circPTPN12/miR-21–5 p/∆Np63α pathway contributes to human endometrial fibrosis

Emerging evidence demonstrates the important role of circular RNAs (circRNAs) in regulating pathological processes in various diseases including organ fibrosis. Endometrium fibrosis is the leading cause of uterine infertility, but the role of circRNAs in its pathogenesis is largely unknown. Here, we provide the evidence that upregulation of circPTPN12 in endometrial epithelial cells (EECs) of fibrotic endometrium functions as endogenous sponge of miR-21–5 p to inhibit miR-21–5 p expression and activity, which in turn results in upregulation of ΔNp63α to induce the epithelial mesenchymal transition (EMT) of EECs (EEC–EMT). In a mouse model of endometrium fibrosis, circPTPN12 appears to be a cofactor of driving EEC–EMT and administration of miR-21–5 p could reverse this process and improve endometrial fibrosis. Our findings revealed that the dysfunction of circPTPN12/miR-21–5 p/∆Np63α pathway contributed to the pathogenesis of endometrial fibrosis.


Introduction
Endometrial fibrosis is clinically characterized with intrauterine adhesions (IUA) and is often secondary to severe injury of endometrium including various uterine operations mainly repeated curettage. Endometrial fibrosis is the most common reason of uterine infertility (Yu et al., 2008;March, 2011a;March, 2011b). The mechanism of endometrial fibrosis remains unclear, which hampers the development of effective therapeutics for the disease (March, 2011b). Recently, our study showed that the ectopic expression of DNp63a, a transcription factor, in the endometria of IUA patients triggers the epithelial mesenchymal transition (EMT) of endometrial epithelial cells (EECs) (EEC-EMT) to promote endometrial fibrosis (Zhao et al., 2020).
In the process of EMT, post-transcriptional regulation has critical roles (Nieto et al., 2016). Studies showed that let-7d is downregulated in EMT of the lung epithelial cells and downregulation of let-7d in mice causes lung fibrosis (Pandit et al., 2010). miR-29b suppresses EMT of lung epithelial cells and prevents mice against lung fibrosis (Sun et al., 2019). In the renal fibrosis, miR-34a induces EMT in renal tubular epithelial cells and knockout miR-34a ameliorates renal fibrosis in mice (Liu et al., 2019a). Moreover, miRNAs may function differently in the pathogenesis of fibrosis in different tissues. For example, miR-27a prevents the fibrosis of detrusor and kidney (Wu et al., 2018;Zhang et al., 2018), but promotes the myocardial fibrosis (Ya-Se et al., 2018). However, the role of miRNAs in the pathogenesis of endometrial fibrosis is less studied (Li et al., 2016).
Recent studies demonstrate that the expression and activity of miRNAs are regulated by circular RNAs (circRNAs). circRNAs are recognized as competing endogenous RNAs (ceRNAs) to sponge miRNAs to modulate the expression and activity of miRNAs and their target genes (Li et al., 2018;Zheng et al., 2016;Cheng et al., 2019). To date, little is known about the biological function of circRNAs in the pathogenesis of endometrial fibrosis. Since we found that the ectopic expression of DNp63a in EECs induces EEC-EMT and promotes endometrial fibrosis (Zhao et al., 2020), and DNp63a is regulated by multiple miRNAs (Candi et al., 2015;Lena et al., 2008;Rodriguez Calleja et al., 2016), we speculated whether circRNAs is involved in the pathogenesis of endometrial fibrosis by interacting with miRNAs. Therefore, we studied the expression changes of circRNAs/miRNAs in endometrial fibrosis and their regulatory relationship with DNp63a to favor the understanding of endometrial fibrosis.

miR-21-5 p is downregulated in EECs and associated with EEC-EMT
To study the spectrum of changes in the levels of DNp63a antagonizer miRNAs, the transcriptomic alterations in the endometria of the patients with IUA were determined. We performed highthroughput sequencing analysis using endometrium samples in late proliferative phase from three severe IUA patients and three normal controls. The histopathological results of the samples used for high-throughput sequencing showed endometrial fibrosis in IUA patients (Figure 1-figure supplement 1A). With the significance threshold of absolute mean fold change > 2 and p-values<0.05, of 1681 detectable miRNAs in total, 56 were upregulated and 150 were downregulated in IUA patients, compared with those in normal controls ( Figure 1A and Supplementary file 1). These differentially expressed miRNAs were also presented in a heat map (Figure 1-figure supplement 1B). The most significantly increased and decreased miRNAs were verified by quantitative reverse transcription-PCR (qRT-PCR) (Figure 1-figure supplement 1C).
Among these dysregulated miRNAs, 40 upregulated and 40 downregulated miRNAs with higher expression abundance were shown in Figure 1B. The expressive abundance of miR-21-5 p was the highest in endometria of normal controls but was significantly downregulated in IUA patients (Supplementary file 2). Interestingly, we found that the 3 0 -UTR of DNp63a has miR-21-5 p binding sites by prediction in Microrna.org database. Thus, we tested the expression change of miR-21-5 p ex vivo and confirmed the decrease of miR-21-5 p expression in endometria of IUA patients ( Figure 1C), which was further proved by RNA-scope assay ( Figure 1D). Meanwhile, RNA-scope assay showed that miR-21-5 p was mainly expressed in EECs ( Figure 1D). Since epithelial cells often switch to matrix-producing myofibroblasts via EMT in tissue fibrosis process (Iwano et al., 2002;Zeisberg et al., 2007) and we proved recently that DNp63a inducing EEC-EMT participates in endometrium fibrosis in IUA patients (Zhao et al., 2020), we speculated that miR-21-5 p may serve as a protector for maintaining endometrium homeostasis and preventing EECs from EMT. Therefore, we performed loss and gain of miR-21-5 p function by transfecting miR-21-5 p mimic and inhibitor into primary EECs respectively. miR-21-5 p expression increased 11 times in EECs transfected with miR-21-5 p mimic, while the expression decreased 2.5 times after miR-21-5 p inhibitor transfection ( Figure 1-figure supplement 2). After miR-21-5 p knockdown, the expression of epithelium-specific marker, E-cadherin, was downregulated, but mesenchymal markers, N-cadherin, a-smooth muscle actin (a-SMA), and fibronectin (FN), were overexpressed in EECs ( Figure 1E). In contrast, overexpression of miR-21-5 p markedly decreased the expression of N-cadherin, a-SMA, and FN and increased the expression of E-cadherin ( Figure 1F). Volcano plots of miRNA-sequencing of the endometria from severe IUA patients (n = 3) and controls (n = 3) based on the high-throughput RNA-sequencing analysis. The red dots represent the miRNAs upregulated, and the green ones represent the downregulated (p<0.05 and fold change > 2). (B) A heatmap showing 40 upregulated and 40 downregulated miRNAs with high abundance in endometrium samples from severe IUA patients (n = 3) and controls (n = 3). (C) miR-21-5 p mRNA levels in endometria of severe IUA patients (n = 18) and controls (n = 18) determined by qRT-PCR. (D) Representative images of RNA-scope assay using specific probes to detect miR-21-5 p in the endometria of severe IUA patients (n = 3) and controls (n = 3). Peptidylprolyl isomerase B (PPIB) serves as the positive control. Arrowhead: epithelial cells; arrow (red dots): PPIB or miR-21-5 p positive dots. Scale bars 50 mm. (E) Left: fibronectin (FN), asmooth muscle actin (a-SMA), E-cadherin (E-cad), N-cadherin (N-cad), and GAPDH protein levels determined by western blotting in miR-21-5 p inhibitor (inh-21)-or negative control (inh-NC)-transfected (48 hr) EECs (n = 3). Right: The quantitative band intensities determined by image J software. (F) Left: FN, a-SMA, E-cad, N-cad, and GAPDH protein levels determined by immunoblotting in miR-21-5 p mimic (miR-21)-or negative control (miR-NC)-transfected (48 hr) EECs (n = 3). Right: The quantitative band intensities determined by image J software. The error bars in (C), (E), and (F) indicate mean ± SD. **p<0.01, ****p<0.0001. The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. qRT-PCR data for miR-21-5 p relative expression. circPTPN12 is upregulated in EECs and negatively correlated with miR-21-5 p Since circRNAs are the main regulators for miRNAs activity and act as ceRNA to sponge miRNAs by complementary base pairing to regulate the expression of target mRNAs (Li et al., 2018;Zheng et al., 2016;Cheng et al., 2019), we analyzed the expression profile of circRNAs in endometrium samples from severe IUA patients and normal controls through high-throughput sequencing. We detected 17,134 distinct circRNAs and 4886 of them have been registered in circBase database ( Figure 2-figure supplement 1A). We annotated these circRNAs and found that 4832 of them derived from protein-coding exons ( Of the 172 miR-21-5 p binding circRNAs, 12 were upregulated and 4 were downregulated in IUA patients (Figure 2A,B). We focused on the upregulated circRNA candidates because miR-21-5 p was downregulated in the endometria of IUA patients ( Figure 1C,D) and circRNAs function as ceRNAs to inhibit miRNAs activities and expression. The most upregulated circRNA was hsa_circ_0003764 (circ-Base ID), and it was 64-fold higher than that in controls ( Figure 2B). qRT-PCR ( Figure 2C) and RNAscope assay ( Figure 2D) showed that hsa_circ_0003764 was significantly upregulated in the endometria of IUA patients and mainly expressed in EECs. Through circPrimer software, we found that hsa_-circ_0003764 is derived from 5 to 8 exons of tyrosine-protein phosphatase non-receptor type 12 (PTPN12) gene. Therefore, we termed it circPTPN12 in following experiments ( Figure 2E).
We designed specific divergent primers to prove the existence of circPTPN12 in the endometria of IUA patients by PCR. The predicted spliced junction of circPTPN12 was validated with agarose gel electrophoresis ( Figure 2F, top). The resultant products of divergent primers were confirmed in line with the sequence of circPTPN12 by sequencing ( Figure 2F, bottom). To investigate the stability of circPTPN12, total RNA isolated from IUA patients' endometria was treated with RNase R for 15 min, then the expression of circPTPN12 and PTPN12 was respectively detected and the results showed that PTPN12 expression was significantly decreased while circPTPN12 expression was not changed ( Figure 2G). Furthermore, we added actinomycin D to inhibit the transcription of EECs. We harvested total RNA at five time points, analyzed the expression of circPTPN12 and PTPN12 by qRT-PCR, and revealed that the half-life of circPTPN12 was exceeding 24 hr, whereas the half-life of linear PTPN12 transcript was less than 4 hr ( Figure 2H). Thus, circPTPN12 was more stable than PTPN12 mRNA, suggesting that circPTPN12 is circular. Separation analysis of nuclear and cytoplasm showed that circPTPN12 was mainly distributed in the cytoplasm of EECs ( Figure 2I).
Since miR-21-5 p was decreased and circPTPN12 was increased in fibrotic endometria of IUA patients, and both were expressed in EECs, we further analyzed the correlation between them. The results showed that circPTPN12 was negatively correlated with miR-21-5 p ( Figure 2J). circPTPN12 functions as ceRNA to sponge miR-21-5 p To prove the interaction between circPTPN12 and miR-21-5 p, we constructed a circPTPN12 plasmid to transfect HEK-293T cells. The transfection efficiency was verified by qRT-PCR ( Figure 3A). We used biotin-labeled circPTPN12 to pull down the circPTPN12 in the lysates of the transfected HEK-293T cells. The electrophoresis showed that miR-21-5 p was pulled down by circPTPN12 ( Figure 3B). To determine the binding site of circPTPN12 for miR-21-5 p, we cloned the full-length sequence of circPTPN12 in pGL3 vector containing a luciferase reporter. Luciferase activity was  Figure 1E. Source data 3. The quantitative band intensities for Figure 1E. Source data 4. Uncropped western blots for Figure 1F. Source data 5. The quantitative band intensities for Figure 1F.  Source data 1. qRT-PCR data for hsa_circ_0003764 relative expression. Source data 2. Uncropped gels for Figure 2F. Source data 3. qRT-PCR data for circPTPN12 and PTPN12 relative expression with or without RNase R treatment. Source data 4. qRT-PCR data for circPTPN12 and PTPN12 relative expression with actinomycin D at the indicated time points. Source data 5. qRT-PCR data for circPTPN12, GAPDH, and U6 relative expression in the nuclear and cytoplasmic fractions of EECs.   substantially reduced in the presence of the miR-21-5 p mimic, which was effectively restored when miR-21-5 p was knockdown ( Figure 3C). We then compared the sequence of circPTPN12 with that of miR-21-5 p using circBank and Circular RNA Interactome databases and noticed that circPTPN12 contains three putative target sites of miR-21-5 p ( Figure 3D, top). The test of delete mutations of each or all of three putative binding sites showed that only the third site was validated to have the capacity of binding by the luciferase assay ( Figure 3D, bottom). The schematic diagram of real binding site is shown in Figure 3E. To further verify the binding of circPTPN12 with miR-21-5 p, we conducted RNA immunoprecipitation (RIP) experiment to precipitate the complex of argonaute (AGO) protein based on a principle that miRNAs exert post-transcriptional regulation through binding with AGO protein to form the targeting module of the miRNA-induced silencing complex (miRISC) (Gebert and MacRae, 2019). The RIP experiment showed that the precipitates of HEK-293T cell lysates stably expressing Flag-AGO2 or Flag-GFP contained not only miR-21-5 p, but also circPTPN12 ( Figure 3F), indicating the bind of circPTPN12 and miR-21-5 p. Fluorescence in situ hybridization (FISH) demonstrated that miR-21-5 p was colocalized with circPTPN12 in the cytoplasm of Ishikawa cell ( Figure 3G).
To investigate the effect of circPTPN12 on the function of miR-21-5 p, we constructed recombinant adenovirus harboring exons 5-8 of PTPN12 (Ad-circPTPN12) along with approximately 1 kb flanking intron sequences containing complementary Alu elements (Figure 3-figure supplement  1A). The expression of circPTPN12 increased seven times in EECs after Ad-circPTPN12 infection ( Figure 3-figure supplement 1B). Importantly, qRT-PCR confirmed that circPTPN12 overexpression decreased the miR-21-5 p level in primary EECs ( Figure 3H). Meanwhile, compared with that of adenovirus containing no circPTPN12 (Ad-control), overexpression of circPTPN12 enhanced protein abundance of N-cadherin, a-SMA, and FN, but decreased the level of E-cadherin in EECs ( Figure 3I). While re-transfected with miR-21-5 p mimic reversed this phenomenon with upregulated E-cadherin expression and downregulated N-cadherin, a-SMA, and FN expression ( Figure 3J).

Downregulation of miR-21-5 p by circPTPN12 promotes EEC-EMT through upregulating DNp63a
Since circPTPN12 inhibited miR-21-5 p, activity to promote EEC-EMT and miR-21-5 p may be the antagonizer of DNp63a that was predicted by Microrna.org database. We conducted the luciferase reporter assay with the full-length 3 0 -UTR of DNp63a wild and DNp63a mut targeted by miR-21-5 p ( Figure 4A). In line with the prediction, transfection with miR-21-5 p mimic significantly reduced luciferase activity in HEK-293T cells transfected with DNp63a wild plasmid, whereas knockout of miR-21-5 p in DNp63a wild HEK-293T cells enhanced the luciferase activity; however, these effects were not observed in HEK-293T cells transfected with DNp63a mut plasmid ( Figure 4B). We used primary EECs transfected with the miR-21-mimic or inhibitor for 48 hr following infection with recombinant adenovirus containing DNp63a (Ad-DNp63a) and empty vector (Ad-CTL), respectively, and found that miR-21-5 p mimic remarkably downregulated the mRNA and protein of DNp63a in DNp63a + EECs ( Figure 4C,D). And the miR-21-5 p inhibitor upregulated DNp63a protein level in DNp63a + EECs ( Figure 4E). Meanwhile, upregulation of circPTPN12 increased luciferase activity in HEK-293T cells transfected with DNp63a wild plasmids, while no luciferase activity change was Source data 1. qRT-PCR data for circPTPN12 relative expression. Source data 2. qRT-PCR data for miR-21-5 p relative expression. Source data 3. Data on luciferase activity in HEK-293T cells transfected with miR-21-5 p mimic or inhibitor. Source data 4. Data on luciferase activity of Luc-circPTPN12 containing single or all mutated sites of three putative miR-21-5 p binding sites in HEK-293T cells transfected with miR-21-5 p mimic. Source data 5. qRT-PCR data for miR-21-5 p relative expression in EECs with Ad-control or Ad-circPTPN12 treatment. Source data 6. Uncropped western blots for Figure 3I. Source data 7. The quantitative band intensities for Figure 3I. Source data 8. Uncropped western blots for Figure 3J. Source data 9. The quantitative band intensities for Figure 3J.  Source data 1. Data on luciferase activity of Luc-DNp63a wild or mut in HEK-293T cells transfected with miR-21-5 p mimic or inhibitor. Source data 2. qRT-PCR data for DNp63a relative expression in EECs transfected with miR-21-5 p mimic. Source data 3. Uncropped western blots for Figure 4D,E. Source data 4. The quantitative band intensities for Figure 4D. Source data 5. The quantitative band intensities for Figure 4E. Source data 6. Data on luciferase activity of Luc-DNp63a wild or mut in HEK-293T cells transfected with circ-control or circPTPN12. observed when HEK-293T transfected with DNp63a mut plasmids ( Figure 4F). Moreover, the upregulation of circPTPN12 counteracted the inhibitory effect of miR-21-5 p on DNp63a expression ( Figure 4G). And that we revealed that miR-21-5 p mimic remitted DNp63a-induced EEC-EMT, upregulated E-cadherin and downregulated N-cadherin and a-SMA at both the mRNA and protein levels ( Figure 4H,I). Furthermore, upregulation of circPTPN12 counteracted the reversal effect of miR-21-5 p on DNp63a-induced EMT in EECs ( Figure 4J).

miR-21-5 p alleviates circPTPN12-induced EEC-EMT in IUA-like mouse model
We developed a mouse model by uterine mechanical injury to simulate IUA in human as described previously (Zhao et al., 2020;Yang et al., 2017). Compared with sham-operated mice, the mechanically injured mice had mildly upregulated N-cadherin and a-SMA in the luminal epithelial cells of endometria, slightly downregulated E-cadherin in luminal and glandular EECs, and minimally changed in Masson staining ( Figure 5-figure supplement 1A).
Based on circBase and circBank database, mouse does not express circPTPN12, which provided us an opportunity to test the role of circPTPN12 in the pathogenesis of endometrium fibrosis by intrauterine injection of a circPTPN12-containing recombinant adeno-associated virus with a green label (AAV-circPTPN12). The flowchart of intrauterine injection is presented in Figure 5-figure supplement 1B. The green fluorescence in mouse endometria was clearly observed after four estrous cycles following virus injection ( Figure 5-figure supplement 1C), indicating that AAV-circPTPN12 infected the mouse endometrium. We further validated the increased circPTPN12 levels in the endometria by qRT-PCR ( Figure 5A). With the increase of circPTPN12 expression, the expression of miR-21-5 p decreased significantly ( Figure 5B). Furthermore, the mice injected with AAV-circPTPN12 showed increased mRNA levels of N-cadherin, a-SMA, and FN and decreased mRNA level of E-cadherin, compared with those injected with adeno-associated virus with empty vector ( Figure 5C). Immunohistochemical staining displayed that AAV-circPTPN12 injection clearly reduced E-cadherin expression in EECs and remarkably upregulated N-cadherin and a-SMA in both epithelial and stroma cells ( Figure 5D). In addition, Masson staining showed strong positive ( Figure 5D). The results of EMT and severe endometrial fibrosis in mice caused by AAV-circPTPN12 intrauterine injection were similar to the results induced by dual-injury, namely mechanical injury with lipopolysaccharide injection (Zhao et al., 2020). DNp63a mRNA was also significantly upregulated after AAV-circPTPN12 administration ( Figure 5E).
Since circPTPN12 served as the ceRNA of miR-21-5 p and the pro-EMT effect of circPTPN12 was reversed by miR-21-5 p in vitro ( Figure 3B-I), we further conducted the mouse experiment by injecting agomir-21-5 p or agomir negative control (agomir-NC) into the uterine cavity of the AAV-circPTPN12 mouse model every 5 days for three times ( Figure 5-figure supplement 1D,E). The results showed that miR-21-5 p in the endometria was overexpressed ( Figure 5F) and the mRNA level of E-cadherin was elevated, while the mRNA levels of N-cadherin, a-SMA, and FN were decreased ( Figure 5G). Immunohistochemical results showed that agomir-21-5 p application significantly increased E-cadherin and reduced N-cadherin and a-SMA in both luminal and glandular epithelia and reduced Masson staining compared with agomir-NC injection ( Figure 5H).

Discussion
In the present study, we revealed that in normal endometrium, the high level of miR-21-5 p in EECs inhibits DNp63a expression to maintain the homeostasis of EECs, and in the pathological circumstance, circPTPN12 is highly expressed, which decreases miR-21-5 p level in EECs, so the suppression effect of miR-21-5 p on the expression of DNp63a is counteracted, leading to EEC-EMT and Source data 7. Uncropped western blots for Figure 4G. Source data 8. The quantitative band intensities for Figure 4G. Source data 9. qRT-PCR data for E-cad, N-cad, a-SMA, and FN relative expression. Source data 10. Uncropped western blots for Figure 4J. Source data 11. The quantitative band intensities for Figure 4J.   Figure 5 continued on next page endometrial fibrosis. Therefore, we propose a working model to explain the role of circPTPN12/miR-21-5 p/DNp63a pathway in the pathogenesis of human endometrial fibrosis ( Figure 6).
Previously, we observed that DNp63a is ectopically expressed in EECs in IUA patients (Zhao et al., 2017), and showed that DNp63a promotes the expression of SNAI1 by DUSP4/GSK3B pathway and induces EEC-EMT and endometrial fibrosis (Zhao et al., 2020). These findings are in agreement with the therapeutic effect of transplantation of mesenchymal stem cells into the uterine cavity (Cao et al., 2018). Since miRNAs are main regulator of DNp63a (Candi et al., 2015;Lena et al., 2008;Rodriguez Calleja et al., 2016) and sufficient abundance of miRNAs transcripts in cells is essential in the regulating target genes (Liu et al., 2019a;Brown and Naldini, 2009), we chose the highest expressed miR-21-5 p in normal endometria to study its role in the current study. We found that miR-21-5 p is located in EECs, and the downregulation of miR-21-5 p can increase the expression of DNp63a, leading to the transition of EECs to cells with features of mesenchymal cells. In IUA mice model, supplementation of exogenous miR-21-5 p reversed the EEC-EMT and endometrial fibrosis, indicating that miR-21-5 p is essential in maintaining EECs properties. Our Source data 1. qRT-PCR data of circPTPN12 relative expression for Figure 5A. Source data 2. qRT-PCR data of miR-21-5 p relative expression for Figure 5B. Source data 3. qRT-PCR data of E-cad, N-cad, a-SMA, and FN relative expression for Figure 5C. Source data 4. qRT-PCR data of DNp63a relative expression for Figure 5E. Source data 5. qRT-PCR data of miR-21-5 p relative expression for Figure 5F. Source data 6. qRT-PCR data of E-cad, N-cad, a-SMA, and FN relative expression for Figure 5G.  findings appear to be different from the observations that in lung and kidney fibrosis researches miR-21-5 p promotes EMT (Liu et al., 2019b;Wang et al., 2019;Luo et al., 2019;Liu et al., 2019c). Thus, we consider that miR-21-5 p may play different roles in different tissues, which merits further study.
CircRNAs are important regulators of miRNAs and play critical roles in the pathogenesis of lung/ renal/liver fibrosis. In this study, we identified a new circPTPN12, which rarely expressed in normal endometrium but significantly upregulated in the endometria of IUA patients. The colocalization of circPTPN12 and miR-21-5P in the cytoplasm of EECs, luciferase assay, and RIP experiments supports the circPTPN12/miR-21-5 p sponge hypothesis. Therefore, the current study for the first time demonstrates the role of a circRNA in endometrial fibrosis.
There are some limitations in our study. First, since the IUA patients enrolled in this study were all associated with repeated curettage, the findings in the present study may not be applicable to the IUA patients associated with infection or uterine artery embolization. Second, we only studied the roles of miR-21-5 p and circPTPN12 based on their highest expressions in the heatmap or volcano plots, yet there may be other less-expressed miRNAs and circRNAs that are potentially associated with the endometrial fibrosis, which remains to be further studied.

RNA-sequencing, bioinformatics analysis for miRNAs and circRNAs
Total RNA extracted from the endometrial samples was subjected to library construction, followed by sequencing on an Illumina Hiseq 250020 (miRNA) or Illumina Hiseq X10 platform (circRNA) by Vazyme (Nanjing, China). The raw reads were filtered to achieve the clean tags through in-house Perl processes to map to the reference genome. miRNAs were identified by mapping to miRBase database and circRNAs were identified by circRNA Finder database (Fu and Liu, 2014). The miRNAs and circRNAs expression level were calculated and normalized to transcripts per million (TPM), TPM = 10 6 C/L, where C is the counts of a miRNA or circRNA, and L is counts sum of all miRNAs or circRNAs. To identify differentially expressed miRNAs and circRNAs between samples, the edgeR package (http://www.r-project.org/) was used. Differentially expressed miRNAs and circRNAs were identified with fold change > 2 or < À2 and p<0.05.

Cell isolation and culture
The isolation and culture of EECs were described previously (Zhao et al., 2017). Endometrial tissues were treated with collagenase type I (Sigma, St Louis, MO), hyaluronidase (Sigma), and DNase (Roche, Indianapolis, IN), followed by filtration through a 40 mm cell strainer (BD Biosciences, San Jose, CA) to remove the stromal cells and then through a 100 mm cell strainer (BD Biosciences) to collect the EECs. The cells were plated on matrigel-coated dishes and cultured with keratinocyte serum-free medium (Gibco, Waltham, MA) containing 2% fetal bovine serum (FBS), 100 U/ml penicillin, and 0.1 mg/ml streptomycin. HEK-293T and Ishikawa cells were grown in Dulbecco's modified Eagle's medium containing 10% FBS, 100 U/ml penicillin, and 0.1 mg/ml streptomycin at 37˚C with 5% CO 2 . They are purchased from ATCC and authenticated using STR profiling by ATCC, and they are tested to be free from mycoplasma contamination.

RNA and qRT-PCR
Total RNA was extracted by Trizol reagent (Invitrogen Life Technologies, Carlsbad, CA). For acquisition of RNA from cytoplasm and nucleus, respectively, the nuclear and cytoplasmic fractions were isolated using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Scientific, Waltham, MA). The circRNAs were reverse-transcribed using PrimeScript reagent Kit with DNA Eraser (TaKaRa BIO, Japan). qRT-PCR was performed using SYBR qPCR Master Mix (Vazyme). U6 and GAPDH were used as internal controls. The relative RNA levels were determined by the 2 ÀDCt or 2 ÀDDCt method. All primers are listed in Supplementary file 5.

Western blotting
After the determination of protein concentration of the cell lysates, each sample with equal amount of protein was subjected to SDS-PAGE (10% or 15%) and transferred to a PVDF membrane (Millipore, Burlington, MA). The membrane was incubated with the specific primary antibody overnight at 4˚C, followed by incubation with HRP-conjugated anti-rabbit IgG (

Immunohistochemistry and immunofluorescence
For immunohistochemistry, mice endometrium tissues were fixed in 10% formaldehyde for 8 hr at 4C , transferred to a tissue processor (Leica ASP300 S, Wetzlar, Germany) for dehydration and wax infiltration, and embedded in paraffin. Paraffin blocks were cut into 2 mm thick slices. After dewaxing and blocking endogenous peroxidase activity in 3% H 2 O 2 , slices were heat-mediated in universal antigen retrieval. The slides were blocked in 2% bovine serum albumin/PBST for 1 hr. After these pretreatments, slides were incubated with primary antibodies at 4˚C overnight and washed by PBST. HRP-conjugated secondary antibodies were incubated for 1 hr. Slides were exposed to 3 0 3-diaminobenzidine to visualize the antigen signals. The positive staining was confirmed in a blinded manner by two independent observers by microscopy (DMi8, Leica, Wetzlar, Germany). For cell immunofluorescence, isolated EECs that had been plated on matrigel-coated coverslips for varying times were fixed in 4% paraformaldehyde for 10 min and permeabilized with cold methanol. Fixed cells were stained with a primary antibody overnight, washed, and incubated with secondary antibodies conjugated to fluorescein or rhodamine. Nuclei of EECs were stained using 4 0 ,6-diamidino-2-phenylindole (DAPI, Abcam). The primary antibodies were as follows: E-cad (1:400, Abcam), N-cad (1:400, Abcam), a-SMA (1:100, Abcam), and DNp63a (1:100, Millipore).

Dual-luciferase reporter assay
The full-length sequence of circPTPN12 or the mutants were inserted into pGL3 luciferase vector to obtain the circPTPN12 wild or circPTPN12 mut constructs (Generay Biotechnology, Shanghai, China). The mutants were those with deletion of each or all the three sites of circPTPN12 for binding miR-21-5 p. 3 0 -UTR or the 3 0 -UTR with miR-21-5 p binding site mutation of DNp63a were inserted into a pGL3 plasmid to obtain the pGL3-DNp63a wild or pGL3-DNp63a mut constructs (Generay). HEK-293T cells were co-transfected with circPTPN12 wild or circPTPN12 mut or pGL3-DNp63a wild or pGL3-D Np63a mut and miR-21-5 p mimic (Ribobio, Guangzhou, China) or miR-21-5 p inhibitor (Ribobio) or circPTPN12 plasmid using Lipofectamine 3000. After 24 hr, the firefly and Renilla luciferase activities were measured consecutively using a Dual Luciferase Reporter Assay kit (Vazyme).

Recombinant dNp63a adenovirus construction
DNp63a adenovirus (Ad-DNp63a) were constructed as previous study (Zhao et al., 2017). The open reading frame (GenBank: AF075431.1) and 3 0 -UTR of Homo sapiens of DNp63a were inserted into a pDC315-3FLAG-SV40-EGFP vector and ligated into a shuttle plasmid. Then, HEK-293A cells were co-transfected with the shuttle plasmid and adenoviral backbone plasmid to produce DNp63a recombinant adenoviral vector. The same vector was used without the DNp63a insertion to construct the control virus (GeneChem, Shanghai, China). circPTPN12 adenovirus (Ad-circPTPN12), adeno-associated virus (AAV-circPTPN12), and plasmid construction circPTPN12 adenovirus were obtained from Genechem. Briefly, the full length of circPTPN12 with its flanking introns including complementary Alu elements was amplified to insert into a pCMV-MCS-EGFP vector, and an empty pCMV-MCS-EGFP vector served as control virus. circPTPN12 adeno-associated virus was obtained from HanBio Biotechnology (Shanghai, China). The full length of circPTPN12 was inserted into pHBAAV-CMV-CircRNA-EF1-ZsGreen vector and then inserted into adeno-associated virus. The empty vector was used as control virus. circPTPN12 overexpression plasmid was obtained from Genepharma (Shanghai, China). The full length of circPTPN12 with its flanking introns was inserted into a pGCMV/ MCS/Neo (pEX-3) vector.

RNA-RNA pull-down assay
A biotin-labeled circPTPN12 probe and an unlabeled probe (5 0 -AGGCCATTACAATGATCTGCAA TGAATAC-3 0 , General Biosystems, Chuzhou, China) were separately incubated with streptavidincoated magnetic beads (Thermo Scientific), followed by incubation with the lysates of HEK-293T cells transfected with circPTPN12 plasmid. The RNA was extracted from the precipitated magnetic beads and subjected to qRT-PCR, and the resultant qRT-PCR products were further analyzed by agarose gel electrophoresis.

RNA-binding RIP
The lysates of HEK-293T cells transfected with circPTPN12 plasmid were incubated with Protein-A/G agarose beads (Millipore) and antibody against Ago2 (Abcam, cat# ab186733). RNA was extracted from the precipitates. The abundance of circPTPN12 and miR-21-5 p in the precipitates was detected by qRT-PCR, and the resultant qRT-PCR products were further analyzed by agarose gel electrophoresis.

Mouse models
All mouse procedures were approved by the Institutional Animal Care and Use Committee at the Nanjing Drum Tower Hospital. BALB/c female mice, at 8-10 weeks of age, 18-20 g, were bred in the Animal Laboratory Center of Nanjing Drum Tower Hospital. Mechanical injury of endometrium was performed as previously reported (Zhao et al., 2020). Briefly, mice at estrum defined by vaginal smears were anesthetized with isoflurane. Laparotomy was performed to expose the uterus, followed by insertion of rough surfaced needle to injure uterine endometrium. Mice in sham-operation were just subjected to laparotomy to expose the uterus without injury. AAV-circPTPN12 or AAV-control (10 ml of 1 Â 10 9 viral genomes/ml) was administered via intrauterine injection 5 days after mechanical injury to study the function of circPTPN12 in endometrium fibrosis. For the therapeutic effect of miR-21-5 p, agomir-21-5 p (Ribobio) or agomir-NC (Ribobio) (5 nmol in 10 ml) was administered every 5 days for three times via intrauterine injection after mechanical injury and AAV-circPTPN12 injection. Mouse model grouping was shown in Supplementary file 6. Uterine tissues were collected at estrum, 24-26 days after first surgery. Two side uteri of each mouse were divided into three parts: 1/3 for RNA isolation, 1/3 for immunohistochemistry analysis, and 1/3 for immunofluorescence.

Statistical analysis
Statistical analyses were carried out by GraphPad Prism software (version 6.0). Student's t-test was used to analyze two experimental groups if data were normally distributed, and one-way ANOVA followed by a Student-Newman-Keuls multiple comparison test was used for comparing three or more groups. Data were presented as mean ± standard deviation (SD) or means ± standard error of mean (SEM) indicated in each figure legend. p<0.05 was considered statistically significant.