Male Sprague-Dawley rats were acquired from the Experimental Animal Center of Central South University (Changsha, China). GPX4 Fos^CreERT2 mice (GPX4 Fos^CreERT2/− and GPX4 Fos^CreERT2/+, GPX4 Fos^CreERT2/− mice were used as controls for GPX4 Fos^CreERT2/+ mice to rule out the influence of the Cre recombinase or other experimental factors) were produced using GPX4-flox mice and Fos^CreERT2 mice (both obtained from GemPharmatech Co., Ltd.). It was confirmed that the expression levels of GPX4 were reduced in the brain tissues of GPX4 Fos^CreERT2/+ mice (Fig. S1). The rats that were aged 8 weeks and weighed 280±20 g underwent Ferrostatin-1 (Fer-1) or short hairpin (sh)RNA treatments, whereas the rats that were aged 6 weeks and weighed 130-150 g were treated with for adeno-associated virus. The mice were aged 8-12 weeks and weighed 22±25 g. All animals were kept in a room with a controlled temperature of 25°C and humidity at 40-60% under a 12-h light/dark cycle, with free access to food and water.

A total of 78 rats and 30 GPX4 Fos^CreERT2 mice were used in the present study. For the Evans Blue extravasation test in vivo, 3 rats were allocated to the sham group, and 6 rats were allocated to the I/R 12-h group and the I/R 24-h group. For the ZO-1 and p23 immunofluorescence assay in vivo, 3 rats were allocated to the sham group and 6 rats were allocated to the I/R 12-h group. To assess the effects of Fer-1 on ZO-1 expression in vivo, 3 rats were allocated to the sham + vehicle group and the sham + Fer-1 group, whereas 6 rats were allocated to the I/R + vehicle group and I/R + Fer-1 group. For the p23 shRNA test in vivo, the sham + shNC group and sham + p23 shRNA group were each allocated 3 rats, and the I/R + shNC group and I/R + p23 shRNA group were each allocated 6 rats. For the p23 overexpression (OE) test in vivo, the sham + vector group and sham + OE p23 group were each allocated 3 rats, and the I/R + vector group and I/R + OE p23 group were each allocated 6 rats. For the GPX4 Fos^CreERT2 immunofluorescence assay in vivo, 3 GPX4 Fos^CreERT2/− and 3 GPX4 Fos^CreERT2/+ mice were allocated to the sham + OE-vector group, and the I/R + OE-vector group and I/R + OE-p23 group were each allocated 6 GPX4 Fos^CreERT2/− mice and 6 GPX4 Fos^CreERT2/+ mice. All experiments and animal procedures were approved by the Experimental Animal Center of Central South University (approval no. 2018sydw0222).

The mouse BMEC line bEnd.3 was procured from The Cell Bank of Type Culture Collection of The Chinese Academy of Sciences. The cells were cultured in DMEM (Gibco; Thermo Fisher Scientific, Inc.) supplemented with 10% FBS (Gibco; Thermo Fisher Scientific, Inc.) and 1% penicillin-streptomycin (Gen-View Scientific, Inc.) at 37°C and 5% CO₂. bEnd.3 cells were used to establish an in vitro model of the BBB, and the oxygen-glucose deprivation/reoxygenation (OGD/R) experiment was used to simulate I/R injury in vitro. As for the administration of Fer-1 (Sigma-Aldrich; Merck KGaA), bEnd.3 cells were exposed to 10 µM Fer-1 at 37°C for 2 h before OGD/R. For the cycloheximide (CHX) chase assay, CHX (10 µM; cat. no. NSC-185; Selleck Chemicals) was applied in OGD/R-insulted b.End 3 cells for 3-12 h at 37°C to inhibit the proteasome pathway for the degradation of TJ proteins. The GPX4 agonist PKUMDL-LC-101-D04 (200 µM; CAS no. 2143896-83-5; MedChemExpress) was used to treat the cells for 1 h at 37°C after OGD/R and p23 siRNA transfection, whereas the GPX4 inhibitor RAS-selective lethal 3 (RSL3; 200 nM; CAS no. 1219810-16-8; MedChemExpress) was used to treat the cells for 16 h at 37°C after OGD/R and p23 oe.

siRNA transfection was performed using the riboFECT™ CP Transfection Kit (Guangzhou RiboBio Co., Ltd.) according to the manufacturer's instructions. Briefly, 1×10⁶ cells/well were plated in 6-well plates and cultured for 12-24 h until the cell density reached 50-60%. The transfection complex was prepared by diluting 5 µl 20 µM siRNA storage solution with 120 µl 1X riboFECT CP Buffer, adding 12 µl riboFECT CP Reagent and incubating at room temperature for 15 min. The transfection complex was then added to the medium and cultured in a CO₂ incubator at 37°C for 24 h. A negative control (NC) siRNA that does not target any known human, mouse or rat genes, and is not chemically modified, was used as a NC (cat. no. siN0000001-1-10; Guangzhou RiboBio Co., Ltd.). The of p23 siRNA sequence was as follows: 5′-GGTCCTGTTTGAACAAAGA-3′. The cell samples were subjected to OGD. After 3 h of OGD exposure, the cells were cultured in normal medium containing 10% FBS within a cell incubator containing 5% CO₂ at 37°C for 12 h.

The p23 OE plasmid, truncated p23 plasmids and blank vector plasmid (GV486; both from Shanghai GeneChem Co., Ltd.) were transfected into cells using the Neofect™ DNA transfection reagent (Neofect Biotech Co., Ltd.) according to the manufacturer's protocol. The truncated p23 sequences are provided in Table I. A total of 1×10⁶ cells/well were plated in 6-well plates and cultured for 12-24 h until they reached a cell density of 60-80%. The DNA diluent was prepared by adding 100 µl Opti-MEM serum-free (Gibco; Thermo Fisher Scientific, Inc.) to 2 µg plasmid. Subsequently, 2 µl Neofect was added to the DNA diluent and incubated at room temperature for 15-30 min. Finally, the complex was added to the cells in 6-well plates and cultured in a CO₂ incubator at 37°C for 48 h.

Cell viability was assessed using the Cell Counting Kit-8 (CCK-8; cat. no. GK3607; Gen-View Scientific, Inc.). A total of 5,000 cells were seeded in a 96-well plate and were cultured for ~24 h. Following treatment, CCK-8 was added to the medium in each well at a 1:10 ratio and the cells were incubated at 37°C for 1-4 h. Subsequently, the colorimetric absorbance of each well was measured at 450 nm using a microplate reader and a ratio was calculated compared with the control group.

Following extraction with TRIzol^® reagent (Invitrogen; Thermo Fisher Scientific, Inc.), 1 µg RNA from cells or animal brain samples was reverse transcribed using 5X Evo M-MLV RT Reaction Mix (Accurate Biology) according to the manufacturer's protocol. Subsequently, qPCR was carried out according to the manufacturer's protocol using the SYBR Premix Ex Taq II kit (Accurate Biology) with the following amplification conditions: Pre-denaturation at 95°C for 30 sec, followed by 40 cycles at 95°C for 5 sec and 60°C for 34 sec. The primer sequences used are presented in Table II. The relative quantification of mRNA was analyzed using the 2^−ΔΔCq method (22).

Proteins from cells or animal brain samples were extracted using radioimmunoprecipitation assay lysis buffer (Guangzhou Dingguo Biotechnology Co., Ltd.) containing 1% PMSF, and were quantified using the BCA method. Subsequently, 20-30 µg cell proteins or 50-80 µg animal proteins were transferred to a PVDF membrane following SDS-PAGE on 10-15% gels. Subsequently, the membranes were blocked with TBS-1% Tween 20 (TBST) containing 5% skim milk powder at room temperature for 2 h. The membranes were then incubated with primary antibodies against β-actin (cat. no. A1978; dilution, 1:5,000; Sigma-Aldrich; Merck KGaA), p23 (cat. no. 160150; dilution 1:200; Cayman Chemical Company), zonula occludens-1 (ZO-1; cat. no. 402200; dilution 1:2,000; Invitrogen; Thermo Fisher Scientific, Inc.), occludin (cat. no. 404700; dilution 1:1,000; Invitrogen; Thermo Fisher Scientific, Inc.), GPX4 (cat. no. ab125066; dilution 1:3,000; Abcam), cyclooxygenase (COX)-2 (cat. no. 160106; dilution 1:200; Cayman Chemical Company), heat shock protein 90 (HSP90; cat. no. 60318-1-lg; dilution 1:2,000; Proteintech Group, Inc.) and Flag (cat. no. 20543-1-AP; dilution 1:2,000; Proteintech Group, Inc.) at 4°C overnight. The membranes were washed with TBST three times at 10-min intervals and were then incubated with the HRP-conjugated goat anti-rabbit IgG (cat. no. BA1055; dilution 1:5,000; Boster Biological Technology) or goat anti-mouse IgG (cat. no. BA1056; dilution 1:5,000; Boster Biological Technology) secondary antibodies at room temperature for 1 h, followed by three further washes with TBST at 10-min intervals. The color reaction was developed using the Clarity Western ECL Substrate (cat. no. 170-5061; Bio-Rad Laboratories, Inc.), and band density was measured using ImageJ software V 1.8.0 (National Institutes of Health) and was normalized to β-actin.

After fixing with 4% paraformaldehyde for 10 min at room temperature, 5,000 bEnd.3 cells grown in 96-well plates were washed three times with PBS (5 min each) and were blocked with BSA (Gen-View Scientific, Inc.) at 37°C for 1 h. The cells were then incubated with a primary antibody against ZO-1 (cat. no. 61-7300; dilution 1:50; Invitrogen; Thermo Fisher Scientific, Inc.), p23 (dilution 1:20; cat. no. 160150; Cayman Chemical Company), CD31 (cat. no. ab9498; dilution 1:1,000; Abcam) and GPX4 (dilution 1:100; cat. no. ab125066; Abcam) at 4°C overnight. The cells were then incubated with a TRITC-conjugatedsecondary antibody (cat. no. AS040; dilution 1:1,000; ABclonal Biotech Co., Ltd.) at 37°C in the dark for 1 h. Nuclei were stained with 1X DAPI working solution at room temperature for 5 min, washed three times with PBS-1% Tween 80 (PBST) and images were captured using a fluorescence microscope (Olympus Corporation).

The rats or GPX4 Fos^CreERT2 mice were anaesthetized through an intraperitoneal injection of 50 mg/kg sodium pentobarbital (China National Pharmaceutical Group). An incision was made in the neck to expose the external carotid, internal carotid, right common carotid and common carotid arteries, and the internal carotid artery was clamped. Following the ligation of the external carotid artery, a small incision was made into the external carotid artery. Monofilaments coated with poly-L-lysine were inserted via the incision, through the internal carotid artery and ultimately to the middle cerebral artery. Following occlusion (2 h for rats and 1 h for mice), the nylon monofilament was withdrawn for reperfusion for 24 h. The sham group underwent the same procedure as the MCAO group without insertion of the monofilament. The survival and health status of the animals was observed every 3 h. The cortex of the infarcted hemisphere was collected for subsequent experiments at 12 or 24 h post-reperfusion. The animals were euthanized by rapid dislocation of the head and neck when it was time to collect the specimens, or when the animals reached the following humane endpoints: They were too weak to eat or drink for up to 24 h; they were in a moribund state, i.e., without anesthesia or sedation, they were showing mental depression accompanied by hypothermia (<37°C). The loss of respiratory movement (the rise and fall of the chest and abdomen), and reflex actions of the eyes, nose and feet were used to verify the death of the animals. Notably, after MCAO treatment, 12 rats and 8 mice were sacrificed before sample collection, and 6 rats and 4 mice died spontaneously.

After anesthetizing with 50 mg/kg sodium pentobarbital and preserving the skin on the top of the head, the rats were positioned horizontally and prone on a stereotaxic frame (RWD Life Science). Subsequently, 40 nmol/kg Fer-1, with saline used as a vehicle control; or p23 shRNA [forward (5′-3′): CCG GGA TCC TAG ACC ATG GAT TTA ACT CGA GTT AAA TCC ATG GTC TAG GAT CTT TTT G, reverse (5′-3′): AAT TCA AAA AGA TCC TAG ACC ATG GAT TTA ACT CGA GTT AAA TCC ATG GTC TAG GAT C; pRNAT-U6.1/Neo, Guangzhou RiboBio Co., Ltd.], with shNC [forward (5′-3′): TAC GAT ACA AGG CTG TTA GAG AG, reverse (5′-3′): TAG AAG GCA CAG TCG AGG used as the control] was infused into the right lateral ventricle at a flow rate of 0.5 µl/min using a syringe pump (KD Scientific) in conjunction with a Hamilton syringe. The stereotaxic coordinates for the right lateral ventricle were -1.5 mm laterally, -1.2 mm anteroposteriorly and -4.5 mm dorsoventrally, with the bregma point serving as the origin. The p23 OE adeno-associated virus (AAV-9; Shanghai GeneChem Co., Ltd.) was injected into the right lateral ventricle -1.2 mm laterally, -0.8 mm anteroposteriorly and -3.5 mm dorsoventrally, with empty AAV-9 used as the vector control. Following the injection, the skull was sealed using bone wax. OE was induced 2-3 weeks after injection. Similarly, the lateral ventricle injection of GPX4 Fos^CreERT2 mice was performed as that performed in rats; the stereotaxic coordinates were as follows: -1.2 mm laterally, -0.8 mm anteroposteriorly and -3.5 mm dorsoventrally. Subsequently, the animals underwent MCAO modeling and the brain tissue was collected for testing.

In rats, the permeability of the BBB was evaluated through Evans Blue extravasation. A 2% Evans Blue solution was prepared using normal saline prior to the experiment. A total of 1 h before the conclusion of reperfusion, 2% Evans Blue (2 ml/kg) was injected into the tail vein of the rats. After 1 h, the rats were anesthetized with 50 mg/kg sodium pentobarbital and underwent cardiac perfusion with PBS. The brains were then collected and 2-mm thick coronal sections were made.

Cells were subjected to a 3-h culture in glucose- and serum-free medium in an anaerobic incubator (37°C, 95% N₂, 5% CO₂). After 3 h of OGD exposure, the cells were cultured in normal medium containing 10% FBS within a cell incubator containing 5% CO₂ at 37°C for 12 and 24 h. The control group cells were cultured under normoxic condition, while the 0 h OGD/R group underwent OGD without reperfusion. The cell samples were then collected for subsequent experiments.

The bEnd.3 cells were cultured on the upper chamber of gelatin-coated Transwell inserts (0.4 µm; Costar; Corning, Inc.) and were exposed to hypoxic conditions (0% O₂) at 37°C for 3 h. Subsequently, 165 µg/ml Evans Blue -0.1% BSA (Sigma Aldrich; Merck KGaA) was added to the upper chamber 1 h prior to measurements. The intensities of the diffused Evans Blue -0.1% BSA in the lower chamber were subsequently measured at 650 nm using a microplate reader. The results are presented as the ratio of the Evans Blue -0.1% BSA concentration in the lower chamber to the total concentration of Evans Blue -0.1% BSA added to the upper chamber at the onset of the experiment.

Using the fluorescent probe diacetyldichlorofluorescein (DCFH-DA) ROS detection kit (cat. no. S0033S; Beyotime Institute of Biotechnology), DCFH-DA was diluted at a ratio of 1:1,000 with serum-free medium to a final concentration of 10 µM. Subsequently, the cell culture medium was removed from the cells, an appropriate volume of diluted DCFH-DA was added and incubation was carried out in a cell incubator at 37°C for 20 min. The cells were then washed with PBS three times, and a 488-nm excitation wavelength and 525-nm emission wavelength were used to detect the optical density using a microplate reader.

The MDA assay was conducted using a Lipid Peroxidation MDA Assay Kit (cat. no. S0131S; Beyotime Institute of Biotechnology). Following homogenization, the brain or cell samples were centrifuged at 13,000 × g and 4°C for 10 min to obtain the supernatant for detection. Subsequently, the prepared thiobarbituric acid working solution was added and heated at 100°C for 15 min. Eventually, the mixtures were centrifuged at 13,000 × g and 4°C for 10 min to obtain the supernatant, which was transferred to a 96-well plate to measure absorbance at 532 nm using a microplate reader. The content of MDA was normalized with respect to the protein amount of the samples, which was determined using the BCA method.

The GSH assay was conducted using a GSH and reduced glutathione (GSSG) assay kit (cat. no. S0053; Beyotime Institute of Biotechnology) according to the manufacturer's protocol. After OGD/R treatment, the cells were collected and resuspended with protein removal agent M. After rupturing the cell membrane through two freeze-thaw cycles, the cells were centrifuged at 10,000 × g at 4°C for 10 min and the supernatant was used for the determination of total GSH. Part of the supernatant was taken for GSSG detection: GSH removal auxiliary solution and GSH removal reagent working solution were added, and the reaction was performed at 25°C for 60 min. The protein removal reagent M solution and total GSH detection working solution were then added to the samples for total GSH or GSSG detection, and were incubated at 25°C for 5 min, before the addition of NADPH. The absorbance of the sample was detected at 512 nm with a microplate reader. The GSH content was calculated as follows: GSH=total GSH-GSSG x2. The GSH content was normalized with respect to the protein amount of the samples, which was determined using the BCA method.

The structures of HSP90, p23 and GPX4 were downloaded from the Protein Data Bank (PDB) database (https://www.rcsb.org/); HSP90 and p23 are in the same protein file (PDB ID: 7L7J), whereasGPX4 (PDB ID: 6HN3) has its own protein structure file. The interaction modes between GPX4, HSP90 and p23 were investigated using the ZDOCK (https://zdock.wenglab.org/) server.

Briefly, 5,000 cells/well were plated on a 96-well slide. After OGD/R, the cells were washed with PBS and successively fixed with 2.5% glutaraldehyde and 1% osmium acid at room temperature for 6 h. Following dehydration using varying concentrations of acetone, the samples were embedded in acetone and embedding solution at room temperature for 12 h, and solidified in an oven at 60°C for 24 h. Subsequently, the slides were double-stained with 3% uranium acetate and lead nitrate at room temperature for 2-3 min. Finally, the samples were observed and images were captured using a Hitachi HT7700 transmission electron microscope (Hitachi, Ltd.).

This assay was performed using agarose beads (Thermo Fisher Scientific, Inc.). Proteins from cell samples were extracted using radioimmunoprecipitation assay lysis buffer (Guangzhou Dingguo Biotechnology Co., Ltd.). After extraction, 300 µg cellular protein was incubated with the p23 (cat. no. 160150; dilution 1:200; Cayman Chemical Company), GPX4 (cat. no. ab125066; dilution 1:3,000; Abcam), and Flag (cat. no. 20543-1-AP; dilution 1:2,000; Proteintech Group, Inc.) antibodies at 4°C for 1 h with agitation. Subsequently, 20 µl suspended agarose beads were added and incubated at 4°C overnight with agitation. The immunoprecipitation products were collected and centrifuged at 2,500 × g for 5 min at 4°C, and the supernatant was discarded. The agarose beads were washed with PBST and centrifuged again; this was repeated four times. Finally, the agarose beads were resuspended with 40 µl electrophoretic loading buffer, boiled at 85°C for 5 min, and 20 µl samples underwent SDS-PAGE and western blotting.

All experiments were independently repeated at least three times. The results, presented as the mean ± SD, were analyzed using one-way or two-way ANOVA along with the multiple-comparison Bonferroni correction test. Statistical analysis was performed using GraphPad Prism 8 software (GraphPad Software Inc.). P<0.05 was considered to indicate a statistically significant difference.

To investigate the effect of the MCAO model on BBB dysfunction, the Evans Blue extravasation test was used to measure cerebral microvascular permeability. As shown in Fig. 1A, the ischemia group had markedly higher Evans Blue leakage than the sham group. TJ proteins are transmembrane proteins located between endothelial cells of the BBB, which serve critical roles in regulating the paracellular permeability of the BBB (23). The TJ protein ZO-1 was markedly decreased after 12-h I/R (Fig. 1B). To investigate the effects of OGD/R on BBB permeability during hypoxia in vitro, a Transwell assay was used to generate an in vitro BBB model, and BBB permeability was measured in BMECs using Evans Blue staining. The content of Evans Blue in the lower Transwell chamber was higher in the 0 and 12 h OGD/R groups than in the normoxia group (Fig. 1C). Subsequently, western blotting was performed to determine whether OGD/R affected the protein levels of ZO-1 and occludin. The expression levels of ZO-1 and occludin were decreased in BMECs in the OGD/R group at 12 h after reoxygenation compared with those in the control group (Fig. 1D-F). In addition, the expression of the TJ ZO-1 was detected in OGD/R-treated BMECs using immunofluorescence. The results showed that the expression of ZO-1 in BMECs in the OGD/R group was significantly decreased and exhibited a discontinuous distribution (Fig. 1I and J).

Increasing evidence has indicated that ferroptosis in BMECs is the main cause of BBB injury induced by I/R (24,25). Pretreatment with Fer-1 was performed via intraventricular injection for 2 h, followed by MCAO modeling, and the results showed that, as compared with in the sham + vehicle (saline) group, ZO-1 expression in brain tissues was decreased in the I/R + vehicle group, whereas it was increased in the I/R + Fer-1 group compared with that in the I/R + vehicle group (Fig. 2A). In addition, BMECs were pretreated with Fer-1, and it was revealed that the reduction in cell viability induced by OGD/R was significantly suppressed (Fig. 2B). Following OGD/R, BMECs exhibited mitochondrial swelling, and a decrease or disappearance in cristae (Fig. 2C), as well as decreased levels of the antioxidant molecules GSH and GPX4, and increased levels of MDA and ROS (Fig. 2D-J). This indicated that intracellular lipid peroxidation levels were increased. In addition, the expression levels of COX-2, another biomarker of ferroptosis, were significantly increased in the 12 h OGD/R group (Fig. 2K-M). These results suggested that ferroptosis may be involved in OGD/R-induced BMECs injury. Inhibiting ferroptosis in BMECs could protect against BBB damage caused by cerebral I/R.

p23 can bind to client proteins and serves a regulatory role in numerous diseases by regulating their biological functions. In the present study, the in vitro results showed that p23 expression was significantly induced by OGD/R in BMECs (Fig. 1D, G and H) Subsequently, the permeability of BMECs was examined following OE or knockdown of p23 in BMECs. First, in the OGD/R model, siRNA was used to knockdown p23 expression (Fig. 3A), the results showed that p23 and ZO-1 expression levels were inhibited by p23 knockdown (Fig. 3C and D), and the permeability of BMECs was clearly increased (Fig. 3B), which indicated that the injury of BMECs induced by OGD/R was aggravated by p23 knockdown. Transfection with the p23 OE plasmid significantly increased the expression of p23 in the OGD/R model (Fig. 4A-C). Compared with in the vector + OGD/R group, the content of Evans Blue in the p23 + OGD/R group was significantly decreased (Fig. 4D and E), and the expression of ZO-1 was increased (Fig. 4F), which indicated that the OE of p23 reversed the OGD/R-induced damage of BMECs. In vivo, p23 knockdown by shRNA reduced the expression of ZO-1 in the MCAO group compared with that in the MCAO + shNC group (Fig. 3E), while p23 OE in the MCAO group could rescue the decrease in ZO-1 compared with in the MCAO + vector group (Fig. 4G); CD31 was used to indicate the location of blood vessels.

Given that ferroptosis is involved in BBB damage in ischemic stroke, the effect of p23 on OGD/R-induced ferroptosis in BMECs was determined. Compared with in the control group, the mortality of BMECs increased by p23 knockdown (Fig. 5A). In addition, p23 was overexpressed at the cellular level, and p23 could rescue cell death induced by OGD/R (Fig. 5B). In the OGD/R group, transfection with the p23 siRNA further decreased the contents of GSH and GPX4, and increased the levels of ROS, MDA, and acyl-CoA synthetase long-chain family member 4 (ACSL4) and subunit solute carrier family 7 member 11 (SLC7A11; Fig. 5C-I) compared with in the OGD/R + NC group, indicating that the increase of lipid peroxidation in BMECs promoted OGD/R-induced ferroptosis. After OGD/R, p23 OE decreased the levels of ROS, MDA, ACSL4 and SLC7A11, and increased those of GSH and GPX4 compared with those in the vector group (Fig. 5J-P), indicating that the levels of antioxidants in BMECs were increased, inhibiting OGD/R-induced ferroptosis. These results suggested that the regulatory role of p23 in OGD/R-induced BMECs may be achieved through ferroptosis.

Notably, p23 can function as an independent molecular chaperone or as a co-chaperone of HSP90 to exert its biological functions (26). To further explore the mechanism of p23-induced inhibition of ferroptosis, protein docking showed that p23 formed a complex with HSP90 and GPX4 through its N-terminal domain (1-90aa) (Fig. 6A). Co-IP experiments demonstrated that anti-GPX4 antibodies could precipitate p23 and HSP90 proteins in cells. Similarly, anti-p23 antibodies could precipitate GPX4 and HSP90 proteins in cells. (Fig. 6B). Full-length and truncated p23 plasmids were then constructed: Flag-p23(1-160), N-terminal Flag-p23(1-90) and C-terminal Flag-p23(91-160) fragments. After transfecting the BMECs with the truncated plasmids, immunoprecipitation was performed with anti-Flag antibody. The results showed that GPX4 could be precipitated by anti-Flag antibody, and the most important binding region was p23(1-90), which indicated that p23 formed a complex with HSP90 and GPX4 to inhibit GPX4 degradation and stabilize the expression of GPX4 through the N-terminal domain (1-90aa) of p23, thus inhibiting ferroptosis (Fig. 6C). In addition, immunofluorescence co-localization verification showed that p23 and GPX4 were co-located in cells (Fig. 6D). The effect of p23 on the stability of GPX4 protein was then tested. The results showed that the stability of GPX4 protein was markedly increased following p23 OE (Fig. 6E). To further verify the experimental results, GPX4 Fos^CreERT2 mice were used to construct a model of MCAO. The results showed that the reduction of ZO-1 in GPX4 Fos^CreERT2/− mice in the MCAO model was partially reversed by p23 OE, while p23 OE did not have this effect on GPX4 Fos^CreERT2/+ mice (Fig. 6F); CD31 was used to indicate the location of blood vessels, indicating that p23 may have a protective role by regulating ferroptosis through GPX4. The present findings demonstrated that p23 may form complexes with HSP90 and GPX4 through its N-terminal domain (1-90aa), thus enhancing the stability of GPX4, inhibiting the degradation of GPX4 and ultimately inhibiting BMEC ferroptosis, thereby protecting against cerebral I/R-induced BBB damage.