A total of 76 pairs of tumor tissues and matched adjacent normal tissues were collected from BCa patients who underwent surgery. The clinical data of 76 bladder cancer patients are shown in ( Table 1 , Supplementary Table S3 ). This study was approved by the Regional Committee for Medical Research Ethics and the Human Ethics Committee of Peking University Shenzhen Hospital. All Bca patients approved this study.

All cell lines, including SV-HUC1 and BCa cell lines, were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). BCa cell lines were cultured in DMEM (Invitrogen, Carlsbad, CA, USA) containing 10% fetal bovine serum (FBS) and 1% antibiotics. SV-HUC1 cells were grown in F12K medium mixed with 10% FBS and 1% antibiotics. All cell lines were cultured in an incubator at a temperature of 37°C with 5% CO₂.

TRIzol reagent (Invitrogen, Carlsbad, CA, USA) was used to extract total RNA. Quantitative real-time PCR (qRT-PCR) was performed according to a SYBR Green PCR kit (Takara) and conducted on a Roche LightCycler^® 480II PCR instrument (Basel, Switzerland). GAPDH or U6 nuclear RNA was applied as an internal control. The sequences of the primers used in this research are listed in Supplementary Table S1 .

Two short hairpin RNAs (shRNAs) targeting circTAF4B (shcirc-1, shcirc-2) and the corresponding control (shCtrl) were devised by GenePharma and cloned into the pGPU6/GFP/Neo vector. The sequences of oligonucleotides are listed in Supplementary Table S3 . miR-1298-5p mimic and its corresponding control (miR-NC) and miR-1298-5p inhibitor and its corresponding control (NC inhibitor) were purchased from RiboBio. The overexpression vector for TGFA (pcDNA3.1-TGFA) and the corresponding control (pcDNA3.1-NC) were synthesized by GenePharma. Lipofectamine 3000 reagent (Invitrogen) was used for all transfections according to the instructions. For stable transfection, SW780 cells were infected with lentivirus with shcirc-1, which was previously verified to be effective, or its corresponding control (shCtrl) and then selected with 3 μg/mL puromycin for two weeks.

Cell Counting Kit-8 (CCK-8) (Beyotime Institute of Biotechnology) experiments and colony formation experiments were used to evaluate cell proliferation ability. For the CCK-8 experiment, the transfected BCa cells were cultured in a 96-well plate at a concentration of 1000 cells per well. The absorbance value of BCa cells was measured at 450 nm using a microplate analyzer (Bio-Rad, Hercules, CA, USA) at 0, 24, 48, 72, and 96 hours. For the colony formation assay, BCa cells were cultured in 6-well plates at a concentration of 1000 cells per well for 15 days. The colonies were stained with crystal violet and rinsed with glacial acetic acid. Finally, the absorbance value was measured at 595 nm by using a microplate reader.

A wound healing assay was used to measure cell migration ability. The transfected Bca cells were grown in 6-well plates. A yellow pipette tip was used to draw scars until cells grew to 90-95% abundance. After 24 hours, we used a digital microscope at 10-fold magnification to capture images of the migrated cells.

Transwell assays were used to evaluate the migration and invasion ability of BCa cells. Transfected cells were mixed in serum-free medium and inoculated in Transwell inserts (Corning) coated with matrix gel (BD Biosciences) or left uncoated at a density of 2-3×10⁴ cells/well. Medium containing 10% serum was added to the lower chamber. After incubation for 24-48 hours, cells were stained with 0.1% crystal violet and washed with PBS. Observation and photography were conducted under a microscope at 20x.

Protein was extracted with RIPA reagent (Beyotime, Beijing, China), separated by 10% SDS-PAGE, and transferred to PVDF membranes. The PVDF membranes were blocked with 5% milk for one hour and incubated with primary antibody at 4°C for over 12 hours and secondary antibody at room temperature for 1 hour. Quantity One software (Bio-Rad) was used for luminescent detection and imaging. N-cadherin, vimentin, E-cadherin, and GAPDH antibodies were provided by Cell Signaling Technology (Danvers, MA, USA), and TGFα antibody was obtained from Abcam (USA).

Nuclear/cytoplasmic fractionation was performed using a PARIS Kit (Life Technologies, MA). After the nuclei and cytoplasm were separated according to the instructions, the expression of the target gene was detected by qRT-PCR, and GAPDH and U6 were used as controls.

SW780 and T24 cells were cotransfected with pcDNA3.1-circTAF4B and biotin-labeled circTAF4B. The pull-down proteins were treated with RNase-free DNase I and an RNeasy Mini Kit (QIAGEN, Germany). microRNAs were extracted and quantitatively detected by qRT-PCR. The primers applied in this experiment and their corresponding sequences are listed in Supplementary Table S1 .

A luciferase reporter plasmid was designed and synthesized by GeneCopoeia. The sequences of wild-type (WT) circTAF4B or mutated (Mut) circTAF4B were cloned into the MT06 vector. The sequences of wild-type (WT) TGFA and mutated (Mut) TGFA were cloned into the MT07 vector. Luciferase reporter plasmids and miR-1298-5p were cotransfected into BCa cells. Finally, the luciferase activity of BCa cells was measured by using a Luciferase Reporter Kit (Promega).

Four-week-old male BALB/C nude mice were used for the tumor xenotransplantation assay in this study. Twelve mice were randomly divided into two groups. SW780 cells expressing shcircTAF4B (5×107) were injected subcutaneously into the backs of mice. The volume of tumor xenografts was measured weekly. Six weeks later, the mice were sacrificed, and the tumors were collected and weighed. Total RNA and total protein were extracted from tumors to detect the expression of target genes.

All data from independent replications were analyzed using SPSS 22.0 (SPSS), and the results are expressed as the mean ± standard deviation (SD). The difference between groups was analyzed by Student’s t test. A P value < 0.05 or a P value <0.01 was considered statistically significant.

Through a bioinformatics website (circBank), we determined that circTAF4B (circ_0047322) is 484 nt in length and formed by the reverse splicing of TAF4B mRNA, which is located at CHR18: 23895192-23915195. The splicing site of circTAF4B is shown in Figure 1A . Sanger sequencing was applied to verify the junction site ( Figure 1B ). The RNase R enzyme digestion assay demonstrated that circTAF4B was resistant to RNase R instead of TAF4B mRNA ( Figure 1C ). Next, we discovered that circTAF4B was significantly augmented in 65.7% (50/76) of BCa tissues ( Figure 1D ). The expression level of circTAF4B in tumor tissues was also significantly increased compared to that in matched adjacent normal tissues ( Figure 1E ). In addition, increased expression of circTAF4B was positively correlated with higher histological grade and lymphatic metastasis ( Table 1 , Figures 1F, G ). However, high expression of circTAF4B was associated with a lower overall survival rate and disease-free survival rate of bladder cancer patients ( Figures 1H, I ), suggesting that circTAF4B may be a cause of the poor prognosis of BCa patients.

Compared to the urinary bladder cell line HT1197, the circTAF4B expression level was obviously increased in BCa cells, especially SW780 and T24 cells ( Figure 2A ). Hence, SW780 and T24 cells were selected for further experiments. To explore the role of circTAF4B in the development of BCa, we synthesized an shRNA targeting circTAF4B to suppress circTAF4B expression ( Figure 2B ). However, knockdown of circTAF4B could not modulate TAF4B mRNA expression, suggesting that circTAF4B could not regulate the transcription of TAF4B ( Figure 2C ). CCK-8 and colony formation assays showed that knockdown of circTAF4B significantly inhibited the proliferation of BCa cells ( Figures 2D, E ). Wound healing and Transwell migration assays demonstrated that inhibition of circTAF4B significantly reduced the migration ability of BCa cells ( Figures 3A, B , 6B ). Transwell invasion assays showed that downregulation of circTAF4B could reduce the invasion ability of BCa cells ( Figure 3C ). Furthermore, knockdown of circTAF4B significantly inhibited the expression of N-cadherin, Snail and Vimentin and increased E-cadherin expression ( Figure 3D ), indicating that circTAF4B promoted the EMT process.

To further investigate the mechanism of circTAF4B in BCa tumorigenesis, we carried out a nuclear/cytoplasmic fractionation assay and found that circTAF4B was mainly distributed in the cytoplasm of BC cells ( Figure 4A ), indicating that circTAF4B may act as a “miRNA sponge” to capture tumor suppressor miRNAs. To verify this hypothesis, we synthesized a circTAF4B-specific probe and found that it could indeed capture circTAF4B upon increased expression of circTAF4B ( Figure 4B ).

RNA pull-down assays showed that miR-1298-5p was significantly enriched in the circTAF4B probe in both BCa cell lines ( Figure 4C ). Furthermore, bioinformatics analysis predicted that circTAF4B was complementary to the seed sequence of miR-1298-5p. Then, luciferase reporter plasmids containing circTAF4B wild-type (WT) and mutant (Mut) were cotransfected with miR-1298-5p. A luciferase reporter assay showed that luciferase activity decreased significantly when the circTAF4B WT plasmid was cotransfected with miR-1298-5p ( Figures 4D, E ). Nevertheless, knockdown or overexpression of circTAF4B did not change the expression of miR-1298-5p ( Figures 4F, G ).

To investigate the target genes of miR-1298-5p, we used miRNA databases (miRtarbase, starBase and TargetScan) to forecast the potential target genes of miR-1298-5p and chose CBX6, FXR1 and TGFA for further experiments ( Figure 5A ). Using the TCGA database, we found that the expression level of miR-1298-5p (MIMAT0005800) in BCa tissues was lower than that in normal tissues ( Supplementary Figure S1A ), while the expression level of TGFA in BCa was significantly higher than that in normal tissues ( Supplementary Figure S1B ). miR-1298-5p expression was negatively correlated with TGFA expression in BCa tissues from the TCGA database ( Figure 5B ). The expression level of TGFA was negatively correlated with miR-1298-5p expression in 76 BCa tissues collected ( Figure 5C ). Moreover, circTAF4B expression was positively correlated with TGFA expression in 76 Bca tissues ( Figure 5D ). Further experiments confirmed that augmented expression of miR-1298-5p suppressed the mRNA and protein expression of TGFA ( Figures 5E, F ).

To further verify the core binding sites between miR-1298-5p and TGFA, we mutated the TGFA 3’UTR (3554-3560), which was predicted to be complementary to miR-1298-5p, and constructed a dual-luciferase reporter plasmid ( Figure 5G ).

Dual-luciferase reporter plasmids containing TGFA wild-type (WT) or mutant (Mut) were cotransfected with miR-1298-5p mimics. A dual-luciferase assay confirmed that miR-1298-5p remarkably reduced the luciferase activity of TGFA wild type (WT) ( Figure 5H ). Finally, qRT-PCR and western blotting experiments proved that inhibition of TGFA expression by circTAF4B knockdown was compensated for by inhibition of miR-1298-5p ( Figures 5I, J ). These results revealed that circTAF4B could upregulate TGFA expression by sponging miR-1298-5p.

To investigate whether circTAF4B promotes proliferation and metastasis by regulating TGFA expression, we conducted a rescue assay between circTAF4B and TGFA. TGFA expression in BCa cells was increased dramatically after transfection with pcDNA3.1-circTAF4B ( Figures 6A , B ). CCK-8 and colony formation experiments showed that forced expression of TGFA could obviously reverse the cell proliferation suppression induced by circTAF4B inhibition ( Figures 6B, C ). Wound healing and Transwell migration assays showed that knockdown of circTAF4B restrained BCa cell migration, while cotransfection of pcDNA3.1-TGFA impaired this effect ( Figures 6D, E ). Transwell invasion assays indicated that the cell invasion ability inhibited by shcircTAF4B was reversed significantly by enhancing TGFA expression ( Figure 6F ). The above results demonstrated that overexpression of TGFA reversed the suppression of BCa cell growth and metastasis induced by circTAF4B knockdown.

To confirm the role of circTAF4B in Bca progression in vivo, we carried out tumor xenograft experiments. SW780 cells stably expressing shcricTAF4B were subcutaneously injected into the backs of nude mice. The nude mice were sacrificed, and the transplanted tumors were collected after 6 weeks ( Figures 7A, B ). As shown in Figures 7C, D the tumor volume and weight of tumors derived from the circTAF4B downregulated group were significantly lower than those in the negative control group. Furthermore, knockdown of circTAF4B inhibited the expression of TGFA, N-cadherin, and vimentin while upregulating the expression of E-cadherin in vivo ( Figure 7E ). Western blotting assays confirmed that downregulation of circTAF4B inhibited the protein expression levels of N-cadherin, vimentin and TGFα while increasing the protein expression level of E-cadherin ( Figure 7F ). In summary, circTAF4B positively regulates TGFA expression and promotes tumor growth and EMT in vivo. The mechanistic diagram of the circTAF4B/miR-1298-5p/TGFA axis is shown in Figure 8 .