All animal procedures were approved by the Jinling Hospital Animal Care and Use Committee and were conducted in accordance with the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health (Publication Nos. 80–23). Male C57/BL6 mice (6–8 w, 22–26 g) were purchased from Nanjing University (Nanjing, China). These mice were kept in an air-conditioned room with air humidity (50 ± 10%) and constant temperature (26 ± 2°C), under a 12 h light/dark cycle. All animals were provided with adequate food and water.

Human astrocytes (HA) cells were purchased from BeNa Culture Collection (Kunshan, China) and were cultured in a medium containing 90% high glucose DMEM, 10% fetal bovine serum (purchased from Gibco BRL, Grand Island, NY, USA). Human cerebral microvascular endothelial cells (hCMEC/D3) were purchased from ABM technology (Zhenjiang, China) and were cultured in a medium containing 89% Prigrow I Medium (TM001), 10% fetal bovine serum, and 1% penicillin-streptomycin. We created an in vitro SAH model by adding 20 μmol/L of oxyhemoglobin (OxyHb) (H2625, Sigma-Aldrich, St. Louis, MO, USA) to the above medium. The OxyHb we purchased from sigma is a qualified product derived from bovine blood. The hCMEC/D3 cell line is a stable, extensively characterized, and well-differentiated human brain endothelial cell line and it is suitable for research on blood–brain barrier function and the response of brain vascular endothelium to inflammation (Weksler et al., 2013).

An experimental SAH model was established by injecting autologous arterial blood into the prechiasmatic cistern of the optic chiasm as described previously, with minor modifications (Deng et al., 2021). The mice were anesthetized using 100% O₂ containing 2% isoflurane and fixed on a stereotaxic apparatus while maintaining anesthesia with 1% isoflurane. Under sterile conditions, we made an incision in the midline of the mouse scalp to expose the skull and subsequently drilled a 0.9 mm depth hole which located 4 mm in front of the bregma. A 27-gauge needle was passed through the burr hole at a 40° angle to the base of the skull. For the SAH group, 50 μl non-heparinized autologous femoral arterial blood was slowly injected into the prechiasmatic cisterna of mice using a syringe pump over 40 s. For the sham group, we injected an equal volume of normal saline. The 27-gauge needle was gently withdrawn 2 min after the blood/saline injections to prevent the backflow of blood and the leakage of cerebrospinal fluid. After sealing the hole using bone wax, the incision was sutured immediately. After injection, we monitored the mice for 40 min for recovery and then returned to their cages.

The schematic diagram of experimental design is demonstrated in Supplementary Figure 1. Our research is containing in vivo and in vitro parts. For the in vivo experiments, we first used laser speckle contrast imaging (LSCI) to compare the CBF of the sham and the 12 h groups after SAH (n = 6 each group). The mice in sham group were euthanized 12 h after operation (n = 6). We measured the time change of NOX2 expression in the temporal lobe of mice, and the change of NO and DHFR levels in endothelial cells extracted from the temporal lobe (12 h, 24 h, and 3 days; n = 6 each group). According to the results of the above experiments, we chose 12 h for further experiments. Then, 12 mice were randomly divided into sham group (n = 6), and 12 h group (n = 6), and collected their brain samples for immunofluorescence staining (Supplementary Figure 1A). To verify the hypothesized pathway, the mice were randomly divided into the following six groups: SAH, Vehicle + SAH, GSK + SAH(1), GSK + SAH(2), AAV-shDHFR + GSK + SAH, and AAV-shCtrl + GSK + SAH groups. For GSK/Vehicle + SAH group, two hours after the injection of GSK/Vehicle into the lateral ventricle of mice, we began to induce experimental SAH in vivo. For AAV-shCtrl/AAV-shDHFR + GSK + SAH group, AAV-shCtrl/AAV-shDHFR were administrated via tail vein injection 14 days before GSK injection. Before euthanasia of the mice in the above experiment, Neurological Score (n = 6 each group) and LSCI (n = 6 each group) were performed. As previously mentioned, their temporal lobe samples were gathered for NO content measurement (n = 6 each group), Western blot (n = 6 each group) and immunofluorescence staining (n = 6 each group) (Supplementary Figure 1B). For in vitro experiments, HA cells contained five groups: Control, OxyHb(1), OxyHb(2), Vehicle + OxyHb, and GSK + OxyHb groups. hCMEC/D3 cells contained the Control group and the OxyHb group. HA cells were co-cultured with hCMEC/D3 cells to explore the effect of astrocytes on endothelial cells, and the co-cultured cells were divided into six groups: AS + OxyHb, AS + Vehicle + OxyHb, AS + GSK + OxyHb(1), AS + GSK + OxyHb(2), LV-shCtrl + AS + GSK + OxyHb, and LV-shDHFR + AS + GSK + OxyHb groups. In the AS + Vehicle/GSK + OxyHb group, after using GSK/Vehicle alone on astrocytes for 2 h, we co-cultured astrocytes with endothelial cells and replaced the medium containing GSK/Vehicle with the medium containing OxyHb without GSK/Vehicle. We used DMSO as the vehicle in the in vitro experiments, and the dose of GSK2795039 was 20 μmol/ml. For LV-shCtrl/LV-shDHFR + AS + GSK + OxyHb groups, 48 h before co-culture, LV-shCtrl/LV-shDHFR were added to hCMEC/D3 cell culture medium. Vehicle/GSK were added to HA cell culture medium 2 h before co-culture. When HA cells were co-cultured with hCMEC/D3 cells, the culture medium containing OxyHb without GSK/Vehicle was replaced. The 6-well plates cultured cells were collected for Western blot (WB, n = 6 each group) while their corresponding culture medium of hCMEC/D3 cells was collected to evaluate the NO level (n = 6 each group). We also collected the medium of astrocytes to measure the amount of H2O2 released by astrocytes (n = 6 each group). In addition, we did immunofluorescence staining (IF)in 12-well plates cultured cells (n = 6 each group). We used 96-well plates to measure the viability of astrocytes and endothelial cells (n = 6 each group) (Supplementary Figure 1C).

Laser speckle contrast imaging is based on blurring the interference pattern of scattered laser light through the flow of blood cells to instantaneously visualize blood perfusion in the circulation. We have been measured CBF and other experimental operations on mice on a constant temperature pad at 37°C, and we used air conditioning to maintain the room temperature at 28°C. Briefly, mice were anesthetized with gas anesthesia (2% isoflurane in 100% O₂), their heads were fixed in a stereotaxic frame, and they were subjected to continuous inhalation of 1% isoflurane throughout the experiment. We made an incision in the midline of the mouse scalp. The blood flow was detected by a CCD camera, and image acquisition was performed using custom software. The same regions of interest were selected for different cortical samples and analyzed by two independent pathological researchers who were blinded to the groupings.

We isolated cerebral blood vessels in the temporal lobe of brain, as previously published, with minor modifications (Bales et al., 2016). After cerebellum and olfactory bulb were removed, we homogenized the remaining brain tissue in 3 ml 4°C sucrose buffer (5 mM HEPES, 0.32 M sucrose, pH 7.4) with homogenizer. Then the homogenate of brain samples was centrifuged at 1,000 g for 12 min, after which the supernatant mainly containing neurons and the dense white myelin sheath on the top of the granules were discarded. Then we resuspended the particles in 3 ml 4°C sucrose solution and centrifuged them at 1,000 g for 12 min. The final pellet, which contained numerous cerebral blood vessels, was lysed in radioimmunoprecipitation assay which contained 1% phenylmethanesulfonyl fluoride (PMSF) for WB or NO content test.

Nitric oxide content of the extracted cortical blood vessels or culture medium was detected using a total nitric oxide assay kit (S0024, Beyotime). In brief, 80 μl of the extracted sample was added in a 96-well plate at room temperature, protected from light. Then, 7.5 μl of 2 mM nicotinamide adenine dinucleotide phosphate (NADPH), 15 μl of flavin adenine dinucleotide (FAD), and 7.5 μl of nitrate reductase were added to the sample sequentially, mixed, and incubated at 37°C for 30 min. Subsequently, 15 μl of lactate dehydrogenase (LDH) buffer and 15 μl of LDH were added, mixed, and incubated at 37°C for an additional 30 min. Finally, 50 μl of Griess reagent I and 50 μl of Griess reagent II were added, mixed, and incubated for 10 min at room temperature. The absorbance of each well was measured at 540 nm using a microplate reader. We quantified the protein content of the different samples and compared the NO levels at the same protein concentration.

For Western blot, cells, and tissues were lysed in radioimmunoprecipitation assay buffer containing 1% PMSF and 1% phosphatase inhibitor cocktail (P5726, Sigma-Aldrich). The protein content was quantified using the Bicinchoninic acid protein assay (BCA) kit (P0011, Beyotime). The molecular weight of NOX2 is 60 kDa, the molecular weight of DHFR is 20 kDa, the molecular weight of eNOS monomer is 140 kDa, and the molecular weight of eNOS dimer is 280 kDa. The samples were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred to polyvinylidene fluoride membranes. After the membranes were blocked in 5% skim milk at room temperature for 2 h, they were incubated with the following primary antibodies at 4°C overnight: anti-β-actin (1:5,000, 81115-1-RR, Proteintech, Chicago, IL, USA), anti-eNOS (1:500, 9572, Cell Signaling Technology, Danvers, MA, USA), anti-NOX2 (1:1,000, ab129068, Abcam) and anti-DHFR (1:2,000, ab124814, Abcam, Cambridge, MA, USA). The membranes were incubated using the appropriate secondary antibody (1:10,000, 7074, Cell Signaling Technology) for 1 h. The membrane is divided into distinct bands containing β-actin and other proteins of interest. Finally, the immunoreactive bands were detected using the enhanced chemiluminescence system. ImageJ software (National Institutes of Health, Bethesda, MD, USA) was used to analyze the relative densities of at least six independent experiments.

The mouse brain sections (6 μm) were incubated with primary antibodies against NOX2 (1:200, ab133303, Abcam), DHFR (1:200, ab124814, Abcam), GFAP (1:200, 2389127, Invitrogen, Carlsbad, CA, USA), and CD31 (1:200, ab24590, Abcam) at 4°C overnight. GFAP is considered a standard marker of mature astrocytes and it is often used to label astrocytes in IF. CD31 is also commonly used to label endothelial cells in IF. So, we used CD31 as a marker to distinguish endothelial cells and GFAP as a marker to distinguish astrocytes. The brain slices were incubated using the corresponding goat anti-rabbit IgG antibodies (1:250, SA00009-2, Proteintech) or goat anti-mouse IgG (1:250, SA00013-1, Proteintech) for 1.5 h at room temperature. Subsequently, the brain sections were sealed with anti-fluorescence solution after incubated DAPI (C1005, Beyotime) for 6 min. Finally, the sections were observed and captured under the fluorescence microscope. Immunofluorescence staining of hCMEC/D3 and HA was similar to brain sections after fixing cells with 4% paraformaldehyde.

The co-culture system consisted of lower and upper chambers separated by a selectively permeable membrane (Corning, Transwell 3450) with 0.4 μm-diameter pores. HA cells were plated in the upper chamber, hCMEC/D3 cells were plated in the lower chamber, and both were co-cultured to mimic neurovascular units.

We measure cell viability using the Cell Counting Kit 8 (CCK-8). HA was seeded in 96-well plates at a density of 5,000 cells/well, and then treated as described in the experimental design for different groups. Because 96-well plates cannot be used for co-cultivation, we diluted different groups of hCMEC/D3 to the same multiple after treatment and seeded them in 96-well plates. After the cells were grown in the incubator for 24 h, they were washed 3 times with medium. Cells were subsequently incubated with CCK-8 solution and culture medium (diluted 1:10) at 37°C for 2 h, and then the absorbance was measured at 450 nm with a microplate reader.

Hydrogen peroxide was measured in vitro using a hydrogen peroxide detection kit (S0038, Beyotime) and 96-well plates as previously described (Gao et al., 2022). Briefly, the media of astrocytes after different groups of treatments were collected. We added 50 μl of sample to each well, followed by 100 μl of detection reagent, mixed them and incubated them at room temperature for 30 min. The absorbance of each well was measured at 560 nm using a microplate reader, and then the hydrogen peroxide level in the medium was calculated according to the standard concentration curve.

The mice were anesthetized using 100% O₂ containing 2% isoflurane and their heads were fixed on a stereotaxic apparatus while maintaining anesthesia with 1% isoflurane. Under sterile conditions, an incision was made in the middle of the mouse scalp to expose the skull, and drills were used to drill 1.0 mm posterior to the bregma and 2.0 mm lateral to the bregma until the dura was penetrated. The vehicle-diluted GSK2795039 (HY-18950, MedChemExpress, NJ, USA) was injected vertically into the lateral ventricle through the hole at a flow rate of 2 μl/min using a syringe pump. For the vehicle group mice, an equal volume of vehicle was injected into the lateral ventricle. In the in vivo experiment, we mixed 10% DMSO, 40% PEG300, 5% Tween-80, and 45% saline as the vehicle, and the dose of GSK2795039 was 3 mg/kg. We kept the vehicle and GSK2795039 in a 37°C water bath before intraventricular injection. The needle was left in place after injection for 5 additional min before immediately sealing the drilled hole using bone wax. We observed the mice for 1 h during recovery before proceeding to the next experimental operation.

Lentiviral vectors which contained shRNA to knock down DHFR (the targeted sequence was CCUCUUCAGUAGAAGGUAATT; Lv-shDHFR) were obtained from Hanbio Biotechnology (Shanghai, China) and 30% proliferating fused hCMEC/D3 was incubated with the Lv-shDHFR mixture for 48 h before the next experiment.

Neurological function was assessed in mice using a modified Garcia Score by two investigators who were blinded to the grouping, using the following six separate tests: locomotion, symmetry of limb movements, forepaw extension, climbing, body proprioception, and response to touch (Gao et al., 2020). After calculating the total of the individual scores of all tests, the lowest neurobehavioral score was 3 points, which represented severely impaired neurological function, and the highest score was 18 points, which represented normal neurological function.

Statistical analysis was performed using Prism 8.02 (GraphPad Software, La Jolla, CA, USA). For each group, the values are expressed as the mean ± standard deviation (mean ± SD). Statistical differences were determined using one-way analysis of variance (ANOVA) when more than two groups were present. When ANOVA showed a significant difference, Tukey’s post-hoc test for multiple comparisons was applied to evaluate that difference. When only two groups were compared, student t-test was utilized. Statistical significance was inferred when the p-value was <0.05.

We applied the prechiasmatic cistern injection model in vivo. We found that the bottom of the brain of mice in the 12 h group was filled with blood compared with the Sham group, demonstrating that the injection model well mimics the situation after SAH in humans (Figure 1A). We performed laser speckle examination on the mice in the Sham and 12 h groups (Figure 1B). The analysis results showed that compared to the sham group, the CBF of the mice in the 12 h group decreased significantly (Figure 1C). We examined the expression of DHFR, eNOS monomer, eNOS dimer, and NO content in cerebral vessels isolated from the temporal lobe. The results showed that compared to the Sham group, the endothelial NO content in the mice in the 12 h groups decreased significantly (Figure 1D). The results of Western blot showed that the eNOS monomer/dimer ratio was significantly increased at 12 h after SAH, which indicated that the uncoupling of eNOS was significantly increased 12 h after SAH (Figures 1E, F). The results of Western blot showed that NOX2 was significantly increased 12 h after SAH (Figures 1G, H). Therefore, we chose to conduct follow-up experiments 12 h after SAH. Likewise, we chose 12 h after SAH as the time point for Immunofluorescence staining analysis. Immunofluorescence staining showed the expression of NOX2 in astrocytes, which was notably increased 12 h after SAH (Figure 1K). The results of Western blot showed that compared to the sham group, DHFR decreased significantly at 12 h, 24 h, and 3 days after SAH, and the decrease was more obvious at 12 h (Figures 1I, J). Furthermore, Immunofluorescence staining showed that DHFR was mainly located in endothelial cells, and the expression of DHFR was significantly decreased 12 h after SAH (Figure 1L).

Subsequently, we used the SAH in vitro model to verify the above results. To explore the effect of astrocytes on endothelial cells, we co-cultured endothelial cells with astrocytes (Figure 2A). The data showed that the NO release from endothelial cells in the lower chamber medium in the OxyHb group was significantly lower than that in the control group and the NO in the medium of AS + OxyHb group was lower than that in the OxyHb group (Figure 2B). According to the results of Western blot, the OxyHb group showed a significant increase in the ratio of eNOS monomer/dimer in endothelial cells compared to the control group (Figures 2C, D). More importantly, the eNOS monomer/dimer ratio was significantly increased in endothelial cells in the AS + OxyHb group when compared to the OxyHb group (Figures 2C, D). The expression of DHFR also showed opposite trends to those of the eNOS monomer/dimer ratios (Figures 2E, F). Consistent with the results of Western blot, Immunofluorescence staining indicated that DHFR expression was significantly decreased in endothelial cells 12 h after SAH (Figure 2G). Taken together, these findings indicate that the presence of astrocytes increases endothelial cell eNOS uncoupling after SAH. Simultaneously, we also found that the expression of NOX2 in astrocytes increased significantly 12 h after adding OxyHb (Figures 2H, I). Moreover, in line with the results of Western blot, Immunofluorescence staining showed similar changes in NOX2 expression in astrocytes from diverse groups (Figure 2J).

Because the above in vivo and in vitro results were similar, we set out to validate our pathway using NOX2 inhibitors and Lv-shDHFR. To further verify the effect of NOX2 in astrocytes on endothelial cells, we used GSK2795039 to inhibit NOX2 in astrocytes and co-cultured astrocytes with endothelial cells. After adding OxyHb for 12 h, we found no significant difference in NO content, eNOS monomer/dimer ratio, and DHFR expression in the AS + OxyHb and AS + vehicle + OxyHb groups but surprisingly found that compared to the AS + OxyHb group, the NO content was significantly increased in the AS + GSK + OxyHb group (Figure 3A). The results of Western blot showed that compared to the AS + OxyHb group, the eNOS monomer/dimer ratio in the AS + GSK + OxyHb group was significantly decreased (Figures 3B, C) and the expression of DHFR was significantly increased (Figures 3D, E). Moreover, compared to the AS + OxyHb group, the DHFR expression was significantly increased in endothelial cells in the AS + GSK + OxyHb group (Figure 3F). We found that there was no significant difference in the viability of astrocytes in the OxyHb group, Vehicle + OxyHb group, and GSK + OxyHb group (Supplementary Figure 2A). However, we found that the viability of endothelial cells in the AS + GSK + OxyHb group was significantly increased compared with the AS + OxyHb group, while there was no significant difference in the viability of endothelial cells between the AS + OxyHb group and the AS + Vehicle + OxyHb group (Supplementary Figure 2B). We found that compared with the Control group, the H₂O₂ released from astrocytes in the medium in the OxyHb group was significantly increased, while compared with the OxyHb group, the H₂O₂ in the GSK + OxyHb group was significantly decreased (Supplementary Figure 3). This proves that a large part of H₂O₂ released by astrocytes after SAH is mediated by NOX2. These findings demonstrate that eNOS and DHFR are downstream effectors of NOX2 in astrocytes.

We demonstrated the knockdown efficacy of Lv-shDHFR in endothelial cells by Western blot (Figures 3G, H). The results showed no significant difference in NO content or eNOS monomer/dimer ratio in the AS + GSK + OxyHb and Lv-shCtrl + AS + GSK + OxyHb groups. The Western blot results showed that compared to the AS + GSK + OxyHb group, the eNOS monomer/dimer ratio was significantly increased in the Lv-shDHFR + AS + GSK + OxyHb group (Figures 3I, J). Compared to the AS + GSK + OxyHb group, the amount of NO released by endothelial cells was significantly decreased in the Lv-shDHFR + AS + GSK + OxyHb group (Figure 3K). These results demonstrate that knockdown of DHFR reduces the protective effect of GSK2795039. These results also confirmed the existence of our conjectured pathway.

We then set out to validate our pathway in vivo. For in vivo experiments, we injected GSK2795039 into the ventricles of mice to inhibit NOX2. The laser speckle diagram showed that there was no significant difference in CBF between the SAH group and the vehicle + SAH group, however, compared to the SAH group, mice in the GSK + SAH group had increased CBF (Figures 4A, B). Simultaneously, we found no significant difference in endothelial NO content, eNOS monomer/dimer ratio, and DHFR expression in the SAH and vehicle + SAH groups. However, compared to the SAH group, the endothelial NO content was significantly increased in the GSK + SAH group (Figure 4C). Moreover, as shown in the cell experiments, compared to the vehicle + SAH group, the ratio of eNOS monomer/dimer and the expression of DHFR were increased significantly in the GSK + SAH group (Figures 4D–G). After injection of GSK2795039, the deteriorating neurological function was also significantly relieved (Figure 4H). Moreover, the results of Immunofluorescence staining showed that compared to the vehicle group, the NOX2 inhibitor significantly increased the expression of DHFR in endothelial cells (Figure 4I). Similar to the results of cell experiments, these findings also demonstrate that eNOS and DHFR are downstream effectors of NOX2.

We first knocked down DHFR by injecting an adeno-associated virus (AAV) into the tail vein of mice, and the knockdown efficacy of AAV-shDHFR was demonstrated by Western blot (Figures 5A, B). The laser speckle diagram demonstrates no significant difference in CBF between the GSK + SAH and AAV-shCtrl + GSK + SAH groups, however, compared to the GSK + SAH group, mice in the AAV-shDHFR + GSK + SAH group had increased CBF (Figures 5C, D). We also found no significant difference in endothelial NO content and eNOS monomer/dimer ratio between the GSK + SAH and AAV-shCtrl + GSK + SAH groups. The data showed that compared to the GSK + SAH group, the endothelial NO content was significantly decreased in the GSK + AAV-shDHFR + SAH group (Figure 5E). The Western blot results showed that the eNOS monomer/dimer ratio in the GSK + AAV-shDHFR + SAH group was significantly increased compared to that in the GSK + SAH group (Figures 5F, G). The protective effect of GSK2795039 in attenuating neurological deficits was also abolished by DHFR knockdown in endothelial cells (Figure 5H). This result proves that DHFR is a downstream factor of NOX2, and its knockdown counteracts the partial alleviation of acute cerebral ischemia in SAH by NOX2 inhibitors.