IGF2BP2-induced circRUNX1 facilitates the growth and metastasis of esophageal squamous cell carcinoma through miR-449b-5p/FOXP3 axis

Fifty-four pairs of ESCC tissues and paracancerous tissues were collected from patients without preoperative chemotherapy or radiotherapy from the First Affiliated Hospital of Zhengzhou University between 2019 and 2021. The tissues were immediately collected after surgery and stored at -80 °C until use. Written informed consents were obtained from each patient in the cohort.

Human normal esophageal epithelial cell line HEEC and human ESCC cell lines (KYSE450, EC109, TE1, KYSE70, KYSE150) and HEK-293T cells were purchased from the Shanghai Institute of life science cell bank center (Shanghai, China). All cells were cultured in 1640 medium (Hyclone, USA) with 10% fetal bovine serum (FBS, VivaCell, China) and 1% penicillin–streptomycin (Solarbio, China). Cell plates were incubated at 37 °C and 5% CO₂ under saturated humidity.

Small interfering RNAs targeting circRUNX1 (si-circRUNX1#1, si-circRUNX1#2, and si-circRUNX1#3) and their relative control (si-NC), miR-449b-5p mimic, miR-449b-5p inhibitor, and miR-NC, FOXP3 small interfering RNA (si-FOXP3) and corresponding control (si-NC) were all purchased from RiboBio (Guangzhou, China). All transfections were finished by using Lipofectamine 3000 (Invitrogen, USA). The full-length circRUNX1 was subcloned into the lentivirus vector (Lv‐circRUNX1) and purchased from GeneChem (Shanghai, China). Stably transfected cell lines were filtered by puromycin (2 μg/mL) for two weeks. RT-PCR was performed to detect the transfection efficiency after 48 h. The sequences of siRNA, shRNA, microRNA mimics, and inhibitors were listed in Table S1.

Total RNA was separated from ESCC tissues and cell lines using RNAiso Plus (Takara, Japan) as per the manufacturer’s instructions. Genomic DNA (gDNA) was extracted from ESCC cell lines using the Blood/Cell/Tissue Genomic DNA Extraction Kit (TIANGEN, Beijing, China). The integrity and purity of extracted total RNA and gDNA were detected by NanoDrop One (Thermo Fisher Scientific, Waltham, USA). The DNA and RNA products were immediately stored at -80 °C until future use.

PrimeScript™ RT reagent Kit (Takara, Japan) was used in reverse transcription of total RNA. After removing the genomic DNA at 42 °C for 2 min by gDNA Eraser, total RNA from ESCC tissues and cells was reverse transcribed into cDNA using a RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, Waltham, USA) through the following procedure: 37 °C for 15 min and 85 °C for 5 s. The product was immediately stored at—20 °C until use. The qRT-PCR was conducted on a QuantStudio 5 Real-Time PCR System (Applied Biosystems, Foster City, USA). The procedure is listed as follows: 95 °C for 30 s, 40 cycles of 95 °C for 5 s, and 60 °C for 1 min. The results were calculated by using the relative quantification 2^−ΔΔCT method normalized by GAPDH. The primers for qRT-PCR were presented in Table S2.

The nuclear and cytoplasmic fractions of RNA were extracted using Cytoplasmic and Nuclear RNA Purification Kit (Norgen Biotek, Japan) according to the manufacturer’s instructions. The extracted RNAs were then used to examine the expression of circRUNX1.

Total RNA (2 μg) extracted from KYSE150 cells was digested with 3 U/μg RNase R (Epicentre, USA) for 15 min at 37℃. The abundance of circRUNX1 and RUNX1 was detected by qRT-PCR.

KYSE150 and TE1 cells were treated with 5 μg/mL actinomycin D (Sigma Aldrich, St. Louis, MO, USA) at 0 h, 4 h, 8 h, 12 h, 24 h, respectively. Then cells were harvested to extract total RNA. The stability of circRUNX1 and RUNX1 mRNA was analyzed using qRT-PCR.

Cy3-labeled specific probes to circRUNX1 and positive control probes (RiboBio, Guangzhou, China) were designed and synthesized to detect the location of circRUNX1 in TE1 cells. In brief, after prehybridization for 30 min at 37 ℃, cell crawling sheets were hybridized with 2.5 μL specific Cy3-labelled circRUNX1 probes and positive control probes (20 μM) at 37 °C overnight, and dyed with DAPI. Slides were photographed with confocal laser scanning microscopy (Zeiss, Jena, Germany).

All cells were lysed with RIPA lysis buffer (Solarbio, China) and protease inhibitors (Beyotime, China). BCA Protein Assay Kit (Biomed, China) was used to adjust proteins to equal concentration. Protein lysates were separated in sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and then transferred to PVDF membranes (Millipore, USA). Subsequently, the membranes were blocked with 5% non-fat milk at room temperature for 2 h and incubated at 4 °C overnight with the primary antibodies against FOXP3 (22228–1-AP, 1:1000, Proteintech, China) or GAPDH (60004–1-Ig, 1:1000, Proteintech, China). ECL chemiluminescent reagent (UE, China) was used for band visualization.

The paraffin sections were baked, dewaxed in xylene, and then soaked in gradient ethanol for hydration. After being washed with PBS, samples were heated in boiling citrate buffer (pH 6.0) for 15 min for antigen retrieval. Then sections were incubated at room temperature with 3% H₂O₂ for 10 min and blocked with 5% normal goat serum at room temperature for 30 min, then incubated with the primary antibodies anti-IGF2BP2 (11601–1-AP, 1:200, Proteintech, China), anti-FOXP3 (22228–1-AP, 1:200, Proteintech, China), anti-Ki67 (ab15580, 1:200, Abcam, USA) at 4 ℃ overnight. After 2–3 rinses in PBS, the sections were incubated with biotinylated anti-IgG secondary antibody and horseradish peroxidase-labeled streptavidin was added to the slides and incubated for 15 min. The final immunoreactivity score of IGF2BP2 was calculated as previously described [19]. For H&E staining, the nucleus and cytoplasm were stained with hematoxylin and eosin, respectively. General histological examination was performed by standard procedures.

Transfected cells were washed twice in pre-cooled PBS, and then lysed in Polysome lysis buffer. Next, the cell lysates were incubated with Pierce protein A/G magnetic beads (Thermo Scientific, USA) conjugated with anti-IGF2BP2 (11601–1-AP, Proteintech, China) or Ago2 (ab186733, Abcam, USA) in rotator at 4 ℃ overnight. After separation and purification, the immunoprecipitated RNAs were further analyzed by qRT‐PCR to measure the expression of circRUNX1 or miR-449b-5p.

CircRUNX1 or FOXP3 fragments with mutant (Mut) or wild-type (WT) miR-449b-5p binding sites were subcloned into the Renilla gene downstream of the psiCHECK2 dual-luciferase reporter vector (Hanbio, China) resulting in the following constructs: circRUNX1-WT, circRUNX1-Mut, FOXP3 3' UTR-WT and FOXP3 3' UTR-Mut. The plasmids were co-transfected into 293 T cells with miR-449b-5p mimics or NC mimics. After 48 h, the Dual-Luciferase Assay Kit (Beyotime, China) was used to measure relative luciferase activity.

For colony formation assays, KYSE150 or TE1 cells were seeded into 6-well plates at 1200 cells per well and incubated at 37 °C for 2 weeks. As for CCK-8 assays, transfected cells were seeded in 96-well plates. 10 μL of cell Counting Kit-8 (Servicebio, Wuhan, China) were added to 96-well at 37 °C for 2 h, then the absorbance was measured at 450 nm by a microplate reader (SpectraMax i3x, Molecular Devices, Shanghai, China). Cell-Light EdU Apollo567 In Vitro Kit (RiboBio, Guangzhou, China) was used in EdU assays and performed as previously described [20]. In brief, KESE150 and TE1 cells were cultured with EdU solution at 37 °C for 4 h. After fixation and penetration, cells were treated with Apollo dyes and DAPI for EdU staining and nuclear staining. Finally, cell images were captured by Inverted Fluorescence Microscope (Olympus, Tokyo, Japan).

For transwell assays, ESCC cells were seeded into the upper chambers with or without Matrigel (Corning, NewYork, USA), then the chambers were put into 1640 medium with 20% fetal bovine serum for 48 h. As for wound-healing assays, 2 × 10⁴ cells were seeded into 12-well plates for 12 h. Then 10 μL pipette tip was used to make a gap, and serum-free medium was added into 12-well plates to culture cells for another 24 h.

Four to five-week-old BALB/c nude mice (Shanghai SLAC Laboratory Animal Co. Ltd.) were used to construct in vivo tumor models. 5 × 10⁶ KYSE150 cells were suspended in 100 μL of PBS and injected subcutaneously into the mice. The size of tumors was measured every 5 days and the volume of tumors was calculated as follows: 0.5 × length × width². After 30 days, the mice were sacrificed and the tumors were isolated. Separated tumors were immediately fixed with 4% paraformaldehyde for subsequent experiments. For in vivo metastasis model, 2 × 10⁶ KYSE150 cells were injected into the tail veins to establish pulmonary metastasis models. These nude mice were monitored weekly and euthanized 60 days later, then pulmonary nodules in each group were counted and fixed.

To explore circRNAs expression profiles in ESCC, circRNA sequencing was performed on cancer tissues and corresponding paracancerous tissues. Sequencing results revealed a total of 581 differentially expressed circRNAs in ESCC tissues (fold change ≥|2.0| and P ≤ 0.05), including 261 up-regulated circRNAs and 320 down-regulated circRNAs. Among the top 8 most high-expressed circRNAs, we found that circRUNX1 (hsa_circ_0002360) was substantially overexpressed in 25 pairs of ESCC tissues (Fig. 1A and Table S3). Then we validated the expression of circRUNX1 in 54 pairs of ESCC tissues and paired normal esophageal tissues by qRT-PCR (Fig. 1B). Taking the median as the boundary, the 54 patients were divided into the high expression group (n = 27) and the low expression group (n = 27). Analysis of clinicopathological features revealed that high circRUNX1 expression was significantly associated with advanced TNM stage and poor differentiation grade (Table S4 and Fig. S1). Based on the expression of circRUNX1 in ESCC tissues, a Receiver Operating Characteristic (ROC) curve analysis was constructed, and the results revealed an Area Under Curve (AUC) of 0.894 (95% CI: 0.829–0.960, P < 0.001), indicating that circRUNX1 could be used to distinguish cancer and paracancerous tissues in ESCC patients (Fig. 1C). Next, the circRUNX1 expression in ESCC cell lines was investigated. As shown in Fig. 1D, circRUNX1 was highly expressed in KYSE150 and TE1 cell lines compared to normal esophageal epithelium cell line, HEEC. As a result, these two cell lines were used for subsequent investigations. As depicted in Fig. 1E, circRUNX1 is derived from exon 5 and exon 6 of Runt-related transcription factor (RUNX1) pre-mRNA by head-to-tail splicing. qRT-PCR revealed that circRUNX1 was more resistant to RNase R digestion than its parental mRNA (Fig. 1F). To verify its ring structure, we designed divergent primers and convergent primers. The agarose gel electrophoresis results showed that circRUNX1 in cDNA could be amplified not only by divergent primers but also by convergent primers. However, no amplification product was found in gDNA amplified with divergent primers (Fig. 1G). After 24 h of actinomycin D treatment, we discovered that the half-life of circRUNX1 was much longer than that of RUNX1 mRNA in KYSE150 and TE1 cell lines (Fig. 1H). To investigate its localization, qRT-PCR was utilized to detect the distribution of circRUNX1 in the nucleus and cytoplasm. The results revealed that a significant portion of circRUNX1 was localized in the cytoplasm (Fig. 1I), and FISH assay further confirmed that the majority of circRUNX1 was located in the cytoplasm of TE1 cells (Fig. 1J). Taken together, our results indicated that increased expression of circRUNX1 correlated with the advanced TNM stage and poor differentiation grade of individuals with ESCC.

To investigate the biological function of circRUNX1 in ESCC cells, circRUNX1 was depleted using three siRNAs that targeted its back-splicing junction sites, and stably overexpressed circRUNX1 cell lines were established utilizing GV-689 lentivirus vector (Fig. 2A and Fig. S2A). Neither siRNAs nor GV-689-circRUNX1 affected linear RUNX1 expression (Fig. 2B and Fig. S2B). CCK-8 assays revealed that ESCC cell viability was inhibited in KYSE150 and TE1 cells after circRUNX1 depletion but significantly increased after circRUNX1 overexpression (Fig. 2D and Fig. S2C). Moreover, colony formation and EdU assays demonstrated that downregulation of circRUNX1 suppressed cell proliferation and limited the colony formation ability of KYSE150 and TE1, while ectopic expression of circRUNX1 showed the opposite trend (Fig. 2C, E and Fig. S2D, E, H). Additionally, the effects of circRUNX1 on the migratory and invasive abilities of KYSE150 and TE1 cells were further examined using transwell and wound healing assays. As shown in Fig. 2F, G and Fig. S2F, G, circRUNX1 knockdown significantly restricted the migration and invasion of ESCC cells, whereas circRUNX1 overexpression had the opposite effect. Collectively, the above data suggested that circRUNX1 played a carcinogenic role in ESCC cells.

To depict the potential downstream regulatory mechanisms of circRUNX1 in ESCC cells, RNA-seq was utilized to determine the differentially expressed transcriptomes in TE1 cells with circRUNX1 knocking down and control cells, which identified 90 downregulated genes and 76 upregulated genes (Fig. 3A). Gene Ontology (GO) functional annotation and KEGG enrichment analysis assessed the differentially expressed genes involved in the regulation of cancers (Fig. S3A-C). Of the 90 downregulated genes, 44 genes have not been annotated and seven genes were long non-coding RNA (lncRNA) (Fig. S3D). To explore the potential target genes of circRUNX1 among the remaining 39 genes, GEPIA (http://​gepia.​cancer-pku.​cn/​) was used to predict the expression of these genes in ESCC patients. Results showed that only three genes (IFI44, CTSE and FOXP3) were upregulated in ESCC (Fig. S3E). Next, we detected the changes in the expression levels of these three genes in circRUNX1 knocked down KYSE150 and TE1 cells. As shown in Fig. 3C and Fig. S3F, FOXP3 was significantly downregulated after interfering with circRUNX1 and was therefore selected for further investigation. RT-PCR assay showed that FOXP3 was highly expressed in 41 pairs of ESCC tissues compared with corresponding paracancerous tissues in our cohort (Fig. 3B). The analysis result of Starbase (https://​starbase.​sysu.​edu.​cn) further validated this finding (Fig. S3G). Moreover, GEPIA website demonstrated that FOXP3 expression was related to the poor prognosis of ESCC patients (Fig. S3H). Subsequently, qRT-PCR and western blot analysis revealed that circRUNX1 could positively regulate the expression of FOXP3 in ESCC cells (Fig. 3C-F). In addition, FOXP3 mRNA level in ESCC tissues was moderately related to circRUNX1 transcripts (r = 0.5730, P < 0.0001) (Fig. 3G). Considering that the role of FOXP3 in ESCC cells remains largely unknown, we knocked down FOXP3 in KYSE150 and TE1 cells (Fig. S4A and B). CCK-8 assays revealed that FOXP3 depletion significantly reduced cell proliferation in KYSE150 and TE1 cells (Fig. S4C). Moreover, transwell and wound healing assays confirmed that FOXP3 knockdown inhibited KYSE150 and TE1 cell migration and invasion (Fig. S4D and E). The above data suggested that FOXP3 acted as a tumor promoter in ESCC cells. Given that circRUNX1 could increase FOXP3 expression, we wondered whether FOXP3 was involved in circRUNX1-mediated ESCC progression. To this end, FOXP3 expression was inhibited in KYSE150 and TE1 cells stably expressing circRUNX1. Overexpression of circRUNX1 increased the proliferative ability of KYSE150 and TE1 cells in CCK-8, colony formation, and EdU assays, while silencing FOXP3 reversed these phenotypes (Fig. 3H-J and Fig. S5A-C). Meanwhile, FOXP3 inhibition reduced circRUNX1-induced ESCC cell mobility, as demonstrated by transwell and wound healing experiments (Fig. 3K-M and Fig. S5D-F). Together, these findings showed that FOXP3 functioned as a downstream target of circRUNX1 in ESCC cells.

One of the most well-known circRNA mechanisms is its role as a competing endogenous RNA (ceRNA) for miRNA binding [21]. Since circRUNX1 was found primarily in the cytoplasm, we hypothesized that it might regulate FOXP3 levels by sponging miRNAs. Two bioinformatics databases (circBank and Targetscan) were utilized to predict potential targeting miRNAs bound by circRUNX1 and FOXP3, and 14 miRNAs were identified in the overlapping list (Fig. 4A). We further examined the expression of these miRNAs in circRUNX1 knockdown ESCC cells. After depleting circRUNX1, only one miRNA, miR-449b-5p, was up-regulated in KYSE150 and TE1 cells, according to qRT-PCR results (Fig. 4B). Correspondingly, ESCC cells stably transfected with circRUNX1 showed a lower expression of miR-449b-5p (Fig. 4C). The dual-luciferase reporter gene assays further showed that miR-449b-5p mimics markedly reduced the luciferase activities of wild-type circRUNX1 reporter but not mutant circRUNX1 in HEK-293 T cells (Fig. 4D). RIP assays confirmed that both circRUNX1 and miR-449b-5p could directly bind to Ago2 (Fig. 4E). Moreover, we analyzed the correlation between circRUNX1 and miR-449b-5p in ESCC tissues, and the result indicated that miR-449b-5p levels were negatively correlated with the expression of circRUNX1 (Fig. 4F). Consistently, miR-449b-5p expression was also significantly decreased in ESCC cell lines (KYSE450, EC109, TE1, KYSE70, KYSE150) compared with normal esophageal epithelium cell line (HEEC) (Fig. 4G). Next, rescue experiments were conducted to determine whether circRUNX1 promoted ESCC progression by competitively interacting with miR-449b-5p. As shown in Fig. 4H-J and Fig. S6A-C, after stably transfecting circRUNX1 in KYSE150 and TE1 cells, cell proliferation and clonogenic abilities were dramatically enhanced. However, these effects could be attenuated by restoring miR-449b-5p expression. Additionally, transwell and wound healing assays showed that the promoting effects of circRUNX1 on cell migration and invasion were abolished after transfecting miR-449b-5p in KYSE150 and TE1 cells (Fig. 4K-M and Fig. S6D-F). Thus, these findings suggested that circRUNX1 acted as a miR-449b-5p sponge to promote ESCC proliferation and metastasis.

Next, we transfected miR-449b-5p mimics in KYSE150 and TE1 cells to examine the change in FOXP3 expression. qRT-PCR revealed that the mRNA levels of FOXP3 were downregulated simultaneously after transfecting miR-449b-5p mimics (Fig. 5A). In addition, miR-449b-5p antagonized the upregulation of FOXP3 protein levels induced by circRUNX1 (Fig. 5B). To test whether miR-449b-5p could bind to the 3’UTR of FOXP3, luciferase reporters with wild-type FOXP3 (FOXP3-WT) and mutant FOXP3 (FOXP3-Mut) were constructed and transfected into HEK-293 T cells. The results of the dual-luciferase assay showed that miR-449b-5p mimics could reduce the luciferase activity of FOXP3-WT reporter rather than FOXP3-Mut reporter (Fig. 5C). Based on the above results, we surmised that circRUNX1 served its biological function via circRUNX1/miR-449b-5p/FOXP3 axis. To test this conjecture, KYSE150 and TE1 cells were treated with miR-NC, miR-449b-5p inhibitor, or miR-449b-5p inhibitor plus si-FOXP3. CCK-8, colony formation, and EdU assays showed that knockdown of FOXP3 resisted the growth-promoting effects caused by miR-449b-5p inhibitor (Fig. 5D-F and Fig. S7A-C). In addition, miR-449b-5p inhibitor led to an obvious enhancement in the migration and invasion of KYSE150 and TE1 cells, while FOXP3 downregulation partially mitigated this effect, as demonstrated by wound healing and transwell assays (Fig. 5G-I and Fig. S7D-F). Collectively, the above findings showed that circRUNX1 acted as a miR-449b-5p sponge to promote ESCC proliferation and metastasis by up-regulating FOXP3 expression.

To understand why circRUNX1 was overexpressed in ESCC, two bioinformatics databases (circAtlas and RBPmap) were searched, and six proteins with the potential to bind to circRUXN1 at its flanking regions were identified (Fig. 6A). Then the expression of these six proteins in ESCC patients was predicted by GEPIA. Interestingly, only the expression of IGF2BP2 differed between ESCC and normal esophageal tissues, as shown in Fig. S8. The results were in line with the ESCC samples that were analyzed using RT-PCR and IHC staining in our cohort (Fig. 6B and C). Besides, we noticed that IGF2BP2 overexpression in ESCC tissues was significantly associated with lymph node metastasis (Fig. 6D and Table S5). Next, we designed the overexpression vector and specific short hairpin RNA (shRNA) targeting IGF2BP2 and transfected them into ESCC cells. As expected, qRT-PCR results showed that knocking down or overexpressing IGF2BP2 could decline or increase the levels of circRUNX1 (Fig. 6E and F). Next, the positive relationship between circRUNX1 and IGF2BP2 was observed in 41 ESCC tissues (r = 0.5001, P < 0.001), and IHC staining also validated this result (Fig. 6G and H). To determine whether IGF2BP2 promoted circRUNX1 formation or inhibited circRUNX1 degradation, KYSE150 cells transfected with scramble or sh-IGF2BP2 were treated with actinomycin D for the specific time. The change in circRUNX1 expression levels was then measured upon IGF2BP2 inhibition. Notably, qRT-PCR analysis revealed that silencing IGF2BP2 significantly reduced the half-life of circRUNX1 (Fig. 6I). Subsequently, RIP assays confirmed the direct interaction between IGF2BP2 and circRUNX1 in ESCC cells (Fig. 6J). Next, we performed rescue experiments aiming to verify the biological effects of IGF2BP2 on ESCC progression. The proliferative abilities of KYSE150 and TE1 cells were tested using CCK-8, colony formation, and EdU assays. As shown in Fig. 6K-M and Fig. S9A-C, upregulation of IGF2BP2 increased cell viability, the number of cell colonies, and the percentage of proliferating cells. However, these promoting effects of IGF2BP2 were evidently impaired by circRUNX1 knockdown. Additionally, transwell and wound healing assays revealed that loss of circRUNX1 diminished Ov-IGF2BP2-mediated enhancement of migration and invasion in KYSE150 and TE1 cells (Fig. 6N-P and Fig. S9D-F). Taken together, these findings suggested that IGF2BP2 acted as an upstream target of circRUNX1 and promoted its expression by increasing RNA stability.

Next, we further investigated the specific roles of circRUNX1 and FOXP3 in vivo based on the findings described above. KYSE150 cells stably expressing circRUNX1 or co-transfected with circRUNX1 and FOXP3 shRNA were injected subcutaneously or intravenously into nude mice (Fig. 7A and B). After 30 days of observation, we noticed that circRUNX1 overexpression groups showed significantly faster tumor growth than the other two groups, while inhibition of FOXP3 in circRUNX1-overexpressing cells restricted tumor growth (Fig. 7C and D). This conclusion was also supported by the final tumor weights of the three groups (Fig. 7E). Next, the levels of FOXP3 and Ki-67 from these tumor tissues were detected by IHC staining. As shown in Fig. 7F, circRUNX1 increased FOXP3 and Ki-67 expression, while FOXP3 deletion reduced this effect. RT-PCR analysis of these xenograft tissues also revealed that overexpression of circRUNX1 increased the endogenous level of FOXP3 (Fig. 7G). Pulmonary metastasis models were then established by inoculating KYSE150 cells into the tail-vein of nude mice. Lung tissues from each group were retrieved at the end of the experiment, and H&E staining revealed that overexpression of circRUNX1 induced KYSE150 cells to form more lung metastases in vivo. In contrast, the number of lung foci was reduced after co-transfecting ESCC cells with sh-FOXP3, indicating that circRUNX1 and FOXP3 have pro-metastatic effects (Fig. 7H and I). Taken together, for the first time, the biological function and specific mechanism of circRUNX1 in ESCC progression have been identified (Fig. 8).