Long Noncoding RNA HAFML Promotes Migration and Invasion of Rheumatoid Fibroblast-like Synoviocytes

Rheumatoid arthritis (RA), a common chronic inflammatory disease, is characterized by permanent inflammation and destruction of the distal joint. The synovial lining layer becomes a mass of “pannus” tissue that aggresses into the adjacent articular cartilage and subchondral bone. Fibroblast-like synoviocytes (FLSs), the major resident cells in synovial tissue, have an essential role in promoting synovial inflammation and aggression in RA (1). RA FLSs particularly exert a surprisingly invasive phenotype, including increased migration and a powerful ability to invade joint cartilage and bone, known as a “tumor-like transformation” or “imprinted aggressors” (2, 3). Emerging evidence implies that selectively targeting FLS might be promising for improving rheumatoid joint damage without suppressing systemic immunity (4, 5). Recent studies indicate that the invasive phenotype of RA FLSs is closely related to epigenetic changes (6, 7). However, it is still unclear what causes FLSs to act as imprinted invaders in promoting joint destruction in RA.

Long noncoding RNAs (lncRNAs), one type of noncoding RNAs that are defined as long transcripts (>200 nt), have been considered a novel layer of modulation in a variety of cellular biological processes. lncRNAs play important roles in modulating not only nuclear transcriptional activation, heterochromatin formation, telomere maintenance, and X chromosome inactivation but also cytoplasmic protein trafficking and mRNA translation and decay (8–11). More importantly, there is a great deal of evidence indicating that dysregulation of lncRNAs contributes to the initiation and development of various human disorders, including cancer, cardiovascular diseases, and neurodegeneration (12–15). Intriguingly, accumulating studies have shown that lncRNAs are emerging as key regulators of the immune system, and they play important roles in the pathogenesis of several autoimmune diseases, such as RA and systemic lupus erythematosus (16–19). Recent reports show that some lncRNAs are involved in RA pathogenesis; for instance, lncRNA LERFS negatively modulates synovial invasion and proliferation (20), lncRNA GAPLINC regulates the tumor-like biological behaviors of RA FLSs as microRNA sponges (21), and lncRNA PICSAR promotes rheumatoid synovial invasion by sponging miR-4701-5p (22); however, their contribution to the pathogenesis of RA remains largely unknown.

In this study, we identified an lncRNA, ENST00000509194, that has higher-than-normal expression levels in rheumatoid FLSs and synovial tissues and explored its critical regulation in the migratory and invasive behaviors of RA FLSs. This lncRNA functions through interactions with human Ag R (HuR, also called ELAVL1), an RNA-binding protein that stabilizes mRNAs by binding to conserved AU-rich elements within 3′ UTRs and preventing gene degradation (23). We further demonstrated that the formation of the ENST00000509194–HuR complex is required for HuR to bind to and stabilize the mRNA of endocytic adaptor molecule APPL2, thereby promoting its protein expression. Considering its close association with HuR and FLS migration, we named ENST00000509194 HAFML (HuR-associated fibroblast migratory lncRNA). Our data suggest that HAFML is an important modulator of FLS migration and invasion, and that increased synovial HAFML levels may promote rheumatoid joint damage. Because fibroblast migration is a critical step for organ fibrosis, our findings also imply that targeting HAFML, to our knowledge, might be a novel potential therapy for dysregulated fibroblast-associated disorders.

The main reagents used in this study are listed as follows: recombinant human TNF-α (210-TA-100; R&D Systems, Bio-Techne), methotrexate (MTX; Pfizer), dexamethasone (DXM; Tianxin), DMEM (06-1055-57-1ACS; Biological Industries), BD Matrigel Basement Membrane Matrix (356234; BD Biosciences); Coomassie Blue (P0017) and 5× Laemmli Sample buffer (P0015 L) from Beyotime; FBS (16000044; Life Technologies) and Lipofectamine 3000 (L3000015; Invitrogen) from Thermo Fisher Scientific; Polybrene (107689), TRI Reagent (T9424), rhodamine-labeled phalloidin (P1951), Fluoroshield Antifade Mounting Medium (F6182), ECL (WBULS0500), actinomycin D (Act D; 5 μg/ml, A9415-5MG), and LPS from MilliporeSigma; and 4% (w/v) paraformaldehyde (E672002), 1× PBS (B540626), xylene (A530011), ethanol (A500737), Proteinase K (A004240), 10% formamide (A600211), DAPI (A606584), methanol (A506806), 0.1% crystal violet (A600331), 0.2% (v/v) Triton X-100 (A600198), 5% nonfat milk (A600669), and TBS (A500027)/Tween 20 (A600560) from Sangon Biotech. The following Abs were used: anti-HuR (1:1000, ab136542; Abcam), anti-APPL2 (1:1000, 14294-1-AP; Proteintech), anti–β-Tubulin (1:10,000, T0198; MilliporeSigma), anti-rabbit IgG (1:5000, 7074; Cell Signaling Technology), and anti-mouse IgG (1:5000, 7076; Cell Signaling Technology).

We obtained rheumatoid synovial tissues from 27 active patients with RA (23 women and 4 men, aged 57.0 ± 9.9 y) who underwent synovectomy of the knee joint at the First Affiliated Hospital of Sun Yat-sen University, Guangzhou, China. RA was diagnosed in accordance with the 2010 American College of Rheumatology/European League Against Rheumatism classification criteria (24). We obtained healthy control synovial tissues from 23 subjects (20 women and 3 men, aged 55.4 ± 10.1 y) who underwent traumatic single above-knee amputation and had no history of arthritis. There were no significant differences in sex or age between the RA patients and the health control subjects (HCs).

Total RNA was isolated with TRI Reagent from FLSs. The RNA quantity and quality were measured by a NanoDrop ND-1000 (Thermo Fisher Scientific). Standard denaturing agarose gel electrophoresis was performed to evaluate the RNA integrity. Sample labeling and array hybridization were performed according to the Agilent One-Color Microarray-Based Gene Expression Analysis protocol (Agilent Technology). After removal of the rRNA, the mRNA was purified from the total RNA (mRNA-ONLY Eukaryotic mRNA Isolation Kit; Epicenter). Then, the purified RNA was amplified and transcribed into fluorescent cRNA along the entire length of the transcripts without a 3′ bias using a random priming method (Arraystar Flash RNA Labeling Kit; Arraystar). The labeled cRNAs were hybridized onto the Human LncRNA Array, version 3.0 (Arraystar). After hybridization and washing, the arrays were scanned using the Agilent Scanner G2505C, and array images were acquired and analyzed by Agilent Feature Extraction software, version 11.0.1.1. Quantile normalization and subsequent data processing were accomplished using the GeneSpring GX v12.1 software package (Agilent Technologies). Volcano plot filtering was used to identify the differentially expressed lncRNAs with statistical significance. The threshold to screen upregulated or downregulated lncRNAs was identified at a fold change of >2.0 and with p < 0.05. A volcano plot was generated using the R package “ggplot2.” Heatmaps representing differentially regulated lncRNAs were generated using the R package. The microarray data discussed in this article were deposited in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database (accession number GSE181614).

The 3′ RACE System and 5′ RACE System for Rapid Amplification of cDNA Ends (18373019, 18374058; Invitrogen, Thermo Fisher Scientific) were used for rapid amplification of the 5′ and 3′ ends of HAFML. RACE was performed according to the manufacturer’s protocols.

The protein-coding potential of HAFML was assessed by the bioinformatics tools CPC2 (http://cpc2.gao-lab.org) and the Coding Potential Assessment Tool (https://wlcb.oit.uci.edu/cpat/). A lower CAPT/Coding Potential Calculator 2 score indicates a low probability of protein-coding potential. Possible open reading frames were analyzed using ORF-Finder (https://www.ncbi.nlm.nih.gov/orffinder/) within HAFML and are listed by nucleotide position and amino acid sequence (Supplemental Table I).

Total RNA was extracted and transcribed from the FLSs using the PrimeScript RT Reagent Kit (RR036A, RR420A; Takara Bio) according to the manufacturer’s instructions. Quantitative real-time PCR (RT-qPCR) was executed using the Bio-Rad CFX96 system (Bio-Rad). To quantify the relative expression of each gene, we normalized the cycle threshold (Ct) values to the endogenous reference values (ΔCt = Ct target − Ct GAPDH) and compared using a calibrator and the ΔΔCt method (ΔΔCt = ΔCt sample − ΔCt calibrator). All experiments were conducted in triplicate.

The cells (1 × 10⁶) were collected using a PARIS kit (AM1921; Thermo Fisher Scientific) to extract the nuclear and cytoplasmic RNA according to the manufacturer’s protocol. The extracted RNA was assessed by RT-qPCR.

HAFML RNA fluorescence in situ hybridization (FISH) was performed using Stellaris FISH probes (LGC, Biosearch Technologies) according to the manufacturer’s instructions. For FLSs, the cells were grown on coverslips (WHB-24-CS; WHB Scientific) in a 24-well culture plate (702001; NEST Biotechnology). The cells were fixed with 4% (w/v) paraformaldehyde in 1× PBS for 10 min at room temperature and then permeabilized with 70% ethanol for 1 h at 4°C. For synovial tissues, the synovial tissue slides were deparaffinized in xylene and hydrated in decreasing concentrations of 100% ethanol to 70% ethanol, in which they were permeabilized for 30 min. Then, the slides were incubated in 70% ethanol for 1 h at room temperature. The tissue sections were further digested by incubation with 10 μg/ml Proteinase K at 37°C. Then, the coverslips (with FLS cells) or slides (with synovial tissues) were washed one time with wash buffer A (SMF-WA1-60; LGC) for 5 min at room temperature and incubated with Stellaris RNA FISH Hybridization Buffer (SMF-HB1-10; LGC) containing 10% formamide and 1000 nM HAFML-specific Stellaris probe labeled with Quasar 570 in the dark at 37°C overnight. The samples were then washed once with wash buffer A and once with wash buffer A containing 1 μg/ml DAPI; both were washed in the dark at 37°C for 30 min, washed with wash buffer B (SMF-WB1-20; LGC) for 5 min at room temperature, and mounted on microscope slides using Fluoroshield Antifade Mounting Medium. Images were acquired with an Olympus BX63 microscope (Olympus Corporation).

FLSs were transfected with 100 nM small interfering RNA (siRNA) using Lipofectamine 3000 according to the manufacturer’s instructions in six-well plates (703001; NEST). After 48 h, the transfection efficiency was measured by Western blotting or RT-qPCR. siRNAs specific for HAFML, HuR, and APPL2 were purchased from RiboBio.

We purchased HAFML shRNA lentiviruses and HAFML- and HuR-overexpressing lentiviruses from Genechem. RA FLSs were cultured in six-well plates until 70% confluence and infected with lentiviruses in the presence of polybrene (10 μg/ml) at an MOI of 20. The cells were cultured for at least 72 h before further experiments were performed. The infection efficiency was examined using RT-qPCR or Western blotting. We yielded a transfection rate of ∼55–72% (see Supplemental Fig. 1A, 1B) using our modified transfection system.

For the in vitro migration assay, a total of 3 × 10⁴ cells were suspended in 300 μl of DMEM without FBS and seeded in the top chamber of 24-well plate Transwell inserts (3353097; Falcon, Corning Life Sciences). The lower chamber contained 600 μl DMEM with 20% FBS. After incubation at 37°C in 5% CO₂ for 6 h, the inserts were fixed in methanol for 15 min and stained with 0.1% crystal violet for 15 min. Cells on the filter’s upper surface were removed using a cotton swab. Migrated cells were quantified using an Olympus BX63 microscope (Olympus Corporation) to count five random fields for each assay. For the in vitro invasion assay, similar experiments were conducted using Transwell inserts coated with BD Matrigel Basement Membrane Matrix, with 6 × 10⁴ cells seeded on the top chamber. Then, the plate was incubated at 37°C in 5% CO₂ for 12 h.

FLSs were cultured in six-well plates in complete media until 80–90% confluence. Then, the cells were serum starved for 12–24 h and wounded with 200-μl pipette tips. The six-well plates were washed three times with 1× PBS to remove any detached cells, and then the remaining cells were grown in DMEM with 10% FBS. After incubation at 37°C for 48 h, migration was quantified by counting the cells that had moved into the wounded area.

RA FLSs were cultured on coverslips in a 24-well culture plate. When the cells were ∼70–80% confluent, they were serum starved for 24 h and then wounded with 200-μl pipette tips. The 24-well plates were washed three times with 1× PBS to remove the detached cells, and then the remaining cells were grown in DMEM with 10% FBS for 6–8 h. Then, the cells were fixed with 4% paraformaldehyde for 10 min and permeabilized with 0.2% (v/v) Triton X-100 for 10 min at room temperature. To detect F-actin and pseudopodia organization, we stained the cells with rhodamine-labeled phalloidin for 30 min at room temperature. The cell nuclei were stained with DAPI, and then the cells were mounted on microscope slides using Fluoroshield Antifade Mounting Medium. Images were acquired with an Olympus BX63 microscope (Olympus Corporation).

Sponges containing FLSs and human cartilage were inserted into the skin of SCID mice to determine the in vivo invasion of RA FLSs. The detailed procedure and scoring method were conducted as described previously (25).

Cells were cultured in 96-well plates (701001; NEST) until 80–90% confluence and then incubated with 50 μM 5-ethynyl-2′-deoxyuridine (EdU) in DMEM containing 10% FBS for 12 h. A Cell-Light EdU Apollo567 In Vitro Kit (C10310-1; RiboBio) was used to measure the FLS proliferation according to the manufacturer’s instructions. Images were acquired with a Leica DMI8 microscope (Leica Microsystems).

We detected the cell viability using a Cell Counting Kit-8 (CCK-8) (CK04; Dojindo Laboratories). In brief, FLSs were cultured in 96-well plates in complete media until they reached 70–80% confluence. Then, the medium was discarded, and 10 μl of the CCK-8 solution was added to 100 μl of complete medium. The plate was incubated for 1–4 h in a humidified incubator (37°C, 5% CO₂). The absorbance was measured at 450 nm using a Sunrise absorbance reader (TECAN).

FLS apoptosis was assessed by staining cells with FITC Annexin V or allophycocyanin Annexin V and propidium iodide (PI) (556547, 550475; BD Biosciences) according to the manufacturer’s instructions. In brief, FLSs were suspended in 1× binding buffer at a concentration of 1 × 10⁶ cells/ml. The cell suspension (100 μl) was then transferred to a 5-ml culture tube, mixed with 5 μl allophycocyanin Annexin V or 5 μl FITC Annexin V and 5 μl PI, and incubated for 15 min at room temperature in darkness. The samples were analyzed by flow cytometry CytoFLEX (Beckman Coulter) within 1 h.

Caspase activity was measured using a commercially available Caspase-Glo 3/7 Assay kit (G8090; Promega) according to the manufacturer’s instructions. Luminescence was measured using a Multimode Microplate Reader Infinite F500 (TECAN).

Full-length sense and antisense HAFML were synthesized using a TranscriptAid T7 High Yield Transcription Kit (K0441; Thermo Fisher Scientific) and labeled with biotin on the 3′ terminus of the RNA strand using a Pierce RNA 3′ End Desthiobiotinylation Kit (20163; Thermo Fisher Scientific). The pulldown assay was performed according to the instructions for the Pierce Magnetic RNA-Protein Pull-Down Kit (20164; Thermo Fisher Scientific). In brief, 50 pmol biotin-labeled sense or antisense HAFML was incubated with 50 μl streptavidin magnetic beads at room temperature for 30 min. Then, cell lysate supernatants (3 mg/ml) from FLSs (1 × 10⁷ cells) were incubated with RNA-bound magnetic beads or beads only (as a negative control [NC]) by rotating them at 4°C for 1 h, followed by washes and protein elution. The retrieved proteins were visualized by SDS-PAGE and Coomassie Blue staining.

After RNA pulldown, equal amounts of samples pulled down by beads only, sense, and antisense HAFML were analyzed by LC–MS/MS (Shanghai Applied Protein Technology). Protein identification was retrieved from the UniProtKB/Swiss-Prot database using Proteome Discoverer 1.4 (Thermo Fisher Scientific).

Protein-binding prediction of HAFML was evaluated by the bioinformatics tools catRAPID (http://www.tartaglialab.com), beRBP (http://bioinfo.vanderbilt.edu/beRBP/predict.html), and RBPmap (http://rbpmap.technion.ac.il).

A bicinchoninic acid protein assay kit (P0010; Beyotime) was used to detect the protein concentrations. Equal amounts of protein were solubilized in 5× Laemmli sample buffer, boiled for 5 min, separated by SDS-PAGE, and transferred onto polyvinylidene difluoride membranes (1620177; Bio-Rad). The membranes were incubated with the following primary Abs: anti-HuR (1:1000, ab136542; Abcam), anti-APPL2 (1:1000, 14294-1-AP; Proteintech), and anti–β-Tubulin (1:10,000, T0198; Millipore Sigma) in TBS–Tween 20 containing 5% nonfat milk at 4°C overnight. The membranes were incubated with the appropriate secondary Abs (anti-rabbit IgG, 7074; anti-mouse IgG, 7076; both from Cell Signaling Technology) for 1 h at room temperature. Immunoreactive bands were visualized using ECL. Each shown image is representative of at least three similar independent experiments.

RIP assays were performed using the EZ-Magna RIP (RNA immunoprecipitation) Kit (17-701; MilliporeSigma) following the manufacturer’s instructions. In brief, 1 × 10⁷ cells were harvested and lysed with RIP lysis buffer. For each Ab, magnetic beads (50 μl) were washed twice, collected, and then resuspended in 100 μl RIP wash buffer. HuR Ab (5 μg, ab136542; Abcam) or normal mouse IgG (used as the NC) was added to the tube and incubated for 30 min at room temperature. The Ab–bead complexes were washed, mixed with 100 μl cell lysates in 900 μl RIP buffer, and then incubated at 4°C overnight. Immunoprecipitated products were then washed, collected, and treated with proteinase K at 55°C for 30 min. Finally, total RNA was extracted from the immunoprecipitated products and used for RT-qPCR analysis.

RNA quantity and quality were measured by a NanoDrop ND-1000. RNA integrity was assessed by standard denaturing agarose gel electrophoresis. Total RNA (2 μg) was used to prepare the sequencing library. After removal of rRNA, mRNA was enriched by oligo (dT) magnetic beads, and then the RNA sequencing library was prepared by a KAPA Stranded RNA-Seq Library Prep Kit (Illumina). The completed libraries were validated by an Agilent 2100 Bioanalyzer and quantified by an absolute quantification quantitative PCR method. The barcoded libraries were mixed, denatured into single-stranded DNA in 0.1 M NaOH, amplified in situ using a TruSeq SR Cluster Kit v3-cBot-HS (GD-401-3001; Illumina), and then sequenced for 150 cycles from both ends on an Illumina X-ten/NovaSeq instrument, to sequence the libraries. After quality control, the raw sequencing data were analyzed to generate the expression profiles, differentially expressed genes, and differentially expressed transcripts. The RNA sequencing data discussed in this article were deposited in the NCBI GEO database (GSE188208).

The RNA–RNA interaction probability of HAFML and APPL2 mRNA was predicted by the bioinformatics tools LncRRIsearch (http://rtools.cbrc.jp/LncRRIsearch/index.cgi) and LncTar (http://www.cuilab.cn/lnctar).

Cells were treated with the given siRNAs for 48 h and treated with Act D (5 μg/ml) to block the synthesis of new RNA. Cells were then harvested to extract total RNA at 0, 3, 6, and 9 h after Act D treatment for RT-qPCR.

Data are presented as the mean ± SD from at least three independent experiments. Two-group comparisons were analyzed using an independent two-tailed Student t test. For comparisons of three or more groups, one-way ANOVA followed by Bonferroni’s post hoc comparisons were used. A p value <0.05 was considered statistically significant. Statistical analyses of all data were performed using SPSS 13.0 (IBM).

All human studies were performed according to the Declaration of Helsinki principles and approved by the Medical Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University. Written informed consent to participate in this study was obtained from the patients. All animal studies were approved by the Animal Care and Ethics Committee of Sun Yat-sen University.

To explore the expression pattern of rheumatoid lncRNAs, we used a microarray to establish the lncRNA expression profiles in FLSs separated from the knee-joint synovial tissues of RA patients and HCs. As shown in (Fig. 1A (left panel) and 1B, the differential lncRNA expression patterns (with fold changes of at least 2.0 and p < 0.05) were obtained using hierarchical clustering analyses. We identified 357 upregulated and 391 downregulated lncRNAs in the RA group compared with the HC group. These data had been deposited in the NCBI GEO database (GSE181614; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE181614). We showed that 20 lncRNAs were upregulated in RA FLS by >4.5-fold (Fig. 1A, right panel).

In this study, we focused on lncRNA HAFML encoded by a gene at position chromosome 4 q34.3 in the genome, also termed ENST00000509194. A schematic graphic of the genomic locus of HAFML is shown in Supplemental Fig. 1C. We further confirmed a significant increase in HAFML expression in RA FLSs compared with HC FLSs using RT-qPCR (Fig. 1C). We verified HAFML to be a 533-nt transcript by using 5′- and 3′-RACE (Fig. 1D). Analysis of its coding potential strongly suggested that HAFML is a noncoding RNA (Supplemental Fig. 1D). Because inflammatory mediators play important roles in modulating the biological functions of RA FLSs, we evaluated the effects of TNF-α and LPS on the expression of HAFML. We found that treatment with TNF-α or LPS mitigated the expression of HAFML in RA FLSs (Fig. 1E). We also demonstrated that treatment of RA FLSs with MTX or DXM, both classical agents for RA treatment, resulted in diminished HAFML expression (Fig. 1F). FISH staining showed that HAFML is located primarily in the cytoplasm (Fig. 1G), which was verified by nuclear/cytoplasmic fractionation (Fig. 1H); this implies that HAFML may exert its biological function in the cytoplasm. Furthermore, FISH staining of paraffin-embedded synovial tissue sections validated the expression of HAFML, which was increased in established RA patients compared with HC subjects (Fig. 1I). Taken together, our data suggest that increased synovial HAFML might be related to joint damage in RA.

To evaluate the potential role of HAFML, we used siRNA to knock down HAFML expression in primary RA FLSs. To exclude nonspecific interference, we constructed three different sequences of siRNA oligonucleotides for HAFML. As shown in Supplemental Fig. 1E, transfection with all three siRNA oligonucleotides reduced the HAFML levels, but the suppressive effect of both siRNA-1 and siRNA-2 was stronger than that of siRNA-3. Accordingly, we used HAFML siRNA-1 and siRNA-2 for subsequent experiments. To explore the role of HAFML in modulating directed migration, we determined the chemotaxis migration of FLSs using a Transwell assay with lncRNA siRNA-transfected RA FLSs. As shown in (Fig. 2A, we observed that RA FLSs with HAFML knockdown had diminished FBS-induced migration compared with the control siRNA. A monolayer wound-scratch assay was also used to evaluate the role of HAFML in regulating FLS migration. We found that HAFML knockdown significantly reduced RA FLS migration compared with control siRNA (Fig. 2B). In contrast, we determined that the migration of RA FLSs overexpressing HAFML was increased compared with that of FLSs transfected with the empty vector control (Fig. 2C, 2D, Supplemental Fig. 1F).

The aggressive behavior of RA FLSs is critical for rheumatoid joint destruction; increase in vitro invasion of RA FLSs is positively correlated with the joint damage rate in patients with RA (26). Thus, we evaluated the effect of HAFML knockdown on the invasion of RA FLSs using Matrigel-coated Transwell membranes. As shown in (Fig. 2E, RA FLSs with HAFML knockdown exhibited reduced invasion compared with that of control siRNA. However, we demonstrated increased invasion in HAFML-overexpressing RA FLSs compared with the empty vector control (Fig. 2F). Because matrix metalloproteinases (MMPs) play roles in controlling the aggressive behavior of RA FLSs, we determined the role of HAFML in regulating MMP expression. However, HAFML knockdown did not affect the TNF-α–induced expression of MMP-3, MMP-9, or MMP-13 (Fig. 2G).

Furthermore, we found that addition of mitomycin, a cell proliferation inhibitor, did not affect the migration or invasion of HAFML-overexpressed RA FLSs (Supplemental Fig. 1H, 1I), indicating that HAFML overexpression-induced migration and invasion of RA FLSs are not associated with their proliferation.

Cell mobility depends critically on dynamic reorganization of the actin cytoskeleton. To investigate the role of HAFML in controlling the actin reorganization of RA FLSs, we used fluorescent phalloidin staining to visualize polymerized actin in migrating cells after wounding in HAFML siRNA-transfected or control siRNA-transfected RA FLSs. We observed that RA FLSs transfected with the control siRNA exhibited ruffling or flat lamellipodia and filopodia at their leading edges; however, cells transfected with HAFML siRNA displayed diminished lamellipodia and filopodia formation (Fig. 2H). This finding suggests that HAFML has an important role in controlling the formation of membrane protrusions in migrating cells.

To further confirm the role of HAFML in synoviocyte migration, we evaluated the effect of HAFML on the migration and invasion of FLSs from HCs. Consistent with our findings in RA FLSs, we demonstrated that HAFML knockdown reduced the migration and invasion of HC FLSs compared with NCs (Fig. 2I–K); however, HAFML overexpression increased the migration and invasion of HC FLSs (Fig. 2L–N). Our data suggest an important role of HAFML in regulating the migration and invasion of FLSs.

Finally, we evaluated the effect of HAFML knockdown on in vivo invasion by RA FLSs. A SCID mouse coimplantation model was used to detect the in vivo effect of HAFML knockdown on the invasion of RA FLSs into cartilage. We coimplanted RA FLSs carrying HAFML shRNA or the control vector into the left or right flanks of the mice, respectively. We demonstrated that RA FLSs infected with HAFML shRNA displayed significant restraint of invasion into cartilage compared with the cells infected with control shRNA (Supplemental Fig. 1G, (Fig. 2O). Collectively, our findings suggest that HAFML promotes the migration and aggression of RA FLSs.

Increased proliferation and reduced apoptosis of FLSs contributes to synovial hyperplasia, ultimately involving the progressive joint destruction of RA. Thus, we evaluated the contribution of HAFML to the proliferation and apoptosis of RA FLSs. We first determined that HAFML knockdown did not affect RA FLS proliferation compared with the control siRNA (Fig. 3A, 3B). The results from the CCK-8 assay also demonstrated that HAFML knockdown did not influence the viability of RA FLSs (Fig. 3C). We also determined that HAFML overexpression did not affect the proliferation and viability of RA FLSs compared with the empty vector control (Fig. 3D–F).

Next, we determined the effect of HAFML knockdown on apoptosis of RA FLSs using Annexin V and PI staining to measure the apoptosis of RA FLSs by flow cytometry. We demonstrated that the total apoptotic cell population did not increase significantly in RA FLSs transfected with HAFML siRNA compared with the cells transfected with control siRNA (Fig. 3G, 3H), nor did we observe any effect of HAFML knockdown on caspase-3/7 activity (Fig. 3I). RA FLSs overexpressing HAFML did not exhibit reduced apoptosis and caspase-3/7 activity compared with those transfected with the control vector (Fig. 3J–L). Taken together, our data suggest that HAFML does not contribute to the proliferation and apoptosis of RA FLSs, and it eliminates the possibility that the effect of HAFML knockdown or overexpression on migration and invasion is associated with either proliferation or apoptosis.

lncRNAs usually function through their binding interactions with other intracellular factors, including proteins, RNAs, and DNAs. Thus, to investigate how HAFML modulates the migration and invasion of RA FLSs, we identified intracellular HAFML binding partners using an RNA pulldown assay. Full-length HAFML was transcribed with biotinylated nucleotides in vitro and then incubated with total cell lysates of RA FLSs. These mixtures were then pulled down with streptavidin. The associated proteins were resolved by SDS-PAGE and visualized by Coomassie Blue staining (Fig. 4A). Furthermore, the isolated pulldown product (Fig. 4B) was analyzed using mass spectrometry (MS) to identify proteins bound to HAFML. We also used three online bioinformatics analysis tools, catRAPID (http://www.tartaglialab.com), beRBP (http://bioinfo.vanderbilt.edu/beRBP/predict.html), and RBPmap (http://rbpmap.technion.ac.il), to predict the HAFML-binding proteins. (Fig. 4C shows a Venn diagram of the overlap of potential HAFML-binding proteins among HAFML MS, catRAPID, beRBP, and RBPmap. We observed that HuR (ELAVL1) was the unique protein in the overlap among the earlier four groups. To confirm that HuR binds specifically to HAFML, we repeated the pulldown assay with biotinylated HAFML and probed for HuR using Western blots. A similar result was obtained in which HuR bound specifically to HAFML (Fig. 4D). To confirm these findings, we used an anti-HuR Ab to immunoprecipitate endogenous HuR from total lysates of RA FLSs and then extracted and analyzed the HAFML bound to HuR. We found a >5-fold enrichment of HAFML in the anti-HuR immunoprecipitates compared with that measured in the IgG control (Fig. 4E). In addition, we determined that HuR was mainly located in the cytoplasm of RA FLSs (Supplemental Fig. 2A). These data confirm that HuR serves as a binding partner of HAFML.

Next, we investigated whether HuR helps regulate the migration and invasion of RA FLSs. siRNA was used to knock down HuR expression. We designed three different sequences of siRNA oligonucleotides for HuR. We found that transfection with all three siRNAs knocked down HuR mRNA levels, but the inhibitory effect of both siRNA-1 and siRNA-3 was stronger than that of siRNA-2 (Supplemental Fig. 2B, 2C). Thus, siRNA-1 and siRNA-3 were used for the subsequent experiments. We determined that HuR knockdown resulted in a significant reduction in the migration and invasion, but not proliferation, in RA FLSs (Fig. 4F–I). Moreover, we found that HuR overexpression increased the migration and invasion of RA FLSs compared with vector controls; however, HAFML knockdown reversed the HuR overexpression-induced increase in the migration and invasion of RA FLSs (Supplemental Fig. 2D, (Fig. 4J, 4K). These data suggest that HuR positively modulates the migration and invasion of RA FLSs. Our results reveal that HAFML is localized in the cytoplasm and indicate that HAFML might bind to cytoplasmic HuR in RA FLSs. Therefore, our findings suggest that lncRNA HAFML and HuR form a cytoplasmic complex to coregulate RA FLS functions.

To explore how the HAFML–HuR complex modulates RA FLS functions, we used RNA sequencing analysis to determine the transcriptome of HAFML siRNA- or HuR siRNA-treated RA FLSs compared with control siRNA. The data were analyzed for differential expression with a p value <0.05. Treatment with HAFML siRNA or HuR siRNA resulted in the downregulation of 79 or 32 genes compared with control siRNA treatment (Fig. 5A).

A Venn diagram of the overlap of the significantly downregulated genes between the HAFML siRNA- and HuR siRNA-treated groups is shown in (Fig. 5B. It was determined that in the overlap of both HAFML siRNA and HuR siRNA treatment, APPL2, WISP1, KRT19, KRTAP1-5, PCDHGA6, and KRTAP2-3 were significantly downregulated genes. We confirmed the significant changes in the APPL2 and WISP1 genes (Fig. 5C), but not in KRT19, KRTAP1-5, PCDHGA6, and KRTAP2-3 (Supplemental Fig. 2E), using RT-qPCR analysis. However, we observed that only APPL2 expression was upregulated in RA FLSs compared with HC FLSs (Fig. 5D). Moreover, we found that APPL2 expression had a positive correlation with HAFML levels in RA FLSs (Supplemental Fig. 2F). We also demonstrated the inhibitory effect of treatment with HAFML siRNA or HuR siRNA on APPL2 protein expression (Fig. 5E). These data suggest that APPL2 might be a downstream target for HAFML and HuR in RA FLSs.

Next, we determined the effect of APPL2 knockdown on the migration and invasion of RA FLSs. We constructed three different sequences of siRNA oligonucleotides for APPL2. As shown in Supplemental Fig. 2G and 2H, transfection with all three siRNA oligonucleotides reduced APPL2 levels, but the inhibitory effect of both siRNA-1 and siRNA-2 was stronger than that of siRNA-3. Accordingly, APPL2 siRNA-1 and siRNA-2 were used for further experiments. We found that transfection with APPL2 siRNAs reduced the migration and invasion of RA FLSs compared with control siRNA transfection (Fig. 5F, 5G). Collectively, our data indicate that APPL2 may mediate the actions of HAFML and HuR in regulating the migration and invasion of RA FLSs.

Because HuR is an RNA-binding protein with functions in mRNA stability, we next observed whether APPL2 and WISP1 could bind to HuR by using a RIP assay. We found a >4-fold enrichment of APPL2, but not WISP1, in the anti-HuR immunoprecipitates compared with that measured in the IgG control (Fig. 5H), suggesting that HuR can bind to APPL2 mRNA. We also observed that transfection with HAFML siRNA or HuR siRNA reduced APPL2 mRNA expression compared with that of the control siRNA in RA FLSs treated with the RNA transcription inhibitor Act D (Fig. 5I). These findings suggest that HAFML and HuR can modulate the mRNA stability of APPL2.

Finally, we searched for the potential mode of action of HAFML–HuR–APPL2. We first evaluated whether HAFML could regulate HuR expression. We found that HAFML knockdown did not influence the mRNA and protein expression of HuR (Fig. 5J, 5K). Because a previous study showed that HuR stabilizes lncRNA HGBC (27), we evaluated whether HuR could regulate lncRNA HAFML expression. However, RT-qPCR analysis showed that HuR knockdown did not affect HAFML expression (Fig. 5L), indicating that HuR does not regulate HAFML expression in RA FLSs. Moreover, the RIP assay showed that HuR could bind to APPL2 mRNA (Fig. 5H). We further demonstrated that HuR overexpression increased the mRNA expression of APPL2 compared with vector controls; however, HAFML knockdown reversed the HuR overexpression-induced increase of APPL2 mRNA expression (Fig. 5M). Intriguingly, we determined that HAFML knockdown reduced the binding of HuR and APPL2 mRNA (Fig. 5N); however, HAFML overexpression increased their binding (Fig. 5O). In addition, bioinformatics analysis predicted that HAFML could not bind directly to APPL2 mRNA. Thus, our data suggest that HAFML modulates RA FLS functions through its interaction with HuR, forming a cytoplasmic HAFML–HuR complex, which then anchors to APPL2 to control the stability of their target mRNAs (Fig. 6).

In this work, we identified lncRNA HAFML in FLSs and validated its increased expression level in RA FLSs. We demonstrated that HAFML functions as a positive modulator of the migration and invasion of RA FLSs. Mechanistically, we identified that HAFML binds to cytoplasmic HuR to form an RNA–protein complex that can anchor to the target mRNA of APPL2. This interaction increases the stability of APPL2 mRNA and accordingly upregulates its protein expression, thereby promoting the aggressive behavior of RA FLSs (Fig. 6).

Some lncRNAs have been shown to play important roles in rheumatoid pathogenesis, especially in regulating RA FLS functions (21, 22, 28); however, their contribution to modulating joint damage in RA remains largely unknown. In this study, we identified the lncRNA HAFML, which had increased expression in FLSs and synovial tissues from RA patients. The lncRNA was knocked down or overexpressed in RA FLSs to explore the contribution of HAFML to RA. We showed that HAFML knockdown reduced the migration and invasion of RA FLSs and HC FLSs; in contrast, HAFML overexpression increased the migration and invasion of RA FLSs and HC FLSs. Our data indicate that HAFML has a positive role in controlling the migration and invasion of FLSs, and that increased HAFML levels in RA FLSs might promote rheumatoid synovial aggression, thereby leading to joint destruction of RA. In fact, in line with our findings, recent studies have indicated that some other lncRNAs are involved in the regulation of the migration and invasion of RA FLSs; for instance, lncRNA NEAT1 promotes the migration and invasion of RA FLSs by regulating miR-410-3p–mediated YinYang 1 (29), LERFS lncRNA negatively modulates RA FLS aggression and proliferation (21), and lncRNA SNHG1 helps to sustain the migration and invasion of RA FLSs (30). In addition, we found that HAFML did not control the proliferation and apoptosis of RA FLSs, suggesting that HAFML is not associated with rheumatoid synovial hyperplasia.

We then addressed the underlying molecular mechanism(s) by which HAFML controls the aggression of RA FLSs. RT-qPCR of nuclear and cytoplasmic fractions and FISH assays revealed that HAFML is mainly located in the cytoplasm, indicating that it may function by interacting with other cytoplasmic proteins. As expected, we demonstrated that HAFML functions by forming a cytoplasmic RNA–protein complex with HuR, an RNA-binding protein that can regulate mRNA stability (31–33). Consistent with our findings, the lncRNA macrophage-associated atherosclerosis lncRNA sequence modulates apoptosis and atherosclerosis by interacting with HuR (34), while lncRNA OIP5-AS1 forms complexes with HuR to regulate proliferative phenotypes (35). The lncRNA HMS also physically interacts with HuR to stabilize HOXC10 mRNA (36). However, we could not completely eliminate the possibility that other unidentified intracellular factors within the complex might mediate the interaction of HAFML with HuR.

As a universal member of a family of RNA binding proteins, HuR can bind to a subset of mRNAs and thereby affect their stability and/or translation (33, 37). HuR plays important roles in modulating many physiological processes, including muscle differentiation, adipogenesis, and responses to stress and inflammatory stimuli (38–40). More importantly, HuR is involved in the pathogenesis of many disorders, such as cancer, cardiovascular diseases, and chronic inflammation (33, 41). Interestingly, increased expression of HuR has been shown in RA FLSs compared with FLSs from osteoarthritis (42). In our study, we found that HuR knockdown reduced the migration and invasion of RA FLSs; conversely, HuR overexpression resulted in increased migration and invasion. These data suggest that HuR is involved in the aggression of RA FLSs. Consistently, recent studies have also shown that HuR regulates the migration of other cell lines, such as vascular smooth muscle cells (43), T cells (44), and cancer cells (45, 46).

We further explored how the HAFML-HuR complex controls the migration and invasion of RA FLSs. RNA sequencing analysis was performed to identify related molecular targets. We demonstrated that knockdown of HAFML or HuR suppressed the expression of APPL2, an adaptor protein containing the pleckstrin homology domain, phosphotyrosine binding domain, and leucine zipper motif (APPL), which plays a role in controlling a series of biological processes (47, 48). Interestingly, a recent study showed the involvement of AAPL2 in regulating HGF-induced migration and invasion of mouse embryonic fibroblasts (49). In this study, we found that APPL2 knockdown reduced the migration and invasion of RA FLSs. These data suggest that APPL2 might mediate the role of HAFML and HuR in controlling RA FLS functions. It will be interesting to explore how APPL2 regulates RA FLS functions in the future.

Next, we investigated the underlying mechanisms by which HAFML and HuR modulate APPL2 expression. First, we determined that HAFML knockdown inhibited the mRNA and protein levels of APPL2. Intriguingly, HAFML knockdown did not influence the expression of HuR, and HuR inhibition also had no effect on HAFML expression. Second, RIP analysis revealed that HuR was bound to APPL2 mRNA, and that HAFML knockdown reduced the binding of HuR and APPL2 mRNA. Third, HAFML knockdown reversed the HuR overexpression-induced increase in APPL2 mRNA expression. Fourth, HAFML or HuR knockdown downregulated APPL2 mRNA expression compared with that of control siRNA in RA FLSs treated with the RNA transcription inhibitor Act D, indicating that HAFML and HuR can stabilize mRNA stability. Moreover, our bioinformatics analysis also revealed that HAFML could not bind directly to APPL2 mRNA (Supplemental Table II). Taken together, our findings indicate that a stable cytoplasmic HAFML–HuR complex is required for HuR binding to APPL2 mRNA. In RA FLSs, elevated HAFML levels result in increased formation of the HAFML–HuR complex, which promotes the binding of HuR with APPL2 mRNA and enhances the stability of its mRNA and subsequent protein synthesis. However, we did not eliminate the possibility that other unidentified factors mediate HAFML regulation of APPL2 expression.

In summary, we identified the cytoplasmic lncRNA HAFML in FLSs and described its role in controlling the migration and invasion of FLSs through interaction with HuR, thereby stabilizing APPL2 mRNA. Our findings suggest that increased HAFML expression in FLSs might contribute to rheumatoid synovial aggression and joint destruction. Furthermore, targeting HAFML in fibroblast subpopulations to inhibit cell migration might be a promising therapeutic approach for a series of disorders related with dysregulated fibroblast activation.

The authors thank Prof. Aishan He for providing parts of clinical samples.

The authors have no financial conflicts of interest.