In contrast to the linear isoforms of HIPK3 mRNA, circHIPK3 was resistant to RNase R (Figure 1A). Fluorescence in situ hybridization (FISH) analysis showed that circHIPK3 was located mostly in the cytoplasm of RA-FLS (Supplementary Figure S1), and it suggested that the circHIPK3 might function as an miRNA sponge. To mimic the (strengthened) microenvironment of the synovial tissue, we set up the RA-FLS and HDMEC co-culture system in the transwell apparatus besides the presence of TNF-α. To further clarify the relationship of circHIPK3 with angiogenesis under synovial inflammatory microenvironment, experiments with circHIPK3 knockdown were carefully performed. The knockdown efficiency of circHIPK3 was determined by qRT–PCR, and 60% knockdown efficiency was achieved, while the expression of linear HIPK3 mRNA was not significantly influenced (Supplementary Figure S2A). The expression level of circHIPK3 was significantly upregulated after TNF-α induction relative to that in control RA-FLS, and this increment could be obviously reduced by the knockdown of circHIPK3 (Figure 1B). Similar changes of VEGF mRNA and protein expression in co-cultured RA-FLS and the VEGF secretion in the supernatants of RA-FLS and HDMECs co-culture were observed, which were in accordance with the expression level of circHIPK3 (Figures 1C–E). Next, we asked if the changed expression of VEGF in human RA-FLS co-culture’s supernatant would have angiogenic activity, and the co-culture’s supernatants were transferred to HDMECs. The transwell assay and tube formation test demonstrated significant upregulation in the migration and capillary-like structure formation of HDMECs, respectively, under the treatment of supernatants from RA-FLS and HDMEC co-culture after TNF-α induction when compared with those from co-culture without TNF-α induction; and the migration and tube formation were significantly decreased in the circHIPK3 knockdown RA-FLS co-culture group whether TNF-α induction was performed or not (Figures 1F,G,I). Interestingly, mouse aortic rings ex vivo showed similar changing trends with the results of the transwell assay and tube formation test (Figures 1H,I). These data suggest that knockdown of circHIPK3 impaired the RA-FLS induced synovial angiogenesis via regulating the expression of VEGF.

Accumulated studies have reported that the circRNA may serve as a competing endogenous RNA (ceRNA) or a molecular sponge to interact with miRNA (Chen, 2020). To further ascertain the characteristics of circHIPK3, we scanned circbase (http://www.circbase.org/), and the spliced sequence of circHIPK3 (hsa_circ_0000284) is shown in Supplementary Figure S3. Considering the miRNA sponge function of circHIPK3, we predicted five miRNAs (miR-149-5p, miR-515-5p, miR-637, miR-619, and miR-215) that interacted with circHIPK3 based on bioinformatics methods and RNA pull-down analysis using a circHIPK3 probe. The pull-down efficiency (Supplementary Figure S4A) and expressions of the five miRNAs detected with qRT–PCR were shown (Supplementary Figure S4B). Pull-down with the circHIPK3-specific probes which were oligonucleotides complementary to the back-splice junction of circHIPK3, but not the control probe, yielded a 11-fold enrichment of miR-149-5p in the pulled down sediments (Figure 2A). As to the five miRNAs, at least one miRNA-binding site of circHIPK3 defined by Arraystar proprietary algorithms was predicted with the help of the CircInteractome database (Supplementary Figure S4C). We then constructed a luciferase reporter plasmid containing the full length of circHIPK3. A luciferase assay that was performed to detect the binding of these five miRNAs showed that only miR-149-5p had strongly reduced the luciferase activity more than 50%, compared with the control (Supplementary Figure S4D, Figure2B). The efficiency of circHIPK3 knockdown/overexpression is shown in Supplementary Figure S2. The qRT–PCR analysis showed that circHIPK3 overexpression or knockdown could further reduce or increase the miRNA expression relative to levels in control RA-FLS, but miR-149-5p was the most significant one (Supplementary Figure S4E, Figure 2C). Moreover, experiments showed that the level of circHIPK3 could be significantly reduced/enhanced by miR-149-5p mimics/inhibitor compared with each control group (Figure 2D), indicating that circHIPK3 could bind miR-149-5p to regulate its level. The potential targets of miR-149-5p were investigated with the help of bioinformatics databases such as Targetscan and Starbase. A putative binding site was found in the 3’ UTR of the FOXO1 mRNA, which was a completely complementary sequence of the seed region of miR-149-5p (Supplementary Figure S4C). To confirm whether the FOXO1 could be regulated by miR-149-5p, we constructed luciferase reporters containing wild-type and mutated putative binding sites of FOXO1 transcripts. Luciferase reporter assays showed that the luciferase activity of the FOXO1 wild-type reporter was significantly reduced when transfected with miR-149-5p mimics, compared with the control reporter or mutated luciferase reporter (Figure 2E). The wild-type and mutated putative binding sites of circHIPK3 and FOXO1 with miR-149-5p are, respectively, shown in Supplementary Figure S5. Furthermore, we applied the biotinylated RNA pull-down assay to confirm the results. It was shown that the FOXO1 mRNA captured by biotin-miR-149-5p was found to be prominently enriched compared with the control group (Figure 2F).

Analyses showed that mRNA and protein levels of FOXO1 were significantly decreased by miR-149-5p mimics and increased by miR-149-5p inhibitor (Figures 3A,B). Moreover, the significant downregulation of FOXO1 expression repressed by miR-149-5p mimics could be significantly rescued by circHIPK3 overexpression; in other words, the significant upregulation of FOXO1 expression induced by circHIPK3 overexpression could be evidently inhibited by miR-149-5p mimics (Figures 3C,D). These results suggest that the circHIPK3 could regulate the FOXO1 expression via miR-149-5p.

The relationship between VEGF and circHIPK3/miR-149-5p/FOXO1 axis was further explored. We first explored the effect of FOXO1 on VEGF expression. Both overexpression and silencing of FOXO1 were constructed in RA-FLS, and the stable transfection efficiency of FOXO1 is, respectively, shown in Supplementary Figure S6. As demonstrated in Figures 3E,F, the overexpression of FOXO1 enhanced VEGF expression, while the silencing of FOXO1 presented a repressed effect on the VEGF level. Interestingly, the upregulation of VEGF expression induced by FOXO1 overexpression can be inhibited by miR-149-5p mimics, which could also significantly reduce the VEGF expression by itself (Figures 3G,H). Further experiments confirmed that the expression of VEGF tended to be consistent with the changing trend of FOXO1 expression (Figures 3C,D). As a matter of fact, these findings indicated that FOXO1 mediated the promotion of VEGF expression in RA-FLS, suggesting that VEGF is the downstream of circHIPK3/miR-149-5p/FOXO1 axis, constituting a functional module.

In our previous study, we found that ATO could inhibit VEGF expression in RA-FLS. Therefore, we wondered whether ATO could play an antiangiogenic role via the inhibition of circHIPK3. The results demonstrated that the level of circHIPK3 was decreased by ATO after RA-FLS was exposed to 2.0 µM ATO for 48 h, and the downregulation of circHIPK3 was more significant under TNF-α induction (Figure 4A). To further verify the intervention of the circHIPK3/miR-149-5p/FOXO1/VEGF functional module in the effect of ATO’s antiangiogenic effect, at the same time of TNF-α induction, we performed experiments of FOXO1 overexpression, miR-149-5p inhibition, and circHIPK3 overexpression, and the simultaneous treatment of ATO, followed by the detection of FOXO1 and VEGF expression. Exposure of RA-FLS to ATO induced decrement of FOXO1/VEGF mRNA and protein levels, and intriguingly, this decrement could be rescued partially by FOXO1 overexpression (Figures 4B,C), miR-149-5p inhibitor (Figures 4D,E), or circHIPK3 overexpression (Figures 4F,G). Apart from that, FOXO1/VEGF levels downregulated by ATO, further partially rescued by circHIPK3 overexpression, however significantly reduced again further upon miR-149-5p mimics addition (Figures 4F,G). These results verified that ATO could exhibit the antiangiogenic effect through regulating circHIPK3/miR-149-5p/FOXO1-mediated VEGF expression.

Our previous study confirmed that ATO has a significant antiangiogenic effect on the CIA synovium via its suppression on the VEGF angiogenic functional module (Zhang et al., 2017). To check if the inhibitory effect of ATO on the VEGF module is partially due to the effect of ATO on circHIPK3, and to verify the anti-circHIPK3/miR-149-5p/FOXO1/VEGF effect of ATO could be replicated in vivo, the collagen II–induced arthritis (CIA) model was established, and AAV-sh-circHIPK3 or AAV-sh-CMV-EGFP viral vectors as a control were intra-articularly administered to CIA mice on the seventh day after first (primary) immunization. AAV effectively transduced into synovial tissue and displayed a local distribution pattern in vivo (Supplementary Figure S7A). We assessed the severity of arthritis from day 24 after primary immunization according to a standard arthritic score system (Zhang et al., 2017). CIA developed rapidly in paws after mice were immunized with collagen II, and showed evident clinical symptoms sustainably (Figures 5A,C). AAV-sh-circHIPK3 significantly alleviated the severity of arthritis and reduced arthritic scores compared with the CIA control group (Figures 5A,C), whereas the control AAV-shRNA treatment had no significant effect on CIA arthritis (Supplementary Figure S7B). Then, the curative effect of ATO treatment plus AAV-sh-circHIPK3 injection was evaluated. ATO treatment at a dose of 2.0 mg/kg/day per mouse lasted from day 26 to day 39 after the first immunization. As shown in Figures 5A,C, ATO exhibited continuously significant suppression from day 26 to day 39. Intriguingly, the effect of ATO treatment in the CIA mice could be enhanced by AAV-sh-circHIPK3 injection (Figures 5A,C), while the control AAV-shRNA injection had no significant effect on the ATO’s curative effect (Supplementary Figure S7B). Process diagrams are demonstrated in Figure 5C. The evaluation of ATO and AAV-sh-circHIPK3 injection toxicity were performed as described before (Zhang et al., 2017), and no side effects, such as diarrhea, behavioral and hair abnormalities, or abnormal death, were observed during the study. Body-weight loss was also recorded to evaluate the curative effects and the toxicity of ATO and AAV-sh-circ-HIPK3 injection. PBS-treated CIA mice showed a significant decrease in weight compared with the normal mice from day 24 to day 39. ATO as well as the combination therapy of ATO and AAV-sh-circ-HIPK3 injection significantly increased body weight compared with the CIA control group. Furthermore, there was no significant difference of body weight between the combination therapy group and normal mice group (Supplementary Figure S7C).

Consistent with the trend toward clinical symptoms, a three-dimensional reconstruction of micro-CT images demonstrated the destruction of bone and cartilage in the paws and knees from CIA mice, and AAV-sh-circHIPK3 intra-articular injection or ATO treatment significantly rescued the destructed joint structure. And joint destruction in the combination group of AAV-sh-circHIPK3 and ATO was more significantly inhibited than AAV-sh-circHIPK3 or ATO therapy alone (Figure 5B, Figure 6A). The mean CT values of the paw and knee joints are evaluated and shown, respectively, in Supplementary Figure S7D–E.

The severity of arthritis in the CIA mice was further evaluated via histological analysis (hematoxylin and eosin stain, i.e., H&E staining) of knee joint sections. Synovial hyperplasia, pannus formation, and massive joint destruction with the erosion of cartilage were observed in the knee joints of the CIA mice, whereas no synovial hyperplasia or pannus formation was found in normal mice. There was significant decreased inflammation, and cartilage and bone destruction in mice after the injection of AAV-sh-circHIPK3 or ATO. More interestingly, bone destruction and inflammatory cell infiltration were significantly improved in the combinatory treatment group of AAV-sh-circHIPK3 and ATO compared with ATO therapy alone (Figures 6B,D).

qRT–PCR were performed to evaluate the expression levels of circHIPK3 and miR-149-5p in the synovium of different groups (Supplementary Figure S8). Synovial circHIPK3/mi-149-5p levels in CIA mice treated with PBS were significantly higher/lower than those in normal mice. Moreover, circHIPK3/mi-149-5p levels in CIA mice could be significantly decreased/increased by AAV-sh-circHIPK3, ATO at a dose of 2.0 mg/kg/day, or AAV-sh-circHIPK3+ATO combination treatment. Most notably, circHIPK3/mi-149-5p levels in CIA mice treated with AAV-sh-circHIPK3+ATO combination therapy were significantly lower/higher than those in CIA mice treated with ATO single therapy.

Immunohistochemical analysis was performed, and the average integrated optical density (IOD) values were calculated to evaluate the effect of AAV-sh-circHIPK3 on the FOXO1 (Figure 6C) and VEGF (Figure 6C) expression in the synovial tissue of CIA mice. Normal mice showed weak immunostaining of FOXO1 and VEGF, while CIA mice under PBS treatment had significantly increased the staining intensity for FOXO1 and VEGF in knee joint synovium, and furthermore, the staining intensity for FOXO1 and VEGF in the CIA mice were significantly decreased by AAV-sh-circHIPK3–injected treatment (Figures 6C,E). On the other hand, the immunostaining intensity for FOXO1 and VEGF were significantly weaker under ATO treatment at a dosage of 2.0 mg/kg/day than those of CIA mice treated with PBS (Figures 6C,E). Interestingly, the immunostaining intensity for FOXO1 and VEGF in the synovium from CIA mice treated with injection of AAV-sh-circHIPK3 combined with ATO were significantly weaker than those of CIA mice treated with ATO alone (Figures 6C,E). In other words, the AAV-sh-circHIPK3 injection enhanced ATO-induced decrement of FOXO1 and VEGF levels in synovial tissue of CIA mice. Together, our results indicated the formation of the circHIPK3/mi-149-5p/FOXO1/VEGF functional module in the synovial tissue of CIA mice, and circHIPK3 suppression is partly involved in the inhibitory effect of ATO on VEGF expression.

RA fibroblast–like synoviocytes (RA-FLS) were purchased from Cell Applications (San Diego, CA, United States), and human dermal microvascular endothelial cells (HDMECs) were obtained from ScienCell Research Laboratories (6076 Corte Del Cedro, Carlsbad, CA, United States). Cells at passages three to six were used for experiments. The co-culture system of RA-FLS and HDMECs was established as described previously (Zhang et al., 2017). Briefly, RA-FLS (4×10⁵ cells/ml) and HDMECs (ranging from 2×10⁵ cells/ml to 3×10⁵ cells/ml) were seeded in the lower and upper chambers of a transwell apparatus (Costar), respectively, and incubated in 2.0 μM arsenic trioxide (ATO; As₂O₃; Yitaida Pharmaceutical Factory, Harbin, Heilongjiang, China) with or without TNF-α (50 ng/ml, Peprotech, Rocky Hill, United States). After being cultured in 1% FBS DMEM for 48 h, the supernatants were harvested. Fresh supernatants were taken out for the subsequent transwell assays, tube formation tests, and ex vivo aortic ring angiogenesis assays. Remaining supernatants were frozen at -20°C until analysis by commercial enzyme-linked immunosorbent assay (ELISA) kit for VEGF (DVE00, R&D Systems). The mRNA and protein expression in RA-FLS co-cultured were, respectively, determined by quantitative real-time PCR (qRT-PCR) and Western blot analysis.

Transwell and mouse aortic ring angiogenesis assays were performed as described previously (Zhang et al., 2017). Briefly, as to transwell assays, the lower chamber was filled with 600 µl of supernatants from the cocultures. HDMECs (3×10⁴ cells) were added to each upper chamber and incubated for 6 h. Cells that migrated to the lower phase were stained and quantified. Aortic rings from thoracic aortas of 8-week-old C57BL/6 mice (SPF, Laboratory Animal Center of the Second Affiliated Hospital of Harbin Medical University) were incubated in 1% FBS EGM-2 on the basis of Matrigel. After verifying sprouts from the aortic rings, the EGM-2 was replaced with 500 μl of fresh supernatants from cocultures, and the aortic rings were further incubated for 3 days. Microvessel outgrowths were examined and analyzed using ImageJ software. In tube formation assays, a µ-Slide (ibidi, GmbH, Munich, German) was used to investigate angiogenesis. 10 μl Matrigel (Corning) was plated per inner well, and 70 μl of cell suspension containing 1×10⁴ cells was added to each upper well for a further 6-h incubation. All images were obtained using microscopy (LEICA DMi8).

Total cellular and tissue RNA was isolated by Trizol (Invitrogen) (Chu and Zhang, 2018). As to RNA preparation for circRNA determination, total RNA (5 µg) was first treated with RNase R (Geneseed Biotech, Guangzhou, China), and GAPDH in the same sample without RNase R treatment was used as an internal control. For the detection of circRNA and mRNA, RNA and PrimeScript RT Master Mix (Takara Biotechnology, Dalian, China) were used in reverse transcription. The qRT–PCR amplification was performed using the iTaq Universal SYBR Green supermix (Bio-Rad). For cellular and tissue microRNA, total RNA was extracted by using the QIAzol Lysis Reagent. Reverse transcription and qRT-PCR were performed with riboSCRIPTTM Reverse Transcription Kit (RiboBio, Guangzhou, China) and Applied Biosystems TaqMan MicroRNA Assay Kit. Relative circRNA, mRNA, or miRNA expression was normalized to GAPDH or U6 snRNA levels using the 2^−ΔΔCt method, respectively. The sequence for each primer is listed in Supplementary Table S1.

FISH assays were performed with Fluorescent In situ Hybridization Kits (Ribobio, Guangzhou, China). Briefly, RA-FLS grown on coverslips were fixed with 4% paraformaldehyde at room temperature for 10 min and treated with 0.5% Triton X-100 at 4°C for 5 min. The samples were treated with a pre-hybridization buffer at 37°C for 30 min and then in hybridization buffer at 37°C for 12–16 h with a CY3-labeled circRNA probe (Ribobio, Guangzhou, China, Supplementary Table S1) in a humid and dark environment. The samples were mounted with a fluorescence mounting medium and imaged by the use of a laser confocal microscope (TCS SP8 X, LEICA).

The miRNA target of circHIPK3 was predicted by bioinformatics databases, namely, Targetscan (http://www.targetscan.org/), CircInteractome (https://circinteractome.nia.nih.gov/), and Rnahybrid (https://bibiserv.cebitec.uni-bielefeld.de/rnahybrid/). The mRNA target of miR-149-5p was predicted by bioinformatics databases such as Targetscan (http://www. targetscan. org/) and Starbase (http://starbase.sysu.edu. cn/).

Adenovirus vector encoding circHIPK3 (Ad-circHIPK3), FOXO1 (Ad-FOXO1), and their GFP control (Ad-Vector) were entrusted to Hanbio Biotechnology (Shanghai, China). In brief, circ HIPK3 and FOXO1 fragments were amplified and ligated with the linearized vector, and the amplified sequence was detected by Sanger sequencing to verify the consistency with circ HIPK3 or FOXO1. Packaging plasmids and viral vectors were co-transfected into HEK-293 T cells using Lipofectamine 3,000 transfection reagent (ThermoFisher). 48 h after transfection, the culture medium was centrifuged (SolarBio). Finally, the culture medium mixture with polybrene was added to RA-FLS.

SiRNAs that target human circHIPK3 (si-circHIPK3) and FOXO1 (si-FOXO1), the vectors that upregulate or downregulate relative miR-149-5p expressions (miR-149-5p mimic and miR-149-5p inhibitor), were designed and constructed by RiboBio (Guangzhou, China). Lipofectamine RNAiMAX (ThermoFisher) was further applied according to the manufacturer’s instructions. Relative sequences are shown in Supplementary Table S1.

As to biotin-coupled miRNA capture, the 3′end biotinylated miR-149-5p mimics or control RNA (Ribio, Guangzhou, China) were transfected into RA-FLS at a final concentration of 50 nM for 48 h before harvest. Then the lysis buffer [0.02 M Tris-HCl (pH 7.5), 0.1 M NaCl, 0.5 mM DTT, 0.5% NP-40, 0.6 U/µl RNase inhibitor (Promega), and complete protease inhibitor cocktail (Sigma, United States) were added into the cell pellets. Then the cell lysates with biotinylated probes were incubated with streptavidin-conjugated magnetic beads (Invitrogen, United States), the beads were washed for five times, and TRIzol reagent (Invitrogen, Thermo, United States) was added to the washed beads for RNA extraction. Then the abundance of circHIPK3 and FOXO1 mRNA in bound fraction pulled down were evaluated by qRT–PCR analysis. The biotinylated circHIPK3 probe was designed and synthesized by RiboBIO (Guangzhou, China). And the RNA pull-down assay was performed using an RNA pull-down kit (BersinBio, Guangzhou, China) according to the manufacturer’s instructions similar to the aforementioned biotin-coupled miRNA capture.

The expression plasmid of circHIPK3 was created by the placement of the human entire circHIPK3 sequence into pcDNA3.1 circRNA Mini Vector (Addgene). RA-FLS were transfected with an miR-149-5p mimic (RiboBio, China) or control mimic combined with luciferase reporter or an empty vector, and cells were also transfected with a pcDNA3.1-circHIPK3 and its mutant type using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s protocol.

The reporter plasmid pGL3-FOXO1 containing the predicted miR-149-5p targeting regions was designed by Ribobio (China). A part of the wild-type and mutated 3′-UTR of FOXO1 was cloned immediately to the downstream of the firefly luciferase reporter. MiR-149-5p mimics and its negative control were transfected into RA-FLS at a final concentration of 100 nM using a riboFECTTM CP Transfection Kit (RiboBio, China) according to the manufacturer’s instructions. The luciferase activity was measured using Dual–Glo luciferase Assay System (Promega) with Cytation five Cell Imaging Multi-Mode Reader (BioTek) 24 h after transfection.

Protein extraction and Western blot analysis were performed as described before (Chu and Zhang, 2018). The antibodies against specific proteins including VEGF (Abcam, ab214424) and FOXO1 (Cell Signaling Technology, 2880) were then added. GAPDH was utilized for normalization.

Experimental procedures were carried out in accordance with the Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of the First Affiliated Hospital of Harbin Medical University. The adeno-associated virus 5 (AAV5) of sh-circHIPK3 was constructed and packaged by HanBio (Shanghai, China), and the AAV5 containing shRNA control was provided at the same time. The collagen-induced arthritis (CIA) mouse model was established using 8-week-old male DBA/1J mice (19 ± 2 g in weight, SLAC, Shanghai, China) as described previously (Zhang et al., 2017). On day 7 after the first immunization, mice were randomly assigned to seven different groups (n = 5 per group): normal control group (mice without immunization and injected with PBS), CIA control group (CIA mice treated with PBS), AAV-sh-circHIPK3 treatment group, control shRNA group, ATO-treated group, AAV-sh-circHIPK3 + ATO group, and control shRNA + ATO group. In the AAV treatment groups, a total of 10 μl (approximately 3.0 × 10¹² vg/ml) of AAV expressing sh-circHIPK3 or control shRNA were injected into the knee joint. To verify the successful AAV infection in the joint cavity, we used the joint frozen section to detect immunofluorescence of AAV5-CMV-EGFP on day 28 after virus injection. The mice were given intraperitoneal injections of ATO at a dose of 2.0 mg/kg/day from day 26 to day 39. The mice were euthanized on day 40, and a part of the synovial tissues were isolated from knee joints of mice for the subsequent tissue qRT–PCR test. A time line for ATO and AAV treatment is provided in Figure 5C. The body weight of each mouse was recorded every 3 days from day 24.

Clinical arthritis was assessed daily, and the arthritic score was evaluated every 3 days from day 24 after primary immunization to the end of the experiment period according to a scoring system (Zhang et al., 2017). Mice were scanned and reconstructed into a three-dimensional (3D) structure via micro-CT imaging (Quantum GX, Perkin Elmer, Waltham, United States) to evaluate joint bone damages on day 40 after the initial collagen injection. The mean CT values of the hind paws were calculated with Caliper Analyze software (Analyze Direct, Kansas, United States) to assess bone loss (Wang et al., 2019).

After being sacrificed, the knee joints of mice were collected and stained with HE staining. To quantitatively evaluate the severity of arthritis, a scoring system was employed referring to the reported protocol (Zhang et al., 2017). Histological changes were examined by microscopic evaluation and scored in a blinded manner by two independent observers.

The immunohistochemical analysis was performed as previously described with some modifications (Zhang et al., 2017). Knee joint sections on slides were incubated with anti-FOXO1 and anti-VEGF antibodies (Wanleibio, Shenyang, China). Subsequently, the sections were stained using the polymer HRP detection system (PV9001, ZSGB-BIO, Beijing, China) and were visualized with DAB Peroxidase Substrate Kit (ZLI-9017, ZSGB-BIO, Beijing, China). Following immunostaining, the synovium area in the joint of each section was evaluated under a microscope (LEICA DMi8) in three randomly selected areas at a magnification of 100×. Image-Pro Plus 6 (Media Cybernetics, Inc.) was used to analyze the average integrated optical density (IOD) according to a previously described protocol (Wang et al., 2019).

Data were presented as means ± S.E.M. Comparisons of means between two groups were performed via Student’s t-test. Comparisons of means among multiple groups were accomplished by one-way analysis of variance (ANOVA), and a multiple range least significant difference (LSD) was used for intergroup comparisons. p values < 0.05 were considered to be statistically significant. All statistical analyses were performed with SPSS 17.0 software.