpressed in activated LX-2 cells, and overexpression of LINC00663 accelerated the progression of HF through sponging miR-3916 to upregulate SF2-FN expression. In addition, previous studies confirmed that LINC00663 promoted the migration abilities of cells by regulating the AKT/mTOR pathway [38]. This is consistent with our study that LINC00663 overexpression can promote the migration of LX-2 cells. The mechanism of hsa-miR-3916 in liver fibrosis has not been reported yet. In prostate cancer, high expression of hsa-miR-3916 reduced the protein level of pyruvate dehydrogenase kinase 1, thereby reducing the risk of prostate cancer [39]. However, in Conjunctival Melanoma patients, upregulation of hsa-miR-3916 was associated with a higher risk of local recurrence [40].
FN is a macromolecular glycoprotein constituting the extracellular matrix, secreted by HSCs after activation and is an important molecule for the formation of HF [8]. As an important substrate of TGF-β, the alternative spliced form of FN, EDA-FN participates in the induction of fibroblast proliferation and differentiation, EMT, fibroblast extracellular matrix production and tissue fibrosis [41,42]. The interaction of TGF-β with EDA-FN promotes the transformation of fibroblasts into α-SMA-expressing myofibroblasts and promotes fibrosis [43,44]. Splicing factor 2 is a kind of serine/arginine rich protein (SR protein) [45]. The capability of SR proteins is to facilitate selection of splice sites by binding to exon splicing enhancers [45]. Just as SF2 is a predominant factor in alternative splicing of FN pre-mRNA, the level of SF2 directly affects the production of EDA-FN [10,11]. SF2 can control alternative splicing of the tumor suppressor to disable their function. Therefore, it also is considered to be an oncoprotein [46].
Long non-coding RNAs are often regulated by transcription factors and are mainly transcribed by RNA polymerase II and 5 -capped, and polyadenylated in the same manner as protein coding mRNAs [31,47]. Forkhead box A is a transcribed protein that has regulatory effects on liver metabolism and development, and is a major transcription factor in epithelial lineage differentiation [48][49][50]. In breast cancer cells, FOXA1 and DSCAM-AS1 form a positive feedback loop to promote cancer cell proliferation [51]. Through our study, we found that FOXA1 induced the upregulation of LINC00663 in LX-2 cells and had a strong affinity to the promoter of LINC00663.
We exhaustively elucidated the relationships between FOXA1/LINC00663/hsa-miR-3916/SF2-FN in the modulation of the progression of cholestatic liver fibrosis. However, limitations existed in our study. We did not verify the expression of related molecules in clinical samples, so we could not evaluate the diagnostic value of LINC00663 as a biomarker, in human hepatic stellate LX-2 cells and in bile duct-ligated cholestasis mice.

Conclusions
In summary, we elaborated the fibrotic effects of LINC00663 in human hepatic stellate LX-2 cells and in bile duct-ligated cholestasis mice. We established a FOXA1/LINC00663/ hsa-miR-3916/SF2-FN axis that provided a potential target for the diagnosis and targeted therapy of cholestatic liver fibrosis.

Data Availability Statement:
The data involved in this study were available through NCBI GEO (GSE34640).

Conflicts of Interest:
The authors declare no conflict of interest.