circHIPK3 promotes pro-fibrotic epithelial reprogramming via the miR-29a-3p/SPARC axis with associated enhancement of TGF-β/SMAD signaling

Abstract

Background

Idiopathic pulmonary fibrosis (IPF) is fatal with limited treatments. circHIPK3 is upregulated in fibrotic lungs, but its role in lung epithelial cells and connection to TGF-β/SMAD signaling are unclear.

Methods

We used TGF-β1-stimulated lung epithelial cells (A549, BEAS-2B) and a bleomycin-induced mouse model. circHIPK3 was silenced with ASOs. Direct interactions were assessed using luciferase reporter and RNA pull-down assays, and functional intervention experiments were performed to investigate the proposed regulatory axis. TGF-β/SMAD activity was measured by CAGA reporter, and SPARC-TGFBR2 association by co-IP/colocalization.

Results

circHIPK3 was upregulated in TGF-β1-treated epithelial cells and fibrotic lungs. Silencing circHIPK3 reduced epithelial reprogramming, migration, and fibrotic markers. circHIPK3 directly bound miR-29a-3p, increasing SPARC. SPARC silencing phenocopied circHIPK3 knockdown or miR-29a-3p overexpression. SPARC overexpression enhanced TGF-β/SMAD signaling and colocalized with TGFBR2. In vivo, miR-29a-3p blockade exacerbated fibrosis, which was attenuated by SPARC knockdown, supporting a downstream role of SPARC in miR-29a-3p-mediated fibrotic responses.

Conclusion

We identify a circHIPK3/miR-29a-3p/SPARC axis driving pro-fibrotic epithelial reprogramming and TGF-β/SMAD enhancement in experimental pulmonary fibrosis, revealing a mechanism and potential therapeutic target.

Graphical abstract

1. Introduction

Idiopathic pulmonary fibrosis (IPF) is a progressive and fatal interstitial lung disease marked by fibroblast activation and excessive extracellular matrix (ECM) deposition, leading to irreversible loss of lung function [1,2]. Current drugs like pirfenidone and nintedanib only slow disease progression, highlighting the need for new therapeutic targets [3,4].

Transforming growth factor-β (TGF-β) is a central pro-fibrotic cytokine that drives fibroblast activation, ECM production, and a pro-fibrotic epithelial phenotype characterized by epithelial reprogramming and enhanced migration [5]. Beyond the core signaling pathways, such as TGF-β, the pathogenesis of IPF is also shaped by multi-layered post-transcriptional regulation, including miRNA-mediated repression, RNA-binding proteins, and non-coding RNAs. In recent years, non-coding RNAs, particularly circular RNAs (circRNAs), have emerged as critical regulators of gene expression in fibrosis [6]. CircRNAs are covalently closed RNAs with high stability. Depending on context, circRNAs can modulate gene expression by sequestering miRNAs, interacting with RNA-binding proteins, and regulating transcription/splicing, or translation [7]. This competitive endogenous RNA (ceRNA) mechanism is implicated in the fibrotic processes in multiple organs [8,9].

Among the circRNAs, circHIPK3 (hsa_circ_0000284 and mmu_circ_0001052), derived from exon 2 of the HIPK3 gene, is known to be dysregulated in cancers and cardiac fibrosis [10,11]. Recent studies have reported that circHIPK3 promotes pulmonary fibrosis in fibroblast-centered settings, functioning as a ceRNA for miR-338-3p (SOX4/COL1A1) or miR-30a-3p (FOXK2) [12,13]. Moreover, two studies in cardiac models have demonstrated that circHIPK3 directly binds to miR-29a and regulates downstream target IGF-1 [14,15]. Our selection of circHIPK3 was further guided by preliminary screening in our laboratory, which identified several TGF-β1-responsive circRNAs, among which circHIPK3 consistently showed upregulation. However, whether circHIPK3 exerts an epithelial cell-context-specific pro-fibrotic program, and how it connects ceRNA regulation to TGF-β receptor/SMAD pathway output, remains unclear.

One of the most significant miRNA targets implicated in fibrosis is the miR-29 family, particularly miR-29a-3p. It is a well-established anti-fibrotic miRNA that represses the expression of multiple ECM components, and its expression is consistently downregulated in fibrotic tissues [16,17]. While miR-29 family downregulation is a consistent feature of fibrotic lungs [16], and TGF-β/SMAD3 signaling can suppress miR-29 expression [18], the cell-type-specific upstream controls and potential ceRNA-mediated buffering of miR-29a-3p activity in IPF remain incompletely understood.

Secreted Protein Acidic and Rich in Cysteine (SPARC) is a matricellular protein enriched in fibrotic lung microenvironments and has been implicated in ECM deposition and aberrant wound repair in IPF [19,20]. Notably, SPARC has been linked to enhanced TGF-β/SMAD pathway activity in multiple systems [20], providing a plausible route by which a ceRNA module could couple post-transcriptional regulation to canonical pro-fibrotic signaling output. However, whether SPARC is regulated by a circRNA/miRNA axis in IPF, particularly in the epithelial context, remains insufficiently explored.

Given the emerging fibroblast-centered roles of circHIPK3 in pulmonary fibrosis, we asked whether circHIPK3 also contributes to an epithelial pro-fibrotic program. Because miR-29a-3p is a key anti-fibrotic miRNA consistently suppressed in fibrotic lungs and SPARC represents a plausible matricellular amplifier of pro-fibrotic signaling, we investigated whether circHIPK3 modulates miR-29a-3p availability to derepress SPARC and thereby enhance TGF-β/SMAD pathway output. Using complementary in vitro and in vivo models, we identify a circHIPK3/miR-29a-3p/SPARC regulatory circuit in lung epithelial cells and explore its functional relevance in pulmonary fibrosis.

2. Materials and methods

2.1. Experimental Animals

Male C57BL/6 mice (6-8 weeks old) were maintained under SPF conditions (12-h light/dark cycle) with ad libitum access to food and water. After 1 week of acclimation, pulmonary fibrosis was induced by a single intratracheal instillation of bleomycin (3 mg/kg in 50 μL) [21,22]. Mice were randomly assigned to six groups (n = 5 per group; total 30 mice): Normal control (Normal), Pulmonary Fibrosis model (PF model), PF + antagomiR-29a, PF + ASO-mmu_circHIPK3, PF + antagomiR-29a + si-SPARC, and PF + ASO-circHIPK3 + si-SPARC. AntagomiR-29a and ASO-mmu_circHIPK3 (targeting the junction of mmu_circ_0001052) were administered intratracheally on days 3, 7, 10, and 14. si-SPARC was formulated with in vivo-jetPEI® (Polyplus) at an N/P ratio of 6 and delivered intratracheally at matching time points. Oligonucleotides were administered at 5 nmol/mouse in 50 μL per dose. Control groups received vehicle instillation at matching time points. On day 21, mice were euthanized by gradual CO₂ asphyxiation followed by cervical dislocation. All procedures were approved by the Animal Ethics Committee of Ningbo University (Approval No. 11373) and conducted in accordance with ARRIVE guidelines.

The sequences of the key oligonucleotides used in this study are as follows: ASO targeting circHIPK3 junction site: 5′-TGA GGC CAT ACC TGT AGT AG-3′; ASO -NC: 5′-GAC AGA GTC AAC GAC CTA GA-3′; si-SPARC-1: 5′-TGA AGA AGA TCC ATG AGA ATG AGA A-3′; si-SPARC-2: 5′-ACC TTC GAC TCT TCC TGC CAC TTC T-3′; si-SPARC-3: 5′-TGC GGG TGA AGA AGA TCC ATG AGA A-3′; si-IGF1-1: 5′-AGC AGA TAG AGC CTG CGC AAT GGA A-3′; si-IGF1-2: 5′-ACA TTG CTC TAA CAT CTC CCA TCT-3′. All oligonucleotides were synthesized by RiboBio (Guangzhou, China).

2.2. RT-qPCR

Total RNA was extracted from tissues or cells using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. For circRNA analysis, 3 μg of total RNA was treated with 3 U/μg RNase R (Epicentre) at 37 °C for 15 min, followed by heat inactivation at 70 °C for 10 min. cDNA was synthesized using the PrimeScript™ RT Kit (Takara). RT-qPCR was performed on an ABI 7500 system (Applied Biosystems) with BeyoFast™ SYBR Green qPCR Mix (Beyotime). For miRNA and mRNA detection, no RNase R treatment was applied. miRNA qPCR reactions were performed using the miRCURY LNA miRNA PCR Assays (Qiagen, #339306), and circHIPK3/mRNA qPCR reactions were performed using HiScript III RT SuperMix for qPCR (Vazyme, #R323). The presence of circHIPK3 was confirmed by gel electrophoresis. The relative expression levels of circHIPK3, miRNAs, and mRNAs were calculated using the 2^−ΔΔCt method, with GAPDH as the internal control for circRNA/mRNA and U6 snRNA for miRNA. The relevant primers are shown in Supplementary Table 1.

2.3. Cell culture

The human lung epithelial cell lines A549 and BEAS-2B were obtained from Zhejiang Ruyao Biotechnology Co., Ltd. Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM; Thermo Fisher Scientific, #11965092) supplemented with 10% fetal bovine serum (FBS; Thermo Fisher Scientific, #10270106) and 1% penicillin/streptomycin (Thermo Fisher Scientific, #15140122). All cell lines were regularly tested and confirmed to be free of mycoplasma contamination. All cell lines were maintained at 37 °C in a humidified incubator with 5% CO₂. All in vitro experiments were performed with a minimum of three independent biological replicates (n = 3) to ensure robustness and statistical power.

2.4. ASO-mediated knockdown of circHIPK3

CircHIPK3 is generated through backsplicing of exon 2 from the HIPK3 transcript (ENST00000303296.4), with a junction sequence of 5′-ACTACAG-GTATGGCCT-3′. An antisense oligonucleotide (ASO) targeting this junction site was designed, along with a non-targeting control ASO (ASO-NC). These junction-specific ASOs were synthesized by RiboBio. A549 and BEAS-2B cells were transfected with the respective ASOs using Lipofectamine 3000 (Thermo Fisher Scientific, #L3000015) according to the manufacturer's instructions. To evaluate the knockdown efficiency, total RNA was extracted from transfected cells, and the expression level of hsa_circHIPK3 was quantified using the previously described RT-qPCR method with junction-specific divergent primers. GAPDH was used as an internal control. In parallel, to confirm that the knockdown was specific to the circular transcript and did not affect the linear HIPK3 mRNA, the expression of linear HIPK3 was measured by RT-qPCR using convergent primers.

2.5. Immunoblotting

Cells (5 × 10⁶) or tissues (100 mg) were collected, and total protein was extracted using RIPA lysis buffer supplemented with a phosphatase inhibitor cocktail (20 μL/mL lysis buffer). The following primary antibodies were used: From Abcam (USA) (dilution: 1:1000–1:2000, unless otherwise specified): Vimentin (ab92547), N-cadherin (ab76011), E-cadherin (ab231303), IGF1 (ab223567), SPARC (ab203284), SNAI2 (ab27568), SMAD2 (ab33875), Phosphorylated SMAD2 (phospho S467) (ab280888), SMAD3 (ab208182), TGFBR1 (ab235578); From ABclonal (dilution: 1:1000): Phosphorylated SMAD3 (phospho Ser423/425) (AP0727), TGFBR2 (A22471), SNAI1 (A24806). Protein expression levels were quantified using ImageJ software, and GAPDH (ab8245) was used as the internal control for normalization. All quantitative analyses of Western blots were performed blindly by researchers unaware of group assignments.

2.6. Cell scratch assay

Cells were seeded in 6-well plates at a density of 5 × 10⁵ cells per well and cultured to full confluence. A 20-μL pipette tip was used to make a scratch in the monolayer. The cells were then cultured in a serum-free medium to minimize cell proliferation effects. Migration was monitored under a phase-contrast microscope at 0 and 48 h, and the migration distance was quantified using Image-Pro Plus 6.0 software. For this and all other microscopy-based assays, images were captured from randomly selected fields across the entire sample to avoid selection bias.

2.7. Bioinformatic analysis

Potential miRNA targets of circHIPK3 were predicted using the circBank database. Experimentally validated and predicted mRNA targets of miR-29a-3p were retrieved from the miRDB and TargetScan databases to increase the reliability of predictions through cross-referencing. To identify genes highly associated with IPF and epithelial-mesenchymal transition (EMT), the GeneCards database was queried using the keywords “IPF” and “EMT”. A list of miRNAs associated with PF was obtained from the HMDD v3.2 database. miRNA expression profiling data from IPF patient samples (GEO accession number: GSE32538) were downloaded and analyzed to identify differentially expressed miRNAs. Venn diagram analysis was performed using an online tool to identify the intersecting miRNAs from the circHIPK3 target set, the IPF-associated miRNA set (from HMDD), and the differentially expressed miRNA set from the GEO dataset. For downstream gene analysis, a similar Venn diagram was constructed to find the intersection between predicted miR-29a-3p targets, the IPF-associated gene set, and the EMT-associated gene set from GeneCards. The conservation of the miR-29a-3p sequence and its binding sites on circHIPK3 across species was analyzed using the UCSC Genome Browser and the NCBI blast website.

2.8. Dual-luciferase reporter assay

Dual-luciferase reporter assays were performed to functionally validate the direct interactions between circHIPK3 and miR-29a-3p, miR-29a-3p and IGF1, and miR-29a-3p and SPARC. The wild-type sequences of circHIPK3 (containing the predicted miR-29a-3p binding sites and the back-splice junction) or the 3′ UTRs of IGF1 and SPARC (containing the predicted miR-29a-3p binding sites) were synthesized and cloned into the pISO vector downstream of the firefly luciferase gene. All constructed plasmids were validated by Sanger sequencing and restriction enzyme digestion to confirm the correct insert size, sequence, and junction location. Site-directed mutagenesis of the seed complementary sites generated mutant (Mut) reporter constructs. For circHIPK3, a mutation was introduced in the predicted 7mer binding site for hsa-miR-29a-3p (as depicted in Fig. 2E). For the 3’ UTRs of IGF1 and SPARC, Mut versions were generated based on predicted miR-29a-3p binding sites identified using miRDB and TargetScan. The pRL-TK vector (Promega) was used as an internal control for normalization. HEK293T cells were seeded in 24-well plates and co-transfected with the constructed reporter plasmids, along with either miR-29a-3p mimic (50 nM) or negative control mimic (NC mimic, 50 nM), using Lipofectamine 3000 (Thermo Fisher Scientific, #L3000015). Each transfection condition was performed in technical triplicate and independently repeated in three biological replicates (n = 3). After 48 h, luciferase activity was measured using the Dual-Luciferase® Reporter Assay System (Promega, #E1910) on a GloMax® Navigator Microplate Luminometer (Promega). Firefly luciferase activity was normalized to Renilla luciferase activity for each sample to control for transfection efficiency, and the relative Firefly/Renilla ratio was calculated and presented. Downstream validation of gene expression changes was performed using RT-qPCR and Western blot analysis.

Fig. 2.

CircHIPK3 directly associates with and negatively regulates miR-29a-3p.

(A) A Venn diagram identifying three candidate miRNAs (miR-29a-3p, miR-34c-3p, miR-146b-3p) by intersecting predicted circHIPK3 targets, IPF-associated miRNAs (from HMDD), and differentially expressed miRNAs in an IPF GEO dataset (GSE32538). (B–D) RT-qPCR analysis of miR-29a-3p (B), miR-34c-3p (C), and miR-146b-3p (D) expression in TGF-β1-induced BEAS-2B and A549 cells transfected with ASO-NC or ASO-circ. (E) A schematic of the predicted canonical 7mer-A1 binding site for miR-29a-3p on the circHIPK3 sequence, as identified by integrated bioinformatic analysis. (F) Dual-luciferase reporter assay in HEK293T cells co-transfected with wild-type (WT) or mutant (Mut) circHIPK3 reporter vectors and miR-NC or miR-29a-3p mimic. (G) RNA pull-down assay followed by RT-qPCR analysis of circHIPK3 enrichment using a biotin-labeled miR-29a-3p probe (Bio-miR-29a-mimic) or NC mimic. (H) RT-qPCR of linear HIPK3 in pull-down material. Data are mean ± SD (n = 3). *P < 0.05, **P < 0.01 vs. Control; #P < 0.05, ##P < 0.01 vs. TGF-β1 or mir-29a probe.

2.9. RNA pull-down assay

To biochemically validate the physical interaction between circHIPK3 and miR-29a-3p, an RNA pull-down assay was performed. A biotin-labeled miR-29a-3p mimic and a negative control (NC) mimic (Sangon Biotech) were transfected into BEAS-2B or A549 cells at a final concentration of 50 nM using Lipofectamine 3000 (Thermo Fisher Scientific, #L3000015) when cells reached 50% confluence. After 24 h, cells were harvested and lysed using RNA lysis buffer (Thermo Fisher Scientific, #AM1931). Cell lysates were incubated with streptavidin-coated magnetic beads (Thermo Fisher Scientific, #11205D) for 3 h at 4 °C with gentle rotation. The beads were then washed three times with wash buffer (20 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 0.5% NP-40, pH 7.5). RNA bound to the biotinylated miRNA was extracted using TRIzol reagent (Thermo Fisher Scientific, #15596026). The enrichment of circHIPK3 in the pull-down material was analyzed by RT-qPCR with specific divergent primers. GAPDH served as an internal control.

2.10. Immunofluorescence

Cell immunofluorescence staining was performed following the protocol described by Zhao et al. [23]. The primary antibodies used included E-cadherin (ab231303, Abcam) and N-cadherin (ab76011, Abcam). 4′,6-Diamidino-2-phenylindole(DAPI, D1306, Thermo Fisher Scientific, USA) was used to stain the nuclei, while phalloidin (CA1610, Solarbio, China) was used to label F-actin and visualize the cytoskeleton. For immunofluorescence colocalization, rabbit anti-human SPARC (ab203284, Abcam, UK) and mouse anti-human TGFBR2 (66636-1-Ig, Proteintech, China) antibodies were used. The corresponding fluorescent secondary antibodies were goat anti-rabbit IgG H&L conjugated with FITC (ab6717, Abcam) and goat anti-mouse IgG H&L conjugated with Alexa Fluor® 594 (ab150116, Abcam). The results were captured and analyzed using a Leica DM500 fluorescence microscope (Leica, Germany). Image acquisition was performed on randomly selected fields, and fluorescence intensity was quantified blindly using ImageJ software.

2.11. Cell transfection

ASO targeting circHIPK3 (ASO-circHIPK3) was synthesized by RiboBio. siRNA sequences targeting IGF1 (si-IGF1 #siG000003479A-1-5, #siG000003479B-1-5, #siG000003479C-1-5), siRNA targeting SPARC (si-SPARC #siG000006678A-1-5, #siG000006678B-1-5, #siG000006678C-1-5), and corresponding negative control siRNA (si-NC) were obtained from RiboBio. miR-29a-3p inhibitor (#miR20000086-1-5) was also obtained from RiboBio. The miR-29a-3p mimic and negative control mimic (miR-NC) were synthesized by Sangon Biotech. Cells (BEAS-2B or A549) were seeded in 24-well plates at a density of 1.5–2 × 10⁵ cells per well and cultured until 50–60% confluent. Transfection was performed using Lipofectamine™ 3000 (Thermo Fisher Scientific, #L3000015) with each nucleic acid construct at a final concentration of 50 nM. After 48 h, transfection efficiency was evaluated by measuring the expression levels of target genes via RT-qPCR or Western blot. Cell viability was assessed using the MTT assay kit (Beyotime, #C0009S) according to the manufacturer's instructions.

2.12. TGF-β/SMAD luciferase reporter assay

To assess canonical TGF-β/SMAD transcriptional activity, a CAGA-box luciferase reporter plasmid (pGL3-4×CAGA-luc) was used. This reporter contains four tandem copies of the SMAD3/SMAD4-responsive CAGA element upstream of firefly luciferase. Cells (BEAS-2B or A549) were seeded in 24-well plates at a density of 1 × 10⁵ cells per well and cultured overnight. Transient transfections were performed using Lipofectamine 3000 (Thermo Fisher Scientific, #L3000015) with 100 ng of CAGA-luc reporter plasmid, 10 ng of pRL-TK Renilla luciferase control plasmid (Promega, #E2241), and, where indicated, 50 ng of dominant-negative SMAD3 expression plasmid (dn-SMAD3, pcDNA3-Flag-SMAD3(3SA)). Total DNA amount was kept constant by adding empty vector.

After 24 h of transfection, cells were serum-starved for 6 h and then treated with or without TGF-β1 (20 ng/mL, R&D Systems, #240-B) for an additional 24 h. For experiments involving the TGF-β receptor kinase inhibitor, cells were pre-treated with 10 μM SB431542 (Selleck Chemicals, #S1067) for 1 h prior to TGF-β1 stimulation. For SPARC overexpression studies, cells were co-transfected with pcDNA3.1-SPARC expression plasmid (Sangon Biotech, custom synthesis) or empty vector. Luciferase activities were measured using the Dual-Luciferase® Reporter Assay System (Promega, #E1910) on a GloMax® Navigator Microplate Luminometer (Promega). Firefly luciferase activity was normalized to Renilla luciferase activity and expressed as fold change relative to the untreated control group. Each experiment was performed in triplicate and repeated independently at least three times.

2.13. Co-immunoprecipitation (Co-IP)

To detect the physical interaction between SPARC and TGFBR2, co-immunoprecipitation was performed using whole-cell lysates from BEAS-2B and A549 cells. Cells were cultured in 10 cm dishes to 80–90% confluence, washed twice with ice-cold PBS, and lysed in IP lysis buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% NP-40, 10% glycerol) supplemented with protease and phosphatase inhibitor cocktails (Thermo Fisher Scientific, #78442). Lysates were incubated on ice for 30 min, centrifuged at 14,000× g for 15 min at 4 °C, and supernatants were collected. Protein concentration was determined using the BCA Protein Assay Kit (Thermo Fisher Scientific, #23225).

For each immunoprecipitation, 500 μg of total protein was pre-cleared with 20 μL of Protein A/G Magnetic Beads (Thermo Fisher Scientific, #88802) for 1 h at 4 °C. Pre-cleared lysates were incubated with 2 μg of anti-SPARC antibody (Abcam, #ab203284), 2 μg of anti-TGFBR2 antibody (Proteintech, #66636-1-Ig), or normal rabbit IgG (Cell Signaling Technology, #2729) as negative control, overnight at 4 °C with gentle rotation. The next day, 30 μL of Protein A/G Magnetic Beads were added and incubated for an additional 3 h at 4 °C. Beads were washed five times with IP wash buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% NP-40). Bound proteins were eluted by boiling in 2× SDS-PAGE sample buffer at 95 °C for 10 min, separated by SDS-PAGE, and analyzed by Western blot using anti-SPARC and anti-TGFBR2 antibodies. Input controls (5% of total lysate) were loaded alongside IP samples. Quantitative analysis of co-immunoprecipitated proteins was performed using ImageJ and expressed as percentage of input.

2.14. In vivo functional validation in animal studies

Pulmonary fibrosis was induced by a single intratracheal instillation of bleomycin (3 mg/kg) as previously described [24]. Treatments were administered as described in the previous “Experimental Animals” section. On day 21, all mice were euthanized humanely. Lung tissues were collected and fixed, followed by paraffin embedding and sectioning. Histopathological evaluation of PF was performed using hematoxylin and eosin (HE) staining and Masson's trichrome staining. To minimize bias, histopathological scoring and subsequent quantitative analyses were performed by two independent investigators who were blinded to the group assignments. Fibrotic changes were quantitatively assessed using the Ashcroft scoring system. Immunohistochemistry (IHC) was carried out following the protocol described by Zhao et al. [23], using the following primary antibodies: collagen I (Abcam, #ab270993), Fibronectin (Abcam, #ab2413), and α-SMA (Abcam, #ab7817). H-score analysis was performed based on IHC staining intensity and distribution using ImageJ software by researchers blinded to group assignments. Total RNA and protein were extracted from lung tissues for further molecular analysis. The expression levels of CircHIPK3, miR-29a-3p, IGF1, and SPARC mRNA were detected by RT-qPCR. Western blot was performed to evaluate the protein expression of vimentin, N-cadherin, E-cadherin, SNAI1, SNAI2, SMAD3, p-SMAD3, SMAD2, p-SMAD2, TGFBR2, SPARC, and GAPDH. Hydroxyproline content in lung tissues was measured using a commercial assay kit (Sigma-Aldrich, #MAK008) according to the manufacturer's instructions. Results were expressed as μg per mg of wet lung tissue.

Sample size (n = 5 per group) was determined based on common practices in the bleomycin-induced pulmonary fibrosis literature and our preliminary experiments [25,26]. No formal a priori power calculation was performed. Post-hoc power analyses for the key endpoints (hydroxyproline content, Ashcroft score, and Western blot data) were conducted to aid interpretation of the observed effect sizes; these analyses indicated that the detected differences were associated with observed statistical power >0.8 for the main comparisons. However, post-hoc power analysis does not confirm the adequacy of the sample size a priori and should be interpreted with caution.

2.15. Statistical analysis

Data are presented as mean ± standard deviation (SD) from at least three independent experiments. The normality of data distribution was assessed using the Shapiro-Wilk test. Comparisons between two groups were analyzed using a two-tailed Student's t-test. Comparisons among multiple groups were analyzed using one-way analysis of variance (ANOVA) or two-way ANOVA, as appropriate for the experimental design. Specifically, two-way ANOVA was applied to datasets involving two independent factors (e.g., the combination of genetic and pharmacological interventions) to test for main effects and interaction effects. For one-way ANOVA, Tukey's post-hoc test was used for multiple comparisons. For two-way ANOVA, Tukey's post-hoc test was used following a significant interaction or main effect. These post-hoc tests control the family-wise error rate. A p-value of less than 0.05 was considered statistically significant. All statistical analyses were performed using GraphPad Prism software (version 8.0, GraphPad Software, USA).

3. Results

3.1. Silencing of circHIPK3 attenuates TGF-β1-induced epithelial reprogramming and cell migration in lung epithelial cells

A schematic overview of the proposed pro-fibrotic mechanism of circHIPK3 is presented in Fig. 1A. To investigate the role of circHIPK3 in PF, we first established an in vitro model by treating human bronchial epithelial cells (BEAS-2B) and alveolar epithelial-like adenocarcinoma (A549) with TGF-β1 (20 ng/mL, 48 h). circHIPK3 (hsa_circ_0000284) is derived from exon 2 of the HIPK3 gene and features a specific back-splice junction (sequence: ACTACAG-GTATGGCCT). Using junction-spanning divergent primers and RNase R treatment to confirm its circular nature (Supplementary Fig. S1), we found that TGF-β1 stimulation significantly upregulated circHIPK3 expression in both cell lines (P < 0.01, Fig. 1B). In contrast, linear HIPK3 mRNA levels were not significantly altered by TGF-β1 treatment in either cell line (P > 0.05, Fig. 1C), indicating that the increase in circHIPK3 occurs independently of transcriptional changes in its host gene. To determine the functional impact of circHIPK3, we employed junction-specific ASOs to achieve efficient and selective knockdown. Transfection with ASO-circHIPK3 significantly reduced circHIPK3 levels to below 50% of the control (P < 0.01, Fig. 1D), without affecting linear HIPK3 mRNA (Fig. 1E), confirming the specificity of our intervention.

Fig. 1.

Silencing circHIPK3 attenuates TGF-β1-induced epithelial reprogramming and migration in lung epithelial cells.

(A) Proposed mechanism of circHIPK3 in pulmonary fibrosis. (B) RT-qPCR analysis of circHIPK3 expression in BEAS-2B and A549 cells after TGF-β1 treatment (20 ng/mL, 48 h). (C) Linear HIPK3 expression under the same conditions. (D) RT-qPCR analysis of hsa_circ_HIPK3 expression in cells transfected with ASO-negative control (ASO-NC) or ASO targeting circHIPK3 (ASO-circHIPK3). (E) RT-qPCR analysis of linear HIPK3 mRNA expression in cells from the same transfection as in (D). (F) Representative Western blot images of E-cadherin, N-cadherin, and Vimentin protein levels in BEAS-2B and A549 cells under different treatments: Control, TGF-β1, TGF-β1 + ASO-NC, TGF-β1 + ASO-circHIPK3. (G-I) Quantitative analysis of the Western blot results for (G) Vimentin, (H) N-cadherin, and (I) E-cadherin, corresponding to the blots shown in (F). (J) Representative images of wound healing assays in BEAS-2B and A549 cells under the indicated treatments. (K) Quantitative analysis of cell migration distance from the wound healing assays shown in (J). Data are mean ± SD (n = 3). *P < 0.05, **P < 0.01 vs. Control; #P < 0.05, ##P < 0.01 vs. TGF-β1.

We next examined the effect of circHIPK3 silencing on the acquisition of epithelial reprogramming, a process characterized by loss of epithelial markers, gain of mesenchymal markers, and enhanced migratory capacity, which is considered a pivotal event in fibrosis progression. Western blot analysis (Fig. 1F–I) revealed that TGF-β1 induced a typical epithelial reprogramming-associated marker pattern, markedly upregulating the mesenchymal proteins Vimentin and N-cadherin, while downregulating the epithelial marker E-cadherin. circHIPK3 knockdown significantly reversed these TGF-β1-induced changes (all P < 0.01). Both cell lines exhibited a marked fibrotic response upon TGF-β1 stimulation. Furthermore, a wound healing assay demonstrated that TGF-β1 markedly enhanced cell migration (P < 0.01, Fig. 1J and K). Silencing circHIPK3 attenuated this pro-migratory effect (P < 0.01). Together, these data indicate that circHIPK3 expression is upregulated by TGF-β1, and its knockdown reduces TGF-β1-induced epithelial reprogramming and cell migration in lung epithelial cells.

3.2. CircHIPK3 directly binds to and regulates miR-29a-3p

To elucidate the mechanism by which circHIPK3 promotes fibrosis, we sought to identify miRNAs that are potentially regulated by circHIPK3. By integrating bioinformatic predictions from the circBank database with miRNA expression profiles from IPF patients (GEO: GSE32538) and a curated set of fibrosis-associated miRNAs (HMDD v3.2), we identified three candidate miRNAs: miR-29a-3p, miR-34c-3p, and miR-146b-3p, which were common to all three datasets (Fig. 2A). We then experimentally assessed the regulation of these candidates in our cellular model. Among them, only miR-29a-3p exhibited consistent and significant regulation by both TGF-β1 and circHIPK3 (Fig. 2B–D). TGF-β1 stimulation markedly suppressed miR-29a-3p expression in both cell lines (P < 0.01). Crucially, this suppression was significantly reversed by circHIPK3 knockdown (P < 0.01). miR-29a-3p is known for its anti-fibrotic functions, and its expression was significantly regulated by both TGF-β1 and circHIPK3. To determine if this regulatory relationship stems from a direct interaction, we employed multiple bioinformatic algorithms (TargetScan, miRDB). The analysis identified a canonical 7mer-A1 binding site for miR-29a-3p within the circHIPK3 sequence (Fig. 2E). The predicted miR-29a-3p binding site on circHIPK3 was highly conserved between humans and mice (Supplementary Fig. S1C and D), supporting the translational relevance of our subsequent in vivo investigations. A dual-luciferase reporter assay confirmed this direct binding: co-transfection with a miR-29a-3p mimic significantly reduced the luciferase activity of the wild-type circHIPK3 reporter (P < 0.01), an effect that was completely abolished upon mutation of this seed complementary site (Fig. 2F). To provide independent biochemical validation of the direct interaction, we performed an RNA pull-down assay using a biotin-labeled miR-29a-3p probe. RT-qPCR analysis revealed a significant enrichment of circHIPK3 in the samples pulled down by the wild-type miR-29a-3p probe compared to a nonspecific control probe (P < 0.01, Fig. 2G). Critically, under the same conditions, the linear HIPK3 mRNA was not enriched by the miR-29a-3p probe (Fig. 2H), confirming the specificity of this interaction for the circular transcript. Together, these results show that circHIPK3 interacts with miR-29a-3p and negatively regulates its availability.

3.3. The circHIPK3/miR-29a-3p axis regulates TGF-β1-induced epithelial reprogramming and cell migration

To determine if miR-29a-3p mediates the pro-fibrotic functions of circHIPK3, we conducted combined-intervention experiments in TGF-β1-treated cells. As previously observed, TGF-β1 enhanced cell migration (P < 0.01, Fig. 3A and B). This pro-migratory effect was significantly attenuated by individual transfection with either ASO-circHIPK3 or a miR-29a-3p mimic (P < 0.01). The combination of ASO-circHIPK3 and miR-29a-3p mimic produced a stronger inhibitory effect on migration than either agent alone (P < 0.01).

Fig. 3.

Combined circHIPK3 silencing and miR-29a-3p overexpression attenuate TGF-β1-induced epithelial reprogramming and migration.

(A) Representative images of wound healing assays in BEAS-2B and A549 cells under the indicated treatments: Control, TGF-β1, TGF-β1+ASO-circ, TGF-β1+miR-29a mimic, TGF-β1+ASO-circ + miR-29a mimic (co-treatment). (B) Quantitative analysis of cell migration distance from the wound healing assays shown in (A). (C) Immunofluorescence of E-cadherin, N-cadherin, and α-SMA (with F-actin) in BEAS-2B and A549 cells. Nuclei were stained with DAPI (blue). (D–F) Quantitative analysis of fluorescence intensity for (D) E-cadherin, (E) N-cadherin, and (F) α-SMA, corresponding to the IF images shown in (C). Data are mean ± SD (n = 3). *P < 0.05, **P < 0.01 vs. Control; #P < 0.05, ##P < 0.01 vs. TGF-β1; $P < 0.05, $$P < 0.01 vs. co-treatment.

We further evaluated the impact on epithelial reprogramming via immunofluorescence staining for the epithelial marker E-cadherin and the mesenchymal markers N-cadherin and α-SMA (Fig. 3C–F). Quantitative analysis confirmed that TGF-β1 induced a pronounced shift toward a mesenchymal-like phenotype, upregulating N-cadherin and α-SMA while downregulating E-cadherin (all P < 0.01). Both ASO-circHIPK3 and the miR-29a-3p mimic individually significantly reversed these changes (P < 0.05). The greatest reversal was observed with the combination treatment. For nearly all markers across both cell lines, the combination therapy was statistically superior to individual treatments (P < 0.01).

These combined-intervention experiments showed that knockdown of circHIPK3 or overexpression of miR-29a-3p each attenuated TGF-β1-induced cell migration and epithelial reprogramming, and the combination of both produced a greater effect.

3.4. Identification of SPARC as a direct downstream target of the circHIPK3/miR-29a-3p axis, with IGF1 as a secondary candidate

To systematically identify the downstream effector genes of the circHIPK3/miR-29a-3p axis, we integrated miRNA target prediction with disease association data. Potential targets of miR-29a-3p from miRDB and TargetScan were intersected with IPF- and epithelial reprogramming-associated genes from GeneCards, yielding five candidate genes: IGF1, COL4A1, COL1A1, SPARC, and PTEN (Fig. 4A). PTEN, a well-known tumor suppressor and anti-fibrotic gene [27], was excluded from further functional validation. We prioritized SPARC and IGF1 for this study to maintain a focused investigation on candidates with established or strongly putative pro-fibrotic functions in ECM remodeling and growth factor signaling. We first validated the expression of these candidates in a bleomycin-induced mouse model of pulmonary fibrosis. RT-qPCR analysis of lung tissues confirmed significant upregulation of all four genes (Fig. 4B). For further screening in vitro, we treated BEAS-2B and A549 cells with TGF-β1 and various interventions. While TGF-β1 upregulated all four genes, only SPARC expression was consistently and significantly suppressed by both ASO-circHIPK3 and miR-29a-3p mimic in both cell lines (P < 0.01, Fig. 4C and D). The combination treatment yielded an even more potent suppressive effect on SPARC (P < 0.01). In contrast, the other genes showed inconsistent responses. IGF1 emerged as a secondary candidate: its expression was significantly downregulated by the miR-29a-3p mimic, and the combination treatment also produced an inhibitory effect (P < 0.01). Based on these findings, SPARC and IGF1 were selected for validation as direct targets. SPARC exhibited the most consistent regulation across all experimental conditions, positioning it as the primary candidate for further functional characterization. IGF1, while also a direct target, was pursued as a secondary comparator given its weaker and less consistent response to circHIPK3 manipulation. Bioinformatics analysis predicted strong binding sites for miR-29a-3p in the 3′UTRs of both genes (Fig. 4E). Dual-luciferase reporter assays confirmed direct binding, as the miR-29a-3p mimic significantly reduced the luciferase activity of the wild-type 3′UTR reporters for both SPARC and IGF1 (P < 0.01), an effect abolished by site-directed mutagenesis (Fig. 4F and G). Furthermore, transfection with the miR-29a-3p mimic significantly reduced the mRNA levels of SPARC and IGF1 (P < 0.01, Fig. 4H and I). Western blot analysis confirmed these results at the protein level (Fig. 4J–L). Thus, both SPARC and IGF1 are direct targets of miR-29a-3p. Among them, SPARC showed more consistent changes in response to circHIPK3 and miR-29a-3p modulation.

Fig. 4.

Identification of SPARC and IGF1 as direct functional targets of miR-29a-3p.

(A) A Venn diagram identifying candidate target genes (IGF1, COL4A1, COL1A1, SPARC, PTEN) by intersecting predicted miR-29a-3p targets, IPF-associated genes, and EMT-associated genes from GeneCards. (B) RT-qPCR analysis of candidate gene mRNA expression in lung tissues from saline-treated (Normal) and bleomycin-induced (PF Model) mice. (C, D) RT-qPCR analysis of candidate gene expression in TGF-β1-induced BEAS-2B (C) and A549 (D) cells transfected with ASO-circHIPK3 or miR-29a-3p mimic. (E) A schematic of the predicted binding sites for miR-29a-3p on the 3′UTRs of human SPARC and IGF1. (F, G) Dual-luciferase reporter assay in HEK293T cells co-transfected with wild-type (WT) or mutant (Mut) 3′UTR reporter vectors of (F) IGF1 or (G) SPARC, along with miR-NC or miR-29a-3p mimic. (H, I) RT-qPCR analysis of (H) IGF1 and (I) SPARC mRNA expression in BEAS-2B and A549 cells transfected with miR-NC or miR-29a-3p mimic. (J) Representative Western blot images of IGF1 and SPARC protein levels in cells from the same treatment as in (H, I). (K-L) Quantification analysis of SPARC (K) and IGF1 (L) protein levels normalized to loading control. Data are mean ± SD (n = 3). *P < 0.05, **P < 0.01 vs. Control; #P < 0.05, ##P < 0.01 vs. TGF-β1; $P < 0.05, $$P < 0.01 vs. co-treatment.

3.5. Functional validation of SPARC as the primary effector downstream of the circHIPK3/miR-29a-3p axis

To functionally validate the roles of SPARC and IGF1 as downstream effectors, we performed loss-of-function experiments in TGF-β1-induced cells. From a screen of three siRNAs per gene, si-IGF1-1 and si-SPARC-1 were selected for all subsequent assays due to their superior knockdown efficacy (P < 0.01; Fig. 5A and B).

Fig. 5.

SPARC silencing exerts broader effects than IGF1 silencing on TGF-β1-induced epithelial reprogramming and migration.

(A) Representative Western blot images (upper) and quantitative analysis (lower) of IGF1 protein levels in BEAS-2B and A549 cells transfected with three different siRNAs targeting IGF1 (si-IGF1-1, si-IGF1-2, si-IGF1-3) or negative control siRNA (si-NC). (B) Representative Western blot images (left) and quantitative analysis (right) of SPARC protein levels in BEAS-2B and A549 cells transfected with three different siRNAs targeting SPARC (si-SPARC-1, si-SPARC-2, si-SPARC-3) or si-NC. (C) Representative images of wound healing assays in BEAS-2B and A549 cells under the following treatments: TGF-β1, TGF-β1 + si-IGF1, TGF-β1 + si-IGF1 + miR-29a inhibitor, TGF-β1 + si-SPARC, TGF-β1 + si-SPARC + miR-29a inhibitor. (D) Quantitative analysis of cell migration distance from the wound healing assays shown in (C). (E) Representative immunofluorescence images for E-cadherin, N-cadherin, and α-SMA (with F-actin counterstaining) in BEAS-2B and A549 cells under the same treatment conditions. Nuclei were stained with DAPI (blue). (F-H) Quantitative analysis of fluorescence intensity for (F) E-cadherin, (G) N-cadherin, and (H) α-SMA, corresponding to the IF images shown in (E). Data are mean ± SD (n = 3). *P < 0.05, **P < 0.01 vs. TGF-β1; #P < 0.05, ##P < 0.01 vs. si-IGF1; $P < 0.05, $$P < 0.01 vs. si-SPARC.

We first assessed the functional impact of silencing these effectors on cell migration. In wound healing assays (Fig. 5C and D), both si-IGF1 and si-SPARC significantly inhibited TGF-β1-induced cell migration (P < 0.01 for both). However, the anti-migratory effect of si-SPARC was more pronounced than that of si-IGF1 (P < 0.01). Notably, co-treatment with a miR-29a-3p inhibitor significantly reversed the anti-migratory effect of si-SPARC (P < 0.01), but failed to rescue the effect of si-IGF1 (P > 0.05). These findings indicate that SPARC, but not IGF1, acts as a primary functional mediator of miR-29a-3p-dependent regulation of cell migration in this context.

We further evaluated the impact on epithelial reprogramming via immunofluorescence staining for E-cadherin, N-cadherin, and α-SMA (Fig. 5E–H). Consistent with the migration data, si-SPARC significantly reversed TGF-β1-induced changes across all three markers in both cell lines (all P < 0.01), and these effects were largely abolished by miR-29a-3p inhibition (P < 0.01). In contrast, si-IGF1 produced weaker and less consistent effects: it significantly upregulated E-cadherin only in BEAS-2B cells (P < 0.05), with limited effects on N-cadherin or α-SMA in either cell line (all P > 0.05). While miR-29a-3p inhibition partially reversed the si-IGF1-induced E-cadherin increase in BEAS-2B cells (P < 0.05), this effect was not observed in A549 cells or for other markers.

In wound healing assays, si-SPARC, but not si-IGF1, reversed TGF-β1-induced cell migration in a miR-29a-3p-dependent manner (Fig. 5C and D). Similarly, si-SPARC, but not si-IGF1, significantly reversed TGF-β1-induced changes in E-cadherin, N-cadherin, and α-SMA expression (Fig. 5E–H). Therefore, SPARC was selected for further mechanistic and in vivo studies.

3.6. SPARC promotes fibrosis in association with enhanced TGF-β/SMAD signaling in a miR-29a-3p-dependent manner

To examine the role of SPARC in driving fibrosis, we first examined its effects on the TGF-β/SMAD pathway. Overexpression of SPARC (OE-SPARC) in both BEAS-2B and A549 cells significantly increased its own protein levels (P < 0.01, Fig. 6A). When cells were stimulated with TGF-β1, OE-SPARC enhanced the expression of key fibrotic proteins, including Vimentin, N-cadherin, the transcription factors SNAI1 and SNAI2, as well as the phosphorylation of SMAD2 and SMAD3 (all P < 0.01, Fig. 6B, C-J). Conversely, silencing SPARC (si-SPARC) suppressed the expression of these same markers (P < 0.01). Critically, the pro-fibrotic effects of OE-SPARC could be significantly rescued by co-expression of the miR-29a-3p mimic, while the anti-fibrotic effects of si-SPARC were reversed by a miR-29a-3p inhibitor (P < 0.05 or P < 0.01, Fig. 6C–J). Modulation of miR-29a-3p alone also influenced these markers, albeit with less consistent and potent effects than direct SPARC manipulation. Notably, total TGFBR1 and total TGFBR2 protein levels in whole cell lysates remained largely unchanged across these manipulations in vitro (Fig. 6B). This observation suggests that, in this acute setting, the observed enhancement of signaling is not due to increased total TGFBR2 abundance, but may involve other mechanisms such as altered receptor complex assembly, trafficking, or stability.

Fig. 6.

SPARC enhances TGF-β/SMAD signaling pathway activity and associates with TGFBR2.

(A) Western blot analysis (upper) and quantitative analysis (lower) demonstrating efficient SPARC overexpression (OE-SPARC) in BEAS-2B and A549 cells. (B) Representative Western blot images showing the expression of fibrotic pathway proteins (Vimentin, N-cadherin, E-cadherin, SNAI1, SNAI2, p-SMAD2, SMAD2, p-SMAD3, SMAD3, TGFBR1, TGFBR2, SPARC, GAPDH) in BEAS-2B and A549 cells under the indicated treatments with TGF-β1 stimulation. (C-J) Quantitative analysis of Western blot results for (C) Vimentin, (D) N-cadherin, (E) E-cadherin, (F) SNAI1, (G) SNAI2, (H) p-SMAD2/SMAD2 ratio, (I) p-SMAD3/SMAD3 ratio, and (J) SPARC, corresponding to the blots in (B). (K) Colocalization of SPARC (green) and TGFBR2 (red) in BEAS-2B and A549 cells. Nuclei were stained with DAPI (blue). Pearson's correlation coefficient is shown. (L) Co-immunoprecipitation (Co-IP) analysis of the SPARC and TGFBR2 association. Cell lysates from BEAS-2B and A549 cells were subjected to IP with the indicated antibodies, followed by Western blot for SPARC and TGFBR2. Input represents 5% of total lysate. (M, N) Quantitative analysis of Co-IP efficiency, presented as the percentage of input co-immunoprecipitated for (M) SPARC and (N) TGFBR2. (O) TGF-β-responsive luciferase reporter assay in BEAS-2B and A549 cells. Cells were transfected with the reporter plasmid and treated as indicated. Relative luciferase activity (fold change) is shown. Data are mean ± SD (n = 3). *P < 0.05, *P < 0.01 vs. TGF-β1 (C-J), vs. IgG control (M-N), or vs. Control (O); #P < 0.05, ##P < 0.01 vs. miR-29a mimic (C-J) or vs. OE-SPARC (O); $P < 0.05, $$P < 0.01 vs. miR-29a inhibitor (C-J) or vs. OE-SPARC + TGF-β1 (O).

To functionally assess the impact of SPARC on TGF-β/SMAD transcriptional activity, we employed a TGF-β-responsive luciferase reporter. OE-SPARC alone significantly activated the reporter in both cell lines (P < 0.01, Fig. 6O), although to a lesser extent than direct TGF-β1 stimulation. This SPARC-induced activation was attenuated by either a TGF-β receptor kinase inhibitor (TGF-β Ri) or a dominant-negative SMAD3 (dn-SMAD3) (P < 0.01). Combined treatment with OE-SPARC and TGF-β1 resulted in the highest level of reporter activation (P < 0.01, Fig. 6O), indicating that the two stimuli can cooperate to enhance pathway activity.

To explore the molecular basis of this functional interaction, we examined whether SPARC and TGFBR2 can be found in the same protein complex. Co-immunoprecipitation (Co-IP) assays showed that endogenous SPARC and TGFBR2 associate in lysates of lung epithelial cells: antibodies against either protein could co-immunoprecipitate the other, whereas control IgG did not (Fig. 6L). These data indicate that SPARC and TGFBR2 reside in a common complex; whether this reflects direct binding or indirect association through additional adaptor proteins remains to be determined. Quantitative analysis confirmed that the specific co-immunoprecipitation signals were significantly higher than background levels in the IgG control (P < 0.01, Fig. 6M and N). These findings were corroborated by immunofluorescence analysis, which revealed colocalization of SPARC and TGFBR2 in cellular compartments (Fig. 6K). Quantitative colocalization analysis showed higher Pearson’s correlation coefficients in BEAS-2B cells (r = 0.647) than in A549 cells (r = 0.196), suggesting cell type-specific differences in SPARC-TGFBR2 colocalization.

SPARC overexpression enhanced TGF-β1-induced expression of Vimentin, N-cadherin, SNAI1, SNAI2, and phosphorylation of SMAD2/3, whereas SPARC silencing suppressed these changes (Fig. 6B–J). The TGF-β/SMAD reporter activity was increased by SPARC overexpression (Fig. 6O). Co-immunoprecipitation showed that SPARC and TGFBR2 can be co-precipitated from cell lysates (Fig. 6L–N), and immunofluorescence revealed their colocalization (Fig. 6K).

3.7. In vivo modulation of the circHIPK3/miR-29a-3p/SPARC axis in bleomycin-induced pulmonary fibrosis

To validate our findings in a pathophysiologically relevant context, we employed a bleomycin (BLM)-induced murine model of pulmonary fibrosis (Fig. 7A). The high conservation of the circHIPK3 junction and the miR-29a-3p sequence between species (Supplementary Fig. S1B–D) underscored the translational relevance of our murine model. Mice were randomly assigned to six groups (n = 5 per group): Normal, PF model, PF + antagomiR-29a, PF + ASO-circHIPK3, PF + antagomiR-29a + si-SPARC, and PF + ASO-circHIPK3 + si-SPARC. Histopathological examination confirmed successful model establishment, with the PF model showing markedly thickened alveolar walls and structural destruction compared to normal controls (Fig. 7B).

Fig. 7.

In vivo validation of the circHIPK3/miR-29a-3p/SPARC axis in a bleomycin-induced murine model of pulmonary fibrosis.

(A) Schematic timeline of the bleomycin-induced pulmonary fibrosis model and therapeutic intervention schedule. (B) Representative H&E-stained lung tissue sections from Normal and PF Model mice. (C) RT-qPCR analysis of circHIPK3 (mmu_circ_0001052) expression in lung tissues from Normal, PF Model, and PF Model + ASO-circHIPK3 treated mice. (D) RT-qPCR analysis of linear HIPK3 mRNA expression. (E) RT-qPCR analysis of miR-29a-3p expression in lung tissues from Normal, PF Model, and PF Model + antagomiR-29a treated mice. (F) Representative Western blot images showing the expression of fibrotic pathway proteins (Vimentin, N-cadherin, E-cadherin, SNAI1, SNAI2, p-SMAD2, SMAD2, p-SMAD3, SMAD3, TGFBR2, SPARC, GAPDH) in lung tissues from the six treatment groups. (G-O) Quantitative analysis of Western blot results for (G) Vimentin, (H) N-cadherin, (I) E-cadherin, (J) SNAI1, (K) SNAI2, (L) p-SMAD2/SMAD2 ratio, (M) p-SMAD3/SMAD3 ratio, (N) TGFBR2, and (O) SPARC. Data are mean ± SD (n = 5). *P < 0.05, **P < 0.01 vs. Normal; #P < 0.05, ##P < 0.01 vs. PF Model; $P < 0.05, $$P < 0.01 vs. PF + antagomiR-29a; &&P < 0.01 vs. PF Model + ASO-circHIPK3 group.

We first confirmed the dysregulation of key axis components in the model. circHIPK3 (mmu_circ_0001052) expression was significantly upregulated in fibrotic lungs (P < 0.01, Fig. 7C), while linear HIPK3 mRNA remained unchanged (Fig. 7D). Similarly, miR-29a-3p expression was significantly downregulated in the model (P < 0.01, Fig. 7E). At the protein level, SPARC was significantly elevated in the PF model (P < 0.01, Fig. 7F–O).

We next evaluated the functional effects of targeting different nodes of the axis on the fibrotic phenotype. As shown in Fig. 7F–O, the PF model led to significant upregulation of pro-fibrotic proteins (Vimentin, N-cadherin, SNAI1, SNAI2, p-SMAD2, p-SMAD3) and downregulation of E-cadherin (all P < 0.05 or P < 0.01). Administration of antagomiR-29a increased the expression of fibrotic markers compared to the PF model (P < 0.01). ASO-circHIPK3 treatment decreased these markers compared to the PF model (P < 0.05 for Vimentin, N-cadherin, SNAI1, SNAI2, p-SMAD2, p-SMAD3; P < 0.01 for SPARC). Co-treatment with si-SPARC reversed the antagomiR-29a-induced increases (P < 0.05 or P < 0.01). In the ASO-circHIPK3 background, si-SPARC co-treatment did not further reduce fibrotic marker expression (P > 0.05). Of note, total TGFBR2 protein levels remained unchanged across all experimental groups (Fig. 7F–N), and TGFBR1 levels were similarly unaffected, indicating that the observed modulation of TGF-β/SMAD signaling occurs through regulation of pathway activity rather than through changes in receptor abundance.

Thus, in this in vivo model, circHIPK3 knockdown reduced fibrotic marker expression, and this effect was associated with increased miR-29a-3p and decreased SPARC. Silencing SPARC alone also reversed the changes induced by miR-29a-3p inhibition.

3.8. SPARC silencing attenuates fibrosis-associated histopathological changes in antagomiR-29a-treated mice

To determine the relative contributions of circHIPK3 and SPARC in regulating ECM remodeling under conditions of miR-29a-3p modulation, we analyzed lung histopathology and ECM deposition via H&E staining, Masson's trichrome staining, and IHC for key ECM proteins (Collagen I, Fibronectin, α-SMA) across the six experimental groups (Fig. 8A).

Fig. 8.

Effects of circHIPK3 and SPARC silencing on lung histopathology, extracellular matrix deposition, and myofibroblast-associated markers.

(A) H&E, Masson's trichrome, and IHC staining for Collagen I, Fibronectin, and α-SMA in lung sections. (B, C) Quantitative Ashcroft fibrosis scores based on H&E staining (B) and Masson's trichrome staining (C). (D-F) Quantitative analysis of IHC staining intensity (H-score) for (D) Collagen I, (E) Fibronectin, and (F) α-SMA. (G) Quantitative analysis of hydroxyproline content (μg/mg wet lung tissue). Data are mean ± SD (n = 5). *P < 0.05, **P < 0.01 vs. Normal; #P < 0.05, ##P < 0.01 vs. PF Model; $P < 0.05, $$P < 0.01 vs. PF + antagomiR-29a.

H&E and Masson's trichrome staining revealed substantial collagen deposition and alveolar destruction in the PF model, which were markedly exacerbated by antagomiR-29a administration. Quantitative Ashcroft scoring (Fig. 8B and C) confirmed these observations: fibrosis scores were significantly elevated in the PF model (P < 0.05 for H&E; P < 0.01 for Masson) and further worsened by antagomiR-29a (P < 0.01). Notably, ASO-circHIPK3 treatment significantly reduced fibrosis scores compared to the PF model (P < 0.01 for Masson scoring). Critically, si-SPARC co-treatment significantly reversed the antagomiR-29a-exacerbated fibrosis (P < 0.01), but showed no additional effect when combined with ASO-circHIPK3 (P > 0.05).

Quantitative IHC analysis (Fig. 8D–F) showed a similar pattern for ECM protein deposition. Expression of Collagen I, Fibronectin, and α-SMA was significantly upregulated in the PF model (P < 0.01) and was further increased by antagomiR-29a (P < 0.01). ASO-circHIPK3 treatment significantly reversed these increases (P < 0.05 for all three markers). si-SPARC co-treatment effectively reversed the antagomiR-29a-induced upregulation of all three ECM markers (P < 0.05 for Collagen I and Fibronectin; P < 0.01 for α-SMA), whereas it showed no additional effect in the ASO-circHIPK3 background (P > 0.05). Quantitative hydroxyproline analysis (Fig. 8G) confirmed these findings: collagen content was significantly elevated in PF model lungs and further increased by antagomiR-29a (P < 0.01). This increase was reversed by ASO-circHIPK3 (P < 0.05) and by si-SPARC co-treatment in the antagomiR-29a background (P < 0.05), while si-SPARC showed no additional effect in the ASO-circHIPK3 group (P > 0.05).

Quantitative analysis of Ashcroft scores (Fig. 8B and C), IHC staining for Collagen I, Fibronectin, and α-SMA (Fig. 8D–F), and hydroxyproline content (Fig. 8G) showed that antagomiR-29a increased ECM deposition compared to the PF model, and this increase was reversed by co-treatment with si-SPARC. In contrast, si-SPARC did not further reduce ECM deposition in the ASO-circHIPK3 background.

4. Discussion

IPF is a fatal lung disease with limited treatment options [2,28]. Here we identify a circHIPK3/miR-29a-3p/SPARC axis that promotes pro-fibrotic epithelial reprogramming in experimental pulmonary fibrosis. This axis amplifies TGF-β signaling, with circHIPK3 acting upstream of miR-29a-3p.

circHIPK3 has been implicated in cancer [10] and cardiac fibrosis [11]. circHIPK3 has been shown to promote pulmonary fibrosis through fibroblast-centered mechanisms [12,13]. In pulmonary fibrosis, previous studies have focused on its role in fibroblast activation, demonstrating that circHIPK3 promotes fibroblast-to-myofibroblast transition via miR-338-3p [12] and regulates glycolysis through miR-30a-3p/FOXK2 [13]. Here, we extend these findings by unveiling a distinct function for circHIPK3 in driving pro-fibrotic epithelial reprogramming via the miR-29a-3p/SPARC axis, suggesting that a single circRNA can regulate multiple pathological processes through different miRNA partners in a cell-type-specific manner [7,10].

A central finding of our study is the identification of miR-29a-3p as the primary functional target of circHIPK3 in this context. The miR-29 family is a well-established master regulator of ECM homeostasis, and its loss is a conserved feature of fibrosis across multiple organs [16,17,29]. The mechanisms underlying miR-29 suppression, however, remain incompletely understood. Our data support a mechanism by which circHIPK3 regulates miR-29a-3p availability, consistent with the ceRNA hypothesis. Direct association between circHIPK3 and miR-29a-3p was validated by dual-luciferase reporter and RNA pull-down assays. Functionally, circHIPK3 knockdown and miR-29a-3p mimic treatment each attenuated TGF-β1-induced epithelial phenotypic changes and migration, and their combination produced a greater inhibitory effect under the tested conditions (Fig. 3), consistent with a ceRNA mechanism. This mechanism is consistent with reports of other circRNAs functioning similarly in fibrosis [8], highlighting a common pathogenic strategy.

The systematic identification of SPARC as a critical downstream effector of the circHIPK3/miR-29a-3p axis is a key contribution. While bioinformatics predicted several candidates, including collagens (COL1A1, COL4A1), which are canonical targets of miR-29 [16], SPARC emerged as the most consistently and potently regulated gene across our cellular and animal models. SPARC is a matricellular protein known to influence cell-ECM interactions and is upregulated in various fibrotic conditions [30,31]. Our data extend these observations by demonstrating that derepression of SPARC via the circHIPK3/miR-29a-3p axis correlates with enhanced TGF-β/SMAD signaling.

Co-immunoprecipitation and colocalization studies revealed that SPARC and TGFBR2 can be found within the same protein complex, providing a potential mechanistic link. Total TGFBR1 and TGFBR2 protein levels remained largely unchanged upon SPARC manipulation in vitro, and both receptors were similarly unaffected in vivo. This indicates that SPARC enhances TGF-β/SMAD signaling without altering the steady-state expression of either receptor component. Rather, the observed signaling enhancement likely reflects modulation of receptor complex function—such as facilitating TGFBR2-containing complex assembly, stability, or downstream signaling kinetics—consistent with SPARC's proposed role as a matricellular modulator of receptor complex activity [32]. This configuration creates a positive regulatory circuit: TGF-β1 induces circHIPK3, which reduces the available pool of miR-29a-3p, leading to SPARC derepression; SPARC in turn amplifies TGF-β signaling. Such a circuit could sustain fibrogenic signaling, although formal demonstration of a self-amplifying loop requires additional kinetic or perturbation experiments. Consistent with this concept, SPARC has been implicated as an important regulator of TGF-β-driven fibrotic responses in pulmonary fibrosis and has been shown to participate in TGF-β-mediated profibrotic signaling in lung fibroblasts [33].

Beyond amplifying TGF-β signaling, the pro-fibrotic actions of SPARC are likely pleiotropic. SPARC can directly regulate ECM assembly [34] and collagen fibrillogenesis [35], influence integrin-mediated cell-ECM adhesion and signaling, and modulate the activity of other growth factors such as PDGF and VEGF [36]. Therefore, the anti-fibrotic effect observed upon SPARC silencing in our study is probably the result of concurrently disrupting both a key TGF-β-amplifying loop and several other TGF-β-independent pro-fibrotic processes. This multifaceted role underscores SPARC's position as a central node in the fibrotic network and enhances its appeal as a therapeutic target.

The anti-fibrotic effect of circHIPK3 silencing was accompanied by restored miR-29a-3p levels and reduced SPARC expression, while SPARC knockdown attenuated the fibrotic effects induced by miR-29a-3p inhibition. These observations support a downstream role of SPARC in miR-29a-3p-mediated fibrotic responses. However, the epistatic relationship between circHIPK3 and miR-29a-3p was not directly tested in the current experimental design and warrants further investigation. Of note, a modest but statistically significant reduction was observed for SPARC expression and SMAD3 phosphorylation, suggesting either incomplete miR-29a-3p inhibition or minor miR-29a-3p-independent mechanisms. This nuanced finding does not detract from the centrality of the circHIPK3/miR-29a-3p axis, but rather highlights the complexity of circRNA biology. This strategy of targeting a pathway node is gaining traction in complex diseases [37]. The high conservation of the miR-29a-3p sequence and its binding sites on circHIPK3 between humans and mice underscores the relevance of this regulatory mechanism to human IPF.

While our initial screening also identified IGF1 as a direct miR-29a-3p target, functional validation revealed that its contribution to epithelial pro-fibrotic phenotypes was limited and inconsistent compared to SPARC. In wound healing assays, si-IGF1 produced a modest anti-migratory effect that was not rescued by miR-29a-3p inhibition, whereas si-SPARC exerted a potent effect that was fully miR-29a-3p-dependent. Similarly, immunofluorescence analysis showed that si-SPARC, but not si-IGF1, consistently reversed TGF-β1-induced changes in epithelial reprogramming markers across both cell lines. These findings align with the well-established role of IGF1 in fibroblast activation and proliferation [38]—processes central to fibrosis but distinct from the epithelial reprogramming examined here—and the presence of IGF-binding proteins that add complexity to its signaling [39]. Therefore, our data identify SPARC as the primary functional effector of the circHIPK3/miR-29a-3p axis in driving epithelial pro-fibrotic reprogramming.

The identification of miR-29a-3p as a key mediator in this axis has potential translational implications. Recent studies have explored lung-targeted delivery of miR-29 mimics as a therapeutic strategy for IPF [40,41]. Our findings suggest that the efficacy of such approaches could be influenced by endogenous circHIPK3 levels, which may act as a molecular sink for the delivered miRNA. Conversely, combining miR-29 mimic therapy with circHIPK3 targeting could potentially enhance therapeutic outcomes by relieving this endogenous buffering capacity. This concept aligns with emerging strategies that aim to restore miRNA activity by simultaneously inhibiting their endogenous sponges [42]. It is noteworthy that A549 and BEAS-2B cells, representing alveolar and bronchial epithelial origins respectively, may possess inherent differences in their signaling networks [43]. Our findings in both cell lines collectively support the proposed pro-fibrotic axis.

Despite these insights, our study has limitations that warrant critical acknowledgement. First, the translational potential of targeting circHIPK3 requires further validation. The in vivo delivery efficiency, biodistribution, and long-term safety profile of circHIPK3-targeting ASOs need comprehensive evaluation before clinical application can be considered. Second, our mechanistic findings are primarily based on two immortalized lung epithelial cell lines: BEAS-2B (SV40-transformed bronchial epithelial) and A549 (alveolar adenocarcinoma-derived). While these cell lines are widely used and have been instrumental for our initial mechanistic dissection, we acknowledge that their immortalized and/or cancer-derived nature may not fully recapitulate the behavior of primary human lung epithelial cells. Future validation in primary human alveolar epithelial cells would be essential to strengthen the pathophysiological relevance of the proposed axis. Third, the precise molecular link between SPARC and TGF-β signaling remains to be fully elucidated. While our data demonstrate an association between SPARC and TGFBR2 within a common complex and show that SPARC manipulation correlates with changes in TGF-β/SMAD pathway activity, the nature of this regulation—whether it involves a direct protein-protein interaction, affects receptor stability, endocytosis, trafficking, or dimerization—is not yet defined. Future studies using purified recombinant proteins (e.g., surface plasmon resonance or co-crystallization) will be required to determine whether SPARC and TGFBR2 interact directly or indirectly through adaptor proteins. The observation that total TGFBR1 and TGFBR2 levels are largely unaffected by SPARC manipulation suggests that SPARC acts primarily at the level of receptor complex function rather than receptor expression, a model consistent with its proposed role as a matricellular protein that modulates signaling complex activity. Consequently, the impact of this axis on TGFBR2 phosphorylation and activation kinetics also remains an important question for future investigation. Fourth, the functional contribution of IGF1 within this network appears to be context-dependent. Its less consistent phenotype compared to SPARC suggests cell-type-specific or process-specific roles, potentially being more prominent in fibroblast activation than in the epithelial reprogramming process central to our study. Fifth, while our data are consistent with a ceRNA-like regulatory relationship for circHIPK3, the full characterization of this interaction—including definitive demonstration of cytoplasmic localization, AGO2 association, and complete rescue of circHIPK3 overexpression effects by miR-29a-3p—awaits future investigation. We also acknowledge that direct biophysical evidence of circHIPK3/miR-29a-3p binding, such as EMSA with alkaline mutagenesis, was not performed. However, our dual-luciferase reporter assay with WT/MUT constructs and RNA pull-down assay provided complementary evidence supporting specific interaction. Therefore, we interpret our findings as consistent with a ceRNA mechanism rather than definitive proof. These studies would further solidify the mechanistic model proposed here. Notably, miR-29a-3p has been reported to target DNMT3A/3B in diabetic kidney disease and lung fibroblasts [44,45]. Our identification of SPARC (and IGF1) as additional targets in lung epithelial cells extends this knowledge. Together, these findings suggest that miR-29a-3p orchestrates its anti-fibrotic effects through a network of multiple downstream effectors in a cell-type-specific manner, which may explain its broad therapeutic potential across different fibrotic diseases. Additionally, we acknowledge that unbiased circRNA profiling was not performed; therefore, we cannot determine whether circHIPK3 is the most highly upregulated or dominant circRNA under TGF-β1 stimulation. Our study was designed as a hypothesis-driven mechanistic investigation rather than a discovery-oriented screen. Unbiased circRNA sequencing will be valuable in future studies to identify the full landscape of TGF-β1-responsive circRNAs in lung epithelial cells. Finally, while our interventions demonstrated anti-fibrotic effects at the molecular, histopathological, and biochemical levels—as evidenced by changes in fibrotic markers, Ashcroft scoring, and hydroxyproline content—the study lacks direct measurements of lung physiological function. We acknowledge that improvements in these markers do not necessarily equate to improved lung function. Future studies with comprehensive functional assessments, including pulmonary function tests, will be essential to determine whether the observed anti-fibrotic effects translate into improved lung mechanics. The upstream mechanism by which TGF-β1 selectively upregulates circHIPK3 without affecting linear HIPK3 mRNA remains unclear and warrants future investigation. In summary, our findings support a model in which TGF-β1 induces circHIPK3, which reduces the available pool of miR-29a-3p and derepresses SPARC, leading to enhanced TGF-β/SMAD signaling and pro-fibrotic epithelial reprogramming.

5. Conclusion

In summary, our data identify a circHIPK3/miR-29a-3p/SPARC axis that promotes pro-fibrotic epithelial reprogramming with associated enhancement of TGF-β/SMAD signaling in experimental pulmonary fibrosis. These findings provide a mechanistic basis for further investigation of this axis in human IPF. We provide experimental evidence that: 1) circHIPK3 is upregulated by TGF-β1 and in fibrotic lungs; 2) it can directly associate with miR-29a-3p and regulate its downstream targets, consistent with the proposed ceRNA hypothesis; and 3) the derepression of SPARC, a direct miR-29a-3p target, is associated with enhanced TGF-β/SMAD signaling, an effect that correlates with its presence in a complex with TGFBR2. The precise molecular mechanism by which SPARC modulates TGF-β receptor complex activity remains to be fully defined. Collectively, our findings support a model in which this axis constitutes a regulatory circuit that enhances fibrotic signaling. In vivo data support a downstream role of SPARC in miR-29a-3p-mediated fibrotic responses, whereas the epistatic relationship between circHIPK3 and miR-29a-3p remains to be established. While this study identifies circHIPK3 as a key upstream regulator in the fibrotic network potentially involving ceRNA mechanisms, and demonstrates anti-fibrotic effects at the molecular and histopathological level, future work is essential to assess whether targeting this node translates into improved lung physiological function in more advanced, chronic disease models. Additionally, full elucidation of the molecular details underlying the SPARC-TGFBR2 association and its role in signal amplification remains an important direction for future investigation.

CRediT authorship contribution statement

Ying Zhou: Writing – original draft, Project administration, Methodology, Conceptualization. Yixin Huang: Writing – original draft, Methodology, Data curation. Jiangpo Ma: Writing – original draft, Resources, Conceptualization. Wei Wang: Investigation, Formal analysis. Zhongkai Tong: Visualization, Formal analysis. Yong Zhou: Writing – review & editing, Resources, Data curation. Yu Cao: Validation, Project administration. Zhaoxing Dong: Writing – review & editing, Supervision, Funding acquisition.

Ethics approval and consent to participate

All animal experiments and studies were conducted following institutional ethical guidelines and were approved by the Ethics Committee of Ningbo University. The study was reviewed and approved by the Ningbo University Animal Ethics Committee (Approval No. 11373) and complied with the ARRIVE guidelines. All animal procedures were conducted in accordance with the guidelines of the respective animal research institutions, and no privately owned animals were used. Therefore, informed consent from individual owners was not applicable.

Data availability

The datasets supporting the conclusions of this article are included within the article (and its additional files). All data supporting the findings of this study are available from the corresponding authors upon reasonable request.

Funding

This study was supported by the National Natural Science Foundation of China (Project ID: 82170074), Zhejiang Provincial Nature Science Foundation of China (Grant No. Y24H010013), Ningbo Natural Science Foundation (Project ID: 2022J241), Ningbo Clinical Research Center for Respiratory Disease (Project ID: 2022L004), and Talent Introduction Project of Yongjiang River (Project ID: 2021B-016-G). Ningbo Key Laboratory of Digital Imaging and Medical-Engineering Integration(Grant No.2024031),Ningbo Top Medical and Health Research Program (Grant No.2024020919). This study was also supported by the Ningbo Municipal Key Clinical Specialty in Respiratory Medicine (No. 2025002) and the Natural Science Foundation of Ningbo (No. 2024J329).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article.

Supplementary Fig. S1.

Validation of circHIPK3 specificity and conservation analysis.

(A) PCR analysis of cDNA from BEAS-2B cells using convergent primers for linear HIPK3 mRNA and divergent primers spanning the back-splice junction for circHIPK3, with or without RNase R treatment. (B) Sequence alignment of mature miR-29a-3p from Homo sapiens and Mus musculus, demonstrating 100% conservation. (C) Schematic illustration of the generation of hsa_circHIPK3 (hsa_circ_0000284) from human HIPK3 gene exon 2 and its murine homolog mmu_circHIPK3 (mmu_circ_0001052), demonstrating consistent size and identical back-splice junction sequence. (D) Sequence alignment of the junction region of human hsa_circHIPK3 and the homologous murine genomic region, showing 89% identity. The predicted miR-29a-3p binding site is highlighted.