Regulation of partial endothelial-to-mesenchymal transition by circATXN1 in ischemic diseases

Abstract

Ischemic injury induces a partial mesenchymal shift in endothelial cells (ECs), contributing to impaired vascular regeneration. However, the molecular regulators of this transitional state remain poorly defined. To address this, we performed circular RNA profiling of endothelial cells under ischemic-like conditions and identified a marked upregulation of a circular RNA, named circATXN1. Functional studies revealed that circATXN1 knockdown modulates endothelial phenotype and vascular response after ischemia. Functional studies have shown that knockdown of circATXN1 can regulate the endothelial cell phenotype and vascular response after ischemia. Mechanistically, circATXN1 knockdown enhances the demethylase protein ALKBH5 to reduce the RNA methylation level of the key transcription factor SLUG, thereby stabilizing SLUG. In animal models, suppression of circATXN1 enhances angiogenesis and improves recovery following ischemic injury. Here, we show that circATXN1 regulates partial endothelial-to-mesenchymal transition (EndMT) and angiogenesis by controlling SLUG mRNA methylation dynamics, highlighting its potential as a therapeutic target in ischemic disease.

Introduction

Ischemic diseases, which include many diseases such as ischemic heart disease, peripheral artery disease, and cerebral ischemia, are marked by the blockage or limitation of blood flow in different tissues and organs caused by hypoxia/ischemia^1,2,3. These disorders have a significant impact on world health, causing high rates of the disorders and death. Endothelial-to-mesenchymal transition (EndMT) is a dynamic biological process occurring in endothelial cells (ECs) that involves the development and progression of ischemic conditions^4,5,6. This process is distinguished by alterations in the genes and proteins expression, encompassing the hindrance in endothelial markers (encompassing VE-cadherin and CD31) and a rise in mesenchymal markers (encompassing SM22α and α-smooth muscle actin (αSMA)). The phenotypic switches associated with EndMT are consistent with molecular changes, leading to the loss of endothelial properties involving cytokine-induced leukocyte adhesion and tube formation and the emergence of mesenchymal traits, including matrix invasion and migratory capacity. EndMT exemplifies endothelium plasticity⁷. The initiation and magnitude of EndMT are impacted by intrinsic factors, such as embryonic lineage^7,8,9 and genetic makeup¹⁰, as well as external elements, including hemodynamics, cytokines, and hypoxia, which can trigger EndMT independently or in combination⁷.

Partial or reversible EndMT, unlike a complete mesenchymal transition, leads to an intermediate phenotype in endothelial cells (ECs) that are undergoing a transition toward a mesenchymal state. This transition retains or partially loses endothelial traits while acquiring mesenchymal properties. Angiogenesis processes, such as tubulogenesis and sprouting, necessitate a partial or reversible EndMT¹¹. During this stage, cells have a temporary loss of apical-basal polarity and acquire the ability to migrate, but they do not fully develop all mesenchymal characteristics or totally lose cell adhesion^12,13,14. Angiogenesis is also crucial for postischemic revascularization and functional recovery in ischemic diseases. Recent studies using single-cell transcriptomics have shown the temporary and changing partial EndMT nature in ischemic models^15,16, encompassing myocardial infarction (MI) and hindlimb ischemia (HLI). These results demonstrate that there is an increase in mesenchymal gene enrichment in vessel sites that have undergone clonal expansion. This indicates that hypoxia-stimulated partial EndMT may help promote neovascularization under ischemic conditions.

Circular RNAs (circRNAs) are a distinct kind of non-coding RNAs that are often produced via the process of backsplicing^17,18. These entities are distinguished by their strong stability and a covalently sealed uninterrupted loop, without 5’ to 3’ polarity and a polyadenylated tail. This allows them to withstand RNA exonucleolytic digestion¹⁹. CircRNAs have been suggested as important regulators in the processes of growth²⁰, development^17,21, and tissue regeneration²² and are the key to revealing the pathological mechanism of diseases and search for therapeutic targets. Their role in the initiation and progression of ischemic diseases has been extensively studied. Emerging evidence indicates that circRNAs may impact ischemic diseases by regulating EndMT. For instance, circDLGAP4 could improve ischemic stroke outcomes by targeting miR-143 to regulate EndMT related to blood-brain barrier integrity²³. Additionally, circUCK2 was proved to upregulate HECTD1 expression by binding with FUS, inhibiting EndMT to mitigate blood-brain barrier damage in ischemic stroke²⁴. However, few investigations have been conducted on the connection between circRNAs and hypoxia-induced partial EndMT in ischemic disease. Considering the similarities in biological characteristics between angiogenesis and partial EndMT, it is important to investigate the molecular mechanism behind hypoxia-induced partial EndMT and its impact on revascularization in ischemic diseases. Our study uniquely investigates the role of circATXN1 in partial EndMT within ischemic cardiovascular contexts. Previous studies demonstrated that circATXN1 promotes angiogenesis in gastric endothelial cells (GECs) by targeting miR-526b-3p and regulating the MMP2/VEGFA axis²⁵. However, this finding is based on a specific tumor microenvironment in GECs, which differs significantly from the ischemic conditions in cardiovascular diseases. In our study, we demonstrated that circATXN1 regulates endothelial function by interacting with ALKBH5 to regulate SLUG mRNA stability, independent of miRNA-mediated mechanisms. This discrepancy highlights the context-specific role of circATXN1 in endothelial cells, providing new insights into the regulation of ischemic disease progression and suggesting potential therapeutic targets.

Results

CircATXN1 was upregulated in partial EndMT endothelial cells

Previous studies have shown that transient or partial EndMT plays a crucial role in the progression of ischemic disease. Two models, myocardial infarction (MI) and hindlimb ischemia (HLI), are often used to study the role of partial EndMT in ischemic disease^16,26,27. Permanent ligation of the anterior descending branch of the left coronary artery in myocardial infarction model mice replicates the pathophysiology of human myocardial infarction and provides valuable insights into the activation of cardiac endothelial cells after myocardial ischemia. Similarly, the HLI model, which involves ligation and excision of the right superficial and deep femoral arteries, is used to simulate peripheral ischemic injury, which is a key factor in conditions such as peripheral artery disease. These models enabled us to examine the role of circATXN1 in modulating angiogenesis and partial EndMT in ischemia disease. Aligned with prior investigation²⁷, immunofluorescence staining (IFS) demonstrated a significant rise in double-positive cells expressing CD31 (an endothelial marker) and αSMA (a mesenchymal marker) in both the myocardial infarction and HLI ligation groups (Supplementary Fig. 1a, b). This confirms the presence of partial EndMT following the ischemic disease. Subsequently, we examined whether ECs received partial EndMT in vitro under hypoxic circumstances. HCMECs were precisely regulated in hypoxic conditions (1% O₂, 5% CO₂, and balanced N₂) and deprived of serum for a duration of 24 h to replicate the very challenging microenvironment of tissue ischemia. To induce complete EndMT, HCMECs were treated with transforming growth factor β (TGF-β) at a concentration of 20 ng/mL to induce complete EndMT. IFS revealed that hypoxic HCMECs exhibited a progressive elevation in the mesenchymal indicator αSMA expression and a hindrance in the endothelial marker CD31 expression (Fig. 1a), suggesting a transition from an endothelial to a mesenchymal phenotype under hypoxic conditions. As a result, exposure to hypoxia significantly elevated the expression of some proteins that indicate the presence of mesenchymal cells, such as fibronectin, αSMA, and fibroblast-specific protein 1 (FSP1). However, the proteins CD31 and VE-cadherin, which are associated with endothelial cells, were reduced but not eliminated (Fig. 1b). We also examined the effects of hypoxia on cell viability, and it was found that 24 h of hypoxia increased cell survival, while TGF-β stimulation decreased cell survival (Fig. 1c). In addition, we examined the tube formation of HCMECs, and found that hypoxia had no significant effect on HCMECs tube formation ability, while complete EndMT caused by TGF-β stimulation caused HCMECs to lose the tube formation ability (Fig. 1d). These results indicate that mesenchymal ECs maintained some characteristics of ECs despite undergoing a partial EndMT. Conversely, the administration of TGF-β resulted in the total disappearance of endothelium markers, a condition known as complete EndMT. The occurrence of circRNA and hypoxia-induced partial EndMT in ischemic disorders is hardly documented. Bioinformatic transcriptome analysis of normoxia- and hypoxia-treated ECs was performed to identify relevant circRNAs (Supplementary Table 1). Among the upregulated circRNAs, there was one circRNA circATXN1 highly conserved and met the screening criteria (p < 0.05, logFC > 1, CPM > 0.2) (Fig. 1e). Sequence alignment of human and mouse circATXN1 showed that the homology was as high as 81.90% (Supplementary Fig. 2). We detected circATXN1 level by RT-qPCR experiments to verify these findings. The outcomes demonstrated that circATXN1 was significantly raised in both hypoxic-treated homo HCMECs and mice MCMECs (Fig. 1f). Next, we examined the circATXN1 expression in ischemic animal models, and compared to the respective controls, RT-PCR reported that both ischemic myocardial and gastrocnemius had a significant rise in circATXN1 expression (Fig. 1g, Supplementary Fig. 3a). Furthermore, immunofluorescence reported an elevation in circATXN1 expression in the CD31-positive cells in the ischemic disease model (Fig. 1h and Supplementary Fig. 3b).

Fig. 1: CircATXN1 was upregulated in partial EndMT endothelial cells.

a IFS of HCMECs cultured under normoxic or hypoxic conditions or treated with TGF-β (20 ng/mL), stained for CD31 (green), αSMA (red), and DAPI (blue) (n = 8 IF images, scale bars = 20 μm). b Western blot analysis of CD31, VE-cadherin, fibronectin, αSMA, and FSP1 protein expression in hypoxia-treated HCMECs, with β-actin as the control (n = 3 biological replicates). c The viability of HCMECs cultured under normoxic or hypoxic conditions or treated with TGF-β (20 ng/mL) was determined by a CCK-8 assay (n = 8 biological replicates). d Tube formation assays of HCMECs cultured under normoxic or hypoxic conditions or treated with TGF-β (20 ng/mL) (n = 4 biological replicates, scale bar, 100 μm). e Volcano plot of differentially expressed circRNAs between normoxic and hypoxic HCMECs. f RT-qPCR of circATXN1 in hypoxic HCMECs and MCMECs, with ANGPTL2 as a positive control and β-actin as an internal control (n = 4 biological replicates). g, h circATXN1 expression in hypoxic heart muscle: RT-qPCR and IFS for CD31 (green) and circATXN1 (red) with DAPI staining (blue) in the infarcted border zone post-MI (n = 8 biological replicates, scale bars = 50 µm). i Expression of circATXN1 in the public RNA-seq dataset GSE100206. FPKM (Fragments Per Kilobase of transcript per Million mapped reads) and TPM (Transcripts Per Million) were used methods for quantifying gene expression level. j circATXN1 expression in peripheral blood exosomes in patients with myocardial infarction and lower limb ischemia compared to the control group (n = 32 patients). Statistical analysis: one-way ANOVA with Tukey’s test for (c, d); Wald test and Benjamini & Hochberg method was used for p-value calculating and adjustment for (e); two-Way ANOVA with multiple comparisons test for (f); unpaired two-tailed Student t test for (g, j). Error bars represent SEM. Source data are provided as a Source Data file.

We identified the presence of circATXN1 in exosomes derived from human plasm and serum a in the public RNA-seq NCBI GEO Dataset GSE100206, showing widespread expression across different individuals (Fig. 1e). Building on the findings from our study, which demonstrated that circATXN1 is predominantly expressed in endothelial cells, we hypothesized that circATXN1 detected in plasma may originate from endothelial cells and be secreted into peripheral blood in the form of exosomes. Furthermore, we utilized publicly available datasets from NCBI to validate this hypothesis (Fig. 1i). To investigate whether circATXN1 expression is significantly elevated in the endothelial cells of patients with ischemic disease and subsequently secreted into the peripheral blood, we collected peripheral blood samples from healthy volunteers, patients with myocardial infarction and non-diabetic patients with a car accident. We extracted exosomes from these samples and measured circATXN1 expression using RT-qPCR. Our results revealed that circATXN1 expression in peripheral blood exosomes was significantly higher in patients with myocardial infarction and lower limb ischemia compared to the control group (Fig. 1j). These findings suggest that circATXN1 might serve as a valuable biomarker for predicting disease progression in ischemic conditions.

Characterization of the hypoxia-induced circATXN1

The bioinformatics analysis revealed that circATXN1 originated from exon 7 of the ATXN1 host gene and had a length of 2078 nucleotides (Fig. 2a). The validity of this discovery was confirmed by experimental verification using RT-PCR with several primers and Sanger sequencing (Fig. 2b). We also identified homologous variants of circATXN1 in mice (circATXN1) to ascertain the conservation of circATXN1 (Supplementary Fig. 4a). Additionally, the amplification of circATXN1 was achieved exclusively by utilizing divergent primers and cDNAs as templates rather than genomic DNAs. This effectively rules out the possibility that circATXN1 was generated by genomic rearrangements or PCR artifacts (Fig. 2c). When comparing the linear transcripts of ATXN1 with circATXN1, it was shown that circATXN1 was not affected by the RNA exonuclease RNase R (Fig. 2c). This provides more evidence that circATXN1 is a circular RNA. The subcellular location of circATXN1 was determined by IFS, revealing its predominant localization in the cytoplasm of HCMEC (Fig. 2d). Simultaneously, RNAs from the cytoplasm and nucleus were isolated for RT-qPCR analysis. Figure 2e and Supplementary Fig. 4b clearly demonstrates that the circATXN1 expression level in the cytoplasm was very high. Considering that an elevated transcription of the host gene would result in an upregulation of circRNA expression, we assessed the circATXN1 host transcript ATXN1 mRNA expression levels in response to hypoxic stress. However, hypoxia did not increase the expression of ATXN1, revealing that the rise of circATXN1 was not raised from the change in host transcript but by another regulatory mechanism (Fig. 2f, Supplementary Fig. 4c). According to reports, around 33% of many circRNAs are subject to dynamic regulation by the alternative splicing factor, Quaking (QKI)²⁸. We wondered if QKI was involved in circATXN1 biogenesis, so we infected HCMECs with Ad QKI-sh, and the knockdown efficiency is shown in Fig. 2g. Then, the RT-qPCR results indicated that QKI knockdown significantly reduced circATXN1 abundance in both normoxic and hypoxic circumstances (Fig. 2h), indicating that QKI may contribute to regulating circATXN1 expression. circSMAD2 was used as a positive control downstream of QKI regulation, and circGNB1 as a negative control.

Fig. 2: Characterization of hypoxia-induced circATXN1.

a Genomic loci of circATXN1. b Sanger sequencing confirming circATXN1 splicing in HCMECs. c RT-PCR of circATXN1 and ATXN1 mRNA from c/gDNA with/without RNase R in HCMECs and MCMECs (n = 3 biological replicates). d circATXN1 localization: RNA-FISH in cytoplasmic/nuclear fractions of HCMECs (n = 3 biological replicates, scale bar: 10 µm). e The results of RT-qPCR analysis showed that the relative expression levels of circATXN1 in both the nucleus and cytoplasm indicated that it was mainly located in the cytoplasm (n = 3 biological replicates). f RT-qPCR of ATXN1 mRNA expression in normoxic or hypoxic HCMECs, with β-actin as internal control (n = 4 biological replicates). g Western blot analysis of QKI expression in normoxic or hypoxic HCMECs with or without QKI-shRNA, with β-actin as the internal control (n = 3 biological replicates). h RT-qPCR analysis of circATXN1, circSMAD2, and circGNB1 mRNA expression in normoxic or hypoxic HCMECs with or without QKI-shRNA, with β-actin as the internal control (n = 4 biological replicates). Statistical analysis: unpaired two-tailed Student t test for (f); two-Way ANOVA with multiple comparisons test for (h).

CircATXN1 inhibits hypoxia-stimulated partial EndMT in vitro

Partial EndMT possesses an essential contribution to ischemic diseases and angiogenesis^11,27. Hence, the relevance of circATXN1 to ischemic/hypoxic signaling prompted us to ascertain the connection between circATXN1 and partial EndMT. In partial mesenchymal transition instances, ECs not only retent or partially lose their original characteristics but also gain some mesenchymal traits, leading to increased motility that could be involved in the (patho) physiological growth of blood vessels²⁹. Therefore, we used adenovirus to achieve the intervention of circATXN1 in vitro to clarify the influence of circATXN1 on cell proliferation, migration, and sprouting. The overexpression/knockdown efficiency of circATXN1 and changes in the host gene are shown in Supplementary Fig. 5a–c. The FUCCI system was successfully integrated into the main ECs with knockdown of circATXN1, as well as the control cells. The cell cycle was analyzed with FUCCI imaging. Following culture with hypoxic circumstances for a duration of 24 h, the S/G2-M cells population in the Ad circATXN1-sh group had a significant up-regulation contrasted with the Ad Scr-sh group (Fig. 3a), while the population at the proliferative stage was significantly reduced when circATXN1 was overexpressed (Supplementary Fig. 6a). These facts showed that circATXN1 knockdown could enhance hypoxia-induced endothelial cell proliferation. Moreover, the circATXN1 implication on cell migration during partial EndMT was estimated with transwell assays. The deficiencies of circATXN1 increased cell migration (Fig. 3b), and the overexpression of circATXN1 further reduced the number of transmembrane cells under hypoxic conditions (Supplementary Fig. 6b). Collective migration is the coordinated migration of endothelial tip and stalk cells, which collaborate to facilitate sprouting angiogenesis. Throughout the process of sprouting angiogenesis, angiogenic ECs have the potential to experience a partial EndMT^11,30,31. This transition allows the ECs to acquire mesenchymal properties, which are essential for developing new vascular sprouts. We conducted spheroid sprouting tests to estimate the circATXN1 consequence of this phenomenon. CircATXN1 knockdown reinforced hypoxia-induced sprouting (Fig. 3c), whereas a decrease in total sprout numbers was observed in circATXN1-overexpressed endothelial cells (Supplementary Fig. 6c). In addition, we found that circATXN1 knockdown enhanced cell viability and inhibited endothelial cell tubularization (Fig. 3d, e), while circATXN1 overexpression did the opposite (Supplementary Fig. 6d, e). We conducted matrigel assay to verify the direct angiogenic potential of circATXN1 in physiological angiogenesis models. The results showed that circATXN1 knockdown could enhance matrigel angiogenesis (Fig. 3f), and overexpression of circATXN1 inhibited this process (Supplementary Fig. 6f). Moreover, the circATXN1 knockdown resulted in an upregulation of mesenchymal markers (fibronectin, aSMA, and FSP1) and a downregulation of endothelial indicators (CD31 and VE-cadherin) (Fig. 3g, h). In contrast, the overexpression of circATXN1 suppressed the development of mesenchymal indicators caused by hypoxia and enhanced the expression of the endothelial marker. However, there was partial preservation of the CD31 and VE-cadherin expression (Supplementary Fig. 6g, h). Under normal oxygen circumstances, both the overexpression and knockdown of circATXN1 did not have a significant impact on the expression of these markers. The above indicates that circATXN1 may hamper the acquisition of mesenchymal characteristics in HCMECs during partial EndMT.

Fig. 3: circATXN1 knockdown enhanced Hypoxia-Induced Partial EndMT in HCMECs.

a FUCCI images of HCMECs infected with Ad Scr-sh or Ad circATXN1-sh for 48 h and exposed to hypoxia for 24 h (red for G1, yellow for G1/S, green for S/G2-M) (n = 3 biological replicates, scale bars = 100 µm). b–c Transwell migration and spheroid sprouting assays of HCMECs infected with Ad Scr-sh or Ad circATXN1-sh, with hypoxia stimulation for 24 h (n = 4 biological replicates for b, scale bars = 100 µm. n = 6 biological replicates for c, scale bars = 100 µm). d The viability of HCMECs infected with Ad Scr-sh or Ad circATXN1-sh for 48 h and exposed to hypoxia for 24 h was determined by a CCK-8 assay (n = 8 biological replicates). e Tube formation assays of HCMECs infected with Ad Scr-sh or Ad circATXN1-sh for 48 h and exposed to hypoxia for 24 h (n = 4 biological replicates, scale bar, 100 μm). f Representative HE stained sections showed luminal structures containing erythrocytes in implants receiving HUVECs infected with Ad Scr-sh or Ad circATXN1-sh after 7 days. Scale bar = 100 μm. Quantification of red blood cells filled microvessel density as microvessels/mm2 in shown in the right panel. (n = 4 mice per group). g Immunofluorescence for CD31 (green) and αSMA (red) in HCMECs infected with Ad Scr-sh or Ad circATXN1-sh, with DAPI counterstaining (blue) (n = 8 IF images, scale bars = 20 μm). h Western blot analysis of Fibronectin, VE-cadherin, CD31, αSMA, and FSP1 in HCMECs infected with Ad Scr-sh or Ad circATXN1-sh, with β-actin as the internal control (n = 3 biological replicates). Statistical analysis: unpaired two-tailed Student t test for (b, c, d, e, f).

CircATXN1 knockdown could enhance partial EndMT in vivo

To explore the role of circATXN1 mediated partial EndMT in endothelial cells in ischemic diseases, we used adeno-associated virus (AAV) to knockdown circATXN1. We selected 2 concentrations, 4 × 10¹¹ VG and 8 × 10¹¹ VG, to inject into tail vein of mice to detect AAV knockdown efficiency. Surprisingly, 3 weeks after AAV injection, the mice in the 8 × 10¹¹ VG group began to die, and all of the mice died at about 4 weeks post injection. No significant abnormalities were observed in the mice of 4 × 10¹¹ VG group in appearance. After 3 weeks of AAV injection, we extracted mRNA of mouse hearts for RT-PCR, and it was found that circATXN1 in both 4 × 10¹¹ VG and 8 × 10¹¹ VG groups decreased significantly (Fig. 4a). Compared with mice in non-injected group and low-dose group, the dead mice in the high concentration group showed significant weight loss and ascites formation. It has been reported that highly expressed shRNA could lead to acute liver failure in mice³². Consistent with literature reports, histological staining results also showed extensive focal necrosis in the liver of dead mice in the high concentration group, with abnormal proliferation of multinucleated cells, necrosis and regeneration of liver cells (Fig. 4b), and no abnormalities in other organs. The results of serum AST and ALT in mice also suggested that there was liver injury after AAV injection (Fig. 4c, d), and the liver injury was especially severe in the high concentration group. The results of serum albumin and direct bilirubin showed liver failure in mice with high concentration, but not in mice with low concentration (Fig. 4e, f). In order to avoid the side effects caused by AAV knockdown of high virus dose, we added 1 × 10¹¹ VG and 2 × 10¹¹ VG groups to detect the knockdown efficiency of AAV circATXN1-sh at 1 week, 2 weeks, 4 weeks, 8 weeks and 12 weeks after AAV injection. The results showed that circATXN1 knockdown efficiency in 1 × 10¹¹ VG group was about 50% at 1 week after injection, but the knockdown efficiency was significantly reduced at 2 weeks after injection, and basically recovered at 4 weeks after injection. In the 2 × 10¹¹ VG group, circATXN1 was knocked down to 30–40% at 1 to 4 weeks after injection, recovered to 80–100% at 8 weeks, and returned to normal at 12 weeks. The 4 × 10¹¹ VG group could continuously inhibit circATXN1 expression to 10-20% from 2 to 8 weeks (Fig. 4g). And we chose 2 × 10¹¹ VG as the appropriate dose to avoid possible side effects of long-term intervention with circATXN1. As for AAV-mediated circATXN1 overexpression, we chose the common concentration 2 × 10¹¹ VG, and no abnormalities were observed in mice throughout the experiment. We administered AAV injections 1 week before surgery on mice. By RT-qPCR and IFS, we detected the overexpression and knockdown efficiency of circATXN1 in the ischemic tissue of these two models, and staining results suggested that the intervention effect of AAV was mainly manifested in endothelial cells (Supplementary Fig. 7a-d, Supplementary Fig. 8a-d). The results from laser Doppler ultrasound showed a reduced blood flow recovery of 7 days in the circATXN1 overexpression group after the HLI surgery (Supplementary Fig. 9a), whereas the circATXN1 knockdown group showed improved perfusion recovery (Fig. 4h). Immunohistochemical and IFS revealed that circATXN1 restrained partial EndMT following ischemic injury, which was marked by downregulated mesenchymal marker (Vimentin) expression in circATXN1-overexpressing mice (Supplementary Fig. 9b). On the contrary, circATXN1 knockdown up-regulated the expression of mesenchymal marker (Fig. 4i). We used an MI model to assess the circATXN1 function in ischemic heart disease. Cardiac function was assessed 4 weeks after MI, and then the hearts were collected for histological section preparation and immunostaining. Knockdown of circATXN1 significantly enhanced cardiac function and remodeling after MI (Fig. 4j). In comparison to the AAV Scr-sh group, the AAV circATXN1-sh mice exhibited smaller infarct sizes in the infarct area (Fig. 4k). Remarkably, circATXN1 knockdown significantly increased the overall survival rate after MI in comparison to the control group of mice. The overall survival rates were 90% (18/20) and 65% (13/20), respectively, during 4 weeks of monitoring. (Fig. 4l). As predicted, immunohistochemical results showed that mesenchymal marker was upregulated (Fig. 4m). We injected AAV Control or AAV circATXN1 into the tail vein before MI to test whether circATXN1 overexpression could aggravate the effects. Contrasted with the AAV Control group, the AAV circATXN1 group reported a weaker heart function, a larger infarct size, worser survival rate (with 50% (10/20) and 70% (14/20) survival), and downregulated mesenchymal marker (Supplementary Fig. 9c–f). Collectively, our outcomes demonstrate that knockdown circATXN1 could promote partial EndMT in vivo, improving the recovery of blood flow and function after ischemic injury.

Fig. 4: CircATXN1 Knockdown Enhances Partial EndMT In Vivo.

a RT-qPCR of circATXN1 in the heart of mice with non-injection or mice injection with 4 × 10¹¹ VG and 8 × 10¹¹ VG AAV circATXN1-sh, 3 weeks after AAV injection, with β-actin as the internal control (n = 6 mice per group). b Representative HE stained sections from the liver of mice with non-injection or mice injection with 4 × 10¹¹ VG and 8 × 10¹¹ VG AAV circATXN1-sh, 3 weeks after AAV injection, the black arrow showed multinucleated cells. c–f Serum AST, ALT, albumin and direct bilirubin of mice with non-injection or mice injection with 4 × 10¹¹ VG and 8 × 10¹¹ VG AAV circATXN1-sh, 3 weeks after AAV injection (n = 6 mice per group). g RT-qPCR of circATXN1 in the heart of mice with non-injection or mice injection with 1 × 10¹¹ VG, 2 × 10¹¹ VG and 4 × 10¹¹ VG AAV circATXN1-sh, 1, 2, 4, 8, 12 weeks after AAV injection, with β-actin as the internal control (n = 6 mice per group). h Blood flow recovery assessed by laser Doppler ultrasound in mice administered AAV Scr-sh or AAV circATXN1-sh (n = 4 mice per group). i Immunohistochemistry stain for Vimentin in gastrocnemius muscle sections 7 days post-HLI (n = 6, scale bars = 100 μm). j Echocardiography of left ventricular ejection fraction (LVEF) at baseline and days 7, 14, and 28 post-MI (n = 6 mice per group). k Scar size measurement at 4 weeks post-MI (n = 6 mice per group, scale bar = 2 mm). l Comparison of survival rates between mice treated with AAV Scr-sh and AAV circATXN1-sh after myocardial infarction (n = 20 mice per group). m Immunohistochemistry stain for Vimentin in heart sections from the infarcted border zone 7 days post-MI. Scale bars = 100 μm. Statistical analysis: unpaired two-tailed Student t test for (k); ordinary one-Way ANOVA test for (c, d, e, f); two-Way ANOVA with multiple comparisons test for (h, j); Logrank test for (i).

CircATXN1 hinders partial EndMT by inhibiting SLUG

There are many mechanisms that might induce EndMT, and these pathways converge on various signaling proteins, including SNAIL, TWIST, SLUG, and ZEB1^31,33,34,35. These transcription factors may be stimulated by intracellular signals and ultimately induce EndMT. The concepts of “EndMT” and “partial EndMT” are not independent, and there is a big interlinked nature of similarity. Subsequently, we first screened these 6 transcription factors to study the mechanism of circATXN1 regulating partial EndMT. WB analysis indicated that under hypoxic conditions, SLUG protein levels were negatively correlated with circATXN1 (Fig. 5a and Supplementary Fig. 10a), but the expression of the other 5 transcription factors was not significantly altered. Hughes et al. discovered that SLUG induces EC to undergo a partial EndMT¹². They proposed that this process is contingent on the presence of endothelial SLUG, suggesting that SLUG levels influence the extent to which cells commit to a mesenchymal identity-determining whether cells undergo partial or complete EndMT. Later, we examined whether SLUG knockdown may restore the partial EndMT caused by circATXN1 lack in a hypoxic environment. The function assays demonstrated that after infection with Ad SLUG-sh, proliferation, migration, and the sprouting effect of circATXN1 knockdown on HCMECs significantly decreased (Fig. 5b–d). Furthermore, alterations in the levels of endothelial and mesenchymal markers were found (Fig. 5e). These results prompted us to believe that the reduced partial EndMT and impaired blood flow recovery caused by circATXN1 in ischemic disease are due to poor SLUG expression. Elevating proof reveals that circRNAs have a role in several biological processes in multiple ways, including functioning as regulating transcription, miRNA sponges, being translated into functional peptides or interacting with proteins^18,36. CircATXN1 also inhibited SLUG mRNA expression (Fig. 5f). Luciferase reporters containing the SLUG promoter sequences were constructed to confirm whether circATXN1 could directly regulate SLUG transcription. The result showed that circATXN1 overexpression possessed no significant impact on the reporter plasmid luciferase activity (Fig. 5g), which suggested that circATXN1 does not regulate SLUG expression by affecting SLUG transcriptional activity. Then we knocked down AGO2, the core protein of the miRNA-stimulated silencing complex (miRISC)^37,38,39, and had no effect on circATXN1 mediated SLUG reduction (Fig. 5h), RIP assay also confirmed that AGO2 could not be combined with circATXN1 (circHIPK3 as a positive control) (Fig. 5i), suggesting that circATXN1 might not act as a miRNA sponge to control SLUG expression. As shown in Fig. 5j, analysis of the circATXN1 sequence revealed the presence of two ATG initiation codons, meaning that there may be two encoded peptides, one containing 186 amino acids and the other 640 amino acids. Next, mCherry was added to the front of the termination codons of these two possible encoding peptide segments, respectively, and then the constructed plasmids were transfected into 293 T cells, and finally, the expression was detected with WB. Contrasted with the mCherry group, no bands were shown in the two recombinant plasmids, indicating that circATXN1 does not encode peptides.

Fig. 5: circATXN1 Inhibits Partial EndMT by Suppressing SLUG.

a Western blot of SLUG, SNAIL, ZEB1, ZEB2, and TWIST in HCMECs with Ad Scr-sh or Ad circATXN1-sh for 48 h (n = 3 biological replicates). b FUCCI images of HCMECs infected with Ad Scr-sh, Ad Scr-sh + Ad SLUG-sh, Ad circATXN1-sh, or Ad circATXN1-sh + Ad SLUG-sh, and hypoxia-stimulated for 24 h (n = 3 biological replicates, scale bars = 100 µm). c, d Transwell and spheroid sprouting assays of HCMECs infected as above, with hypoxia stimulation for 24 h (n = 4 biological replicates for c, scale bars = 100 µm. n = 6 biological replicates for d, scale bars = 100 µm). e Western blot of Fibronectin, VE-cadherin, CD31, αSMA, and FSP1 in HCMECs infected and stimulated as above (n = 3 biological replicates). f RT-qPCR of SLUG in HCMECs infected with Ad Control or Ad circATXN1 (n = 3 biological replicates). g Dual luciferase reporter assay of SLUG promoter activity in HCMECs infected with Ad Control or Ad circATXN1 (n = 5 biological replicates). h Western blot SLUG in HCMECs infected with Ad Control + Ad Scr-sh, Ad circATXN1 + Ad Scr-sh, Ad Control + Ad AGO2-sh, or Ad circATXN1 + Ad AGO2-sh (n = 3 biological replicates). i RIP-qPCR analysis of circATXN1 and circHIPK3 in HCMECs with anti-IgG or anti-AGO2 (n = 3 biological replicates). j Western blot of mCherry expression in HEK293T cells transfected with Control, circATXN1-peptide1-mCherry, circATXN1-peptide2-mCherry, or mCherry for 48 h. β-actin served as the internal control (n = 3 biological replicates). Statistical analysis: ordinary one-Way ANOVA test for (c, d); unpaired two-tailed Student t test with Welch’s correction for (f); unpaired two-tailed Student t test for (g); two-Way ANOVA with multiple comparisons test for (i).

CircATXN1 Inhibits Partial EndMT via interaction with ALKBH5

To clarify the mechanism of circATXN1 regulating SLUG expression, we conducted an RNA pulldown assay to search the circATXN1-interacting proteins. This included introducing biotinylated DNA oligomers, either in the sense (S) or antisense (AS) direction, that crossed the back-splicing junction site of circATXN1 into cells. Subsequently, the circATXN1-bound proteins were separated using SDS-PAGE and visualized using silver staining. Figure 6a demonstrates the particular capture of a protein band with a molecular weight ranging from 40 to 55 kDa (shown by a black arrow) using the AS probe and excised for LC-MS/MS analysis (Supplementary Table 2). Among them, ALKBH5, a demethylase that often acts as an eraser for mRNA m6A modification^40,41,42, was of great interest to us (Supplementary Fig. 1l). The association of circATXN1 with ALKBH5 was validated in circATXN1 pulldown samples; the amounts of pulled circRNA and the interacting protein ALKBH5 were both positively correlated with the concentration of antisense probes used for incubation (Fig. 6b). In addition, the association between ALKBH5 and circATXN1 was further validated using RNA immunoprecipitation (RIP) assay employing an anti-ALKBH5 antibody (Fig. 6c). Immunofluorescence assay reported that circATXN1 was localized in the cytoplasm, and ALKBH5 was expressed in both the nucleus and cytoplasm (Fig. 6d). By knocking down ALKBH5, the SLUG-mediated partial EndMT effect induced by down-regulated circATXN1 was effectively weakened (Fig. 6e), which means that circATXN1 regulates the expression of SLUG and downstream effectors through ALKBH5. On this basis, when we supplement the SLUG protein, the impaired proliferation, migration, and sprouting ability of endothelial cells are restored (Fig. 6f and Supplementary Fig. 12a, b). ALKBH5^fl/fl mice were breed with Cdh5-Cre mice to establish ALKBH5 endothelial-specific knockout mice (ALKBH5^CKO mice) (Supplementary Fig. 13). ALKBH5^fl/fl mice and ALKBH5^CKO mice were injected with AAV cricATXN1-sh to knockdown circATXN1, and then a myocardial infarction model was established. The cardiac function of mice was detected by ultrasound at 7, 14, and 28 days after surgery. It was found that when ALKBH5 endothelial specificity was knocked out, circATXN1 inhibition lost its protective effect on cardiac function in mice. Masson staining of heart of sacrificed mice at 28 days after surgery also showed that the reduction of cardiac scar size from circATXN1 knockdown disappeared. The bad effects of ALKBH5-CKO could be improved by SLUG overexpression (Fig. 6g–h). The mouse HLI model also showed that the improved effect of circATXN1 knockdown on blood flow recovery was dependent on ALKBH5, and SLUG supplementation could improve the inhibitory effect of ALKBH5^CKO on blood flow recovery (Supplementary Fig. 14). These results suggest that the diminished partial EndMT and impaired endothelial function caused by circATXN1 in ischemic conditions are attributed to reduced SLUG expression mediated through its interaction with ALKBH5.

Fig. 6: CircATXN1 Inhibits Partial EndMT via interaction with ALKBH5.

a Silver stain showing proteins captured with biotin-labeled sense (S) or antisense (AS) probes targeting circATXN1 junction sites in HCMEC lysates (n = 3 biological replicates). b Western blot (upper panel) and RT-PCR (lower panel) of ALKBH5 and circATXN1 using biotin-labeled S or AS probes in HCMECs (n = 3 biological replicates). c RIP (upper panel) and Western blot (lower panel) showing the interaction between ALKBH5 and circATXN1 in HCMECs transfected with circATXN1 (n = 3 biological replicates). d Representative images of dual RNA-FISH and immunofluorescence (n = 3 biological replicates). e Western blot of Fibronectin, VE-cadherin, CD31, αSMA, FSP1, SLUG, and ALKBH5 in HCMECs infected with various constructs and hypoxia-stimulated for 24 h (n = 3 biological replicates). f FUCCI imaging of HCMECs infected with various constructs and hypoxia-stimulated for 24 h (n = 3 biological replicates, scale bars = 100 µm). g Echocardiography of left ventricular ejection fraction (LVEF) at baseline and days 7, 14, and 28 post-MI (n = 6 mice per group). h Scar size measurement at 4 weeks post-MI (n = 6 mice per group, scale bar = 2 mm) (* compared with AAV Scr-sh / ALKBH5^fl/fl group, # compared with AAV circATXN1-sh / ALKBH5^fl/fl group, & compared with AAV circATXN1-sh / ALKBH5^CKO group). Statistical analysis: two-Way ANOVA with multiple comparisons test for (g); ordinary one-Way ANOVA test for (h).

CircATXN1 reduced SLUG mRNA stability by Preventing ALKBH5 Nuclear Translocation

As shown in Fig. 7a, b and Supplementary Fig. 15a, not only the mRNA expression of SLUG but also the protein level was altered in ALKBH5-intervened cells. Previous results reminded us that circATXN1 does not collaborate with ALKBH5 in the transcriptional regulation of SLUG (Fig. 5g). Actinomycin D was used to evaluate the stability of SLUG mRNA, and ALKBH5 knockdown significantly shortened the half-life of SLUG mRNA (Fig. 7c). Since ALKBH5 often acts as an m6A demethylase, we wondered if ALKBH5 was involved in circATXN1-induced downregulation of SLUG expression in this way. Indeed, as indicated by methylation quantification assay, ALKBH5 knockdown could increase m6A modification of SLUG mRNA (Fig. 7d), while the increased expression of the wild-type ALKBH5 gene decreased the m6A content, the mutant ALKBH5 H204A, which lacks catalytic activity⁴³, did not have the same effect (Supplementary Fig. 15b). Meanwhile, the wild-type ALKBH5 overexpression, but not the mutated ALKBH5 H204A gene, could bring back SLUG expression (Fig. 7b and Supplementary Fig. 15c). This suggests that ALKBH5 primarily influences SLUG expression via its demethylation activity. Besides, immunofluorescence colocalization analysis and nuclear plasma separation experiment found that when circATXN1 expression was down-regulated, ALKBH5 was translocated to the nucleus, whereas when circATXN1 expression was increased, ALKBKH was more localized in the cytoplasm (Fig. 7e, f and Supplementary Fig. 15d, e). Notably, circATXN1 only regulated the space and time change of ALKBH5 but had no effect on its expression (Fig. 6e). It has been reported that ALKBH5 could exert m6A demethylase by binding pre-spliced mRNA⁴⁴. We hypothesized that ALKBH5 might bind to pre-SLUG mRNA after entering the nucleus, instead of mature SLUG mRNA in the cytoplasm. Firstly, we analyzed the localization of pre-SLUG mRNA, which was mainly located in the nucleus (Supplementary Fig. 15f). CircATXN1 also decreased the expression of pre-SLUG mRNA (Supplementary Fig. 15g). RIP assay also confirmed that ALKBH5 could bind pre-SLUG mRNA, but could not bind mature SLUG mRNA (Fig. 7g). To determine whether ALKBH5 binds to pre-SLUG mRNA or mature-SLUG mRNA, we employed AlphaFold2 to predict the structural conformations of ALKBH5 in complex with both the precursor and mature forms of SLUG mRNA (Fig. 7h). The predictions suggested a potential interaction between the protein and both mature and pre-mRNA. Further molecular docking analysis using ChimeraX revealed that pre-SLUG mRNA forms a greater number of potential interactions with ALKBH5, whereas mature-SLUG mRNA establishes fewer potential interactions, with binding sites located exclusively in exonic regions (see Supplementary Table 3). These findings suggest that binding specificity depends more on structural differences than on mRNA sequence. To further characterize the interaction between ALKBH5 and SLUG mRNA, we analyzed the nucleic acid–protein interactions in greater detail. The results showed that, in mature mRNA, only nucleotides located between positions 698–802 and 1870–2016 could interact with the protein, with each nucleotide forming relatively few amino acid contacts. Additionally, binding across the mature mRNA sequence was relatively uniform, with no distinct interaction hotspots, whereas pre-mRNA exhibited more binding sites with a concentrated distribution. (Fig. 7i). To validate this trend, we further employed omiXcore to predict the binding preferences of ALKBH5 for pre-SLUG versus mature-SLUG mRNA. The results demonstrated a stronger propensity for ALKBH5 to bind pre-mRNA, with site-specific binding scores that aligned well with the AlphaFold predictions, thereby corroborating our hypothesis (Supplementary Fig. 15h).

Fig. 7: circATXN1 reduced SLUG mRNA stability by Preventing ALKBH5 Nuclear Translocation.

a RT-qPCR of circATXN1 and SLUG mRNA expression in HCMECs with or without circATXN1 overexpression (n = 4 biological replicates). b Western blot of SLUG and ALKBH5 in HCMECs infected with different constructs (n = 3 biological replicates). c RT-qPCR of SLUG mRNA half-lives in HCMECs with Act D stimulation, and infected with Ad Scr-sh or Ad ALKBH5-sh (n = 3 biological replicates). d MeRIP-qPCR analysis of m6A levels of circATXN1 in HCMECs infected with Ad Scr-sh or Ad ALKBH5-sh (n = 4 biological replicates). e Dual FISH/IF of circATXN1 (red) and ALKBH5 (green) in HCMECs with DAPI staining (blue) (n = 3 biological replicates, scale bar: 10 µm). f Western blot of ALKBH5 in nuclear and cytoplasmic fractions of HCMECs, with GAPDH and LMNB as controls (n = 3 biological replicates). g RIP-qPCR of pre- and mature SLUG mRNA in HCMECs using anti-IgG or anti-ALKBH5 (n = 3 biological replicates). h Structural predictions of ALKBH5 binding to Mature-SLUG mRNA (top) and Pre-SLUG mRNA (bottom). ALKBH5 is shown in blue, while SLUG mRNA is represented in pink. Magnified views highlight the binding interface, showing that pre-mRNA forms more extensive interactions with ALKBH5 compared to mature mRNA. i The distribution of interaction sites along the mRNA sequence for both mature (cyan) and pre-mRNA (yellow). The curves represent the relative binding affinity, while the bars indicate the number of interactions at each position. j Phosphorylated tyrosine residues of ALKBH5 protein in HCMECs with or without circATXN1 knockdown (n = 3 biological replicates). k Phosphorylated tyrosine residues of ALKBH5 protein in HCMECs with different constructs (n = 3 biological replicates). l Immunoprecipitation of ALKBH5 and CRM1 in HCMECs with or without circATXN1 knockdown (n = 3 biological replicates). m Western blot of ALKBH5 protein expression in nuclear and cytoplasmic extracts of HCMECs infected with different constructs (n = 3 biological replicates). Statistical analysis: two-Way ANOVA with multiple comparisons test for (a, c, g); unpaired two-tailed Student t test for (d).

Studies have shown that the nuclear translocation of ALKBH5 is associated with O-GlcNAcylation modification⁴⁵. However, we found that knockdown of circATXN1 had no significant effect on cellular O-GlcNAcylation levels (Supplementary Fig. 15i), suggesting that the effect of circATXN1 on ALKBH5 subcellular localization might not be dependent on glycosylation. Previous reports indicate that EGFR and LATS2 can phosphorylate ALKBH5, inhibiting its interaction with the nuclear export protein CRM1 and thereby promoting its nuclear retention^46,47. We observed that circATXN1 knockdown enhanced ALKBH5 phosphorylation (Fig. 7j). Further experiments showed that knockdown of EGFR significantly reduced ALKBH5 phosphorylation induced by circATXN1 deficiency, whereas knockdown of LATS2 had no such effect (Fig. 7k). This suggests that circATXN1 knockdown may release ALKBH5, making it more susceptible to phosphorylation via the EGFR pathway. Given the crucial role of CRM1 in promoting ALKBH5 subcellular localization^46,47, we investigated whether circATXN1 intervention affects the interaction between CRM1 and ALKBH5. Co-immunoprecipitation (IP) experiments demonstrated that circATXN1 overexpression enhanced ALKBH5-CRM1 binding, while circATXN1 knockdown inhibited this interaction (Fig. 7l and Supplementary Fig. 15j). Additionally, cytoplasmic fractionation assays confirmed that CRM1 overexpression significantly reduced ALKBH5 nuclear retention induced by circATXN1 knockdown (Fig. 7m). These findings suggest that circATXN1 directly binds to ALKBH5, reducing its phosphorylation and thereby promoting its interaction with the nuclear export protein CRM1. This prevents ALKBH5 nuclear retention, allowing it to exert its demethylation function to destabilize pre-SLUG mRNA, ultimately impeding the partial EndMT process.

Low dose of TGF-β (1–2 ng/mL) could induce valva ECs to resemble partial EndMT transformation, showing that interstitial properties are acquired while endothelial cell properties are retained or partially lost⁴⁸. We examined the role of circATXN1/ALKBH5 axis in this model. IFS and Western blot results confirmed that circATXN1/ALKBH5 axis could also affect partial EndMT in valva ECs. CircATXN1 knockdown could increase the expression of mesenchymal marker and inhibit the expression of endothelial cell marker. On this basis, ALKBH5 knockdown can reverse the effect of circATXN1 knockdown (Supplementary Fig. 16a, b), suggesting that the regulation of circATXN1/ALKBH5 axis on partial EndMT may be universal in partial EndMT. Valproic acid (VPA) is an HDAC inhibitor that can affect gene expression by regulating RNA modification and has been reported to induce EndMT, inhibit cell activity, and inhibit angiogenesis, similar to complete EndMT⁴⁹. We stimulated EC with VPA and interfered with circATXN1/ALKBH5 axis. The results showed that after intervention with circATXN1/ALKBH5, VPA had no effect on the inhibition of cell activity and angiogenesis (Supplementary Fig. 16c, d), suggesting that circATXN1/ALKBH5 may have a specific effect on partial EndMT.

Therapeutic and long-term effect of circATXN1 knockdown for tissue recovery following ischemia

To explore the therapeutic and long-term effect of circATXN1 knockdown on tissue ischemia, we injected AAV 2 × 10¹¹ VG into the tail vein of myocardial infarction mice 1 day after surgery, and measured the cardiac function of mice at 4 and 12 weeks after injection, respectively. Injections of AAV circATXN1-sh could improve heart function (Fig. 8a). The mice were sacrificed 12 weeks after injection and the hearts were taken for Masson staining. Compared with the mice in the non-injection group, the degree of fibrosis in the hearts of the circATXN1 knockdown mice was significantly reduced 12 weeks after operation (Fig. 8b). Considering our previous findings that AAV circATXN1-sh could cause liver damage, we measured serum ALT, AST, albumin, and direct bilirubin at 4 and 12 weeks after injection, and no significant abnormalities were observed (Fig. 8c–f). Sustained knockdown of circATXN1 might have unintended consequences, and further studies were needed to evaluate whether long-term inhibition of circATXN1 lead to fibrosis. Masson staining was performed on parenchymatous organs such as lungs, kidneys, spleen, liver, and gastrocnemius muscle, and no significant fibrotic changes were observed (Fig. 8g). In addition, we locally injected AAV circATXN1-sh, 2 × 10¹¹ VG, into the muscle of lower limb ischemia mice, and conducted ultrasonic Doppler detection at 2 and 12 weeks after injection. Compared with non-injected mice, blood flow of mice in the AAV group was improved at 2 weeks after injection, but the lower limb blood flow of mice in the 2 groups almost returned to normal at 12 weeks (Fig. 8h). No difference was seen, possibly due to the strong recovery function of the mice. Besides, the degree of fibrosis in the gastrocnemius was also reduced at 12 weeks after AAV injection (Fig. 8i). These results suggest that intervention with circATXN1 after ischemia is safe and effective for potential clinical use.

Fig. 8: Therapeutic effect and long-term effect of circATXN1 knockdown for tissue recovery following ischemia.

a Echocardiography of left ventricular ejection fraction (LVEF) before injection, 4 weeks and 12 weeks post-injection (n = 6 mice per group). b Representative Masson stained sections from heart of non-injected MI mice and MI mice injected with AAV circATXN1-sh after 12 weeks. c–f Serum AST, ALT, albumin and direct bilirubin of mice with non-injection or mice injection wit 2 × 10¹¹ VG AAV circATXN1-sh, 12 weeks after AAV injection (n = 6 mice per group). g Representative Masson stained sections from lung, kidney, spleen, liver, and gastrocnemius muscle of non-injected MI mice and MI mice injected with AAV circATXN1-sh after 12 weeks (n = 6 mice per group, scale bar = 200 μm). h Blood flow recovery assessed by laser Doppler ultrasound in non-injected MI mice and MI mice injected with AAV circATXN1-sh after 12 weeks (n = 4 mice per group). i Representative Masson stained sections from gastrocnemius muscle of non-injected MI mice and MI mice injected with AAV circATXN1-sh after 12 weeks (n = 4 mice per group, scale bar = 200 μm). Statistical analysis: two-Way ANOVA with multiple comparisons test for (a, h); ordinary one-Way ANOVA test for (c, d, e, f).

Discussion

We conducted a sequencing analysis on endothelial cells with hypoxia and serum deprivation to discover molecules that are linked to the partial or temporary activation of mesenchymal cells in ECs following tissue ischemia/hypoxia. The findings revealed a significant rise in the circATXN1 expression. Considering circRNA’s covalently closed loop structure, which is highly stable and resistant to exonuclease, detecting circRNA expression through noninvasive sampling in body fluids could serve as a reliable biomarker, making it useful for routine clinical testing^20,36,50,51. We observed that circATXN1 expression is elevated in the peripheral blood of patients with myocardial infarction and lower limb ischemia, suggesting its potential as a biomarker to predict disease severity and prognosis. Biomarkers like soluble CD40 ligand and MCP-1, which increase in ischemic disease patients, are associated with advanced stages of atherosclerotic cerebrovascular disease and higher risk of cardiovascular events⁵². Given its unique structure, circATXN1 may serve as a superior biomarker for ischemic diseases. A partial mesenchymal transition can potentially contribute to the growth of blood vessels in (patho) physiological conditions by promoting a pro-migratory and -invasive phenotype in endothelial cells^7,13,14,53. Endothelial tip cells, which guide the formation of new sprouts, can also be considered to have undergone a partial EndMT^11,16,54. This transition involves the loss of apical-basal polarity, separation from the endothelium of established blood vessels, and guiding the invading sprout. Therefore, by overexpressing circATXN1, we examined the proliferation, migration, and sprouting phenotypes of endothelial cells, and the outcomes revealed that circATXN1 hampered the acquisition of mesenchymal characteristics in HCMECs during partial EndMT. In contrast, circATXN1 knockdown raised the expression of the mesenchymal marker and mitigated but partially preserved the expression of endothelial markers, which drove ECs toward a partial EndMT. Subsequently, to ensure knocking down efficiency and avoid unpredictable side effects, we have used different concentrations of AAV circATXN1-sh, and finally selected the appropriate concentration to carry out in vivo experiments. Outcomes suggested that circATXN1 knockdown had an improved effect on angiogenesis post-ischemia and function recovery, while circATXN1 overexpression inhibited the above effects.

Then, we examined the effects of circATXN1 on several specific transcription factors recognized to be involved in EndMT regulation, and results showed that, compared with other transcription factors, cirATXN1 only affected SLUG expression and was negatively correlated. Nevertheless, the luciferase reporter experiment demonstrated that circATXN1 did not impact the SLUG promoter activity, suggesting that circATXN1 did not directly regulate SLUG transcription. CircATXN1 also did not affect ATXN1 expression, largely avoiding other potential regulatory pathways mediated by the host gene. Since circRNAs regulate various biological processes in different ways, and here other possible mechanisms of action (such as miRNA sponges, translating into functional peptides) were excluded one by one through RNA pulldown assays and LC-MS/MS analysis, we ultimately identified a protein that interacts with circATXN1 to mediate the regulation of SLUG, which was ALKBH5. The interaction was further confirmed by RIP assays.

Since ALKBH5 usually acts as an eraser for mRNA m6A modification, we hypothesized that it may also regulate SLUG mRNA expression in this way^55,56. With the addition of actinomycin D, which interfered with mRNA synthesis, ALKBH5 could indeed weaken the m6A modification of SLUG mRNA, thereby consolidating the mRNA stability and increasing the mRNA expression, while the reported catalytic inactive mutant ALKBH5 H204A abolished this function. We conducted immunofluorescence co-localization analysis and nuclear plasma separation experiment to explore further how circATXN1 regulates ALKBH-mediated SLUG mRNA demethylation, and the results proved that circATXN1 precisely regulated the space and time change of ALKBH5, but had no effect on its expression. More specifically, we found that circATXN1 directly binds to ALKBH5, reducing its phosphorylation and thereby promoting its interaction with the nuclear export protein CRM1. This prevents ALKBH5 nuclear retention, allowing it to exert its demethylation function to destabilize pre-SLUG mRNA, ultimately impeding the partial EndMT process. Additionally, we observed that circATXN1 knockdown increased ALKBH5 phosphorylation, and further mechanistic investigations revealed that this effect was significantly reduced by EGFR knockdown but not by LATS2 knockdown, suggesting that circATXN1 may regulate ALKBH5 subcellular localization via the EGFR pathway. Thus, when the expression of circATXN1 was downregulated, more ALKBH5 entered the nucleus. By knocking down ALKBH5, the SLUG-mediated partial EndMT effect induced by down-regulated circATXN1 is effectively weakened, and if we supplement SLUG protein, the impaired angiogenic abilities (such as proliferation, migration, and sprouting) of endothelial cells were restored. Small interfering RNAs inhibit ncRNA expression through complementary base pairs. Targeting ncRNAs using RNAi techniques is particularly advantageous because ncRNAs are difficult to target using standard small compounds and antibody drugs. The FDA has approved several RNAi drugs, including Patisira, Givosira, Lumasiran, Lumasiran, Vutrisiran, and Zilebesiran, demonstrating the viability of the RNAi approach. We explored various concentrations of AAV circATXN1-sh and finally determined the safe concentration. It is safe and effective to knockdown circATXN1 in time after ischemia. The translational potential of our findings is significant.

Ischemic diseases such as ischemic heart disease, lower limb ischemia, and ischemic cerebrovascular diseases are closely linked to aging^57,58,59, which impairs endothelial function and, consequently, vascular health. Improving endothelial cell function is a critical target for treating cardiovascular diseases⁶⁰, and understanding whether circATXN1 contributes to age-related endothelial dysfunction will be a focus of our future studies. Additionally, metabolic disorders, including atherosclerosis, obesity, non-alcoholic fatty liver disease (NAFLD), and diabetes, often exhibit early endothelial dysfunction⁶¹, which subsequently leads to vascular abnormalities. Early intervention in lipid metabolism has been shown to improve outcomes in ischemic diseases⁶², and our mouse model suggests that early intervention with circATXN1 could improve prognosis. Exploring the potential of circATXN1 in early metabolic disease intervention is of considerable interest to us.

Methods

Animal experiments were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee of Shouzhengpharma Biotechnology Co., Ltd., which approval all animal study protocols. All mice were housed under a controlled 12-h light-dark diurnal cycle with regulated temperature (20 °C–25 °C) and relative humidity (40% to 60%).

All human studies were approved by the Ethics Committee of The First Affiliated Hospital of Zhengzhou University, and written informed consent was obtained from all participants before enrollment. All procedures were conducted in accordance with the ethical principles outlined in the Declaration of Helsinki.

Cell culture

The HEK293T cells were purchased from Zhongqiaoxinzhou Biotechnology (Shanghai, China) and cultivated in high-glucose DMEM enriched with 10% FBS and 100 U/mL penicillin-streptomycin (P/S). Human cardiac microvascular endothelial cells (HCMECs) at passage 3, sourced from Lonza Bioscience (Basel, Switzerland), were cultivated in endothelial culture media (ECM; ScienCell, Carlsbad, CA, USA). Primary mouse CMECs (MCMECs) were obtained from the left ventricles of C57BL/6 J mice using a previously established method^63,64. After the epicardial, endocardial, and coronary arteries were removed, the left ventricular tissue was chopped and digested with liberase. The cell suspension was incubated for 30 min at 4 °C with microbeads conjugated to anti-CD31 antibody (ab7388, Abcam, UK; Thermo Fisher Scientific, USA) under gentle rotation. The first step was the isolation of primary MCMECs using a magnetic separator. These cells were subsequently cultivated on fibronectin-coated plates using a complete ECM (ScienCell, USA). Following the removal of serum, the cells were cultivated in hypoxic conditions consisting of 5% CO₂, 1% O₂, and the remaining balance of N₂ for a period of 24 h, and the cells in the control group were kept in the normoxia mixture (5% CO₂ and 21% O₂), both the hypoxic cells and normoxic cells were derived from the same maternal generation. HUVECs were isolated from umbilical veins as previously described and cultured using ECM⁶⁵. Primary HUVECs were cultured in an incubator with a temperature of 37 °C and a CO2 concentration of 5%. The experiments used HUVECs between passages 2 and 6. Valve endothelial cells (Valve ECs) were isolated from aortic valve leaflets with relatively normal echocardiographic features collected during Bentall procedures using the collagenase I digestion method as previously described⁶⁶. Briefly, normal aortic valve leaflets were digested with collagenase I (2 mg/ml) for 15 min, Valve ECs on the valve surface were gently scraped into a petri dish using a sterile cotton swab and cultured in a specialized medium (EGM-2) supplemented with 2% FBS, 100 μg/mL streptomycin and 100 U/mL penicillin. TGF-β (Cat. No.:CA59; Novoprotein, Shanghai, China) was used to stimulate ECs.

Recombinant adenovirus and adeno-linked virus generation

In vitro experiments, recombinant adenovirus was used to infect endothelial cells to interfere with cirAXTN1, SLUG, and ALKBH5. Recombinant adeno-associated virus (AAV) type 9 systems, including vectors with control/circATXN1 driven by an endothelial-specific promoter (Tie promoter) or scramble shRNA/circATXN1 shRNA, were administrated into the mice tail veins to modify circATXN1 levels in the heart. The AAV was administered to the mice via tail vein injection.

Mice lower limb ischemia model

The protocol for the hindlimb ischemia model was in accordance with our previously published literature⁶⁵. Briefly, eight-week-old male C57BL/6 J mice were used to establish the hindlimb ischemia (HLI) model. After anesthesia with inhaled 2% isoflurane, mice were subjected to a femoral artery surgery in the right hind limb. The hair of the hind-limb was removed. Fixed in the supine position, the mouse was prepared to undergo a left hind-limb ischemia surgery. The surgery started with a ∼10 mm incision in the groin to expose the femoral artery, then the femoral vein and nerve were separated from the femoral artery. After that, a ligation proximal to the outlet of the profundal femoris artery was made, and another was made proximal to the outlet of the saphenous artery. Finally, the femoral artery segment between two ligations was excised, and the surgical wound was then closed by using 4-0 prolene sutures. All the veins and nerves were carefully cared for and were not damaged during the operation. The blood flow was calculated by laser Doppler scanning system (Perimed, Sweden) on postoperative days 1, 3, 7, and 14. The perfusion ratio was quantified from the ischemic limb (right) and the control limb (left), which reflected the recovery of blood flow. The mice were sacrificed at a certain point in time, the gastrocnemius was excised, paraffin embedding or −80 °C preservation. Immunostaining, RT-qPCR analysis and other follow-up experiments were performed

Mice myocardial infarction model

8-week-old male C57BL/6 J mice were used for myocardial infarction model following the procedure as previously reported⁶⁷. Briefly, after anesthesia with inhaled 2% isoflurane, the mice received a thoracotomy with the assistance of a small animal respirator. Under the microscope, the left anterior descending coronary artery was ligation with 10–0 nylon wire, and the color of the surrounding myocardial tissue turned immediately pale. Mice in the sham group received all the surgical procedures induced by MI except the ligation step. For gain and loss of function experiments, the mice were administered with AAV virus via the tail vein 1 week before MI surgery. Left ventricular ejection fraction was measured on pre- and postoperative days 7, 14, and 28 by a small animal ultrasound instrument.

RT-qPCR

Aligning with manufacturer recommendations, the total RNA was extracted via the SPARK easy Improved Tissue/Cell RNA Kit (Shandong Sparkjade Biotechnology Co., Ltd, Jinan, Shandong, China). The PrimeScriptTM RT reagent Kit (Takara, Japan) was implemented to convert 1 μg of total RNA into cDNA with reverse transcription, and random hexamer primers were used specifically for the reverse transcription of circRNAs. If RNase R pretreatment is required, 5 μg of total RNAs and 10 units of RNase R were incubated at 37 °C for 15 min. The qPCR experiment was conducted with the SYBR Green master mix (Vazyme, China) and examined using the QuantStudio 6 Operating Software (Life Technologies, USA). Primer sequences are shown in the Supplementary Table 4.

RNA sequencing for circRNA

Total RNA was obtained to find differentially expressed circRNAs in HCMECs under normoxic and hypoxic conditions. rRNA Depletion and VAHTS Universal V6 RNA-seq Library Prep Kits (Vazyme, China) were utilized to produce circRNA-sequencing libraries based on the manufacturer’s recommendations. The CIRC2 algorithm was employed to identify circRNAs, which were then annotated by comparing sequences in circBase. circRNAs with a log fold change greater than 1 and a p-value less than 0.05 between the two conditions were significantly differentially expressed.

Study population

Patients with acute myocardial infarction (AMI) and lower limb ischemia, were compared with healthy volunteers. All participants were recruited from the First Affiliated Hospital of Zhengzhou University. The control group consisted of healthy volunteers. These volunteers had no significant systemic diseases, such as ischemic heart disease, cancer, pulmonary disease, or infectious diseases. Patients diagnosed with acute myocardial infarction (AMI) were included in the study. Blood samples were collected prior to any surgical intervention, primarily on the first day of admission. Patients with clinically diagnosed lower limb ischemia were also included. Blood samples were taken upon admission and prior to any surgical or interventional treatments. For all groups, blood samples were collected under standard clinical conditions. Clinical characteristics, vital signs, laboratory test results, and radiological reports were extracted from electronic medical records to complement the blood sample data. A total of 32 healthy volunteers, 32 acute myocardial infarction patients and 32 nondiabetic patients with a car accident were recruited in this study as well. Clinical characteristics, vital sign, laboratory tests, and radiological reports were extracted from electronic medical records between January 2023 and March 2024. All participants provided written informed consent prior to enrolment in the study. In cases where participants were unable to provide consent directly, consent was obtained from their Legally Authorised Representatives. The main clinical and laboratory data of human subjects were summarized in Supplementary Table 5.

The mRNA half-life determination

The cells were treated with 5 µM actinomycin D (Sigma, USA) and afterward digested at specified time intervals for the purpose of RNA extraction. At each time point, the initial level (0 h) was implemented to adjust the remaining RNA levels. The decay kinetics of mRNA were analyzed by GraphPad Prism software, and the result was rendered with a one-phase exponential decay curve.

Cytosolic/nuclear fractionation

Typically, 1 × 10⁷ cells were gathered with trypsin-EDTA buffer. After washing and centrifugation 2-3 times, the cell pellets were operated according to the instructions of the NE-PER Nuclear and Cytoplasmic Extraction Reagents step by step. The obtained nuclear/cytosolic extracts were used for subsequent Western blot (WB) analysis. As references for the cytoplasmic and nuclear fractions, GAPDH and LMNB were chosen, respectively.

RNA fluorescence in situ hybridization (RNA-FISH)

A biotin-labeled oligonucleotide probe was developed in TSINGKE, either antisense (AS) or sense (S), that specifically targets the circ-ATXN1 junction. The biotin RNA Labeling Mix and T7 RNA polymerase (Roche, Germany) were employed to generate the probes for 18S and U6 through in vitro transcription of PCR fragments. The hybridization occurred inside a humidified chamber at 37 °C for 16 h, either with or without treatment with RNase R at a concentration of 3 U/mg. The Fluorescent In Situ Hybridization kit (RiboBio, Guangzhou, China) was implemented for signal detection, whereas 4′,6-diamidino-2-phenylindole (DAPI, Sigma) was implemented for nuclei counterstaining. Primer sequences are shown in the Supplementary Table 4.

Immunofluorescence assays

HCMECs were planted on the cell slide to a suitable density and treated with 4% paraformaldehyde for 15 min to fix them. The cell specimens were afterward rinsed three rounds with pre-chilled PBS and subjected to 0.1% Triton X-100 at ambient temperature for 15 min to allow for permeabilization. Following a 1-2 h blocking with a 1% BSA solution at ambient temperature, the slides were treated with primary antibodies and kept in a humidified container overnight at 4 °C. After rinsing the cell slides with PBS three rounds for 5 min each. They were treated with fluorescent dye-labeled secondary antibodies for 1 h at ambient temperature in the darkness. Afterward, the slides were incubated with DAPI for 15 min at ambient temperature without exposure to light. For immunofluorescence staining (IFS) of tissues, the tissues need to be fixed, paraffin-embedded, and sectioned. Sections were successively deparaffinized, rehydrated, and treated in a pressure cooker with sodium citrate antigen repair solution. The subsequent steps, including blocking, primary and secondary antibody incubation, and DAPI staining, refer to the procedures described above. A fluorescence microscope (Zeiss, Germany) was used to visualize and acquire images, and ImageJ software was deployed to analyze images.

Immunohistochemistry

Immunohistochemical staining was conducted as earlier established⁶⁸. Tissue samples were embedded in paraffin and sectioned, Vimentin immunohistochemical staining was performed. Photographs were taken using an optical microscope (Olympus, Japan).

Western blot analysis

The cells were gathered and disrupted with RIPA buffer containing a protease inhibitor. Typically, a quantity of 20 μg of protein was isolated with a 10% SDS-PAGE method and then electro-transferred onto PVDF membranes (Millipore, USA). The membranes were obstructed with a 5% solution of skim milk and thereafter incubated overnight at 4 °C with primary antibodies. Following the washing step with TBS-T solution, HRP-conjugated secondary antibodies were introduced and incubated at ambient temperature for 2 h. The blots were processed with Super ECL plus reagents (Vazyme, China) and were displayed using automated chemiluminescence image analysis equipment (Tanon 5200, China). Antibodies used are shown in the Supplementary Table 6.

Migration assays

After 24 h infection, HCMECs were enzymatically broken down and resuspended in serum-free ECM. A suspension containing 1 × 10⁴ cells was placed into the upper chambers, while a full medium with 10%FBS was introduced to a 24-well plate lower chambers. The entire Transwell® device (Corning, USA) was incubated in the cell incubator for 24 h, and the upper chamber was successively transferred to 4% paraformaldehyde for fixation and 0.1% crystal violet for staining. Next, a cotton swab was utilized to delicately cleanse the non-migratory cells located inside the top cavity, while the migrated cells were left intact on the underside of the polycarbonate membrane. The migrating cells were captured with microscopy and quantified.

FUCCI reporter cell line and Proliferation assays

The plasmid pBOB-EF1-FastFUCCI-Puro (RRID: Addgene 86849) was employed to create the cell cycle reporter (FastFUCCI reporter). This plasmid encodes two fluorescent probes (mKO2-hCDT1 and mAG-hGEM) that are used to distinguish different phases of the cell cycle. The FastFUCCI reporter was stably transduced into endothelial cells by lentivirus to form the FUCCI reporter cell line. The indicator cells exhibited red-, yellow-, and green-emitting populations, representing cells in the G1, G1/S-, and S- or G2-to-M phases, respectively. Through the gain- or loss-of-function experiments on the FUCCI reporter cell line, we could explore the effect of targeted gene(s) on cell cycle and cell proliferation.

Spheroid sprouting assay

HCMECs were seeded in a round-bottom 96-well plate and cultured with a complete medium comprising 0.25% (w/v) methylcellulose for 24 h using hanging drops to create spheroids. The spheroids were embedded in a mixed collagen solution (formula as described previously), and the volume of the solution was calculated based on the number of spheroids, resulting in an approximate concentration of 50 spheroids/ml. The mixture was then introduced to the 24-well plate (1 ml/well) and polarized for 30 min. After incubation with ECM containing 50 ng/ml VEGF for 24 h, the collagen with spheroids was fixed with 4% paraformaldehyde, photographed, and analyzed.

CCK-8 Assay

In each well of a 96-well plate, a seeding of 1.2 × 10⁴ HCMECs was carried out. The HCMECs were treated with the indicated condition. Following the 24-h treatment, the cell culture medium was aspirated, and 90 µl of cell culture medium was added to every well. Subsequently, each well received 10 µl of the CCK-8 reagent. The plates were then incubated for 2 h to allow for proper colorimetric reaction. Absorbance at 450 nm was assessed with a microplate reader.

Tube formation assay

For the tube formation assay, each well of 96-well plates was first precoated with 50 μl of Matrigel (356234, Corning). Next, HCMECs were infected with the indicated adenovirus in a six-well plate for 24 h. Then, these HCMECs were detached from the six-well plate, mixed with human VEGF (10 ng/ml, Peprotech), and suspended into each well of the Matrigel-precoated 96-well plate at a density of 1 × 10⁴ cells/well. The plates were then incubated at 37 °C for 4 h. Tube formation was observed under a microscope (Olympus). And the number of total length was measured by using NIH ImageJ software.

Matrigel plug assay

500 μl Matrigel containing cells that were pre-infected with indicated adenovirus, heparin (60 U, Sigma-Aldrich) and bFGF (150 ng/ml, Peprotech) were subcutaneously injected into the groin of 8-week male C57BL/6 J mice. Seven days later, the mice were euthanized, and the implants were harvested. Then, Matrigel plugs were embedded in paraffin for hematoxylin and eosin (H&E) staining.

Dual-luciferase reporter assay

The genomic DNA was implemented to amplify the SLUG promoter region (–2089/  +  235) while the One Step Cloning Kit (C112-02, Vazyme Biotech Ltd., China) was deployed to insert the pGL3-Basic vector (Promega). The pRL-TK plasmid (Renilla luciferase reporter plasmid, Promega), which operates as an internal control, was co-transfected into HEK293T cells along with the Luciferase reporter structures. The Dual-Luciferase Reporter Assay Kit (Promega) was implemented to ascertain the luciferase activity, aligning with the manufacturer’s recommendations.

Pulldown assay with Biotin-labeled RNA probes and mass spectrometry

Typically, 3 μg of biotin-labeled sense and antisense RNA probes were incubated at 4 °C for 1 h with streptavidin magnetic beads. Subsequent to a first mild wash with the buffer solution (20 mM Tri-HCl, pH7.5, 1 mM EDTA, and 450 mM NaCl), the beads were then exposed to cell lysates at 4 °C for an interval of 4 h. This was afterward followed by five further washes. Following that, the binding proteins were subjected to immunoblot analysis or mass spectrometry (Bioprofile Technology Co., Ltd, Shanghai, China). Probe sequences are shown in the Supplementary Table 4.

RNA-IP assay

The cells were cross-linked using UV radiation at a dosage of 200 J/cm² and a wavelength of 254 nm. Next, the cells were lysed with a solution comprising protease inhibitors and RNase inhibitors. The resulting mixture was then homogenized and centrifuged. The supernatant, was afterward incubated with anti-ALKBH5 and rabbit IgG for a duration of 4 h with gentle rotation. Next, Protein A/G Agarose beads (Millipore, USA) were introduced into the mixes and left to rotate for an additional 1 h. The immunoprecipitated RNA was purified using Trizol Reagent after being washed three times. It was then subjected to RT-qPCR analysis using primers. All processes were conducted on ice or at a temperature of 4 °C.

RIP-qPCR assay

RNA immunoprecipitation was performed with the Magna RIP™ RNA-Binding Protein Immunoprecipitation Kit (Millipore) according to the manufacturer’s instructions. Briefly, magnetic beads coated with 5 μg of antibodies against IgG, AGO2 or ALKBH5 were incubated with pre-frozen cell lysates overnight at 4 °C. Associated RNA-protein complexes were collected and washed 6 times and then subjected to proteinase K digestion and RNA extraction by TRIzol. The relative interaction between protein and RNA was determined by qPCR and normalized to input.

MeRIP-qPCR assay

The MeRIP test was derived from a previously published methodology⁴⁴. To summarize, the undamaged poly-A-purified RNA was denatured at 70 °C for 10 min. Subsequently, the specimen was rapidly chilled with ice and mixed with m6A antibody in a 1 ml solution comprising RNasin Plus RNase inhibitor 400 U (Promega, USA), 50 mM Tris-HCl, 750 mM NaCl, and 0.5% (vol/vol) Igepal CA-630 (Sigma Aldrich, USA). The combination was incubated for 2 h at 4 °C. The mixture was incubated at 4 °C with rotation for 2 h, following the Dynabeads Protein G (Invitrogen, USA) were purified and introduced to it. The m6A RNA was eluted two times using a solution containing 6.7 mM N6-methyladenosine 5′-monophosphate sodium salt. The rinses were conducted at 4 °C for 1 h. Afterward, it was precipitated with 5 μg of glycogen, one-tenth of the amount of 3 M sodium acetate, and 2.5 times the volume of 100% ethanol. The precipitation procedure was conducted overnight at –80 °C. The m6A enrichment degree was evaluated using qPCR analysis. The fragmented mRNA was immediately incubated with a solution comprising m6A antibodies and subjected to comparable treatment. Primer sequences are shown in the Supplementary Table 4.

Extraction of exosomes

Extraction of exosomes was performed as described previously with minor modifications⁶⁹. Briefly, Cellular components in fresh blood samples were eliminated with centrifugation (two times 2500 × g, 4 °C, 15 min). Supernatant (i.e., platelet-free plasma) was diluted to 2× with phosphate-buffered saline (PBS), and was filtered through a 0.8 μm filter (Merck, Darmstadt, Germany) by hydrostatic pressure to remove remaining platelets and apoptotic bodies. Then, the samples were centrifuged at 13,200 × g for 22 min at 4 °C. The supernatant was filtered through 0.22 μm filters, and exosomes were pelleted by ultracentrifugation at 120,000 × g for 1 h, 3 h, 6 h, or 14 h at 4 °C. Additional conditions included ultracentrifugation at 37 °C with protease inhibitors or using 5× diluted plasma. The pellets were washed with PBS and re-centrifuged at 120,000 × g. Exosomes were used fresh to maintain integrity.

Protein-mRNA interaction site prediction

Protein-mRNA interaction sites were predicted using AlphaFold2⁷⁰ (https://alphafoldserver.com/). Since AlphaFold does not directly provide binding site information, we performed post-processing analysis on the CIF files generated by AlphaFold. For structural modeling, UCSF ChimeraX (v 1.9)⁷¹ was used. Protein-mRNA interactions were analyzed using the Structure Analysis tool under Tools > Structure Analysis > Contacts, with the following parameters: 1). Van der Waals overlap threshold ≤ −0.40. 2). Center-to-center distance ≤ 4.00. 3). At least one selected atomic pair for filtering. 4). Ignoring interactions between atoms separated by four or fewer bonds. 5). Ignoring interactions between residues separated by five or fewer positions. 6). Including interactions across models, within the same model, the same molecule, and within the same residue. 7). Ignoring hidden models. The interacting residues and atoms were visualized, with interactions displayed as blue pseudo bonds of 0.075 radius. The output included all intermolecular interactions, and specific interactions between the A-B strands were further analyzed. The complete interaction dataset and the filtered results were compiled in the Supplementary Table 3. To further analyze the binding sites, we used R (ggplot2) for statistical analysis and visualization of protein-mRNA interactions. Additionally, interaction propensity scores corresponding to RNA positions were downloaded from omiXcore⁷², followed by further analysis and visualization.

Statistical analysis

The data were presented as the mean value ± the standard error of the mean (s.e.m.). The cutoff values for gene expression were determined based on the median or mean levels. We used Shapiro-Wilk test to evaluate data normality. The differences were evaluated using Student’s t-test, analysis of variance (ANOVA). For non-normally distributed data across multiple groups, we employed the Kruskal-Wallis test. The expression correlation was assessed with the Pearson correlation coefficient analysis. The log-rank test was deployed to evaluate the disparity in survival. The two-sided t-test was implemented as a regular test method, and the p-values below 0.05 were deemed significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.