Raw circRNA expression matrices for HCC were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo) by using ‘circRNA’ and ‘hepatocellular carcinoma’ as search terms. Two GEO datasets, GSE97332 (circRNA expression in HCC and matched non-tumor tissues) (22) and GSE242797 (construct a ceRNA regulatory network to explore potential pathogenesis and therapy options of human hepatocellular carcinoma) (23), were used for differential expression analysis. The names of the circRNAs were standardized according to the circBase database (http://www.circbase.org/). The differential expression of circRNAs was analysed by the R package ‘limma’ (version 4.2.2; Posit Software, PBC) with log2 (fold-change) >1.5 and P<0.05 (24). Host gene promoter sequences were obtained from the University of California, Santa Cruz database (https://genome.ucsc.edu/) and subsequently submitted to the JASPAR database (https://jaspar.elixir.no/) to predict potential AR binding sites. The Cancer Genome Atlas (TCGA) database (https://www.cancer.gov/tcga) data were analyzed using the University of Alabama at Birmingham Cancer data analysis portal online tool (http://ualcan.path.uab.edu) to assess differential gene expression. The median expression value was used as the cutoff to define high- and low-expression groups.

THLE-2 normal hepatocytes and liver cancer cell lines MHCC97H, HCCLM3 and HepG2 were obtained from iCell Bioscience Inc. Short tandem repeat profiling was performed externally by the cell supplier to confirm the identity of each line, and all tested negative for Mycoplasma. Using DMEM (Gibco; Thermo Fisher Scientific, Inc.) supplemented with 10% FBS (Gibco; Thermo Fisher Scientific, Inc.), cells were kept at 37°C with 5% CO₂ in a humidified incubator. For overexpression, the hsa_circ_0011385 sequence was cloned into the pCD2.1-ciR circRNA expression vector (Guangzhou Geneseed Biotech. Co., Ltd.) using 50 ng of vector per ligation reaction and a 3:1 molar ratio of insert to vector, as recommended by standard protocols. Cells transfected with the corresponding empty vector were used as the negative control in these experiments. Small interfering RNA (siRNA), overexpression plasmids and corresponding control vectors were manufactured by Wuhan GeneCreate Biological Engineering Co., Ltd. Lipofectamine™ 2000 (Invitrogen; Thermo Fisher Scientific, Inc.) was used for cell transfection in complete compliance with the manufacturer's instructions. The sequences used were as follows: AR-si1 sense: 5′-GGUUCAACAGUCCACAATT-3′, antisense: 5′-UUGGUGAGUGGUAGAAGCTT-3′; AR-si2 sense: 5′-CUGAUCUGUGGAGAUGAATT-3′, antisense: 5′-UCAUCUCACAGAUCAGTT-3′; AR-si3 sense: 5′-GCAAGGUUGUGCUAGUATT-3′, antisense: 5′-UACUAGCACAAGCUUGCTT-3′; si-negative control sense: 5′-UUCUCCGAACGUGUCACGUTT-3′. Among these, si-AR-3 exhibited the most potent AR knock-down and was therefore selected for all subsequent functional experiments, which were performed at 48 h post-transfection.

Lenvatinib was purchased from Beyotime Biotechnology. A lenvatinib-resistant cell line was generated using a stepwise concentration-increasing method. MHCC97H cells at logarithmic growth phase were treated with a starting concentration of ~10 µM (half of the 72-h IC₅₀ value). The concentration was doubled every 2–3 weeks until reaching 80 µM and cells were then stably passaged at this concentration for 4 weeks. Throughout the procedure, cells were maintained in a 5% CO₂ humidified incubator at 37°C, with drug-containing medium refreshed every 48 h. Cells were passaged upon reaching ~90% confluence. The entire induction process lasted ~4 months. The final resistant cell line obtained was named MHCC97H-LR (lenvatinib-resistant).

MHCC97H cells were seeded in 6-well plates at a density of 3×105 cells per well and cultured overnight. Cells were then treated with 10 µM enzalutamide (SC0074-10 mM; Beyotime Biotechnology) at 37°C for 48 h. Control cells were treated with DMSO (cat. no. ST038; Beyotime Biotechnology) at a final concentration of 0.1% at 37°C. After 48 h, cells were harvested for RNA extraction.

Cells from all experimental groups (including normal hepatocytes, HCC cell lines, lenvatinib-resistant cells and transfected cells) were lysed for total RNA or miRNA isolation with the Animal Total RNA Isolation Kit (cat. no. RE-03011; Chengdu Fuji Biotechnology Co., Ltd.) according to the manufacturer's instructions and the yield stored at −80°C. miRNA was reverse-transcribed with the SeqHunt^® First Strand cDNA Synthesis Kit (cat. no. CA01; Seq-Hunt Biotechnology Co., Ltd.). mRNA and microRNA levels were then quantified by RT-qPCR using 2× Blue Universal SYBR qPCR Master Mix (cat. no. AF07; Seq-Hunt Biotechnology Co., Ltd.) with GAPDH and U6 as the respective internal controls. The thermocycling conditions were as follows: 95°C for 5 min, followed by 40 cycles of 95°C for 15 sec and 60°C for 40 sec. Expression ratios were calculated using the 2^−ΔΔCq method (25) and primer sequences are provided in Table I.

To perform this assay, an optimal density of cells was seeded into 96-well plates and incubated overnight at 37°C under 5% CO₂ to ensure adherence. Following a 24-h exposure to 100 µl serially diluted drug (under identical culture conditions), cell viability was assessed. Specifically, 10 µl CCK-8 reagent (cat. no. CK04; Dojindo Laboratories, Inc.) was introduced per well and the plates were returned to the 37°C incubator for a 1 to 2-h development period. The reaction was then stopped by immediate mixing after adding 10 µl stop solution to each well. Finally, absorbance at 450 nm was recorded using a microplate reader (TMR-100; Tuoh Electromechanical Technology Co., Ltd.).

Following digestion and harvesting, cells were resuspended in serum-free DMEM (Gibco; Thermo Fisher Scientific, Inc.) for subsequent counting. A cellular suspension containing 1×10⁴ cells in 100 µl was applied to the upper chamber of a Transwell insert. Meanwhile, the lower compartment received 600 µl of complete medium supplemented with 10% FBS. After a 24-h incubation at 37°C and 5% CO₂, the insert was carefully withdrawn. Stationary cells remaining on the upper membrane surface were then cleared away. Subsequent steps for the migrated cells on the lower membrane involved sequential processing: First, fixation in 4% paraformaldehyde (15 min), followed by staining with crystal violet (20 min), both at room temperature, followed by two washes with PBS (cat. no. BL302; Biosharp Life Sciences). Finally, the membranes were examined and photographed using an inverted MI40 microscope (Guangzhou Micro-Shot Technology Co., Ltd).

Apoptosis was detected by flow cytometry using the Annexin V-FITC/PI double-staining method, with experimental procedures strictly following the kit instructions (cat. no. KTA0004; Abbkine, Inc.). Briefly, after discarding the cell culture medium and collecting the supernatant, cells were gently washed twice with pre-chilled PBS. Digestion was performed using 0.25% EDTA-free trypsin, which was then terminated by adding complete medium containing 10% FBS, followed by gentle pipetting to obtain a single-cell suspension. The suspension was centrifuged at 120 × g for 5 min at room temperature. After discarding the supernatant, the cell pellet was resuspended in pre-chilled 1× PBS and washed twice again. Subsequently, cells were resuspended in pre-chilled 1× Annexin V binding buffer and adjusted to a density of 1×10⁶ cells/ml. A 100 µl aliquot of the cell suspension was taken and 5 µl Annexin V-FITC and 10 µl PI staining solution were added sequentially, followed by incubation at room temperature in the dark for 15 min. Finally, 400 µl binding buffer was added to each tube and samples were immediately analyzed using a flow cytometer (CytoFLEX; Beckman Coulter, Inc.). Data analysis was performed using CytExpert software (version 2.5; Beckman Coulter, Inc.).

For hsa_circ_0011385 and the Akt3 3′-untranslated region (UTR), luciferase reporters carrying either wild-type (WT) or mutant (MT) fragments were generated and named circ_0011385-WT, circ_0011385-MT, Akt3-WT and Akt3-MT, respectively. All constructs were sequence-verified prior to use. Subsequently, 293T cells were co-transfected with these reporters together with hsa-miR-212-5p mimics using the same transfection protocol as aforementioned (Lipofectamine 2000; Invitrogen; Thermo Fisher Scientific, Inc.); 48 h later, firefly and Renilla activities were quantified with the Dual-Luciferase Reporter Assay Kit (cat. no. JKR23008; Wuhan GeneCreate Biological Engineering Co., Ltd.). The miR-212-5p mimics (sense: 5′-ACCUUGGCUCUAGACUGCUUACU-3′, antisense: 5′-AGUAAGCAGUCUAGAGCCAAGGU-3′) and NC mimics (sense: 5-UCACAACCUCCUAGAAAGAGUAGA-3′, antisense: 5′-UCUACUCUUUCUAGGAGGUUGUGA-3′) were utilized in the present study.

Chromatin immunoprecipitation was carried out using the ChIP Kit (cat. no. JKR23002A; Wuhan GeneCreate Biological Engineering Co., Ltd.). After 10 min cross-linking with 1% formaldehyde, the cross-linking reaction was quenched through a 5-min incubation with 0.125 M glycine at room temperature. Subsequent to two washes with PBS, cells were pelleted by centrifugation at 1,000 × g for 5 min at 4°C and rapidly frozen at −80°C. For chromatin lysis, cells were resuspended in Lysis Buffer provided in the ChIP Kit (cat. no. JKR23002A; Wuhan GeneCreate Biological Engineering Co., Ltd.), supplemented with 1X Protease Inhibitor and 5 µl DTT according to the manufacturer's protocol. A 10-min lysis step was performed on ice. After centrifugation at 12,000 × g for 10 min at 4°C to pellet debris and remove the supernatant, the pellet was subjected to a second lysis and sonication. Chromatin fragments were then incubated with either AR antibody (cat. no. 22089-1-AP; Proteintech Group, Inc.) or control immunoglobulin G (normal rabbit IgG; cat. no. F040301; Beyotime Biotechnology), both used at a final concentration of 1 µg/ml, overnight at 4°C. After elution and reversal of cross-links (65°C, overnight), DNA was purified according to the kit protocol and analysed by RT-qPCR with primers specific for the EIF3I promoter region. The primer sequences used were as follows: Primer 1: forward 5′-TGCCAGGCACTCTCCTAGAT-3′ and reverse 5′-GTTGGTTTGAAGCCTGGCAG-3′; primer 2: forward 5′-CTGCCAGGCTTCAAACCAAC-3′ and reverse 5′-GTTTGAGGCCACCTGGAAGA-3′; and primer 3: forward 5′-CACAGAGACGCACCTCAGTAT-3′ and reverse 5′-TGACTCACTCGTCTGCATTC-3′.

Western blotting was performed as follows: Proteins were extracted from cells using RIPA buffer (cat. no. BL504A; Biosharp Life Sciences), quantified by BCA (BL521A; Biosharp Life Sciences) and 20 µg was denatured and separated by 10% SDS-PAGE (80/120 V). Proteins were wet-transferred (200 mA; 90 min; ice bath) to PVDF membranes (0.22 µm). After blocking (5% skimmed milk in TBST with 0.1% Tween-20; 1 h; room temperature), membranes were probed with primary antibodies (1:1,000-1:50,000 in 1% BSA-PBST; 4°C; overnight) and HRP-secondary antibodies (1:10,000 in 5% skim milk-PBST; 1 h; room temperature), with PBST washes between steps. Detection was conducted with the SuperKine™ West Pico PLUS Chemiluminescent Substrate (cat. no. BMU101-CN; Abbkine, Inc.; 1.5 min) and a ChemiScope 6100T imager (Clinx Science Instruments Co., Ltd.). Detailed information on the antibodies used is provided in Table II.

After determining parametric assumptions (normality by Shapiro-Wilk test and homogeneity of variances by Levene's test), pairwise comparisons were analyzed using unpaired Student's t tests. Multi-factor datasets were analyzed using one-way ANOVAs followed by Tukey's multiple-comparison tests, except for CCK-8 data which were analyzed by two-way ANOVA followed by Tukey's test. Data are presented as the mean ± SD from at least three independent experiments. P<0.05 was considered to indicate a statistically significant difference. All statistical analyses were performed using GraphPad Prism (version 7.0; Dotmatics), R software (version 4.2.2; Posit Software, PBC) or SPSS Statistics (version 24; IBM, Corp.).

Preliminary bioinformatic analysis of GEO microarray datasets (GSE97332 and GSE242797) identified 16 significantly upregulated and 16 significantly downregulated circRNAs in HCC tissues compared with paired normal tissues (log₂ FC >1.5; P<0.05), visualized as a heatmap and volcano plot (Fig. 1A). Analysis of TCGA data revealed high expression of two circRNAs previously implicated in HCC (13–16) along with their corresponding host genes: hsa_circ_0001955 [casein kinase 1 γ-1 (CSNK1G1)] and hsa_circ_0011385 (EIF3I; Fig. 1B). RT-qPCR analysis of AR expression in the same cell lines showed that, compared with THLE-2 normal hepatocytes, MHCC97H and HCCLM3 cell lines expressed significantly higher AR mRNA levels, whereas HepG2 cells exhibited the lowest AR expression (Fig. 1C). RT-qPCR determined that, relative to normal THLE-2 hepatocytes, hsa_circ_0001955 (CSNK1G1) and hsa_circ_0011385 (EIF3I) exhibited markedly elevated levels in MHCC97H and HCCLM3 cells, whereas HepG2 cells exhibited only faint expression (Fig. 1D).

MHCC97H-LR, a lenvatinib-resistant HCC derivative, was obtained by stepwise escalation of drug concentration. Relative to the parental population, the selected line exhibited markedly reduced lenvatinib sensitivity: IC₅₀ increased from 19.7–25.7 to 63.0–79.6 µM, giving a 3.1-fold resistance index (Fig. 2A). In these resistant cells, AR, Akt3 and hsa_circ_0011385 (EIF3I) were upregulated, while hsa-miR-212-5p expression was downregulated (Fig. 2B).

RT-qPCR showed that compared with the THLE-2 cell line, MHCC97H cells exhibited significantly higher expression of AR, hsa_circ_0011385, its host gene EIF3I and the downstream target gene Akt3, while hsa-miR-212-5p expression was significantly lower (Fig. 3A). RT-qPCR demonstrated an inhibitory effect after AR siRNA silencing (Fig. 3B) and showed that AR knockdown in MHCC97H cells led to significant downregulation of Akt3 and hsa_circ_0011385 (EIF3I) expression and significant upregulation of hsa-miR-212-5p expression (Fig. 3C). Western blotting analysis determined significantly higher protein expression of AR, EIF3I and Akt3 in MHCC97H cells compared with THLE-2 cells and demonstrated that AR knockdown relatively reduced EIF3I and Akt3 protein levels (Fig. 3D).

To define how AR regulates hsa_circ_0011385, the EIF3I promoter (source of circ_0011385) was scanned in JASPAR for AR motifs (Fig. 4A). ChIP-qPCR with three primer pairs verified strong AR enrichment at the predicted sites (Fig. 4B). Dual-luciferase assays showed that miR-212-5p directly bound circ_0011385-WT and Akt3-WT, reducing their reporter signals (Fig. 4C). To validate AR transcriptional regulation, MHCC97H cells were treated with the AR antagonist enzalutamide. Enzalutamide significantly reduced EIF3I mRNA and hsa_circ_0011385 levels compared with DMSO control (Fig. 4D), consistent with the ChIP-qPCR results. These findings determined a targeted regulatory cascade within the AR/hsa_circ_0011385/miR-212-5p/Akt3 axis.

RT-qPCR validated the overexpression effect of hsa_circ_0011385 (Fig. 5A). CCK-8 assay results showed that compared with parental MHCC97H cells, lenvatinib-resistant cells exhibited significantly enhanced proliferation capacity, which was significantly weakened upon AR interference (Fig. 5B). Transwell assays exhibited a similar trend with lenvatinib-resistant cells displaying strong migration capacity, which was reduced after AR interference (Fig. 5C). Flow cytometric apoptosis assays indicated that compared with MHCC97H, resistant cells exhibited a decreased apoptosis rate, which significantly increased after AR interference (Fig. 5D). Overexpression of hsa_circ_0011385 significantly reversed the inhibitory effects caused by AR knockdown.

To provide direct functional evidence for the role of miR-212-5p in lenvatinib resistance, MHCC97H-LR cells were transfected with miR-212-5p mimic or inhibitor. RT-qPCR and western blotting analyses showed that miR-212-5p mimic significantly downregulated Akt3 expression at both mRNA and protein levels, whereas miR-212-5p inhibitor upregulated Akt3 expression (Fig. 6A and B). The CCK-8 assay demonstrated that the miR-212-5p mimic suppressed the proliferation of lenvatinib-resistant MHCC97H-LR cells, while miR-212-5p inhibitor further enhanced proliferation (Fig. 6C). Flow cytometric apoptosis analysis revealed that miR-212-5p mimic increased the apoptosis rate, whereas miR-212-5p inhibitor decreased apoptosis in resistant cells (Fig. 6D). These results determined that miR-212-5p directly regulated Akt3 expression and lenvatinib sensitivity, supporting the proposed competing endogenous RNA (ceRNA) mechanism.