Aquaporin-4 deficiency induced white matter injury via upregulated complement component 3 in cerebral small vessel disease

Abstract

Background

Cerebral small vessel disease (CSVD) is the most common neurological disorder associated with a high incidence of stroke and dementia. For the heterogeneity of the pathogenesis, effective preventive and therapeutic strategies remain limited. Reduced aquaporin-4 (AQP4) was reported in the development of CSVD, while its role and mechanisms have not been fully elucidated.

Methods

We employed logistic regression analysis to assess the association between AQP4 gene single-nucleotide polymorphisms (SNPs) and CSVD presence. The functional impact of SNP mutations on AQP4 gene was evaluated using a Luciferase assay. Subsequently, Aqp4^−/− mice were subjected to bilateral carotid artery stenosis (BCAS) surgery to detect the CSVD phenotype of Aqp4 reduction. Aqp4^+/+ and Aqp4^−/− mice in the Sham and BCAS groups were subjected to MRI, histological examinations, and behavioural tests. AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} were used to investigate Aqp4 expression and translocation ability after BCAS. Furthermore, RNA-sequencing was performed on the corpus callosum (CC) and Aqp4^−/− astrocytes to identify transcriptional changes associated with Aqp4 deficiency.

Results

We identified four SNPs (rs335929, rs335930, rs335931 and rs455671) were significantly associated with CSVD presence. Among these, rs335929 variant was correlated with reduced AQP4 mRNA expression. Compared with Aqp4^+/+ mice, fractional anisotropy values were decreased in the Sham and BCAS groups in Aqp4^−/− mice. Under BCAS conditions, Aqp4^−/− mice exhibited more severe demyelination and myelin density in the CC. Intraventricular delivery of AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} attenuated BCAS-induced white matter injury and cognitive function, no significant differences were observed between the two AAV treatment groups. RNA sequencing analysis indicated upregulation of inflammatory responses and complement cascades in the CC of Aqp4^−/− mice. Complement component 3 (C3) mRNA was notably elevated in astrocytes isolated from Aqp4^−/− mice. Treatment with a C3a receptor antagonist in Aqp4^−/− mice improved myelin integrity and reduced MBP loss following BCAS.

Conclusion

Reduced AQP4 expression was associated with CSVD presence in clinical studies. Experimentally, Aqp4 deficiency has been shown to exacerbate white matter injury, that effect may be mediated by upregulation of C3 mRNA in Aqp4-deficient astrocytes. Targeting C3 activation may represent a promising strategy to mitigate AQP4 loss-induced white matter injury in CSVD.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12974-025-03688-w.

Background

Cerebral small vessel disease (CSVD) is an age-related neurological disorder characterized by pathological changes in small arteries, arterioles, capillaries, and venules within the brain [1]. It accounts for approximately 25% of all ischemic strokes and contributes to more than 20% of all dementia cases [2]. Currently, CSVD diagnosis relies primarily on neuroimaging features, including white matter hyperintensities (WMH), cerebral atrophy, lacunar infarcts, enlarged perivascular spaces (EPVS), and cerebral microbleeds (CMBs) [3]. The etiology of CSVD remains incompletely understood and is thought to be multifactorial, involving vascular endothelial dysfunction [4], blood-brain barrier disruption [5], impaired perivascular clearance [6], and neuroinflammation [7]. Given its complex pathogenesis, effective therapeutic options remain limited.

Aquaporin-4 (AQP4), is widely expressed on the astrocytic endfeet, regulating cerebrospinal fluid (CSF) homeostasis, metabolite clearance, neuroinflammation, cellular calcium influx and water transport in the nervous system [8–10]. Under physiological conditions, AQP4 trafficking to the astrocytic membrane follows a circadian rhythm [11]. However, in pathological states, both AQP4 protein synthesis and polarized localization are impaired. Reductions in astrocytic AQP4 expression and its mislocalisation have been linked to glymphatic dysfunction in mouse models of Alzheimer’s disease (AD) [12] and Parkinson’s disease (PD) [13]. In AD patients, mislocalised AQP4 exacerbates β-amyloid deposition and accelerates cognitive decline [14]. Mechanistically, AQP4 mislocalisation leads to stagnation of cerebral interstitial fluid, resulting in glymphatic impairment [15] and subsequent accumulation of metabolic wastes. Furthermore, loss of AQP4 promotes astrogliosis and microglial activation. In PD mouse models, nuclear factor kappa B (NF-kB) activity is upregulated during AQP4 deficiency [16], whereas in neuromyelitis optica spectrum disorder (NMOSD), loss of AQP4 triggers microglia-mediated complement component 3 (C3) upregulation and oligodendrocyte detachment [17, 18]. These findings suggest that AQP4 participates in astrocyte–oligodendrocyte crosstalk via inflammatory pathways.

In stroke-prone renovascular hypertensive rat (RHRSP) models, reduced perivascular Aqp4 expression has been observed [19]; however, the intrinsic mechanisms underlying Aqp4 reduction in CSVD remain unclear. In the present study, we investigated the effects of AQP4 gene expression on CSVD in humans using AQP4 single-nucleotide polymorphisms (SNPs). We then employed BCAS-treated Aqp4^−/− mice to assess CSVD-related imaging and histological alterations, and rescue therapies were performed using AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} delivery. Furthermore, we demonstrated that C3 upregulation in Aqp4-deficient astrocytes contributed to white matter injury following BCAS. Collectively, these findings suggest a novel therapeutic strategy targeting AQP4 loss-induced white matter injury in CSVD.

Methods

AQP4 gene and CSVD in clinical research

Study design

Discovery cohort: The study population was derived from the Third China National Stroke Registry (CNSR-III registry), a prospective registry study of hospitalized patients with acute ischemic cerebrovascular disease between August 2015 and March 2018 [20]. Participants in the CNSR-III study were recruited from mainland China, across 163 grade III and 38 grade II hospitals nationwide. In this registry, 10,914 patients were prespecified in the genetic sub-study that was undertaken for acquiring qualified blood samples for whole-genome sequencing (WGS) [21]. The WGS procedure and data quality-control process have been previously described [22]. Validation cohort:The study population was derived from the PolyvasculaR Evaluation for Cognitive Impairment and vaScular Events (PRECISE) cohort. The PRECISE cohort is a community-based prospective study conducted between 2017 and 2019 among older adults residing in Lishui, Zhejiang Province, China. A total of 3067 participants aged 50 to 75 years were enrolled from 6 villages and 4 communities [23]. The study aimed to investigate the prevalence of clinical and subclinical polyvascular disease in community-dwelling older adults and to examine its association with cognitive impairment, cerebrovascular events, and mortality.

AQP4 SNPs acquisition

In the WGS data of the CNSR-III study, genomic region from Chr18: 26,842,038 to Chr18: 26,875,803 was screened, which includes the human AQP4 gene along with 10 kilobase pairs of upstream and downstream sequence. All sequenced reads were aligned to the human reference genome GRCh38. Single nucleotide polymorphisms (SNPs) within this gene region were extracted, aligned, and subjected to quality control based on minor allele frequency (MAF). Variants with a MAF of less than 0.01 were excluded. Subsequently, eligible SNPs within the human AQP4 gene region were included for association analysis. Then, AQP4 gene SNPs (rs335929, rs335930, rs335931, rs455671 and rs162008) were acquired from the PRECISE study in the same method.

MRI data acquisition

A 3.0-T MRI scan was performed wherein the sequences included 3-dimensional T1-weighted magnetisation prepared by rapid acquisition gradient-echo, axial T2-weighted, fluid-attenuated inversion recovery, and axial susceptibility-weighted imaging. The MRI markers of CSVD were defined according to the Standards for Reporting Vascular Changes on Neuroimaging Criteria [24]. WMH was defined as the brightness on T2 images, and WMH severity was rated according to the Fazekas rating scale [25]. Lacunes were defined as rounded or ovoid lesions of CSF signal with a diameter of 3–20 mm. CMBs comprised rounded, hypodense lesions (2–10 mm) on a gradient-recalled echo- or susceptibility-weighted image. The total numbers of lacunes and CMBs were recorded. PVS was defined as small (< 3 mm) punctate or linear hyperintensities on T2 images and, in the basal ganglia, was rated using a semi-quantitative rating scale developed by the Edinburg group [26]. Each CSVD imaging marker was rated by two trained raters blinded to the patients’ clinical data. Images with inconsistent results were assessed by a senior neurologist who was blinded to the initial results. Based on the grading system above, we rated CSVD burden (Wardlaw CSVD burden score) on an ordinal scale from 0 to 4 by allocating one point to WMH burden (periventricular WMH Fazekas 3 or deep-WMH Fazekas 2–3), presence of lacunes, CMBs, and moderate-to-severe basal ganglia PVS (n = 10–20) [3]. Furthermore, we also evaluated the modified CSVD burden score using a recently validated score ranging from 0 to 6. In the modified CSVD burden score (Rothwell CSVD burden score), 1 point was allocated for the presence of lacunes, CMB burden (n = 1–4), severe basal ganglia PVS (n > 20), and modified WMH burden (total periventricular + subcortical WMH grade 3–4), and 2 points were allocated for CMB burden (n ≥ 5) and modified WMH burden (total periventricular + subcortical WMH grade 5–6) [27].

Clinical outcomes

In this study, the outcome variables included the CSVD presence, WMH, EPVS, CMB, and lacunar infarcts. The CSVD presence was defined using both the CSVD burden score (Wardlaw) and a modified CSVD burden score (Rothwell). Patients were categorized into the CSVD burden = 0 group (no CSVD manifestations) and the CSVD burden > 0 group (presence of CSVD manifestations) [28]. For the outcome of WMH, the criteria were set as follows: periventricular white matter hyperintensity at Fazekas grade 3 or deep white matter hyperintensity at Fazekas grade 2–3. Patients meeting either criterion were classified as having WMH; those not meeting these criteria were classified as without WMH. For the outcome of CMB, patients with ≥ 1 CMB were defined as having CMB, while those with no CMB were defined as without CMB. For the outcome of lacunar infarcts, patients with ≥ 1 lacunar infarct were classified as having lacunar infarcts, and those with 0 were classified as without lacunar infarcts. For the outcome of EPVS, patients with > 10 EPVS were defined as having moderate-to-severe EPVS, while those with ≤ 10 EPVS were defined as having none or mild EPVS.

Statistical analysis

Demographic variables of patients without (CSVD burden = 0) and with (CSVD burden ≥ 1) CSVD were compared using the Student’s t-test for continuous variables and the chi-square test for categorical variables (e.g., sex and medical history). Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

The association analyses of AQP4 SNPs with CSVD outcomes (CSVD presence, WMH, lacune, EPVS and CMBs) were performed using logistic regression models. Odds ratios (ORs) with their 95% confidence intervals (CIs) were calculated. The significance threshold for association was set at P < 0.05. For further analyses, correction for multiple comparisons was done by applying the false discovery rate (FDR) approach, and the result was presented as P_FDR. Statistical significance was set as a P_FDR < 0.05. Associations not reaching P_FDR < 0.05 but showing a P < 0.05 were also considered suggestive of an association.

The meta-analysis in this study was conducted using data from the PRECISE and CNSR-III cohorts, as both provided genetic data from Chinese populations. Heterogeneity between the two studies was first assessed to determine the appropriate model. The I² statistic was calculated: if I² > 50%, significant heterogeneity was considered present, and a random-effects model was selected; otherwise, a fixed-effects model was applied. Effect sizes and study weights were then pooled accordingly. As only two studies were included, no sensitivity analysis was performed.

Animal and cell experiments of the AQP4 gene in CSVD

Experimental animals and bilateral carotid artery stenosis surgery

C57BL/6J mice (aged 7-8-week; weighing 22–25 g) were purchased from Vital River. The male Aqp4 knockout (Aqp4^−/−) mice line was reared at Gempharmatech (China). The mice were housed in groups under a 12-h:12-h light: dark cycle. All experiments were approved by Beijing Tiantan Hospital, which was affiliated with the Capital Medical University Committee on Animal Resources, Beijing, China.

Bilateral carotid artery stenosis surgery (BCAS) was used as the animal model to mimic CSVD phenotype [29]. In the exploration of the influence of the AQP4 on the BCAS model, Aqp4^+/+ and Aqp4^−/− mice were subjected to either sham or BCAS surgery. The sham group served as the control, in which mice were anaesthetised, underwent a neck skin incision, and had their common carotid arteries exposed, but no microcoils were implanted. BCAS was performed on male 7–8-week-old mice that were anaesthetised with 2,2,2-tribromoethanol (120 mg kg^{− 1}, intraperitoneally [i.p.]) and placed in the supine position. The common carotid artery was carefully exposed through a midline incision. A microcoil (inner diameter 0.18 mm, pitch 0.50 mm, total length 2.5 mm; Sawane Spring Co.) was wrapped around the common carotid artery to induce chronic cerebral hypoperfusion, and 60 min later, the same procedure was performed on the other common carotid artery. Sham-operated mice underwent all procedures, except the microcoil implantation. One month after surgery, the mice were subjected to behavioural tests and MRI and sacrificed for the subsequent experimental evaluations.

In the evaluation of overexpressing AQP4^M1 or AQP4^{7×cMyc−Kras} delivery in sham or BCAS assignment, Aqp4^+/+ mice with AAV2/9-Gfap-Egfp were defined as control group, and mice with AAV2/9-Gfap-AQP4^M1-Egfp or AAV2/9-Gfap-AQP4^{7×cMyc−Kras}-Egfp delivery were defined as treatment group. The sham or BCAS procedure were the same in the previous part, and the AAV delivery procedure was detailed in AAV preparation and injection part.

For the C3aR antagonist rescue assay, Aqp4^−/− mice underwent BCAS surgery were applied to the placebo treatment or the C3 treatment. Each mouse was intraperitoneally administered 1 mg/kg C3a receptor antagonist (Merck Millipore, #559410) or 400 ul cold PBS daily for 30 days after BCAS. The treatment method was followed by a previous study [7].

The experimental unit in the animal part was one individual mouse. The sample size for each experiment was informed by our prior experience with the respective techniques and the availability of age-matched controls. Sample sizes were not predetermined using statistical methods. We randomly allocated the animals to the different treatment groups or the different surgery group within each experiment by allocating mice from different cages to the same experimental group.

Animal MRI

In the mouse MRI experiments, structural brain changes and corpus callosum damage were evaluated in Aqp4^+/+ and Aqp4^−/− mice following sham or BCAS surgery and subsequent AAV injection experiment. MRI was performed using an 11.7-T Bruker MR system. The mice were anaesthetised by isoflurane inhalation (1.5−2%) and monitored to maintain constant physiological parameters. T2-weighted images were acquired using 2D rapid acquisition with a relaxation enhancement sequence along with the following parameters: repetition time (TR), 2800 ms; echo time (TE), 30 ms; field of view (FOV), 20 × 16 mm; image size, 256 × 205; and 29 adjacent slices of 0.5-mm slice thickness. Diffusion-weighted images were acquired using a spin-echo echo-planar imaging sequence with the following parameters: two b-values (b = 0 and 1796 s/mm²) along 30 non-collinear directions; TR, 5000 ms; TE, 17 ms; FOV, 20 mm×16 mm; image size, 110 × 88; and 50 adjacent slices of 0.3-mm slice thickness. Imaging data were converted to the NIFTI format using MRIcron software. Fractional anisotropy (FA) maps were generated using FSL (version 5.0.9). The FA was calculated after eddy and motion corrections for each mouse in the FSL. Raw DTI data were transferred to the NIFTI format using DSI Studio (http://dsistudio.labsolver.org/) and processed using the FSL Software v6.0. The corpus callosum (CC), wherein white matter tracts are most abundant in the mouse brain, was selected as the ROI. Using the FSLSTATS function, representative FA values were calculated for these ROIs. MRI was performed by one researcher and MRI data was analysed by another researcher.

Immunofluorescence staining and quantification

Mice were anaesthetised with 2,2,2-tribromoethanol 120 mg kg^− 1 i.p., and intracardially perfused with PBS and 4% paraformaldehyde (PFA) in PBS. Mouse brains were dissected and fixed in 4% PFA for 24 h, then cryo-protected in 30% sucrose concentration in PBS and embedded in the optimal cutting temperature (OCT); compound brain Sect. (15-µm thick) were selected for staining from each mouse. The slices were permeabilised with 0.3% Triton-X-100 in PBS, blocked with Immunol staining blocking buffer (Beyotime Biotechnology, Shanghai, China, #P0102, 1:1), incubated with primary antibody overnight, washed three times in PBS, and incubated with fluorophore-linked secondary antibodies for 1 h at 25 °C. The primary antibodies used were goat anti-CD31 (R&D, #AF3628, 1:100), Aquaporin4 polyclonal antibody (Proteintech, #16473-1-AP, 1:500), and Myelin Basic Protein (Abcam, #ab40390, 1:100). The vessel density with CD31 immunostaining was captured using an Olympus confocal microscope under a 10 × field of view and analyzed statistically with Fiji software. For each mouse, staining images were acquired from three sections of the same brain region, and a consistent 400 × 400 pixels area was selected from each brain slice as the statistical sample. The fluorescence threshold was set to a range of 30–255, and the “Analyze Particles” tool was applied with a size range of 10–15 to infinity. Through data analysis, the count of CD31-positive blood vessels in the same brain region of each image was obtained. The mice myelin with MBP immunostaining was also captured using an Olympus microscope. For each mouse, images were taken from three staining sections of the mid-corpus callosum. A consistent area of 250 × 250 pixels in the mid-corpus callosum was selected for statistical analysis of immunofluorescence intensity, which was performed using Fiji software. For the AQP4 immunostaining quantification, images were captured using an Olympus microscope system. During image analysis, a consistent fluorescence threshold range of 30–255 was applied to each brain section image. Each mouse was captured from three sections of the same brain region, and each slice with a sampling interval of 120 μm between sections. For the AQP4 and vascular colocalization experiments, data acquisition was performed using Olympus microscope, and colocalization analysis was conducted with the Colocalization module in Fiji. All images were captured by a single investigator, and the analyses were performed by a separate investigator. All quantitative results of the assay are shown as the mean and the standard error of the mean (SEM) and were analysed using GraphPad Prism 9.5.0 software. Significant intergroup differences were determined using a two-tailed unpaired Student’s t-test; P-values of 0.05(*), 0.01(**), and 0.001(***) were set as the levels of significance for the statistical tests.

Glymphatic function assessment

Aqp4^+/+ mice underwent sham or BCAS surgery were applied to the assessment of glymphatic function. For the glymphatic influx assay, 5% FITC-CM-Dextran-4000Da (FITC-CM-Dextran, average mol wt 4,000, Sigma-Aldrich, #68059) in 10 µL of PBS was stereotactically injected into the cisterna magna for 5 min using a syringe pump (Hamilton, Bonaduz, Switzerland). The pump speed remained 2 µL/min. Syringe pump was linked with a 10 cm-length PE50 tube. And the tube was linked with an injector needle (Injector, Single / O.D, 0.30 mm-30G, RWD, #800-0288-01). The injector needle was injected into cisterna magna for 1 mm depth. After injection, the needle was kept in the cisterna magna with 1 cm length tube. After 15 min post injection, mice were sacrificed. Then the mice brain was observed under a Stereo Fluorescence microscope under 7 magnification.

For the glymphatic efflux assay, 0.5% Tetramethylrhodamine isothiocyanate-Detran (average mol wt 155 kDa; Sigma-Aldrich, #T1287) in PBS was stereotactically injected into striatum (AP: +0.8 mm from Bregma, ML: -2.0 mm and DV: -3.6 mm) using a syringe injector (701 N, volume 5 µL, needle size 26G, Hamilton). The pump speed remained 0.1 µL/ min. Injection volume was 1 µL. After injection, the syringe was kept in striatum for 5 min. After 1 h post injection, mice were sacrificed and then perfused with PBS and 4%PFA. Fixed for 24 h, mice brains were sliced into coronal sections (100 μm-thick). Six layers of mice brain were detected for a single animal, and three brain slices were used as three replicates for one brain layer. All brain slices were captured with a PE scanning microscopy. During detection, fluorescence intensity remained the same and were observed using the same threshold for 50–255. An average of fluorescence intensity and fluorescence positive area of 18 images per mice were calculated.

Luxol fast blue staining

Luxol Fast Blue (LFB) staining was used to assess myelin integrity. Staining was performed using a Luxol Fast Blue Myelin Stain Kit (Solarbio Science and Technology, Beijing, China, #G3245). LFB-stained sections were imaged using a Zeiss microscope system under 10× magnification. For each mouse, three regions in the mid-corpus callosum were captured. Average optical density (AOD) quantification was performed in Fiji: the mid-corpus callosum was outlined using the polygon selection tool, and the AOD value of the selected area was measured. Three sections per mouse were analyzed.

Luciferase assay and transfection

The pT7-hRluc-HSV TK-Fluc vector was constructed in Shanghai GeneBio Co., Ltd.; 1000 bp-length (+ 1− + 1000) of AQP4 3’UTR sequence, including the SNP rs335929 with reference allele (wild-type) or risk allele (variant), were cloned into the vector, which contained those downstream of the firefly luciferase reporter gene. HEK-293 cells were used as tool cells, which were cultured in 12-well plates and transfected with 1 µg pT7-hRluc-HSV TK-Fluc-rs335929-WT or pT7-hRluc-HSV TK promoter-Fluc-rs335929-MT. The pT7-hRluc-HSV TK-Fluc vector was used as a system control. Using a Dual-Luciferase Reporter Assay kit (#RG027; Beyotime Biotechnology, Shanghai, China), Luciferase activity was measured 24 h after transfection. The measurements were performed using a multifunctional microplate reader. Each experiment comprised four replicates.

AAV preparation and injection

AAV plasmid construction (pAAV2/9-Gfap-Egfp, pAAV2/9-Gfap-AQP4^M1, pAAV2/9-Gfap-AQP4^{7×cMyc−Kras}) and AAV production were performed using GENERAL BIOL (Anhui, China). Recombinant AAV expression plasmids were generated by GENERAL BIOL (Anhui, China), with 1 × 10¹³ vg/ml concentrations of AAV2/9-Gfap-Egfp, AAV2/9-Gfap-AQP4^M1, and AAV2/9-Gfap-AQP4^{7×cMyc−Kras}. Intracerebroventricular AAV injection was administered through a 10-µL Hamilton syringe with a 34G needle (ML: +1.0/− 1.0 mm, AP: − 0.8 mm, and DV: -2.0 mm, relative to the bregma). The intracerebroventricular injection was administered bilaterally at 1 µL/min for a final volume of 5 µL on one side. After the injection on one side, the needle was left in place for 5 min before the injection on the other side.

Animal behaviour

In the animal behaviour test, Aqp4^+/+ and Aqp4^−/− mice subjected to the sham or BCAS surgery, and the subsequent AAV delivery mice subjected to the sham or BCAS group were applied to the open field test (OFT) and novel object recognition test (NOR).

The open field test was used to evaluate locomotor activity and exploratory behaviour and was performed using an open-field experimental system. Mice were habituated to the experimental environment for 10 min and then tested for another 10 min. The total distance travelled, ratio of time spent in the outer ring to the total distance travelled, distance travelled in the outer zone, ratio of frequency spent in the outer ring, and total area were calculated.

The novel object recognition test was used to evaluate non-spatial recognition memory. Before the test, the mice were habituated individually for 10 min to allow them to open their arenas. For training, mice were allowed to explore two matching objects placed in the arena for 10 min. The mouse-movement curves were captured using a computer-interfaced camera. The NOR test was performed 6 h after training. The mice were allowed to explore the novel object for 10 min. The recognition index (novel number / (novel number + familiar number)) was calculated to indicate the novel exploration ability.

The experimenter conducting the behavioral tests was blinded to the group allocation of the mice. The only identifying information during testing was the ear tag code.

CC and astrocyte bulk-RNA sequencing

For the CC RNA sequencing, Aqp4^+/+ mice and Aqp4^−/− mice underwent one-month sham or BCAS surgery were used as samples. Total RNA from the CC of 12-week-old male mice was extracted using the RNeasy Mini Kit and quantified using a Nanodrop2000 (Thermo Scientific). High-throughput RNA-seq was performed using Illumina HiSeq 2500 (Illumina, San Diego, CA, USA) at MajorBio (Shanghai, China). Raw sequencing data were aligned to the mouse reference genome (GRCm38/mm10). The htseq count function was used to accumulate the number of aligned reads that fell under the exons of the gene to determine the expression of each gene. PCA was used to determine the sample homogeneity. Differential gene analysis was performed using the DESeq2 package. Differentially expressed genes (DEGs) were identified with a false discovery rate of 1%. Genes were classified according to an adjusted P-value < 0.005 and log₂FC ≥ 1. GO and Kyoto Encyclopaedia of Genes and Genomes analyses were performed to determine the biological functions of the DEGs.

For the astrocyte RNA sequencing (RNA-seq), total RNA from astrocytes sorted from adult WT and Aqp4^−/− mice was extracted and quantified as described previously. High-throughput RNA-seq was performed using Illumina HiSeq 2500 (Illumina, San Diego, CA, USA) at MajorBio (Shanghai, China). Furthermore, the DEGs were classified according to the adjusted P < 0.001 and log2FC ≥ 1. The other analytical methods were the same as those described in the previous paragraph of CC RNA sequencing.

Membrane staining

Twelve-well plates and glass sheets were coated with poly-D-lysine (Beyotime Biotechnology, #ST508) for 1 h before use. The glass sheets and wells were washed three times with PBS. The 293T cells (after transfection with AQP4 plasmids) were plated overnight on glass sheets to ensure that they were fixed on glass slides, and then the 293T cells were fixed with 4% PFA for 20 min and washed three times with PBS. The cell membrane was stained with Vybrant Dil dye (Invitrogen, #V22885, 1:200) for 15 min at 37 ℃ and washed with PBS three times before taking images.

Recombinant AQP4 lentiviral plasmid construction and transfection

Plasmids were generated by GENERAL BIOL (Anhui, China) and Sangon Biotech (Shanghai, China). pLenti-EF1α-Egfp, pLenti-EF1α-AQP4^M1-Egfp, pLenti-EF1α-AQP4^M23-Egfp, pLenti-EF1α-AQP4^S276D, pLenti-EF1α-AQP4^Kras-BSD, and pLenti-EF1α-AQP4^{7xcMyc − Kras} were constructed. HEK-293T cells and U87 cell lines were used as tool cells, which were cultured in 12-well plates and transfected with 1 µg of recombinant AQP4 plasmids. Lipofectamine 3000 (Invitrogen, #L300075) was used as the transfection agent per well following the manufacturer’s protocol. After 48 h of transfection, AQP4 overexpression was observed using confocal microscopy.

Astrocyte sorting

Astrocytes were isolated from five 8-week-old Aqp4^+/+ and Aqp4^−/− mice using ACSA-2 (Miltenyi Biotechnology, Anti-ACSA-2 Microbead Kit, mouse, #130-097-678) magnetic-associated cell separation. Five mice were used in each test. Mice brains were dissected and prepared as single-cell suspensions using a High-Activity Adult Brain Enzymatic Digestion Kit (Mouse and Rat) (RWD Life Science, #DHABE-5003). A single-cell suspension was incubated with ACSA2⁺ microbeads at 4 °C for 15 min. The cell suspension was loaded onto an MS Magnetic Column (Miltenyi Biotech) and placed in the field of a magnetic MACS Separator. ACSA2⁻ cells were washed off, while the ACSA2⁺ astrocytes were retained. After collection in an EP tube, ACSA2⁺ astrocytes were counted using an automatic cell counter for descending experiments.

Oligodendrocyte progenitor cell sorting and recombinant C3 stimulation test

Oligodendrocyte progenitor cells (OPC) were isolated from 6 to 7 day wild-type mice followed by PDGFRα magnetic-associated cell separation (Miltenyi Biotechnology, PDGFRα Microbead Kit, mouse, #130-101-502).

In the recombinant C3 stimulation part, OPCs were pre-seeded in 12-well plates with 1 mL of differentiation medium per well. OPCs were stimulated with recombinant human Complement component 3 protein (rhC3, Sigma-Aldrich, #204885) at a concentration of 100 ng/µL. The rhC3 or PBS was added to the cell culture wells (six replicates per group). On the second day of differentiation, the differentiation medium was replaced, and rhC3 or PBS stimulation was repeated. RNA sample was collected from the cells after four days post differentiation.

RNA extraction and quantitative PCR

Total RNA was extracted from mouse brain tissue and astrocytes using the FastPure Cell/Tissue Total RNA Isolation Kit (Vazyme, #RC112-01). Quantitative real-time PCR analyses were performed using the SYBR Green PCR master mix on an ABI 7500 PCR instrument. The relative quantification of genes was analysed by the 2^−ΔΔCt method and normalised with GAPDH. The primer sequences used were as follows:

Gene name	GeneBank accession no.	Primer sequences (5’-3’)
Aqp4 (Mus)	11,829	F: ATCAGCATCGCTAAGTCCGTC
		R: GAGGTGTGACCAGGTAGAGGA
Gapdh (Mus)	14,433	F: AGGTCGGTGTGAACGGATTTG
		R: GGGGTCGTTGATGGCAACA
Mbp (Mus)	17,196	F: ACACACGAGAACTACCCATTATGG
		R: AGAAATGGACTACTGGGTTTTCATCT
C3(Mus)	12,266	F: GAGCGAAGAGACCATCGTACT
		R: TCTTTAGGAAGTCTTGCACAGTG
Cspg4 (Mus)	121,021	F: ACCCAGGCTGAGGTAAATGC
		R: ACAGGCAGCATCGAAAGACA
Plp (Mus)	18,823	F: CCCACCCCTATCCGCTAGTT
		R: CAGGAAAAAAAGCACCATTGTG
Mog (Mus)	17,441	F: CTGTTTGTTATTGTGCCTGTTCTTG
		R: AGTCTTCGGTGCAGCCAGTT

Statistical analysis

All animal data were analysed with GraphPad Prism v.9 and presented as mean ± SEM. For two-sample comparisons, the Student’s t-test was applied. One-way ANOVA was used to compare multiple samples. Statistical significance was set at P < 0.05. All mice statistical analysis was performed by a researcher didn’t engage in any mice experiments. Statistical analyses of the human data were performed using population statistics.

Results

AQP4 reduction is associated with CSVD presence in the population

The flow chat of the clinical part was shown in Fig. 1A. In the CNSR-III genetic subgroup of AQP4 gene analysis, data were available for assessing WMH in 8469 patients, CMBs in 4782 patients, lacunar infarcts in 8427 patients, EPVS in 8327 patients, CSVD burden score (Wardlaw criteria) and modified CSVD burden score (Rothwell criteria) were available for 4740 patients. Baseline data for 4740 patients included in this study were detailed in Additional file 1: Table S1 (Wardlaw criteria) and Additional file 1: Table S2 (Rothwell criteria). Based on the Wardlaw burden score, patients were categorized into 1575 individuals with CSVD burden score = 0 and 3165 individuals with CSVD burden score > 0. Among them, patients with CSVD burden score = 0 had a mean age of 58.7 ± 11.4 years, with male patients accounting for 64.6%. Patients with a CSVD burden score > 0 had a mean age of 63.9 ± 10.6 years, with male patients accounting for 70.5%. The baseline characteristics were similar in modified CSVD burden score.

Fig. 1.

Reduced AQP4 expression is associated with the presence of CSVD. A Schematic depicting the study design. B Forest plot of meta-analysis of association between AQP4 SNPs with the presence of CSVD (Rothwell criteria). C Physical map of the AQP4 gene and the location of the 4 SNPs. D Map of the dual luciferase reporter plasmid design and Luciferase assay results. ***P < 0.001; SNP, single nucleotide polymorphism

In the CNSR-III study, 37 SNPs were identified within the AQP4 gene region (Additional file 1: Table S3). Analysis based on the Wardlaw criteria for the presence of CSVD (Additional file 1: Table S4) revealed six loci significantly associated after adjusting for age and gender: rs335929 (OR, 1.14 [95% CI 1.00–1.30]), rs335930 (OR, 1.15 [95% CI 1.01–1.32]), rs335931 (OR, 1.14 [95% CI 1.01–1.30]), rs455671 (OR, 1.14 [95% CI 1.003–1.30]), rs162008 (OR, 1.14 [95% CI 1.01–1.30]), and chr18_26869996 (OR, 0.75 [95% CI 0.57–0.98]). When applying the Rothwell criteria (Additional file 1: Table S5), eight loci showed significant associations, including rs335929, rs335930, rs335931, rs455671, rs63514, rs162009, rs162008 and chr18_26869996. Following multiple testing correction, only rs335930 remained significantly association with the presence of CSVD (P_FDR = 0.04). To minimize the potential oversight during exploratory analyses, we still regard carriers of the rs335929 (C), rs335930 (C), rs335931 (G), rs455671 (G) and rs162008 (T) variants as having an elevated likelihood of developing CSVD, whereas carriers of chr18_26869996 (A) had a reduced risk. These SNPs were therefore carried forward as candidates for further validation of their association with the presence of CSVD.

In the validation cohort (PRECISE study), analyses of the association between AQP4 SNPs and CSVD presence included 2918 subjects. The baseline characteristics of subjects with CSVD burden score = 0 versus those with score > 0 subjects were detailed in Additional file 1: Table S6 (Wardlaw criteria) and Additional file 1: Table S7 (Rothwell criteria). Due to the differences in sequencing depth relative to the CNSR-III study, Chr18_26869996 variant was not observed in the PRECISE cohort. Furthermore, rs335929 (C), rs335930 (C), rs335931 (G), rs455671 (G) and rs162008 (T) variants did not show significant associations with the presence of CSVD in this cohort (Additional file 1: Table S8 and Table S9).

Given the relatively low proportion of CSVD cases in the PRECISE cohort (30.3% by the Wardlaw criteria and 40.9% by the Rothwell criteria), which limited the statistical power, we conducted a meta-analysis pooling data from both cohorts. In the combined analysis of the CNSR-III and PRECISE studies, rs335931 (OR, 1.10 [95% CI 1.00–1.22]) and rs162008 (OR, 1.10 [95% CI 1.00–1.21]) showed a borderline association with CSVD presence (Wardlaw criteria) (Additional file 1: Fig. S1). While the meta-analysis revealed significant associations between CSVD presence (Rothwell criteria) and rs335929 (OR, 1.13 [95% CI 1.03–1.25]), rs335930 (OR, 1.16 [95% CI 1.04–1.28]), rs335931 (OR, 1.15 [95% CI 1.04–1.27]), and rs455671 (OR, 1.15 [95% CI 1.04–1.27]) (Fig. 1B).

Then, we examined how the identified AQP4 SNPs influence AQP4 gene expression. The human AQP4 gene (18q11.2; AQP4), with 323 amino acids, is composed of five exons and four introns [30]. Previous linkage disequilibrium analysis indicated strong correlations among the SNPs rs335929, rs335930, rs335931 and rs455671 [31]. Among these, rs335929 resides in the 3’untranslated region (UTR) of the AQP4 gene, whereas the other three SNPs are situated in the intron region (Fig. 1C). To assess the impact of rs335929 mutation, we performed a dual-luciferase reporter assay. The wild-type AQP4 3’UTR sequence (carrying the A genotype at rs335929) and a mutant 3’UTR sequence (carrying the C genotype) spanning positions + 1 to + 1000 were cloned downstream of a firefly luciferase reporter, with Renilla luciferase serving as an internal control. The assay revealed that the rs335929 mutation (carrying the C genotype) reduced AQP4 mRNA expression to approximately 70% of that observed with the wild-type sequence (carrying the A genotype) (P < 0.001, Fig. 1D), indicating that this SNP contributes to decreased AQP4 expression. Together, these findings suggest that reduced AQP4 expression was associated with CSVD presence in the studied population.

For further investigating the role of AQP4 in CSVD, we examined the associations between AQP4 SNPs and CSVD imaging markers (Table 1). In the meta-analysis combing data from both cohorts, rs335930 and rs335931 showed borderline associations with WMH, and rs455671 had a borderline association with lacune. In the CNSR-III study, rs335930 was associated with WMH (OR, 1.19 [95%CI 1.03–1.38]). Collectively, these findings suggest that AQP4 reduction might be associated with WMH in patients with CSVD.

Table 1.

Association between AQP4 SNPs and MRI imaging markers

SNP	Cohort	WMH		CMBs		Lacune		EPVS
		OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value
rs335929	CNSR-III study	1.13 (0.98–1.30)	0.09	1.07 (0.93–1.22)	0.36	1.06 (0.97–1.16)	0.22	1.02 (0.93–1.13)	0.64
	PRECISE study	1.05 (0.85–1.30)	0.64	1.04 (0.81–1.34)	0.74	1.36 (0.96–1.91)	0.08	0.96 (0.74–1.25)	0.76
	Meta-analysis	1.10 (0.98–1.23)	0.12	1.06 (0.94–1.19)	0.34	1.07 (0.98–1.17)	0.12	1.01 (0.92–1.11)	0.83
rs335930	CNSR-III study	1.19 (1.03–1.38)	0.02	1.11 (0.97–1.28)	0.13	1.07 (0.97–1.17)	0.17	1.03 (0.93–1.14)	0.56
	PRECISE study	1.01 (0.81–1.25)	0.94	0.94 (0.73–1.21)	0.62	1.24 (0.88–1.75)	0.23	1.04 (0.80–1.36)	0.77
	Meta-analysis	1.11 (0.99–1.25)	0.07	1.06 (0.94–1.20)	0.31	1.07 (0.98–1.17)	0.15	1.02 (0.93–1.12)	0.71
rs335931	CNSR-III study	1.15 (1.00-1.32)	0.058	1.08 (0.95–1.24)	0.25	1.06 (0.97–1.16)	0.20	1.02 (0.92–1.13)	0.69
	PRECISE study	1.06 (0.86–1.31)	0.59	1.05 (0.82–1.35)	0.69	1.33 (0.94–1.87)	0.11	1.01 (0.78–1.32)	0.93
	Meta-analysis	1.11 (0.99–1.25)	0.07	1.07 (0.96–1.21)	0.24	1.08 (0.99–1.17)	0.10	1.02 (0.93–1.11)	0.73
rs455671	CNSR-III study	1.12 (0.98–1.29)	0.11	1.07 (0.94–1.23)	0.30	1.07 (0.98–1.71)	0.13	1.03 (0.93–1.14)	0.60
	PRECISE study	1.05 (0.85–1.29)	0.68	1.04 (0.81–1.33)	0.79	1.27 (0.90–1.78)	0.17	0.94 (0.72–1.22)	0.63
	Meta-analysis	1.09 (0.98–1.23)	0.12	1.07 (0.95–1.20)	0.28	1.08 (0.99–1.18)	0.07	1.02 (0.93–1.11)	0.75

WMH White matter hyperintensity, CMBs Cerebral microbleeds, EPVS Enlarged perivascular space, OR Odds ratio, CNSR-III China National Stroke Registry-III, PRECISE PolyvasculaR Evaluation for Cognitive Impairment and vaScular Events

Aqp4^-/- mice were vulnerable to white matter injury

Based on the previous finding that AQP4 reduction was likely to associate with WMH, we used Aqp4^−/− mice to model AQP4 deficiency in vivo and subjected them to BCAS to simulate CSVD. Structural imaging revealed no significant changes between Aqp4^−/− and Aqp4^+/+ mice (Fig. 2A). White matter integrity was then assessed by measuring FA values in the CC. In the sham group, mean FA value was 0.45 ± 0.007 in Aqp4^+/+ mice and 0.43 ± 0.005 in Aqp4^−/− mice, with the latter showing a statistically significant reduction (P = 0.01; Fig. 2B). Following 1-month BCAS, mean FA values decreased to 0.39 ± 0.012 in Aqp4^+/+ mice and 0.35 ± 0.010 in Aqp4^−/− mice. Respectively, this decline was more pronounced in Aqp4^−/− mice (P = 0.02), indicating their vulnerability to white matter injury.

Fig. 2.

Aqp4^−/− mice were vulnerable to white matter injury after BCAS treatment. A Representative T2 and DEC maps of mice on the 30th day after BCAS in Aqp4^+/+-Sham, Aqp4^+/+-BCAS, Aqp4^−/−-Sham and Aqp4^−/−-BCAS groups. The area within the white dashed lines represents the corpus callosum of the mouse. B FA value of corpus callosum in Aqp4^+/+-Sham, Aqp4^+/+-BCAS, Aqp4 ^−/−-Sham and Aqp4^−/−-BCAS group (n = 10–14 per group). C Representative LFB staining images of corpus callosum. Comparisons of average optical density of the medial of corpus callosum between four groups (n = 4 per group). Scale bar: 200 μm. D Representative MBP immunostaining images of corpus callosum in four groups. Comparisons of MBP mean intensity in the medial of corpus callosum between four groups (n = 5 per group). Scale bar: 250 μm. E Representative images and comparisons of cortical CD31 immunostaining in four groups (n = 4–5 per group). Scale bar: 100 μm. F In the open field test, representative trajectory chart of each group was shown. Comparisons between total distance travelled and ratio of movement in the out ring and total area were analysed among the groups (n = 17–22 per group). G In the novel object recognition test, total distance travelled and recognition index (novel object / (novel + old object) exploring frequency %) were analysed (n = 10–12 per group). *P < 0.05, **P < 0.01, ***P < 0.001. DEC map, directionally encoded color map; BCAS, bilateral common carotid artery stenosis; FA, fractional anisotropy; LFB, Luxol Fast Blue; MBP, myelin basic protein

We next evaluated myelin density and integrity in the CC histologically. Based on the results of LFB staining, a post-BCAS sparseness of the myelin sheath was noted (P = 0.007 in Aqp4^+/+ mice and P < 0.001 in Aqp4^−/− mice; Fig. 2C). Even under sham conditions, Aqp4^−/− mice exhibited a trend toward less compact myelin, though this difference was not statistically significant (P = 0.20). Following BCAS, myelin was significantly sparser in Aqp4^−/− mice compared with Aqp4^+/+ mice (P = 0.02). Similarly, MBP expression in the CC region was lower in sham-operated Aqp4^−/− mice, albeit without reaching significance (P = 0.08, Fig. 2D). After BCAS, Aqp4^−/− mice presented more severe demyelination than Aqp4^+/+ mice (P = 0.04). Vascular integrity, assessed by CD31 staining in cortical brain sections (Fig. 2E), showed that BCAS reduced CD31⁺ vessel density in both Aqp4^+/+ (P = 0.02) and Aqp4^−/− mice (P < 0.001); however, compared with Aqp4^+/+ mice, Aqp4^−/− mice did not experience significant vascular loss (P = 0.07). These observations indicate that AQP4 deficiency confers greater vulnerability to white matter damage during ischaemia and hypoxia.

The OFT showed no significant differences in total distance travelled across the four groups (Fig. 2F), indicating that Aqp4 reduction does not influence motor ability in mice. Assessment of emotional variation revealed no significant anxiety-like behavior in Aqp4^−/− mice, either with or without BCAS (P = 0.16 and P = 0.25). In the NOR test, no significant differences in the total distance travelled were detected across the four groups either. However, Aqp4^−/− mice displayed a mild recognition deficit relative to Aqp4^+/+ mice in the sham (P = 0.057) and BCAS (P = 0.23) conditions. Notably, Aqp4^−/− mice subjected to BCAS showed a significantly lower recognition index than sham-operated Aqp4^+/+ mice (P = 0.003, Fig. 2G), indicating increased susceptibility to cognitive impairment following BCAS.

These experimental observations align with the clinical association between reduced AQP4 expression and WMH, demonstrating that Aqp4^−/− mice develop more severe white matter injury under CSVD conditions.

BCAS mice lost AQP4 translocation

Previous study has reported the reduced perivascular Aqp4 expression in CSVD rat model [19], whether BCAS model is a nice model to explore impaired Aqp4 expression and function of CSVD is not known. In this way, we investigated Aqp4 expression in the mice cortex, CC, and striatum tissue (Additional file 1: Fig.S2A−C). We observed that Aqp4 expression in the cortex and CC significantly reduced after BCAS treatment, whereas striatal Aqp4 expression remained unaltered. Aqp4 mRNA levels changed in a manner consistent with that of protein levels (Additional file 1: Fig. S2D). Furthermore, in our 1-month BCAS model, we found that the co-localisation between Aqp4 and vascular tissue was reduced (Fig. 3A). Given that glymphatic function is closely associated with AQP4 translocation [32], we performed glymphatic influx and efflux assays [33]. Through cisterna magna injection, we found that the glymphatic influx was significantly reduced after BCAS (P = 0.01 in mean intensity and P = 0.008 in the mouse brain area; Additional file 1: Fig.S2E). We did not observe any alterations in the glymphatic efflux assay in the 1-month BCAS model (Additional file 1: Fig. S2F).

Fig. 3.

Design of membrane-anchoring AQP4 plasmid. A Representative co-immunostaining of AQP4 (green) and lectin (red, indicating vascular) of wide type mice in Sham and BCAS group. Co-localization assay of AQP4 and vascular in Sham and BCAS group. Scale bar: 100μm. B Membrane-anchored plasmid design and instant transfection assay in 293T and U87 cell line. pLenti-EF1α-Egfp, pLenti-EF1α-AQP4^M1, pLenti-EF1α-AQP4^M23, pLenti-EF1α-AQP4^S276D, pLenti-EF1α-AQP4^Kras, pLenti-EF1α-AQP4^7×cMyc-Kras was constructed. Scale bar: 40μm. C Membrane staining of 293T after transfection of wide type AQP4^M1 and engineered AQP4^7×cMyc-Kras plasmid. Membrane was staining by DID dye (Red). Scale bar: 10μm. D Schematic plot of AAV delivery. AAV2/9-Gfap-Egfp, AAV2/9-Gfap-AQP4^M1, AAV2/9-Gfap-AQP4^7×cMyc-Kras was applied to mice through bilateral ventricular injection (AP: -0.8 mm, ML: +1.00/-1.00 mm, DV: -2.0 mm). GFAP was co-immunostaining with GFP in mice brain post 14^th day injection. Scale bar: 40μm

Given the importance of AQP4 polarity in neurodegenerative diseases [14], we therefore sought to determine whether AQP4 polarity or expression level is more critical in CSVD. To address this question, we designed a membrane-anchored AQP4 plasmid to specifically investigate the functional consequences of polarized AQP4 localisation. A previous study reported that the AQP4^M23 type was membrane-localised [34]; therefore, we transfected the pLenti-EF1α-AQP4^M23 plasmids into 293T cells and U87 cells directly and found that the fusion protein could not completely translocate to the cell membrane (Fig. 3B). Subsequently, we applied the point mutation of AQP4 at 276 sites, which facilitates AQP4 protein translocation to the membrane after phosphorylation [35]. Previous studies reported that the K-ras gene with a CAAX motif and a polybasic domain of six consecutive lysine residues comprised signals for the subcellular localisation of proteins to the membrane [36]. Therefore, we cloned 17 C-terminal amino acids of K-ras onto the C terminus of the AQP4 gene [37]. To detect the membrane-anchoring function of the engineered AQP4, we inserted a 7-folded C-Myc tag sequence (EQKLISEEDL) in the first extracellular loop of AQP4. The engineered plasmid maps and structures were shown in Additional file 1: Fig.S3. The 293T and U87 cell lines showed the same level of AQP4 expression after transfection with the wild-type and engineered plasmids (Fig. 3B). pLenti-AQP4^{7×cMyc−Kras} was the best membrane-anchoring type. The in vitro co-localisation assay results were shown in Fig. 3C, where pLenti-EF1α-AQP4^{7×cMyc−Kras} colocalised with the cell membrane prior to the wild-type pLenti-AQP4^M1 structure. Then, we performed intraventricular injections of the engineered AAV to assess its in vivo effects. Brain sections stained for AQP4 and GFAP following injection were presented in Additional file1: Fig.S4A and S4B. GFAP immunostaining confirmed that EGFP was successfully expressed in astrocytes after treatment with AAV2/9-Gfap-Egfp. Moreover, both AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} delivery resulted in AQP4 expression in the perivascular space (Fig. 3D). These results demonstrated that AQP4^{7×cMyc−Kras} was effectively localized to the membrane both in vivo and in vitro.

Intraventricular delivery of AAV2/9-Gfap-AQP4^M1 rescued post-BCAS white matter injury

Following AAV delivery, BCAS was performed 14 days post-injection to evaluate whether AQP4 overexpression or membrane-anchored AQP4 could mitigate BCAS-induced white matter injury (Fig. 4A and B and Additional file 1: Fig.S4C). The control group received AAV2/9-Gfap-Egfp. One month after BCAS, CC integrity was improved in both the AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} treatment groups (Fig. 4C). FA values were significantly higher in the AAV2/9-Gfap-AQP4^M1 group (0.426, P = 0.03) and the AAV2/9-Gfap-AQP4^{7×cMyc−Kras} group (0.425, P = 0.04) compared to the control group, with no significant difference between the two experimental groups (P = 0.96). Myelin density, assessed by LFB staining (Fig. 4D), showed significantly increased average optical density (AOD) in the CC region of both AAV2/9-Gfap-AQP4^M1 (P < 0.001) and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} (P = 0.048) groups relative to controls under BCAS conditions. Although FA values and myelin density were slightly elevated in the sham group, the difference was not statistically significant. To further evaluate demyelination, we performed MBP immunostaining. Quantification of MBP expression revealed that neither AAV2/9-Gfap-Aqp4^M1 nor AAV2/9-Gfap-Aqp4^{7×cMyc−Kras} treatment significantly reduced demyelination in BCAS mice (Fig. 4E). NOR test showed that both AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} delivery led to improved cognitive performance in BCAS mice (Fig. 4F; P = 0.002 for each group). In summary, AAV2/9-Gfap-AQP4^M1 delivery ameliorated white matter injury and cognitive deficits following BCAS.

Fig. 4.

Overexpression of AQP4 helped rescue white matter injury in BCAS. A A schematic map of the AAV treatment experiment. B Representative image of GFP in cortex in AAV2/9-Gfap-Egfp, AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} group in Sham and BCAS groups. Scale bar: 40 μm. C Comparisons between FA value of corpus callosum in AAV2/9-Gfap-Egfp, AAV2/9-Gfap-AQP4^M1, AAV2/9-Gfap-AQP4^{7×cMyc−Kras} in both Sham and BCAS group (n = 8–10 per group). D Representative images and comparisons of corpus callosum LFB immunostaining between six groups (n = 3–4 per group). Scale bar: 200 μm. E Representative images and comparisons of corpus callosum MBP immunostaining in six groups (n = 3 per group). Scale bar: 100 μm. F Total distance travelled and recognition index (novel / (novel + old) exploration time %) were compared in the six groups. *P < 0.05, **P < 0.01, ***P < 0.001

Blockade of C3 activation reduced white matter injury in Aqp4^-/- mice post-BCAS

To investigate the mechanism by which AQP4 reduction contributes to white matter injury, we extracted CC tissues from Aqp4^+/+ and Aqp4^−/− mice in the sham and BCAS groups for RNA sequencing (Fig. 5A). In Aqp4^+/+ mice, significant alterations in gene expression were observed between sham and BCAS groups, with 484 genes upregulated and 230 genes downregulated (Fig. 5B). Inflammatory and immune pathways were markedly activated in the BCAS group, whereas pathways related to cognition, learning, and memory pathways were downregulated. In Aqp4^−/− mice, immune pathways were upregulated, while those associated with axon terminus and neuron projection were significantly suppressed, showing 545 upregulated and 44 downregulated genes, respectively (Fig. 5C); indicating greater susceptibility to neuronal loss. Further comparison between Aqp4^+/+ and Aqp4^−/− mice revealed enhanced activation of the NF-kB pathway, JAK-STAT signalling, and complement and coagulation cascades in Aqp4^−/− mice (Fig. 5D), suggesting pronounced immune and inflammatory responses post-BCAS. Among these, C3 expression was significantly elevated in Aqp4^−/− mice under the sham and BCAS conditions (Fig. 5E). qPCR confirmed upregulation of C3 mRNA in the CC from Aqp4^−/− mice (Fig. 5F, P = 0.002).

Fig. 5.

C3 upregulation contributed to white matter injury in Aqp4^-/- mice. A RNA sequencing of mice corpus callosum of Aqp4^+/+ and Aqp4^-/- in Sham and BCAS group (n=3 per group). PCA analysis of mice corpus callosum. B Heatmaps of differential expressed genes in corpus callosum between Sham and BCAS group in Aqp4^+/+ mice. KEGG terms were annotated in each gene cluster. C Heatmaps of differential expressed genes in corpus callosum between Sham and BCAS group in Aqp4^-/- mice (n=3 per group). KEGG terms were annotated in each gene cluster. D Comparison of KEGG enrichment pathways based on the regulating gene cluster between Aqp4^-/- and Aqp4^+/+ mice in BCAS group. E Heatmap of differential expressed genes in corpus callosum of Aqp4^+/+-Sham, Aqp4^+/+-BCAS, Aqp4^-/--Sham and Aqp4^-/--BCAS groups were shown (n=3 per group). F mRNA expression of C3 in Aqp4^-/- and Aqp4^+/+ mice in Sham group. G Schematical depict of mice oligodendrocyte progenitor cell differentiation test under recombinant human C3 treatment. Cspg4 and Mog mRNA expression with or without C3 treatment (n=6 per group). rhC3, recombinant human Complement component 3. H Ratio between Cspg4 mRNA expression and Mog mRNA expression. I Experiment design of C3aR antagonist on Aqp4^-/- mice under BCAS. Representative images and quantitative comparisons of corpus callosum in LFB staining (n=4-5 per group) and MBP immunostaining between C3aR antagonist-treated or untreated group (n=3 per group). Scale bar for LFB staining: 100μm. Scale bar for MBP immunostaining: 100μm. C3aRA, C3a receptor antagonist. *P < 0.05, **P < 0.01

Given that complement activation triggers oligodendrocyte death in NMO models [38], we hypothesized that increased C3 expression may contribute to white matter injury in Aqp4^−/− mice. Based on reported roles of C3 in myelination [39], we treated OPCs with rhC3 in vitro (Fig. 5G). Cspg4 mRNA levels showed an increasing trend after 4 days of rhC3 treatment (P = 0.13), and Mog mRNA levels were significantly reduced in rhC3-treated group (P = 0.044). The ratio of Cspg4 to Mog mRNA was elevated in the rhC3-treated group (P = 0.02; Fig. 5H), suggesting a shift toward OPC persistence over oligodendrocyte differentiation. To assess whether Aqp4^−/− mice exhibited impaired oligodendrocyte maturation, we examined oligodendroglia lineage markers in the CC, hippocampus, and cortex separately (Additional file 1: Fig. S5A and S5B). The CC of Aqp4^−/− mice presented significantly lower Mog expression (P = 0.007) and a non-significant increase in Cspg4 mRNA (P = 0.40). While Cspg4 / Mog ratio was higher in Aqp4^−/− mice in the CC (P = 0.02), which indicated disrupted oligodendrocyte maturation, potentially explaining the myelin integrity deficits in Aqp4^−/− mice. Administration of a C3aR antagonist to Aqp4^−/− mice under BCAS conditions (Fig. 5I) ameliorated myelin damage, as shown by improved LFB staining (P = 0.003) and increased MBP expression (P = 0.04).

To identify the source of elevated C3 expression, we considered prior evidence that microglial ablation did not reduce C3 in an NMO model [17], and that astrocytes upregulated C3 upon exposure to NMO patient serum [40]. We therefore isolated astrocytes from Aqp4^+/+ and Aqp4^−/− mouse using MACS-based separation (Additional file 1: Fig.S6A). RNA sequencing revealed 989 upregulated and 240 downregulated genes in Aqp4^−/− astrocytes. Specifically, C3 mRNA was significantly upregulated elevated (Additional file 1: Fig.S6B) and KEGG enrichment highlighted complement and coagulation activation (Additional file 1: Fig.S6C). Subsequent qPCR confirmed astrocyte-specific C3 upregulation in Aqp4^−/− astrocytes (Additional file 1: Fig.S6D, P < 0.001). These findings support that astrocytic C3 upregulation, driven by AQP4 deficiency, promotes white matter injury.

Discussion

Our study provides evidence linking AQP4 downregulation to the presence of CSVD in humans. Supporting a pathogenic role, Aqp4 knockout induced white matter injury in CSVD mouse model. Notably, delivering AAV2/9-Gfap-AQP4^M1 could mitigate this injury and help restore myelin integrity. Further analysis revealed a potential mechanism, as astrocytic C3 mRNA expression was significantly upregulated in Aqp4-deficient astrocytes, which may contribute to white matter injury-associated CSVD.

Given the established involvement of AQP4 in glymphatic regulation, impaired glymphatic function has been reported in clinical studies of CSVD patients [41]. In our study, no significant decline in glymphatic efflux function was observed in Aqp4^+/+ mice one month after BCAS modeling. This discrepancy may be due to differences in the timing of observation, which could influence the detection of glymphatic activity. It is also noteworthy that previous studies found no statistically significant difference in glymphatic function between Aqp4^−/− and Aqp4^+/+ mice [11], suggesting that AQP4 deficiency may not directly lead to glymphatic dysfunction, or that compensatory mechanisms may remain active in mice aged 8–12 weeks. In this study, we propose that although BCAS model mice show decreased AQP4 protein expression and loss of polarization, the resulting changes in glymphatic efflux function were not markedly evident.

Notably, Aqp4^−/− mice exhibited a significant decrease in MRI-derived FA values and a reduction in the AOD value based on LFB staining, indicating a pronounced loss of myelin integrity and volume. In the assessment of vascular injury, we found no significant differences in vascular impairment between Aqp4^−/− and Aqp4^+/+ mice. Furthermore, no notable differences in vascular damage were observed between the two groups after BCAS. Taken together, these findings indicated that at one-month time point, vascular injury was less pronounced than white matter injury in the BCAS model, irrespective of AQP4 status. This suggests that AQP4 deficiency-induced white matter injury may precede blood-brain barrier disruption.

Both AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^{7×cMyc−Kras} delivery ameliorated white matter injury in the CC and improved cognitive function following BCAS, although neither significantly reduced the extent of demyelination. While AAV2/9-Gfap-AQP4^{7×cMyc−Kras} did not outperform AAV2/9-Gfap-AQP4^M1 delivery as assessed by LFB staining and MBP immunostaining, both vectors supported myelin repair in the BCAS model. We propose that AQP4 deficiency impairs the capacity for myelin formation, thereby delaying reparative processes after BCAS, though further investigation is required. Importantly, although loss of perivascular AQP4 was observed in the BCAS model, AQP4-based interventions alleviated white matter injury independently of its membrane anchoring function. These findings suggest that AQP4 expression, rather than its proper translocation to the astrocytic membrane, underlines white matter injury in BCAS. Thus, we postulate that in the BCAS mice model, the white matter injury might be more attributable to reduction of AQP4 gene expression rather than to its mislocalisation. Given previous reports linking impaired AQP4 translocation to glymphatic function in ageing-related neurodegeneration disorders, we further suggest that AAV2/9-Gfap-AQP4^{7×cMyc−Kras} delivery may represent a promising strategy for restoring glymphatic function in these conditions.

In AQP4 antibody-positive NMOSD, loss of astrocytic AQP4 leads to demyelination in the spinal cord and optic nerve [42]. Although astrocyte-microglial crosstalk has been implicated in C3 upregulation during demyelination [17, 43], one study reported that this alteration persisted even after microglial ablation in vivo. In the present study, we classified astrocytes from Aqp4^−/− and Aqp4^+/+ mice and observed that C3 mRNA was significantly upregulated in Aqp4^−/− astrocytes in vivo. Consistent with this, an in vitro assay showed increased C3 protein levels in culture of primary mice astrocyte treated with NMOSD patient serum [43], suggesting that loss of astrocytic AQP4 promotes glial activation. Similar to the astrocytic AQP4 loss in NMOSD, our results confirmed that C3 mRNA upregulation is related to astrocytic AQP4 reduction. AQP4 has been linked to astrocytic swelling and dysregulation of cytosolic Ca²⁺ [44, 45], which may initiate inflammatory responses and complement activation in astrocytes; however, the exact mechanism requires further investigation. Given that C3 was reported to induce white matter injury following hypoperfusion [7, 46], and is necessary for neurodegeneration [47]. As Aqp4^−/− mice had dysfunctional myelin formation, we propose that AQP4 deficiency might disrupt the differentiation of OPC into oligodendrocytes. Our experiments further revealed that OPC differentiation is adversely affected by C3 exposure. We therefore conclude that astrocytic C3 upregulation contributes to white matter injury in Aqp4^−/− mice. Since clinical delivery of AQP4 via AAV vectors remains challenging, C3aR antagonists may represent a promising therapeutic strategy for mitigating white matter injury associated AQP4 loss in CSVD.

This study has certain limitations. First, in the clinical study part, the association analyses were confined to Chinese population. Considering the heterogeneity in allele frequencies across ethnic groups, the findings may have limited generalizability to other populations. Additionally, in the PRECISE cohort, the associations between the key SNPs and CSVD burden were failed to reach due to the limited sample size. Therefore, to enhance statistical power, we performed a meta-analysis combining data from the CNSR-III and PRECISE cohorts. As the standardised procedures for information collection and criteria for CSVD imaging were consistent across both studies, which ensured the reliability of the association results to some extent. Second, although white matter injury was detected in our BCAS model, other CSVD phenotypes—including CMBs, lacunes, and EPVS—were not observed. Thus, the relationship between AQP4 and these specific pathological features remains to be determined. Third, a significant distance travelled was observed in mice administered AAV2/9-Gfap-AQP4^M1 and AAV2/9-Gfap-AQP4^7×cMyc-Kras delivery. The mice were likely to be in a state of irritation with increased AQP4 expression, although hydrocephalus, brain oedema and liver or renal impairment were not explored. Therefore, AQP4-based gene therapies for the treatment of these diseases requires additional safety tests.

Conclusions

In conclusion, our study demonstrates that decreased AQP4 expression contributes to the pathogenesis of white matter injury-associated CSVD. That C3 mRNA upregulation in Aqp4^−/− astrocytes suggests that C3 may serve as a potential therapeutic target for treating white matter injury with AQP4 deficiency in CSVD.

Abbreviations

Aqp4

Aquaporin-4

CSVD

Cerebral small vessel disease

CNSR-III

Third China National Stroke Registry

PRECISE

Poly-vasculaR Evaluation for Cognitive Impairment and vaScular Events

WMH

White matter hyperintensities

EPVS

Enlarged perivascular space

CMB

Cerebral microbleed

C3

Complement component 3

CC

Corpus callosum

BCAS

Bilateral common carotid artery stenosis

MRI

Magnetic resonance imaging

KO

Knockout

Mbp

Myelin basic protein

Mog

Myelin oligodendrocyte glycoprotein

Cspg4

Chondroitin sulfate proteoglycan

Authors’ contributions

YC, LLJ, YLW and YSP contributed to the conception and design of the study. All authors (YC, LLJ, DXY, DDL, YSP and YLW) contributed to the acquisition and analysis of data. YC and LLJ contributed to preparing the tables and figures, drafting the manuscript. YLW and YSP provided critical feedback and helped shape the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by grants from The National Natural Science Foundation of China (No. 82425101), the Beijing Municipal Science & Technology Commission (No. Z231100004823036), Capital’ Funds for Health Improvement and Research (2022-2-2045), and the National Key R&D Program of China (2024YFC3044800, 2022YFF1501500, 2022YFF1501501, 2022YFF1501502, 2022YFF1501503, 2022YFF1501504, and 2022YFF1501505).

Data availability

The data that support the findings of this study are available from the corresponding author, upon reasonable request.

Declarations

Ethics approval and consent to participate

The protocol of the CNSR-III study was approved by ethics committee at Beijing Tiantan Hospital (IRB approval number: KY2015-001-01) and all participating centres. The protocol of the PRECISE study was approved by ethics committee at Beijing Tiantan Hospital (IRB approval number: KY2017-010-01) and lishui hospital (IRB approved number: 2016-42).

Consent for publication

All authors have approved this manuscript.

Competing interests

The authors declare no competing interests.