To investigate the role of neuronal ferroptosis in cerebral ischemia/reperfusion injury, we established a mouse transient middle cerebral artery occlusion (tMCAO) model and confirmed infarction by triphenyl tetrazolium chloride (TTC) staining at 24 h postinjury. Representative images are shown in Fig. 1A, with the white region indicating infarcted tissue. Ferroptosis-associated markers, including 4-hydroxynonenal (4-HNE), malondialdehyde (MDA), total iron, GSH, and GSH/glutathione disulfide (GSSG) ratio, were assessed in sham animals as well as in the ipsilateral (Ipsi) and contralateral (Contra) hemispheres of tMCAO mice. No significant differences were observed between the sham group and the contralateral hemisphere of tMCAO mice (Fig. S1A), indicating minimal oxidative damage in nonischemic tissue. In contrast, compared to the contralateral side, the ipsilateral hemisphere exhibited significantly elevated levels of 4-HNE, MDA, and total iron, along with decreased GSH levels and a reduced GSH/GSSG ratio (Fig. 1B). Immunofluorescence staining for NeuN, 4-HNE, and MDA further demonstrated colocalization of lipid peroxidation products with neurons in the ischemic cortex (Fig. 1C and D, lower panels). In contrast, no significant staining of 4-HNE and MDA was observed in the contralateral cortex (Fig. 1C and D, upper panels) and the cortex of sham animals (Fig. S1B), supporting that cerebral ischemia/reperfusion induces neuronal lipid peroxidation and ferroptosis-associated injury.

Mitochondria are a primary source of ROS, and impaired mitochondrial function leads to excessive mtROS production, a known trigger of ferroptosis. To explore the role of mtROS in neuronal ferroptosis, we established an oxygen-glucose deprivation (OGD) model followed by reoxygenation in Neuro2a (N2a) cells. Fluorescent probes revealed a marked increase in mtROS (MitoSOX) and mitochondrial lipid peroxidation (MitoPeDPP) in OGD-treated cells (Fig. 2A and B). Additionally, the levels of cytosolic ROS (cROS; measured by 2′,7′-dichlorofluorescein diacetate [DCFH-DA]) and the oxidized/reduced fluorescence ratio of C11-BODIPY also increased, confirming elevated oxidative stress and lipid peroxidation. Treatment with Mito-TEMPO (Mito-T), an mtROS inhibitor, significantly reduced mtROS, cROS, and lipid peroxidation, supporting an important role for mtROS in OGD-induced ferroptosis.

mPTP opening is a key regulator of mitochondrial function and mtROS production [16,17]. Using the calcein–CoCl₂ quenching assay, we demonstrated that cyclosporin A (CSA), an mPTP inhibitor, effectively blocked mPTP opening (Fig. 2C). CSA also reduced mtROS, mitochondrial lipid peroxidation, cROS, and C11-BODIPY oxidation while restoring mitochondrial membrane potential (ΔΨm), as assessed by JC-1 staining (Fig. 2D and Fig. S2A and B). Furthermore, both Mito-T and CSA treatment significantly improved cell viability under OGD conditions, as measured by the Cell Counting Kit-8 (CCK-8) assay and Live/Dead cell staining (Fig. S2C to E). Collectively, these findings demonstrate that mtROS production and mPTP dysfunction are closely associated with ferroptotic injury under OGD conditions and that their inhibition mitigates neuronal oxidative damage.

Recent studies have highlighted the significant roles of mt-circRNAs in regulating mitochondrial function, and their dysregulation has been implicated in various pathophysiological conditions [19,23,24]. Given the high abundance of mitochondria in the brain, we performed mt-circRNA sequencing in mouse brain tissue and identified 65 mt-circRNAs, which were distributed across both the heavy and light strands of the mitochondrial genome (Fig. 3A). Among these, circRNAs derived from the mitochondrial gene MTCO2 have been shown to influence mtROS production in human liver fibroblasts and to be associated with lymphocytic leukemia [25,26]. To confirm its presence in the brain, we designed divergent and convergent primers targeting the backsplice junction of circMTCO2. Sanger sequencing of the polymerase chain reaction (PCR) product confirmed the existence of circMTCO2 (Fig. 3B). This was further validated by PCR analysis of complementary DNA (cDNA) and genomic DNA (gDNA) from mouse brain, as divergent primers amplified a product in cDNA, but not in gDNA, confirming the circular nature of circMTCO2 (Fig. 3C). To assess the stability of circMTCO2, we treated total RNA with RNase R. As expected, circMTCO2 was relatively resistant to RNase R treatment, whereas linear MTCO2 messenger RNA (mRNA) and β-actin mRNA were nearly completely degraded (Fig. 3D). Similarly, 4 h of transcriptional inhibition with actinomycin D (Act D) confirmed that circMTCO2 is more stable than its linear counterpart in N2a cells (Fig. 3E). To determine its relevance to ischemia, we measured circMTCO2 levels by quantitative reverse transcription polymerase chain reaction (qRT-PCR) in OGD-treated N2a cells and observed a significant reduction (Fig. 3F). In vivo, circMTCO2 expression in the ipsilateral hemisphere of tMCAO mice was significantly decreased at 24 h postinjury and remained reduced through day 3 (Fig. 3G). Taken together, these findings identify circMTCO2 as a stable, mitochondria-derived circRNA whose expression is dynamically regulated. Importantly, the significant down-regulation of circMTCO2 in both in vitro and in vivo models of ischemic stroke suggests a potential role in the pathophysiology of ischemic brain injury.

To investigate the functional role of circMTCO2 in neuronal ferroptosis and ischemic stroke, we developed a dual-targeting EV delivery system capable of transporting circMTCO2 into neuronal mitochondria. This approach utilized the rabies virus glycoprotein (RVG) peptide, which specifically binds to acetylcholine receptors, facilitating selective targeting of EVs to neuronal cells. The RVG peptide was fused to the EV membrane protein lysosome-associated membrane glycoprotein 2b (Lamp2b). Lentiviruses expressing RVG–Lamp2b and circMTCO2 with or without green fluorescent protein (GFP) were cotransduced into suspension 293F cells, followed by puromycin and hygromycin B selection to establish a stable cell line. EVs were isolated from the culture supernatant by sequential ultracentrifugation and filtered through a 0.22-μm filter to ensure sterility. EV^Ctrl refers to EVs harvested from the supernatant of untransduced 293F cells, serving as the control group in our experiments.

To further direct circMTCO2 to mitochondria, we employed triphenylphosphonium (TPP)–poly-d-lysine (PDL) conjugates, which are known to exhibit low cytotoxicity and high affinity for RNA [25,27]. TPP–PDL was electroporated into RVG-EV^RNA to generate the mitochondria-targeted RVG-EV^mt-RNA. Transmission electron microscopy confirmed that EV morphology was unchanged by these modifications (Fig. 4A). Nanoparticle size distribution analysis showed similar diameters across EV groups: 115.1 ± 1.3 nm (EV^Ctrl), 120.1 ± 2.1 nm (RVG-EV^RNA), and 112.9 ± 1.8 nm (RVG-EV^mt-RNA) (Fig. 4B). Western blotting for classical EV markers (Lamp2b, CD63, and tumor susceptibility gene 101 protein [Tsg101]) and the Golgi marker 130 kDa cis-Golgi matrix protein 1 (GM130) validated EV identity and purity (Fig. 4C). Next, we measured the amount of circMTCO2 encapsulated in EVs by absolute quantitative polymerase chain reaction (qPCR). The copy number of circMTCO2 per EV particle was 5.10 ± 0.33 in RVG-EV^RNA and 4.59 ± 0.11 in RVG-EV^mt-RNA (Fig. 4D). To assess mitochondrial delivery, N2a cells were treated with EV^Ctrl, RVG-EV^RNA, or RVG-EV^mt-RNA for 24 h, after which mitochondrial RNA was extracted. qRT-PCR revealed that RVG-EV^mt-RNA efficiently delivered circMTCO2 into mitochondria (Fig. 4E). To further confirm circMTCO2 delivery and subcellular localization, we generated GFP-labeled circMTCO2 constructs to obtain RVG-EV^RNA-GFP and RVG-EV^mt-RNA-GFP, both of which exhibited characteristics similar to those of their unlabeled counterparts (Fig. S3). In cells treated with RVG-EV^mt-RNA-GFP, circMTCO2-GFP localized specifically to mitochondria (Fig. 4F, right panel), whereas in the absence of the TPP–PDL moiety, circMTCO2-GFP was more diffusely distributed in the cytoplasm (Fig. 4F, left panel). Together, these findings demonstrate that the sequential neuron-to-mitochondria targeting EV platform enables efficient and specific circMTCO2 delivery into neuronal mitochondria, providing a useful tool for investigating its role in oxidative injury and ferroptosis.

To investigate the effects of circMTCO2 on mitochondrial function and neuronal ferroptosis, we treated OGD-exposed N2a cells with EV^Ctrl, RVG-EV^RNA, or RVG-EV^mt-RNA. Using the calcein–CoCl₂ quenching assay, we found that treatment with RVG-EV^mt-RNA significantly restored calcein fluorescence intensity, indicating that circMTCO2 mitigates mPTP opening under OGD conditions (Fig. 5A). MitoSOX and MitoPeDPP staining showed that RVG-EV^mt-RNA reduced mtROS and mitochondrial lipid peroxidation. Similarly, DCFH-DA and oxidized/reduced C11-BODIPY probes demonstrated reduced cROS and cellular lipid peroxidation. JC-1 staining further confirmed that RVG-EV^mt-RNA partially restored mitochondrial membrane potential (ΔΨm), indicating improved mitochondrial integrity (Fig. 5A to C). Moreover, CCK-8 assays and Live/Dead cell staining demonstrated enhanced cell viability in RVG-EV^mt-RNA-treated cells under OGD conditions, indicating that RVG-EV^mt-RNA effectively mitigates cell injury (Fig. 5D and Fig. S4A). To evaluate whether TPP–PDL itself exerts any intrinsic protective effects, we also included a control group treated with RVG-EV^mt (TPP–PDL-modified vesicles lacking circMTCO2). This group was tested across multiple assays under OGD conditions (Fig. S5). In all assays, RVG-EV^mt and RVG-EV^RNA produced comparable results, suggesting that TPP–PDL itself does not produce measurable mitochondrial or redox-related protection.

To further validate the protective effect of RVG-EV^mt-RNA, we established an erastin-induced ferroptosis model. The results demonstrated that RVG-EV^mt-RNA significantly alleviated ferroptosis-associated injury, as evidenced by reduced oxidative stress and lipid peroxidation markers together with improved cell viability (Fig. S4B to D). These findings collectively demonstrate that mitochondria-targeted circMTCO2 delivery via RVG-EV^mt-RNA protects neurons from ferroptosis-associated injury by restoring mitochondrial function and redox balance, highlighting its therapeutic potential for ischemic stroke.

CircRNAs are emerging as key regulators of RNA-binding proteins and cellular processes. To elucidate the molecular mechanisms underlying the protective effects of circMTCO2, we conducted RNA pull-down assays in N2a cells using a biotinylated circMTCO2 probe, followed by liquid chromatography–tandem mass spectrometry (LC–MS/MS) analysis. ANT1, a key regulator of the mPTP, was identified as a potential binding partner of circMTCO2 (Fig. 6A). Western blotting confirmed the presence of ANT1 in the circMTCO2-associated protein complex (Fig. 6B). RNA immunoprecipitation (RIP) assays showed that circMTCO2 was specifically enriched in ANT1 immunoprecipitates, further supporting interaction between circMTCO2 and ANT1 (Fig. 6C). To investigate their subcellular localization, we performed circMTCO2 fluorescence in situ hybridization (FISH) combined with MitoTracker and ANT1 staining. Although submitochondrial resolution was not achieved, the results revealed substantial colocalization of circMTCO2 with ANT1 in the mitochondrial region (Fig. 6D).

Next, to test whether OGD impairs the interaction between endogenous circMTCO2 and ANT1, we performed RNA pull-down assays under OGD conditions. The results showed that OGD significantly reduces the binding between circMTCO2 and ANT1 (Fig. S6). To examine the regulatory dynamics of this interaction under oxidative stress, we overexpressed circMTCO2 in OGD-treated N2a cells. RNA pull-down and RIP assays demonstrated that circMTCO2 overexpression significantly enhanced the association between ANT1 and circMTCO2, restoring the interaction impaired by OGD (Fig. 6E and F). To identify the specific region of circMTCO2 involved in ANT1 binding, we predicted its secondary structure and interaction propensity using RNAfold and catRAPID (Fig. S7) and constructed a series of truncated circMTCO2 variants. In vitro RNA–protein pull-down assays identified the 36- to 120-nt region of circMTCO2 as essential for ANT1 binding (Fig. 6G). We then overexpressed full-length and mutant circMTCO2 in N2a cells. RIP assays showed that mutations disrupting the paired region of circMTCO2 markedly reduced the ANT1–circMTCO2 interaction (Fig. 6H). Finally, we assessed the functional significance of this interaction in regulating mPTP opening using the calcein–CoCl₂ quenching assay. The results showed that the protective effect of RVG-EV carrying full-length circMTCO2 was significantly diminished when the binding-deficient mutant circMTCO2 was used (Fig. 6I). Together, these findings demonstrate that circMTCO2 binds ANT1 through a defined structural region and that this interaction is required for the suppression of mPTP opening under oxidative stress.

We next evaluated the therapeutic efficacy and safety of circMTCO2 delivery in vivo using a tMCAO mouse model, guided by the experimental timeline shown in Fig. 7A. DiR (1,1′-dioctadecyl-3,3,3ʹ,3ʹ-tetramethylindotricarbocyanine iodide)-labeled EVs were injected intravenously at a dose of 200 μg. In vivo imaging system (IVIS) imaging demonstrated that RVG modification markedly enhanced the relative brain enrichment of EVs (Fig. 7B and Fig. S8A). TTC staining at 1 d postinjury (dpi) revealed a significant reduction in infarct volume in the RVG-EV^mt-RNA-treated group compared with that in controls, indicating effective neuroprotection (Fig. 7C). Biochemical analyses revealed significant reductions in 4-HNE, MDA, and total iron levels in the RVG-EV^mt-RNA group, along with marked increases in GSH levels and the GSH/GSSG ratio, indicating effective suppression of oxidative stress and lipid peroxidation in the ischemic brain (Fig. 7D). Behavioral tests were performed to assess neurological function over 21 d. RVG-EV^mt-RNA treatment markedly improved modified neurological severity score (mNSS), beam walking performance, rotarod endurance, and corner test performance beginning at day 3 postinjury (Fig. 7E). These improvements persisted through day 21, indicating sustained functional recovery.

To assess the safety profile of RVG-EV^mt-RNA therapy, we performed comprehensive hematological and biochemical analyses at 7 dpi. Routine parameters, including red blood cell, white blood cell, platelet, alanine aminotransferase, aspartate aminotransferase, alkaline phosphatase, blood urea nitrogen, and creatinine, remained within normal ranges across all treatment groups (Fig. S8B). Additionally, hematoxylin and eosin (H&E) staining of the heart, liver, spleen, lung, kidney, and brain at 7 dpi revealed no histological abnormalities (Fig. 7F). Collectively, these results support that RVG-EV^mt-RNA-mediated delivery of circMTCO2 confers neuroprotection in ischemic stroke by mitigating oxidative stress and lipid peroxidation, reducing infarct size, improving neurological outcomes, and exhibiting a favorable safety profile.

Male C57BL/6 mice, aged 8 to 9 weeks and weighing 20 to 22 g, were used in this study. All animal procedures were approved by the Animal Care and Use Committee (No. 202201013). Mice were anesthetized with an intraperitoneal injection of sodium pentobarbital (50 mg/kg). The right common carotid artery, external carotid artery, and internal carotid artery were exposed. A silicon-coated monofilament (diameter 0.20 to 0.21 mm) was inserted through the external carotid artery and advanced into the internal carotid artery to occlude the middle cerebral artery. CBF was continuously monitored using laser Doppler flowmetry (RWD Life Science, China), with successful occlusion confirmed by a >70% reduction in baseline CBF. After 60 min of occlusion, the filament was withdrawn to allow reperfusion. Sham surgery was performed with identical procedures, except for the insertion of the monofilament. All procedures followed the ARRIVE guidelines for animal research. Mice were randomly assigned to groups, and sample sizes were determined based on prior experience to ensure statistical power. Data analysis was conducted in a double-blinded manner with no exclusions.

MDA levels in mouse brain were quantified using an MDA assay kit (Solarbio Life Science, China). Brain tissues were homogenized in lysis buffer and centrifuged, and 100 μl of the supernatant was mixed with 300 μl of MDA detection solution. The mixture was incubated at 100 °C for 60 min, and optical density (OD) was measured at 532 nm using a microplate reader. The MDA content was calculated using the following formula according to the manufacturer’s instructions: MDA content (nmol/mg protein) = (53.763 × ΔA)/C_pr, where ΔA = ΔA₅₃₂ − ΔA₆₀₀, ΔA₅₃₂ = A_532T − A_532B, and ΔA₆₀₀ = A_600T − A_600B. A_532T and A_600T are the sample absorbances at 532 and 600 nm, respectively, and A_532B and A_600B are the blank absorbances. C_pr is the protein concentration.

4-HNE levels in mouse brain were analyzed using a 4-HNE assay kit (Abcam, UK). Brain tissues were lysed with radioimmunoprecipitation assay (RIPA) lysis buffer and centrifuged to collect the supernatant. A 50-μl standard or sample was added to a 4-HNE conjugate-coated plate and incubated for 10 min. After adding 50 μl of diluted anti-4-HNE antibody, the plate was incubated for 1 h. After 3 washes with 1× wash buffer, 100 μl of the diluted secondary antibody–horseradish peroxidase (HRP) conjugate was added, followed by another 1-h incubation. After another wash, 100 μl of substrate solution was added and incubated for 15 min. The reaction was stopped with 100 μl of stop solution, and OD was measured at 450 nm. 4-HNE concentration was determined using a standard curve derived from the logarithmic concentration and OD₄₅₀ values.

GSH and GSSG levels in mouse brain were measured using GSH and GSSG assay kits (Solarbio Life Science, China) following the manufacturer’s instructions. Brain tissues were homogenized in lysis buffer, and the supernatant was collected. Standard solutions of GSH and GSSG were prepared at various concentrations. GSSG content was calculated as follows: GSSG content (μmol/g tissue) = GSSG concentration × V/W × 612.62, where V is the total supernatant volume and W is the tissue wet weight. Similarly, GSH content was calculated using GSH content (μmol/g tissue) = GSH concentration × V/W × 307.33.

Iron levels were measured using an iron assay kit (Solarbio Life Science, China). Brain tissues were homogenized in extraction solution, and the lysates were centrifuged to collect the supernatant. A total of 120 μl of the supernatant was mixed with 60 μl of Reagent 1 and 120 μl of Reagent 2 and then incubated at 100 °C for 5 min. After incubation, 60 μl of chloroform was added, and the mixture was centrifuged. The upper inorganic phase was transferred to a 96-well plate, and OD was measured at 532 nm. The total iron content was calculated according to the assay kit instructions.

Mouse neuroblastoma N2a cells were obtained from the National Infrastructure of Cell Line Resource and cultured in minimum essential medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin (Thermo Fisher Scientific, USA). Cells were maintained at 37 °C in a 5% CO₂ humidified incubator and passaged every 3 d. For OGD, N2a cells were cultured in glucose-free medium and exposed to hypoxia for 3 h in a chamber with a gas mixture of 95% N₂ and 5% CO₂. After OGD, cells were returned to normoxic conditions (5% CO₂) and complete glucose-containing medium for 24 h of reoxygenation to model ischemia/reperfusion injury in vitro. OGD/reoxygenation was used as an in vitro ischemia/reperfusion paradigm to examine intrinsic mitochondrial dysfunction and ferroptosis-related responses under controlled conditions. Unless otherwise specified, OGD refers to a 3-h OGD followed by 24-h reoxygenation.

To induce ferroptosis, cells were treated with 40 μM erastin (MedChemExpress, USA) for 24 h. To modulate mitochondrial function, cells were pretreated for 2 h with either 1 μM Mito-T, a mitochondria-targeted antioxidant, or 10 μM CSA (MedChemExpress, USA), an inhibitor of the mPTP. To assess treatment efficacy, N2a cells were incubated in a 24-well plate with 40 μg of either EV^Ctrl, RVG-EV^RNA, or RVG-EV^mt-RNA for 24 h.

Cell viability was assessed using CCK-8 (Beyotime, China). N2a cells were seeded into a 96-well plate and cultured for 24 h at 37 °C in a humidified incubator with 5% CO₂. After the indicated treatments, cells were washed with phosphate-buffered saline (PBS) and incubated with fresh culture medium containing 10% (v/v) CCK-8 reagent. After 3 h of incubation at 37 °C, the absorbance at 450 nm was measured using a microplate reader (Multiskan FC, Thermo Fisher Scientific, USA). For Live/Dead cell staining, N2a cells were seeded into 24-well plates and treated as indicated. Live and dead cells were stained using the Live/Dead Viability/Cytotoxicity Kit (Beyotime, China) according to the manufacturer’s instructions. Briefly, cells were incubated with calcein-AM and propidium iodide, and fluorescence images were acquired by fluorescence microscopy. The percentage of live cells was quantified from 5 random fields per well.

C11-BODIPY 581/591 (Thermo Fisher Scientific, USA) was used to measure intracellular lipid peroxidation; DCFH-DA (MedChemExpress, USA), intracellular ROS; MitoPeDPP (Dojindo Laboratories, Japan), mitochondrial lipid peroxidation; and MitoSOX (Thermo Fisher Scientific, USA), mtROS. C11 BODIPY 581/591 was diluted to 10 μM, DCFH-DA to 5 μM, MitoSOX to 5 μM in serum-free medium, and MitoPeDPP to 0.5 μM in Hank’s solution with 10 mM HEPES. After treatment, N2a cells were incubated with the corresponding probes at 37 °C for 30 min in the dark, followed by nuclear staining with 4′,6-diamidino-2-phenylindole (DAPI). Images were captured using a fluorescence microscope (BX53, Olympus, Japan). Fluorescence excitation/emission wavelengths were as follows: C11-BODIPY 581/591, 581/591 nm (reduced) and 488/510 nm (oxidized); DCFH-DA, 488/525 nm; MitoPeDPP, 470/525 nm; MitoSOX, 396/610 nm; and DAPI, 405/461 nm. The fluorescence intensity was quantified using the ImageJ software by measuring the mean gray value (average fluorescence intensity) within defined regions of interest, across at least 5 randomly selected fields per group. Unless otherwise specified, this method was applied for cell fluorescence quantification throughout the study.

mPTP opening was assessed using a calcein–CoCl₂ quenching assay kit (Beyotime, China) following the manufacturer’s instructions. N2a cells were incubated with a 1× calcein-AM and 1× CoCl₂ fluorescence quenching solution for 40 min at 37 °C. Calcein was excited at 494 nm, and emission was recorded at 517 nm. Fluorescence images were captured using a fluorescence microscope, and intensity was quantified using the ImageJ software.

Mitochondrial transmembrane potential was assessed using the lipophilic cation probe JC-1 (MedChemExpress, USA), which accumulates in a potential-dependent manner. A decrease in the red (JC-1 aggregates) to green (JC-1 monomers) fluorescence intensity ratio indicates mitochondrial depolarization. Cells were incubated with 2 μM JC-1 solution for 30 min at 37 °C in the dark. After resuspension in ice-cold PBS, cells were analyzed by flow cytometry (FACSAria III, Becton Dickinson, USA), collecting data from at least 100,000 cells. The fluorescence excitation/emission wavelengths were as follows: JC-1 monomer, 510/527 nm (green), and JC-1 aggregate, 585/590 nm (red). The red/green fluorescence ratio was used to quantify mitochondrial dysregulation.

Total RNA was extracted from mouse brain tissue using RNAiso Plus (Takara Bio Inc., Japan). Ribosomal RNA (rRNA) was removed using the KAPA RNA HyperPrep Kit with RiboErase for Illumina (HMR) (Kapa Biosystems, Inc., USA). PolyA tailing was performed on the rRNA-depleted RNA, followed by purification with CleanNGS magnetic beads (GC Biotech, the Netherlands). An RNase R reaction was then conducted at 37 °C for 30 min, and the RNA was purified again. Rolling circle reverse transcription was performed using the SMARTer PCR cDNA Synthesis Kit (Takara Bio Inc., Japan), and cDNA amplification was carried out with the KAPA HotStart mix (Kapa Biosystems, USA). Fragments between 200 and 2,000 bp were selected for sequencing. End repair and dA-tailing were performed using the NEBNext Ultra II End Repair/dA-Tailing Module (NEB, USA). Libraries were loaded onto MinION Mk1B flow cells and sequenced on the MinION platform (Oxford Nanopore Technologies, UK) using standard sequencing buffers. Full-length circRNAs were detected using CIRI-long as previously reported [43].

gDNA was extracted from the mouse cerebral cortex using the MiniBEST Universal Genomic DNA Extraction Kit (Takara Bio Inc., Japan) according to the manufacturer’s protocol. The DNA concentration was measured with a NanoDrop One Microvolume spectrophotometer (Thermo Fisher Scientific, USA).

Total RNA from cells and tissues was extracted using the TRIzol method (Thermo Fisher Scientific, USA). cDNA synthesis was performed with PrimeScript RT Master Mix (Takara Bio Inc., Japan). qRT-PCR was conducted using TB Green Premix Ex Taq II (Takara Bio Inc., Japan), with primers designed via National Center for Biotechnology Information Primer-BLAST (detailed in Table S1).

For RNase R treatment, 2 μg of RNA was incubated at 37 °C for 5 min with or without 3 U/μg RNase R (LGC Biosearch Technologies, UK). Reverse transcription was performed using random hexamer primers, followed by qRT-PCR. For Act D treatment, N2a cells were treated with 5 μg/ml Act D (MedChemExpress, USA) to inhibit transcription and collected at specified time points. RNA was extracted, and the stability of circRNA and linear mRNA was analyzed by qRT-PCR.

Plasmids for overexpressing circ MTCO2 were constructed using the pLO5-ciR vector, while GFP-tagged circMTCO2 plasmids were generated with the pLC5-ciR vector. A mutant plasmid of circ MTCO2, featuring a deletion between positions 36 and 120 nt, was also constructed (GenScript, China). The ANT1 plasmid, designed to express the ANT1 protein, was created by cloning the coding sequence of the mouse SLC25A4 gene (NM_007450.5) into the pCDNA3.1(+) vector (GenScript, China). All plasmids were confirmed by DNA sequencing and transfected using Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific, USA).

Suspension 293F cells (Thermo Fisher Scientific, USA) were cultured in Nalgene single-use Erlenmeyer flasks on an orbital shaker at 100 rpm. 293F cells were cotransduced with lentiviruses expressing RVG–Lamp2b and circMTCO2, with or without GFP (OBiO Tech, Inc., China). Forty-eight hours posttransduction, hygromycin B and puromycin (MedChemExpress, USA) were added to the culture medium to establish a stable cell line, which was passaged every 3 d. EV^Ctrl and RVG-EV^RNA were harvested from the supernatant of untransduced 293F cells and the stable cell line via sequential ultracentrifugation. The culture medium was centrifuged at 500 × g for 10 min to remove cells and at 3,000 × g for 10 min to remove cell debris and underwent ultracentrifugation at 100,000 × g for 2 h at 4 °C using Avanti JXN-30 (Beckman Coulter, USA). To ensure sterilization, the resuspended EVs were filtered with a 0.22-μm filter. To produce RVG-EV^mt-RNA, the mitochondria-targeting moiety TPP–PDL (Ruixibio, China) was loaded into RVG-EV^RNA. Electroporation was performed with 0.5 μM TPP–PDL in 4-mm electroporation cuvettes at 700 V (Bio-Rad Laboratories, USA), and the samples were placed on ice for 30 min. EVs were then centrifuged at 100,000 × g for 1 h at 4 °C to collect RVG-EV^mt-RNA. For RVG-EV^{mt-mut RNA}, the lentivirus expressing circMTCO2 was replaced with one expressing the mutant circMTCO2 (36 to 120 nt), and the procedure followed was identical to that for RVG-EV^mt-RNA. The EV protein concentration was measured using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, USA). EV size distributions (intensity-weighted mode) were determined by dynamic light scattering using Zetasizer Ultra (Malvern Panalytical, UK). Transmission electron microscopy was performed as previously described [44].

Samples were lysed in RIPA buffer, separated by 10% sodium dodecyl sulfate–polyacrylamide gel electrophoresis and transferred to a polyvinylidene fluoride membrane. The blots were incubated with primary antibodies overnight at 4 °C, followed by HRP-conjugated secondary antibodies. Detection was performed using Super ECL Plus (Applygen Technologies Inc., China), and the signal intensity was quantified with the ChemiDoc MP Imaging System (Bio-Rad Laboratories, USA). The antibodies used are listed in Table S2, with glyceraldehyde-3-phosphate dehydrogenase serving as the loading control.

Absolute qPCR was performed to determine the copy number of circMTCO2 in EVs. A 352-bp fragment containing the backspliced junction of circMTCO2 was inserted into the pcDNA3.1(+) vector (GenScript, China). Serial 10-fold dilutions of the plasmid were used to generate a standard curve through real-time PCR. The copy number was calculated using the following formula: copy number (copy/μl) = 6.02 × 10²³ × plasmid concentration (ng/μl) × 10⁻⁹/[(molecular weight of vector + molecular weight of inserted fragment) × 660]. The standard curve equation was derived as Y = aX + b, where X is the log-transformed initial copy number and Y is the corresponding cycle threshold value. The cycle threshold values of circMTCO2 were used to determine its copy number in EVs by applying the standard curve equation.

To evaluate mitochondrial targeting, N2a cells were incubated with 40 μg of RVG-EV^RNA-GFP or RVG-EV^mt-RNA-GFP in a 24-well plate for 24 h. The cells were then stained with 500 nM MitoTracker Red (Thermo Fisher Scientific, USA) in serum-free culture medium for 40 min at 37 °C in the dark. After 3 washes with PBS, nuclei were stained with DAPI. The fluorescence excitation/emission wavelengths were as follows: MitoTracker Red, 581/644 nm; GFP, 488/509 nm; and DAPI, 405/461 nm. Images were acquired using a Zeiss LSM770 confocal microscope (Zeiss ZEN 2 software).

N2a cells cultured in 10-cm dishes were washed twice with ice-cold PBS and detached using a cell scraper. The cell pellet was resuspended in ice-cold IB-1 homogenization buffer (225 mM mannitol, 75 mM sucrose, 0.1 mM EGTA, and 30 mM Tris–HCl) containing a 1× protease inhibitor cocktail (Beyotime, China). Homogenization was performed with 90 strokes in a Dounce homogenizer (Sigma-Aldrich, USA). The homogenate was centrifuged at 740 × g for 5 min at 4 °C to collect the supernatant, which was then centrifuged at 9,000 × g for 10 min at 4 °C to isolate crude mitochondria. The mitochondrial pellet was resuspended in ice-cold IB-2 buffer (225 mM mannitol, 75 mM sucrose, and 30 mM Tris–HCl) and purified by centrifugation at 10,000 × g for 10 min at 4 °C. The purified mitochondria were stored at −80 °C.

RNA pull-down assays were performed using the Magnetic RNA-Protein Pull-Down Kit (Thermo Fisher Scientific, USA) according to the manufacturer’s instructions. N2a cells were cross-linked with 1% formaldehyde for 10 min. The cells were then lysed in Pierce IP Lysis Buffer (Thermo Fisher Scientific, USA), supplemented with 2 mmol/l dithiothreitol (Solarbio Life Science, China), 1 mM phenylmethylsulfonyl fluoride (Beyotime, China), and 200 units/μl RiboLock RNase Inhibitor (Thermo Fisher Scientific, USA) for 20 min on ice. Lysates were sonicated for 3 min using JY92-IIN Ultrasonic Homogenizer. The lysates were incubated overnight at 37 °C with 100 pmol of 3′-biotin-labeled circMTCO2 probe or a negative control probe (Genepharma, China) (detailed in Table S1). Following incubation, 50 μl of Dynabeads M-280 streptavidin (Thermo Fisher Scientific, USA) was added, and the mixture was rotated end to end at room temperature for 1 h. The beads were washed 5 times with wash buffer. Proteins pulled down by the probes were analyzed by Western blotting or LC–MS/MS.

LC–MS/MS analysis was performed using ORBITRAP ECLIPSE Mass Spectrometer (Thermo Fisher Scientific, USA). Peptides were loaded onto a C18 trap column, connected to a C18 reverse-phase analytical column, and eluted over a 60-min gradient (6% to 95% acetonitrile, 0.1% formic acid in water) at a flow rate of 300 nl/min. MS1 spectra were acquired at 120,000 resolution (350 to 1,500 m/z), and precursor ions were selected for MS2 analysis with higher-energy collisional dissociation collision energy (33%). Spectra were acquired at 15,000 resolution (first mass 110 m/z).

To identify the binding region between circMTCO2 and ANT1, truncated circMTCO2 probes of varying lengths were synthesized by GenScript (see Table S1 for details). In an in vitro RNA–protein pull-down assay, 50 pmol of the truncated circMTCO2 probes were incubated with mitochondria isolated from ANT1-overexpressed N2a cells, followed by coincubation with Dynabeads M-280 streptavidin, as previously described. Western blotting was performed to detect the bound proteins.

RIP was performed using the Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Merck, Germany). N2a cells were washed with ice-cold PBS and lysed in Pierce IP Lysis Buffer containing a protease inhibitor and an RNase inhibitor (Thermo Fisher Scientific, USA) for 20 min on ice. Protein A/G magnetic beads (Thermo Fisher Scientific, USA) were incubated with antibodies against ANT1 (ABclonal, China) or control rabbit immunoglobulin G (Cell Signaling Technology, USA) for 1 h at room temperature. The cell lysate supernatant was then incubated with the antibody–bead complexes overnight at 4 °C. After immunoprecipitation, the complexes were washed 5 times with wash buffer. Protein digestion was carried out with proteinase K buffer at 55 °C for 30 min. RNA was extracted using the TRIzol method and analyzed by qRT-PCR.

RNA FISH was performed to detect circMTCO2 using the Fluorescent In Situ Hybridization Kit (RiboBio, China) according to the manufacturer’s instructions. Mitochondria in N2a cells were stained with 400 nM MitoTracker Deep Red (Thermo Fisher Scientific, USA) for 1 h, followed by washing with PBS. Cells were fixed with 4% formaldehyde for 15 min at room temperature and permeabilized with 0.05% Triton X-100 for 10 min at 4 °C. After 30 min of prehybridization at 37 °C, hybridization was performed overnight at 60 °C in a dark, humid chamber with 20 nM digoxin-labeled oligonucleotide probe targeting the circMTCO2 backsplice sequence (SinoGenoMax Technologies, China). Cells were washed with 4×, 2×, and 1× saline-sodium citrate (Beyotime, China) and incubated with anti-digoxigenin–fluorescein isothiocyanate (FITC) antibody (1:300) (Roche, Switzerland) for 1 h at room temperature.

To assess colocalization of circMTCO2 and ANT1 protein, cells were incubated with an ANT1 antibody (ABclonal, China) for 4 h at room temperature. After 3 PBS washes, cells were incubated with Alexa Fluor 594 goat anti-rabbit immunoglobulin G secondary antibody (Thermo Fisher Scientific, USA) for 1 h. Nuclei were stained with DAPI, and slides were mounted with ProLong Gold antifade reagent (Thermo Fisher Scientific, USA). The fluorescence excitation/emission wavelengths were as follows: anti-digoxigenin–FITC, 488/517 nm, and Alexa Fluor 594, 594/615 nm. Images were captured using a Leica MICA microscope (Software version: Leica LAS X 3.10.0).

The secondary structure of circMTCO2 was predicted using the RNAfold web server, part of the ViennaRNA Web Services (Institute for Theoretical Chemistry, University of Vienna), based on the ViennaRNA Package [45]. Predictions were generated using the minimum free energy model, with color scales indicating the confidence level of each base-pair prediction; red shades denoted regions of high prediction confidence (website: http://rna.tbi.univie.ac.at/). To predict potential interaction sites between circMTCO2 and ANT1, we utilized the catRAPID omics platform (version 2.1).

EVs were labeled with DiR as previously described [44]. DiR-labeled EVs (200 μg) were intravenously administered into mice via the tail vein. Fluorescence imaging of organs was performed at 4 h postinjection using a small-animal imaging system (Xenogen IVIS Spectrum, PerkinElmer, USA) with excitation and emission wavelengths of 745 and 800 nm, respectively. IVIS was used as a semiquantitative method to compare relative biodistribution among EV formulations under standardized conditions. Regions of interest were defined, and the average radiant efficiency was quantified in (p/s/cm²/sr)/(μW/cm²).

Mice were sacrificed 24 h post-tMCAO injury, and their brains were quickly harvested and sectioned into 1-mm-thick coronal slices. The tissue slices were incubated in 2% TTC (Merck, Germany) at 37 °C for 20 min, followed by fixation in 4% paraformaldehyde. Infarct areas appeared white, while viable tissue stained red. The infarction percentage was calculated using the following formula: Infarct area (%) = [(contralateral hemisphere area − non-infarct area of the ipsilateral hemisphere)/contralateral hemisphere area] × 100. Because all coronal sections were 1 mm thick, infarct area summed across sections was expressed as infarct volume. Images were analyzed using the ImageJ software (National Institutes of Health, USA).

Frozen 12-μm coronal brain sections were prepared using a Leica CM1950 cryostat (Leica Biosystems, Germany). Sections were blocked with 3% bovine serum albumin and 0.3% Triton X-100 in PBS for 1 h at room temperature. After overnight incubation with primary antibodies, sections were incubated with secondary antibodies for 2 h at room temperature. Details of the antibodies used are provided in Table S2. Sections were mounted using Antifade Mounting Medium with DAPI (SouthernBiotech, USA). Images were acquired using a BX53 microscope (Olympus, Japan).

Tissue samples (brain, liver, heart, lung, spleen, and kidney) were fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned at 4-μm thickness using a microtome (Leica Biosystems, Germany). H&E staining was performed using an H&E staining kit (Beyotime, China). Images were captured using a BX53 microscope (Olympus, Japan).

Blood samples were collected from the tail vein at 7 d post-tMCAO injury. Hematological parameters, including red blood cell count, white blood cell count, and platelet count, were measured using a TEK-II Mini hematology analyzer (Tecom Science Corporation, China). Biochemical markers, including alanine aminotransferase, aspartate aminotransferase, alkaline phosphatase, blood urea nitrogen, and creatinine, were measured using a TBA-40FR biochemical analyzer (Toshiba Medical Systems Corporation, Japan).

Mice were randomly assigned to 5 groups and pre-trained for 3 d. Testing was conducted on the baseline (presurgery) day and on postsurgery days 1, 3, 5, 7, 10, 14, and 21. Behavioral tests were performed and analyzed by 2 independent, blinded technicians. Neurological function was evaluated up to 21 d post-tMCAO using mNSS, which evaluates motor, sensory, and balance functions on a scale from 0 (normal) to 14 (most severe impairment).

For the beam walking test, mice were placed at the start of a 1-m-long, 12-mm-wide wooden beam elevated 50 cm above the ground. Performance was scored on a 6-point scale: 0 = fall off the beam; 1 = unable to cross or place the affected limb on the beam but can maintain balance; 2 = unable to cross but can place the affected limb on the beam and balance for ≥5 s; 3 = crosses <50% of the beam with missteps; 4 = crosses >50% of the beam with missteps; 5 = crosses the beam with ≤ 2 limb slips; and 6 = normal crossing without limb slips.

For the rotarod test, mice were placed on a rotarod system (Nanjing Calvin Biotechnology Co., Ltd., China) with a rod accelerating from 5 to 40 rpm over 300 s, followed by a constant speed for 2 min. The trial ended when the mouse fell off, and the latency to fall was recorded. Each trial was repeated 3 times, with a 30-min interval.

For the corner turn test, 2 wooden boards (dimensions: 30 × 20 × 1 cm) formed a 30° corner apparatus. Mice were placed 12 cm from the corner and allowed to walk into it, where the direction of turning was recorded to assess asymmetry. Each mouse participated in 10 trials. The asymmetry percentage was calculated as follows: Asymmetry (%) = (number of left turns/(number of left turns + number of right turns)) × 100%.

Data are presented as mean ± standard error of the mean. For relative comparisons, data were normalized to the control group, as specified in the figure legends. Statistical analyses were conducted using GraphPad Prism v8.0 (GraphPad Software Inc., USA). The Shapiro–Wilk test was used to assess data normality. For normally distributed data, differences between 2 groups were analyzed using an unpaired 2-tailed Student t test. Comparisons among 3 or more groups were performed using one-way analysis of variance (ANOVA) or 2-way ANOVA followed by Tukey’s multiple-comparisons test, as appropriate. Statistical significance was defined as *P < 0.05, **P < 0.01, and ***P < 0.001.