Inflamed vessel–anchored release of H₂ across the blood-brain barrier for ischemic stroke neuroprotection

Abstract

Ischemic stroke followed by reperfusion urgently requires safe and efficient cytoprotective strategies, a need still unmet by current pharmacotherapies. Nanotechnology holds promise for improved drug delivery to the brain, yet the efficacy of nanomaterials crossing the blood-brain barrier (BBB) is quite limited, and long-term intracranial retention of nanomaterials may provoke neurotoxicity. Leveraging the anti-inflammatory, BBB-crossing, and biosafe properties of hydrogen (H₂), we develop an inflamed vessel–targeted/anchored H₂–producing system by modifying ZrSi₂ nanoparticles with a P-selectin–binding peptide (ZSNP), mimicking P-selectin/P-selectin glycoprotein ligand–mediated innate immune recruitment. Rather than relying on nanoparticle penetration into the brain parenchyma, this design enables ZSNP to anchor at the BBB vasculature, where it locally and continuously generates H₂ via hydrolysis. The released H₂ traverses the BBB, exerting cytoprotection through antioxidant and immunomodulatory mechanisms that coordinate multicellular recovery processes. Furthermore, ZSNP promotes microglia-mediated angiogenesis and neurogenesis, guides axonal projections along neovascular trajectories, and facilitates microglia-neuron interaction via the noncanonical Wnt/Ca²⁺ pathway. This reconstruction of the neurovascular network supports the reintegration of functional neural circuits, leading to structural and functional recovery that surpasses the effects of edaravone. By enabling sustained H₂ release at the BBB interface without requiring nanoparticle intracranial accumulation, this strategy represents a promising and low-burden neuroprotective approach for ischemic stroke.

The BBB vasculature–anchored H2–releasing platform, without requiring nanoparticle penetration, provides superior neuroprotection.

INTRODUCTION

Ischemic stroke represents a critical global health concern, significantly contributing to morbidity and mortality due to its devastating impact on neurological function (1). Although reperfusion therapies such as intravenous thrombolysis and endovascular thrombectomy have improved the management of ischemic stroke, they fail to adequately address the pathological cascade initiated by ischemic reperfusion. Consequently, there has been a resurgence of interest in cytoprotective strategies aimed at preserving vulnerable neurons and other brain cells during the ischemic cascade (2). Despite extensive research into cytoprotective agents (such as excitotoxicity inhibitors, antioxidants, immunomodulators, etc.), their therapeutic efficacies have been limited by several issues, including restricted blood-brain barrier (BBB) penetration, narrow therapeutic time window associated with monotherapy, and potential systemic toxicity (2, 3). A prominent example is edaravone, a free radical scavenger that has been widely used for stroke therapy (4). Nevertheless, its low bioavailability, limited BBB penetration, and potential hepatotoxicity hinder its overall therapeutic effectiveness (5). To date, no pharmacological agent has demonstrated unequivocal efficacy in improving functional recovery in large-scale stroke trials, underscoring the need for innovative therapeutic strategy.

Nanotechnology has emerged as a transformative approach for stroke treatment, leveraging precision targeting, enhanced BBB permeability, and controlled drug delivery (6, 7). However, following critical issues urgently demand a solution (8). First, the penetration of BBB continues to pose a significant challenge. Although receptor-targeted systems (e.g., transferrin receptor–mediated) (9, 10) and biomimetic platforms (e.g., membrane-coated nanoparticles, exosomes) (11, 12) demonstrate improved BBB permeability, their delivery efficiency remains suboptimal. Focused ultrasound–mediated BBB opening can enhance penetration efficiency within a specific time window (13, 14), but this method carries substantial risks as BBB disruption directly correlates with poor neurological outcomes. Second, rapid and undifferentiated tissue distribution of therapeutic agents remains a critical but unresolved challenge for achieving optimal treatment efficacy in nanomedicine (15, 16). Currently, no agents have been identified to be able to achieve BBB traversal and rapid brain parenchyma distribution within minutes. Third, even when successful penetration of the BBB is achieved, excessive accumulation of nanomaterials within the brain will induce endoplasmic reticulum stress and trigger neurotoxicity (17). Consequently, overcoming these obstacles to facilitate stroke recovery remains a challenge.

Compared to traditional drug molecules, molecule hydrogen (H₂) offers several distinct advantages, such as exceptional BBB permeability, rapid tissue distribution, high biosafety, and antioxidative as well as immunomodulatory effects (18–20). Recent studies in both animal models and stroke patients have demonstrated that H₂ can reduce cerebral infarct volume, alleviate brain edema, mitigate brain injury severity, and enhance neural cell viability and neurological function (21). Furthermore, H₂ has been shown to improve 90-day survival in out-of-hospital patients with cardiac arrest without neurological deficits (22). No adverse effects have been reported in cellular or animal studies (21, 23). These findings highlight the favorable safety profile and translational potential of H₂ in neuroprotective applications. However, because of its rapid diffusion and lack of targetability, neither intravenous/oral administration of H₂-rich water nor inhalation of H₂ gas can achieve a sustained, high local concentration of H₂ at the lesion, significantly limiting H₂ therapy efficacy. Recent work has reported the use of stereotaxic intracranial injection of H₂-powered microswimmers, based on poly(lactic-co-glycolic acid)-encapsulated Mg nanoparticles, as a treatment for acute ischemic stroke. (24). In rat models of transient middle cerebral artery occlusion (tMCAO), these microswimmers reduced infarct volume and improved spatial learning and memory by scavenging reactive oxygen species (ROS) and mitigating inflammation via localized H₂ release. Despite their promising therapeutic potential, critical challenges remain, such as iatrogenic tissue injury caused by the injection procedure, potential long-term safety issues related to nanoparticle retention in the lateral ventricle, and local alkalization resulting from the accumulation of Mg(OH)₂ byproducts. Currently, achieving sustained H₂ release at the site of stroke lesion without intracranial nanoparticle accumulation remains a challenge.

To overcome above limitations, we propose a strategy to anchor H₂-releasing nanoparticles to inflamed BBB vasculature without requiring BBB penetration. This approach enables localized H₂ generation, allowing H₂ to diffuse across the BBB and protect brain cells from the detrimental effects of cerebral ischemia (Fig. 1A). Inspired by the innate immune recruitment response during inflammation, where inflamed endothelium rapidly up-regulates P-selectin within minutes and subsequently tethers neutrophils (25, 26), we modified H₂-producing zirconium silicide nanoparticles (ZrSi₂, ZSN) with P-selectin–binding peptides (ZSNP) to mimic the tethering interactions between neutrophils and inflamed endothelium via the P-selectin/P-selectin glycoprotein ligand 1 (PSGL1) bond (25). Consequently, these H₂-producing nanoparticles are recruited and anchored onto activated vasculature, allowing for in situ hydrolytical generation of H₂. The sustainedly released H₂ from ZSNP can readily diffuse across the BBB, enabling rapid distribution in brain while circumventing nanoparticle penetration and associated intracranial retention. On the basis of this design, we evaluated the therapeutic efficacy of ZSNP relative to edaravone and elucidated H₂-mediated neuroprotective mechanisms poststroke.

Fig. 1. Design and characterization of the inflamed BBB-anchored H₂ release system for stroke treatment.

(A) Schematic illustration of ZSNP with inflamed BBB targeting and H₂ release capabilities for ischemic stroke therapy. (B) Synthetic illustration for preparation of ZSNP. (C) FTIR spectra of ZSN, P-selectin–binding peptide (P-peptide) and ZSNP. The spectra were vertically offset for clarity. (D) Representative TEM image of ZSNP. (E and F) Size distribution and zeta potential of ZSN and ZSNP. (G) The H₂ release profile of ZSNP in deionized water by H₂ microelectrode. (H) Schematic illustration of the mouse tMCAO model and timeline for evaluating in vivo targeting efficacy. Created in BioRender. Jack, C. (2025) https://BioRender.com/2eq1hf1. (I) Photographs of the brain tissues. Red triangle indicates ZSNP accumulated in the ischemic hemispheres. (J and K) Ex vivo fluorescent images (J) with quantified fluorescence intensities (K) of Cy5-labeled ZSN and Cy5-labeled ZSNP in the ischemic hemisphere from tMCAO mice and sham-operated mice (n = 4 mice). The quantitative measurements are derived from the ratio of fluorescence intensity in the ipsilateral hemisphere relative to that in the contralateral hemisphere. (L) Representative confocal images of Cy5-labeled ZSN or Cy5-labeled ZSNP (gray) colocalized with P-selectin (green) and vasculature (lectin, red). White lines in the fluorescence images indicate colocalization analysis of fluorescence intensities along the line segments. 3D reconstruction images were generated from high-resolution confocal images using Imaris software. (M) Concentration of H₂ released from ZSNP in the ischemic hemispheres (n = 3 mice). Scale bars, (D) 500 nm; (I) and magnified images, 50 and 10 μm, respectively. Data are presented as means ± SD. Statistical significance was determined by one-way analysis of variance (ANOVA) with Tukey’s post hoc test (K). h, hours.

RESULTS

Design and characterization of the inflamed brain endothelium–anchored H₂-releasing nanoparticles

ZSN were synthesized through a physical ball-milling process followed by centrifugal separation. Transmission electron microscopy (TEM) characterization determined that ZSN had an average size of approximately 100 nm (fig. S1). X-ray diffraction analysis confirmed that the chemical composition and crystal structure of ZSN remained consistent with those of the raw ZrSi₂ material (fig. S1). Furthermore, because of their nanoscale dimension and high specific surface area, ZSN exhibited markedly enhanced H₂ release compared to raw and microsized ZrSi₂ (fig. S2, A to C). Specifically, ZSN generates H₂ through hydrolysis with water, as represented by the following redox reaction: ZrSi₂ + 6H₂O → Zr⁴⁺ + 2SiO₃²⁻ + 6H₂. The sustained H₂ release profile of ZSN indicated a rapid release phase within the first 48 hours with the release rate peaking around 6 hours (fig. S2D).

To enable ZSN to target the stroke lesion, a P-selectin–binding peptide (DAEWVDVS, referred to as P-peptide; fig. S3) derived from PSGL1, was conjugated onto ZSN. This conjugation leverages the critical role of P-peptide in tethering neutrophils to inflamed endothelium during ischemic stroke conditions. P-peptide was first grafted with 3-isocyanatopropyltrimethoxysilane (IPTMS, a silane coupling agent) through nucleophilic addition between the amino group of P-peptide and the isocyanate group of IPTMS, and then covalently modified onto the surface of ZSN (Fig. 1B and fig. S4). Fourier transform infrared spectroscopy (FTIR) analysis indicated that the ZSNP exhibited the disappearance of the N─H stretching vibration at 3272 cm⁻¹, which corresponds to the primary amine from P-peptide (Fig. 1C). Concurrently, an absorption band at 1594 cm⁻¹, attributed to the C═O stretching of a urea bond (─NH─CO─NH─) was observed, confirming the reaction between P-peptide and IPTMS. The peak at 1636 cm⁻¹, which is associated with the amide band (C═O stretching), was also visible in the modified ZSNP. This indicates the presence of the peptide backbone structure. Furthermore, the appearance of peaks at 1090 cm⁻¹ (Si─O─Si stretching) and 772 cm⁻¹ (Si─C stretching) provided additional evidence for the covalent grafting of P-peptide onto ZSN via the silane coupling.

TEM and dynamic light scattering (DLS) analyses demonstrated that the ZSNP exhibited excellent dispersion in water, without notable change in hydrodynamic size compared to unmodified ZSN (Fig. 1, D and E). Zeta potential analysis showed that the charge of the nanoparticles shifted from +7.24 mV for ZSP to −10.22 mV for ZSNP (Fig. 1F). Moreover, the modification of ZSN with P-peptide did not compromise its H₂ production capacity (Fig. 1G). ZSNP exhibited a sustained H₂ release of behavior over a duration of 5 days, as confirmed by gas chromatography (GC) (fig. S5). Furthermore, in vitro cytocompatibility evaluation revealed that ZSNP, including its extracts and hydrolysis byproducts, all exhibited high biocompatibility at all tested concentrations (0 to 400 μg ml⁻¹). Specifically, cell viability exceeded 94% at concentrations up to 300 μg ml⁻¹ (fig. S6). The average hemolysis ratio observed at the maximum concentration of 400 μg ml⁻¹ was below 1.5%, which is lower than the established safety threshold of 5% (fig. S7). Bright-field microscopy further confirmed this, showing no obvious erythrocyte rupture, thus verifying the hemocompatibility of ZSNP.

Since P-peptide is able to target inflamed endothelium, we next evaluated the targeting efficacy of ZSNP in a murine ischemia-reperfusion stroke model induced by tMCAO (Fig. 1H). Mice were randomly administered with ZSN or ZSNP via tail vein injection at 4 hours post-tMCAO. Notably, a clear accumulation of black ZSNP was observed in the ischemic hemisphere after 6 hours of nanoparticle injection, distinguishing it from both the ZSN group and sham injection of ZSNP (Fig. 1I). For better visualization, both ZSN and ZSNP were labeled with a cyanine dye Cy5 (fig. S8). Ex vivo fluorescence imaging revealed markedly higher accumulation and retention of Cy5-labeled ZSNP in the ischemic hemisphere compared to Cy5-labeled ZSN (Fig. 1J). Quantitative analysis indicated an approximately 2.0-fold increase in signal intensity for ZSNP in the stroke model compared to the sham model, demonstrating its excellent targeting capability toward stroke lesion (Fig. 1K). Furthermore, the distribution of nanoparticles in major organs (heart, liver, spleen, lung, and kidney) demonstrated that ZSNP exhibited superior ischemic brain targeting efficacy compared to ZSN, with a significant accumulation of ZSNP observed in the brain and kidney (fig. S9).

To further investigate the high-resolution spatial localization of nanoparticles within ischemic hemisphere, mice treated with ZSN or ZSNP were subjected to lectin injection (a biomarker of perfused blood vessels), followed by immunostaining of brain section for P-selectin. As anticipated, a notable up-regulation of P-selectin expression was observed in the lectin-labeled vascular endothelium of the ischemic hemisphere, whereas minimal P-selectin expression was detected in the sham group (Fig. 1L). Notably, a substantial proportion of Cy5-labeled ZSNP colocalized with P-selectin–positive vascular endothelium in the tMCAO stroke model (Fig. 1L). In contrast, only minimal ZSNP accumulation was observed in the sham group, which exhibited low P-selectin expression. Furthermore, we performed colocalization analysis and three-dimensional (3D) reconstruction of confocal images. The colocalization of P-selectin and lectin dual-labeled vascular endothelium with Cy5-labeled ZSNP clearly demonstrated that ZSNP was anchored onto the P-selectin–positive endothelium (Fig. 1L). In addition, a relatively stable concentration of H₂ was detected in the ischemic hemisphere of tMCAO mice using a H₂ microelectrode (Fig. 1M), with values calculated on the basis of a preestablished H₂ calibration curve. These results suggest that ZSNP was able to effectively anchor onto ischemic vascular endothelium for in situ–sustained H₂ release.

H₂ released from ZSNP traverses the BBB to exert immunomodulatory effects surpassing edaravone

To validate the concept of “instead of attempting to cross the BBB via nanoparticles themselves, H₂ released from nanoparticles anchoring onto the blood vessels can traverse the BBB,” we established three experimental models: 3D cell spheroids, a microfluidic chip, and a Transwell BBB model, to confirm the BBB-crossing capability of H₂. First, using a 3D mesenchymal stem cell (MSC) spheroid model that mimics solid tissue, we compared the delivery efficiency of nanoparticles and H₂ by studying their penetration kinetics. As illustrated in Fig. 2A, we observed a gradual accumulation of Cy5-labeled ZSNP on the surface of MSC spheroids over time. However, the ZSNP failed to penetrate the spheroids, with most remaining on the surface. In contrast, H₂ released from ZSNP infiltrated the spheroid core quickly, as evidenced by a time-dependent increase in H₂ probe fluorescence intensity. This phenomenon highlights a common challenge for nanoparticles during tissue infiltration and directly demonstrates the superior penetrative capability of H₂.

Fig. 2. H₂ released from ZSNP traverses the BBB and exhibits immunomodulatory effects.

(A) Time-dependent permeability of H₂ in MSC spheroids. (B) Schematic illustration of the microfluidic chip design for the in vitro BBB model. Created in BioRender. Jack, C. (2025) https://BioRender.com/syxk6ge. The left channel contained cocultured bEnd.3 cells and pericytes to establish the BBB. Brain cells pretreated with H₂ probes were inoculated in Matrigel on the middle channel. (C) Bright-field microscopy image of cells cultured in the microfluidic chip. (D) Immunostaining of VE-cadherin expression in the left BBB channel, confirming the formation of tight junctions. (E) Confocal image of H₂ signal in the right brain cells channel. Cy5-labeled ZSNP was injected into the left BBB channel, and H₂ probe expression in the right channel was observed 15 min postinjection. (F) Schematic of the design for in vitro Transwell BBB model. Created in BioRender. Jack, C. (2025) https://BioRender.com/0si1rry. (G and H) Representative confocal images (G) and quantitative analysis (H) of H₂ permeation through the Transwell BBB model (n = 5 per group). (I) Confocal images of microglia in the lower chamber, stained for inflammation-related biomarkers (ROS, iNOS, ARG1, and γH2AX), and (J) the corresponding quantitative analysis of their expression (n = 6 per group). Created in BioRender. Jack, C. (2025) https://BioRender.com/tqxrut5. Scale bars, (A) 100 μm; [(C) to (E)] 100 μm; (G) 20 μm; (I) 40 μm. Data are presented as mean ± SD. Statistical significance was determined by one-way ANOVA with Tukey’s post hoc test. d, day.

To demonstrate H₂ penetration through biological barriers, we used a well-established in vitro BBB model within a microfluidic chip (27). The chip comprised three channels: the left channel, which established a stable BBB structure by incorporating bEnd.3 cells and pericytes; the middle channel, which contained brain cells embedded in Matrigel to mimic brain parenchyma; and the right channel, which was used to supply the essential medium required for cell survival (Fig. 2, B and C). The cocultured endothelial cells and pericytes formed tight intercellular junctions, as evidenced by robust VE-cadherin expression (Fig. 2D), validating the successful establishment of the BBB model. Upon injection of Cy5-labeled ZSNP into the left BBB channel, the released H₂ efficiently traversed the BBB, leading to an enhanced green fluorescence signal in the brain parenchyma–mimicking channel within 15 min (Fig. 2E).

To further evaluate the superiority of ZSNP, we established a Transwell BBB model consisting of a monolayer of bEnd.3 cells in the upper chamber and H₂ probe–pretreated microglia in the lower chamber (Fig. 2F). Consistent with our above observations from cell spheroids and microfluidic chip experiments, Cy5-labeled ZSNP was unable to cross the bEnd.3 monolayer even after 12 hours of incubation. In contrast, a clear H₂ signal (green fluorescence) was detected in the lower chamber containing H₂ probe–pretreated microglia within 10 min (Fig. 2, G and H). Moreover, the H₂ signal intensity increased progressively over time (Fig. 2H), and the average dissolved H₂ concentration was determined to be 16.1 μM at 12 hours (fig. S10). These findings validate the feasibility of our proposed concept.

Next, to verify the function of H₂ released from ZSNP, we investigated its ROS scavenging and anti-inflammatory properties using the same Transwell coculture involving a bEnd.3 monolayer and microglia (Fig. 2, I and J). Microglia in the lower chamber were initially stimulated with tumor necrosis factor–α (TNF-α) for 6 hours, after which each treatment group was applied to the upper chamber for a coculture period of 3 days. To assess the effects of the soluble components released by ZSNP, we prepared ZSNP extracts and evaluated their effects. As shown in Fig. 2I, TNF-α stimulation provoked a proinflammatory activation in microglia, reflected by substantial up-regulation of ROS, inducible nitric oxide synthase (iNOS), and γH2A.X. ZSNP extracts showed limited efficacy in mitigating this response, whereas both edaravone and ZSNP exerted clear anti-inflammatory effects. Notably, ZSNP, with its H₂-releasing capability, surpassed edaravone across all measured parameters: It scavenged 2.3-fold more ROS and reduced iNOS and γH2A.X expression by 1.3-fold and 1.6-fold, respectively. Furthermore, ZSNP increased ARG1 levels by 2.4-fold relative to edaravone. Among all treatments, only ZSNP restored the microglial state to a level comparable to the negative control following TNF-α treatment. These findings indicate that H₂ released from ZSNP not only penetrates cell spheroids and the BBB efficiently but also exceeds the efficacy of edaravone in alleviating TNF-α–triggered inflammatory cascades, highlighting its therapeutic promise.

ZSNP facilitates brain repair by blocking poststroke inflammation

Having testified ZSNP’s targeting capability to inflamed cerebral vessels and its H₂-releasing capability to penetrate brain parenchyma, we endeavored to evaluate whether this approach could mitigate the neuroinflammatory milieu in ischemic stroke and facilitate brain regeneration. Given that ischemic stroke initiates a complex pathological cascade involving neuronal apoptosis, BBB disruption, vascular network impairment, and microglial activation (28), we systematically evaluated ZSNP’s therapeutic potential involving these key events (Fig. 3A). Evans blue staining revealed BBB damage with vascular leakage in phosphate-buffered saline (PBS)–treated tMCAO mice, indicating that ischemic-reperfusion injury severely impaired poststroke recovery (Fig. 3B). Such conditions pose a high risk of hemorrhagic transformation and a life-threatening complication that adversely affects neurological outcomes and rehabilitation. In contrast, ZSNP significantly maintained the integrity of BBB and demonstrated slightly superior outcomes compared to edaravone (Fig. 3, B and C). Immunostaining further confirmed extensive leakage of AF647-labeled dextran in the PBS group, whereas no fluorescence leakage was observed from lectin⁺ vessels in both ZSNP and edaravone groups (fig. S11). Moreover, ZSNP treatment enhanced the formation of laminin⁺ vascular matrix membrane surrounding lectin⁺ vessels (fig. S11), suggesting the development of a more mature and structurally integrated vasculature.

Fig. 3. ZSNP mitigates brain tissue necrosis and attenuates neuroinflammatory activation following ischemic stroke.

(A) Schematic diagram of the experimental design. Created in BioRender. Jack, C. (2025) https://BioRender.com/k8rku4n. Edaravone (1 μg g⁻¹ b.w., daily) and ZSNP (10 μg g⁻¹ b.w., every two other days) were administered intravenously over 14 days. (B and C) Representative images (B) of Evans blue extravasation in the stroke region in tMCAO mice from different treatment groups after 3 days, with quantification (C) of Evans blue per gram of brain by spectrophotometry (n = 4 mice). (D and E) Representative images (D) of TTC-stained brain slices with quantification (E) of relative infarct area after each treatment for 3 days in tMCAO mice (n = 7 mice). Green dash circles indicate the infarct region. (F and G) Representative laser Doppler scans (F), and quantification (G) of blood perfusion percentage (n = 3 mice). (H) Representative immunofluorescence images of GFAP⁺ astrocytes (green) and Glut-1⁺ vessels (red) in and around the ischemic lesion following 14 days of respective treatments. The white asterisks in panel (H) indicate the infarct region. (I) Corresponding quantification of the Glut-1–positive area (n = 5 per treatment). (J) Representative immunostainings of Iba-1⁺ microglia in and around the ischemic lesion after 7 days of treatment. (K) Quantification of the Iba-1⁺ microglia surface area and volume (n = 5 per treatment). (L) Quantification of the relative area of Iba-1⁺ microglia in the peri-infarct region (n = 5 per treatment). (M) Levels of inflammatory cytokines at day 7 poststroke (n = 3 mice). Scale bars, 100 μm (H and J); 5 μm [magnified image in (J)]. Data are presented as means ± SD. Statistical significance was determined by one-way ANOVA [(C), (E), (I), (K), (L), and (M)] or two-way ANOVA (G) with Tukey’s test.

We then quantified the infarct areas in tMCAO mice after a 3-day treatment. ZSNP administration significantly reduced tissue necrosis following reperfusion compared to the PBS group and modestly surpassed edaravone in therapeutic efficacy (Fig. 3, D and E). Moreover, both ZSNP and edaravone exhibited significant neuroprotective effects, as evidenced by a marked reduction in terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick end labeling–positive apoptotic cells compared to the PBS group (fig. S12). To further elucidate the functional benefits of ZSNP on the cerebral vasculature following stroke, we evaluated its effects on poststroke cortical perfusion using laser speckle imaging (Fig. 3F). The perfusion rate in all groups demonstrated a 70% reduction relative to the contralateral site during tMCAO, indicating successful modeling of ischemic stroke. Over time, blood perfusion progressively declined in the ischemic hemisphere of PBS group, suggesting continuous deterioration of cerebral cortical vessels following stroke (Fig. 3, F and G). In contrast, ZSNP treatment significantly improved intracranial perfusion relative to the PBS group and demonstrated a modest advantage over edaravone at day 3 and 7. Given the substantial evidence from both animal models and clinical studies demonstrating that H₂ inhalation reduces cerebral infarction size, attenuates neurological deficits, and improves functional outcomes (21, 29), we compared the therapeutic efficacy of ZSNP with that of inhaled H₂ in the acute phase using 2,3,5-triphenyl tetrazolium chloride (TTC) staining on day 3 posttreatment (fig. S13). Both treatments significantly reduced infarct size relative to the PBS control. Compared to the PBS control (36.8%), the infarct area was reduced by 55% in the H₂ inhalation group (16.4%) and by 69% in the ZSNP group (11.4%). Moreover, ZSNP demonstrated statistically superior infarct reduction over H₂ inhalation (P = 0.0493).

To further evaluate vascularization and astrocytic responses, we performed immunostaining for Glut-1 (a brain endothelial-specific glucose transporter) and glial fibrillary acidic protein (GFAP, an astrocyte marker) in both the infarct core and the peri-infarct area (ischemic penumbra) of tMCAO mice. PBS-treated mice exhibited pronounced astrogliosis with dense glial scar formation and sparse vascularity (Fig. 3, H and I). In contrast, ZSNP treatment significantly reduced astrogliosis while markedly increased Glut-1⁺ neovascularization in both the infarct core and the peri-infarct regions, suggesting progressive penumbral salvage. Moreover, ZSNP demonstrated a 4.2-fold higher density of Glut-1⁺ cerebral microvessel within the infarct core compared to edaravone (P = 0.0019), providing direct morphological evidence of its superior angiogenic capacity.

The extent of stroke damage is closely associated with microglial activation in peri-infarct penumbra, which drives inflammatory signaling, axonal damage, and neuronal death. To determine whether ZSNP modulates microglial responses, we performed morphological analysis of Iba1⁺ microglia in peri-infarct area 7 days after stroke. Confocal 3D reconstructions showed that microglia in PBS-treated mice displayed an activated state, characterized by reduced surface area and volume (Fig. 3J). In contrast, both ZSNP and edaravone promoted a shift in microglia toward a steady state, defined by a ramified and elongated morphology, with ZSNP inducing the most pronounced transition (Fig. 3K). Regarding microglial density in the ischemic penumbra, edaravone showed only a marginal reduction compared to PBS, whereas ZSNP significantly reduced microglial accumulation (Fig. 3L).

We next investigated the phenotype of microglia by costaining with iNOS (proinflammatory phenotype) and ARG1 (proreparative phenotype). ZSNP-treated mice exhibited the highest ARG1⁺/iNOS⁺ ratio of 2.7, representing 5.2-fold and 1.8-fold increases compared to PBS and edaravone, respectively (fig. S14). Given that activated microglia release inflammatory neurotoxic cytokines, they create a detrimental circumstance for neuronal survival and function. To assess this response, we measured cytokine levels in the ischemic hemisphere (Fig. 3M). At day 7 poststroke, ZSNP treatment significantly reduced the level of proinflammatory cytokines [interleukin-1α (IL-1α), IL-1β, IL-2, and TNF-α] compared with PBS treatment, while concurrently up-regulating the expression of the anti-inflammatory cytokine IL-4. These effects were comparable to those observed with edaravone, a clinically recognized free radical scavenger. Together, these results suggest that ZSNP could down-regulate microglial proinflammatory activation, creating a favorable microenvironment for stroke recovery.

ZSNP promotes ischemic brain repair contingent upon microglia

Given that microglia are the most rapid responders to ischemic neurons and the major myeloid cells exhibiting long-lasting reparative phenotypes (28), we subsequently aimed to examine the impact of microglia on ZSNP-mediated stroke repair by selectively depleting them using a PLX5622 diet (Fig. 4A). PLX5622, a colony-stimulating factor 1 receptor (CSF1R) inhibitor, was administered to deplete microglia. The results demonstrated highly specific and effective microglial depletion (figs. S15 and S16). The proportion of microglia decreased by 27.4-fold (from 5.23 to 0.19%) in the PLX5622-treated mice. Consistent with established literature, PLX5622 also led to a reduction in the macrophage population (30), which shares a CSF1R-dependent lineage with microglia (31), while the proportions of T cells showed no significant changes following PLX5622 treatment. TTC staining revealed extensive necrotic tissue in microglia-depleted brains (Fig. 4, B and C), indicating that microglia depletion severely impaired ZSNP’s therapeutic efficacy in ischemic brain injury compared to mice with intact microglia. Immunostaining further showed that microglia depletion impeded Glut-1⁺ vessels ingrowth (Fig. 4, D and E) and axonal sprouting (NF200⁺, a neurofilament biomarker) in the infarct area (Fig. 4, F and G).

Fig. 4. ZSNP boosts ischemic brain repair via microglia modulation.

(A) Schematic illustration of the experimental design for microglia depletion using a PLX5622-containing diet. Created in BioRender. Jack, C. (2025) https://BioRender.com/bugmtx5. (B) Representative images of TTC-stained brain slices from ZSNP-treated tMCAO mice with or without microglia depletion. (C) Quantification of the infarct area (n = 5 mice). (D and E) Representative images (D) and quantitative analysis (E) of Glut-1 immunostaining in ZSNP-treated tMCAO mice with or without microglia depletion (n = 7 per treatment). (F and G) Representative images (F) of NF200 immunostaining with (G) quantification of relative NF200 area in ZSNP-treated tMCAO mice with or without microglia depletion (n = 6 per treatment). (H) Principal components analysis (PCA) of gene expression profiles (n = 3 to 4 mice). (I) Top 15 enriched biological processes in ZSNP-treated microglia compared to PBS-treated microglia isolated from tMCAO mice. (J) Enriched gene sets identified by GSEA. The normalized enrichment scores (NES), nominal P value, and false discovery rate (FDR) of each gene set are shown. (K and L) Heatmap of inflammatory genes (K) and brain repair–related genes (L) in ZSNP-treated microglia versus PBS-treated microglia. (M) Fragments per kilobase of transcript per million mapped reads (FPKM) values of brain repair–related genes in ZSNP-treated microglia compared to PBS-treated microglia. (N) Schematic representation of H₂-mediated microglia-neuron interaction. Created in BioRender. Jack, C. (2025) https://BioRender.com/xy48u79. Primary microglia in the upper chamber were pretreated with LPS (100 ng ml⁻¹) for 24 hours and then cocultured with neurons (lower chamber) for an additional 3 days in either a conventional incubator or a H₂ incubator. To inhibit Wnt/β-catenin and CaMKII signaling pathways in microglia, 5 μM MSAB and 20 μM KN-93 were used, respectively. Scale bars, 100 μm [(D) and (F)]; 50 μm (N). Data are presented as means ± SD. Statistical significance was determined using an unpaired two-tailed Student’s t test.

To explore the mechanism underlying the ZSNP-mediated microglial contribution to post-stroke recovery, we performed SMART-seq (Switching Mechanism at 5′ End of RNA Template sequencing) on microglia isolated from ischemic brains at day 7 post-tMCAO. Principal components analysis revealed distinct transcriptional profiles for ZSNP-treated group compared to edaravone and PBS groups (Fig. 4H). Differential gene expression analysis identified 1449 differentially expressed genes (DEGs) between ZSNP and PBS, and 2919 DEGs between ZSNP and edaravone (fig. S17A). Notably, 681 genes were commonly up-regulated in ZSNP-treated microglia compared to both edaravone and PBS groups. In addition, 1609 genes were specifically up-regulated in ZSNP-treated microglia compared to edaravone, while 405 genes were specifically up-regulated compared to PBS group.

Gene ontology (GO) analysis revealed enrichment of biological processes in ZSNP-treated microglia, including angiogenesis, wound healing, nervous system development, inflammatory response, etc. (Fig. 4I and fig. S17B). Enrichment score plots and a heatmap further demonstrated distinct differences in gene expression profiles associated with angiogenesis, wound healing, brain development, and synaptic vesicle exocytosis between ZSNP and edaravone treatments (fig. S18). In contrast, edaravone exhibited a gene expression pattern that closely resembles that of PBS. To further explore the biological processes modulated by ZSNP in microglia, we performed gene set enrichment analysis (GSEA). GSEA results showed positive enrichment in pathways related to angiogenesis, axonogenesis, neuron differentiation, and wound healing (Fig. 4J). Moreover, ZSNP-treated microglia displayed positive enrichment in “cyclic adenosine monophosphate–mediated signaling,” a pathway known to play a critical role in neuroprotection and neurite growth. Conversely, the term “inflammatory response” showed negative enrichment in ZSNP-treated microglia. Specifically, proinflammatory genes, including Pld3, Ccl5, Ccl12, Clec7a, and Ccl8, were down-regulated, while brain repair–related genes, such as Wnt5a, Vegfa, Ntrk2, and Camk2g, were up-regulated (Fig. 4, K and L). These findings collectively indicate that ZSNP treatment promotes a phenotypic shift in microglia toward a reparative state.

We observed that ZSNP modulates the activity of poststroke microglia and influences the expression of genes associated with Wnt signaling pathway, including Wnt5a and Camk2g (Fig. 4M). Moreover, it effectively triggers both canonical Wnt/β-catenin and noncanonical Wnt/Ca²⁺ pathways (fig. S19). To further elucidate the regulatory role of H₂ in microglia-mediated brain repair, we established a coculture system to further explore its effect on neuronal function (Fig. 4N). To directly attribute the observed signaling modulation to H₂ and provide specific evidence for its role in regulating the Wnt pathways, we used a H₂ incubator for the in vitro validation rather than administering ZSNP directly. Further, we used MSAB, a canonical Wnt/β-catenin inhibitor, and KN-93, a CaMKII inhibitor that targets the noncanonical Wnt/Ca²⁺ pathway, to distinguish their respective contributions. Neuronal architecture and synaptic integrity were evaluated using microtubule-associated protein 2 (MAP2) and synaptophysin (SYP), respectively. Lipopolysaccharide (LPS)–activated proinflammatory microglia induced pronounced neuronal damage and synaptic loss, as evidenced by reduced MAP2⁺ staining and diminished SYP⁺ signal. In contrast, H₂ markedly attenuated the proinflammatory microglia–induced neuronal damage, as indicated by the preservation of MAP2⁺ neuronal structures and the maintenance of SYP⁺ puncta density. Notably, H₂-mediated protection of neuronal integrity was unaffected by canonical Wnt pathway inhibition but was abolished by CaMKII inhibition, which led to severe structural and synaptic deficits. Subsequent molecular docking revealed only weak, nonspecific binding of H₂ to CaMKIIγ and Wnt5a (binding energy ≈ −1.44 kcal·mol⁻¹ for each; fig. S20), indicating that its modulation of the pathway is not through direct binding but through its established function as antioxidant and immunomodulator. Collectively, our data indicate that ZSNP-released H₂ drives a reparative shift in microglial phenotype, which in turn enhances neuronal synaptic integrity via the noncanonical Wnt/Ca²⁺ pathway. This facilitated microglia-neuron interaction is essential for poststroke functional recovery.

ZSNP induces neurogenesis and axonogenesis

Neurogenesis is a critical process for poststroke recovery and repair of injured brain regions (32–34). However, poststroke neurogenesis is limited by several factors, including the restricted migration of neuroblasts from the subventricular zone to the injury site, as well as the hostile microenvironment at the lesion site that inhibits neuronal survival and integration (34). To explore the potential role of ZSNP in enhancing neurogenesis for stroke treatment, we assessed the integration of neuroblasts into the ipsilateral hippocampus and cortex 14 days postinjury. Immunostaining with doublecortin (Dcx) was used as it is expressed in neuronal precursor cells and immature neurons, serving as a marker for assessing neurogenesis. ZSNP-treated mice exhibited a significant increase in the number of proliferative neuroblasts (Ki67⁺/Dcx⁺) in the cortex compared to both PBS and edaravone groups (Fig. 5, A and B). Although edaravone demonstrated anti-inflammatory effects during the early phase (Fig. 3, J to M), its impact on neurogenesis was minimal and comparable to PBS. Furthermore, ZSNP treatment enhanced the number of hippocampal Dcx⁺ neuroblasts and extended their migration distance toward the ischemic cortical region (Fig. 5, C and D), reaching approximately 2.3-fold longer than that observed in the edaravone group. Thus, these data suggest that ZSNP effectively promotes the migration of neuroblasts into the damaged tissue adjacent to the stroke cavity during the initial stages of tissue recovery following a stroke.

Fig. 5. ZSNP promotes neurogenesis, axonal sprouting, and angiogenesis against ischemic brain injury.

(A and B) Representative immunostaining images of the proliferating neuroblasts (Ki67⁺Dcx⁺) in the ipsilateral cortex (A) and quantification for Ki67⁺Dcx⁺ proliferating neuroblasts [(B) n = 6 per treatment]. (C and D) Representative immunostaining images of Dcx⁺ neuroblasts migrating toward the peri-infarct cortex (C) and quantification of migration distance of Dcx⁺ cells from the hippocampus [(D) n = 5 per treatment]. (E) Representative immunostaining images of axonal neurofilaments (NF200) within and around the stroke lesion (*) at day 14 poststroke. (F to H) Quantification of the axonal area (NF200) within and surrounding the stroke lesion, as well as the infiltration distance at day 14 post-stroke (n = 4 per treatment). (I and J) Representative H&E staining (I) and Nissl staining (J) of brain tissue following different treatments at week 12 poststroke. (K) Representative immunostaining images of axonal neurofilaments (NF200) within and around the stroke lesion (*) at week 12 poststroke. (L to N) Quantification of the axonal area (NF200) within and surrounding the stroke lesion, as well as the infiltration distance at week 12 poststroke (n = 4 per treatment). (O) Representative immunostaining images of vessels (Glut-1) and axonal neurofilaments (Tuj-1) in the infarct region at week 12 poststroke. (P) Representative immunostaining images of pericytes (PDGFR-β) costained with vessels (Glut-1) and astrocyte end-feet (aqua-4) in the stroke lesion at week 12 poststroke. Scale bars, 100 μm [(A), (C), (E), and (K)]; 50 μm [(O) and (P)]. Data are presented as means ± SD. Statistical significance was determined using one-way ANOVA with Tukey’s post hoc test.

To check whether ZSNP can promote axonal infiltration into the stroke lesion site, we assessed axonal growth using NF200 immunostaining in the infarct and peri-infarct regions. At day 14 poststroke, both ZSNP and edaravone treatments increased axonal density in the peri-infarct area (Fig. 5, E and F). In contrast, ZSNP treatment resulted in significantly greater NF200⁺ staining intensity and a more extensive presence of NF200⁺ axons within the infarct area compared to PBS and edaravone groups (Fig. 5, E and G), with deeper axonal penetration observed in the infarcted tissue (Fig. 5H). These results demonstrate that ZSNP promotes effective axonal connectivity within the infarct region, which is likely attributed to the enhanced proliferation and migration of Dcx⁺ neuroblasts, as evidenced by above findings (Fig. 5, A to D). This aligns with recent studies showing that neuroblast recruitment and migration play a critical role in axonal remodeling and functional recovery after ischemic injury (35, 36).

To evaluate the long-term effect of ZSNP on poststroke recovery, we administered ZSNP and edaravone intravenously for the first 14 days poststroke, followed by histological analyses at 12 weeks to assess sustained therapeutic outcomes. Hematoxylin and eosin (H&E) staining revealed extensive necrosis and tissue degradation in the infarct area of the PBS group, whereas the ZSNP group exhibited minimal necrotic tissue (Fig. 5I). Although edaravone reduced tissue necrosis compared to PBS, its reparative effects were inferior to those of ZSNP. Nissl staining further demonstrated severe neuronal loss and disorganized morphology in the infarcted areas of PBS- and edaravone-treated tMCAO mice (Fig. 5J). In contrast, ZSNP treatment preserved structurally intact neurons throughout the ischemic hemisphere at week 12. In addition, the ZSNP group demonstrated a significantly greater neurofilament area and enhanced infiltration toward the infarct region at 12 weeks compared to the other groups (Fig. 5, K to N), surpassing the levels observed at day 14 posttreatment (Fig. 5, E to H).

Since neurogenesis and angiogenesis are closely intertwined (33, 37), we next examined the relationship between vascular and axonal networks by costaining of vessels (Glut-1) and neurofilament networks (Tuj-1) at 12 weeks. The ZSNP group exhibited a spatial correlation between blood vessels and axons (Fig. 5O), accompanied by a significantly higher density of Tuj-1 expression (fig. S21). These findings are consistent with above observations regarding the elevated expression of Glut-1 (Fig. 3, H and I), as well as data obtained from RNA sequencing (RNA-seq) analyses (Fig. 4, I and J), further supporting the notion that ZSNP enhances nerve repair through modulation of angiogenesis. Moreover, immunostaining revealed mature vascular coverage by pericytes [platelet-derived growth factor receptor β (PDGFR-β)] and astrocyte end-feet (aquaporin-4, Aqua-4), indicating that ZSNP treatment restores the normal architecture of the neurovascular unit (Fig. 5P).

ZSNP improves motor functional recovery in the subacute phase of stroke

To determine whether ZSNP-induced neuroregeneration translates to functional recovery, we assessed contralateral forelimb dexterity using a comprehensive battery of behavioral tests (beam-walking, grid-walking, cylinder/asymmetry, and pasta handling tests) (37), which collectively model clinical motor deficits in stroke patients. The modified neurological severity score (mNSS) was determined through beam-walking performance. Both ZSNP and edaravone improved crawling stability over time, while ZSNP-treated mice demonstrated accelerated recovery compared to edaravone (Fig. 6A). Moreover, ZSNP treatment significantly reduced limb-use deficits compared to edaravone at weeks 4 and 12 poststroke. In the grid-walking task, ZSNP-treated mice showed fewer contralateral foot-faults than edaravone-treated mice, with significant differences emerging by week 4 (Fig. 6B). In the cylinder test, edaravone showed a limited effect on forelimb asymmetry (Fig. 6C). In contrast, ZSNP treatment significantly increased contralateral forelimb use during the test. Furthermore, during the pasta handling test, tMCAO mice treated with ZSNP exhibited significantly reduced manipulation and consumption times starting from week 1 poststroke. (Fig. 6D).

Fig. 6. ZSNP restores motor function by enhancing axonal connectivity and neuronal activity poststroke.

(A to D) Behavioral assessments of tMCAO mice at weeks 1, 4, 8, and 12 poststroke (n = 6 mice). Created in BioRender. Jack, C. (2025) https://biorender.com/tthhp7v. The beam-walking test was used to evaluate overall motor ability and balance (A), the grid test was used to assess foot faults of the contralateral hindlimb (B), the cylinder test measured the dexterity of the contralateral forelimb (C), and the pasta test was used to examine the sensitivity of the contralateral forepaw (D). (E) Behavioral tests in ZSNP-treated mice with neuronal silencing induced by sequential AAV9 and CNO injections. Created in BioRender. Jack, C. (2025) https://BioRender.com/j6qjskq. (F and G) Representative c-Fos immunostaining images (F) and quantification (G) of neuronal activity in the ischemic region at week 12 poststroke (n = 7 per treatment). (H and I) Representative immunostaining images (H) and quantitative analysis (I) of the BDA-traced neurons in the ipsilateral hemisphere at week 12 poststroke (n = 4 per treatment). (J) Schematic diagram illustrating the proposed mechanisms of ZSNP in treating ischemic stroke. Scale bars, [(F) and (H)] 100 μm. Data are presented as means ± SD. Statistical significance was determined by two-way ANOVA with Tukey’s post hoc test [(A) to (D)], one-way ANOVA with Tukey’s post hoc test [(G) and (I)], or an unpaired two-tailed Student’s t test (E).

To investigate whether the functional recovery observed in ZSNP attributes to the axonal connections formed within the infarct, we used chemogenetic silencing of ZSNP-induced axonal projections. Using adeno-associated virus 9 (AAV9) that expresses the neuronal silencing receptor hM4D DREADD, we retrogradely transfected the neurons that extend axons into the infarct site (38). Neurons forming connections with the damaged tissue induced by ZSNP will be selectively inhibited upon administration of clozapine-N-oxide (CNO), which acts as a ligand for hM4D. When CNO treatment was applied to ZSNP-treated mice, the previously improved motor functions were abolished and animal deficits worsened significantly (Fig. 6E). This finding provides direct evidence that the established axonal network is responsible for the observed functional improvement.

On the basis of these findings, we used c-Fos immunofluorescence to evaluate neuronal activation in response to external stimuli in the ischemic cortex. C-Fos, an immediate early gene product, serves as a well-established marker of neuronal activity (39, 40). At week 12 poststroke, both PBS and edaravone treatments exhibited minimal effects on c-Fos expression (Fig. 6, F and G). In contrast, ZSNP administration significantly enhanced c-Fos expression in the infarct region, revealing a 3.9-fold increase relative to edaravone-treated mice (Fig. 6, F and G). These results demonstrate that enhanced c-Fos expression in the peri-infarct region correlates with improved functional outcomes in the tMCAO mice. To further observe the neuronal population that projects axons to the stroke site, we performed local injection of biotinylated dextran amine (BDA) at week 12 poststroke for bidirectional axonal tracing. BDA uptake was visualized using streptavidin-fluorochrome detection, and brain sections were stained with the astrocytic marker GFAP to delineate the lesion site. While PBS- and edaravone-treated mice showed limited axonal regeneration within the lesion, ZSNP-treated mice exhibited robust retrograde labeling of neuronal cell bodies in the ipsilateral peri-infarct cortex projecting toward the injection site (Fig. 6, H and I).

To evaluate the in vivo biosafety of ZSNP, we conducted integrated biodistribution and histological analyses. Biodistribution was assessed by measuring zirconium (Zr) levels in major organs at various time points postinjection using inductively coupled plasma mass spectrometry (ICP-MS). Compared with blank control mice (no ZSNP injection), ICP-MS results confirmed the brain-targeting capability of ZSNP (fig. S22). Consistent with typical nanoparticle pharmacokinetics, ZSNP initially accumulated in reticuloendothelial system (RES) organs, including the liver, spleen, and lungs, yet this accumulation proved transient, as Zr signals gradually decreased and eventually returned to baseline. H&E staining revealed no visible pathological damage in these RES organs despite the transient accumulation, with tissue integrity comparable to the PBS group (fig. S23). Serum biochemistry further corroborated these findings, showing no impairment of liver or kidney function, as indicated by normal levels of alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), creatinine (CREA), and urea (UREA) (fig. S23). In addition, the detection of Zr in kidney tissues indicated renal clearance of ZSNP or its metabolites. Meanwhile, the clearance of the Zr signal from the brain over time demonstrated successful elimination (fig. S22), thus mitigating concerns regarding potential nanoparticle accumulation and the associated intracranial burden.

Collectively, these results demonstrate that ZSNP promotes functional recovery following stroke by facilitating the formation of an interconnected axonal network from adjacent brain regions. The observed improvements in both structural connectivity and behavioral outcomes position ZSNP as a promising therapeutic strategy for stroke treatment, potentially addressing the critical need for treatments that enhance neuroplasticity during the subacute phase of stroke recovery.

DISCUSSION

The pathophysiological cascade following acute ischemic stroke represents a spatiotemporally orchestrated sequence of events, where excitotoxicity, neuroinflammation, and oxidative stress form an interdependent triad driving secondary injuries (3, 28). Our findings establish H₂ as a multimodal therapeutic agent capable of simultaneously addressing these pathogenic axes through its unique physicochemical properties and biological activities. We propose a therapeutic strategy involving the modification of H₂-releasing nanoparticles with P-selectin/PSGL1 anchorage points to specifically target activated endothelium (Fig. 6J). Under inflammatory conditions, activated endothelial cells rapidly up-regulate P-selectin from Weibel-Palade bodies within minutes, facilitating the tethering of free-flowing leukocytes to the endothelial surface (25). This mechanism is particularly relevant in stroke, where endothelial activation is a hallmark of the early inflammatory response. By leveraging this natural targeting mechanism, ZSNP can effectively anchor onto inflamed endothelium, enabling local sustained H₂ release.

Although most intense during the initial hyperacute phase, deleterious processes after ischemia-reperfusion do not cease abruptly. Instead, the initial ischemia-reperfusion injury evolves into a persistent state of sterile inflammation that can last for weeks (41, 42). Driven by activated microglia and infiltrating immune cells, this inflammatory response continuously exacerbates tissue damage, thereby establishing a vicious cycle of secondary neuronal loss and long-term functional impairment (43, 44). The generation of H₂ from ZSNP serves as a sustained antioxidant and anti-inflammatory agent (24, 29, 45, 46), facilitating the suppression of these damaging cascades throughout the critical acute and subacute periods, even in regions with restricted access for routine therapeutic agents. Moreover, our study demonstrated a superiority of ZSNP in infarct reduction over H₂ inhalation. We attribute this enhanced efficacy to the capacity of ZSNP for localized and sustained H₂ release. Although arterial H₂ concentration rises rapidly following inhalation, reaching its peak within 20 min, the concentration declines swiftly after the cessation of inhalation. Specifically, it decreases to below 50% of the plateau level in the venous circulation within 6 to 8 min and reaches approximately 10% by 18 min (21, 47). Thus, the intermittent nature of the inhaled H₂ regimen likely renders it insufficient to match the benefits of continuous H₂ delivery provided by ZSNP.

In addition, the poststroke inflammatory response is biphasic (28, 41). Following the hyperacute phase, a shift toward a reparative inflammatory response is crucial for functional recovery (28). This phase, mediated by reparative microglia, is vital for processes like synaptogenesis, axonal growth, and neural circuit reorganization (30, 36). By providing a sustained supply, H₂ can help steer this transition from a proinflammatory to a proreparative state, thereby actively facilitating the brain’s endogenous recovery mechanisms over several weeks. Our findings demonstrate that in situ H₂ treatment suppresses microglial activation (Figs. 2, I and J, and 3, J and K) and concurrently promotes a transcriptional program conducive to repair, encompassing angiogenesis, axonogenesis, neurogenesis, and wound healing (Fig. 4, I to M). This coordinated response ultimately fosters a protective neuron-microglia interplay that supports poststroke recovery.

Notably, in situ H₂ treatment demonstrated superior neuroprotective effects compared to edaravone, particularly in preserving the neurovascular unit. ZSNP-treated tMCAO mice exhibited more mature vascular structures and improved BBB integrity (Fig. 3, B, C, and H, and 5, O and P, and fig. S11). Even a 1% increase in BBB permeability can reduce the likelihood of favorable functional outcomes after thrombolysis by up to 75% (28, 48). Although edaravone reduces the ischemic area and promotes vascularization during the acute phase compared to PBS, ZSNP treatment further enhances the formation of mature microvessel with intact neurovascular unit architecture in the peri-infarct region, contributing to improved functional recovery. Moreover, ZSNP facilitated the migration of Dcx⁺ neural progenitor cells to the injury site, enhancing neurogenesis and functional recovery in tMCAO mice, outperforming edaravone. These findings are consistent with recent studies demonstrating that H₂ promotes angiogenesis and neurogenesis by modulating vascular endothelial growth factor and brain-derived neurotrophic factor signaling pathways (49).

In summary, our approach presents a promising strategy for addressing challenges associated with traditional drug delivery across the BBB while ensuring efficient H₂ delivery without accumulation of nanoparticles in brain parenchyma. However, although significant differences in brain structure, immune system, and pathophysiology exist between rodents and humans (1, 50), there are several limitations to this work that warrant further investigation. Specifically, leukocyte composition in peripheral blood differs considerably between humans and rodents. In humans, approximately 50 to 60% of leukocytes are neutrophils and 15 to 30% are lymphocytes. In contrast, rats and mice have only about 15 to 20% neutrophils while having around 60% lymphocytes (51). Given the pivotal role of the immune system in stroke pathology, these disparities could pose a significant hurdle when translating results from rodent stroke models into clinical trials involving human patients (52). Furthermore, the tMCAO model primarily replicates focal arterial occlusion. In contrast, human ischemic stroke is etiologically heterogeneous and includes subtypes such as penetrating artery infarcts linked to small vessel disease and watershed infarcts that often result from cardiovascular dysfunction (1, 53). More critically, more than 80% of strokes occur in elderly populations where hypertension stands out as the single most powerful risk factor, accounting for approximately 45% of population-attributable risk (1, 53). The prevalence of hypertension escalates dramatically with age, and this combination creates a unique pathological milieu characterized by exacerbated neuroinflammation and preexisting endothelial dysfunction, a condition not fully represented in young healthy rodent models.

MATERIALS AND METHODS

Preparation and characterization of ZSNP

ZrSi₂ nanoparticle (ZSN) was synthesized via the ball-milling combined centrifugal separation method. Briefly, 200 mg of ZrSi₂ raw powder (Aladdin, Z302607) was added into 35 ml of 28.6 mg ml⁻¹ polyvinylpyrrolidone (Sigma-Aldrich, 856568, in ultradry ethanol), followed by a ball-milling process (400 rpm) at room temperature for 4 days. To remove the microsized particles, the above ball-milled ZrSi₂ particles were centrifuged at 4000 rpm for 20 min. Then, the supernatant was collected and further centrifuged at 12,000 rpm for 20 min to collect the nanosized ZrSi₂ particles. After washing with ultradry ethanol thrice, ZSN was finally obtained and stored in super-dry ethanol under a nitrogen atmosphere.

To modify ZSN with P-selectin–targeting peptide (P-peptide), 2.30 mg of P-peptide (DAEWVDVS, M_w = 919 Da, purity >98%, China Peptides) was dissolved in 15 ml of ultradry N,N′-dimethylformamide (DMF, Meryer, M33593) and reacted with 513 μg of IPTMS (Bide Pharmatech, BD87579) at a 1:1 molar ratio for 1 hour. After that, 10 ml of 0.45 mg ml⁻¹ of ZSN (in DMF) was added into above product and reacted under stirring (500 rpm) and reflux conditions at 60°C for 2 hour. Then, the ZSNP was collected via centrifugation at 12,000 rpm for 20 min after alternately washing with ultradry DMF and ultradry ethanol twice. The obtained ZSNP was stored in super-dry ethanol and nitrogen atmosphere at 4°C for the subsequent study.

The morphology of ZSN and ZSNP were characterized by TEM (HITACH, HT7700) and scanning electron microscopy (Thermo Fisher Scientific, APREO FEI). The hydrodynamic diameter and zeta potential of ZSN or ZSNP were analyzed by DLS (Malvern, Zetasizer Nano ZS90). The presence of P-peptide on ZSNP was determined by FTIR (Thermo Fisher Scientific, Nicolet Nexus 670).

H₂ release kinetics of ZSNP in vitro and in vivo

The in vitro H₂ release behavior of ZSN and ZSNP in deionized water were detected by H₂ microelectrode (Uni-sense, Denmark) and GC (Agilent Technologies, 7890B). Specifically, the real-time H₂ concentration was measured by directly inserting the H₂ microelectrode into a ZSNP solution (100 μg ml⁻1) in an open environment. This method provides a direct readout of dissolved H₂, allowing for continuous monitoring of the kinetic release profile and determination of the peak concentration. While the GC data were obtained by sampling the headspace air (6-ml aliquots) from sealed vials containing 4 ml of the ZSNP solution (100 μg ml⁻¹). At the indicated time points, 1 ml of headspace gas was withdrawn for H₂ concentration analysis by GC.

The in vivo analysis of H₂ release from ZSNP was conducted by measuring the generation of H₂ at the ischemic stroke lesion using a H₂ microelectrode (Uni-sense, H2-N-009982). Briefly, ZSNP was intravenously injected after the ischemia-reperfusion stroke model was induced for 4 hours. Subsequently, the skin was sterilized and removed, creating a 2-mm gap with a scalpel. A skull drill was used to create a 0.8-mm diameter notch on the infarct side. Then, the H₂ microelectrode was inserted into the stroke lesion at 3, 6, 9, 12, and 24 hour to quantify H₂ production.

Synthesis of Cy5-labeled ZSNP

Cy5–polyethylene glycol (PEG)–NH₂ (1.7 mg) and 1.84 mg of P-peptide were dissolved in 7.5 ml of ultradry DMF, respectively. Subsequently, 102.6 μg of IPTMS was added into Cy5-PEG-NH₂ solution and reacted with Cy5-PEG-NH₂ for 1 hour to form the IPTMS-conjugated Cy5-PEG (named Cy5-PEG-IPTMS). Similarly, 410.6 μg of IPTMS was added into the P-peptide solution and reacted with the P-peptide for 1 hour to form the IPTMS-conjugated P-peptide (named P-peptide–IPTMS). Then, the above obtained Cy5-PEG-IPTMS and P-peptide–IPTMS were transferred into 5 ml of ZSN (0.9 mg ml⁻¹) in DMF) and reacted at a fixed molar ratio of 1:1:60. After stirring (500 rpm) under reflux conditions at 60°C for 2 hours, the Cy5-labeled ZSNP was collected via centrifugation at 12,000 rpm for 20 min. After washing with ultradry ethanol, the obtained Cy5-labeled ZSNP was stored in super-dry ethanol and nitrogen atmosphere at 4°C. The Cy5-labeled ZSN was prepared by a similar procedure but without addition of the P-peptide–IPTMS.

Cell isolation and culture

The bEnd.3 cells (FH0356) and pericytes (FH-M193) were purchased from FuHeng. Alpha mouse liver 12 (AML12) cells were purchased from Procell (CL-0602). All cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, C11995500BT) with 10% fetal bovine serum (FBS, Gibco, 10099141C) and 1% penicillin-streptomycin, unless otherwise specified.

The microglia were isolated from forebrain of 3-day-old C57BL/6 mouse as previously described (54). Briefly, the cerebellum and meninges were carefully removed, and the cerebral tissues were dissociated using the neural tissue dissociation kit (Miltenyi Biotec, 130-092-628) according to the manufacturer’s protocol. The tissue was resuspended in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. The resulting cell suspension was filtered through a 70-μm strainer and centrifuged at 500g for 5 min. Cells were then seeded onto poly-d-lysine (20 μg ml⁻¹)–precoated flasks and maintained in a humidified incubator at 37 °C with 5% CO₂. Once mixed glial cultures reached full confluence, the microglia were separated by orbital shaking at 100 rpm for 1 hour at 37 °C, followed by collection of the supernatant. The supernatant containing microglia was transferred to a new culture vessel for further expansion to obtain highly purified microglial populations.

Primary hippocampal neurons were prepared following established protocols with modifications (55, 56). Briefly, hippocampi were dissected from fetal rats at embryonic day 18, followed by digestion in 0.25% trypsin for 15 min at 37°C. The digested tissues were then washed twice with Hank’s balanced salt solution buffer and triturated with a fire-polished glass pipette in neurobasal medium (Gibco, 10888022) supplemented with 1% Glutamax (Gibco, 35050061), 2% B27 (Gibco, A1486701), and 1% NaCl. Neurons were plated at densities of 2 × 10⁵ cells per well on poly-l-lysine–coated glass-bottom 24-well Transwell plates (Cellvis, P24-1.5H-N) and grown in incubators at 37°C and 5% CO₂ for 20 days. Thirty % of the culture medium was replaced with fresh maintenance medium every 2 days.

In vitro biocompatibility

Cell viability was assessed in vitro using both AML12 and bEnd.3 cells. Briefly, cells were seeded in 96-well plates at a density of 1 × 10⁴ cells per well and allowed to adhere overnight. Subsequently, the medium was replaced with fresh complete medium containing serial concentrations (0 to 400 μg ml⁻¹) of ZSNP, ZSNP extracts (the supernatants obtained by incubating ZSNP in medium for 24 hours followed by centrifugation), or ZSNP by-products. Following 24 hours of incubation, cell viability was determined using CCK-8 assay kit (Beyotime, C0048L) according to the manufacturer’s instructions. The absorbance at 450 nm was measured with a microplate reader, and the data are presented as relative viability normalized to the blank control (0 μg ml⁻¹).

Hemocompatibility was evaluated using an erythrocyte hemolysis assay. Erythrocytes were isolated from C57BL/6 mice (Slac Laboratory) whole blood by centrifugation at 1500 rpm for 10 min, washed three times with PBS, and resuspended in PBS to a 5% (v/v) concentration. Aliquots (1 ml) of the erythrocyte suspension were transferred into microcentrifuge tubes. After centrifugation and removal of the PBS, the pelleted erythrocytes were resuspended in 1 ml of the following test solutions: PBS (negative control), 0.1% Triton X-100 (positive control), or serial concentrations of ZSNP, its extracts, or hydrolysis byproducts. Following incubation at 37°C for 1 hour, the samples were centrifuged at 8000 rpm for 15 min. The absorbance of the supernatant was measured at 540 nm to quantify hemoglobin release. The hemolysis ratio was calculated using the following equation: hemolysis (%) = (OD_sample – OD_{negative control}) / (OD_{positive control} – OD_{negative control}) × 100. In addition, erythrocyte morphology following each treatment was examined by microscopy.

Permeability of H₂ in MSCs spheroid, microfluidic chip BBB, and Transwell BBB model

For permeability of the H₂ in MSCs spheroid, MSCs were isolated from rat bone marrow. The passage one MSCs were collected, incubated with H₂ probes for 12 hours using our previously established method (19), and subsequently seeded in ultralow attachment 96-well V-shaped bottom plates (PrimeSurface, MS-9096VZ) at a density of 2 × 10⁴ cells per well. After culturing for 12 hours, MSCs formed spheroids with an average diameter of 200 μm. Then, the MSC spheroids were coincubated with Cy5-labeled ZSNP at a concentration of 50 μg ml⁻¹ for varying durations. The permeability of H₂ in the MSC spheroids was observed using two-photon microscope (NIKON, A1MP).

We also established an in vitro BBB model on a microfluidic chip (Dxfluidics, China) as previously described (27). Briefly, brain cells were isolated from neonatal mouse brain tissue, pretreated with H₂ probes, and subsequently suspended in Matrigel at a density of 8 × 10⁶ per ml before being injected into the brain channel (middle channel) of the chips that was placed onto a cold pack. The chips were then incubated at 37°C for 30 min to allow gelation. After gelation, neuron culture medium was injected into the adjacent right channel to maintain the survival of the cells in the middle chip. The bEnd.3 cells and pericytes were suspended in DMEM supplemented with 10% FBS at a density of 1 × 10⁶ cells/ml. Next, bEnd.3 cells and pericytes were mixed at a 9:1 cell number ratio based on previous literature, and then 10 μl of the cell suspension was injected into the BBB channel (left channel) of the chip. After culturing for 7 days to facilitate BBB formation, 10 μl of Cy5-labeled ZSNP (100 μg ml⁻¹) was injected into the left BBB channel, the tissue permeability of H₂ in the middle brain channel was observed with confocal microscope (Leica TCS SP8). All culture medium was refreshed every other day during culturing.

For the permeability of H₂ in a Transwell BBB model, we established a coculture Transwell (Corning, 3470) system consisting of bEnd.3 cells and microglia. The upper chamber was used to simulate BBB function by allowing bEnd.3 cells to reach a monolayer with transendothelial electrical resistance (TEER) value above 300 Ω·cm², whereas microglia pretreated with H₂ probes (1.5 × 10⁵ cells per well) were seeded in the lower chamber. After that, the Cy5-labeled ZSNP was added into the upper bEnd.3 chamber at a concentration of 100 μg ml⁻¹, and the fluorescence of H₂ probes in the bottom microglia was observed by confocal microscope.

Immunomodulation of ZSNP in the TNF-α–stimulated Transwell BBB model

The microglia in the lower chamber were stimulated with TNF-α at a concentration of 100 ng ml⁻¹ for 6 hours. Subsequently, PBS, edaravone, or ZSNP was added to the upper bEnd.3 monolayer chamber for an additional incubation of 24 hours, followed by further observation of inflammatory expression in the microglia present in the lower chamber. To evaluate the impact of ionic dissolution, extracts of ZSNP were also tested on microglia. Intracellular ROS levels were assessed using 10 μM 2ʹ,7ʹ-dichlorodihydrofluorescein diacetate (DCFH-DA) (MedChemExpress, HY-D0940, Ex/Em = 488/525 nm). For detection of the inflammatory marker, the microglia were fixed with 4% paraformaldehyde (PFA), washed with PBS, permeabilized with 0.1% Triton X-100, and subsequently incubated with iNOS (Abcam, ab15323, 1:200 dilution), ARG1 (NOVUS, NBP2-03618, 1:100 dilution), or γH2AX (CST, 9718 T, 1:300 dilution) antibodies. After that, the cells were stained using corresponding secondary antibodies, and fluorescence images of the stained cells were captured using a confocal microscope. The following secondary antibodies were used: Alexa Fluor 647–conjugated goat anti-mouse immunoglobulin G (IgG) H&L (Abcam, ab150115, 1:1000 dilution) and Alexa Fluor 488–conjugated goat anti-rabbit IgG H&L (Abcam, ab150081, 1:500 dilution).

Murine models of transient cerebral ischemia

The induction of transient cerebral ischemia was achieved by intraluminal occlusion of the left middle cerebral artery for a duration of 60 min as previously described (57). The sham-operated animals received identical anesthesia and arterial exposure without occlusion of the middle cerebral artery. Briefly, male C57BL/6 mice (10 weeks old, Slac Laboratory) were anesthetized with 2% isoflurane at a flow rate of 0.4 liter min⁻¹. The external carotid artery opening was cannulated with an 8-0 monofilament equipped with a silicon-coated tip, which was advanced through the internal carotid artery to reach the middle cerebral artery. The filament remained in place for 60 min to induce middle cerebral artery blood flow restriction. Throughout the surgical procedure, the mice’s body temperature was maintained at 37.0° ± 0.5°C using a temperature-regulated heating pad. The cerebral blood flow was assessed in all stroke animals using a laser speckle contrast imaging system (RWD, Shenzhen, China). Successful models were defined by achieving a 70% reduction in perfusion rate relative to the contralateral site during the tMCAO. Blood flow in the ischemic hemispheres was also monitored for the subsequent 7-day period. The intravenous administration of edaravone and ZSNP was initiated 4 hours after reperfusion. Unless otherwise specified, the following dosing regimen was applied: Edaravone was administered daily at a dose of 1 μg g⁻¹ body weight (b.w.), while ZSNP was administered every two other days at a dose of 10 μg g⁻¹ b.w. To compare ZSNP with inhaled H₂, a separate experiment was conducted: One group received 2% H₂ (v/v/ in air) for 2 hours daily over 3 days, while the ZSNP group received a single intravenous injection.

Animals used in this study were handled humanely and in accordance with established experimental protocols. All animals were obtained legally, and their housing, use, and care complied fully with applicable federal, state, and local regulations, as well as the guidelines of the Institutional Animal Care and Use Committee of East China University of Science and Technology for the Care and Use of Laboratory Animals. Proper husbandry practices were maintained throughout the study, including standardized housing conditions, controlled feeding regimens, and regular sanitation. The animal experiments were approved by the Animal Ethics Committee of East China University of Science and Technology (approval number: ECUST-2024-041). Animals were randomly assigned to experimental groups to ensure unbiased group allocation. All data were included in the analysis without exclusion, and the number of biological replicates is explicitly stated in the corresponding results sections.

In vivo targeting and distribution

The Cy5-labeled ZSN or ZSNP were intravenously administered via tail vein injection at 4 hours postreperfusion. After 6 hours of administration, transcardial perfusion was performed to eliminate unbound nanoparticles from peripheral blood using 20 ml of heparinized saline solution (0.9% NaCl containing heparin sodium at a concentration of 100 U ml⁻¹). Then, the fresh heart, liver, spleen, lung, kidney, and brain were collected and observed under fluorescence imaging system (IVISLumina III, PerkinElmer).

To confirm the inflamed endothelium targeting property, DyLight 594–conjugated Lycopersicon Esculentum (Tomato) lectin (1 mg ml⁻¹, 100 μl, Vector Laboratories, DL-1177) was intravenously injected after the administration of Cy5-labeled ZSN or ZSNP for 6 hours. After lectin injection for 30 min, transcardial perfusion with heparinized saline solution was conducted before immunofluorescence histology on the brain infarct region.

BBB permeability assessment

At day 3 post–ZSNP administration in tMCAO mice, 2% Evans blue solution was intravenously injected into the tail vein to label the circulating blood. After 30 min of circulation, the mice were anesthetized, and their brains were meticulously dissected after transcardial perfusion with heparinized saline solution. Then, digital photographs of the brains were recorded. To quantify the degree of BBB disruption, hemisphere samples were weighed and homogenized with 400 μl PBS, followed by overnight precipitation with 50% trichloroacetic acid (Sigma-Aldrich, T0699). Subsequently, all samples were centrifuged at 300g for 30 min to separate brain tissue pellets before measurement. The quantification of Evans blue in the brain was performed by measuring absorption at 625 nm using a plate reader (SpectraMax M3, Molecular Devices).

Histological and immunofluorescence analysis

For analysis of necrotic brain tissue, the fresh brain tissues were sectioned using a 1-mm brain matrix (RWD, 68707) and subsequently stained with 2% TTC (Sigma-Aldrich, T8877) in PBS at 37°C for 20 min until viable tissue exhibited a bright-red coloration. Digital images were captured using a Nikon camera, and infarct volume was quantified using ImageJ software. The proportion of infarct volume was calculated as follows: (contralateral uninfarct area − ipsilateral uninfarct area)/total contralateral uninfarct area × 100%.

For immunofluorescence, the mice were transcardially perfused with 20 ml of heparinized saline. Subsequently, fresh brains were harvested and fixed in 4% PFA for 4 hour at 4°C. After overnight immersion in cryoprotective solution consisting of 20% sucrose and 2% polyvinylpyrrolidone, all samples were embedded in OCT compound (Leica) and sectioned. Next, sections were permeated with 0.1% Triton X-100 for 20 min and then blocked with 5% goat serum at room temperature for 1 hour before incubation overnight at 4°C with the following antibodies: P-selectin (NOVUS Biologicals, NB100-65392SS, 1:100 dilution), Glut-1 (Abcam, ab652, 1:200 dilution), Iba-1 (WAKO, 016-26721, 1:500 dilution), GFAP (Invitrogen, 13-0300, 1:400 dilution), NF200 (Sigma-Aldrich, N4142, 1:500 dilution), Dcx (Cell Signaling Technology, 4604S, 1:500 dilution), laminin (Invitrogen, PA1-16730, 1:500 dilution), iNOS (Abcam, ab15323, 1:200 dilution), ARG1 (NOVUS Biologicals, NBP2-03618, 1:100 dilution), PDGFR-β (Novus Biologicals, NBP1-43349, 1:50 dilution), Aqua-4 (Santa Cruz Biotechnology, sc-32739, 1:50 dilution), Ki67 (eBioscience, 14-5698-82, 1:50 dilution), and c-Fos (Cell Signaling Technology, 2250S, 1:1000 dilution). Primary antibodies were then visualized with following species-appropriate secondary antibodies: Alexa Fluor 488–conjugated goat anti-rat IgG H&L (Jackson ImmunoResearch, 112- 545-062, 1:400), Dylight488-conjugated goat anti-mouse IgG H&L (Abcam, ab96879, 1:400), Alexa Fluor 568–conjugated goat anti-rabbit IgG H&L (Jackson ImmunoResearch, 111-575-144, 1:400), and Alexa Fluor 647–conjugated goat anti-mouse IgG H&L (Jackson ImmunoResearch, 115-605-003, 1:400). The sections were then mounted using ProLong Gold Antifade reagent with 4′,6-diamidino-2-phenylindole (CST, 8961S). In some experiments, Alexa Fluor 647–conjugated dextran (10 kDa, Invitrogen, D22914) was intravenously injected 30 min before euthanasia, and then the implants were retrieved for fluorescence labeling and further analysis. All immunofluorescence stainings were observed using confocal microscope.

For H&E and Nissl analysis, the mice were transcardially perfused with 20 ml of heparinized saline solution. Subsequently, the brains were harvested and fixed in 4% PFA for 72 hours, followed by dehydration, embedding, and sectioning into slices with a thickness of 7 μm. The sections were then subjected to H&E staining as well as Nissl staining (Solarbio, G1434).

Image reconstruction and morphology study

For morphology analysis of the Iba-1⁺ microglia and the spatial distribution of ZSNP in P-selectin–expressed vessels, Imaris (Bitplane) software was used to reconstruct 3D images. Briefly, Z stack confocal images were imported into Imaris software. After applying a smoothing value of 0.300 μm and setting a threshold to distinguish the target signal from background noise, a 3D structure was reconstructed. All images were captured using consistent camera and microscope settings and processed with identical adjustments and parameters.

Cytokine analysis

For cytokine analysis, on day 7 after conducting the tMCAO model, fresh brains were harvested, and the ischemic hemispheres were dissected and homogenized. Subsequently, 500 μl of PBS with 0.2% FBS was added and centrifuged to collect the supernatant. The levels of cytokines in the supernatant were analyzed using a series of enzyme-linked immunosorbent assay kits: IL-1α (Neobioscience, EMC009a), IL-1β (Neobioscience, EMC001b), IL-2 (Neobioscience, EMC002), IL-4 (Neobioscience, EMC003), interferon-γ (Neobioscience, EMC101g), and TNF-α (Neobioscience, EMC102a).

In vivo microglia depletion

For microglia depletion, PLX5622 was administered to mice through their diet (Biopike) at a concentration of 1200 ppm (equivalent to 1200 mg kg⁻¹ of chow), starting 7 days before tMCAO surgery and continuing until the completion of the experiments.

Flow cytometric analysis

For analysis of the effect of PLX5622 administration on microglial clearance in tMCAO mice, flow cytometric analysis was performed to determine the proportions of relevant cell populations at 7 days postsurgery. Briefly, single-cell suspensions were prepared from fresh infarcted brain tissue via enzymatic digestion using the Neural Tissue Dissociation Kit for 15 min at 37°C. The resulting suspension was filtered through a 40-μm strainer and washed with cell staining buffer (BioLegend, 420201). Murine cells were blocked with CD16/CD32 monoclonal antibody (eBioscience, 14-0161-81) for 30 min at 4°C, followed by staining with LIVE/DEA Fixable Near IR Viability Dye (Invitrogen, L34994), phycoerythrin-conjugated anti-CD45 (BioLegend, 147712), BV510-conjugated anti-Ter119 (BioLegend, 116237), BV421-conjugated anti-CD11b (BioLegend, 101235), and fluorescein isothiocyanate–conjugated anti-CD3 (BioLegend, 100204) for 45 min at 4°C. Samples were acquired using an LSRFortessa^T X-20 flow cytometer (BD Biosciences) and analyzed with FlowJo software (Tree Star).

RNA-seq and data analysis

The RNA-seq analysis was conducted by LC-Bio Technology Co. Ltd. Briefly, a single-cell suspension was obtained from fresh infarcted cerebral tissue through enzymatic digestion using the neural tissue dissociation kit at day 7 postsurgery. Subsequently, microglia were isolated using CD11b (Microglia) MicroBeads (Miltenyi Biotec., 130-093-634) following the manufacturer’s standard protocols. Total mRNAs were extracted and purified from the sorted microglia using TRIzol reagent (Thermo Fisher Scientific, 15596018CN), and cDNA and indexed cDNA libraries were prepared with the SMART-Seq Stranded Kit. Library construction and sequencing were performed on the Illumina 6000 sequencing platform at LC-Biotechnology. Approximately 50 million reads were obtained for each sample.

We used the HISAT2 package (version: HISAT2-2.2.1) to align reads from all samples to the reference genome, which initially filtered out low-quality reads based on accompanying quality information and subsequently mapped them to the reference genome. The assembled mapped reads of each sample were then processed using StringTie (version: stringtie-2.1.6) with default parameters. Subsequently, a comprehensive transcriptome was reconstructed by merging all transcriptomes from different samples using gffcompare software (version: gffcompare-0.9.8). Following the generation of the final transcriptome, expression levels for all transcripts and mRNAs were estimated using StringTie and ballgown (http://bioconductor.org/packages/release/bioc/html/ballgown.html), respectively, by calculating FPKM values. Differentially expressed mRNAs were identified using DESeq2 R package with fold change ≥ 2 or P value < 0.05 criteria, followed by performing GO enrichment analysis and KEGG pathway enrichment analysis for these differentially expressed mRNAs.

In vitro Transwell model for coculturing microglia and neurons

To analyze the impact of H₂-mediated microglia on neuronal synaptic expression, we performed immunostaining for synaptic markers in a coculture system consisting of microglia and neurons under H₂ conditions. After primary neurons reached maturity, a microglia-neuron coculture system was established as follows. Microglia were seeded into the upper Transwell insert at a density of 5 × 10⁴ cells per well. To model the inflammatory microenvironment of ischemic stroke, microglia were first prestimulated for 24 hours under one of the following conditions: LPS (100 ng·ml⁻¹, Sigma-Aldrich, L2630) alone, LPS combined with 5 μM MSAB (a Wnt/β-catenin inhibitor, MedChemExpress, HY-120697), or LPS combined with 20 μM KN-93 (a CaMKII inhibitor, MedChemExpress, HY-15465). Following stimulation, microglia were thoroughly washed and cocultured with neurons in the lower chamber. The coculture system was then maintained for 3 days in either a conventional incubator or a H₂ incubator. Neuronal integrity was evaluated through immunostaining of MAP2 (Sigma-Aldrich, M9942, 1:500 dilution) and SYP (Synaptic Systems, 101002, 1:500 dilution).

Behavioral assessment

The beam-walking test was conducted in accordance with established protocols (58). The mNSS was evaluated using a 5-point scale: 0 points indicated smooth traversal of the balance beam without any risk of falling; 1 point denoted crossing the beam with a chance of falling less than 50%; 2 points represented crossing the beam with a chance of falling greater than 50%; 3 points indicated successful traversal, but with impaired movement in the paralyzed hind leg; 4 points signified inability to cross the balance beam, yet ability to maintain sitting position on it; and last, 5 points described immediate loss of balance resulting in fall upon placing the mouse on the balance beam. The cylinder test/spontaneous forelimb task and grid-walking test were performed as previously described (59). For the pasta test, the mice underwent food restriction 1 day before testing and then were placed in a cylinder and given four pieces of uncooked vermicelli (4 cm in length, 1.5 mm in diameter, De Cecco Capellini). The time taken to consume one piece of pasta (in seconds) was recorded.

Molecular docking

The peptides were predicted using AlphaFold3 (60). The structural files used in molecular docking were processed using AutoDock Tools-1.5.7 (61). Molecular docking study of receptor with ligand was performed by autodock4. The docking grid was centered on the active site (center_x = 0, center_y = 0, center_z = 0) with dimensions of 50 Å by 50 Å by 50 Å, sufficiently covering each peptide. Docking calculations were performed with the number of output binding modes set to 100. The resulting poses were ranked according to predicted binding affinity, and the pose with the strongest affinity was selected for further analysis.

In vivo silence neurons with AAV9 injection and CNO administration

In a separate cohort, tMCAO mice treated with ZSNP were subjected to AAV9-hM4D/CNO–mediated neuronal silencing. The CNO/DREAAD system has been demonstrated to inhibit the action potential of transfected cells by activating outward K⁺ efflux and membrane hyperpolarization, thereby reducing neuronal excitability and neurotransmitter release to achieve functional silencing of transfected neurons (62). At 9 weeks poststroke, the mice were anesthetized and placed in a stereotaxic frame (RWD, Shenzhen, China). Following a craniotomy, 1 μl of AAV9-hM4D (OBIO Technology) was infused into the motor cortex of the ischemic hemisphere (anterior-posterior = 0.26 mm, medial-lateral = 3 mm, and dorso-ventral = 1 mm) at 0.1 μl/min. After a 3-week period for viral expression, neuronal silencing was achieved by intraperitoneal injection of CNO (5 mg kg⁻¹ b.w., TargetMol, T4494) at 12 weeks poststroke. Behavioral assessments were performed during the CNO-induced neuronal silencing window (0.5 to 2 hours postinjection) to evaluate the functional necessity of the ZSNP-induced circuits.

In vivo bidirectional axonal tracing with intracranial BDA injection

At 12 weeks poststroke, tMCAO mice treated with PBS, edaravone, or ZSNP were anesthetized with 2% isoflurane and placed in a stereotaxic frame. Following a craniotomy, 1 μl of 10% BDA (M_W = 10,000, Thermo Fisher Scientific, D1956) was microinjected into the motor cortex of the ischemic hemisphere (anterior-posterior = 0.26 mm, medial-lateral = 3 mm, and dorso-ventral = 1 mm) at a rate of 0.1 μl/min. After a 5-day period for axonal transport, the mice were transcardially perfused, and brains were harvested. The tissues were fixed, embedded in OCT compound, and sectioned into slices. BDA-labeled axons were finally visualized using streptavidin conjugated to Alexa Fluor 555 (Thermo Fisher Scientific, S32355, 1:400).

Systemic biosafety evaluation

To investigate the biodistribution of ZSNP, tMCAO mice administered a single dose were euthanized at designated time points (12 hours and 7, 14, and 28 days). The brain and other major organs were harvested and immediately freeze-dried to obtain dry weights. The lyophilized tissues were digested in freshly prepared aqua regia (a 1:3 v/v mixture of HNO₃ and HCl) at 37°C for 48 hours. The resulting solutions were evaporated to near dryness, and the residues were redissolved in 5% (v/v) HNO₃ for elemental analysis. Zirconium (Zr) content was quantified using ICP-MS (PerkinElmer, NexION 2000G) and normalized to the dry tissue weight, expressed as microgram Zr per gram of dry tissue. Healthy mice without ZSNP injection served as blank controls.

For in vivo biocompatibility assessment, tMCAO mice treated with ZSNP were euthanized 14 days posttreatment. Blood samples were collected for biochemical analysis of ALT, AST, ALP, CREA, and UREA using a biochemical analyzer (Olympus, AU640). Major organs (heart, liver, spleen, lung, and kidney) were harvested, processed for histological sectioning, and stained with H&E for examination.

Statistical analyses

The results were analyzed using GraphPad Prism software (version 9.4.1). All data were presented as means ± SD. Comparisons between two groups were performed using unpaired, two-tailed Student’s t tests. Comparisons among multiple groups were conducted using one-way analysis of variance (ANOVA) or two-way ANOVA, followed by Tukey’s post hoc test. In all analyses, a P value less than 0.05 was considered statistically significant. All the experiments were repeated three times independently with similar result.

Supplementary Materials

This PDF file includes:

Figs. S1 to S23