We obtained discovery datasets GSE16561 (10) and GSE195442 (8) and validation datasets GSE110993 (11) and GSE22255 (12) from the GEO database.

The GSE16561 dataset included 39 cerebral infarction patients and 24 healthy controls, while GSE195442 comprised 10 patients and 10 controls. For validation, GSE110993 contained 20 cerebral infarction patients and 20 healthy controls, and GSE22255 included 20 patients and 20 controls. Detailed characteristics of these datasets are presented in Table 1.

Following normalization of raw data, differential expression analysis was performed using the R Bioconductor limma package (v3.52.3). Genes meeting the criteria of p < 0.05 and |log2fold change| > 0.58 were identified as differentially expressed mRNAs (13), while circRNAs with p < 0.05 and |log2fold change| > 1.2 were classified as differentially expressed. Subsequently, DEcircRNAs were analyzed using the circbank database¹ (14) to predict potential miRNA interactions. The resulting miRNAs were further filtered based on differential expression (p < 0.05 and |log2 fold change| > 1.2) using GSE110993 data. Heatmaps and volcano plots were generated using the R pheatmap package (v1.0.12).

Predicted miRNAs from the circbank database (see text footnote 1) were validated against GSE110993 data. miRNA-mRNA interaction datasets were obtained from miRcode (v6.0, http://www.mircode.org/), miRTarBase (v8.0, http://mirtarbase.mbc.nctu.edu.tw/), and TargetScan (v7.2, http://www.targetscan.org/), with potential interactions identified using Perl software (v5.36.0, https://www.perl.org/). Only mRNA-miRNA pairs confirmed by all three databases were included in subsequent analyses.

Using cerebral infarction diagnosis as the dependent variable, we analyzed 21 DEGs identified through intersection analysis of GSE16151 dataset and GSE110993 predictions. Ferroptosis-associated genes (including drivers, suppressors, markers, and unclassified genes) were retrieved from FerrDb (15).² Key ferroptosis DEGs were visualized using the R “Venn” package (v1.11).

PC12 cells (rat pheochromocytoma cell line; Shanghai Institute of Biological Science, Chinese Academy of Sciences) were maintained in high-glucose DMEM (C11995500BT, Gibco, China) containing 10% fetal bovine serum (13011-8611, TIANHANG, China), along with 100 U/mL penicillin and 100 μg/mL streptomycin (C100C5, NCM Biotech, China), at 37°C under 5% CO₂ humidified conditions. To establish the OGD/R model, cells were subjected to incubation in glucose/serum-free DMEM (11,966,025, Gibco) using a tri-gas incubator (37°C, 94% N₂/5% CO₂/1% O₂) for 4 h, followed by a 24 h recovery period in complete DMEM. The experimental design comprised two groups: Control (standard culture conditions) and OGD/R.

PC12 cells (10,000 cells/well) were seeded in 96-well plates with 100 μL complete medium. After OGD/R, cells were incubated with 10 μL CCK-8 reagent (G1613-5ML, Servicebio, China) in 100 μL fresh DMEM for 1 h at 37°C. Cell viability was calculated as: [(Aexperimental − Ablank)/(Acontrol − Ablank)] × 100%.

The ROS were measured using dihydroethidium (DHE; CA1420, Solarbio, China). Cells were incubated with DHE (1:1000 dilution in DMEM) at 37°C for 30 min in the dark, washed with PBS (G4202-500, Servicebio, China), and visualized using an inverted fluorescence microscope (DMi8, Leica).

The content of intracellular iron content, malondialdehyde (MDA), and reduced glutathione (GSH) levels were quantified using commercial kits (BC5315, BC0025, and BC1175 respectively; Solarbio, China) following manufacturer protocols. Tissue samples from ischemic penumbra were processed identically. Protein content was determined using a BCA kit (BL521A, Biosharp, China), with absorbance measurements normalized to protein concentration.

Total RNA was extracted from both cultured cells and brain tissue samples using RNAiso Plus reagent (9,109, Takara, Japan). Reverse transcription was performed using HiScript II Reverse Transcriptase (R223, Vazyme, China) following the manufacturer’s protocol. qRT-PCR amplification was carried out using ChamQ Universal SYBR qPCR Master Mix (Q311, Vazyme, China). The miRNA was reverse transcribed using the miRNA 1st Strand cDNA Synthesis Kit (by stem-loop) (MR101-01, Vazyme, China), and quantified using the miRNA Universal SYBR qPCR Master Mix (MQ101-01, Vazyme, China). All reactions were performed on a CFX Opus Deepwell Real-Time PCR Detection System (Bio-Rad, United States). GAPDH was used as the endogenous reference gene, and relative gene expression levels were calculated using the 2^−ΔΔCt method. The primer sequences used in this study are provided in Table 2.

CIL56 is a selective ferroptosis inducer that can induce ferroptosis by generating iron-dependent ROS. We chose CIL56 (T4309, TargetMol, America) to construct the ferroptosis model under the condition of 0.7 μM for 24 h.

(E)-2-benzylidene-3-(cyclohexylamino)-2,3-dihydro-1H-inden-1-one (BCI) is a commonly used inhibitor of MKP-1. We used BCI (T10486, TargetMol, America) to inhibit DUSP1 in PC12 cells under the condition of 10 μM for 24 h.

Adult male Sprague–Dawley rats (250–280 g; Experimental Animal Center of Xi’an Jiaotong University, License: XJTUAE2024-076) were housed under standard conditions. The MCAO/R was performed under 2% pentobarbital sodium anesthesia (40 mg/kg, i.p.). After exposing the carotid sheath, the common carotid artery and external carotid artery were ligated. A nylon monofilament (2636A4, Beijing Cinontech, China) was inserted 18–20 mm into the internal carotid artery to occlude the middle cerebral artery for 90 min, followed by reperfusion. Sham-operated rats underwent identical procedures without filament insertion. Twelve rats were included in each group (sham and MCAO/R), with animals dying before the 24 h endpoint being excluded and replaced sequentially. Ischemic penumbra sampling followed established protocols (16), as illustrated in Figure 1A.

Neurological function was assessed 24 h after surgery using the modified Zea Longa 5-point scale (0 = no deficit; 4 = involuntary movement with consciousness disturbance). Animals were excluded if they: (1) scored 0 (model failure) or 4 (excessive injury), (2) showed subarachnoid hemorrhage at necropsy, or (3) died before the sampling endpoint.

At 24 h post-reperfusion, rats were perfused transcardially with PBS followed by 4% paraformaldehyde. Brains were dehydrated, paraffin-embedded, and sectioned (4 μm) for hematoxylin and eosin staining and histopathological examination.

Protein extraction was conducted using RIPA lysis buffer (PL001, ZHHC, China) supplemented with protease inhibitors. Protein concentrations were quantified using a BCA assay (BL521A, Biosharp, China), with 30 μg of protein per sample mixed with 5 × loading buffer (ZS306-1, Zomanbio, China) at a 4:1 ratio. Following denaturation (100°C, 5 min), samples were resolved on 10% SDS-PAGE gels (PG212, Epizeme, China) and transferred to PVDF membranes (ISEQ00010, Millipore, United States) at 300 mA for 90 min. Membranes were blocked with RapidBlock solution (PS108, Epizeme, China) for 15 min at room temperature, followed by overnight incubation at 4°C with primary antibodies diluted in antibody dilution buffer (PS119, Epizeme, China). The following primary antibodies were used: DUSP1 (ET1701-82, Huabio, China; 1:1000), GPX4 (ET1706-45, Huabio, China; 1:10000), TFRC (ab269513, Abcam, United Kingdom; 1:5000), β-tubulin (66240-1-lg, Proteintech, China; 1:20000), PTGS2 (TP53818, Abmart, China,1:500), β-actin (66009-1-Ig, Proteintech, China; 1:20000). After TBST washes, membranes were incubated with HRP-conjugated secondary antibodies (SA00001-1/SA00001-2, Proteintech, China; 1:10000) for 1 h at room temperature. Protein bands were visualized using a chemiluminescence imaging system (GelView 6000Plus, BLTLUX, China). For reprobing, membranes were stripped with antibody stripping solution (WB6500, NCM Biotech, China). Band intensities were quantified using ImageJ software (v1.53t, National Institutes of Health, United States).

Statistical analyses were performed using R software (v4.3.0, https://www.r-project.org/) and GraphPad Prism (v9.0, San Diego, California, United States). Continuous variables with normal distribution were compared using Student’s t-test, while non-normally distributed data were analyzed using the Mann–Whitney U test. Brown-Forsythe tests with Dunnett’s T3 post hoc correction were used for multiple comparisons. A p-value < 0.05 was considered statistically significant. The study flowchart is presented in Figure 2.

Through comprehensive analysis of the GSE195442 and GSE16561 datasets, we identified 45 differentially expressed circRNAs (DEcircRNAs) and 201 differentially expressed mRNAs (DEmRNAs), respectively. The expression patterns of these differentially regulated circRNAs and mRNAs were clearly visualized through volcano plots (Figures 3A,B). Further intersection analysis between the DEmRNAs from GSE16561 and the potential mRNA targets predicted by miRNAs in GSE110993 revealed 21 overlapping genes. The expression profiles of these 21 DEmRNAs across different samples were subsequently illustrated using a heatmap (Figure 3C).

From the FerrDb database, we compiled a comprehensive set of 487 ferroptosis- associated genes, consisting of 235 driver genes, 196 suppressor genes, 102 unclassified genes, and 9 marker genes. By intersecting these genes with our 21 DEmRNAs, we identified two candidate ferroptosis-associated genes: the ferroptosis marker gene PTGS2 and the unclassified gene DUSP1 (Figure 4A). In the GSE16561 dataset, DUSP1 demonstrated robust diagnostic potential, with an area under the curve (AUC) of 0.8579, sensitivity of 82.05%, and specificity of 75% (Figure 4B). PTGS2 showed moderate diagnostic performance, with an AUC of 0.7853, sensitivity of 84.62%, and specificity of 58.33% (Figure 4C). These findings were further validated in the external dataset GSE22255, where DUSP1 maintained significant diagnostic value (AUC = 0.735, sensitivity = 95%, specificity = 60%; Figure 4D), while PTGS2 exhibited slightly reduced but still notable efficacy (AUC = 0.655, sensitivity = 95%, specificity = 45%; Figure 4E).

Following OGD/R treatment in PC12 cells, we observed a marked reduction in cell viability (Figure 5A). Concurrently, characteristic indicators of ferroptosis were evident, including decreased intracellular reduced GSH levels, elevated MDA and iron content, and increased ROS production (Figures 5B–E). The result of qRT-PCR further revealed upregulated TFRC and downregulated GPX4 expression (Figures 5F,G). The mRNA expression of DUSP1 was increased (Figure 5H). The result of Western blotting corroborated these findings at the protein level, demonstrating increased TFRC and decreased GPX4 levels, and significantly elevated DUSP1 protein expression (Figures 5I–L).

Meanwhile, the mRNA expression of PTGS2 was increased (Figure 5M). The result of Western blotting demonstrated that the protein expression of PTGS2 was increased (Figures 5N,O).

Successful establishment of the MCAO/R model in rats was confirmed through both behavioral and molecular assessments (Figure 1B). Consistent with the in vitro findings, the MCAO/R group showed decreased GSH and increased MDA levels in the ischemic penumbra (Figures 1C,D). The ischemic penumbra region, delineated in Figure 1A, exhibited significant pathological alterations, including neuronal shrinkage, reduced cell density, and vacuolar degeneration (Figure 1E). Molecular analyses revealed upregulated transcription of DUSP1 and TFRC, coupled with downregulated GPX4 expression (Figures 1F–H). Western blotting confirmed these trends at the protein level, with increased TFRC and decreased GPX4, and significantly elevated DUSP1 protein expression (Figures 1I–L).

Meanwhile, the mRNA expression of PTGS2 in ischemic penumbra of rat was increased (Figure 1M). The result of Western blotting demonstrated that the protein expression of PTGS2 in ischemic penumbra was increased (Figures 1N,O).

Treatment of PC12 cells with CIL56 resulted in decreased cell viability. Notably, co-treatment with CIL56 and the DUSP1 inhibitor BCI led to a more pronounced reduction in cell survival compared to CIL56 treatment alone (Figure 6A). Following the induction of ferroptosis by CIL56, intracellular MDA levels were significantly elevated. Importantly, the BCI-treated group exhibited even higher MDA content than cells treated with CIL56 alone (Figure 6B). Consistent with these findings, CIL56 treatment increased intracellular ROS levels in PC12 cells. This effect was markedly enhanced when CIL56 was combined with BCI (Figure 6D).

Western blot analysis revealed that CIL56 successfully induced ferroptosis in PC12 cells, as evidenced by decreased GPX4 expression and increased TFRC levels. Interestingly, DUSP1 expression was upregulated in response to CIL56 treatment (Figure 6C). The ferroptosis effects were further aggravated upon BCI co-treatment, with more pronounced changes in both GPX4 and TFRC protein levels compared to the CIL56-only group (Figures 6E–G). These results collectively suggest that pharmacological inhibition of DUSP1 potentiates ferroptosis in PC12 cells, indicating that DUSP1 may function as a negative regulator of ferroptosis.

Through systematic analysis, we constructed a ceRNAs network. Validation using the GSE110993 and GSE195442 datasets confirmed the involvement of these regulatory molecules, as illustrated by the Sankey plots (Figure 7A). Notably, miR-101-3p emerged as a key regulator capable of simultaneously targeting both DUSP1 and PTGS2. Subsequent experiments in both the OGD/R and MCAO/R model demonstrated significant downregulation of miR-101-3p (Figures 7B,C), establishing its role in ferroptosis regulation following CI/RI.

Further investigation revealed that circ_0093708, through competitive binding with miR-101-3p, promotes the expression of both DUSP1 and PTGS2. Remarkably, circ_0093708 exhibited exceptional diagnostic potential for cerebral infarction in the GSE195442 dataset, with an AUC of 0.93, sensitivity of 90%, and specificity of 90% (Figure 7D). The proposed regulatory mechanism of the circ_0093708/miR-101-3p/DUSP1/PTGS2 axis is comprehensively illustrated in Figures 7E,F. This integrated approach not only confirms the central role of DUSP1 in ferroptosis following CI/RI but also identifies a novel ceRNAs network with significant diagnostic and therapeutic potential.